flood warning, forecasting and emergency response ||
TRANSCRIPT
Flood Warning, Forecasting
and Emergency Response
Kevin Sene
Flood Warning, Forecasting and Emergency Response
ISBN 978-3-540-77852-3 e-ISBN 978-3-540-77853-0
Library of Congress Control Number: 2008927074
© 2008 Springer Science + Business Media B.V.
No part of this work may be reproduced, stored in a retrieval system, or transmitted in any form or by any
means, electronic, mechanical, photocopying, microfilming, recording or otherwise, without written
permission from the Publisher, with the exception of any material supplied specifically for the purpose
of being entered and executed on a computer system, for exclusive use by the purchaser of the work.
Printed on acid-free paper
9 8 7 6 5 4 3 2 1
springer.com
Kevin Sene
United Kingdom
Preface
This book provides an introduction to recent developments in the area of flood
warning, forecasting and emergency response. The topic spans a wide range of
disciplines, including weather forecasting, meteorological, river and coastal
detection systems, river and coastal flood forecasting models, flood warning dis-
semination systems, and emergency response procedures. The text deals mainly with
general principles and concepts, but also includes references to a number of manu-
als, guidelines and papers which provide more detailed information on factors to
consider in designing and implementing a flood warning system.
Although informal flood warning systems have existed ever since people settled
near to rivers and coastlines, improvements to communication and computer systems
in recent years have opened up a range of possibilities in many aspects of the flood
warning process. These include developments in remote sensing techniques, ensemble
forecasting, automated flood warning systems and decision support systems. Some
recent research and operational developments in these areas are discussed, although
specific brands of equipment (software, instrumentation etc.) are not considered. The
topics of performance monitoring, risk based design and prioritisation of investment
are also considered in several chapters, with recent developments driven in part by ris-
ing public expectations, and by an increasing need for organisations to justify invest-
ments in new equipment and procedures.
Early warning systems are often described in terms of the detection, warning
dissemination, response, recovery and review stages. In many cases, a forecasting
component will also be included, and preparedness is essential for an effective
emergency response. This structure is also adopted here, although with only a short
discussion of the recovery phase, since flood warning and forecasting has a less
important role to play once flood levels start to recede, such as estimating when
floodwaters will drain, or if any further flooding is imminent. By contrast, the
warning aspect is discussed in several locations, including a chapter on the decision
criteria used for issuing flood warnings (often called thresholds) and sections on
decision support and decision-making under uncertainty.
The book is presented in three main sections as follows:
● Part I – Flood Warning, which discusses the topics of detection, thresholds and
dissemination
v
● Part II – Flood Forecasting, which discusses general principles, specific types of
river and coastal forecasting models, and examples of specific applications
● Part III – Emergency Response, which covers the topic of preparedness,
response and review
The types of flooding which are discussed include river flooding, coastal surge,
snowmelt, ice-jams, urban drainage, flash flooding, and geotechnical risks, such
as Tsunami, dam breaks, and debris flows. The impacts of tropical cyclones, hur-
ricanes and typhoons are also discussed from a flooding perspective, although the
meteorological aspects are only considered briefly. Examples of operational sys-
tems are also provided from several countries, which in places has led to a need to
decide on the most appropriate terminology to use. So, for example, the term
catchment is used to describe what in some countries is known as a river basin or
watershed, the term cell phone describes mobile or cellular phones, and the term
flood defence is used in place of the terms levees or dikes. A glossary provides
more detail on the terminology used.
Although the book is primarily about real time flood warning, forecasting and
emergency response, some of the techniques described have evolved from those
used in other applications, such as flood simulation, water resources, hydromete-
orology, and ocean modelling, and may be of wider interest. The main difference
in flood warning applications is the requirement for rapid decision making, often
with incomplete or uncertain information. Supporting tools, such as forecasting
models, also need to operate sufficiently quickly and reliably to be of value in the
process, again often with less input data than would be available in simulation
modeling, although with the option of updating outputs in real time to help to
correct for differences between observed and forecast values. There is also often
a greater emphasis on the resilience of systems, and on documenting any design,
operational and other decisions made during model operation. These differences
all add an interesting dimension to this diverse and wide ranging subject.
vi Preface
Acknowledgements
This book has benefited from discussions with many people. Following several years
working in fluid mechanics, I joined the Centre for Ecology and Hydrology in
Wallingford (formerly the Institute of Hydrology) which provided the opportunity to
work on a wide range of research and consultancy projects on flood-related, hydromete-
orology, water resources and hydrometry topics in more than twenty countries. The
many discussions with colleagues during that time provided a useful grounding for the
topics discussed in this book.
Subsequently, as part of a large engineering consultancy, I have had the benefit
of many meetings, site visits and discussions with operational staff as part of flood
warning and forecasting improvement projects and strategies, and on projects to
develop best practice guidelines in flood forecasting for the Environment Agency
and SEPA.
In a rapidly developing field such as flood warning, forecasting and emergency
response, much information can also be obtained from internet searches, and many
organizations place conference proceedings, reports, manuals, and other useful
documents in the public domain. In presenting figures, references and quotations
from internet and published sources, both the publisher and myself have attempted
to identify and provide citations to the appropriate sources, although we apologise
if there have been any unintentional errors.
Many people assisted with providing comments on short extracts from the draft
text and providing figures, and I hope that I have included their comments accu-
rately. Michael Robbins and Steve Jebson from the Met Office, and Ian Marshall
and Hazel Phillips from the Environment Agency, were also very helpful with my
requests to use a range of figures and tables in the book.
I am also grateful to a number of colleagues for discussing aspects of the text,
or providing figures, including Marc Huband, Nick Elderfield, Jayne Lamont, Tom
Rouse, and Graham Clark. Also, Yiping Chen for many useful discussions on
hydraulic modelling for real time forecasting applications, and Jonathan Wright for
his general advice and support.
Finally, from Springer, I would like to thank Robert Doe and Nina Bennink for
their help and advice throughout the process of writing the book and bringing it to
production.
vii
Additional organizations that I would like to thank include:
● Environment Agency for Figures 3.5, 4.2, 8.7, 9.3 and Tables 3.1, 3.3, 5.1, 5.5,
and 6.1
● Federal Emergency Management Agency (FEMA) for Figure 10.1
● Her Majesty’s Stationary Office for the text cited in Box 9.2
● KNMI, Royal Netherlands Meteorological Institute for Figure 3.1
● Met Office for Figures 2.1, 2.3 and 2.4
● NOAA/National Weather Service for Figures 7.5 and 7.6
● Proudman Oceanographic Laboratory/National Tidal and Sea Level Facility for
Figures 7.3 and 7.4
● Royal Meteorological Society for Figure 2.2
● Scottish Hydraulics Study Group for Table 8.1
● STOWA for Figures 10.4 and 10.5
● World Meteorological Organisation for Figures 1.2, 1.3, 3.4, 3.6, 7.1, 7.2, 8.2,
8.5 and 11.1
Note that any text/material regarding TCP/WMO does not imply the expression/
endorsement of any opinion whatsoever on the WMO Secretariat concerning the
legal status of any country, territory, city or area or of its authorities, or concerning
the delimitation of its frontiers or boundaries.
Also, I would like to thank the following people for assisting with providing
figures, or comments on draft text:
● S. Baig, National Hurricane Centre, for comments on Box 7.3
● C. Carron, Environment Agency, for comments on Box 8.2
● P. Durrant for Figure 10.3
● K. Horsburgh, Proudman Oceanographic Laboratory, for comments on Box
7.2
● H. Lewis, North Cornwall District Council, for comments on Box 10.1
● J. Nower, Environment Agency, for comments on Box 8.2
● T. Peng, World Meteorological Organisation, for comments on Box 7.1
● A. Richman, Virtual Environmental Planning, for Figure 9.4
● B. Stewart, Bureau of Meteorology, for comments on Boxes 1.1, 7.1 and 8.1
● K. Stewart, Urban Drainage and Flood Control District (UDFCD), for Figure 4.4
● A. Tyagi, World Meteorological Organisation, for comments on Boxes 1.1 and 8.1
● D. Vogelezang and colleagues, Royal Netherlands Meteorological Institute
(KNMI), for comments on Box 3.1 and for providing Figure 3.1
● H. Vreugdenhil and colleagues, STOWA, for comments on Box 10.2
● D. Whitfield, Environment Agency, for comments on Boxes 4.1 and 5.1
viii Acknowledgements
Contents
Preface ............................................................................................................. v
Acknowledgements ........................................................................................ vii
1 Introduction .............................................................................................. 11.1 The Flood Warning Process .............................................................. 1
1.2 The Nature of Flood Risk ................................................................. 8
1.2.1 Flooding in Context .............................................................. 8
1.2.2 Assessing Flood Risk ............................................................ 9
1.3 Emergency Response ........................................................................ 13
1.4 The Role of Flood Forecasting ......................................................... 15
Part I Flood Warning
2 Detection ................................................................................................... 212.1 Meteorological Conditions ............................................................... 21
2.1.1 Site Specific Observations .................................................... 22
2.1.2 Remote Sensing .................................................................... 28
2.1.3 Weather Forecasting ............................................................. 33
2.2 River and Coastal Conditions ........................................................... 36
2.2.1 River/Tidal Level Monitoring ............................................... 37
2.2.2 River Flow Monitoring ......................................................... 39
2.2.3 Wave Monitoring .................................................................. 42
2.3 Instrumentation Networks ................................................................. 44
2.3.1 Telemetry Systems ................................................................ 44
2.3.2 Network Design .................................................................... 47
3 Thresholds ................................................................................................ 513.1 Rainfall Thresholds ........................................................................... 51
3.2 River and Coastal Thresholds ........................................................... 56
3.2.1 Introduction ........................................................................... 56
3.2.2 Simple Forecasting Techniques ............................................ 61
3.3 Performance Monitoring ................................................................... 67
ix
4 Dissemination ........................................................................................... 714.1 Flood Warning Procedures ............................................................... 71
4.1.1 Introduction ........................................................................... 71
4.1.2 Flood Warning Areas ............................................................ 73
4.1.3 Organisational Issues ............................................................ 75
4.1.4 Control Rooms ...................................................................... 77
4.2 Dissemination Techniques ................................................................ 79
4.2.1 Introduction ........................................................................... 79
4.2.2 Role of Information Technology .......................................... 81
4.2.3 Warning Messages ................................................................ 84
4.3 Design and Implementation .............................................................. 87
Part II Flood Forecasting
5 General Principles .................................................................................... 935.1 Model Design Considerations ........................................................... 93
5.2 Forecasting Systems ......................................................................... 97
5.3 Data Assimilation ............................................................................. 104
5.3.1 Error Prediction ..................................................................... 106
5.3.2 State and Parameter Updating ............................................... 107
5.3.3 Other Techniques .................................................................. 108
5.4 Model Calibration and Performance ................................................. 108
5.4.1 Basic Concepts ...................................................................... 108
5.4.2 Model Calibration ................................................................. 110
5.4.3 Performance Measures .......................................................... 113
5.5 Model Uncertainty ............................................................................ 114
6 Rivers......................................................................................................... 1236.1 Model Design .................................................................................... 123
6.1.1 Forecasting Requirement ...................................................... 124
6.1.2 Data Availability ................................................................... 126
6.1.3 Type of Model ....................................................................... 128
6.2 Rainfall Runoff Models .................................................................... 132
6.2.1 Introduction ........................................................................... 132
6.2.2 Process-Based Models .......................................................... 135
6.2.3 Conceptual Models ............................................................... 137
6.2.4 Data-Based Methods ............................................................. 139
6.3 River Channel Models ...................................................................... 141
6.3.1 Introduction ........................................................................... 141
6.3.2 Process Based Models ........................................................... 142
6.3.3 Conceptual Models ............................................................... 145
6.3.4 Data Based Methods ............................................................. 146
7 Coasts ........................................................................................................ 1497.1 Model Design Issues ......................................................................... 149
7.2 Process-Based Models ...................................................................... 156
x Contents
7.2.1 Astronomical Tide Prediction ............................................... 156
7.2.2 Surge Forecasting .................................................................. 157
7.2.3 Wave Forecasting .................................................................. 165
7.2.4 Shoreline Processes ............................................................... 167
7.3 Data-Based Methods ......................................................................... 169
7.3.1 Artificial Neural Networks ................................................... 169
7.3.2 Other Techniques .................................................................. 171
8 Selected Applications ............................................................................... 1758.1 Integrated Catchment Models ........................................................... 175
8.1.1 Introduction ........................................................................... 175
8.1.2 Modelling Approach ............................................................. 177
8.1.3 Ungauged Inflows ................................................................. 178
8.2 Flash Flood Forecasting .................................................................... 181
8.3 Snow and Ice ..................................................................................... 185
8.3.1 Snowmelt Forecasting ........................................................... 185
8.3.2 River Ice Forecasting ............................................................ 188
8.4 Control Structures ............................................................................. 190
8.4.1 Dams and Reservoirs ............................................................ 190
8.4.2 River Control Structures ....................................................... 195
8.4.3 Tidal Barriers ........................................................................ 198
8.5 Urban Drainage ................................................................................. 199
8.6 Geotechnical Risks ........................................................................... 202
8.6.1 Structural Risks ..................................................................... 203
8.6.2 Earth Movements .................................................................. 205
Part III Emergency Response
9 Preparedness ............................................................................................. 2099.1 Flood Emergency Planning ............................................................... 209
9.1.1 General Principles ................................................................. 209
9.1.2 Risk Assessments .................................................................. 214
9.1.3 All-Hazard Approaches ........................................................ 217
9.1.4 Validation and Testing of Plans ............................................ 219
9.2 Resilience .......................................................................................... 220
9.2.1 Introduction ........................................................................... 220
9.2.2 Analysis Techniques ............................................................. 224
9.3 Role of Information Technology ...................................................... 226
9.3.1 Introduction ........................................................................... 226
9.3.2 Geographical Information Systems ....................................... 227
9.3.3 Visualisation and Simulation ................................................ 228
10 Response .................................................................................................. 23110.1 Flood Event Management ............................................................. 231
10.1.1 Preparatory Actions ........................................................ 231
10.1.2 Timelines ........................................................................ 234
Contents xi
10.2 Decision Support Systems ............................................................ 237
10.3 Dealing with Uncertainty .............................................................. 244
11 Review ..................................................................................................... 24911.1 Performance Monitoring ............................................................... 249
11.2 Performance Improvements .......................................................... 253
11.2.1 Detection ......................................................................... 254
11.2.2 Thresholds ....................................................................... 255
11.2.3 Dissemination ................................................................. 256
11.2.4 Forecasting ...................................................................... 257
11.2.5 Preparedness ................................................................... 258
11.2.6 Response ......................................................................... 258
11.3 Prioritising Investment .................................................................. 260
11.3.1 Cost Benefit Analysis ..................................................... 261
11.3.2 Multi Criteria and Risk Based Analysis ......................... 265
Glossary .......................................................................................................... 267
References ....................................................................................................... 275
Index ................................................................................................................ 299
xii Contents
Chapter 1Introduction
Recent flood events have shown the devastating impact that flooding can have on
people and property. Flood warning and forecasting systems can help to reduce the
effects of flooding by allowing people to be evacuated from areas at risk, and to
move vehicles and personal possessions to safety. With sufficient warning, tempo-
rary defences can also be installed, and river and tidal control structures operated
to mitigate the effects of flooding. Many countries and local authorities now operate
some form of flood warning system, and the underlying technology requires knowledge
across a range of technical areas, including rainfall and tidal detection systems,
river and coastal flood forecasting models, flood warning dissemination systems, and
emergency response procedures. This introductory chapter provides a general overview
of the flood warning process, approaches to flood forecasting and emergency
response, and the nature of flood risk.
1.1 The Flood Warning Process
Flood warning systems provide a well-established way to help to reduce risk to life, and
to allow communities and the emergency services time to prepare for flooding
and to protect possessions and property. Actions may also be taken to reduce or prevent
flooding; for example, by operating river control structures, and floodfighting activities
such as reinforcing flood defences, and installing temporary or demountable barriers.
Informal flood warning systems have existed ever since people started to live and work
near rivers and coastlines. Heavy rainfall, high river levels, unusual sea states and other
cues, such as the sound of running water, all provide useful information on impending
flooding, with traditional methods for providing warnings including word of mouth, mes-
sengers, and raising flags and storm cones. These approaches still have a valuable role to
play, particularly where flooding develops rapidly, and communities must rely on their
own resources for the initial response. For example, in remote parts of Australia, farmers
may alert others further downstream if river levels are high or flooding has started
(Emergency Management Australia 1999) and, following the December 2004 Tsunami,
several community leaders were praised for recognising the abnormal sea conditions and
issuing an alert in time to prevent major loss of life (e.g. UNESCO 2006).
K. Sene, Flood Warning, Forecasting and Emergency Response, 1
© Springer Science + Business Media B.V. 2008
2 1 Introduction
The use of a more technological approach started to become widespread with
the introduction of telegraph transmission of river levels in the mid to late 19th
century in countries such as the USA, France and Italy (e.g. Smith and Ward
1998), followed by telephone and radio telemetry early in the 20th century, and
accelerated in the 1950s and 1960s as the computer and electronic industries
developed. Developments have included the introduction of operational computer
models of the atmosphere (from the 1950s), weather radar and satellite based
observations of rainfall (from the 1970s), and automated and internet based meth-
ods of warning dissemination (from the 1990s). The widespread ownership of tel-
evisions, radios and telephones and, more recently, cell (mobile) phones and
computers, has increased the range of methods which can be used for issuing
warnings, supplementing traditional door knocking, loud hailer, siren and other
techniques.
Many countries and local authorities now operate some form of flood warning
system, and Box 1.1 summarises estimates by the World Meteorological Organisation
for the status of flood warning and forecasting services worldwide. Flood warning is
also increasingly considered as part of a multi-hazard response to natural, technological
and other risks (e.g. United Nations 2006a). If the performance meets the required
levels of accuracy, reliability and lead time, flood warning can also be one of a range
or portfolio of non-structural measures which can be used to manage or reduce flood risk
in river catchments or along coastlines, together with other measures such as land use
planning, and tax and insurance incentives to limit development in flood prone areas.
A flood warning system can include rainfall and tidal detection systems, river
and coastal flood forecasting models, flood warning dissemination systems, and
emergency response procedures. Each link in this chain is important, and the modern
emphasis is on a Total Flood Warning System (Emergency Management Australia
1999) or people-centred approach, in which communities provide inputs to the
design of flood warning systems, and help with their continuing operation (e.g.
Parker 2003; ISDR 2006; Basher 2006; Martini and de Roo 2007).
The various components considered in this book (Fig. 1.1) are shown in Table 1.1,
although the recovery component is only discussed briefly, since flood warning and
forecasting has a lesser role to play at this stage of a flood event (for example,
advising on when flood waters will recede). Also, mitigation measures (e.g. land
use planning, insurance) are not discussed.
Of course, the terminology used varies between countries and organisations, and
some aspects may overlap (e.g. Alexander 2002). For example, the US Army Corps
of Engineers (1996) identifies the following stages in the flood warning process:
Flood-Threat Recognition; Warning Dissemination; Emergency Response;
Postflood Recovery; and Continued Plan Management whilst, for tropical cyclone
forecasting (Holland 2007) the following ten phases are identified in a typical
cyclone season: Pre-Season Check; Routine Monitoring (at least twice daily);
Cyclone Information (about 48 hours from estimated landfall); Cyclone Watch or
Alert (landfall within 36–48 hours), Cyclone Warning (landfall within at least 24
hours); Imminent-Landfall; Post-Landfall; Impact Assessment; Documentation;
and System Review.
Table 1.1 Typical components in the flood warning, forecasting and emergency response process
Item Component Examples
Flood warning Detection Monitoring of meteorological, river and tidal conditions;
and meteorological forecasting (e.g. nowcasting,
numerical weather prediction)
Thresholds The meteorological, river and coastal conditions under
which decisions are taken to issue flood warnings (some-
times called triggers, criteria, warning levels or alarms)
Dissemination Procedures and techniques for issuing warnings to the public,
local authorities, emergency services, and others
Flood forecasting Rivers, coasts Conceptual, data based and process based models for fore-
casting future river and coastal conditions
Emergency
response
Response Emergency works, temporary barriers, flow control, evacuation,
rescue, incident management, decision support
Recovery Repairs, debris removal, reuniting families, emergency
funding arrangements, providing shelter, food, water,
medical care, counselling, support to businesses, restoration
of services if interrupted
Review Review of the performance of all components of the
system, and recommendations for improvements
Preparedness Emergency planning, public awareness campaigns, training,
systems improvements, business continuity/resilience
assessments, flood risk mitigation etc.
Detection
FLOODFORECASTING
Thresholds
Dissemination
Response
Recovery
Review
PREPAREDNESS
For example
Stakeholder meetings/consultations
Customer satisfaction surveys
Public Awareness campaigns
Media briefings
School/outreach campaigns
Interagency coordination meetings
Flood Hazard Mapping
Flood Emergency Plans
Table-top and full scale exercises
Business Continuity/Resilienceassessments
Staff Training
Forecasting model improvements
System improvements(instrumentation, communications,dissemination etc)
Inputs to flood mitigation projects
EMERGENCY RESPONSEFLOOD WARNING
Fig. 1.1 Illustration of the components of a flood warning, forecasting and emergency response system
4 1 Introduction
The resilience of flood warning systems to failure is also an important consideration,
and risk based techniques from other technical sectors and types of emergency are
gradually being introduced to help to identify potential points of failure, and appropriate
risk reduction measures.
There is also much debate about the effectiveness of flood warnings (e.g.
Drabek 2000; Handmer 2002; Parker 2003) and of computer models and infor-
mation systems (e.g. Fortune 2006). Clearly, a warning is successful if it initi-
ates action which prevents flooding which might otherwise have occurred in the
absence of that warning; for example by triggering the closure of a tidal barrier,
or installation of a temporary defence. However, research suggests that success
with providing warnings to the public is mixed, although in some countries has
improved markedly in recent years through a combination of using flood fore-
casting models to extend the lead time and accuracy of warnings, a better
understanding of how to communicate warnings, and an increased emphasis on
community participation and inter-agency collaboration. For example, one
recommendation (Emergency Management Australia 1999) is that the flood
warning task can be boiled down to providing appropriate responses to the following
five questions:
● How high will the flood reach, and when?
● Where will the water go at the predicted height?
● Who will be affected by the flooding?
● What information and advice do the people affected by the flooding need to
respond effectively?
● How can the people affected by the flooding best be given the appropriate
information?
A particular issue to consider is that of the requirements for warning lead time, which
can range from a few minutes or less for people on a steep sloping river bank to reach
higher ground, to many hours or days for some situations, such as raising temporary
defences, evacuating large numbers of people, or drawing down a reservoir in advance of
flooding. Similarly, the requirements for accuracy, and tolerance to false alarms, will
vary between organisations and communities, and can be influenced by education and
public awareness exercises. This topic is discussed in more detail in later chapters.
One early success story is that of Bangladesh (World Meteorological Organisation
2006b) in which a programme of investment in education, early warning systems,
establishing a volunteer network, and emergency planning has led to a significant
reduction in the number of casualties from tropical cyclones, storm surges, and tidal
and river floods. For example, in 1998, a major storm surge led to about 140 deaths
but, in a storm of similar magnitude in 1991, approximately 130,000 people lost
their lives. Flood forecasting and warning systems have also led to major reductions
in casualties in China in recent years (e.g. Huaimin 2005). Similar improvements
can also be cited in many other countries where, due to improvements in flood
warning systems, the risk to loss of life from flooding has reduced markedly.
Approaches to flood warning, forecasting and emergency response are constantly
evolving as technical advances are made, lessons are learned from flood events, and
Box 1.1 The WMO Flood Forecasting Initiative
The WMO Flood Forecasting Initiative aims to improve the capacity of meteoro-
logical and hydrological services to jointly deliver more timely and accurate
flood forecasting and warning products and services for use in emergency
preparedness and response. The initiative was launched in April 2003, and the
main expected outcomes are (World Meteorological Organisation 2006a):
● Improved quantitative and qualitative weather forecasting products are avail-
able in such a way that these can be directly used for flood forecasting.
● Medium-range weather forecasting and climate prediction tools can be
applied to extend warning times and produce pre-warning information.
● National Meteorological and Hydrological Services have improved their
capacity to cooperate to jointly deliver timely and accurate flood forecasting
information.
● Integrated weather, climate and hydrological forecasting information are
available in a relevant format for use by civil organizations responsible for
disaster preparedness and mitigation.
Between 2003 and 2006, a series of regional workshops on “Improved
Meteorological and Hydrological Forecasting for Floods” was held in West,
Central and South Africa, Latin America, Asia, Europe and the Mediterranean
1.1 The Flood Warning Process 5
ideas are adapted from other technical disciplines. For example, technological
developments in recent years have included the introduction of short range rainfall
forecasting techniques (nowcasting) which typically combine weather radar obser-
vations with the outputs from Numerical Weather Prediction models, and of multi-
media systems for issuing warnings. Much social and behavioural research has also
been performed into public understanding of, and response to, flood warnings, in
some cases building on research in other disciplines, such as health care and emer-
gency response for other natural hazards. Improvements can also be driven by
national legislation, rising public expectations, customer satisfaction surveys, per-
formance monitoring, and the introduction of level of service targets (e.g.
Andryszewski et al. 2005). Risk based and probabilistic approaches are also
increasingly being evaluated and used operationally, building on ideas from meteoro-
logical forecasting and elsewhere; for example, in techniques for prioritising invest-
ment, and ensemble forecasting. Increasingly, improvements are performed within
a framework of targets for flood warning performance at a national level.
Chapters 2–4 discuss the topics of detection, threshold setting and dissemination
for flood warnings, whilst Chapters 9–11 discuss the preparedness, response and
review stages. The remaining chapters (Chapters 5–8) cover flood forecasting for
rivers and coastlines.
(continued)
6 1 Introduction
Box 1.1 (continued)
basin countries. These meetings involved hydrologists and meteorologists
from about 85 countries and a number of regional and river basin organisa-
tions, as well as universities and research institutions. These meetings helped
identify the status of flood forecasting and warning in the countries which can
be categorised as: (Fig. 1.2)
● Level I – flood forecasting and warning services are limited or not opera-
tional, and a significant upgrading and strengthening of the basic data
collection and transmission networks is required, together with improve-
ments in the coordination between meteorological and hydrological serv-
ices and in the dissemination of flood warnings.
● Level II – the basic infrastructure is in place for flood forecasting and
warning services but improvements are needed in data management and
flood forecasting modelling, with training in advanced modelling tech-
niques, and some improvements in coordination between meteorological
and hydrological services.
● Level III – well established flood forecasting and warning services using the
latest observation and forecasting techniques, and with warnings generally
communicated through various media to Government and Civil Protection
Agencies, industry and the public. The main requirement identified here
was for improved training and staff capacity in some cases.
These workshops were followed by an international conference in Geneva in
November 2006 to identify gaps in current procedures and to establish and
agree on a framework and action plan to improve national and regional capac-
ities for flood forecasting. The action plan addresses flooding due to flash
Fig. 1.2 Overall status of national flood forecasting and warning services (sample-86
countries) (Reproduced from the WMO Strategy and Action plan for the enhancement of
cooperation between National Meteorological and Hydrological Services for improved
flood forecasting, courtesy of WMO)
1.1 The Flood Warning Process 7
Box 1.1 (continued)
floods, riverine floods, coastal floods, snowmelt floods, ice-jams glacier, lake
outburst floods, landslides and mud flows.
The review of existing techniques showed a wide range of capabilities,
ranging from well developed systems using the latest Numerical Weather
Prediction, weather radar, and satellite and modelling technologies, through
to some countries lacking the technical or institutional capacity to operate
flood forecasting and warning systems (Fig. 1.3).
However, for some of the countries with limited capacity (14–16%),
hydrological forecasts are provided by a regional transboundary river basin
authority, and activities were underway in a further 28–33% of countries to
improve and modernise existing monitoring and forecasting systems. More
than half of countries surveyed (55%) identified a lack of monitoring equip-
ment (automatic weather stations, weather radars, satellite imagery) as an
issue, including some 27% of countries which required a significant upgrad-
ing of basic meteorological and hydrological networks and telemetry sys-
tems for flood forecasting applications. More general requirements which
were identified included the need for improved coordination and coopera-
tion between organisations and countries, guidance materials for a range of
subjects including data exchange, warning dissemination, and forecast
products, and improved training and capacity building.
Fig. 1.3 Main symptoms of insufficient or non-existent national flood forecasting capabil-
ity (sample-86 countries) (Reproduced from the WMO Strategy and Action Plan for the
enhancement of cooperation between National Meteorological and Hydrological Services
for improved flood forecasting, courtesy of WMO)
8 1 Introduction
1.2 The Nature of Flood Risk
1.2.1 Flooding in Context
Flooding is a threat to many communities and businesses, and flood risk is increas-
ing in some locations due to development on floodplains, migration to urban areas
at risk from flooding, and artificial influences on flow regimes; for example, urban
developments can sometimes increase flood risk through changes to runoff charac-
teristics and the drainage paths of floodwater. Climate change may also be increas-
ing the likelihood of flooding in some places through changes in the frequency and
severity of storms, patterns of snowfall and snowmelt, and rising sea levels.
Estimates by the Centre for Research on the Epidemiology of Disasters (CRED)
suggest that, in the period 1974–2003, there were more than 200,000 victims of flood-
ing, with many more people affected for every casualty. In that period, the July 1974
and December 1999 floods in Bangladesh and Venezuela each accounted for about
30,000 deaths, and flood events in India and China accounted for seven of the ten dis-
asters which were identified as affecting more than 100,000,000 people (all figures from
Guha-Sapir et al. 2004). During 2007, flooding due to heavy rainfall affected approxi-
mately half of all African countries, affecting more than 1,000,000 people with about
400 victims whilst, in India, Bangladesh and Nepal, the death toll from monsoon rains
exceeded 2,000 and affected some 30,000,000 people.
Compared to other types of natural disaster, floods account for approximately
20–40% of the events which are reported. Floods can also cause extensive damage
to property, infrastructure and crops, and can cut across administrative and national
boundaries. For example, the 1998 floods in China were estimated to have sub-
merged more than 200,000 km2 of farmland (e.g. Kundzewicz and Jun 2004) whilst,
for Hurricane Katrina in August 2005, in addition to causing more than 1,000
deaths, hundreds of thousands of people were evacuated, and billions of dollars of
damage was caused to property, businesses and infrastructure, much of this flood
related. Other examples include the Midwest floods of 1993 on the Missisippi and
Missouri rivers in the USA, which affected more than 15% of the country, damag-
ing or destroying some 50,000 homes, with approximately 54,000 people evacuated
(Smith 2004) whilst, in Europe, the flood events of 2002, 2005 and 2006 affected
thousands of people in central Europe, and caused more than 100 deaths.
The causes of flooding are mainly atmospheric or geotechnical (Table 1.2).
Atmospheric hazards include heavy rainfall, causing rivers to flood, sometimes
linked to snowmelt and ice-jams in colder climates, and coastal and estuarine flood-
ing due to surge, wave and wind effects, most notably in tropical cyclones, hurri-
canes and typhoons. Geotechnical factors such as landslides, debris flows and
earthquakes can also lead to raised river levels causing inland flooding, and
Tsunami waves resulting in coastal flooding. Secondary effects may include over-
topping or breaches of river and sea defence structures, debris blockages at bridges
and other structures, surcharging of drainage networks in urban areas, and dam
failure or overtopping. Due to the short time available for people to react, fast
developing floods present a particular risk to life, including flash floods, dam or
defence breaches, and some ice-jam and local surge and wave overtopping events.
Tropical cyclones, hurricanes and typhoons are all forms of tropical storm, with
the term tropical cyclone used in the Indian Ocean, hurricane in the Atlantic and
Eastern Pacific Oceans, and typhoon in the Western Pacific. Frontal depressions are
most common in mid-latitudes, and can cause prolonged rainfall, as can monsoons
which are driven by seasonal variations in temperature between sea and land
masses. Thunderstorms can occur at most latitudes, and can cause intense rainfall
for periods of typically up to a few hours. Snow and ice related problems affect
many high latitude regions on all continents, and high mountain ranges else-
where. Dam and defence risks are possible anywhere that reservoirs or polders have
been constructed, or dams built across lakes, as are breaches in river or coastal
flood defences (often known as levees or dikes). Tsunami can affect all ocean
basins, but are most prevalent in the Pacific Ocean and in South East Asia (although
the December 2004 Tsunami was in the Indian Ocean). Debris flows are a major
problem in Central Asia and the Caucasus and in parts of the USA.
1.2.2 Assessing Flood Risk
Flood risk is often expressed as the combination of two factors; probability (or
hazard) and consequence (or impact). The probability expresses the likelihood of
damaging flood levels or flows being reached, whilst the consequence can
be expressed in terms of indicators such as the numbers of properties affected, loss
of life, or economic damages.
Table 1.2 Examples of flooding mechanisms
Type Example Typical types of flooding
Atmospheric Frontal depressions Extensive river flooding, coastal surge and wave
overtopping, estuary and delta flooding, urban
and pluvial (surface water) flooding
Thunderstorms Fast response/flash flooding and urban and pluvial
(surface water) flooding
Monsoon Extreme prolonged rainfall causing a range of river
and urban flooding issues
Tropical cyclones Coastal surge and wave overtopping, inland flooding,
estuary and delta flooding
Snowmelt Extensive river flooding
Ice jams Rapid rises in river levels
Glacial lake outburst
flows
Fast moving, deep river flows
Geotechnical Dam break Fast moving, deep river flows
Defence breach Extensive inundation of coastal or inland areas
Tsunami Extensive inundation of coastal margins
Debris flow Destructive flows with high mud and rock content
1.2 The Nature of Flood Risk 9
10 1 Introduction
Estimates for the numbers of people at risk from flooding, and affected in indi-
vidual events, are of course subject to many uncertainties, including the degree to
which events are reported, the approach taken to flood risk assessments and, for
international comparisons, differences in the datasets and recording methods which
are used. However, some studies (e.g. Parker 2000; Smith 2004) suggest that the
percentages of people at risk from flooding range from 3% to 5% of the population
in the UK and France, to about 12% in the USA, 50% in the Netherlands, and
70–80% in Vietnam and Bangladesh. Estimates are also complicated by transient
populations, which can include tourists, hikers, temporary workers, business travellers,
and the homeless. Indeed, in some countries, such as the USA, one of the main risks
to life from flooding is from people in cars and other vehicles being trapped or
swept away by floodwater (e.g. Henson 2001).
The link between flood risk and social, political and economic factors, particularly
risk to life, is well documented, and can arise from issues such as a lack of public
awareness of flooding issues, or controls on floodplain development, limited funds
available for flood control and protection (e.g. river and sea defences), low resilience
of buildings to flooding (e.g. temporary compared to permanent settlements), and a
lack of investment in flood warning, forecasting and emergency response systems.
Where these factors are significant, the numbers of people affected by a flood event
can be much higher than equivalent events in locations without these problems.
Measures of vulnerability to flooding are also increasingly considered in flood
risk studies: for example, combining the following factors (e.g. Wade et al.
2005):
● Flood hazard (depth, velocity, debris)
● Area Vulnerability (effectiveness of flood warning, speed of onset of flooding,
and type of buildings e.g. low rise/high rise)
● People Vulnerability (ability to ensure own safety and that of dependents e.g. the
elderly, infirm, children)
Of course, vulnerability to flooding can depend on a wide range of physical, environ-
mental, social, economic, political, cultural and institutional factors, and can vary
widely between individuals, households and communities; for example, the length of
time that people have lived in the floodplain (or if they are visiting the area e.g. tourists),
recent experience of flooding, and local institutional capacity to respond to flooding.
Some alternative definitions (e.g. World Meteorological Organisation 2006c) express
vulnerability in terms of physical, material, constitutional, organisational, motivational
and attitudinal conditions or, for tropical cyclones (e.g. Holland 2007), include the avail-
ability of existing community level plans and organisational structures, the proportion
of cyclone resistant property, the state of protective works (river and coastal defences
etc.), and the likely protection from coastal forests and mangroves.
When designing a flood warning scheme, a starting point is often to make an
assessment of the locations and numbers of people and properties at risk from
flooding. Vulnerability studies can also highlight where to target effort in public
awareness campaigns, developing flood emergency plans, and in emergency
response. Methods for assessing risk include interviews with people who know the
area well, examination of historical flood records (trash mark surveys, aerial and
other photographs, newspaper reports, satellite images etc), and hydrodynamic and
other modelling techniques.
Interviews and historical records can provide useful information, although may
give a false impression if any significant changes have occurred since the last major
flood event in the level of flood risk or key flooding mechanisms (e.g. construction
of flood defences, dredging, urban development). Also, people may not be aware
of more serious flooding before they moved to the area. Ground survey and remote
sensing techniques can also provide detailed maps of flooding extent, although not
necessarily for the peak of the flood, and satellite observations are increasingly
being used to monitor flood extents using both optical and microwave frequencies,
and to build up databases of flood extent information.
Models provide a more formal way of assessing flood risk, and can range from
simple correlation and other methods for single locations, through to detailed
hydraulic models for river and coastal processes. Some countries (e.g. the USA,
Japan and various European countries) have programmes in place to systematically
assess flood risk at a national scale through detailed hydraulic modelling of locations
with a significant flood risk (Box 1.2).
Box 1.2 Flood risk modelling
The national flood risk mapping programmes in many countries use a range of
modelling techniques to estimate flood depths, velocities and extents. For rivers,
for example, actual or synthetic rainfall events can be fed into a network of rainfall
runoff models representing major sub-catchments, whose outputs provide the
inputs to a model for the river network and significant features such as floodplains
and reservoirs. In areas prone to flooding, the model detail may include all signifi-
cant controls on river levels and flows, such as bridges, culverts, gates, defences
and other features, as well as the main details of the floodplain, using construction
and topographic information obtained from conventional survey and remote sens-
ing techniques (e.g. Light Detection and Ranging LIDAR equipment, or Synthetic
Aperture Radar SAR equipment). In increasing order of complexity (and, in prin-
ciple, accuracy), process-based methods for modelling river levels, flows and, in
some cases, velocities, on the floodplain can include:
● One-dimensional models for the main river channel, with projection of
levels onto the floodplain, or separate pathways for main channel and
floodplain flows
● One-dimensional models including floodplain pathways represented via
spill units, compartments and/or cells
● Two dimensional models of the floodplain using ‘bare earth’ digital ter-
rain models based on mass conservation only, or including momentum
effects as well
(continued)
1.2 The Nature of Flood Risk 11
12 1 Introduction
However, whatever the technique used to assess flood risk, one problem is
always to assess the extent of mobile and transient populations who may not appear
in conventional property and census databases. Examples can include vehicle users,
shopping centres, supermarkets, tourists, hikers, outdoor events, and locations such
as caravan or mobile home parks, and camp sites. Local visits, and discussions with
people who know the area well, may be the best way of determining the extent of
this risk, and the options (if any) for providing warnings to these groups, or prevent-
ing access in time to minimise the flood risk.
Some other problems which can arise with property databases are that they may
omit some commercial properties with significant numbers of occupants during
working hours, since the correspondence address is at another location (e.g. head
Box 1.2 (continued)
● Fully two or three dimensional models of the floodplain incorporating
features on the floodplain such as buildings, embankments, gulleys etc.,
and possibly urban drainage networks
Hydrodynamic techniques can also be used for modelling inundation of
coastal floodplains due to high tidal levels, wave action and surge. Maps may
be developed either with or without flood defences, with the no defence case
sometimes being used to study the worst case flood extent; for example, if a
defence is breached, overtopped or bypassed. Later chapters show several
examples of the results from flood risk mapping studies including plan view
and virtual reality representations.
Having estimated river flows and levels, and possibly depths and veloci-
ties, within flooded areas, the resulting flood outlines can then be intersected
with information on property locations, and lists generated of properties at
risk. The resulting extents can also be related to the gauge heights used for
triggering flood warnings (see Chapter 3). These property lists then form the
basis for deciding which properties need to receive flood warnings.
Vulnerability maps can also be generated to assist in developing emergency
plans, although this is performed much less frequently than mapping of flood
extent. The resulting flood outlines may also be expressed in terms of proba-
bility or return periods, with presentations of maps at 1 in 50, 100, 200 and
1,000 year return periods perhaps the most widely selected.
Some sources of uncertainty in flood risk mapping can include the accuracy
of input data and high flow rating curves (for river modelling), the various
modelling assumptions and parameters, survey data accuracy, local influences
around structures, and other factors. Methods for assessing the uncertainty in
flood extent estimates are increasingly being explored (e.g. Pappenberger and
Beven 2006; Pappenberger et al. 2007), and can potentially feed into decision
support and other systems used in preparing for and managing flood events.
Probabilistic techniques are also increasingly being used to consider the risk
from failures or overtopping at flood defences (e.g. Sayers et al. 2002).
office), and that some locations with many residents (e.g. apartment blocks) may
appear as only a single property. Also, some high-risk locations may not be clearly
identified, such as water treatment or industrial works and critical locations such as
hospitals, power stations, telecommunications hubs etc. Again, local visits and
discussions can help to resolve some of these issues.
1.3 Emergency Response
Emergency response is the process of responding to a flood event, ideally on the
basis of a flood warning received. In many countries, there is a separation in
responsibilities between the flood warning and forecasting service, and emergency
responders such as the police, fire service and local authorities. However, the
organisation of a flood warning service can vary widely, with warnings being issued
by the meteorological service in some countries, and a range of river management,
coastal and local authorities in others.
Privately developed systems also operate in some locations, with applications ranging
from community based warning systems through to systems operated by owners of major
infrastructure such as railways and hydropower schemes. Sometimes warnings may also
be restricted to specific types of flooding, such as river flooding or coastal flooding, and
exclude other types, such as flooding in urban areas from drainage problems.
A major flood event often requires a multi-agency response, involving local
authorities, the emergency services, transport operators (road, rail etc.), utility
operators (water, electricity, gas, telecommunications), the military, coastguard,
medical services, voluntary services, humanitarian aid organizations, and others.
The response can include closing transport routes, protection of key installations, such
as power stations and water treatment works, reinforcing flood defences, providing
rest centers and shelters for people evacuated from properties, and rescue of people
and livestock stranded in flood waters. Difficult decisions may also need to be made
on issues such as the need to evacuate hospitals and nursing homes (with the evacuation
itself presenting risks), precautionary shutdown of power or water supplies, and
ordering widespread evacuations of property.
During a flood event, individual property owners can also take action to reduce
the damage caused by flooding by moving (as appropriate) vehicles, furniture,
electrical equipment, personal possessions, valuables, animals and livestock to
safer locations, and using sandbags, flood boards and other flood resilience meas-
ures to protect their property (if available). For example, in a post event survey of
flooding in parts of the Elbe and Danube catchments (Thieken et al. 2007), emergency
measures which were reported by residents included:
● Put moveable contents upstairs
● Drive vehicles to a flood-safe place
● Safeguard documents and valuables
● Protect the building against inflowing water
● Switch off gas/electricity
● Disconnect household appliances/white goods
1.3 Emergency Response 13
14 1 Introduction
● Gas/electricity was switched off by public services
● Protect oil tanks
● Install water pumps
● Seal drainage/prevent backwater
● Safeguard domestic animals/pets
● Redirect water flow
Businesses can also take actions to reduce damage to stock, equipment and systems
and, depending on the time of day, may also be able to advise employees not to
come in to work, or to leave early, in order to minimise risk.
Flood warnings can also assist river management and coastal authorities with the
operation of structures and in other actions to help to reduce or prevent flooding
and some examples (Fig. 1.4) include:
● Flood barriers – installation or operation of temporary or demountable barriers
to protect properties and infrastructure from flooding
● Flood gates – closing gates which at low to medium flows are normally kept
open to allow for drainage, access, navigation etc.
● Flow diversion – diversion of river flows into off-line storage areas to reduce
flows further downstream (e.g. washlands, flood retention areas)
● Pumping – use of high volume pumps to reduce water levels
Fig. 1.4 Examples of river and coastal flood defences and a flood gate for washland drainage
● Reservoirs – draw down of reservoir levels in advance of high inflows to provide
flood storage to reduce flows further downstream
● Sandbags – placing sand bags to raise the level of flood defences, fill gaps in
defences, or to protect properties
● Temporary works – emergency repairs to flood defences (levees and dikes) and
other locations which might provide a flow route for flood water
● Tidal barriers – closing barriers or gates to reduce the risk of inland flooding due
to surge or high tides
Temporary and demountable barriers are increasingly used for flood prevention, and
consist of metal, plastic, rubber and other types of panels, bags or tubes which can be
placed at locations where flooding is anticipated, if a flood warning is received in time.
Chapters 9–11 describe emergency response in more detail, including the devel-
opment of flood emergency plans, decision support systems, dealing with uncer-
tainty, and performance monitoring.
1.4 The Role of Flood Forecasting
Although flood warnings can be issued on the basis of observed meteorological,
river and coastal conditions alone, the development of flood events can often only
be anticipated a short time into the future, and it can be difficult to translate what
is observed into estimates of flooding extent. Interpretation can also be complicated
by other effects, such as operations at river control structures, storm surges, and
inflows from major tributaries.
Flood forecasting models can help with these issues, and are increasingly used
to improve the lead time and accuracy of warnings provided by a flood warning
service. Typically, forecasts are based on observations of river levels and rainfall
higher in a catchment (for river flooding), or of tidal levels, wave heights, wind
speed and other parameters (for coastal flooding). Rainfall and surge forecasts from
atmospheric and oceanographic models may also be used as inputs to further extend
the lead time of flood forecasting models. Forecasts may also have wider applica-
tions in areas such as river navigation, hydropower generation, water resource
management, and pollution incident control.
‘What if’ scenarios can also be performed; for example, using scenarios for
future rainfall or snowmelt, or for operational actions such as closing a tidal barrier.
Flood forecasts can also be used to automatically trigger the issuing of warnings,
or the operation of flow control structures, although the decision to use an automated
approach depends on confidence in the model outputs, policy, and other factors, and
the vast majority of current systems still require interpretation of outputs by an
experienced forecaster.
There are many approaches to flood forecasting, ranging from simple empirically
based methods to fully integrated catchment or coastal models which, increasingly,
incorporate real time hydrodynamic modelling components. Different types of
models, or model components, may also be developed for different flooding mechanisms;
1.4 The Role of Flood Forecasting 15
16 1 Introduction
for example, for the situation shown in Fig. 1.5, a range of rainfall runoff, reservoir
(dam), river, estuary (delta), and offshore, nearshore and wave overtopping models
might be required, optimised to provide forecasts at towns, infrastructure, and
transport routes where they are at risk from flooding.
One distinguishing feature of forecasting models, compared to off-line simula-
tion models, is the ability to use observed (telemetered) data to modify forecasts as
they are generated. Thus, if the forecast at the present time is in error, it can be
adjusted to account for the current observed values, and also into the future, based
on assumptions about the cause of errors up to the present time (‘time now’), and
likely future trends. This real time updating of forecasts (or data assimilation) can
significantly improve the accuracy of model outputs, and many techniques have
been developed, including error prediction methods and techniques which adjust
the internal state of model components, or model parameters.
One reason for needing updating techniques is the uncertainty in model outputs,
which can arise from many sources. For example, outputs can be affected by errors
and uncertainties in measurements of rainfall and levels and flows (for rivers), and
wind speed, wind direction, wave height and tidal levels (for coastlines).
If meteorological forecasts are used to extend warning lead times, additional
uncertainties arise from that component of the system.
These issues have long been recognised (e.g. World Meteorological Organisation
1994; Emergency Management Australia 1999; Beven 2008) and it is widely
accepted that flood forecasts should be issued with an indication of confidence or
uncertainty. The case for probabilistic forecasts in hydrology has been concisely
summarised by Krzysztofowicz (2001), which is that:
Town
Power Station
RailwayDam
Village
FarmsChemicalFactory
Town
TownSurge,Tide,
Waves
Surge,Tide,Wave
River, tide,surge
Town
WaveMajor roadDamOvertopping
River Outof Bank
DefenceOvertopping
Caravan Park
River Outof Bank
Fig. 1.5 Illustration of flooding issues which might be included in a regional flood forecasting
model
● First, they are scientifically more ‘honest’ than deterministic forecasts: they allow
the forecaster to admit the uncertainty and to express the degree of certitude
● Second, they enable an authority to set risk-based criteria for flood watches,
flood warnings, and emergency response; and they enable the forecaster to issue
watches and warnings with explicitly stated detection probabilities
● Third, they appraise the user of the uncertainty; and they provide information
necessary for making rational decisions, enabling the user to take risk explicitly
into account
● Fourth, they offer potential for additional economic benefits of forecasts to every
rational decision maker and thereby to society as a whole
One widely quoted example of the potential use of uncertainty information in
decision making (e.g. Krzysztofowicz 2001) is for the 1997 flooding on the Red
River in Grand Forks, North Dakota, which caused flooding to some 5,000 homes,
and is described in more detail in Chapter 10. Post event analysis showed that the
actual peak was higher than the forecast values, with the question arising that, if the
uncertainty in the forecast been known, would sandbagging of levees have been
continued to higher levels, avoiding the flooding which occurred? By contrast, an
example of deriving economic benefits from ensemble forecasts is in the hydro-
power industry, where some operators in the USA and Canada gain significant
savings from using probabilistic forecasts of seasonal flows (e.g. Howard 2004).
In meteorology, probabilistic forecasting techniques have been used since the
1990s, and are nowadays seen as an indispensable tool in weather forecasting.
The basis of the method is to adjust the initial conditions for the computer models
used to forecast atmospheric and ocean conditions over a range reflecting the uncertainty
in current conditions, and possibly model parameters. Stochastic methods, consisting
of statistical sampling of inputs or outputs, can also be used. The resulting scenarios, or
ensembles, are then used to guide forecasters in the information that they issue to the
public and, in some cases (e.g. the Netherlands), the range of estimates is presented
in some national television weather forecast bulletins.
Similar techniques are also starting to be used in flood forecasting, perturbing
the meteorological and other inputs to models (e.g. river flows), and possibly internal
model parameters and other factors (e.g. the high flow ends of stage-discharge
relationships). The issue of how to communicate the resulting information on
uncertainty to decision makers, including the public, is also an active research area
(e.g. Todini et al. 2005; National Research Council 2006; Pappenberger and Beven
2006). Chapters 5–8, and 10, describe these topics in more detail.
Forecasting models usually also require a computer platform on which to operate,
capable of data gathering, the scheduling and control of model runs, alarm handling,
and post-processing of model outputs into a form which is useful to forecasters.
During a widespread flood event, a purpose-made system may provide the only
practicable way of operating and interpreting the output from large numbers of
models, particularly if data assimilation is used. Modern systems increasingly make
use of spatial techniques for analysing and presenting data and forecasts, and func-
tionality can include map-based indicators (e.g. flashing symbols) of locations
1.4 The Role of Flood Forecasting 17
18 1 Introduction
where flooding is expected, overlays of property locations, street maps, aerial
photographs, terrain etc., and the facility to ‘drill down’ for additional information
and detail at any location. The facility to perform real time inundation mapping
during an event is also increasingly available.
Although much of the focus nowadays is on automated techniques, simpler tech-
niques still have an important role to play, particularly where budgets are limited, the
level of flood risk does not justify investing in a complicated approach, or as a
backup to a more sophisticated approach. Low cost community based systems are
also widespread, and typically involve nominated members of the community moni-
toring raingauges, marker boards or river gauges, and issuing warnings by loud
speaker, community billboards, and door knocking as appropriate. Information may
also be passed to local experts to decide on the appropriate action to take, who may
also have access to paper based or computerised forecasting models (e.g. FEMA
2005). Informal systems are also widely used in some countries (e.g. Parker 2003).
Another example is the flood forecasting system that was trialled for the two major
rivers in Somalia (the Jubba and Shebelli) between 1988 and 1990 (Institute of
Hydrology). The system operated on a stand-alone personal computer, with data
entered manually based on observations by government workers at some 20 locations
along the two rivers, and transmitted verbally to Mogadishu over the government radio
network. A low cost approach was used for model development, using a range of sim-
ple correlation, flow routing and overtopping models, and a more detailed model for
an off-line storage reservoir.
Real time updating was included using a simple interactive method which
allowed operators to adjust forecasts visually to account for the trend in forecast
errors over recent model runs. In operational use, information was received three
times per day, and the model runs were used to provide forecasts of future river
levels and flows up to 7 days ahead to farmers, irrigation scheme operators, and
engineers engaged in river works, together with warnings of high flow conditions.
Forecasts were also included in weekly agricultural situation reports.
Chapters 5–8 describe forecasting techniques in more detail, including general
principles (Chapter 5), river forecasting methods (Chapter 6), coastal forecasting
methods (Chapter 7) and a range of applications (Chapter 8), including integrated
catchment modeling, and forecasting for flash floods, the effects of snow and ice,
control structures, urban flooding, and geotechnical risks such as dam break,
defence breach, and tsunami.
Part IFlood Warning
K. Sene, Flood Warning, Forecasting and Emergency Response, 21
© Springer Science + Business Media B.V. 2008
Chapter 2Detection
Most flood warning systems use near real time measurements of meteorological
and river or coastal conditions to guide operational decision making. Depending on
the application, this may include information on rainfall, wind speeds, sea state,
tidal levels, river levels and other parameters, such as snow cover. Remote sensing
techniques such as weather radar and satellite may also be used, together with the
outputs from Numerical Weather Prediction models and nowcasting techniques.
This chapter provides a general introduction to these and other techniques for moni-
toring meteorological, river and coastal conditions for flood warning applications.
Telemetry systems are also discussed, together with approaches to designing telemetry
networks for flood warning applications.
2.1 Meteorological Conditions
With only a few exceptions, such as geotechnical risks (see Chapter 8), most flood-
ing problems are linked to atmospheric conditions, and observations or forecasts of
rainfall and other parameters often provide the first indication of potential flooding.
The main types of meteorological information which are useful in flood warning
and forecasting applications include:
● Site Specific (or Point) Observations – measurements at a specific location using
rain gauges, automatic weather stations etc.
● Remote Sensing (or Areal) Observations – based on satellite observations,
weather radar etc.
● Computer Model Outputs – from Numerical Weather Prediction (NWP) models,
nowcasting techniques, and other approaches
Note that weather forecasting techniques are included in this chapter as a form of
detection since, as with site specific and remote sensing techniques, the outputs
represent another source of information for the operation of flood warning and
forecasting systems.
When considering these approaches, there are various trade-offs in terms of the
spatial resolution, accuracy and lead times of each technique. For example, site
22 2 Detection
specific observations provide an indication of actual conditions at certain locations
in a catchment or coastal reach, but may be unrepresentative of the overall condi-
tions which lead to flooding. By contrast, remotely sensed data provide an overall
picture of the distribution of the parameter being observed (e.g. rainfall, snow
cover), but require some assumptions or a model to translate observations to condi-
tions at the ground or sea surface. This introduces an additional source of uncer-
tainty, and measurements are sometimes of too coarse a resolution to be useful.
Weather forecasting techniques provide additional lead times, and usually also pro-
vide detailed spatial information for the parameters being forecast (rainfall, wind,
soil moisture etc.), but obviously rely on the outputs from computer models, which
again can introduce an additional source of uncertainty.
Chapters 5–8 discuss some of these various trade-offs between lead time and
accuracy, whilst other factors which need to be considered include the likely relia-
bility during flood events, and the choice of an appropriate degree of instrumenta-
tion in relation to the level of flood risk; a topic which is discussed further in
Section 2.3 and Chapter 11 when considering techniques for prioritising investment
in flood warning schemes.
For river flooding applications, rainfall is often a key parameter, although other
meteorological parameters which may be required include observations or estimates
for air temperature, wind speed, net and solar radiation, soil moisture, snow cover, river
ice cover and ice jam locations, and reservoir and lake evaporation. For coastal flood-
ing applications, information on atmospheric pressure, and wind speed and direction,
is often a key input to surge and wave forecasting models and, for tropical cyclones
(and hurricanes and typhoons), information on storm size, intensity, track and speed is
also important. Table 2.1 summarises these and some other requirements for the vari-
ous threshold based and forecasting techniques described in later chapters.
The remainder of this section discusses the technological background to these
various techniques.
2.1.1 Site Specific Observations
The techniques for observing meteorological parameters at specific locations are
well established (e.g. World Meteorological Organisation 1994b, 2000; Strangeways
2007). For flood warning and forecasting applications, near real time measurements
are usually required, and methods suitable for telemetry include tipping bucket
raingauges, cup or ultrasonic anemometers (wind speed), wind vanes (wind
direction), radiometers (solar, net and reflected radiation), hygrometers (humidity)
and neutron or capacitance probes (soil moisture). Instruments can be installed on
land, or on moored buoys, ships and other offshore locations. These and other
devices can of course also be used if values are transmitted manually; for example,
by voice, email, fax or telegraph.
Evaporation can be measured directly using evaporation pans, or, less often, by
turbulence monitoring techniques which integrate water vapour transport through a
fixed pathway between two sensors. Alternatively, methods such as the Penman
equation are widely used for estimating open water evaporation from wind speed,
temperature, humidity and (possibly) net radiation measurements, with the Penman
Monteith approach used for estimating evapotranspiration from grass, vegetation
etc.
Often the various sensors can be combined into an automatic weather station,
which may monitor some or all of the following parameters:
Table 2.1 Some common requirements for meteorological data and forecasts in flood warning
and forecasting applications
Parameter Category Examples of techniques
Rainfall Site specific observations Raingauges, disdrometers
Remote sensing Satellite, weather radar,
microwave links
Weather forecasting Numerical Weather Prediction
models, nowcasting
Soil moisture Site specific observations Capacitance probes, neutron
probes, lysimeters
Remote sensing Satellite (e.g. Synthetic
Aperture Radar)
Weather forecasting Numerical Weather Prediction
models, nowcasting
Snow cover Site specific observations Ablation stakes, snow pillows,
snow cores
Remote sensing Satellite based optical or
infrared channels
Weather forecasting Numerical Weather Prediction
models, nowcasting
Atmospheric conditions (air
temperature, humidity,
wind speed etc.)
Site specific observations
Remote sensing
Weather forecasting
Automatic Weather Stations
Satellite based infrared sensors
(temperature)
Numerical Weather Prediction
models, nowcasting
Storm scale information Remote sensing Satellite, weather radar
Weather forecasting Storm scale/mesoscale Numerical
Weather Prediction models,
nowcasting
Radiation Site specific observations Solar radiation, net radiation (or
individual components), soil
heat flux
Evaporation Site specific observations Evaporation pan, turbulence
measuring devices (also
indirect methods using
wind speed, radiation and
humidity)
Weather forecasting Numerical Weather Prediction
models, nowcasting
2.1 Meteorological Conditions 23
24 2 Detection
● Rainfall
● Air temperature
● Humidity
● Wind speed and direction
● Solar and net radiation
● Soil or water temperature
Figure 2.1 shows two examples of automatic weather stations, consisting of a tem-
porary installation above a tropical lake in Southeast Asia for a study into long term
trends in lake evaporation (Sene et al. 1991) and a buoy mounted instrument being
inspected off the coast of the UK.
Of the various meteorological parameters which could be monitored, for flood
warning and forecasting applications, perhaps the two of most interest are rainfall
and snowmelt, and these are described in more detail below.
2.1.1.1 Rainfall
For measuring rainfall, tipping bucket raingauges are probably the most widely
used method for flood warning and forecasting applications, and record rainfall
when the depth reaches a sufficient amount (or weight) to cause a bucket mecha-
nism to tip. Typical bucket sizes are equivalent to rainfall depths in the range
0.1–2.0 mm, with the choice of tip size often based on the maximum rainfall intensities
Fig. 2.1 Examples of inland and offshore automatic weather stations (Kevin Sene and © Crown
Copyright 2007, the Met Office)
anticipated at the site. Each tip is recorded, together with the time of the tip, and
can be reported by telemetry directly, or accumulated to fixed time intervals before
transmission.
Weighing raingauges, by contrast, use springs, vibrating wires or balance
weights to record the weight, and hence depth, of rainfall, whilst drop-counting
gauges (e.g. Stow et al. 1998) use optical techniques or electrodes to record indi-
vidual drops of a fixed size released through a constriction. Depth type gauges
accumulate rainfall and use an electrode (e.g. Oi and Opadevi 2006) or float
mechanism, linked to a recording device, to record the depth of rainfall and hence
the incremental changes in given time intervals.
Disdrometers, which use a laser or ultrasound beam to detect falling rainfall, are
a newer technique for recording rainfall. These devices work on the principle of
detecting the passage of raindrops through a beam of light (e.g. Nemeth 2006) or
ultrasound, with appropriate signal processing to estimate rainfall amounts. In prin-
ciple this approach requires less maintenance than traditional raingauges since
there is no capture of rainfall. Factors to consider include processing for a range of
droplet sizes (and fall velocities), for different types of precipitation (rainfall, snow,
hail etc.), and for wind driven effects as rain passes through the beam. Low cost
(micro) vertically pointing precipitation profilers are also another recent develop-
ment for single site measurements of rainfall.
Manually operated (storage) raingauges can also be useful for providing rainfall
information to assist with post event evaluations of flooding, and more generally to
improve understanding of the rainfall distribution in a region or catchment when
developing rainfall runoff forecasting models. Measurements are typically made on
a daily or monthly basis.
Best practice in the installation and use of raingauges, and the strengths and
limitations of different designs, is well documented (e.g. World Meteorological
Organisation 1994b, 2000) but some specific problems which can arise in high
wind and rainfall conditions include:
● Splashing – both into and out of the gauge
● Exposure – sheltering by obstacles such as trees or buildings
● Wind influences – from the airflow over the gauge and at the site
● Snowfall – blocking by snowfall (raingauges only)
● Flooding – submergence of the gauge if it is installed in a flood prone site
For flood warning or forecasting applications, the recording interval to use will
depend on the capability of the equipment and associated electronics, but should
ideally be sufficiently frequent to resolve the key features of events. Typically a
5 minute, 15 minute or hourly value is used.
More generally, for all types of site specific measurement, the question arises of
how representative the measurements are of overall catchment conditions, and of
appropriate techniques to use for estimating catchment average rainfall for input to
rainfall runoff forecasting models. Box 2.1 describes some techniques for estimating
catchment average rainfall.
2.1 Meteorological Conditions 25
26 2 Detection
Box 2.1 Catchment rainfall estimation
Raingauges give estimates of rainfall at a single point whereas, for many
flood warning and forecasting applications, area averaged values are
required; for example, for catchment average rainfall. For small catchments,
or reasonably uniform rainfall, a single raingauge may be representative.
However, usually a number of gauges within the catchment, and possibly
from nearby catchments, will be used in the averaging process, and some
techniques include:
● Arithmetic mean – which simply takes the average value for all selected
raingauges, giving equal weight to all gauges without considering their
spacing or the rainfall distribution in the catchment.
● Thiessen polygons – in which polygons are derived by joining the mid
points of the lines between adjacent raingauges, with the weights based on
the proportion of the catchment area attributed to each gauge within the
catchment, divided by the catchment area.
● Isohyetal method – which derives lines of equal rainfall based on the
observed values, from which a catchment average value can be
derived.
● Surface fitting methods – which include a range of automated techniques,
such as multiquadratic, inverse distance, triangular planes (TIN) and
polynomial methods.
● Geostatistical techniques – such as Kriging which also interpolate values
but using functions for the dependence of values on distance between
gauges for all combinations of gauges. Methods such as co-Kriging also
bring in auxiliary variables such as elevation or aspect.
Additional factors such as topography, aspect, runoff coefficient, and soil type
can also be brought into some of these weighting schemes.
The methods are presented in approximately increasing order of complex-
ity and accuracy and there have been numerous studies into the merits of the
various approaches (e.g. Creutin and Obled 1982; Goovaerts 2000). In partic-
ular, elevation and rain shadow effects can be significant, as illustrated in Fig.
2.2 for average annual rainfall estimates for the Lesotho Highlands, whose
peaks rise to approximately 3,500 m in places.
Analyses of weather radar data and computer modelling can also assist
with understanding storm characteristics, such as typical storm scales, pre-
ferred directions of travel, local topographic influences etc., and in developing
appropriate catchment averaging schemes. Sometimes it is found that, where
there are no anticipated major spatial variations in flood generating rainfall
(e.g. frontal events in low lying areas), the simpler fixed weight methods can
provide reasonable results. However, where spatial and topographic variations
(continued)
2.1.1.2 Snow Cover
Information on snow depth, water equivalent and snow extent can be required for
input to the snowmelt forecasting component of flood forecasting models, and also
for more empirical techniques for estimating the consequences of snowmelt.
The challenge in snow monitoring is that the depth and type of snow cover can
vary significantly over small distances compared to typical catchment scales, so
that only a limited sample of values can usually be obtained.
Satellite monitoring can assist in assessing snow coverage, whilst observation
techniques for estimating depth include snow (or ablation) stakes and snow cores.
Traditional depth measuring techniques rely on an observer sending values by tele-
phone, email, radio etc., but automated techniques have also been developed. For
example, radio isotype methods can be used in which the water equivalent of snow
is estimated from absorption of gamma radiation in the vertical or horizontal plane
and, in principle, can provide real time estimates of water equivalent snowfall (e.g.
World Meteorological Organisation 2000). Tipping bucket raingauges can also be
used to measure the water equivalent snow depth if they are fitted with heaters,
although may under-record the true amount of snowfall, and show a lag between
are significant, more complex methods may be required if real time systems
and software are available to support this type of analysis.
Fig. 2.2 Annual rainfall distribution along a transect through the Lesotho Highlands
(Royal Meteorological Society/Sene et al. 1998)
2.1 Meteorological Conditions 27
Box 2.1 (continued)
28 2 Detection
snowfall and the recorded values. Measurements of air temperature may be used to
help in interpreting the readings from raingauges when snowfall is thought to be a
factor.
Perhaps the most extensive ground based monitoring of snow depth and extent
is by the SNOTEL (SNOwTELemetry) monitoring network in the USA (Schaefer
and Paetzold 2000). Observations were started in the mid-1970s and the network
consists of more than 700 automatic sensors installed in mid and high elevation
areas of the western United States and Alaska to record snowpack, precipitation and
temperature, typically at hourly intervals. For the snowpack component, a typical
installation includes one or more snow pillows, consisting of a flat circular con-
tainer filled with non-freezing fluid, and a pressure sensor to record the changes in
hydrostatic pressure due to the weight of the snow layer. A downward looking sen-
sor may also be used to monitor snow depth. Some sites also include automatic
weather stations, soil moisture and soil temperature sensors. Data transmission is
by meteorburst telemetry (see Section 2.3) and power is from battery packs and
solar panels. A snow pillow network is also used in Norway, supplemented by satel-
lite and manual observations, to monitor snow conditions for flood forecasting and
other applications (Rohr and Husebye 2005).
2.1.2 Remote Sensing
2.1.2.1 Weather Radar
Ground based weather radars use electromagnetic waves to detect precipitation in
the form of raindrops, snowflakes and hail (often called hydrometeors). A typical
installation consists of a radome, a tower, and buildings housing the computer and
generator equipment needed to operate the device. Figure 2.3 shows a typical radar
installation from the United Kingdom.
There are many books, review papers and guidelines describing the principles of
weather radar operation (e.g. Collier 1996; World Meteorological Organisation 2000;
Cluckie and Rico-Ramirez 2004; Meischner 2005) and only a few details are provided
here. For most types of radar, the beam is rotated about a vertical axis and the type and
quantity of precipitation is inferred from the power of the back-scattered energy, whilst
the location (distance) is inferred from the time of travel of the signal (the beam is
pulsed to provide an interval in which to detect the returned signal). Rainfall intensity
is estimated using relationships between drop size density and the power of the
received signal. Many different hardware options are available including:
● Wavelength – attenuation by rainfall reduces with increasing wavelength, but
longer wavelength radars require a larger dish and are usually more expensive.
In order of increasing wavelength are X-band (3 cm), C-band (5.5 cm) and
S-band (10 cm) radars, where typical wavelengths are shown in brackets. Most
weather radars use C or S bands, although X-band radars have been used opera-
tionally in some applications.
● Dual polarisation – use of horizontal and vertical polarisation to help with iden-
tifying hydrometeor shapes and hence types (e.g. with larger drops showing
more deformation between planes).
● Multiple beams – to detect vertical variations in reflectivity to assist in correct-
ing radar outputs for gradient and other effects, including the option of multiple
level scans.
● Doppler – to detect the direction of motion of hydrometeors to help in filtering
out ground clutter and estimating wind speed and direction.
The power of the reflected signal decreases with the range of the precipitation from
the radar due to attenuation by droplets, dust and other factors, and the spread of
the beam. The beam may also be transmitted at a positive angle to the horizontal,
and may overshoot precipitation at lower levels, including orographic growth of
rainfall in hilly regions and evaporation and wind drift/dispersion at low levels.
The use of a slight negative beam angle is also used for some weather radars in
mountainous regions. However, the beam will eventually either overshoot rainfall due
to the curvature of the earth, or encounter terrain causing anomalous reflections.
The accuracy of a weather radar therefore decreases with range, so a regional or
national radar network is often designed to achieve an acceptable coverage at a rea-
sonable cost, perhaps focused on areas with the highest rainfall or flood related
risks. For flood warning and forecasting applications, another option is to use a
denser network of low cost short range radars to infill gaps in the main radar
network in areas of interest such as major population centres, or as the main component
Fig. 2.3 View of the internal workings of a weather radar, and the Chenies Radar in the UK
(© Crown Copyright 2007, the Met Office)
2.1 Meteorological Conditions 29
30 2 Detection
of the radar network; for example, for the Local Area Weather Radar network in
Denmark (Pedersen et al. 2007).
In addition to considerations of range and beam angle, the outputs from a
weather radar may in addition be affected by meteorological factors (e.g. Collier
1996) such as:
● Bright band – due to the beam intersecting the melting layer increasing reflectivity
● Anaprop – caused by distortion of the beam in strong temperature or humidity
gradients, causing the beam to intersect the ground surface causing false returns
● Ground clutter – intersection of the beam with hills, mountains and other obsta-
cles (e.g. buildings, masts)
● Hail – causing an increase in the strength of the reflected signal compared to
rainfall
● Drop size distribution – assumptions about typical distributions may be less
valid in certain conditions, such as drizzle
Many of these factors can be reduced by post processing of the received signal,
including linking into other sources of information such as the outputs from
Numerical Weather Prediction models. Areas of research include making use of
real time information on rainfall from vertically pointing radars, microwave com-
munication links and disdrometer installations, and using Digital Terrain Models to
help in identifying sources of ground clutter.
The signals from individual radars are also often combined to produce a so-called
composite or mosaic image. Measurements are usually presented on a gridded basis,
after being transformed from the original polar coordinates. Many radar systems have
a range of sophisticated visualisation and analysis software, for example allowing
rainfall estimates to be accumulated at catchment level, sequences of radar images to
be animated, and values to be sent to other systems (e.g. flood forecasting systems).
Examples of composite images at a continental scale include the outputs from the
NEXRAD system of radars in the USA, and the OPERA project, which combines
more than 150 radar outputs for countries across Europe (e.g. Harrison et al. 2006).
If the raingauge network is of sufficient density and quality, radar estimates of
rainfall may also be adjusted to take account of raingauge measurements of rainfall,
in an attempt to correct for low level and other effects missed by the radar. The methods
used include multi-quadratic, Bayesian and other techniques (e.g. Moore et al. 2004;
Todini 2001). In applying these techniques, of course, there are also uncertainties in
the accuracy of the raingauge measurements, and in particular how representative
they are of the spatial distribution of rainfall. In addition to using raingauges, the radar
outputs can also be improved using outputs from Numerical Weather Prediction models,
satellite imagery, wind profiles and other sources of information (e.g. lightning
detectors), and this type of nowcasting product is described later.
The resolution at which radar data is provided depends on the type of signal
processing algorithms used, and the distance from the radar. In the UK, for exam-
ple, values are available at grid lengths of 1, 2 and 5 km, and 5 or 15 minute time
intervals, depending on the distance from the radar (the corresponding ranges
are up to 50, 100 and 250 km). Figure 2.4 illustrates the appearance of images at
these scales for a heavy rainfall event in the south of England.
In the figure, the distance between the towns of Oxford and Watford is approxi-
mately 60 km. The figure shows that the degree to which a weather radar can resolve
the spatial distribution of rainfall depends on the grid resolution, which in turn depends
on the distance of the catchment from the nearest radar. Signal quality may also be
influenced by topographic and other effects, particularly in mountainous regions.
Fig. 2.4 Illustration of weather radar images at a resolution of 1, 2 and 5 km (© Crown Copyright
2007, the Met Office)
2.1 Meteorological Conditions 31
32 2 Detection
Radar coverage maps, typically estimated using digital terrain models, and line of
sight calculations, can give an indication of the likely resolution and coverage of
radar for a given location or catchment.
2.1.2.2 Satellite
Satellite observations offer the potential to provide estimates of rainfall and other
parameters (e.g. sea state) for input to river and coastal flood forecasting models, and
are routinely assimilated into Numerical Weather Prediction and nowcasting models.
Images of cloud cover are also widely used by flood warning services, although
the quantitative use of information, such as estimates of rainfall rate, is much less
widespread, in part due to the relatively coarse resolution of measurements com-
pared to the scale of some smaller catchments. Some examples of geostationary
satellite systems which have been used in flood forecasting applications include:
● Meteosat – operated by Eumetsat and the European Space Agency for weather
forecasting including infrared, visible and radiation budget sensors
● Geostationary Operational Environmental Satellites (GOES) – operated by the
National Oceanic and Atmospheric Administration (NOAA) for weather fore-
casting with multiple sensors
Geostationary satellites maintain a fixed position relative to the earth’s surface,
whilst polar orbiting satellites are at lower altitudes (hence with better resolutions)
although may only pass overhead a given location every few days. The sensors used
vary depending on how long ago the satellite was built, and the primary applica-
tions. However, in general, most meteorological and oceanographic satellites are
able to monitor cloud cover, the radiation budget (radiation, reflected energy), sur-
face and cloud top temperatures, snow cover and other parameters.
The GOES and Meteosat systems form part of the World Meteorological
Organisation (WMO) World Weather Watch (WWW) Global Observing System
(GOS) programme (World Meteorological Organisation 2003) which also includes
a number of other satellites launched as part of national programmes for environ-
mental and meteorological monitoring (e.g. GMS-Japan, METEOR and GOMS –
Russian Federation; FY-1 and FY-2 – China).
For flood forecasting applications, one technique of interest is the estimation of
rainfall intensities from cloud top temperatures, with cooler temperatures indicating
cloud tops at greater altitudes, and therefore possibly with greater depths.
Observations of cloud temperature relative to surrounding regions, and of cloud
morphology, can also be used to help to discriminate between clouds with similar
cloud top temperatures but different rainfall producing potential, such as cirrus and
cumulonimbus clouds (e.g. Golding 2000; World Meteorological Organisation
2000; Grimes et al. 2003). Methods may use single images (cloud indexing meth-
ods), or sequences of images (life history methods) to assess cloud development
and movement. These techniques have also been used to estimate rainfall for input
to rainfall runoff forecasting models (e.g. Grimes and Diop 2003)
Passive and active microwave measurements also show potential in estimating
rainfall intensity and the soil moisture at land surfaces (e.g. Crow et al. 2004; Love
2006), although the algorithms which are used need to interpret the signals from dif-
ferent types of land surface including open areas, forest, water, ice, snow and urban
areas. Rainfall rates at the surface may also be inferred from the radiation received
from sources such as liquid water droplets or suspended ice particles. For active sys-
tems, similar principles are used to ground based weather radar, and are actively being
developed as part of NASA’s Tropical Rainfall Measuring Mission (TRMM) and the
planned international Global Precipitation Measurement Mission. Satellites can also
be used for monitoring snow cover, and the formation and break up of ice in rivers
and lakes, to provide advance warning of likely flooding problems.
2.1.2.3 Other Techniques
Some other remote sensing techniques which have been considered for rainfall
detection in flood forecasting applications include:
● Microwave techniques – use of horizontally transmitted beams to detect rainfall
● Lightning detection – inferring rainfall amounts from lightning activity
Microwave techniques estimate the path averaged rainfall rate from the attenuation
in the signal, and could potentially make use of the extensive transmitter networks
used by cell phone operators (e.g. Leijnse et al. 2008). For example, as part of the
MANTISSA project (Rahimi et al. 2003; Holt et al. 2005), experiments were per-
formed using dual frequency microwave links with path lengths from 9 to 23 km for
a catchment in the northwest of England, and results compared with raingauge and
weather radar estimates of rainfall. Uncertainties can arise from unknowns such as
the drop shape, temperature and size distributions.
Lightning detection methods (e.g. Price et al. 2007) aim to provide forecasts from
the short term up to a few hours ahead for heavy rainfall linked to thunderstorms.
Lightning activity can be monitored remotely at a global level using space and
ground based observations, and in principle can be used to track the progression of
thunderstorms. Historical rainfall-lightning relationships can be established from
past records, and used together with satellite observations and Numerical Weather
Prediction models to estimate rainfall intensity in real time. Lightning data is also
assimilated into some forms of nowcasting model as described in the next section.
2.1.3 Weather Forecasting
The topic of weather forecasting covers a wide range of numerical, empirical,
observational and other techniques, and in operational forecasting the final deci-
sions on the forecasts to issue are often taken based on a combination of these
approaches. For flood forecasting and warning applications, the following two
approaches are of particular interest:
2.1 Meteorological Conditions 33
34 2 Detection
● Numerical Weather Prediction models – primarily for rainfall forecasts
(Quantitative Precipitation Forecasts) for input to rainfall runoff models, and
wind fields for input to coastal models, but possibly also for a range of other
surface variables which may be calculated (e.g. soil moisture, air temperature).
Typical maximum useful lead times can be 3–5 days or more for deterministic
forecasts, and up to 10–15 days for ensemble forecasts, although can be consid-
erably less for events such as thunderstorms.
● Nowcasting systems – which provide short term forecasts based on a combina-
tion of weather radar, satellite and other observations and, increasingly, the out-
puts from Numerical Weather Prediction models.
The distinction between these two approaches is not clear cut, since Numerical
Weather Prediction models also make extensive use of observed data from the sea,
ground, air and space to initialise model runs via a process called data assimilation.
A simple definition here is that a nowcast is a short term forecast, based primarily
on weather radar, typically for times of up to 3–6 hours ahead.
Seasonal forecasting systems, combining statistical and other modelling
approaches, are also increasingly being used, and have been used in forecasting
snowmelt, for example (see Chapter 8).
2.1.3.1 Numerical Weather Prediction
Numerical Weather Prediction models form the basis of the forecasting service
offered by many meteorological services, and solve approximations to the equa-
tions describing mass, momentum and energy transfer in the atmosphere (e.g.
World Meteorological Organisation 2000).
The equations may be solved over a global domain, or domains limited by hori-
zontal extent. The boundary conditions for the limited area models are then derived
from the larger scale models (i.e. the models are nested). The equations are usually
solved on a layered grid, with typical horizontal scales of 10–100 km for global
scale operational models, and 1–10 km for local models, and up to 100 layers rep-
resenting vertical development in the atmosphere. Local models may be called local
area, mesoscale or storm-scale models, depending on the type and spatial extent of
modelling approach adopted.
Sub-models may be included for a range of processes, including cloud devel-
opment and decay, energy and water transfer at the ocean and land surfaces, and
interactions with topography and other obstacles. As noted earlier, models are
initialised using a process called data assimilation which can be a major under-
taking, using measurements taken from raingauges, weather stations, weather
radar, lightning detectors, aircraft, ships, wave buoys, radiosondes, satellites, and
other sources. Models typically run on a 1, 6 or 12 hourly timestep, and the data
assimilation component can often take a significant proportion of the time
between model runs.
Model outputs can include the wind field, rainfall, potential temperatures, specific
humidity, surface pressure, evapotranspiration, snow depth, surface and soil temperatures,
soil moisture, cloud water and ice, and other variables. Other outputs can include
the convective cloud base and cloud top elevations, sea surface roughness, vertical
velocities, and other parameters.
Results are usually processed further into specific ‘products’ which vary from
country to country but may include a general outlook, synoptic charts, surge fore-
casts, daily forecasts, strong wind warnings, heavy rainfall warnings, flash alerts,
and other forms of output tailored to meet each user’s requirements. Other types of
output which may be useful in hydrological applications include estimates for soil
moisture conditions and snow cover.
Due to the intrinsic uncertainties in both the models, and the data assimilation
process, it is now standard practice in many meteorological services to use an
ensemble forecasting approach, in which the initial model state is perturbed and
multiple realisations of model runs are performed to provide an indication of the
uncertainty in the forecasts.
With current computing power, typically of the order of 10–100 ensemble runs
are performed at each time step. In some countries (e.g. the Netherlands), the
ensemble outputs may be presented as part of national weather forecasts on televi-
sion in the form of an estimated range of values for parameters such as air tempera-
ture or rainfall, or as probabilities of occurrence. Multi-model techniques are also
used, in which the outputs from several models are displayed in a common format
to see the variability between different formulations (e.g. Garcia Moya et al. 2006;
Rotach et al. 2007).
Probabilistic and ensemble forecasts are also increasingly being introduced into
flood forecasting applications, and are discussed further in Chapters 5 and 8.
For river flood forecasting applications, a key requirement is often to translate the
meteorological model outputs to a scale more appropriate to hydrological modelling.
Both statistical and dynamical techniques are used (e.g. Rebora et al. 2006;
Schaake et al. 2005). Statistical techniques can include multi-fractal cascades, non-
linear autoregressive models, and processes based on the superposition of rainfall
cells at different scales (cluster models). Dynamical techniques can include nesting
of higher resolution atmospheric models for the catchment or region of interest
within models with a coarser resolution but wider spatial extent (e.g. Environment
Agency 2007). For large catchments, upscaling may also be required to help to pre-
serve hydrological spatial characteristics over large distances, particularly where
there are significant topographic or climatic variations.
2.1.3.2 Nowcasting
The term Nowcasting covers a range of techniques which use spatial extrapolation
of current observations of rainfall from weather radar, sometimes guided by or
combined with the outputs from Numerical Weather Prediction models. For short
2.1 Meteorological Conditions 35
36 2 Detection
lead times, these techniques can perform better than Numerical Weather Prediction
models, and their relative simplicity allows more frequent model runs (e.g. every
few minutes) and higher model grid resolutions (e.g. 1–10 km). The maximum lead
times provided can be several hours, although values of 3–6 hours are often quoted.
Nowcasting methods often use the assumption that, if the speed, size and direction
of travel of a storm is known at the present time, then the future development can be
estimated by extrapolation, at least at short time scales. Methods range from simple
extrapolation of current conditions, neglecting possible growth or decay, to techniques
using a wide range of sources of information to help to estimate the future evolution
of a rainfall event or thunderstorm or tropical cyclone (e.g. Franklin et al. 2003).
A more sophisticated approach is to use the outputs from Numerical Weather
Prediction models. Forecasts can then be generated by extrapolating the motion of
areas of rainfall, using the wind fields and other forecast outputs from these models
to guide this response, including allowance for the development and dissipation of
rainfall (e.g. Golding 2000; Wilson 2004). Sub-models may be included to forecast
the development of convective cells (thunderstorms) using conceptual (life-cycle)
models and probabilistic techniques.
As with Numerical Weather Prediction models, ensemble and probabilistic
approaches are increasingly being used in Nowcasting, with research also consider-
ing how seamless ensembles can be generated covering a range of timescales, from
nowcasting through to Numerical Weather Prediction and seasonal forecasting.
For example, the Short Term Ensemble Prediction System STEPS (e.g. Bowler
et al. 2006) recognises the inherent uncertainty in forecasts over a wide range of
scales, including the fact that smaller scales are shorter lived and less predictable,
and blends extrapolation, stochastic noise and Numerical Weather Prediction model
outputs on a hierarchy of scales. The system generates 50 member ensembles of
rain rate and accumulation at a 2 km grid, 5 minute resolution to provide forecasts
at lead times of typically up to 6 hours ahead.
2.2 River and Coastal Conditions
Near real time measurements of river and tidal levels, wave conditions, and river
flows are important in many flood warning and forecasting applications. There is
much in common between the techniques used for river and coastal monitoring,
although river gauges may be affected by debris and sediment loads, whilst tidal
gauges may experience a harsher environment in terms of salinity and wind and
wave loading. Various types of instruments are also deployed in the open oceans
although are not described here, including free drifting floats, ocean gliders, and
ship-borne measurements.
For river monitoring, depending on the nature of the catchment, information
may also be required on levels in reservoirs and off-line storage reservoirs, on flow
depths on floodplains, and for ice conditions, pump settings or flows, borehole levels,
and other parameters. For both river and coastal monitoring, additional information
may also be required on gate settings (e.g. at reservoirs, or tidal barriers), and the
condition of river and sea defences and other key assets, particularly if there is a
suspected risk of breaching or overtopping. Some monitoring techniques for these
applications are discussed briefly in Chapter 8.
The techniques used for river and coastal monitoring are well established (e.g. World
Meteorological Organization 1980, 1998; Hershey 1999; Intergovernmental Oceano-
gra phic Commission 1994) and only a few key points are presented here, together with
some recently developed techniques. It is convenient to categorise techniques as follows:
● River/tidal level monitoring – shaft encoder (float), pressure transducer, bubbler
gauges, downward looking devices, sensor networks, satellite altimetry
● River flow monitoring – ultrasonic and electromagnetic devices, gauging struc-
tures, particle imaging velocimetry
● Wave monitoring – recording of wave heights, periods etc.
● Position monitoring – applicable to gates, ice monitoring, flood defences etc.
Position monitoring devices are not described in detail but include shaft encoders,
ice motion detectors (e.g. doppler radar, or instruments linked by wire to plates
anchored in the ice), and strain gauges. Fixed or panning CCTV and webcams for
visible light, low light or infrared are also increasingly used for monitoring loca-
tions prone to blockages, ice formation, or other problems, and for estimating
parameters such as wave overtopping rates at sea defences.
For flood forecasting and warning applications, unless manual observations are
used, all devices require a means of translating movement into an electrical signal
for data logging, and onward transmission by telemetry. Instruments should also be
installed with electronics above the highest likely flood levels. Only instruments
suitable for telemetry are described in the following sections.
2.2.1 River/Tidal Level Monitoring
Level monitoring devices record water levels using a range of techniques including:
● Float recorders – a float contained in a stilling well, installed either in a down-
pipe within the water body, or set into the ground and connected by a horizontal
pipe to the river, reservoir or sea. The float moves up and down with water levels,
causing the pulley from which it is suspended to rotate, and the rotation is
detected electronically by a shaft encoder.
● Pressure transducers – are typically submerged at the end of a downpipe which
acts as a protective conduit for the wire connecting the device to the data logger.
The pressure which is recorded depends on the depth of water above the sensor.
Pressure sensors have also been used in urban areas to detect flooding on roads,
for example.
2.2 River and Coastal Conditions 37
38 2 Detection
● Bubbler gauges – typically release bubbles of inert gas (e.g. nitrogen) or air from
an orifice and are supplied from a gas canister or compressor. The pressure
required to displace water from the submerged orifice depends on the water
pressure and hence the depth of water above the device. Such devices also
include a pressure sensor, and are sometimes called pneumatic gauges.
● Downwards looking devices – include radar, ultrasound and acoustic devices
suspended above the water surface by a purpose-made frame, or from existing
structures (e.g. bridges, piers), which estimate levels based on the time of travel
of transmitted and reflected signals. Although acoustic devices can be operated
in the open, usually they are contained within a narrow sounding tube contained
within a stilling well (e.g. for tidal monitoring), whilst radar and ultrasound
devices do not normally require a protective conduit for the beam.
● Sensor networks – are a newer technology (pervasive or grid computing)
consisting of networks of small low power pressure sensor devices with integral
microprocessors and transmitters (e.g. radio or microwave), programmed to
collaborate to form networks which can reconfigure automatically if any one
sensor fails, and are much cheaper to install and operate than conventional instruments
(e.g. Hughes et al. 2006).
● Satellite altimetry – is widely used for monitoring ocean levels, and shows
potential for monitoring of river levels, particularly on large rivers (e.g. Beneviste
and Berry 2004; Xu et al. 2004; Zakharova et al. 2005).
For river applications, the time interval for measurements can be set at a value based
on the expected rate of rise and fall of river flow hydrographs or reservoir levels, and
ideally would provide several values on the rising limb in a flood warning application
(although this may not always be practicable in a fast responding river).
Figure 2.5 shows a float in stilling well and a pressure transducer installation.
The examples are for a river float in stilling well device, and a reservoir pressure
transducer installation with radio mast and a staff gauge for manual observations.
Each method has its own strengths and limitations. Devices installed below the
water surface face the risk of damage by debris during a flood event, or blockage of
the equipment by sediment or ice. In some countries, heaters or other forms of pro-
tection may be required to ensure operation in ice conditions. For tidal applications,
and to a lesser extent reservoirs and lakes, the gauge output may also be affected by
wave action. Individual sensor types of course have their own limitations, and may
require corrections for drift, temperature effects, density effects and other factors.
In rivers, downward looking devices may return ambiguous signals when there
is significant debris floating on the water surface (trees etc.). For all types of device,
data recording and transmission may be affected by floodwater if the data logger
and telemetry electronics are installed at too low a level or in a location prone to
erosion or impact by debris. Also, datum values need to be established and regu-
larly checked so that water level measurements are consistent over time and can be
related to national datum values. Tidal gauges may also incorporate a datum probe
or switch which operates at a known sea level so that datum offsets or errors in the
tidal record can be identified.
Instruments may also need to be able to record over considerable ranges in
levels; for example, ranges of 10 m or more are not uncommon in some tidal and
river monitoring locations. Also, if water levels can fall below the height of the
sensor (e.g. in low flow periods), then the instrument needs to be able to cope
with dry conditions, and possibly high air temperatures, and blowing sand.
However, in tidal applications, sometimes a gauge is designed to become
exposed to the air at low tides to allow the datum or instrument to be checked at
regular intervals.
2.2.2 River Flow Monitoring
For flood warning applications, measurements of levels may sometimes be all that
is required. Indeed, compact self contained units are available commercially com-
bining a pressure transducer or float recorder with a solar power or battery supply
and a direct connection or telemetry link to a warning device (e.g. a bell, siren or cell
phone) which is triggered if one or more preset levels is exceeded (see Chapter 4).
However, for many river monitoring applications, an estimate of flow is
required and, unless a purpose made flow monitoring gauge is installed (see
later), values must be obtained by calibration of a stage-discharge relationship or
Fig. 2.5 Examples of water level recording devices
2.2 River and Coastal Conditions 39
40 2 Detection
rating curve established based on concurrent measurements of levels and flows or
discharge (i.e. spot gaugings). Most hydrometric services have programmes in
place for regular gaugings, and often also for high flow measurements during
flood events.
The technique for measuring flows typically involves taking a series of measure-
ments of depths and velocities across the river, using one or more velocity measure-
ments at each location, and integrating values to estimate the flow. Velocity
measuring devices include propeller meters (current meters), and Acoustic
Doppler Current Profilers (ADCP). Except at low flows, when wading can be
used, current meters must normally be suspended from a bridge, boat or cableway,
whilst ADCP devices can be floated across the river surface from the river banks
or using towed, manned or radio-controlled floats and boats. On a large river, a
single measurement of flow can be a time consuming operation, with potential
health and safety and access issues during a flood event, particularly for measure-
ments taken at night and in fast flowing, debris laden water, although remotely
controlled winch systems (e.g. Park et al. 2006) are a possible way of helping to
avoid these risks.
Other less widely used methods include dilution gauging, in which the change
in concentration of a tracer such as salt water or dye is recorded along a river reach,
and slope-area methods, in which the change in water surface elevation is measured
along a reach and the flow estimated from hydraulic formulae.
The procedures for establishing a stage discharge relationship are well estab-
lished (e.g. International Standards Organisation 1996, 1998) and Fig. 2.6 shows a
simple example.
Stage discharge relationships are often represented by power law equations with
parameters obtained from a least squares fit regression to observed values. Look up
tables and polynomial functions are also used, although to a lesser extent.
0.1
1
10
0.1
Discharge (cumecs)
Sta
ge
(met
res)
1 10 100 1000
Fig. 2.6 A simple example of a stage discharge relationship
The interpretation of stage discharge relationships can be complicated by a
number of factors including:
● Weed growth – seasonal or intermittent growth of weeds causing channel
constrictions
● Backwater influences – influences from downstream of the site such as from
gate operations, high flows in tributaries etc.
● Tidal influences – tidal influences affecting water levels at the gauge, perhaps
only for exceptional tides
● Channel profile changes – changes in the channel bed profile at or near the site
due to erosion, scour, sedimentation, dredging etc., and changes in the river
cross section as levels rise (e.g. flows going out of bank onto a floodplain or
bypassing the station)
● Ice cover – formation of ice constricting river flows to varying degrees at certain
times of the year, causing backwater effects for ice downstream, throttling flows
for ice cover next to the instrument, and with a range of effects for ice cover
upstream
● Hysteresis – differences in flow values for rising river levels and falling river
levels
These and other factors can cause curves to change with river depth, season and
over time, leading to multiple equations valid only for given periods or seasons.
Also, errors in the high flow end of stage-discharge relationships are perhaps one
of the main sources of uncertainty in flood forecasting models. However, the stage
discharge approach is probably the most widely used method internationally for
estimating flows in rivers.
Some techniques for extending curves include hydraulic modeling (1D, 2D or
3D), slope-area methods, based on peak water levels estimated from photographs,
maximum level recorders, or from trash left after flood events, and velocity-area
methods, based on direct survey of the river cross section area, and extrapolation or
estimates of velocity.
Given the uncertainties in estimating river flows from levels alone, various tech-
niques have also been developed to provide a more direct measurement of flows,
although are often significantly more expensive in terms of initial capital costs.
The main types of device include:
● Ultrasonic devices – which measure flows at one or more depths by the travel
times of ultrasound waves between senders and receivers set at angles (typically
in the range 30–60 degrees) to the main river flow. The average velocity at each
depth can be estimated from the difference in time of travel between pulses with
an upstream and downstream component, and can be integrated to provide an
estimate of overall flow.
● Electromagnetic devices – which record the electromotive force induced by
water flowing over a coil buried in the river bed, which is notionally proportional
to the water velocity. The signal is detected by electrodes in the river banks, and
additional probes may be required to allow corrections to be made from other
electromagnetic sources near to the station.
2.2 River and Coastal Conditions 41
42 2 Detection
● Gauging structures – purpose made structures designed or adapted for flow
measurements, including various types of weirs, flumes and other structures,
with estimates of flow obtained from levels recorded at one or more prescribed
locations in the structure and a corresponding theoretical formula.
For all three techniques, it is usually necessary to make a number of spot gauging
measurements when the instrument is first installed to check or establish the cali-
bration. Also, for gauging stations in particular, but also for ultrasonic and electro-
magnetic devices, occasional spot gaugings are often made during routine operation
to check for any drift in the calibration.
All three types of device can be affected by the problems of ice, algae, weed
growth, sediment and damage from debris, although electromagnetic instruments are
less affected by weed growth. In addition, for gauging structures, the usual assumption
is that the depth at the structure is controlled at the structure, and is independent of
levels downstream. However, under high flow conditions, structures may drown out,
so that the theoretical relationship no longer applies. One approach to estimating flows
at structures where this occurs is to install a second level recorder downstream and to
use theoretical or modelled relationships to estimate flows in these conditions.
To provide better sensitivity to changes in depth, some structures also include
changes in channel cross section, with additional channels becoming effective at
higher flows, and only a narrow channel in operation at low flows. V notch weirs
also achieve a similar effect. In flood warning and forecasting applications, another
option is sometimes to calibrate an existing structure, built for other applications
(e.g. navigation, irrigation), if it provides a stable control on water levels.
One newer technique which shows some promise is the use of automated com-
puter analysis of video camera images of existing tracers (e.g. foam, flotsam) on the
water surface (e.g. Creutin et al. 2003). The technique, Particle Image Velocimetry,
gives an estimate of surface flow velocities, which can be related to overall
flows either using standard formulae or previous current meter measurements.
The method relies upon suitable tracers being present on the water surface, and can
be affected by shadows and reflections; however, it offers the promise of being able
to estimate flows at a low cost from remote locations. Similar trials have also been
performed using hand held or bridge mounted radar devices.
2.2.3 Wave Monitoring
In flood warning applications, estimates of wave height and direction are useful to
assist in deciding whether to issue coastal flood warnings, and for input to coastal
flood forecasting models. Typical wave periods might only be a few seconds, so the
information is usually recorded over intervals of a few minutes or more and
expressed in terms of the spectral properties of the wave field, from which key
statistics such as significant and maximum wave heights, dominant and average
wave period and direction, wave spread, mean water level, wind speed and direction,
and wave spectra, can be estimated. The output across a number of wave monitoring
locations provides a spatial picture of wave distributions.
The main techniques for monitoring waves include (Massel 1996; World
Meteorological Organisation 1998):
● Measurements from below the sea surface
● Measurements at the sea surface
● Measurements from above the sea surface
For sub-sea devices, the signal can be transmitted by cable to the shore, or to a
nearby buoy to be transmitted by radio or satellite. Pressure transducers are the
most widely used method, with the pressure at the instrument varying with wave
height. The resulting spectrum is then corrected for hydrodynamic attenuation with
depth. However, depth corrections start to be of a similar magnitude to typical wave
pressure signals for water depths much beyond 10–15 m, and also tend to filter out
higher frequency signals, limiting the depths at which pressure transducers can be
used (e.g. World Meteorological Organisation 1998). Vertically pointing echo
sounders can also be used, although the signal may be affected by bubbles from
breaking waves.
Measurements at the sea surface are typically made from wave buoys, in which
the vertical acceleration is measured using an accelerometer mounted on a gyro-
scopically stabilised platform, although solid state techniques are increasingly
used. Wave heights can then be inferred from the acceleration terms. Motion can
also be monitored in the two horizontal planes (roll and pitch) to provide spectral
estimates of wave direction. Some devices also use Global Positioning Systems
(GPS), and solid state inertial motion sensors which provide combined values for
surge, sway, heave, roll, pitch, yaw and heading. Telemetry can consist of radio or
satellite links. Lightships can also be used, in which the accelerometer output is
combined with pressure sensor outputs to detect horizontal motions. A ship pro-
vides a more stable platform, although is less sensitive to smaller waves, whilst a
buoy needs to be carefully installed so that the mooring does not influence the
motion significantly.
These methods are more appropriate for deep water, and shallow water tech-
niques include capacitance probes and resistance probes, which can be mounted on
structures such as piers or platforms. These devices consist of a series of sensors
along a board (wave staff), where the signal depends on the depth of wave immer-
sion, although can be affected by breaking waves. Devices which use ultrasonic or
electromagnetic velocity meters can also be used to measure the two horizontal
components of wave orbital velocity which, in conjunction with a pressure recorder
or capacitance or resistance probe, can provide useful directional information.
Downward looking devices of the types described earlier, such as laser, infrared,
and acoustic range finding devices, can also be used to monitor waves if a suitable
platform is available, although can be affected by reflections and other influences
from the structure.
For model calibration, satellite based estimates of long term wave state can be
derived using synthetic aperture radar and other spaceborne instruments, although
for polar orbiting satellites observations at a given location are only made once
every orbit. Shore based high frequency radar also provides a method for monitor-
ing wave states and sea surface currents over large areas.
2.2 River and Coastal Conditions 43
44 2 Detection
2.3 Instrumentation Networks
Flood warning and forecasting systems usually rely on a network of meteorologi-
cal, river and/or coastal instruments. Individual types of instrumentation may also
be combined; for example, an automatic weather station may be installed on a wave
buoy, or a raingauge at a river gauging station.
Monitoring networks can also serve a range of purposes in addition to flood
warning and forecasting, such as water resources monitoring, marine forecasting,
and climate change monitoring, requiring a compromise between these different
applications. For example, a water resources gauge may be installed close to a river
confluence to monitor the entire runoff from a catchment but, at high flows, suffer
from backwater influences from the main river, possibly making it unsuitable for
use in a flood forecasting application.
For new sites, issues of site permissions, power supply, access for installation
and maintenance, and other factors may lead to gauges being installed in locations
that are not ideal. The requirements for telemetry connections may also influence
the locations at which gauges are installed. The following sections discuss some
options available for telemetry of real time information, and give a brief introduc-
tion to the design of networks for flood warning applications. Chapter 11 also dis-
cusses some of the economic considerations in network design, and in choosing an
appropriate solution tailored to the level of flood risk.
2.3.1 Telemetry Systems
For telemetry of real time data, the following options are widely used in flood warn-
ing and forecasting applications (e.g. World Meteorological Organisation 1994b):
● Telephone lines (PSTN) – connections via land-lines using the public switched
telephone network. Each instrument has a unique telephone number which can
be dialled to retrieve data or check the condition of the instrument.
● Mobile telephone (GSM, GRPS) – similar to PSTN lines but using cell phone
technology.
● Radio – Ultra High Frequency (UHF) or Very High Frequency (VHF) commu-
nication links.
● Satellite – transmission from the instrument to an orbiting or geostationary satel-
lite for relay to a ground station.
● Meteorburst – use of naturally occurring ionisation in the atmosphere left by meteor
trails to reflect radio waves between a base station and outstation. Meteor impacts
are sufficiently frequent that reasonable data transfer rates can be achieved.
● Internet – broadband, Ethernet and wireless connections.
Each approach has advantages and limitations and Table 2.2 provides some exam-
ples of these considerations.
Other considerations can include power consumption, licence requirements, and pur-
chase and installation costs. Some generic examples of use of these techniques include:
● WHYCOS – a World Meteorological Organisation initiative to install river and
climate monitoring stations on the main rivers worldwide, which is being devel-
oped as a series of regional projects. Instruments typically use Data Collection
Platforms (DCPs) transmitting via the Meteosat satellite system and other sys-
tems (World Meteorological Organisation 2005).
● ALERT (Automated Local Evaluation in Real Time) – a set of radio based com-
munication protocols, sensing technologies and data formats which is widely
used in the USA and elsewhere for locally operated flood warning systems
incorporating raingauges, river level and other sensors (NOAA/NWS 1997).
There are also many national and regional examples of applications of these tech-
niques; for example, in the United Kingdom, the public switched telephone network
Table 2.2 Examples of strengths and possible limitations in telemetry methods
Method Strengths Possible limitations
Telephone, broadband,
local wireless ethernet
Uses an existing network May incur connection and usage
charges
Simple to set up and operate Requires a reliable public network
Land lines and exchanges can be
damaged by flooding and high
winds etc. if not designed to
avoid these problems
Cell phone Uses an existing network May incur connection and usage
charges
Simple to set up and operate Possible data drop-outs in heavy
rainfall
Networks can be affected by power
cuts during flood events
Radio Probably no connection
charges other than radio
licence fees once the
network is established
User needs to establish and main-
tain the network (equipment,
permissions, licences etc.)
User retains full control of
the network
Line of sight required for transmis-
sion possibly requiring repeater
stations, or limiting the range
for coastal applications
May be affected by interference
Satellite Instruments can be installed
anywhere visible to the
satellite
May incur data transmission
charges
Possibly no suitable satellite visible
No requirement to establish
a network
Transmission may be restricted to
the time of overpass (for orbit-
ing satellites) or transmission
time slots determined by the
operator
Meteorburst No requirement to establish
a network
Relatively high power transmitter
required
Signals can be transmitted
over long ranges
Possible delays whilst waiting for
suitable transmission conditions
2.3 Instrumentation Networks 45
46 2 Detection
(PSTN) is used almost exclusively for data links to river and raingauge instrumen-
tation whilst, in the USA, the Meteorburst system is used for transmitting data from
the snow monitoring SNOTEL network described in Section 2.1. Manual systems,
in which levels are relayed by telephone or radio, are still widely used in some
countries.
For the limitations which are listed, many potential solutions have been devised by
suppliers, and can work well in some situations. Also, careful design can eliminate
some problems. Networks consisting of more than one type of telemetry link are also
an option if no single method is appropriate, or if backup transmission routes are
required at each instrument in case of failure of any one method (e.g. radio backed up
by cell phone). Interfaces may also be required to locally operated systems, such as
the SCADA (Supervisory Control and Data Acquisition) systems which are some-
times used at reservoirs, hydropower schemes and other control structures.
The connections to individual instruments typically consist of a data logger, to
keep a record of values which can be downloaded at each visit, and a modem, to
translate the signal into a form suitable for transmission by telemetry. The logger
and data link may allow for multiple sensors, as with an automatic weather station,
for example. Additional channels may also be used for sensors internal to the
instrument or the logger/modem housing to monitor the status and environmental
conditions of the instrument; for example, battery or solar panel condition, air tem-
perature, and humidity, and sometimes a GPS unit for time and location informa-
tion (e.g. for satellite telemetry).
Telemetry connections can be bi-directional or one way only. Simplex connec-
tions are links in which the instrument sends packets of data at predefined times, or
when a critical threshold is exceeded, whilst duplex connections allow downloading
of data on demand. A duplex system provides the flexibility to increase the sampling
(polling) rate of instruments when required (e.g. as a flood starts to develop) and also
allows the operational status of the instrument to be checked remotely. By contrast,
simplex systems are cheaper to install and operate, although with the risk of com-
munications clashing between instruments if they transmit at the same time.
Overall control of a telemetry network is typically from one or more central
computers which will periodically poll, or update, values from the network. Modern
data gathering systems typically include a wide range of functionality including:
● Interfaces to a range of data sources and systems
● Map based displays of instrument locations and data values, and spatial data
(e.g. weather radar data)
● Summary presentations of instrument status and data returned
● Data validation facilities (possibly)
● Report and graph generation facilities
● Alarm Handling options with transmission of alerts by email, SMS, fax etc.
● A wide range of options for onward transfer of data (e.g. to a flood forecasting
model, automated dialling system, or permanent database system)
● Database options for short term on-line storage of data, and longer term off-line
storage of data
Alarms can include rainfall depth duration values, river level thresholds, tidal level
thresholds, and other types of threshold (see Chapter 3). Some systems may also be
programmable, so that simple flood forecasting models such as level to level corre-
lations can be operated on the telemetry system as a back up to the main forecasting
system. Also, multicriteria alarms and rules might be included (e.g. IF X > Y AND
Z > A THEN…). In control rooms, large wall mounted ‘mimic panels’ can help
with providing an overview of current system status against a backdrop of key
information, such as reservoir locations, towns and catchment boundaries, although
are increasingly being replaced by computer displays.
A hydrometeorological database is usually either an integral component of the
system, or may be operated alongside the system as a long term repository for the
near real time data. Many such systems are available commercially or have been
developed by national hydrological and meteorological services, and the function-
ality might typically include:
● Database summary options including key metadata for individual stations
● Statistical reporting functionality (e.g. hydrological year books, extreme value
statistics)
● Data validation tools for checking, correcting and infilling erroneous data values
● A wide range of map based, reporting and graphical options for display and
printing of data
● A range of data conversion options (e.g. from river levels to flows, or hourly
values to daily values)
● Possibly a range of data analysis options (e.g. for stage discharge relationships,
flood frequency analysis)
For database and telemetry systems, most modern systems provide options to facili-
tate the exchange of spatial and time series data through agreed data formats (e.g.
XML) and metadata standards.
2.3.2 Network Design
The topic of network design for river flood warning and forecasting applications is
covered in a number of guidelines, manuals and papers (for example, World
Meteorological Organisation 1994a, 1998; USACE 1996; NOAA/NWS 1997;
Environment Agency 2002, 2004; Sene et al. 2006), although recommendations can
be specific to local meteorological conditions (desert, mountain, tropical etc.), the
types of flooding mechanisms experienced, and other catchment and coastline
specific factors.
Some general issues to consider in network design include:
● The accuracy, reliability and lead time requirements for flood warnings
● The likely performance of any new or existing instrumentation under flooding
conditions
2.3 Instrumentation Networks 47
48 2 Detection
● The reliability of any existing or proposed telemetry links under flooding
conditions
● The level of flood risk at the location, or locations, for which flood warnings are
required
Here, flooding conditions can include high river or tidal levels, and the associated
high winds and heavy rainfall which often accompany flood events, and an assess-
ment of likely performance under these conditions usually forms part of the design
study (e.g. is the instrument range sufficient to monitor all likely conditions, and
are the electronics above likely maximum flooding thresholds). Backup power units
and lightning conductors may also be needed and, in cold climates, heaters may be
needed to ensure operation in snow or ice.
The lead time requirement for flood warning can also influence network design.
For example, for river monitoring sites, to assess local conditions, ideally an instru-
ment would be installed at or near the location for which flood warnings are
required. However, typical rates of rise of river levels in flood events may be so fast
that the flood warning threshold level would have to be set to a low value to achieve
a useful lead time, causing too many false alarms.
Some ways of extending the lead time would therefore be to install a gauge fur-
ther upstream, or to develop a forecasting model to the original proposed gauge
location. Both methods introduce some uncertainty into the flood warning process,
and both approaches might be used to help to reduce that uncertainty, possibly also
using data assimilation and a probabilistic approach for the forecasting component,
as described in later chapters.
Similarly, for coastal locations, a tide gauge may be available at or near the loca-
tion of interest, but if, for example, several hours of advance warning are needed to
evacuate properties or to operate a tidal barrier, then locations further afield would
need to be considered (or additional instruments installed), probably combined with
use of surge forecasting models. Offshore monitoring also provides early warning
of deep swell and Tsunami events not linked to local storms. For locations with
complex wave and surge patterns (e.g. some harbours, and coastal reaches), on site
monitoring is often the only way to resolve these effects.
For raingauges, the flood warning or forecasting requirement may be simply to
give an idea of rainfall in the general area, or to provide estimates of catchment
rainfall or rainfall distribution for lumped or distributed rainfall runoff models. If
the raingauge density is insufficient, then additional raingauges might be installed,
including gauges in nearby river catchments. Existing alternatives, such as weather
radar, might also be considered, if the coverage and accuracy is sufficient in the
locations of interest.
Given that major operational decisions may be taken based on the data provided,
the issue of reliability (or resilience) is also important, and often one or more
backup instruments may be identified in case of failure of any one instrument dur-
ing a flood event. For a river monitoring gauge, that might be a gauge further
upstream, whilst for a tide gauge another gauge might be selected from the same
coastal reach. Backup instruments might also be installed at the same site or nearby
locations, particularly in high risk locations (e.g. city centres).
A number of techniques can assist with network design including:
● Digital terrain models – for radio path or line of sight studies, for estimating
catchment characteristics (area, slopes, elevations etc.), and for viewing poten-
tial instrument locations against a backdrop of topography, flood risk locations,
and other factors
● Hydraulic and hydrological analyses – to study the likely response of the catch-
ment or coastal reach at the proposed instrument locations (e.g. rate of rise of
levels for typical events, typical depth-duration values for rainfall, times of travel
of flood or surge waves from distant locations, possible backwater and conflu-
ence influences etc.)
● Meteorological analyses – using historical raingauge data, weather radar data and
possibly Numerical Weather Prediction model outputs to help in developing an under-
standing of flood generating conditions, with the likely scale, speed, and direction of
storms all being important factors in deciding on appropriate raingauge locations
● Temporary gauges – installation of temporary gauges, maybe without telemetry, to
investigate river or coastal characteristics at potential sites, and to check site secu-
rity and feasibility (e.g. risk of vandalism, objections from nearby residents etc.)
More generally, it is often worthwhile considering other current or planned applica-
tions of the data; for example, for other purposes (e.g. water resources monitoring,
ocean climate monitoring, port and harbour operations), or for providing flood
warnings to additional locations. For example, considerable cost savings can some-
times be realised by considering opportunities to share data between departments
or organisations, or by finding alternative nearby site locations which would serve
more than one purpose.
Another consideration, particularly for flood forecasting applications, is the level
of uncertainty which can be tolerated. For example, for river forecasting models, it
is often not economically feasible to place raingauges and river gauges in all major
subcatchments, with the result that some inflows to the model (lateral inflows) will
need to be estimated, introducing a source of uncertainty into the process. Also, it
might be desirable to install more raingauges to obtain a better idea of rainfall distri-
bution in and around the catchment, and to upgrade gauging stations so that they are
better able to record accurate values at high flows. These various trade-offs and
compromises are all part of the process of network design, and in part are one of the
motivations for the increasing interest in using probabilistic and ensemble tech-
niques to help to quantify the uncertainty (see Chapters 5–8 for examples).
2.3 Instrumentation Networks 49
Chapter 3Thresholds
Flood warning thresholds define the meteorological, river and coastal conditions at
which decisions are taken to issue flood warnings, whilst flooding thresholds are
the values at which flooding occurs. Normally, a flood warning threshold will be
set to achieve an acceptable lead time before the flooding threshold is reached, or
may be time based (as with tropical cyclones, for example). Alternative names for
flood warning thresholds include triggers, criteria, warning levels, critical condi-
tions, alert levels and alarms, and sometimes a range of values will be required as
warnings are escalated from advisories (or watches, or pre-warnings) through to
full warnings. Threshold values may be set based upon experience or analysis of
historical data, or using conceptual, data based or process based modelling studies.
Values may be fixed (static) for all flood events, or dynamic, varying depending on
how each event unfolds. This chapter describes a range of techniques ranging from
simple fixed flood warning thresholds through to probabilistic approaches, together
with several examples of approaches to performance monitoring of thresholds.
3.1 Rainfall Thresholds
Observations or forecasts of heavy rainfall often provide the first indication of likely
river flooding. Some typical uses of rainfall thresholds are for the initial mobilisation
of staff (e.g. opening an incident room), and moving to an increased frequency of
monitoring river conditions and operation of flood forecasting models.
Rainfall values can be obtained from observations (e.g. raingauges, weather radar,
satellite) or forecasts (e.g. nowcasts, Numerical Weather Prediction models), with
observed values usually providing higher accuracy, but with a shorter lead time before
the onset of flooding. Best practice is to calibrate methods directly to the type of input
data (or forecasts) to be used operationally, to account for any systematic or other
differences between rainfall measurement and estimation techniques. Rainfall
amounts can also be used directly to initiate flood warnings although, due to the
various uncertainties in how rainfall translates into river flows (see later), this
approach is used much less widely, with a greater risk of a high false alarm rate com-
pared to warnings based on river levels.
K. Sene, Flood Warning, Forecasting and Emergency Response, 51
© Springer Science + Business Media B.V. 2008
52 3 Thresholds
For some types of rainfall inputs, such as weather radar observations, or rainfall
forecasts, rainfall values will usually be available on a gridded basis, so that the
criteria for raising or displaying an alarm might apply to a single grid square, or to
the average value across a region or in a river catchment. Information on rainfall
amounts and accumulations can also be presented spatially; for example, as maps
of rainfall amounts with overlays of catchment boundaries, rivers, topography and
flood risk locations, and in terms of probabilities of exceedance (if using ensemble
rainfall forecasts). Spatial estimates for rainfall distribution can also be derived for
raingauges, if required, using the techniques described in Chapter 2.
Rainfall threshold (or alarm) criteria are often expressed in terms of the quantity
(depth) of rainfall in a given period (duration) which has the potential to cause
flooding. A range of depth-duration values may be used; for example, an alarm
might be raised if rainfall is forecast to exceed 25 mm in any 3 hour period, or
40 mm in any 6 hour period. Alternatively, thresholds may be expressed in rainfall
frequency terms calculated from a statistical analysis of historical records, such as
the 1% or 10% exceedance probability, or the 1 in 100 year or 1 in 10 year return
period. Values can be tested by analysis of long term historical rainfall records; for
example, by comparing the number of alarms which would have been raised com-
pared to the number of actual flooding events (or near misses), and estimating the
number of false alarms which would have occurred (see Section 3.3).
Of course, rainfall values alone do not provide a full indication of flooding
potential, since the catchment characteristics (topography, land use etc.), current
catchment state (e.g. soil moisture, snow cover) and other factors (e.g. reservoir
levels) may also influence the magnitude and timing of flooding. Rainfall thresh-
olds are therefore often combined with indicators of catchment response and the
current catchment state.
Table 3.1 illustrates a simple approach of this type, and is adapted from one of
several examples in Environment Agency (2002).
Here, the codes and locations are for three hypothetical Flood Warning Areas, and
the Forecast Rainfall values could be either best estimates, or worst-case scenarios.
The depth/duration pairs (e.g. 20/6) are in units of mm of rainfall, and hours, and
Table 3.1 Illustration of rainfall alarm criteria (Adapted from Environment Agency 2002;
© Environment Agency copyright and/or database right 2008. All rights reserved)
Code and location
Forecast Rainfall (mm)
Flood Watch Criteria based
on SMD
6
hours
12
hours
18
hours
24
hours
SMD
(mm) <5 5–20 21–40 >40
FW021 Bridgetown 3.8 5.8 8.8 10.8 39.0 20/6 25/6 30/6 30/6
25/12 30/12 40/12 45/18
FW022 Southford 3.0 5.0 8.0 10.0 50.3 24/6 28/6 32/6 35/6
30/12 35/12 40/12 45/18
FW023 Northtown 3.0 5.0 8.0 10.0 50.8 24/6 28/6 32/6 35/6
30/12 35/12 40/12 45/18
catchment conditions are expressed in terms of the soil moisture deficit (SMD), which
is the depth of rainfall which would be required to bring the catchment to saturated
conditions (i.e. the amount of water required to bring the soil to field capacity).
Some other possible indicators for catchment conditions (e.g. World
Meteorological Organisation 1994; USACE 1996) include recent rainfall, current
river flows, Catchment Wetness Index, Base Flow Index, Antecedent Precipitation
Index, and borehole levels. Where, as is often the case, direct observations are not
available, values are often computed from the soil moisture accounting compo-
nent of rainfall runoff models (see Chapter 6) or as a secondary (diagnostic) out-
put from the land-atmosphere component of Numerical Weather Prediction
models (e.g. Cox et al. 1999). Satellite based methods also show potential for
remote sensing of soil moisture.
Another approach to setting thresholds is to use a catchment rainfall runoff and
flow routing model to explore the rainfall amounts required to achieve flooding for
a range of durations and catchment initial conditions. One approach is to first derive
a typical storm profile from historical data, describing the variation in rainfall
during the course of an event. These values are then scaled by magnitude and dura-
tion, and the resulting synthetic storms used as input to the catchment model. For
each duration, the depth required to reach flooding thresholds is noted, perhaps for
a range of catchment conditions, and the resulting table of values can then be used
as the basis for estimating the rainfall thresholds for that location. Other factors,
such as reservoir drawdown at the start of an event, or the depth of water in off-line
storage areas, might also be considered in setting thresholds. Some possible criteria
for flooding thresholds include bank full flows, peak river levels exceeding a
threshold at which flooding commences, or flood flows of a given probability
(return period). The latter method is often used for ungauged catchments and in
ensemble forecasting approaches (see Chapter 5).
For example, these types of method form the basis of the Flash Flood Guidance
concept (FFG) developed by the National Weather Service in the USA (Sweeney 1992).
Flash Flood Guidance is defined as the amount of rainfall of a given duration over a
small basin needed to create minor flooding (bank full) conditions at the outlet of the
basin. The approach has been used operationally since the 1970s and was integrated into
a system called the Flash Flood Guidance System in 1992, and has more recently been
considered for providing early alerts for debris flows (NOAA-USGS 2005). Chapter 8.2
provides some examples of international initiatives using this approach.
In the original version of the method, threshold values of runoff were estimated
based on the outputs from a lumped rainfall runoff modelling approach. More
recent developments (National Weather Service 2003) have included improvements
to the method for areas of the country where rainfall intensity and land characteristics
have more influence on flash flooding than soil moisture (e.g. some desert regions),
and the introduction of a distributed (grid based) hydrological modelling approach
for estimating thresholds and for real time soil moisture accounting. Operationally,
estimates of rainfall depths and durations (e.g. 1, 3, 6, 12 and 24 hours) are
compared with the threshold values appropriate to the estimated soil moisture
conditions. The resulting exceedance over threshold values can then be mapped.
3.1 Rainfall Thresholds 53
54 3 Thresholds
Spatial analysis tools are also available to examine rainfall accumulations, rainfall
intensities and guidance values at point or catchment scale.
Some other developments in the area of rainfall threshold based approaches (e.g.
Martina et al. 2006; Georgakakos 2005, 2006; Reed et al. 2006; Collier 2007;
Fouchier et al. 2007) have considered or implemented systems which include vari-
ous permutations of the following techniques:
● Bayesian techniques requiring optimisation of a utility function combining the per-
ception of stakeholders, historical losses, and perhaps the losses from false alarms
● Alternative soil accounting approaches using a variety of conceptual and process-
based catchment models
● Artificial neural network methods which improve forecasting skill by ‘learning’
from meteorological and streamflow response
● Thresholds based on the return periods/recurrence intervals of model flows
based on long term simulations using historical or synthetic rainfall data
● Development of indicators of flash flood potential which can be searched in real
time including soil moisture, channel constriction/debris risks, storm depth-
duration, direction and velocity
Ensemble approaches are also increasingly being used, in which the probability of
rainfall is displayed on graphs, maps and tables and compared to probability based
thresholds. Risk-based approaches, combining probability and consequence, can
also be used, and Box 3.1 provides an example of an operational system in the
Netherlands which uses ensemble forecasts of rainfall to provide rainfall alarms to
assist with water management operations in polder regions.
Chapter 5 provides further information on ensemble forecasting techniques and
Chapter 10 gives more background on risk-based and cost loss approaches to
decision making.
In addition to the use of rainfall thresholds, various other meteorological indicators
have been considered for use in providing early warning of flooding, with an
emphasis on probabilistic techniques.
One of the earliest methods was a combined deterministic/stochastic approach
which was developed for application in the Mediterranean areas of France (Obled and
Datin 1997; Bernard 2004). Observations and forecasts of rainfall and other parame-
ters at lead times of 2–3 days or more are linked to an archive of rainfall and other
parameters for past events; for example, geopotential or temperatures. The technique
can also be used at shorter lead times, using stochastic modelling to link observations
up to time now with likely future scenarios (again based on an historical archive),
with the option of conditioning forecasts on nowcasts and likely limits on daily rain-
fall for the catchment for the type of storm being observed.
Another approach is to use operational mesoscale and other Numerical Weather
Prediction models to monitor parameters which are thought to be good precursors
of flooding, including:
● Potential vorticity – as an indicator of atmospheric stability
● Convective Available Potential Energy (CAPE) – an indicator of the energy
available for a storm to develop
Box 3.1 The KNMI precipitation alert system
The Netherlands is a low lying country with extensive areas of reclaimed land,
known as polders, and more than half of the population lives in areas below sea
level. The main flood risk in the polder areas arises from heavy rainfall falling
directly on the polders, often combined with high river levels or sea levels (due
to surge, wind and wave action), which may limit the ability to remove excess
water. A network of pumping and drainage systems is used to manage water
levels in polders, and rainfall observations and forecasts play an important role
in optimising these operations. Rainfall alarms are used to mobilise staff, alert
third parties, obtain emergency pumping equipment (if required), and to help
in deciding when to start pumping operations.
Since 2003, as part of a collaborative project with the Union of Water
Boards, the Royal Netherlands Meteorological Institute (KNMI) has been
issuing probabilistic rainfall alarms to selected Water Boards based on ensem-
ble forecasts and observed rainfall data. For lead times up to 36 hours ahead,
deterministic forecasts are used from the national HIRLAM Numerical
Weather Prediction model whilst, at longer lead times, 50 member ensemble
outputs are obtained from the European Centre for Medium-Range Weather
Forecasts (ECMWF)
Values for rainfall depth and duration are estimated for each polder based on
weather radar observations of rainfall in the previous 5 days, and rainfall
forecasts for the next 9 days (Fig. 3.1). Critical rainfall depth-duration values
are defined by each Water Board on the basis of historical rainfall records and
flooding histories and can depend on soil type, the proportion of urban areas,
storage capacity, pumping capacity, the time of year, and other more subjective
Fig. 3.1 Example of a 9 day probabilistic rainfall forecast (KNMI)
(continued)
3.1 Rainfall Thresholds 55
56 3 Thresholds
● Precipitable water – or precipitable water vapour
● Intensity and direction of the low level flow
These indicators can be calculated from wind, moisture, pressure and other fields
(e.g. Collier et al. 2005; Environment Agency 2007). These developments are often
linked to the general trend to develop higher resolution (storm scale) models better
able to forecast the development of convective and other events. Lightning activity
has also been considered as a possible indicator of the likelihood of heavy rainfall
and flash flooding (e.g. Price et al. 2007).
3.2 River and Coastal Thresholds
3.2.1 Introduction
River and tidal thresholds (or triggers) are a key component in many flood warning
systems, and define the levels and possibly other variables (e.g. wind speed and direction)
at which the decision to issue a flood warning should be taken, or other actions initiated
Box 3.1 (continued)
factors. The probabilities at which alarms are issued are calculated from a risk
assessment which compares the cost of mitigating actions (pumping etc.)
with the estimated losses (damages) if no action is taken. The so-called cost
loss ratio gives an indication of the appropriate probability thresholds to use
for each polder, which can be refined based on experience.
Thus the criteria for issuing warnings are defined in terms of a risk profile
for each polder consisting of a range of depth, duration and probability combi-
nations (e.g. 50 mm in 72 hours with a 33% exceedance probability). Over a
number of events, consistent use of these profiles should help to minimise the
economic impacts from flooding. The criteria are checked automatically on an
hourly basis and, when individual values are exceeded, the relevant Water
Board managers receive an email alert or text message, and the alert is also
published on a secure website.
A user group meets at regular intervals to share experience on use of the
system, and in particular to discuss approaches to the setting and verification of
appropriate risk profiles, since this is a new and developing application for ensemble
forecasting. Planned developments include use of radar rainfall nowcasts for
shorter term rainfall forecasts, and use of wind and tide forecasts, since wind effects
can add to the flood risk in larger polders via wave and local surge effects.
Reference: Kok C J, Vogelezang D H P, Wichers Schreur B, Holleman I
Description and use of the automated warning system for the Dutch water boards,
KNMI.
(e.g. mobilisation of staff, more frequent monitoring). They are sometimes called
Action Thresholds. Some systems may also automatically issue a warning at these
levels without any human intervention (e.g. using email, sirens or pagers) although
there are many issues to consider in taking an automated approach of this type; for
example, the likely false alarm rate, and the possibility of missed warnings (see
Chapter 4). Other types of thresholds can include parameters such as ice motion and
river flows. Time based criteria may also be used in some situations; for example, the
time before landfall for hurricanes, typhoons and tropical cyclones (see Chapter 9 for
further discussion of this topic).
Observations are normally made by telemetry but, where this is not available, or
extra safeguards are required, observers and patrols may be deployed on site,
or other methods such as CCTV or webcams used. Community representatives may
also monitor conditions in some flood warning schemes. On site observations can
be particularly useful where site specific flood risks can occur, such as defences
breaching, waves overtopping at sea defences, or bridges being blocked by debris,
and as additional backup for high risk locations such as town centres.
Threshold values are normally defined based on a combination of experience,
analysis of historical data, and possibly detailed hydraulic and other modelling of
river or coastal response. Values are usually chosen to achieve the required warning
lead time, without causing an unacceptable number of false alarms and, for instru-
ments at the location of flooding, are set in relation to the flooding threshold, as
illustrated in Fig. 3.2 for the case of a river level threshold.
Here, the flooding threshold is the gauge reading at which flooding impacts begin
(and for which a warning is required), such as property flooding, or flooding of roads,
or overtopping of flood defences (levees), and is sometimes called a Result
Threshold. In practice, the actual warning lead time will be less than the potential
TimeWarning Lead Time
Flooding Threshold
Flood WarningThreshold
River Level
Fig. 3.2 Illustration of a flood warning threshold for an at site gauge
3.2 River and Coastal Thresholds 57
58 3 Thresholds
value indicated in the figure since factors such as decision times, and flood warning
dissemination times, must be accounted for, as described in Chapters 4, 5 and 10.
Also, it is advisable to include some allowance (contingency) in the setting of values
to allow for uncertainty in data, models and event specific factors.
The terminology and approaches used vary between organisations and countries,
but some typical types of threshold (or trigger) include:
● At site or local values – where the flood warning is issued based on values at or
near the location for which the flood warning is required
● Upstream or remote values – where the flood warning is issued based on values at
a site further upstream in the river network, or further offshore or around the coast
in the case of coastal triggers, to provide additional lead time at the site of interest
● Forecast values – where the flood warning is issued based on the output from a
river or coastal forecasting model for the site or other location of interest
For each type of threshold, there is a trade off between the accuracy, reliability and
timeliness which can be achieved; for example, if a threshold is lowered, this normally
increases lead time, but may also increase false alarm rates (e.g. USACE 1996), whilst
forecast values may be set at a higher threshold (e.g. a flooding or result threshold) than
for at site values due to the additional lead time available from model outputs.
To provide additional lead time and resilience, a site may have more than one
type of threshold, with warnings being issued on the basis of exceedance of any one
value, or other permutations. Values may also be nominated as the primary, second-
ary (or backup) or failsafe threshold, with the choice depending on the relative per-
formance of each type of threshold. Within each category of threshold, there may
also be a range of values for different operational and warning conditions. For
example, a site might have standby or alarm values which are set at a low level for
early warning of possible events, and mobilisation of staff, and a range of flood
warning values to escalate the severity of the warning as river or sea levels rise.
Also, as flood hazard mapping techniques improve, and flood warning dissemi-
nation systems become more sophisticated, it is increasingly becoming possible to
target warnings to smaller areas, or even to individual properties, with the advan-
tages of reducing the number of false alarms experienced by property owners, and
allowing for a more phased approach to warning and evacuation of properties.
If this approach is used, then each zone or sub-area will have its own warning
threshold level, both for the At-Site gauge (if available) and for any Remote gauges
or Forecast values. Of course, the terminology and formats used for flood warning
procedures, and the criteria for escalating and downgrading alerts, differ widely
between organisations but the general principle of escalation of warnings, followed
by confirmation that the threat has passed, is widespread. Figure 3.3 provides an
example of this general approach for a river flood warning application.
In this hypothetical example, the Flood Warning Area at Newtown is divided into
four sub-areas or zones, identified by codes FW001 to FW004. The corresponding
warning threshold levels are shown for the At-Site gauge, and the Remote gauge would
also have its own set of values (not shown). If a forecasting model output is available,
that too would have a set of values based on a consideration of flooding thresholds,
FW001–Riverside Paths
FW002–Riverfront Apartments
FW003–Town Centre
At Site Gauge
FW004–Power Station
Remote Gauge
FW004FW003FW002FW001
RIVER
To Bridgeham
Fig. 3.3 Illustration of at site and upstream thresholds (not to scale)
model lead times, and other factors. Table 3.2 provides a simplified illustration of how
these values might be implemented into a set of operational Flood Warning Procedures
for the At-Site gauge (sometimes called an Action Table or Flood Intelligence Card),
although it is important to note that the details of warning messages and operational
responsibilities differ widely between countries and organisations.
Values are expressed in terms of gauge readings, but absolute values might also
be included, relative to a national datum level. Other thresholds (sometimes called
Information Thresholds) might also be included to indicate other useful informa-
tion, such as the highest level recorded at the site, and peak levels for historical
flood events. A similar table would also be produced for the Remote Gauge, with a
separate set of values.
As illustrated, the gauge at Newtown might also be a Remote gauge for another
Flood Warning Area further downstream, and an example is included for the town
of Bridgeham. Following the initial standby alarm, a series of flood warnings is
issued as the event escalates, and operational instructions are also issued where this
requires direct contact with other organisations or the public. As noted in Chapter 4,
this allows an audit trail of actions and decisions to be maintained during the event,
including any departures from the agreed approach as each warning level is
exceeded (for example, based on other information which may be available, such
3.2 River and Coastal Thresholds 59
60 3 Thresholds
as information from operational staff observing river levels on site). Finally, an All
Clear is issued when the river levels have dropped back to standby levels, and the
forecast indicates that levels will continue to drop.
In this example, a forecasting model output is also available, and the Forecast
Thresholds are included in the procedures. The extent to which the Forecast Thresholds
are formally integrated will depend on organisational policy, the confidence in model
outputs, and the expertise and background of duty officers. Some examples of the way
that the forecast outputs could be used include:
1. Issue the warning either if the observed value is exceeded, or if the forecast value
is exceeded
2. Issue the warning if both the observed and forecast values are exceeded
3. Consider issuing the warning if the observed value is exceeded, using the fore-
cast outputs to take the final decision
4. Generate warnings to individual properties or groups of properties from real
time forecasts of the inundation extent
Table 3.2 Illustration of flood warning thresholds for the example in Fig. 3.3
Observed level (m) Forecast level (m) Action required
3.2 >3.8 STANDBY ALARM
Issue FLOOD WATCH: WW001 – Newtown and
Bridgeham
Issue OPERATIONAL INSTRUCTION: OP001
– Loud Hailer Patrol crew standby for Newtown
Record confirmation of receipt of FLOOD WATCH:
WW001 from Newtown Town Council, Power
Station at Newtown, Bridgeham City Council
3.5 >4.0 Issue FLOOD WARNING: FW001 – riverside paths
at Newtown
3.8 >4.2 Issue FLOOD WARNING: FW002 – riverside apart-
ments at Newtown
Issue OPERATIONAL INSTRUCTION: OP002
– start patrols along flood defences in Newtown
Town Centre, assign liaison officer to Newtown
Police Emergency Command Centre
4.0 >4.4 Issue FLOOD WARNING: FW003 – Newtown Town
Centre
Issue OPERATIONAL INSTRUCTION: OP003
– Loud Hailer Patrol in Newtown Town Centre
4.2 >4.5 Issue FLOOD WARNING FW004 – power station at
Newtown
Issue FLOOD WARNING FW011 – riverside proper-
ties in Bridgeham
3.1 <2.5 Issue ALL CLEAR
FW001 – Riverside Paths at Newtown
FW002 – Riverside Apartments at Newtown
FW003 – Newtown City Centre
FW004 – Power Station at Newtown
FW011 – Riverside Properties at Bridgeham
Other possibilities might also be envisaged. A maximum forecast lead time (horizon)
or minimum observed level might also be specified, with forecasts at longer lead
times or for lower levels given less weight or not considered at all due to the
increase in uncertainty with increasing lead time.
The first approach takes advantage of the potential additional lead time from
forecasts, but raises the prospect of more false alarms if the model overestimates
levels. The second approach helps to guard against the risk of a model providing
erroneous outputs, but would possibly lead to less lead time in warnings, and could
result in missed warnings if the forecasting model underestimates levels.
The third approach relies mainly on the forecasting model outputs, with the observed
value acting as an initial alert level above which issuing a warning should be considered.
Another possibility is to introduce the concept of soft and hard limits, or a contingency
for uncertainty, in which there is a range of levels in which the duty officer can provide
an input to the decision making process, but once the hard limit is reached a warning
must be issued. This again increases the number of decision criteria, although some
forecasting and telemetry systems can help in automating application of this approach.
The fourth approach requires a forecasting model which is able to estimate flood
inundation extents in real time, such as a one-dimensional or two-dimensional
hydraulic model. The resulting extents can then be intersected with maps of prop-
erty locations, and lists of property addresses and contact details generated for a
range of forecast lead times. In principle, these can then be used to automatically
generate warnings to individual properties (e.g. by telephone or cell phone) once
the forecast has been approved. However, this is a new and developing area, and the
issue of confidence in the model outputs again needs to be carefully considered
before implementing this approach.
Another factor to consider in deciding on the degree of automation is the worst-case
scenario of a widespread flood event. For a small number of locations, it may be prac-
ticable for a duty officer to inspect every output (observed levels, and forecast values,
if available) and take a decision based on experience and judgment. However, in a
major event, many hundreds of threshold levels may be exceeded during the course of
the event, and duty staff will have less time to consider the accuracy or appropriateness
of each value, except possibly in high risk locations, where major decisions need to be
taken (e.g. on evacuating population centres, or closing down key utilities). Similar
considerations apply for very fast responding catchments or coastal reaches.
3.2.2 Simple Forecasting Techniques
Simple forecasting techniques provide an alternative or supplement to observed
thresholds, and typically use information from remote sites to estimate conditions
at the site of interest, or information on the rate of rise at the site itself. This cate-
gory of methods is sometimes called a ‘Simple Triggers’ approach (e.g. Graham
and Johnson 2007), and may introduce dynamic thresholds, which vary between
events, rather than fixed values.
3.2 River and Coastal Thresholds 61
62 3 Thresholds
The distinction between these methods, and forecasting models of the type
described in Chapters 5–8, is not clear cut, and threshold based techniques, as
described in the previous section, can also be viewed as a simple type of forecast-
ing, based on assumptions about typical rates of rise and travel times for flood
events. Here, the distinction is taken to be:
● The methods do not attempt to model physical processes.
● The methods can easily be applied using non-computerised approaches such as
graphs, look up tables and charts (although can be computerised if required).
● Where a computerised approach is needed, the calculations are simple enough
to perform on a telemetry system, if required, rather than a dedicated flood
forecasting system.
The advantage of using paper based techniques is that the methods can be applied
by staff without computer skills, and are quick and cheap to implement. Also, where
more sophisticated techniques are available, the methods can be used as a backup in
case of system failure (e.g. due to power cuts), and as a cross check on the plausibility
of the outputs from more advanced techniques. For example, an observer on site
might relay observations of river levels by telephone or hand held radio for a duty
officer to use, even if both the telemetry and forecasting systems have failed.
Similarly, if the methods are implemented on a telemetry system, then this can pro-
vide additional backup and resilience to the flood forecasting system, and modern
telemetry systems are often capable of running simple types of model.
Simple forecasting techniques include correlations (single and multiple regres-
sions), multicriteria approaches (e.g. look-up tables, carpet plots, nomograms),
transformation matrices, rate of rise triggers, and time of travel (isochrone) maps.
3.2.2.1 Correlations
Correlations are widely used in flood warning applications, and relate parameters
at the location where an estimate is required to real time observations or forecasts
at one or more remote locations.
For rivers, correlations are usually performed in terms of river levels or flows,
and can be calculated using either peak values for a number of representative his-
torical events, or values for the full flow range. Peak to peak correlations are some-
times called Crest Stage forecasts, and Fig. 3.4 shows two examples (World
Meteorological Organisation 1994).
The first figure shows a peak level to level correlation between two river sta-
tions, whilst the second shows a set of correlations in which the choice of relation-
ship depends on current runoff conditions. Other secondary variables may be used
including snow cover, air temperature, soil moisture, and time of year.
Whole hydrograph correlations are usually estimated assuming a typical lag time
between upstream and downstream stations. An optimum value can also be estimated by
repeating the calculation for a range of assumed lag times and finding the value which
gives the smallest correlation coefficient. Correlations are normally performed in
terms of levels since this avoids the need for a rating equation or stage discharge relation-
ship at each site to convert levels to flows (see Chapter 2). However, levels at an individ-
ual gauge can be affected by other influences, such as backwater effects from tributaries,
gate operations or tidal influences, which can introduce additional uncertainty into the
relationship. One of the risks in this approach is also that these effects may only become
apparent in large flood events, beyond the range of calibration. Flow based correlations
help to avoid this problem, but as noted introduce the uncertainties arising from stage
discharge relationships (where used), and can only be applied between sites for which
flow estimates are available.
Correlations can also be affected by inflows, storage and losses between the two
stations being used (e.g. tributaries, or spills onto floodplains, or lakes and reservoirs),
and by the extent and motion of storm events over the catchment. One option in this
situation is to use a multiple regression approach to include other records in the rela-
tionship, such as gauges further upstream, and on tributaries or floodplains (e.g. Torfs
2004), or lake or reservoir levels. Another consideration is whether a correlation
derived from peak values should be applied over the full flow range to derive a fore-
cast hydrograph. This approach is often used operationally and sometimes works
well, although it is important to note that lag times and wave speeds vary over the
flow range, so that the shape of the rising limb of the hydrograph can be considerably
in error, which is important if the errors are close to flood warning threshold values.
Correlations can also be used in tidal applications, with examples including sin-
gle or multiple regressions for the following situations, often including a time delay
factor between the observed or forecast point and the location for which the esti-
mate is required:
● Estuaries – relating levels at a point in the estuary to tidal conditions and/or river
levels upstream
● Tide Gauges (observations) – relating levels at a tide gauge to conditions at gauges fur-
ther around the coast; for example, to estimate the likely development of surge events
Fig. 3.4 Examples of level to level and flow to flow correlations (Reproduced from the WMO
Guide to Hydrological Practices – Data Acquisition and Processing, Analysis, Forecasting and
Other Applications, courtesy of WMO)
3.2 River and Coastal Thresholds 63
64 3 Thresholds
● Tide Gauges (forecasts) – relating offshore forecasts of level from a surge model
at one or more node points to conditions at the shoreline
Additional parameters which may be included are wind speed and direction, and
possibly wave heights. Relationships are often expressed in the form of look up
tables advising on when to issue warnings for various combinations of actual or
forecast tide levels, wind speed, wind direction, maximum surge, and maximum
wave heights.
Some other ways of presenting information include equations, nomograms or
carpet plots. For example, Fig. 3.5 shows an example of a flood warning plotting
chart for an estuary location in North West England where water levels are influ-
enced by both fluvial flows and tidal levels (although in this case the graph repre-
sents the theory behind the model and has now been replaced by automated
forecasting techniques, including use of tidal forecasts). The chart gives estimates
of levels at Lancaster Quay based on forecasts for tidal levels at Fleetwood Dock
and river flows at Caton.
For both rivers and coastlines, hydrodynamic and other numerical models can
also be used to guide the development of correlations and other types of threshold,
such as time based thresholds. Multiple model runs can be performed for a wide
range of scenarios to explore the key influences on flood response at the location(s)
of interest, possibly deriving a range of relationships for different initial conditions
Fig. 3.5 Example of a flood warning plotting chart for a coastal location (Environment Agency
2004, © Environment Agency copyright and/or database right 2008. All rights reserved)
for snow cover, reservoir storage and other factors, and examining response for
levels beyond the range of the historical data. In some cases, a considerable amount
of exploratory work may be required to determine the appropriate combinations of
variables to use in real time.
This approach is often used for generating operational look up charts and tables
to assist with the operation of tidal barriers, for example. As another example,
transformation matrices derived from detailed off-line scenario modelling for
wave transformation and overtopping are used operationally in coastal flood warning
procedures for some locations around the coast of England and Wales (Environment
Agency 2004a). Models of this type can also be implemented in real time, if the
model run time is short enough, and the model stability and convergence is acceptable,
and Chapters 6 and 7 discuss this topic in more detail.
3.2.2.2 Time of Travel Maps
For river catchments, time of travel or isochrone maps (e.g. World Meteorological
Organisation 1994) can be a useful aid in flood warning applications, and show
estimated or typical travel times from the onset, centroid or peak of a rainfall event,
to the peak of flows being observed at various points in the river network, or travel
times between locations in the network. Values can be presented in the form of
tables, graphs or as shaded or contour maps of equal travel times. Figure 3.6 shows
an example of a time of travel map based on times to the lowermost point in the
catchment (World Meteorological Organisation 1994).
Fig. 3.6 Example of an isochrone map (Reproduced from the WMO Guide to Hydrological
Practices – Data Acquisition and Processing, Analysis, Forecasting and Other Applications, cour-
tesy of WMO)
3.2 River and Coastal Thresholds 65
66 3 Thresholds
In this case, the lines of equal time (isochrones) were estimated assuming an
average velocity of flow in the river channels, whilst some other methods for estimating
the time of travel include:
● Analysis of historical rainfall and river level and flow data
● Area based methods using empirical overland flow models to estimate velocities
● Unit hydrograph rainfall runoff modelling techniques
● Catchment models combining rainfall runoff, flow routing and/or hydrodynamic
modelling techniques
Values can also be estimated using Geographical Information Systems, with other
options for presenting information including overlays of catchment boundaries,
gauging stations, flood risk locations etc., and annotations giving information on
catchment response times and peak flows for historical flood events.
If the estimates are based on historical data, then a range of representative events
needs to be selected, and mean or median values for lag times derived across all events,
or individual values derived for different types of event (e.g. snowmelt, floodplain flows
etc.). For modelling based approaches, assumptions also need to be made about the
magnitude, distribution, speed and direction of the storm events which are used. For all
types of analysis, it may be advisable to also consider the scenarios likely to give the
smallest lag times in a catchment to indicate the worst case for flood warning
applications.
Although mainly used for river flood warning, this technique can also be useful
in coastal applications to give an idea of the timing of the peak in astronomical tides
around a coastline, and the typical movement of surge peaks.
3.2.2.3 Persistence Methods
Another simple forecasting technique is to assume that some aspects of the current
observed response will persist into the future. Some approaches which have been
applied operationally include:
● Rate of rise methods – which extrapolate the rate of rise of a hydrograph or tidal
levels as levels rise towards threshold values. Typical or fastest likely rate of rise
values can be estimated from historical data and/or hydrodynamic modelling
results for a number of events and then used operationally to forecast the likely
time of crossing of thresholds. Rate of rise values can also be estimated dynami-
cally during an event, in which case parameters which can be varied to optimise
the performance of the model include the averaging time over which the rate of
rise is calculated, and the required lead time. An optimisation can then be per-
formed to maximise the success rate of warnings, and minimise the number of
false alarms (e.g. Graham and Johnson 2007). Similar techniques can also be
applied to reservoirs, in which different categories of warning are issued based
on the current water surface elevation, and the rate of rise of levels.
● Constant offset – in which a constant value is applied to observed data to com-
pensate for event specific factors. For example, Tissot et al. (2005) compare a
simple persistence based method with a range of more sophisticated modelling
techniques in which the differences between observed tidal levels, and the esti-
mated tidal harmonics, are assumed to persist for the duration of the forecast.
3.3 Performance Monitoring
Threshold based approaches to flood warning are widely used and may be progres-
sively improved using experience gained over successive flood events.
For example, for a river gauging station, if post event analysis shows that a warn-
ing was issued too late at a site, then the warning threshold might be lowered to
allow more lead time, although possibly at the expense of an increase in false alarm
rates. Similarly, if a threshold is resulting in too many false alarms, then after care-
ful analysis the value might be increased, provided that this does not increase the
risk of missing actual events. Alternatively, more accurate approaches might be
investigated, such as development of a flood forecasting model.
As part of the development of a flood warning service, it is usual to review the
performance of flood warning thresholds on a regular basis, and after each major
flood event, and when other changes occur which may influence performance (e.g.
flood defence construction work, dredging, instrument replacements, changes in
forecasting models etc.).
Flooding thresholds should also be regularly reviewed, although it is less likely that
they will need adjusting. However, some examples of when this might be required
include when a flood defence is raised or repaired changing the level at which it is
likely to be overtopped, and if additional properties require adding to the warning
system. As with all components of an operational flood warning system, any changes
to thresholds and alarms should be fully tested and documented before implementa-
tion, and discussed with key stakeholders who may be affected by those changes.
A wide range of methods can be used for monitoring the performance of thresholds.
Similar techniques can also be used at the design stage of a flood warning scheme to
explore how values would have performed based on the historical data available to date.
In practice, the values used are often a compromise between the need for an adequate
warning lead time, to avoid missing flood events, and to minimise false alarms rates.
Values for warning lead times can be estimated by examining historical records to
determine the time difference between crossing of the warning and flooding thresholds.
Note that this time is not the same as the lead time provided to recipients of flood warn-
ings which, as noted earlier, may include additional time delays; for example the time
taken for decision making, or in issuing a warning. Estimates of these actual lead times
are more difficult to obtain, but methods which are used include post event surveys of
people who were flooded, and examination of the records (logs) maintained during the
event by flood warning duty officers, the emergency services and others (e.g. for the
times of phone conversations, and for the estimated time of onset of flooding).
One simple way to present information on lead times is as a histogram showing
the lead time performance across a number of events. Values might also be tabu-
lated by gauge, Flood Warning Area, or catchment or coastal reach, as illustrated in
Table 3.3 for a single gauge across a number of flood events (adapted from an
example in Environment Agency 2002).
3.3 Performance Monitoring 67
68 3 Thresholds
Table 3.3 Example of a lead time summary for several flood warning areas (Adapted from
Environment Agency 2002; © Environment Agency copyright and/or database right 2008. All
rights reserved)
Flood warning area
After start
of flood
<2
hours
2–4
hours
4–6
hours
6+
hours
Modal
value
(hours)
Target
(hours)
FW001 – riverside paths at
Newtown
0 0 7 4 0 2–4 2
FW002 – riverside apartments
at Newtown
1 3 2 0 0 <2 2
FW003 – Newtown City
Centre
0 0 2 0 0 2–4 3
FW004 – power station at
Newtown
0 0 0 1 0 4–6 4
FW005 – riverside properties
at Bridgeham
2 0 1 0 0 After 2
Table 3.4 Simple 2 x 2 contingency table for flood warning threshold evaluation
Flooding threshold exceeded
Yes No
Flood warning threshold exceeded Yes A B
No C D
In this hypothetical example, which is based on Fig. 3.4, the warning lead times
for FW001, FW003 and FW004 were satisfactory for all events, but were late for
one event at FW002, and below the target value for three events. Also, for FW005,
on two occasions the warning threshold was not reached until after flooding started
at Bridgeham. This might indicate the need to adjust the value for FW002, and pos-
sibly for a new approach for FW005.
An alternative way of examining performance is using a contingency table
approach as shown in Table 3.4 for the case of a river or tidal level gauge. The
example uses information on the crossing of flood warning and flooding thresholds
but, as described earlier, could also be extended to include an evaluation of the dis-
semination component of the system, based on information obtained from post
event surveys and incident logs (e.g. was a warning received? was your property
flooded?).
Based on this table, a number of parameters (categorical statistics) can be
defined, including the following three performance statistics which are widely used
in flood warning and forecasting verification studies:
● Probability of Detection (POD) = A/(A + C)
● False Alarm Ratio (FAR) = B/(A + B)
● Critical Success Index (CSI) = A/(A + B + C)
The CSI parameter is sometimes called a ‘Threat Score’. These statistics can be
accumulated across a number of flood events at a single site, or a number of sites,
and give an indication of overall performance, and are discussed further in Chapter 5.
Histograms or cumulative frequency plots can also be produced as a guide to the
thresholds which provide the best compromise between probability of detection and
the number of false alarms.
Similar techniques can also be used for rainfall threshold values; for example,
Table 3.5 shows a hypothetical analysis to assist with verification, or setting, of
rainfall thresholds for two raingauges, based on analysis of 10 years of historical
rainfall data.
The Flood Warning column indicates the number of flood warnings issued each
year at this frequent flooding location, which of course may not always indicate that
flooding actually occurred. For Raingauge 1, the analysis indicates that the False
Alarm Ratio is high for the shorter duration thresholds, but is of the order 50–75%
for the 12 hour duration, which may be acceptable for applications such as providing
an initial alert to duty officers of the need to start monitoring rainfall and river levels
more closely. For the newer gauge (Raingauge 2), false alarm rates are approximately
twice those of the other gauge, so the depth-duration thresholds could possibly be
adjusted, or perhaps the gauge is not representative of rainfall in the catchment.
Additional checks would also be required to confirm that the successful alarms are
linked to the same rainfall events which led to the flood warnings.
For evaluation of flood warning performance, it can also be useful to introduce
the concept of a ‘near miss’ (e.g. Environment Agency 2004b), in which levels are
within some defined tolerance of the threshold value. For example, if flood warn-
ings are being provided for an area behind a flood defence, then the tolerance might
be set equal to (or to some factor of) the design freeboard, on the basis that any level
within that tolerance is a cause for concern.
Some additional examples of approaches to verification are presented in Chapter
5 for the case of evaluation of flood forecasting model outputs, and in some cases
these methods might also be used for verification of warning thresholds. It is also
worth noting that many of the ideas used in flood warning verification have been
developed from other fields, in particular meteorology, for which the science of
Table 3.5 Example of a verification analysis for raingauge data
Year
Flood warning
issued
Raingauge 1 Raingauge 2
10 mm in
3 hours
15 mm in 6
hours
20 mm in
12 hours
10 mm in
3 hours
15 mm in
6 hours
20 mm in
12 hours
1998 1 5 5 2
1999 2 8 5 5
2000 2 8 5 3
2001 1 3 4 4
2002 1 6 5 2
2003 2 6 3 5
2004 1 6 5 3
2005 2 8 6 5 11 15 11
2006 1 6 2 2 10 6 4
2007 1 4 5 3 8 10 7
3.3 Performance Monitoring 69
70 3 Thresholds
forecast verification is perhaps better established (e.g. Stanski et al. 1989; Jolliffe
and Stephenson 2003).
One outcome of performance monitoring may be that a decision is taken either
that a more sophisticated approach is needed (e.g. a flood forecasting model, or
additional instrumentation), or that the level of service hoped for cannot be achieved
in practice with current budgets or technology. For reporting at organisational,
regional or national level, it may also be useful to aggregate performance statistics
across large numbers of Flood Warning Areas, which in turn can be used as a basis
for deciding on future investment and other requirements to improve performance.
Chapter 11 discusses some of these issues in terms of the economic performance of
an overall flood warning, forecasting and emergency response system.
Chapter 4Dissemination
Although organisational structures can differ widely between countries, a regional
or national flood warning service typically has a wide range of responsibilities,
which can include monitoring meteorological, river and coastal conditions, development
and operation of flood forecasting models, and dissemination of flood warnings to
the emergency services, local authorities and the public. Other responsibilities may
include operation of control structures to mitigate flooding impacts, assisting with
or coordinating the emergency response (evacuation, sandbags etc.), and contributing
to post event assessments. These various activities may be performed within an
overall framework of flood warning targets and performance monitoring, so that the
lessons learned from each flood event guide future investments and technological
improvements. This chapter discusses some of these organisational and procedural
aspects to providing a flood warning service, and gives an overview of techniques
for disseminating flood warnings and for implementing a flood warning system.
Later chapters describe how the flood warning service fits into the wider emergency
response to a flood event, which can potentially involve participants from many
different organisations.
4.1 Flood Warning Procedures
4.1.1 Introduction
Flood warning procedures define the actions that flood warning staff should take as
a flood event develops. Some reasons for establishing clearly defined procedures
include:
● During the pressure of a major flood event, there may be little time available for
analysis and discussion regarding whether to issue individual flood warnings
● Given that floods can occur any time of day or night, less experienced staff may
be on duty, and need clear guidance on the actions to take
● If procedures are not available, or do not cover all likely eventualities, vital
actions may be overlooked
K. Sene, Flood Warning, Forecasting and Emergency Response, 71
© Springer Science + Business Media B.V. 2008
72 4 Dissemination
● Increasingly, there a need for organisations to maintain an audit trail of actions
during a flood event, and to perform detailed performance analyses and post
event reviews after the event
The format of Flood Warning Procedures varies widely between organisations and
can range from short documents or charts through to detailed manuals and compu-
terised decision support systems. Procedures can cover just a single location,
through to large numbers of flood warning areas in a region, and Table 4.1 illus-
trates some of the topics which may be covered.
Table 4.1 Some typical items covered in flood warning procedures
Items Description
Detection Procedures for routine monitoring and telemetry of
meteorological, river and coastal conditions against pre-
defined alert criteria (thresholds), and for other forms of
monitoring such as CCTV or webcams and control gate
settings (Chapter 2)
Pre-warning activities Actions to take when monitoring a potential flood
incident (e.g. opening the incident room, calling in
additional staff)
Action or flood intelligence tables Summaries of the threshold values and other criteria
under which flood warnings should be issued and other
actions taken (Chapter 3)
Dissemination Details on who flood warnings should be provided to,
and by what means, including operating instructions
Flood forecasting The operation of flood forecasting models and interpre-
tation of the model outputs (Chapters 5–8)
Systems Guidance on operation of key systems, equipment and
other facilities, and backup plans in case of failure
Reporting Requirements for recording information on warnings issued,
actions taken, maintenance of communication logs, report-
ing etc., and external liaison with the public, media and other
organizations both during and following the event
Flood Event Recording Other actions which may be useful to perform during
and after a flood event if possible to help with post event
analysis and reporting (e.g. aerial photography, high
flow calibration of equipment, surveys of flood extent
and properties affected etc.)
Health and safety Guidance on safe working near water, and dealing with
emergencies
Contingency plans For failure of any key aspect of the flood warning system,
including the need to relocate the incident room if there is a
threat of flooding affecting access, escape or equipment
Contacts Names, addresses, phone numbers for key individuals
from various organisations, including representatives for
vulnerable people or communities at risk from flooding
(hospitals, care homes, the elderly etc.)
Different sets of procedures may also be available for different types of flood event;
for example, river flooding and coastal flooding.
Threshold values for individual sites are often a key component of a flood warning
system and are described in more detail in Chapter 3. They summarise the conditions
under which flooding may occur, and the meteorological, river or coastal conditions
for issuing warnings or for operational response (if applicable). Some examples of
operational response can include:
● Vehicle or foot patrols of flood defences at risk from breach or overtopping
● Initiation of sandbagging, raising of temporary defences and barriers etc.
● Visual observations of gauge board levels at rivers, reservoirs, coastal reaches
● Operation of sluice gates, tidal gates, and other flow control structures
● On site (visual) verification of closure of canal gates, flap gates etc.
● On site (visual) checks for blockages by debris at culverts, bridges and other
structures, and clearance of blockages (if possible)
● Operation of pumps, flow diversion structures etc.
Depending on the organisational structure, there may also be a requirement for staff
to coordinate the on site dissemination of warnings by loud hailer, door knocking,
portable sirens etc., although in some countries that task may be performed by local
authorities, the police, community representatives, or other groups.
Some additional information which may be included in procedures includes
photographs of the sites described, safe access routes under normal conditions and
for various flooding scenarios, descriptions of flooding mechanisms at specific
locations, and detailed instructions on the operation of structures. A single Flood
Warning Manual can cover many different sites, each with its own set of Action
Tables, and so can be a lengthy document.
In developing procedures, if resources are available, it is also useful to test them
regularly using table top or full scale response exercises. As described in Chapter 9,
a table top exercise attempts to mimic the decision making processes and pressures
which occur during a real flood event, and may make use of computer generated
visualisations, simulated television news reports, and other items to add to the realism.
The coordinator will introduce a range of scenarios and complications during the
course of the exercise following a timeline for the event.
4.1.2 Flood Warning Areas
A major task in developing flood warning procedures is often to define the districts
or properties for which warnings will be provided. Locations can be identified from
consultations, street maps, and site visits, making use of flood risk maps as well if
these are available (see Chapter 1). However, if a map based approach is used, then
an additional step is to convert these results into operationally useful units for pro-
viding flood warnings.
For example, the flood outlines derived from modelling studies may cut through
individual properties or groups of properties (e.g. an industrial site or hospital
4.1 Flood Warning Procedures 73
74 4 Dissemination
grounds), or indicate that access and escape routes to some properties might be cut
by flood water, even though the properties themselves are not at risk. Also, there
may be advantages in extending areas so that warnings go to identifiable groups of
people, or people outside the area which might be flooded, if they are in a position
to help those at risk (e.g. the elderly, hard of hearing etc.).
Issues like these often need to be discussed with other participants in the flood
warning process and with community representatives, and the flooding outlines may
need to be adjusted based on those discussions. Also, as described in Chapter 3, with
the increasing sophistication of warning dissemination techniques, another option is
to subdivide areas based on the probability of flooding, so that flood warnings are
progressively provided to more and more people as levels rise. This subdivision can
be both linear (along the river or coastal reach) and lateral (moving laterally away
from the river or coastline). Each sub area would then have its own set of threshold
values to allow warnings to be extended to more properties as river levels rise.
For example, Fig. 4.1 shows a simple example in which the flood risk outlines,
estimated from hydraulic modelling, are used as a guide to the development of a
Flood Warning Area for a hypothetical town called Newtown. The following three
zones or sub areas are established in the Flood Warning Area:
● Sub Area A – Riverside paths, sports ground and road at Newtown
● Sub Area B – Town Centre and Southside District at Newtown
● Sub Area C – Northside District at Newtown
In this example, the sub areas are extended in some places outside the main flood
risk zones to cover complete communities for which access may be affected, or for
which there may be a history of flooding not represented in the hydraulic model.
Also, for Sub Area B, rather than escalate warnings to a small number of properties
in the 1–2% risk band, all properties to the south of the river are combined into one
sub area. There are of course many issues to consider in defining the extent of warn-
ing areas, including whether property owners in areas not at direct risk would wel-
come information on potential flooding (or not), and national policy on this issue.
A possible extension of this method is to derive estimates of the probability of
flooding in real time, where the probabilistic component arises from the uncertain-
ties in observations and forecasts, and other unknowns. The resulting probabilistic
flood outlines can then be combined with consequence to give a measure of risk
Fig. 4.1 Example of defining the extent of a Flood Warning Area from flood risk mapping outputs
Road
Road
Town boundary
1 in 100 year (1%)flood outline
1 in 50 year (2%)flood outline
Sub Area B
Sub Area A
Sub Area C
(probability x consequence), opening the way to a risk-based approach to issuing
flood warnings. This is a new and developing area which is discussed further in
Chapters 5 and 10, and users with differing risk tolerances might wish to be warned
at different levels; for example, utility operators might require warnings at a low
probability so that staff can be mobilised and contingency planning started. By
contrast, some property owners might wish to avoid false alarms and so only
require warnings when the likelihood of flooding is better defined.
4.1.3 Organisational Issues
The organisation of a flood warning service varies widely between countries and,
depending on the scale of the overall system, duties might include some or all of
the following activities:
● Detection – design, installation and operation of rainfall, river level, reservoir,
tidal level, wind, wave, and other monitoring equipment (e.g. for snow cover,
soil moisture)
● Design – design of flood warning schemes, including contributing to decisions
on who should receive flood warnings, setting flood warning thresholds, decid-
ing how warnings should be disseminated, and under what circumstances
● Dissemination – monitoring measurements and forecasts against thresholds, and
issuing warnings following agreed procedures, and public awareness activities
● Operational – taking actions to mitigate flooding, such as patrols, channel clear-
ance, operation of river control structures, or installing temporary barriers
● Management – general management activities including defining staff rotas, pro-
curement, performance monitoring and reporting, research and development etc.
● Forecasting – development and operation of flood forecasting models to provide
estimates of river levels, river flows, tide levels, wave overtopping etc.
Of course, some of these tasks might be unnecessary for a small-scale community
based system, where the primary needs are for detection and the dissemination of
warnings. However, for a regional or national flood warning service, most of these
tasks will usually be necessary, although some might be shared with other organisa-
tions. For example Box 4.1 provides an introduction to the flood warning service
operated by the Environment Agency for England and Wales.
A common example of shared responsibilities is a separation between the mete-
orological service, which provides weather forecasts, and the organisation respon-
sible for operating flood forecasting models and issuing flood warnings (although
in some countries these functions are combined). Another example is a split in the
responsibilities for issuing warnings; for example, responsibility may only extend
to issuing warnings to other organisations such as local authorities or the police, or
may extend to issuing warnings directly to the public. Other approaches can also be
found, as in the United States for example, where, in addition to providing warnings
at a national scale, the National Weather Service works in partnership with many
smaller scale ALERT, IFLOWS and other local flood warning systems (NOAA/
National Weather Service 1997).
4.1 Flood Warning Procedures 75
76 4 Dissemination
Box 4.1 The flood warning service in England and Wales (Environment Agency)
Recent estimates suggest that in England and Wales approximately five mil-
lion people in two million properties are at risk from flooding from a 1% (1
in 100 years) event, including almost 400,000 businesses. More than 70% of
properties currently receive a flood warning service, with a target lead time
for warnings of at least 2 hours, where this is technically feasible.
The flood warning service in England and Wales is operated by the
Environment Agency, whose responsibilities include installation and opera-
tion of raingauges, river gauges, tide gauges, and other instrumentation, the
development and operation of river and coastal flood forecasting models, the
implementation of flood warning schemes, monitoring weather radar outputs,
and issuing flood warnings to local authorities, the emergency services and
the public. The Environment Agency also has many wider responsibilities;
for example, in flood defence, and water resources.
Flood warnings (Fig. 4.2) are issued by more than 20 local offices sup-
ported by flood forecasts provided by eight regional offices, and meteorolog-
ical forecasts (rainfall, wind, surge etc.) from the UK Meteorological Office.
Some limited real time information from other organisations, such as water
companies and canal operators, is also available and used in the flood fore-
casting and warning process.
Fig. 4.2 Flood Warning codes in England and Wales (Environment Agency, © Environment
Agency copyright and/or database right 2008. All rights reserved)
(continued)
4.1.4 Control Rooms
For a flood warning authority, monitoring of rainfall, river and coastal conditions is
usually performed from one or more control rooms. Typically these are equipped
with computers to monitor rainfall, river and tidal conditions, and the outputs from
forecasting models, together with telephone, fax, and automated communications
systems for disseminating warnings and liaison with other organisations.
Box 4.1 (continued)
Flood warnings are issued using an automated dissemination system
called Floodline Warnings Direct. This allows alerts to be issued by email,
Internet, text messages, fax, telephone and cell phone, and includes text-to-
speech conversion software for phone based methods. All messages are
available in a range of languages. The system has the capability to target
warnings to small groups of properties and individuals, and is supplemented
by a range of community-based methods which, depending on location, can
include sirens, flood wardens, loud hailers and door knocking. Television,
radio, and Internet based approaches are also used, including the Environment
Agency’s web based Floodline service (http://www.environment-agency.
gov.uk/subjects/flood/floodwarning/)
Flood Warning Areas are defined using a community based approach and,
in some high risk locations, are divided into zones so that the number of prop-
erties warned can be increased as the extent of flood inundation increases.
Individually worded messages may also be issued to key contacts in the emer-
gency services, local authorities, utilities, and communities. Public awareness
activities, media training, and community and inter-agency liaison all have a
high profile between and during flood events, with local approaches guided by
targets and procedures established within the national Flood Awareness
Campaign.
Several methods are used to provide early warning of potential flooding,
including heavy rainfall, daily rainfall, flash warning, and surge tide fore-
casts from the UK Meteorological Office, and catchment-based alarm values
set on weather radar and raingauge observations and forecasts. River and
coastal conditions are monitored against a range of pre-defined threshold
levels and, if these are exceeded, a warning will normally be issued. Flood
forecasting models assist throughout this process and, where the model out-
puts are known to be reliable, are formally integrated into flood warning pro-
cedures. Operational threshold levels are also used to initiate operational
response, such as establishing foot patrols in high risk areas, raising tempo-
rary defences (barriers, sandbags etc.), and operating river and coastal con-
trol structures to help to mitigate or avoid flooding.
4.1 Flood Warning Procedures 77
78 4 Dissemination
The control centre could just be a single room in an office, or a separate building
dedicated to providing flood warnings. The location should be outside any possible
area which may flood, or for which access may be impeded by flood waters, with
an alternate location available in case of problems. Some other equipment which
may be available includes:
● Maps – large scale maps on walls or on chart tables
● Mimic boards – large wall mounted displays showing flood risk areas, gauges,
and other features (e.g. control structures, gates)
● Whiteboards – wall mounted boards for drawing sketches etc., possibly includ-
ing electronic whiteboards to transmit images to other offices
● Media kit – equipment to assist staff with providing television and radio brief-
ings to the media
● Television/radio – to keep up to date with news reports and how the event is
being reported
● Briefing area – an area for staff briefings and for visitors from government, the
media etc. to catch up with and observe operations
● Hot desks – networked workstation areas for temporary visitors to work
Figure 4.3 shows one possible example for the layout for the incident room for a
small regional centre.
In the figure, the numbers of computers and telephones shown are illustrative
only and other devices such as printers and fax machines are not shown; also most
key systems are likely to have complete backups in case of failure. A separate loca-
tion for media briefings is also usually advisable (Holland 2007).
Wall Maps
Meteorology
Telemetry
IncidentManager
FloodForecasting
OperationsManager
Communications
Whiteboards
Dissemination
Meeting Area / Table
Warning Status
Fig. 4.3 Example layout for a Flood Incident Room
Since floods can occur at any time of day or night, a rota of key staff will usually
be established, equipped with laptop computers, cell phones, radios or pagers so that
they can easily be contacted when away from the office. The incident room may be
permanently staffed, or limitations may be placed on how far duty officers can travel
from the incident centre when on call. At the handover between shifts, a briefing may
be held for incoming staff on the current situation, and a package of key manuals,
equipment, situation reports and other information formally handed over.
During normal operations, when no flooding is occurring or anticipated, the
duty officer might only monitor weather, river or coastal conditions daily, or a
few times a day, and perhaps have other duties unrelated to flood warning. For a
full time flood warning service, routine day to day duties can include review and
improvements to existing flood warning schemes, issuing routine river and
coastal situation reports and bulletins, development of new flood warning
schemes, training, post event reporting, reporting against organisational or
national targets, planning and liaison with other flood response organisations
(local authorities, emergency services etc.), public awareness campaigns (news-
paper, television, radio, meetings, leaflets etc.), commissioning public satisfac-
tion surveys, installation of monitoring equipment, system improvements
(telemetry, forecasting, dissemination etc.) and other activities.
When flooding conditions appear possible, the frequency of monitoring (and fore-
casting model runs, if available) will typically be increased, and additional staff put
on standby or called in to the incident room. The number of staff required for a fully
operational incident room can be high, and can include representatives from local
authorities, the police, and other organisations. The organisational structure differs
considerably between countries, but could include a general incident manager, a
manager for operational staff deployed on site, monitoring and forecasting specialists,
communications experts, a press or public relations officer (or technical staff trained
in media relations), and other specialists from within and outside the flood warning
team with detailed knowledge of the catchments or coastal reaches at risk.
As described in Chapter 9, other organisations, such as the emergency services
and local authorities, may establish their own command and control centres. Ideally
one centre will be designated to lead the response, with representatives from all
other key services and functions present, with clearly documented procedures
describing the division of responsibilities between different organisations.
4.2 Dissemination Techniques
4.2.1 Introduction
Flood warnings may need to be issued to the public, emergency services, local
authorities and others with an interest in when and where flooding is likely to occur,
or who are involved in the emergency response.
4.2 Dissemination Techniques 79
80 4 Dissemination
Again, terminology differs, but dissemination techniques can broadly be
separated into indirect methods, community based methods, and direct methods,
with some alternative classification schemes including use of the terms General and
Specific (Emergency Management Australia 1999), individual, community and
broadcast (e.g. Andryszewski et al. 2005a, 2005b) and Pull or Proactive Mode, and
Push or Reactive Mode (Martini and De Roo 2007).
Direct methods provide a warning targeted at specific individuals or organisa-
tions, and have the advantage in some cases of confirmation that the recipient has
received the warning, which is particularly important for the representatives of
communities, local authorities, emergency services and other key responders.
Indirect warnings by contrast provide a more general warning and can potentially
reach large numbers of people, whilst community based methods fall in between
these two extremes. Some examples of these techniques include:
● Indirect – television, radio, Internet, teletext, telephone help line, RSS,
newspapers
● Community – sirens, fixed, mobile or helicopter loud hailers, megaphones and
public address systems, bells, storm cones, flags, cascade systems, motorcycles,
billboards/signs (electronic/manual), road barriers, flood wardens
● Direct – telephone, cell phone (voice, text), door knocking, fax, telex, pagers,
two way radio, email, leaflet drops
For some methods, the distinction between these approaches is blurred. For example,
sirens may be installed at one or more strategic locations to provide complete cov-
erage of an area, but can be operated either locally, or indirectly from a control
centre over a telemetry network. Similarly, some cell phone networks have the
capability both for direct communications, and to broadcast emergency messages
to all phones within range of the nearest network tower.
Where loud hailers or hand operated sirens are used, these are typically operated
either by people patrolling the streets on foot, or from vehicles, often following a
pre-planned, timed route covering all areas for which a warning is required. Fixed
installations may also be used in locations where there is a regular flood risk.
Cascade systems (or telephone trees) may also be used, in which contacts are ini-
tially with one small group of key people by telephone or in-person, who in turn
each warn a second tier of people, and so on until all intended recipients have
received the warning.
All methods have their own advantages and potential drawbacks, and many
organisations use at least two alternate approaches, both in case of failure of any
one method, and because research has shown that people are more likely to
respond if they receive information in varying ways and from more than one
source, including any existing informal networks (e.g. Parker 2003; Andryszewski
et al. 2005a, 2005b).
Table 4.2 illustrates some potential issues with approaches to issuing direct
warnings (although note that all of the methods shown can work well in many situ-
ations, and much depends on the institutional and cultural setting, and the resilience
built into the design of dissemination procedures).
4.2.2 Role of Information Technology
Developments in computer and communication technology in recent years have led
to a range of new approaches for issuing warnings which complement existing
techniques. They also provide the opportunity to issue direct or community warn-
ings to much larger numbers of people than has been feasible in the past using
manually based methods alone.
Table 4.2 Some issues to consider with a selection of direct warning methods
Issue Examples of methods affected
Transmission/telemetry network can be affected
by power failure
Telephone/cell phone, sirens (remotely
controlled), two way radio (if using
repeater stations), Internet
High workload for staff (control centre) Telephone/cell phone (if manually operated)
High workload for staff (on site) Loud hailer, door knocking, leaflet drops
High record keeping requirements between
flood events
Telephone/cell phone, internet (email)
Possibly a significant time delay between
issuing a warning, and all recipients
receiving the warning
Telephone/cell phone (if manually operated),
loud hailer, door knocking, leaflet drops
Recipient of the warning may not be the
decision maker (e.g. on evacuation)
Telephone/cell phone (automated messaging),
loud hailer, sirens, leaflet drops, internet
(email)
Relies on recipient having device with them,
and switched on
Cell phone, two way radio
Message will probably not be received by
recipients away from the property
(e.g. at work)
Telephone, loud hailer, door knocking, siren,
leaflet drop
Message is not voice or text based and relies
on recipients understanding the meaning
and that it applies to a flood event
Sirens, storm cones, flags
No direct confirmation that a warning has been
received and understood
Telephone/cell phone (automated messaging),
loud hailer, siren, bell ringing, leaflet drop,
others
Transient populations cannot easily be added to
the list of recipients (e.g. people in vehicles)
Telephone/cell phone (automated messaging),
internet (email)
Less effective in rural areas with widespread
properties
Loud hailer, siren, bell ringing, door knocking,
others
Reception may be affected by high ambient
noise (e.g. factories), high winds, window
insulation etc, and/or be an issue for the hard
of hearing
Loud hailer, siren, bell ringing
Dissemination may be affected by flood waters
interrupting access.
Possible health and safety issues for staff and
volunteers
Loud hailer, door knocking, leaflet drop
Loud hailer, door knocking, leaflet drop
4.2 Dissemination Techniques 81
82 4 Dissemination
Perhaps the most widely used indirect approach is the Internet, with information
on web addresses provided via television, radio, newspaper and other public aware-
ness campaigns. Table 4.3 lists examples of Internet based flood warning systems
from several countries.
Some typical functionality can include information on the time of issue of the
warning, text, graphical and map based representations of the areas at risk, contacts
for more information, advisory information on acting upon the warning, and search
facilities by location, river, town etc.
For example, Fig. 4.4 shows a display used by the Urban Drainage and Flood
Control District in Colorado, which combines information on river levels, likely
impacts, historical flood heights, and site specific issues, and provides links to
maps, additional information on rainfall, and a range of tabulated outputs.
Increasingly, multimedia dissemination systems are also being used to
allow warnings to be targeted more precisely at specific groups of people (see
Box 4.1, for example). Some typical functionality for this type of system
might include:
● Information stored on people and properties at risk from flooding, including
preferred methods of contact, and alternate contact methods
● Information stored on flood warning codes and the messages to use (fax, voice,
other)
● The facility to link flood warning codes to specific groups of people or proper-
ties, including map based displays and definition of groups
● Automated dialling of phone numbers and sending of emails etc., perhaps
with computer generation of text and voice messages, including multiple
languages
● Automated logging of warnings issued, including time of issue and time of
receipt, including call back facilities in case of no response
● Automated generation of summary statistics
Table 4.3 Examples of Internet based flood warning systems
Country Operator Name or link
Australia Bureau of Meteorology http://www.bom.gov.au/hydro/flood/
Bangladesh Flood Forecasting and Warning
Centre
http://www.ffwc.gov.bd/
Finland Finnish Environment Institute http://www.environment.fi/
France Ministry of Ecology and
Sustainable Development
http://www.vigicrues.ecologie.gouv.fr/
Germany Rhineland Palatinate Flood
Warning Centres
http://www.hochwasserzentralen.de/
Japan Japan Meteorological Agency http://www.jma.go.jp/en/warn/index.html
United Kingdom Environment Agency, SEPA Floodline
USA NOAA/National Weather
Service
http://www.nws.noaa.gov/
In some systems, summary information can also be generated for the overall num-
bers of people or properties warned, for statistics on the average time delays expe-
rienced between issuing and receiving warnings, and other performance measures,
such as the percentage of people who acknowledged messages. Systems may be
opt-in, with people identified as being at risk of flooding choosing to be included,
or opt-out, with the default being to include people unless they confirm otherwise.
A particular problem for any warning system is that of so-called transient
populations, such as road users, pedestrians on riverside or coastal paths, business
travellers, hikers, and people in campsites. Options for automated transmission of
location dependent warnings include using the traffic alert systems available with
digital radio, targeting voice and text messages to cell phones within a given range
of a transmitter, and remotely activated electronic warning signs to warn of potential
flooding of roads and river and coastal footpaths.
One other approach to dissemination, which is sometimes used if false alarms
are not a major issue, or floods develop very rapidly, is to link the detection and
dissemination components of the system directly, without human intervention.
For example, river level detectors might be linked to an alarm bell or siren to alert
gate operators to the need to take action. Some other examples include:
● A community based heavy rainfall warning system in Central America, in which
rainwater piped into a container is detected using three electrodes set at different
Fig. 4.4 Greyscale example of water level templet from the Information Services and Flood
Warning Program, Urban Drainage and Flood Control District, Denver, Colorado
4.2 Dissemination Techniques 83
84 4 Dissemination
depths, which trigger audible alarms, and send automated messages by cell
phone or landline to selected community representatives (Oi and Opavedi
2006).
● Pressure transducers installed at low level in road kerbs with a direct link to
nearby electronic road signs to warn road users of potential flooding, as used in
parts of Texas, for example. Also, automated road barriers in some parts of the
USA (e.g. canyons).
● An alert system in Nepal for Glacial Lake Outburst Floods which used the out-
puts from river level sensors to trigger warnings in turn from a series of sirens
further downstream, connected by radio telemetry.
However, the decision to use a fully automated approach will depend on a range of
factors, including system reliability, the consequences of failure, and tolerance to
false alarms, and most operational systems rely on expert inputs from duty officers
at some point in the warning process.
Recent developments have also considered how to provide warnings to rural or
dispersed communities, with particular emphasis on a low cost, sustainable
approach, as illustrated in Box 4.2.
More generally, research on flood warning technologies increasingly aims to
improve the targeting of warnings, allowing more effective use of staff and other
resources, and avoiding the unnecessary of evacuation of properties. Forecasting
models can play a useful role here; for example, as described in Chapters 5 and 10,
systems and models are now available which, when combined with property data-
bases and digital terrain models, are capable of mapping likely flood inundation
extents at each time step in a model run, and generating lists and maps of properties
likely to flood at specific times into the future. When coupled to automated dis-
semination systems (such as voice messaging systems), warnings can in principle
be issued to individual properties, together with estimates of the likely depth, start
time and duration of flooding. However, a cautionary note is that much relies on the
model accuracy, so manual intervention is still likely to be required at some point
in the process to provide a check on model outputs.
4.2.3 Warning Messages
The content and wording of warning messages again varies widely between coun-
tries and much research has been done on the most effective ways to issue warnings
to the public and non-specialists, and on how warnings are perceived (see Chapter
11 for examples).
Some general principles are to provide a clear and accurate description, in familiar
(non technical) language, and ideally contrasting the severity of the current situation
to recent events which people may remember or can relate to. For example, in some
countries, colour coded marker boards are used on river banks and buildings illustrat-
ing the flood levels likely to be reached for different stages of warning.
Box 4.2 Examples of international developments in dissemination
technologies
Recent years have seen the development of a number of techniques which
combine the latest communication technologies with a low cost, sustainable
approach. These methods are often aimed particularly at rural communities,
and include the option to broadcast warning messages for many types of natu-
ral hazards. Some examples include:
● RANET – the RANET (Radio and Internet Technologies for the
Communication of Hydro-Meteorological and Climate Information for
Rural Development) project is an international initiative supported by
National Meteorological and Hydrological Services, Non Governmental
Organisations, and others. It aims to make early warnings about natural
hazards and other climate and weather related information available to
rural populations and communities, and operates in Africa and parts of
Asia and the Pacific. Activities also include identification of appropriate
dissemination technologies, training, and capacity building. Information
is sent via uplink to satellites for broadcasting every hour and can be
received by computer, digital radio, and cell phone, with other techniques
under development. Additional information is also transmitted on topics
such as general health, agriculture and basic education by a range of
information providers (Sponberg 2006).
● Village knowledge centres (Tamil Nadu, India) – are focal points for
information on various issues of interest to rural and coastal communi-
ties, such as fish movements, wave heights, weather forecasts, health,
education, agriculture and other community activities. The centres pro-
vide access to computer, radio, telephone and Internet equipment and
include public address systems. Information kiosks are also widely avail-
able or planned in large numbers of villages in India (UNESCO 2006).
● The ISLAND project – was a collaborative exercise between organisa-
tions in Vietnam, Cambodia, Laos and several European countries to
explore communication needs for hazard-related information (floods,
pollution, epidemics, forest fire etc.), particularly in rural communi-
ties. Initial consultations showed that sustainability could be improved
if the scope was widened beyond crisis forecasting and management
to provide other information of daily interest to rural communities,
such as weather forecasts and market data (e.g. for crops and live-
stock), together with longer term information on hazards (e.g. how
long fields were likely to remain flooded). The project also explored
new ways of presenting information such as community electronic
billboards and multimedia information sent directly to cell phones
(Morel and Despres 2006).
4.2 Dissemination Techniques 85
86 4 Dissemination
The warning should also include the time of issue and the location and expected
time and duration of flooding, recommended actions, and the time for the next
update. Locations are better expressed in terms of places where people live or work
(communities etc.), rather than in terms of river or coastal reaches or monitoring
locations. Messages should also be from a single authority, or intermediaries (e.g.
community representatives) or, if not, following an agreed plan for different organi-
sations to issue different components of the message.
The following items show a suggested generic format for a warning bulletin
based on a comparative study of methods used in several European countries
(Martini and De Roo 2007):
● Header – a short title describing the event and/or its development
● Date and time – of the bulletin’s delivery and its time of validity
● Name – of the bulletin’s provider (the organisation)
● Core message – short and clear description of the current situation, and its fore-
casted development
● Data – observed and forecasted data; comparison with past and historical events;
flood warning level if available; time and level of the forecasted peak
● Uncertainty – level of forecasting uncertainty together with explanations
● Local/personal advice – where appropriate; feared impacts on public life (trans-
port, communication networks, …..) and advice to face them; refer to seasonal
activities if appropriate (holidays, sports events, ….)
● Information – about further broadcast/information technologies to ease tele-
phone service
● General permanent information – flood warning level scale if available; emer-
gency or other useful contact points for more information; links to other infor-
mation providing systems (Internet addresses, telephone numbers,…); where to
find general information about flood risk in the area
● Date and time – of the next information/forecast
Where lead times are sufficiently long, experience suggests that warnings are more
effective if a staged approach is used, allowing people to prepare for the possibility
of flooding before needing to take action. For example (World Meteorological
Organisation 2006), a generic set of warnings for natural hazards such as typhoons,
hurricanes and floods is:
● An advisory informs people within a designated area of probable weather or
hydrological conditions that could lead to hazardous situations, but they are not
yet severe enough to move to the next stage of alert. People should take note of
an advisory and be aware of any change in conditions.
● A watch alerts the public of the possibility of a particular hazard and provides
as much information as is available on its intensity and direction. Such forecasts
are issued well in advance of a weather event such as a cyclone, when conditions
are suitable for development of severe conditions. When a watch is announced,
people should take steps to prepare to protect their lives and property. Depending
on the circumstances they may need to prepare for evacuation.
● A warning is a forecast of a particular hazard or imminent danger issued when
extreme conditions have developed and are occurring or have been detected. It
is time to take appropriate action.
In some situations, the initial stages of warnings may only go to emergency services
and governmental organizations. Different countries use different approaches; for
example, Box 4.1 illustrates the four stage warning system used in the UK which
is defined as follows:
● Flood Watch – flooding of low lying land and roads is expected.
● Flood Warning – flooding of homes and businesses is expected. Act now!
● Severe Flood Warning – act now! Severe flooding is expected with extreme dan-
ger to life and property.
● All Clear – no further flooding is expected. Water levels will start to go down.
Some conditions which might lead to issuing a Severe Flood Warning include high
risk to life (due to depth, velocity etc.), large numbers of properties likely to be
flooded, severe disruption to infrastructure and the ability of responders to act, and
the risk of flood defences failing or overtopping.
More generally, there is an increasing trend to provide estimates of uncertainty
with flood warnings and forecasts, and Chapters 5, 10 and 11 discuss this topic,
together with a more detailed discussion on how effective warnings can contribute
to reducing risks during a flood event.
4.3 Design and Implementation
The design of a flood warning system can require many stages, including consulta-
tion with community representatives, local authorities and the emergency services,
installation of monitoring equipment, development of forecasting and dissemina-
tion systems, writing procedures, and a range of management, training and other
activities. Several guides have been published on the main steps involved both for
flood warning systems, and other types of early warning system, and some exam-
ples are summarised in Table 4.4.
Some typical areas to consider in designing and implementing a flood warning
system can include:
● User requirements – from consultations, consideration of flood warning per-
formance targets (if any), and other criteria
● Risk assessment – formal assessments of the locations at risk from flooding
based on historical flood events, modelling and consultations
● Detection – assessment of the availability, quality and reliability of existing real
time data on rainfall, rivers, tides etc. (as appropriate), and installation of new
sites if required
● Thresholds – definition of the criteria under which warnings will be issued
● Flood forecasting – review of any existing flood forecasting models and devel-
opment of new models (if required)
4.3 Design and Implementation 87
88 4 Dissemination
● Dissemination – decisions on the most appropriate methods to use, preparation
of appropriate messages, implementation of databases etc. (as required)
● Procedures – development of flood warning procedures
● Preparedness – liaison with other organisations involved in flood response, and
development of flood emergency plans (Chapter 9)
● Resilience – assessment of all aspects of the proposed system for possible points
of failure, particularly during extreme weather and flood events (Chapters 2, 3,
5, 9, 11)
● Communication – design and implementation of public awareness campaigns
(Chapter 11)
● Performance Monitoring – developing procedures for ongoing monitoring and
evaluation of the scheme (Chapter 11)
An important first step is usually to decide on the main aims of the flood warning
scheme or system and any performance requirements or targets. Consultees can
include members of the public, or their representatives, the emergency services,
local authorities and others. Approaches to consultation can include workshops,
town meetings, site visits, household surveys, telephone surveys and question-
naires. Requirements may be expressed in terms of lead time, accuracy, ways of
receiving information, and other choices. This approach, particularly if it is driven by
members of the communities involved, also raises awareness of the proposed scheme
and builds a sense of ownership (Emergency Management Australia 1999).
Table 4.4 Examples of guidelines on designing and implementing various aspects of flood warning
and other early warning systems
Country Organisation Document(s) Reference
Australia Emergency
Management
Australia
Australian Emergency
Management Manuals;
Flood Warning volume
Emergency Management
Australia (1999)
England & Wales Environment Agency A series of best practice
guidelines on river,
estuary and coastal
flood forecasting
Tilford et al. (2007),
Environment Agency
(2002, 2004)
USA NOAA/NWS Automated local flood
warning systems hand-
book
NOAA/National
Weather Service
(1997) (see also
USACE 1996)
Generic World Meteorological
Organisation
Guide to hydrological
practices (forecasting
sections)
World Meteorological
Organisation (1994)
Generic World Meteorological
Organisation
Global guide to tropical
cyclone forecasting
Holland (2007)
Generic ISDR Developing early warning
systems: a checklist
ISDR (2006)
Another key question is the likely budget available, and the economic case for
the scheme, and this point is discussed in Chapter 11, together with a range of
techniques for prioritising investment in schemes (cost benefit, multi-criteria meth-
ods etc.). Some aspects of the design may also be guided by organisational or
national targets and standards, again as discussed in Chapter 11. Later chapters also
discuss the design of flood forecasting systems (Chapters 5–8), techniques for
examining resilience (Chapter 9), and performance monitoring (Chapter 11).
4.3 Design and Implementation 89
Part IIFlood Forecasting
Chapter 5General Principles
Flood forecasting models are an important component in many flood warning and
emergency response systems. Models can assist by providing advance warning of
the likely timing and magnitude of flooding, and in helping to understand the complexities
of a flood event as it develops. Models outputs may also be used in decision support
systems for flood event management and the operation of flow control structures.
The techniques used for flood forecasting have many similarities to the methods
used for simulation modelling of river and coastal processes. However, the design
may be constrained by the availability of real time data, and computer systems on
which to operate the model, although there is the advantage that model outputs can
be updated to help to account for differences with observed values; a process which
is often called data assimilation. Ensemble and probabilistic techniques are also
increasingly being developed to provide information on model uncertainty to users
of model outputs, and to allow a more risk-based approach to decision making. This
chapter provides a general introduction to these issues and to the topic of flood
forecasting model calibration and performance monitoring.
5.1 Model Design Considerations
In designing a flood forecasting model, some important factors to consider include:
● The forecasting requirement
● The real time data available to support operation of the model
● The forecasting system on which the model will be operated
● The required model performance
● The time, budget and skills available for model implementation
The overall design will often be a compromise based on these various considera-
tions. Section 5.2 discusses forecasting systems, whilst the requirements for data
depend on the particular application, and examples are provided in Chapters 6–8 for
a range of forecasting applications. The issues of time, budget and skills, and the
level of flood risk, often form part of a wider decision making process on the eco-
nomic justification for a flood warning system, and are discussed in Chapter 11.
K. Sene, Flood Warning, Forecasting and Emergency Response, 93
© Springer Science + Business Media B.V. 2008
94 5 General Principles
The forecasting requirement depends on the needs of users of the model outputs,
and Table 5.1 lists some examples of applications which might be considered. For
a given situation, a model may help in addressing some or all of these requirements.
Although it is difficult to generalise, in moving down Table 5.1, the types of model
required will generally be more complex, and take longer to develop.
For example, for a river flow model, to estimate the flood peak at a single location,
a simple rainfall runoff model may in some cases be sufficient whilst, to estimate flow
depths and velocities at property and street locations, a real time hydrodynamic model
of the floodplain flow would ideally be required. Similarly, for coastal forecasting, to
estimate inundation extent behind a sea defence, offshore-nearshore wave
transformation, wave overtopping, surge and floodplain models might be required. In
particular, models to assist with control structure operations, and decision support
applications, can sometimes require considerable exploratory work to implement. Of
course, exceptions can always be found and some of the choices available are dis-
cussed in Chapters 6–8 when considering the strengths and limitations of alternative
modelling approaches for river and coastal forecasting applications.
Another aspect of the forecasting requirement is to specify where forecasts are
required. These locations are often known as Forecasting Points, and can include:
Table 5.1 Some typical applications of flood forecasting models (Adapted from Environment
Agency 2002; © Environment Agency copyright and/or database right 2008. All rights reserved)
Requirement Typical applications
To provide additional warning lead time Providing the emergency services and public
with advanced warning of the likely times for
the onset of flooding or the peak of the flood
event
To estimate peak levels or flows As above; also providing estimates for peak
values, perhaps also linking to maps of likely
flooding extent based on off-line simulation
modelling or past experience
To estimate flooding duration As above; also considering when to issue the
message that the flood risk has reduced or
passed
To model flooding depths, extent and,
possibly, velocities
As above, but providing real time updates to the
likely spatial extent of flooding, together with
depths and velocities at general or specific
locations (districts, roads, houses etc.)
To provide information to assist with
operation of control structures
Optimising response to mitigate flooding (e.g. at
dams, river control structures, tidal barriers
etc.) and possibly to reduce penalty payments
and opportunity losses
To provide information to assist with
event-specific factors
Providing advice on the implications of various
problems arising during an event; for exam-
ple, failure of a flood defence, blockages by
debris, pumps failing etc. Also exploring dif-
ferent options for response (e.g. controlled
diversion of flows)
● Locations where there is a flood risk
● Locations where real time information is available to evaluate or update model
outputs
● Locations where forecasts are required to assist with operations of structures
Some alternative names for Forecasting Points include Forecast Points, or Flood
Forecast Points or, for coastal applications, Coastal Cells or Units. One option availa-
ble for the development of a forecasting model is to focus the modelling effort on
achieving acceptable model performance at these locations, but to use a simpler
approach elsewhere provided that this does not affect performance at those points.
This approach can result in considerable savings in the cost and time for model devel-
opment and/or reductions in model run times. For example, for a hydrodynamic river
model, detailed survey data may only be required at and around the Forecasting
Points, rather than for the whole catchment whilst, for a nearshore hydrodynamic
coastal model, the grid resolution can be tailored to the areas which have the most
influence on levels, surge and wave action at the Forecasting Points.
The performance requirements for the model can also be defined in terms of the
requirements at Forecasting Points; for example, the required accuracy of forecasts
of peak river levels, or the lead time provided for surge forecasts. Section 5.4 dis-
cusses some other possible model performance and calibration criteria. However,
these requirements may not always be achievable with the data, models and budget
available, and this needs to be factored into the overall design, and potential users
of model outputs warned of any such limitations. Chapter 11 discusses a range of
approaches to prioritising the development of forecasting models and other compo-
nents in the flood warning process.
In many cases, the forecast lead time is one of the key design criteria, and may
dictate the overall design of the model. For example, for a river forecasting model,
if the catchment response time is less than the required lead time, then rainfall
forecasts will usually be required as inputs to the model. Additional tasks required
in this case could include an investigation of rainfall forecast accuracy (and whether
this is suitable for use with the model), establishing a real time data feed of those
forecasts to the model and (ideally) calibration of the model to a historical archive
of forecast values, if this is practicable.
Also, in a real time application, it is important to distinguish between the poten-
tial lead time provided by the forecasting model, and the likely lead time for flood
warnings. The warning lead time can be considerably shorter than the potential lead
time due to time delays in the system including:
● Polling time – the time delay between a parameter (e.g. rainfall) being measured,
and being available for transfer to the forecasting system
● Forecasting system run frequency – the time delay whilst waiting for the fore-
casting system to initiate the next model run
● Pre-processing time – the time taken to prepare and validate the data for input to the
model, and to perform any preparatory analyses (e.g. infilling of missing values)
● Model run time – the time taken to initialise model states (if required), run the
model(s) in the forecasting environment, including any probabilistic simulations,
and perform any real time updating required
5.1 Model Design Considerations 95
96 5 General Principles
● Post-processing time – the time taken to collate and process the model outputs
into reports, graphs, maps etc., and to raise alarms (if required)
● Decision time – the time taken for the forecaster to review the outputs and decide
whether to approve the forecast, and for the recipients of that forecast to decide
whether to issue a warning (including time for discussions etc.)
● Dissemination time – the time taken between a flood warning being issued, and
all of the intended recipients receiving it
Figure 5.1 illustrates these various potential time delays for the idealised case of a
rainfall runoff model providing flow forecasts for a single Forecasting Point follow-
ing a short duration, intense rainfall event (see USACE 1994; Environment Agency
2002; Carsell et al. 2004 for similar examples).
In the figure, the relative magnitudes of the various time delays are illustrative
only and are exaggerated in places; also the catchment response time is assumed to
apply from the mid-point of the rainfall event, although various other definitions are
used (for example, starting from the centroid of the rainfall event). However, for
this particular example, the warning lead time available would be considerably less
than the catchment response time or the forecast lead time. This lead time could be
extended by using rainfall forecasts although, as described in Chapters 2 and 6,
with the accuracy likely to decrease with increasing rainfall forecast lead time.
In practice, for a simple model like this, operating on a dedicated forecasting
system, the overall time delay due to these factors can be small compared to the
physical response times of the river or coastal reach. However, if many models
are being operated in parallel (e.g. as on a regional forecasting system), or if more
complex model types are being used (e.g. hydrodynamic models), or probabilistic
or ensemble techniques are being used, these time delays can be significant, and
may in some cases lead to the decision to use an alternative modelling approach.
For example, hydrodynamic models of flood defence systems developed for design
Floodingthresholdexceeded
RainfallEvent
Peak rainfalldata received
Pre-processingcompleted
Model Runcompleted
Post-processingcompleted
Decision toissue warning
Warningreceived byall property
owners
Warning lead time
Forecast lead time
Catchment response time
Floodpeak
reached
Fig. 5.1 Illustration of the time delays in issuing a warning for a single rainfall runoff model
studies can sometimes take many hours to run, unless they have been optimised for
real time use, in which case improvements in run time of one or two orders of
magnitude or more are sometimes possible (e.g. Chen et al. 2005).
Chapters 6–8 discuss some run time issues associated with various types of river
and coastal forecasting techniques, and Chapter 9 discusses some other sources of
time delays in the emergency response process. However, for the forecasting com-
ponent alone, some possible options for obtaining solutions more quickly include
(e.g. Chen et al. 2005; Environment Agency 2007):
● Model emulators – simpler models, such as transfer functions, which can emu-
late the behaviour of more complex models
● Restructuring of models – so that the more computationally intensive compo-
nents of the model can be run on demand, or at a less frequent time step (e.g. a
hydraulic model for a specific flood risk area)
● Filtering or clustering of ensembles – use of only a subset of values if ensemble
techniques are used (although with many issues to consider regarding the repre-
sentativeness of the sampling technique)
● Computer processing – using faster processors, or structuring the model so that
computing effort can be shared between more than one processor
● Model rationalisation – improvement of the underlying model to improve run
times, convergence and stability
The issue of whether the decision time can be eliminated by automatically issuing
warnings based on forecasts is an interesting question, and ties in with various topics
which are discussed in Chapters 4 and 11; for example, the uncertainty in the model
outputs, the time available for an emergency response (e.g. for flash flood forecast-
ing), performance targets, and tolerance to false alarms. Many (but not all) modellers
take the view that some human intervention and interpretation is essential in the over-
all process, where there is time to do this. Also, based on estimates of model uncer-
tainty, or ensemble forecasts, or experience, the forecaster may choose to wait for
additional model runs before deciding whether to issue the forecast, in the expectation
that by then the model uncertainty will have reduced (see Chapter 10).
5.2 Forecasting Systems
A flood forecasting system provides the operating environment within which flood
forecasting models can be operated, and is sometimes called the system environment.
Table 5.2 shows some of the key functionality which is typically available in modern
forecasting systems and Box 5.1 provides an example of an operational system.
The precise options available will depend on the system developer or vendor.
Systems typically run continuously all year around (24/7) and are required to meet
specified standards for availability, reliability and downtime. For additional resilience,
many systems offer multiple redundancy in computer hardware, software, and data
transfer routes in case of failure of any one component.
5.2 Forecasting Systems 97
98 5 General Principles
Table 5.2 Some typical functionality in modern flood forecasting systems
Item Function Description
Pre-processing Data gathering Polling of instruments directly, or receiving data
from a separate telemetry system (see data inter-
facing)
Data interfacing Interfacing to a range of real time data feeds and
forecast products from various sources (mete-
orological, river, coastal)
Data validation Real time validation using a range of time series,
statistical, spatial and other validation methods
Data transformation Transformation of input data into the values
required by the modelling system (e.g. catch-
ment rainfall estimates), including infilling miss-
ing values by interpolation and other methods
Model runs Model run control Scheduling and control of model runs, and error
handling
Data assimilation Application of real time updating and data assimila-
tion algorithms
Data hierarchy Automatic fall-back to alternative options in case of
failure of one or more components (models, data
inputs etc.)
Post processing Model outputs Processing of model outputs into reports, maps,
graphs, web-pages etc.
Inundation mapping Intersection of inundation extents (if computed)
with street and property maps etc. to generate
information on areas at risk at each time step
Alarm handling Raising alarms when thresholds are forecast to be
exceeded, using map based displays, email,
pager, text messaging etc.
Performance
monitoring
Automated calculation and reporting of information
on model performance and system availability
Audit trail Maintenance of a record of data inputs, model run
control settings, model forecast outputs, operator
identities etc.
Replay The facility to replay model runs for post event
analysis, operator training and emergency
response exercises
User interface Model outputs Map based, graphical and other displays of input
data, forecast outputs, alarms etc., including
overlays of aerial and satellite photography
What if functionality For running scenarios defined during the design
phase or in real time (e.g. for future rainfall,
defence breaches, gate operations etc.)
System configuration Interactive tools for off-line configuration of
models, data inputs, output settings, alarms etc.
Model calibration Off-line tools for calibration of models
Although models can be operated without a forecasting system, this can rapidly
become complicated if there are multiple Forecasting Points to consider, or forecast
lead times are short, or real time updating is required, or forecasting duty officers
have other tasks to perform. Organisations are also increasingly required to provide
(continued)
5.2 Forecasting Systems 99
Box 5.1 The National Flood Forecasting System in England and Wales
The National Flood Forecasting System (NFFS) is the operational flood fore-
casting system used by the Environment Agency in England and Wales. The
system can run a wide range of river and coastal forecasting models, and pro-
vides forecasts to numerous Forecasting Points on rivers and coastal reaches
across England and Wales.
Operationally, the system gathers real time data from a wide range of sources,
including regional telemetry systems (rainfall, river, reservoir, tide data etc.), and
Met Office weather radar data and weather forecasts (rainfall, surge, wind speed
and direction etc.). An extensive range of data validation, aggregation and trans-
formation tools is available. The range of model types and options includes:
● Conceptual rainfall runoff models
● Transfer function rainfall runoff models
● Flow routing models
● One-dimensional hydrodynamic models
● Snowmelt models
● Reservoir models
● Coastal wave transformation models
● Specialised models related to structure operations
● Data assimilation tools
Hydrodynamic models are optimised for run times, stability and convergence
before loading onto the system, and several thousand kilometres of river net-
work is represented in this way, allowing complex backwater, confluence,
control structure and tidal effects to be modelled in real time. Real time inundation
mapping is also being evaluated for some high risk locations.
A range of simpler methods, such as correlations and look-up tables, is
also included as a backup, and sometimes as the primary model type where a
more expensive approach is not justified.
Models are typically operated at least daily to monitor for potential
flood risk. If flooding seems likely they are run on a more frequent basis
to keep forecasters up to date on when and where floods are expected. The
latest model outputs are normally available to inspect via the map based
user interface, and a variety of graphical and other reporting formats.
The system includes an extensive range of forecasting performance moni-
toring tools, including contingency table, statistical, and skill-score
approaches.
The system includes the facility to run ‘what if’ scenarios to explore the
effect of different rainfall forecasts, control structure operations, and event
specific factors (e.g. defence breaches), and includes data hierarchies in
case of instrument or telemetry failure. There is also the facility to raise
alarms when rainfall and level thresholds are exceeded but, for operational
reasons, these are normally raised on the regional telemetry systems.
100 5 General Principles
an audit trail of decisions made during a flood event, including use of the outputs
from forecasting model runs, and a forecasting system can help to provide this
functionality. Automation of model runs can also free up skilled staff to spend more
time on interpretation and discussion of data and model outputs, rather than routine
data processing and analysis. Some possible exceptions are situations where only a
small number of models needs to be considered, the data entry requirements are
modest, or model runs are only required at irregular intervals (for example, for
some types of coastal or groundwater flooding).
Figure 5.2 illustrates a typical configuration for a forecasting system operating
both catchment and coastal flood forecasting models.
In this example, real time data flows are received from a network of raingauges,
river gauging stations, automatic weather stations, weather radars, and tide gauges.
Rainfall and surge forecasts and composite weather radar data are also received from
a meteorological service or department. Satellite, reservoir and snowcover/snowmelt
information might also be included, although this is not shown on the figure. The
data inputs are handled by a separate telemetry system, which feeds data to the fore-
casting system and to an off-line hydrometric database system (not shown).
Two independent forecasting systems are shown; the operational (Duty) system
and a backup (Standby) system, which operates in parallel and can switch automati-
cally to being the live system in case of problems. Normally, the hardware for these
independent systems would be located at different sites, both out of the floodplain,
and ideally separated by a sufficient distance that both would not be adversely
affected at the same time by widespread catastrophic events (fire, flood, earthquake,
Box 5.1 (continued)
Ensemble and probabilistic flood forecasting techniques are actively under
development.
The software and database systems are hosted on two server clusters more
than 100 km apart to provide resilience in case of local problems (flooding,
power cuts etc.), with automated switching from duty to standby servers, or
parallel running at both locations. The system operates ‘around the clock’
without user intervention, and stand-alone versions are available to allow test-
ing and integration of new models, and user training. A web server also allows
outputs to be viewed. A novel feature of the system is that it is open source,
in the sense that any model complying with the open XML-based published
interface can be ‘plugged’ into an existing network of models, provided that
it is physically sensible to do so.
The development of NFFS (e.g. Werner and Van Djik 2005) was a major
undertaking and included migrating existing models from a range of older
legacy systems. The system became fully operational in 2006 and passed its
first major test in June and July 2007, when flood events affected more than
20% of the catchment area in England and Wales.
major power failure etc.). The telemetry system may also have backup facilities,
although this is not shown here.
Many other configurations can be used, of course; for example, the forecasting sys-
tem might also manage the polling (collection) of data, avoiding the need for a separate
telemetry system, although introducing the risk of interruptions in data collection if
there are problems with the forecasting system. Also, the forecast outputs might be fed
back to the telemetry system to allow both forecasts and observed data to be displayed
there, and for the alarm (threshold) handling on forecasts to be performed in parallel
with that for observed data, providing greater consistency and avoiding duplication of
functionality, although at the expense of some loss of system redundancy. Various other
combinations can also be envisaged, depending on the relative roles and responsibilities
of the various organisations involved in the data collection and forecasting process.
Internationally, many types of forecasting system have been developed for indi-
vidual forecasting services, or are available commercially, and the system which is
used can be a key factor in choice of modelling approach since usually only a spe-
cific range of model types can be configured on a particular system. There may also
be model run time, licencing and other issues to consider. Increasingly, though,
forecasting systems use a toolkit approach, with several choices of model, which
can be configured as appropriate for each modelling problem, perhaps using an
openly published interface which allows any type of model to be used which con-
forms to this interface (e.g. Fortune 2006). Table 5.3 presents several examples of
NumericalWeather
Prediction
Pre-Processor
Post-Processor
Alarm Handling
Model RunControl &Database
Telemetry System
Pre-Processor
Post-Processor
Alarm Handling
Model RunControl &Database
Flood WarningDisseminationSystem
Weather StationsWeather Radar Tide GaugesRainfall Forecasts
Composite Radar Rainfall
Surge Forecasts
ForecastingSystem(Duty)
ForecastingSystem
(Standby)
Raingauges River Gauges
Fig. 5.2 Illustration of a possible configuration for a flood forecasting system
5.2 Forecasting Systems 101
102 5 General Principles
systems operated at a national or regional scale, and further examples can be found
in Chapters 6–8 and in the references to those chapters.
Later sections and chapters describe the process of selecting and calibrating an
appropriate model, or network of models, for use on a forecasting system. Once
those stages have been completed, the next step is usually to configure the models
for real time use. The approach to configuration varies between systems, but might
include creating a database or hypertext file (e.g. XML) describing how models link
together, and how data flows through the chain of models. Alternatively, some sys-
tems offer the option of setting up the configuration using a graphical user interface
and Fig. 5.3 shows an example of how an interactive configuration editor might
appear for a simple catchment forecasting model.
The symbols used are for illustration only, but the principle of abstracting the
chain of models into a network through which data can flow, is common to many
systems. In particular, the representation in the system need not be to the same scale
and layout as the physical situation.
In this example, the evaporation function could consist simply of a typical seasonal
profile, or be estimated using real time data from an automatic weather station. The
flow routing component might consist of a single ‘black box’ model, or be represented
by a series of nodes at cross sections along the river reach (e.g. a hydrodynamic
model). In the latter case, the inflow components could be connected to the appropriate
nodes so as to better represent the timing and attenuation of the hydrograph at locations
further downstream, with additonal components to represent lateral (ungauged)
Table 5.3 Examples of national or regional scale flood forecasting systems
Country or region Abbreviation System Reference
Australia AIFS Australian Integrated Forecast
System - flood forecasting
component
Elliott et al. (2005)
Bangladesh National Flood Forecasting and
Response System
Paudyal (2002) http://
www.ffwc.gov.bd/
Central America CAFFG Central America Flash Flood
Guidance system
Sperfslage et al. (2005)
China NISFCDR National Information System
for Flood Control and
Drought Relief
Huaimin (2005)
Finland WSFS Watershed Simulation and
Forecasting System
Vehviläinen et al.
(2005)
Norway National Flood Forecasting
Service
Røhr and Husebye
(2005)
Del Plata river basin Del Plàta Basin Hydrological
Warning System (Argentina,
Bolivia, Brazil, Paraguay,
Uruguay)
Goniadzki (2006)
United Kingdom NFFS/FEWS National Flood Forecasting
System, FEWS Scotland
Werner and van Djik
(2005), Cranston
et al. (2007)
USA AHPS Advanced Hydrological
Prediction Service
http://www.nws.noaa.
gov/oh/ahps/
inflows. The river gauging station locations might also be selected as real time updat-
ing points, with the choice of updating procedure selected via the user interface.
In addition to the appearance of graphs, maps, tables and other forms of output, some
key considerations in configuring a model onto a forecasting system typically include:
● The choice of time step (or model run frequency) – this is unlikely to be less than
the polling or transfer intervals for real time data or rainfall and surge forecasts,
but can be multiples of those values. Ideally, a value will be chosen that is suffi-
cient to resolve the details of any event, particularly for the rising limb of a
hydrograph, as flooding thresholds are approached.
● The approach to model initialisation – how will rainfall runoff, flow routing,
hydrodynamic, reservoir, coastal and other models (as required) be initialised for
routine operation and when starting after a gap or interruption in operations, and
what choices will be made for saving model states between runs (if required)?
● Integration of the system into operational procedures – how will the forecast
outputs be used in the process of issuing flood warnings, and will the system be
operated continuously, or on demand as a flood event develops? Are there cer-
tain times of the day at which model outputs will be required for inspection, and
are there benefits in operating the model less frequently (e.g. once per day) when
flooding is unlikely and, if so, what is the process for switching from normal
operation to a raised state of alert (e.g. more frequent model runs)?
● Data storage requirements – data volumes can increase rapidly when both input
data and forecast runs are archived, particularly if a probabilistic or ensemble
approach is being used. Recent values may need to be kept in a ‘rolling barrel’ store
which after each run stores the oldest values off-line, then overwrites them with the
latest values.
Many systems also allow a data hierarchy to be defined in case of failure of one or
more input data streams, such as an instrument or a telemetry link, or the overall
telemetry system. A hierarchy or choice of models can also be defined to run in
River Gauging Station
Town
A
C
D
Raingauge
1
2
Evaporation Function
Flow Routing Model
Rainfall Runoff Model
AC
B
1
2
D
B
Fig. 5.3 Example of model configuration in a flood forecasting system
5.2 Forecasting Systems 103
104 5 General Principles
parallel on some systems; for example, if a hydrodynamic model run fails to converge,
then the outputs from a simpler backup hydrological routing or correlation model
might be used instead. One common example of a data hierarchy is for raingauges
and, as a simple illustration, the following set of rules might be one possibility for
the rainfall runoff models shown in Fig. 5.3:
● If both raingauges are operating, use raingauges 1 and 2.
● If raingauge 1 fails, use raingauge 2.
● If raingauge 2 fails, use raingauge 1.
● If raingauges 1 and 2 fail, use weather radar data (if available).
● If the weather radar data feed fails, use a standard storm profile.
This sequence could be configured to run automatically on the system, with alerts
provided to the user that replacement data streams have been used in model runs.
Also, to account for possible calibration and other differences in each data
stream, ideally the model should be calibrated and optimised for each combina-
tion of inputs, with a hierarchy of parameter sets also available, although this may
not be practicable in many cases. More complicated scenarios, using other more
distant raingauges, and combinations of gauges, could also be envisaged.
A related option provided by some systems is the functionality to perform what-if
scenarios during a flood event; for example, to explore the impact on flood magnitude,
timing and extent for situations such as operating a river control structure, or event-
specific problems occurring such as a bridge being blocked by debris, or a breach
occurring in a flood defence. One commonly used example is a set of scenarios for
future rainfall and some options could include:
● The forecast rainfall follows a pre-defined profile.
● The forecast rainfall matches values from a significant historical flood event.
● The forecast rainfall continues at the current intensity.
● Rainfall stops.
Normally, a set of pre-defined scenarios will be calibrated and configured ready for
use during an event although, in some cases (e.g. gate operations, blockages), the
forecasting system may provide users with direct access to model parameters and
initial conditions via the user interface (e.g. dialog boxes) so that these can be
changed manually. Ensemble and probabilistic techniques might also be used as
described in Section 5.5.
5.3 Data Assimilation
Although there are many complicating factors in real time modelling, one advantage
compared to off-line simulations is that observations of river or coastal conditions
are usually available to compare with model outputs. If the observations are reliable,
then the forecasts can be updated to help to take account of the differences between
observed and forecast values.
Many different techniques have been developed for forecast updating, and these
are often separated into the following three main categories (e.g. Reed 1984; World
Meteorological Organisation 1994; Moore 1999), although the distinctions between
methods can sometimes be blurred:
● Error prediction – in which forecast outputs are adjusted directly, based on models
calibrated to the time series of differences between observed and forecast values
● State updating – in which the initial conditions of the model are adjusted to
achieve a better match between the observed and forecast values
● Parameter updating – in which the parameters in the model are adjusted to
achieve a better match between observed and forecast values
State updating is often called data assimilation by meteorologists and coastal mod-
ellers, whilst the terms ‘real time updating’ or ‘real time adaptation’ are often used
by hydrologists to describe all three approaches. Error prediction is sometimes also
called error correction, output updating or real time adjustment.
Although data assimilation can significantly improve the accuracy of flood fore-
casts, and is often recommended as best practice, a few issues to consider include
(e.g. Environment Agency 2002):
● Updating does not remove the need to have a well calibrated model, able to rep-
resent response for a wide range of types of event. In particular, some forms of
updating algorithm can struggle with correcting errors in the timing of peaks.
● The quality of the updated forecast will depend on the quality of the input data,
and erroneous data can degrade, rather than improve, the accuracy of forecasts.
Usually, it is advisable to validate data inputs either automatically or manually
before they are used for data assimilation.
● For real time control applications, the use of updating needs to be factored into
the system design from the start, since otherwise unwanted feedback effects can
develop; for example, control gates ‘hunting’ for optimum settings.
If data quality remains an issue after validation, then one solution might be to
restrict updating to ranges in which values are known to be reliable; for example, if
the high flow end of a stage discharge relationship is suspect, or levels start to
become insensitive to flows (e.g. for some floodplain flows), then updating could
be performed only up to a certain threshold value.
At the simplest level, one approach to updating a forecast is to inspect the
observed and forecast values on a graph, and to adjust the forecast ‘by eye’ to com-
pensate for any differences. Figure 5.4 illustrates this process for a river flow
hydrograph.
In this example, the forecast is below the observed values throughout the period
up to the start of the forecast period and, for the earlier event, the forecast peak was
later than the observed peak. There are also minor differences in hydrograph shape
or volume. The adjusted forecast attempts to visually compensate for these errors
by applying both timing and magnitude corrections.
The human eye is remarkably good at distinguishing between errors in timing
and magnitude, and in deciding on the appropriate adjustments to make. However,
5.3 Data Assimilation 105
106 5 General Principles
although systems have been developed which allow a forecaster to make adjustments
of this type by editing or blending graphical outputs on a computer display, this
approach can be impractical to apply if there are many Forecasting Points to
consider, or there are many other demands on a forecaster’s time. In practice, therefore,
the majority of updating procedures operate automatically without user intervention,
although the original and updated forecasts are often displayed together, with the
option to accept or reject the adjusted forecast.
5.3.1 Error Prediction
One distinguishing feature of error prediction methods is that they are usually inde-
pendent from the forecasting model, and can be applied as part of the post-processing
of model outputs. State and parameter updating techniques, by contrast, tend to be
specific to the type or ‘brand’ of model, although with some exceptions.
Error prediction methods to some extent mimic the visual adjustments described
earlier, although may use a sophisticated range of time series analysis techniques. The
basis for the approach is that, although the sequence of errors (residuals) is unknown
before the start of the event, the natural persistence in catchment and coastal processes
will tend to cause the outputs to be consistently higher or lower than observed values
for at least part of the event, with timing errors also showing persistence over time.
A time series model can then be fitted either to historical datasets, resulting in a
fixed set of parameter values, or can be dynamically fitted during an event as it
progresses. The forecast values from ‘time now’ are then adjusted based on the
output from the time series model. The effects of these adjustments vary between
approaches but tend to force the forecast values to match observed values at ‘time
Time
‘Time Now’
Observed Flow
Flow
Original Forecast
Adjusted Forecast
Fig. 5.4 Illustration of a visual approach to forecast updating
now’, with the magnitude of the adjustments decaying into the future as the infor-
mation content of the observed values reduces.
Some examples of automated approaches to error prediction include autore-
gressive techniques (e.g. AR, ARMA), transfer function, and neural network tech-
niques (e.g. Reed 1984; Rungo et al. 1989; Serban and Askew 1991; Moore 1999;
Goswami et al. 2005).
5.3.2 State and Parameter Updating
State updating techniques aim to adjust the initial conditions of the model to obtain
a better match between observed and forecast values at the start of the forecast
period (‘time now’), and possibly over a number of time steps in the period leading
up to that point (e.g. Wohling et al. 2006). The approaches used vary between dif-
ferent categories of model and can be very model specific.
For conceptual models (see Chapter 6), state updating typically involves updating
the internal stores in the model (for a rainfall runoff model) or, in the case of a reser-
voir, the initial level for the simulation. For example, the adjustment may be made
using a gain factor which redistributes water between selected stores in the model
(e.g. Moore 1999). By contrast, process-based models such as hydrodynamic river
models (see Chapter 6) and coastal surge models (see Chapter 7) usually represent
the catchment or coastal response using a grid-based approach. In this case, the
updating problem is to distribute the forecast errors across the whole domain of grid
nodes based on what is usually a relatively small number of observation points (river
gauges, tide gauges etc.), whilst preserving conservation of mass and momentum,
and without producing unwanted transient effects from the input of potentially large
corrections at the grid squares closest to the points of observation.
For coastal models, many different approaches have been considered, including
empirical, sequential and variational methods, sometimes involving minimising of a
cost function which depends on functions of the differences between modelled and
observed values. A major international initiative (the Global Ocean Data Assimilation
Experiment: GODAE) is also exploring new and improved techniques for data
assimilation in ocean forecasting models (http://www.godae.org/), including the use
of satellite altimetry to widen the extent of data available for assimilation.
Perhaps the simplest state updating technique of all is simply substitution of
observed values for forecast values; a technique which has been used for both river
and coastal models with some success, although at the risk of introducing oscilla-
tions and other unwanted behaviour into the model forecasts. Techniques which
adjust the input data to the model, such as inflows or rainfall values (e.g. Serban
and Askew 1991) are also sometimes considered as a form of state updating.
For example, for hydrodynamic models of rivers, one approach is to distribute the
error into the tributary and main river inflow components (e.g. Rungo et al. 1989).
For the future, low cost sensor networks, of the type described in Chapter 2, may
also provide the possibility of using many more updating points than is possible at
present, due to their small (unobtrusive) size and low cost of installation.
5.3 Data Assimilation 107
108 5 General Principles
By contrast, parameter updating techniques seek to update the model parameters,
rather than the state variables. For process-based models, and some types of
conceptual models, parameter updating is little used, and might at first sight seem
inappropriate, since the basis of these modelling approaches is to use parameter
values which have some physical meaning. However, due to measurement, calibra-
tion and other uncertainties in the modelling process, model parameters can never
be known precisely, and parameter updating therefore seeks to compensate for this
lack of knowledge by adjusting parameters within defined bounds. An example of
use of this approach is in adjusting the roughness coefficient which is often used to
parameterise friction losses in a river channel in hydrodynamic models.
5.3.3 Other Techniques
For data-based models, such as transfer functions (see Chapters 6 and 7), the distinction
between approaches is perhaps less clear cut, and methods which have been used
for data assimilation include Recursive Least Squares, Adaptive Gain and Instrumental
Variable techniques, and various forms of Kalman filter, including extended
Kalman filter and ensemble Kalman filter approaches (e.g. Young and Tomlin
2000; Beven 2001; Romanowicz et al. 2006).
Kalman filter techniques have also been applied to other types of model; for
example, to operational coastal surge forecasting models in the Netherlands
(Verlaan et al. 2005), and to both rainfall runoff and hydrodynamic modelling
applications in the form of the ensemble Kalman filter (e.g. El Serafy and Mynett
2004; Weerts and El Serafy 2005; Butts et al. 2005).
5.4 Model Calibration and Performance
5.4.1 Basic Concepts
The principles of model calibration for catchment and coastal models are well
established and are discussed in many books and technical papers (see Chapters 6
and 7 for examples). However, for flood forecasting applications, some additional
factors need to be considered, and these are briefly described in this section. These
include the overall structure of the model (if the model configuration can be varied),
and the optimisation criteria to be used in the calibration. Once a model is opera-
tional, the performance will also need to be monitored, both to advise users of the
likely accuracy, and to guide future improvements to the model and of the real time
sources of data used in its operation.
As with off-line models, it is useful to know how well the model can represent
the timing and magnitude of peak levels, flows, surge, wave, wind and other out-
puts, and the time history of values during the event (e.g. the shape of a river flow
hydrograph). However if the model is used in a flood warning application, then the
performance in the time leading up to the crossing of flood warning threshold levels
may also be of interest, together with other criteria, such as the success rate of issu-
ing warnings, and the number of false alarms.
A related issue is that, due to model errors, some forecasts may not reach critical
threshold values, although the flows which are subsequently observed pass those
thresholds. Figure 5.5 shows three different examples of forecasts for a single hypo-
thetical event (for example, from three different modelling approaches).
In this example, Forecast A correctly indicates that the flooding threshold will
be exceeded, although at a slightly later time than actually occurred, whilst
Forecasts B and C do not. Also, Forecasts A and B indicate that the flood warning
threshold will be exceeded, but Forecast C does not. Another scenario which can
occur, of course, is that the actual flows do not reach critical thresholds, but the
forecast predicts that those thresholds will be exceeded, in which case a false alarm
would be triggered by the forecast output.
These examples consider the case of single forecasts, issued at a given time (‘time
now’). The start time of a forecast is often called the Forecast Origin, whilst the
maximum lead time which the forecast can reliably provide is called the forecast lead
time or forecast horizon. For river flow forecasting, the maximum useful forecast
lead time is of the order of the catchment response time, if observed rainfall values
are being used, but can be extended by using forecasts of catchment rainfall. The
model performance at each lead time can be assessed, leading to the concept of fixed
lead time forecasts, as illustrated in Fig. 5.6 for the case of a river flow forecast.
The figure shows seven forecasts at successive time intervals, with the Forecast
Origins shown as circles. In this example, the Forecast Origins coincide with the
actual flow values (as might be the case for some forms of real time updating),
although this need not necessarily be the case. As an example, the figure also shows
the values three time steps ahead for each forecast which, for hourly model runs,
Time
Level
Flood Warning Threshold
A
Flows subsequently observed
B
C
Flooding Threshold
Fig. 5.5 Illustration of forecasting issues related to threshold levels
5.4 Model Calibration and Performance 109
110 5 General Principles
would be called the 3-hour ahead forecasts. These values can be joined by a smooth
(interpolated) curve to create the so-called 3-hour fixed lead time forecast. In this
example, at this lead time, the forecast flows are both later, and lower than, the values
which were subsequently observed (and which are shown by the dashed line).
Fixed lead time forecasts can be constructed for a range of lead times, and calibration
criteria and performance measures can be developed which make use of these
values. However, if data assimilation techniques are used, then the values obtained
may be significantly improved at shorter lead times, so any performance values
quoted should note whether or not updating was in operation.
An additional consideration is whether a probabilistic or ensemble approach is
used, since additional calibration and performance measures may be required as
discussed later.
5.4.2 Model Calibration
The objective in model calibration is to develop a model which best meets the
requirements of the modelling study. In flood forecasting applications, examples
might include the forecast lead time requirement, and the ability to estimate the
timing and magnitude of flood peaks, the times of crossing of thresholds, or levels
at a river control structure. Chapters 6 and 7 give some examples of how the require-
ment can influence the design of river and coastal flood forecasting models.
The approach to calibration depends partly on the choice of model, and can
include calibration in off-line/simulation mode, assuming perfect foresight of input
time series data (e.g. rainfall), or calibration in real time mode, only making use of
data up to the time at which each evaluation is performed.
In simulation mode, the model is optimised against a number of historical flood
events, possibly also including values for the intervening periods to assess the full
range of performance. In real time mode, the focus of calibration is typically on the
performance of fixed lead time forecasts.
Time
Flow
Fig. 5.6 Illustration of Fixed Lead Time forecasts (circles are fixed origins, triangles are 3-hour
ahead forecasts)
Simulation mode is widely used for calibrating process-based and conceptual
models, with any real time updating component calibrated at a later stage. Data-
based models, by contrast, are often calibrated in real time mode, with the data
assimilation aspects sometimes integral to the overall model formulation. Also, as
discussed in Chapters 6 and 7, many data-based models are event based, in the sense
that they are calibrated only to specific flood events whilst, for some process-based
and conceptual models, it is essential to calibrate over much longer periods of data.
Flood forecasting models can also consist of networks of interconnected models;
for example, integrated catchment models consisting of rainfall runoff, flow routing
and other model types (e.g. reservoir models), or coastal models comprising
offshore, nearshore and foreshore components. For each model in the network, the
steps in model calibration can include:
● Model identification – choice of an appropriate structure for the model (if there
is a choice)
● Choice of calibration criteria – decisions on which criteria to use in model cali-
bration, and the relative importance of each choice
● Model optimisation – optimisation using best fit, trial and error and other approaches
● Model validation – tests of model performance using additional datasets, not
used in the original calibration
Model Identification techniques can include trial and error for different configura-
tions of model (e.g. alternative choices of stores in a conceptual rainfall runoff
model), or automated searches of a wide range of configurations (e.g. for some
types of transfer function model).
There are many approaches to model calibration and validation, including auto-
mated optimisation techniques, such as hill climbing, genetic algorithm, Monte
Carlo, and simulated annealing approaches (e.g. Beven 2001; Anderson and Bates
2001). Optimisation criteria can include the timing and magnitude of peak values,
measures of the overall shape of the hydrograph (bias, mean absolute error, root
mean square error, Nash Sutcliffe efficiency etc.), and threshold crossing measures
(e.g. the mean timing error in threshold crossings). Multi objective or multicriteria
techniques can also be used. For some of the more physically based model types,
some model parameters may also be fixed, or restricted to certain ranges, depending
on catchment, river or coastal characteristics.
Values can also be calculated on a fixed lead time basis, giving an indication of
how model performance changes with increasing lead time. For example, plots can
be produced of how mean square error, or Nash Sutcliffe Efficiency, decreases with
increasing lead time, and a cut-off value identified beyond which performance
drops below an acceptable level.
Additional threshold based criteria can also be defined using a contingency table
approach, as illustrated in Table 5.4.
Based on this table, the following parameters can be defined:
● Probability of Detection (POD) = A/(A + C)
● False Alarm Ratio (FAR) = B/(A + B)
● Critical Success Index (CSI) = A/(A + B + C)
5.4 Model Calibration and Performance 111
112 5 General Principles
Each type of criterion has its own strengths and limitations; for example, measures
based on peak values obviously overlook many other aspects of model performance,
whilst some whole hydrograph measures (e.g. mean square error) can be sensitive to
small timing errors or outliers, and may not consider how performance varies with
magnitude (unless they are reported only above certain thresholds), or the sign of
errors (e.g. bias). Threshold based measures are closely linked to the operational
flood warning requirement, but are based on individual threshold values, although
these methods can be extended to sample over a range of possible choices of thresh-
old values (e.g. Environment Agency 2004). Some trial and error may be required to
achieve a reasonable compromise across a number of flood events and criteria,
together with adoption of some best practice principles for model calibration for
flood forecasting, which include (e.g. Environment Agency 2002):
● Data validation – use validated, quality controlled data, particularly for values
during flood events (e.g. assessing the high flow performance of stage-discharge
relationships).
● Data sources – calibrate models to the same sources of data that will be used in
real time to avoid bias and other problems (e.g. if using radar rainfall data in real
time, calibrate to historical radar not raingauge data).
● Type of event – choose calibration datasets for events of the type(s) which the
model is required to represent (e.g. frontal events, thunderstorms, snowmelt).
● Data currency – use datasets representative of current conditions (e.g. since
flood defences were constructed, instruments installed, channels dredged etc.).
● Run frequency – set the model run frequency (if possible) to adequately resolve
the type of flood events being modelled, particularly in the time leading up to
flood warning thresholds.
● Data assimilation – use real time updating where the model type supports this,
and the data quality is good enough.
● Model initialisation – for types of models where this is important, focus on pro-
viding realistic initial conditions.
● Model validation – validate the model outputs against a number of flood events
not used in the original model calibration.
Also, for combinations of models, the performance of the overall network should be
assessed as well as for individual models within the network. As part of the model docu-
mentation, flooding mechanisms and other factors not represented in the model should
be highlighted to operational staff, together with the likely uncertainty in model outputs,
and the acceptable limits of model performance (maximum flows, lead times etc.).
Table 5.4 Simple 2 × 2 contingency table for flood forecasting model
evaluation
Threshold crossed (observed)
Yes No
Threshold crossed (forecast) Yes A B
No C D
5.4.3 Performance Measures
Once a flood forecasting model has been developed, and integrated into the opera-
tional forecasting environment, a period of pre-operational testing will often follow,
during which the performance and reliability of the model is assessed, before using
it in the flood warning process.
This monitoring will then continue as part of the routine assessment of the model,
and in particular to help to identify areas for improvements to the calibration and the
quality of the input data. Also, any sudden deterioration in performance can be identi-
fied; for example, due to changes in the accuracy of input data (e.g. stage-discharge
relationships), or in catchment or coastal characteristics (e.g. recently constructed
flood defences). Usually, model performance will need to be evaluated following
each major flood event, both for post-event reporting and to detect any event-related
problems which need to feed into the model development programme.
Modern flood forecasting systems often have the facility to automatically calculate
a wide range of model performance measures, including some of the real time meas-
ures described in the previous sections (e.g. contingency tables, and fixed lead time
performance). The types of methods used, and the scale of the assessment which is
possible will, in part, depend on the availability of observed data to use in the assess-
ment, and may be limited for some types of model (e.g. process-based models). Also,
some types of information which it would be desirable to measure in real time (e.g.
inundation extent and depths) are difficult to capture other than by post event survey,
and may not be practicable to assess for every event.
For assessment of individual flood events, many of the calibration criteria
discussed in the previous section can provide useful information on model perform-
ance; for example (Environment Agency 2002, 2004; Werner and Self 2005):
● Graphs of model outputs at fixed lead times, compared to the (subsequently)
observed values
● Tables summarising errors in the timing and magnitude of peaks, and event-based
measures of performance, such as the R2 coefficient or root mean square error
● Threshold based measures such as Probability of Detection, False Alarm Rate,
Critical Success Index, Over-Prediction Ratio, and timing errors in crossing of
thresholds expressed in absolute, mean square or probability of detection terms
● Lead time based measures such as the bias, average or median error in lead
times, or the distribution of errors, and the first forecast of threshold time
● Map based comparisons of inundation extent (if applicable)
The extent of validation will depend on the application, and in some cases several
measures will be evaluated to provide an overall view of the different aspects of model
performance, and, if relevant, may extend to analyses of performance with respect to
specific aspects of model performance; for example, closures at a flow control struc-
ture, or flow diversions to an off-line storage reservoir. Values can also be normalised
to facilitate comparisons of trend and variability between different locations. Also,
since real time updating can have a significant influence on performance, it is impor-
tant to indicate whether or not updating was used in the model runs (if applicable).
5.4 Model Calibration and Performance 113
114 5 General Principles
For evaluating long term performance over a number of events, or when considering
the performance of a number of models in a single event (e.g. for post event report-
ing), it is useful to consider methods for aggregating outputs to allow an overall
picture of performance to be obtained. In this situation, it can also be useful to
introduce some more operationally focussed performance measures; for example,
the success rate for issuing flood warnings, or for property owners acting upon
warnings received. However, measures of this type depend on factors beyond just
the model performance, and are discussed in more detail in Chapter 11.
Some examples of aggregation methods for evaluating forecasting model
performance include:
● Histograms of performance measures across a number of events
● Plots of performance measures against lead time
● Tabulated values of performance measures by river reach and frequency
● Maps of performance statistics to highlight spatial trends
Ideally, values will be presented in non-dimensional form to facilitate comparisons
between models, locations and events.
For example, for performance monitoring of the Storm Tide Forecasting
Service (STFS) operated by the Meteorological Office in the UK (see Chapter
7), some performance statistics which are used include contingency tables based
on crossing of alert levels, histograms of surge forecast lead times for alerts
raised relative to a threshold, and tables summarising the estimated mean, maxi-
mum, root mean square and standard deviation of errors (in metres) across a
number of events.
For evaluating the performance of probabilistic and ensemble forecasts, options
available for meteorological forecasts (e.g. Jolliffe and Stephenson 2003) include
the Brier Skill Score (including continuous versions), Ranked Probability Score,
Relative Operating Characteristic, Reliability and Sharpness, and these types of
measure can also be adapted and extended for flood forecasting applications (e.g.
Laio and Tamea 2007). The general aim is usually to assess aspects of the ensemble
such as how close the median is to the deterministic value, how well the probability
of an event occurring is represented (over the long term), and the confidence which
can be placed in the probabilistic estimates, and the value or utility of the forecast
(e.g. expressed in terms of cost-loss functions).
5.5 Model Uncertainty
As noted in Chapter 1, uncertainties in the flood forecasting process can arise from
many sources, and it is widely agreed that flood forecasts should be issued with an
indication of confidence or uncertainty (e.g. Krzysztofowicz 2001). Information on
uncertainty can also help with deciding where to focus effort on future model develop-
ment and data improvement programmes. Some sources of uncertainty in flood
forecasting models can include (e.g. Butts et al. 2005):
● Random or systematic errors in the model inputs (boundary or initial conditions)
● Random or systematic errors in the observed data used to measure simulation
accuracy
● Uncertainties due to sub-optimal parameter values
● Uncertainties due to incomplete or biased model structure
Table 5.5 provides examples of some additional sources of uncertainty in river and
coastal forecasting models.
Table 5.5 Some of the main sources of uncertainty in river flood forecasting models (Environment
Agency 2007; © Environment Agency copyright and/or database right 2008. All rights reserved)
Component Typical sources of uncertainty
River modelsCatchment averaging proce-
dures (for raingauges)
Representation of physical processes (topography,
elevation etc.)
Type of rainfall event (convective, frontal, orographic etc.)
Rain gauge density and distribution
Instrumental problems at one or more of the rain gauges used
Choice of model type
and structure
Lumped, semi-distributed, distributed rainfall inputs
Representation of catchment runoff processes
River channel and floodplain representation
Under/over parameterisation (parsimony)
Flood defence loading/fragility (if represented)
Gate operations
Representation of ungauged inflows
Representation of abstractions/discharges
Representation of groundwater influences
Model calibration Effectiveness of optimisation routines
Choice of optimisation criteria
Availability of sufficient high flow events for calibration
Skill of person calibrating the model
Operational Changes in catchment/channel characteristics since model was
calibrated
Use of different input data streams from those used in the
original model calibration (e.g. radar rainfall or forecasts
instead of raingauges)
Events outside the range of the model calibration
Model stability problems
Representation of initial/antecedent conditions
Representation of snowmelt (if applicable)
Instrument/telemetry downtime problems (rainfall)
Real time updating procedures Appropriateness for the type of model used
Sophistication of calibration software
Quality of the high flow data used both for calibration and in
real time
Event specific problems (backwater, bypassing, debris etc.)
Instrument/telemetry downtime problems (flows)
(continued)
5.5 Model Uncertainty 115
116 5 General Principles
The uncertainty in weather radar data, satellite data, meteorological forecasts,
and other inputs could also be considered where applicable.
There are many techniques for estimating the uncertainty in model outputs in
real time and these include:
● Sensitivity tests – simple tests of alternative data inputs, model structures,
parameter values etc.
● Multi-model approaches – in which the outputs from several models are com-
pared to examine the range of estimates (e.g. Georgakakos et al. 2004).
● Probabilistic approaches – in which multiple forecast realisations are produced,
typically based on stochastic sampling from probability distribution functions
for the model parameters, input data, and/or boundary conditions.
● Ensemble approaches – which use a collection of forecasts obtained by perturbing
the input data and/or model parameters for a model over plausible ranges (and
can include multi-model ensembles).
Some data assimilation techniques, such as the Kalman Filter, also automatically pro-
vide an estimate of uncertainty as part of the assimilation process (see Section 5.3).
One general distinction between approaches is whether the likely uncertainty in input
data, parameters etc. is defined beforehand based on previous forecasts, or assessed
dynamically, during a flood event, based on current forecasts and observed data.
Sensitivity tests can include ‘what if’ scenarios, of the type described in Section
5.2, whilst Rotach et al. (2007) provide an example of a real time multi-model
approach in which the outputs from a wide range of atmospheric models, and con-
ceptual and process-based catchment flood forecasting models, can be compared in
a common format, with colour coding on maps and tables to show where threshold
levels have been exceeded for specific locations.
Coastal modelsModel boundary conditions Magnitude and timing of changes in wind direction and storm
track
Subgrid scale/secondary depressions
Peak values for astronomical tides
Choice of model type and
structure
Grid resolution – inadequate representation of local bathy
metric and topographic features that cause changes in local
water levels
Coupling of offshore and nearshore models
Calibration Availability of sufficient extreme events for model verification
Influence of mobile/shingle beaches
Operational Changes in characteristics since model was calibrated
Events outside the range of the model calibration
Instrument/telemetry downtime problems
Values for trigger levels
Data assimilation Extent of improvement and limitations on data quality
Component Typical sources of uncertainty
Table 5.5 (continued)
Probabilistic and ensemble approaches are an active area for research in flood
forecasting, with several operational systems under development worldwide.
Probabilistic approaches generate multiple scenarios for model outputs by random
sampling from probability distributions for the model parameters, initial condi-
tions, boundary conditions, or input data, and many hundreds or thousands of sce-
narios might be generated offline, for subsequent analysis, or in real time.
For example, Pappenberger et al. (2004) describe Monte Carlo sampling of rating
curve and roughness coefficient uncertainty in a hydraulic model, whilst Pappenberger
et al. (2007) describe studies into the influence of uncertainty on flood inundation
extent using multiple combinations of effective model parameters for a 2D flood
inundation model. By contrast, ensemble techniques employ a more focussed sam-
pling technique (sometimes on account of constraints on model run times) to derive
a smaller number of realisations (typically of the order 10–100) which span the
likely range of outcomes (Box 5.2).
5.5 Model Uncertainty 117
(continued)
Box 5.2 Some general principles of ensemble flood forecasting
Figure 5.7 shows a simple example of an ensemble approach for a catchment
flood forecasting problem.
In this example, five possible tracks are shown for an idealised storm
approaching the catchment. The resulting catchment rainfall estimates for each
realisation could then be propagated through a rainfall runoff and flow routing
model to estimate the uncertainty in flows in the lower parts of the catchment.
The figure also shows some other sources of uncertainty, including uncertainty
in catchment antecedent conditions, stage discharge relationships and river
channel survey data. The ensemble rainfall inputs might be combined with
models to represent the uncertainties in these components as well, so that these
various sources of uncertainty are also included in the flow estimates.
Ensemble outputs can be presented in a wide range of formats, including
graphical, map-based and tabulated outputs. The raw outputs can also be used
as input to a decision support system, as discussed further in Chapter 10.
Figure 5.8 shows some idealised examples of graphical methods for present-
ing information on uncertainty in a flood hydrograph.
The figure shows the following four types of display together with the
deterministic forecast (single line):
• Spaghetti plot – the raw ensemble outputs (of which 6 are shown for
illustration)
• Plume – flow values within defined probability bounds (which often
include additional sets of bounds with appropriate colour coding)
• Whisker plot – giving the median, 25 and 75 percentile values (say) and the
maximum and minimum values
• Stacked histogram or bar chart – for a pre-defined set of values (e.g. 25%,
50%, 75% and 100%)
118 5 General Principles
Box 5.2 (continued)
River Gauging Station
Town
Raingauge
Level
Flow
Level
Storm
Rating
Rating
Antecedent conditions
River flows
Flow
Fig. 5.7 Illustration of some sources of uncertainty for a catchment flood forecasting problem
Fig. 5.8 Illustration of graphical presentations of ensemble outputs for a flow hydrograph
Time
Flow
Time
Flow
Time
Flow
Time
Flow
(continued)
Box 5.2 (continued)
Note that the probability distributions shown are idealised, and are not directly
comparable between the different graphs; for example, spaghetti plots usually
show a complex array of overlapping possible scenarios. Many other ways of
presenting probabilistic information are also possible, including tabulated, map
based and other formats, such as persistence tables, and strike probability plots
for hurricane landfalls.
Experience from other areas (e.g. meteorology) suggests that forecast
products which include measures of uncertainty should be developed in close
consultation with end users to help choose the most appropriate form of pres-
entation, and to provide advice on how the information can be interpreted.
Some current research themes in ensemble forecasting include:
● Downscaling – of the outputs from Numerical Weather Prediction models to the
scales of interest for hydrological modelling using both statistical and dynamic
techniques (e.g. Rebora et al. 2006), and (if applicable) blending ensembles
generated at different time scales into a seamless ensemble. Also, generation of
higher resolution and shorter lead time ensembles (e.g. local area and storm
scale NWP models, and probabilistic nowcasting techniques)
● Computational efficiency – in producing multiple ensemble flood forecasting
model outputs in a time which is operationally useful, with potential solutions
including emulators, parallel processing, simplification and rationalisation of
models, and filtering or clustering of ensembles
● Decision support – how to use stochastic and ensemble forecasts in making deci-
sions on issuing flood warnings, operating control structures etc., and to com-
municate information on uncertainty in flood forecast products to the public and
emergency response organizations (see Chapters 8 and 10)
● Hindcasting – reanalysis or derivation of long term ensemble simulations of the
outputs from Numerical Weather Prediction models in their present day form
(resolution, parameterizations etc.) to provide test datasets for use in developing
ensemble flood forecasting techniques
Two major research programmes which are considering these and other topics are:
● Hydrological Ensemble Prediction Experiment (HEPEX) – an international col-
laboration in ensemble flood forecasting techniques involving scientists from the
National Weather Service in the USA, the European Centre for Medium-Range
Weather Forecasts (ECMWF) and the European Joint Research Centre, Canada,
Italy, Brazil, Bangladesh and elsewhere (Schaake et al. 2005)
● COST-731 – uncertainty in advanced meteo-hydrological forecast systems – a
European initiative to examine meteorological and hydrological techniques for ensem-
ble flood forecasting, including the use of forecasts in decision making (e.g. flood
warning), and involving meteorological and hydrological services from more than ten
European countries (European Cooperation in Science and Technical Research)
5.5 Model Uncertainty 119
120 5 General Principles
Table 5.6 also presents several examples of research and operational studies in flood
forecasting applications, whilst later chapters provide additional examples for coastal
applications, and of using probabilistic forecasts to assist in optimising decision making
for flood control and emergency response (Box 5.3).
Box 5.3 Decision making using probabilistic forecasts
Probabilistic and ensemble flood forecasts provide a range of possible future
flow scenarios and, for flood warning applications, it is interesting to con-
sider how this information can be used to help in making decisions about
whether or not to issue a warning.
The development of techniques for defining probabilistic thresholds is an
active research area and, in some cases, builds on ideas which are already
well established in ensemble forecasting in meteorology (e.g. Jolliffe and
Stephenson 2003). Perhaps the simplest approach, and one which is widely
used, is to use a qualitative (“eyeball”) assessment of the spread of forecasts
to provide an indication of uncertainty, and whether any of the ensemble
members exceed thresholds of interest. Outputs can be viewed in a range of
formats, including spaghetti plots, plumes, whisker plots and histograms.
For example, when ensemble rainfall forecasts are used as inputs to a rain-
fall runoff model, the uncertainty in runoff estimates would be expected to be
greater for some types of events, such as convective rainfall events, and with
increasing lead time. Over a number of events, a forecaster could build
Table 5.6 Examples of research and operational applications of probabilistic and ensemble flood
forecasting techniques
System
Probabilistic or ensemble
component Reference
Extended Streamflow
Prediction System, USA
Statistical sampling based on long
term hydrological records;
and development of short term
ensembles for a range of sources
of uncertainty
Schaake et al. (2005)
European Flood Alert System
(EFAS)
ECMWF ensemble rainfall forecasts Thielen et al. (2004)
Bureau of Meteorology,
Australia
Ensemble Quantitative Precipitation
Forecasts, stochastic nowcast-
ing techniques, and multi-model
assessments
Elliott et al. (2005)
Environment Agency, UK Ensemble Surge Forecasts, stochas-
tic wave modeling techniques
Tozer et al. (2007)
Lower Severn, UK Data assimilation and propagation
of uncertainty in a network of
transfer function models
Romanowicz et al. (2006)
Blue River, USA; Welland
and Glen catchment, England
Ensemble Kalman filtering Butts et al. (2005)
(continued)
5.5 Model Uncertainty 121
Box 5.3 (continued)
up experience in how the uncertainty varies between locations and the type
and magnitude of event, and of the confidence to attach to forecasts in differ-
ent situations and from different models. Other criteria, such as the clustering
of ensembles, have also been shown to be useful in some studies.
A logical next step is to interpret the distribution of forecasts in terms of key
flooding and other thresholds, ideally leading to a single binary (yes/no) deci-
sion. As for deterministic (single valued) forecasts, a flooding threshold can be
defined for the parameter of interest (e.g. rainfall, levels, flows) but, in addition,
an exceedance probability needs to be defined for the proportion of ensemble
members which exceed that value. Action could then be taken (e.g. issuing a
warning) when the appropriate percentage of ensemble members exceeds the
critical value. Additional criteria could also be introduced; for example, some
studies have shown that false alarm rates can be reduced if the persistence in
threshold exceedance between two or more model runs is also considered. The
interpretation of ensembles in terms of thresholds also allows other forms of
output to be made available to the forecaster, such as colour coded maps show-
ing the locations at which flows (or other parameters) exceed threshold values.
To estimate appropriate values for probabilistic thresholds, perhaps the
simplest approach is to assess forecasting performance over a number of
events by trial and error. The optimum value can be deduced in terms of sta-
tistical measures such as Probability of Detection, average lead time, and the
number of false alarms (see Section 5.4). More generally, several studies
have suggested (e.g. Roulin 2007) that techniques developed in meteorology
for assessing the economic value of forecasts (e.g. Richardson 2003) might
also be adapted for flood forecasting applications. In this approach, the value
of a forecast can be expressed as the economic benefit, over a number of
events, from having the forecast available, compared to the situation of only
having climatological information available.
The economic value (or reduction in mean expense) typically depends
on the performance of the forecast (expressed in terms of Probability of
Detection, and False Alarm Rate), the probability threshold which is
selected, and the average costs and losses (or risk profile) for the recipient(s)
of the forecast, and is often expressed as a ratio to the equivalent value for
a perfect forecast. Probability thresholds can then be selected to provide
the maximum economic value for each individual user (or group of users),
introducing the concept that the value of a forecast depends not only on its
performance, but also on the risk tolerance and economic circumstances of
the user. Here, the costs are those incurred in taking mitigating action,
whilst the losses are those due to flooding in the absence of a warning. For
example, for a temporary or demountable flood defence barrier, assuming
that appropriate action is always taken when a forecast is issued, a cost
(e.g. in staff and transportation costs) is incurred each time that the barrier
is installed, including occasions when the forecast provides a false alarm.
(continued)
122 5 General Principles
Box 5.3 (continued)
However, losses (e.g. damage to property and vehicles) are only incurred
when flooding occurs but is not forecast.
Chapter 10 provides some further discussion of this topic, and illustrates how
costs and losses can be summarised in an expense table, similar to the con-
tingency table shown in Table 5.4. Some extensions of the approach are to
consider other factors, such as only partial mitigation of losses, a partial
response to warnings (e.g. only some people take action), and variations in
costs and losses with forecast lead time. Utility functions can also be used to
bring in other factors which cannot easily be expressed in monetary terms,
such as the differing risk profiles of recipients, and tolerance to false alarms.
For the more complex approaches, thresholds need to be computed dynamically
for each event, and the calculations are probably best performed within a
decision support system.
Cost-loss analyses of this type show potential for flood forecasting appli-
cations, particularly where flooding is frequent, or decisions need to be taken
on a regular basis, and are already used in some of the real time control and
decision support systems which are discussed in Chapter 8 for water supply
reservoirs and hydropower schemes. However, as with many other types of
flood-related analysis, the issue of extreme (rare) events needs particular con-
sideration in order to provide statistically meaningful samples, and techniques
developed in other areas, such as for design flood estimation (e.g. regional
pooling groups and continuous simulation), might provide one route to fur-
ther development of these techniques for application to low probability, high
impact events. The social and behavioural aspects of response to extreme
events also need to be considered (for example, tolerance to false alarms, and
modifications to response when faced with large or catastrophic losses).
K. Sene, Flood Warning, Forecasting and Emergency Response, 123
© Springer Science + Business Media B.V. 2008
Chapter 6Rivers
Flood forecasting models for rivers can range from simple empirical relationships
to complex integrated catchment models. Forecasts may be based simply on river
level or flow observations at locations upstream of the site of interest, or use obser-
vations and possibly forecasts of rainfall to gain additional lead time. This chapter
begins with a discussion of the main factors which can influence the design of a
river flood forecasting model, including the forecasting requirement and the availa-
bility of real time data, and then describes two main categories of model; rainfall
runoff models, and river channel (flow routing) models. Examples are provided for
a range of process-based and conceptual modelling techniques, and for data-based
approaches such as transfer function and artificial neural network models.
6.1 Model Design
River forecasting models aim to estimate river conditions at or near sites of interest
in a river basin (or catchment), such as locations at which there is a flooding history,
or where studies suggest that there may be a significant flood risk. Forecasts may
also be required at specific structures, such as reservoirs or flow control gates, to
assist with real time control of river flows to mitigate flooding.
The modelling approaches used in flood forecasting applications have many
similarities with the techniques used for catchment simulation, and the factors
which need to be considered during the initial catchment conceptualisation include
the catchment response (response times, lakes, reservoirs etc.), local influences on
river levels (tidal, backwater, confluences, structures etc.), any artificial controls on
flows (dams, river gates, tidal barriers, abstractions/discharges etc.), and other fac-
tors, such as snowmelt and groundwater influences. Chapter 8 discusses modelling
approaches for several of these possible catchment forecasting problems.
For real time applications, the following factors also need to be considered when
considering the most appropriate approach to use:
124 6 Rivers
● Forecasting requirement – the intended use of the forecasts
● Data availability – the availability of data and other information in real time and
for model calibration
● Forecasting system – the system(s) on which models will be operated
● Performance requirements – for run time, accuracy, lead time and other
variables
● Type of model – process-based, conceptual or data-based
Model design is therefore often a compromise, and the overall approach may also
be constrained by additional factors such as costs and system limitations, as dis-
cussed in Chapter 11. Chapter 5 discusses some of the factors to consider regarding
forecasting systems and model performance requirements, whilst the following sec-
tions present a brief summary of the remaining topics.
6.1.1 Forecasting Requirement
The forecasting requirement is often one of the main criteria in selection of an
appropriate modelling approach. Some factors to consider can include the level of
flood risk in the catchment, the specific locations at which forecasts are required,
and the intended operational use of the forecasts.
As noted in Chapter 5, the locations at which forecasts are required are often called
Forecasting Points, and the model (or models) used will typically be optimised to
achieve the required performance at those locations. For example, if the forecasting
model is based on an existing simulation model, it may be desirable to remove some
of the model complexity away from these key locations to make the model run faster,
or to ensure model stability under all combinations of flow conditions. This is a
commonly used approach with real time hydrodynamic models, for example.
For river flood forecasting applications, some examples of possible locations for
Forecasting Points include:
● Flood warning areas (to assist with issuing warnings)
● High risk locations (power stations, hospitals etc.)
● River control structures (to assist with structure operations)
● River gauging stations (for real time evaluation and updating of forecasts)
● Reservoirs and dams (to assist with operations to reduce flood risk)
For a given location, there may be more than one Forecasting Point; for example, a
single Flood Warning Area could include points at several anticipated overtopping
locations in a flood defence system.
Where Forecasting Points are separated by some distance, another consideration
is whether to develop individual models appropriate to each point, or an integrated
catchment model covering the whole catchment (or, at least, the catchment above
the furthest point downstream). The integrated solution may be more complex and
expensive to develop initially, but may reduce telemetry requirements, and may
allow solutions to be developed for intermediate Forecasting Points where the risk
and benefits are too low to justify development of a forecasting model specifically
for that location, such as small numbers of properties or farmland. Chapter 8 pro-
vides more background on integrated catchment modelling techniques.
The intended use of forecasts can also strongly influence the types of model
which are selected. For example, it can be useful to view the problem from the point
of view of potential users of the forecasts, such as the emergency services, local
authorities, and the public, and Table 6.1 provides some examples of how user
requirements can translate into a modelling requirement.
Here, the term hydrograph refers to the variation in river levels or flows over
time. These examples, although simplified, illustrate that the complexity of the
modelling approach can potentially vary considerably between applications.
Models may also be required for purposes other than flood forecasting; for exam-
ple, for forecasting water resource availability, or water depths and velocities for
navigation, and again Chapter 8 discusses this topic in more detail.
In some of these applications, the main interest is in forecasts of river levels and
flows in the time leading up to crossing of a threshold value which, as described in
Chapter 3, can be at levels some way below the flood peak. This requirement can
have implications for the choice of model; for example, for hydrodynamic models,
if thresholds are set at river levels for which flows remain in channel, then there
may be no particular requirement to model the details of floodplain flows or flood
Table 6.1 Illustration of how user requirements can translate to a modelling requirement
(Adapted from Environment Agency 2002, © Environment Agency copyright and/or database
right 2008. All rights reserved)
Question Possible minimum modelling requirement
Will flooding occur? Expected value for the peak level
When will the flood be at its worst? Magnitude and timing of the peak
When will flooding begin? Time at which a threshold level is reached
What depths (and, possibly, velocities)
will be reached at this street, road,
railway etc.?
The peak level reached and/or the flow on the
floodplain
How long will the flooding last? Times of crossing thresholds on the rising and
falling hydrograph
When will flood levels drop? Time of dropping below a threshold; possibly
also a floodplain and/or reservoir/storage
drainage model
Which properties will be flooded? Flows and volumes on the floodplain and
location of any overtopping
Where will flood defences be overtopped
first (or where should sandbags
be placed?)
The peak level reached at one or more locations
in a flood defence system, and possibly
defence breach risk modelling
When should this control gate be operated?
Does the reservoir level need to
be reduced?
Can vary widely, from very simple models
to decision support systems incorporating
optimisation algorithms
6.1 Model Design 125
126 6 Rivers
defence overtopping in the neighbourhood of Forecasting Points, which can sim-
plify the overall model development (although it is usually useful to know the likely
timing and magnitude of peak levels at telemetry sites).
If there are no limitations on budget, then one option is to produce the best
model that is technically feasible, including floodplain flows, river control struc-
tures, and other factors, as appropriate. However, even the best models make some
assumptions, and may not include all flooding mechanisms (e.g. urban drainage),
so it is important to be realistic about what the model can (and cannot) achieve in
discussions with potential users of the forecast outputs. Also, sometimes a simple
model, such as a regression relationship, may meet the requirement, representing a
considerable saving in development time and costs compared to a more sophisti-
cated approach.
In many cases, the choice of modelling approach is also influenced by the level
of flood risk at the selected Forecasting Points. For example, if only a few proper-
ties are at risk from occasional flooding, there may be less justification for develop-
ing a complex model than for a major city, with thousands of properties at risk.
These economic aspects of model selection are discussed in Chapter 11 as part of
a wider discussion of the costs and benefits of flood warning.
6.1.2 Data Availability
In addition to the availability of historical calibration data, and survey and other
static data, a key consideration in designing river forecasting models is the availa-
bility of near real time data for model operation and real time updating. Whilst his-
torical records are useful for exploring catchment rainfall distributions and flow
response, and in model calibration, ultimately the model needs to be configured to
operate using the real time data feeds which are available.
As discussed in Chapter 2, these can include rainfall observations from rain-
gauges, weather radar and satellite, and rainfall forecasts from nowcasting and
Numerical Weather Prediction models, and observations of river level and flows,
but may also include information on control structure settings, pumping operations,
reservoir levels, evaporation, catchment conditions, and other parameters. Also, as
described in Chapter 10, the forecasting model may form part of a wider decision
support system incorporating information on flood defence condition, locations of
temporary defences (barrier, sandbags etc.), and the current flooding situation
(floodplain depths etc.).
The requirement for forecast lead time is an important consideration in the
choice of modelling approach, since this may influence whether a river channel
(flow routing) model relying simply on information from locations further upstream
is sufficient, or whether a rainfall runoff model, representing the relationship
between rainfall and flow, is required. Generally there is a trade-off between
increasing lead time and decreasing forecast accuracy (e.g. Reed 1984; Environment
Agency 2002), as illustrated in Table 6.2.
Various attempts have been made to link the choice of model to catchment
response; for example, Reed (1984) proposed the following criteria based on the
characteristic response time of the catchment (TP):
TP ≤ 3 hours Rainfall runoff modelling plus quantitative rainfall forecasts
3 ≤ TP ≤ 9 hours Rainfall runoff modelling
TP ≥ 9 hours Flow routing
However, this was only proposed as a rule of thumb, and the many assumptions in this
approach were noted. Later approaches have included additional factors such as a dis-
tinction between overland and river channel flows (e.g. Lettenmaier and Wood 1993),
and allowances for the various time delays in the data processing, model run, and dis-
semination chain which are discussed in Chapters 5 and 9 (e.g. Tilford et al. 2007).
Map based presentations of catchment response can also help in deciding on an
appropriate choice of models; for example, by plotting lines of equal response
times based on analyses of historical data, or modelling studies, together with the
locations of key Forecasting Points. Figure 6.1 illustrates one example (Box 6.1)
and Chapter 3 provides a further example, and more sophisticated contoured or grid
based analyses are easily performed using Geographical Information Systems.
When estimating response times using historical rainfall and flow data, care
should be taken to choose events of a similar magnitude and type to those for which
the model is required, if this information is available. For example, response times
can vary significantly depending on catchment antecedent conditions, and whether
river flows are in-bank or extend onto the floodplain.
For a given forecasting problem, some issues which may influence the choice of
real time data inputs used, and the overall modelling approach, include:
● Spatial density – Chapter 2 includes a short discussion on network densities for
raingauges and river gauges for various applications. For example, for flood
forecasting applications, it is often desirable to have a river gauge at or near each
of the key Forecasting Points of interest. For rainfall runoff models, another
consideration is whether to use process-based (distributed) models when the
main sources of input data are also gridded (e.g. weather radar data, Numerical
Weather Prediction model outputs), or to aggregate the gridded outputs to catch-
ment scale, and use a conceptual (lumped) or data-based approach (see later).
● Data quality – The accuracy and consistency of model outputs will depend on
the quality of data inputs (‘rubbish in-rubbish out’), so difficult choices may
need to be made on whether to include instruments which are sited conveniently
Table 6.2 Illustration of the trade-off between forecast lead time and accuracy
Approach in order of decreasing lead time Typical modelling approach
Rainfall forecasts (e.g. Numerical Weather Prediction) Rainfall Runoff Models
Rainfall observations (e.g. raingauge, weather radar) Rainfall Runoff Models
River flow forecasts upstream of the Forecasting Point Flow Routing Models
River flow observations upstream of the Forecasting Point Flow Routing Models
6.1 Model Design 127
128 6 Rivers
for inclusion in the model, but have poor or unreliable outputs. In particular,
performance during heavy rainfall and high flow conditions will need to be con-
sidered, particularly for river gauges; for example, the risk of flows bypassing
the gauge, backwater influences, and the accuracy of the high flow end of stage-
discharge relationships (see Chapter 2).
● Data reliability – Since a primary requirement of a forecasting model is to oper-
ate during flood events, the reliability of real time data feeds during heavy rain-
fall is of particular importance; for example, is a monitoring station likely to be
flooded, or telemetry links interrupted? A fall-back hierarchy of data inputs may
need to be considered, and the model calibrated or tested for each of these sce-
narios (see Chapter 5).
● Model initialisation requirements – Certain types of model may have particular
requirements for real time information to initialise the model state. Examples
include reservoir models, for which information on reservoir levels and gate set-
tings (if applicable) is usually required, hydrodynamic models, and some types
of rainfall runoff model, which may need long runs of historical data to reinitialise
after a break in operations.
● Data assimilation – If real time updating is used, data quality issues are of par-
ticular importance for any river gauging stations which are used as updating
locations. If data quality is a concern, then other options may need to be consid-
ered, such as not using updating at those locations, or only updating within
specified flow ranges (see Chapter 5).
● Ensemble forecasting – If a probabilistic or ensemble approach is to be used for
rainfall or other inputs (see Chapter 5), then multiple model runs may need to be
performed at each forecasting time step, with possible issues of model run times
and post processing of model outputs.
6.1.3 Type of Model
In choosing an appropriate modelling solution, there are many types of model
which could potentially be used, and one widely used classification scheme is as
follows:
● Process or physically based models – which model the spatial variations in catch-
ment or river response in detail, typically using physically based equations or
functions, on a regular or irregular grid (sometimes called distributed models)
● Conceptual models – which, although to a certain extent physically based, con-
ceptualise the overall catchment or river response, whilst still representing the
main features of the response
● Data based models – which use systems analysis concepts, such as transfer func-
tions and artificial neural networks, to capture the main features of river response
(and are sometimes called data driven, metric or black box models)
In some applications, the choice of approach may be decided by external factors,
such as the time and budget available, local policy, the model calibration software
6.1 Model Design 129
Box 6.1 Model selection example
Figure 6.1 illustrates a simple example of the model selection problem for a
small catchment. There are two towns along the main river with a significant
flood risk, and a Forecasting Point is required in each location. There are two
raingauges in the catchment, and four river gauging stations, all of which are
on telemetry. However, there is no weather radar data available of suitable
accuracy for rainfall estimates.
A wide range of modelling solutions could be proposed, ranging from a sim-
ple correlation based approach, to a full hydrodynamic modelling solution.
3 hrs
2 hrs
1 hr
2 hrs 5 hrs
7 hrs
Reservoir
River Gauging Station
Town
A
C B
D
Raingauge
1
2
Fig. 6.1 Example of a simple catchment modelling problem
Configuration 2
RainfallRunoff 2 Gauge BRain 2
RainfallRunoff 1Rain 1
RainfallRunoff 3 Gauge C
Routing 1 Gauge DGauge A
Configuration 1
RainfallRunoff 1 Gauge DRain 1
Fig. 6.2 Some possible model configuration options
(continued)
130 6 Rivers
available, the ability to run models in real time, and the availability of real time data for
model operation. Some types of model may also be better suited to certain types of
forecasting problem than others, and Chapter 8 presents examples of a variety of fore-
casting approaches for flash floods, snowmelt, river ice, reservoirs, control structures,
urban drainage and geotechnical risks such as dam break and flood defence breaches.
The skills and modelling preferences of individuals can also be a factor and,
whilst it might be anticipated that the more complex types of model would have
better performance, this is not necessarily always the case in real time applications
(e.g. Beven 2001; Arduino et al. 2005). For example, for rainfall runoff models,
Table 6.3 provides some examples of the advantages which are sometimes stated
for each general type of model when used for flood forecasting applications.
A similar table could also be produced for flow routing models. Here, a parsi-
monious model is one which is no more complex than necessary to predict the
observations sufficiently accurately to be useful, and links to the concept of equifi-
nality, which is that there may be many models of a catchment (e.g. parameter sets
for an individual type of model) that are acceptably consistent with the observations
available (e.g. Beven 2001).
Box 6.1 (continued)
Figure 6.2 shows two intermediate modelling solutions; Configuration 1,
which uses a single rainfall runoff model to Gauge D, with flows at sites fur-
ther upstream estimated by scaling on catchment area or correlations (not
shown), and Configuration 2, which uses all four river gauges in the catch-
ment, with a hydrological or hydrodynamic flow routing model extending
from Gauges A, B and C to Gauge D. In Configuration 2, the use of a hydro-
dynamic model would allow forecasts to be extracted at internal node points
in the model if required; for example at the uppermost Forecasting Point
whilst, if a hydrological routing approach is used, the model could extend
from Gauge A to Gauge D, with the inflows from Gauges B and C included
at appropriate node points.
As is often the case, the models are configured around the locations
of telemetry gauges, rather than to natural features in the catchments
such as confluences. This also allows the forecast model outputs to be
updated based on real time data from those locations. Configuration 1
ignores possible differences in catchment response between the various
subcatchments, whilst Configuration 2 includes some representation of this
effect, and of confluence influences at the uppermost Forecasting Point.
Of course, several other configurations could be envisaged, with one
obvious choice being to estimate the rainfall in individual tributary
subcatchments using a catchment averaging approach based on both rain-
gauges, and possibly additional raingauges outside the catchment, and
another option being to include a sub model for the influence of the
reservoir.
Later sections also discuss some other factors to consider in the choice of
approach, including model run times, model performance outside the range of cali-
bration, model stability, the representation of variations in runoff within a catch-
ment, and of catchment initial conditions, and the number of parameters which
need to be estimated or calibrated.
Another approach to the question of model selection is to use more than one
model for a given forecasting problem and to compare the outputs to see if they
agree on likely future flows and, in particular, on the possibility of flooding thresh-
olds being exceeded (e.g. Rotach et al. 2007).
Some additional factors to consider in choosing the optimum modelling
approach include:
● Is the availability and quality of real time data sufficient to develop a useful
model?
● Are the catchment and flooding processes understood well enough to develop a
suitable model?
For example, it is often the case that the near real time information available is not
as complete or reliable as would ideally be required, or that there are uncertainties
in how the catchment responds to rainfall, or the precise conditions which cause
flooding to occur. If rain gauge information on rainfall is insufficient, then one
option is to use inputs from other techniques with an improved spatial coverage,
such as weather radar data, Numerical Weather Prediction or nowcasting model
outputs, and satellite data. Best practice would then be to investigate the perform-
ance of these inputs using historical records for the catchment (or nearby catchments)
Table 6.3 Some potential advantages of different rainfall runoff modelling approaches
Type Description
Process based models Well suited to operate with spatially distributed inputs (weather
radar, Numerical Weather Prediction model outputs, multiple
inflow locations etc.)
Can represent variations in runoff with both storm direction and
distribution over a catchment
Parameter values are often physically based and can be related to
catchment topography, soil types, channel characteristics etc.,
including (possibly) the potential to represent events outside
the range of calibration data, and for ungauged catchments
Conceptual models Fewer parameters to specify or calibrate than in the process based
approach
Fast and stable for real time operation
Easier to implement real time state updating than for process
based models
Data based models Parsimonious, run times are fast, and models are tolerant to data
loss
Can be optimised directly for the lead times of interest
The model fitting or data assimilation approach automatically pro-
vides a measure of uncertainty for some types of model
6.1 Model Design 131
132 6 Rivers
and, if satisfied with the performance, to calibrate the model directly to the inputs
which will be used in real time.
If the river gauge coverage is incomplete, one option might be to delay imple-
mentation of the model until sufficient real time data is available. Temporary moni-
toring equipment, such as water level recorders, might also be installed, and
exploratory modelling studies performed to better understand the catchment
response. An interim model could also be developed, providing a basic framework
for flood forecasting, but with some components scheduled for improvement when
additional information becomes available.
If the decision is taken to install new monitoring equipment, this introduces a
delay into the model development programme whilst site permissions are obtained,
site works are completed, and a period of reliable data is collected to use in model
calibration. Some options for accelerating this process include adding a telemetry
link to existing instruments which are known to perform reliably, or upgrading
existing equipment to resolve known problems, such as with the high flow end of
stage discharge relationships (see Chapter 2).
The approach used will depend on the time and budget available, and Chapter
11 describes a range of approaches to prioritisation of investment. For the telemetry
and modelling solution aspects, some additional guidance can also be found in Sene
et al. (2006) and Tilford et al. (2007).
6.2 Rainfall Runoff Models
6.2.1 Introduction
Rainfall runoff models aim to estimate flows in a river channel from observations
or forecasts of rainfall, and are sometimes called hydrological or hydrologic models.
If observed values of rainfall are used, the maximum lead time provided is typically
similar to the average response time of the catchment to rainfall, although may be
influenced by factors such as initial catchment conditions, snowmelt, storm speed
and direction, and reservoir storage.
In many cases, the lead time provided by using observed rainfall from rain-
gauges or weather radar can be sufficient, but can be extended by using rainfall
forecasts, although usually with a trade off between increasing lead time and
decreasing accuracy and a coarser spatial resolution. Some general categories of
rainfall runoff model include:
● Process based models – which typically attempt to model catchment processes
in some detail, using partial or ordinary differential equations and possibly sim-
pler, more conceptual relationships, to represent overland flows, soil infiltration
and percolation, groundwater flow, evapotranspiration, and other factors. Models
are often grid based, using regular or irregular grids, ideally with the grid scale
small relative to the scale of the catchment. The gridded approach is well suited
to use with weather radar or satellite data, and nowcasting and Numerical
Weather Prediction model forecasts, although rainfall fields derived from rain-
gauge observations can also be used. Models of this type can, in principle, rep-
resent the spatial variations in runoff which occur as a storm passes over a
catchment, including modelling the influence of storm path and direction on
flood response. Some alternative names which are used include distributed models,
and physically based models (although the degree to which processes are mod-
eled can vary between each approach).
● Conceptual models – sometimes called lumped conceptual models, also use a
physically based approach in the sense that they attempt to capture the main fea-
tures of catchment response through empirical and simple physically-based equa-
tions. However, models are typically formulated at the scale of the whole
catchment, and use rainfall inputs averaged (or lumped) at that scale, hence losing
some of the detail of the flood response provided by a fully distributed approach.
● Data based models – such as transfer functions and artificial neural networks typi-
cally view the rainfall-flow or rainfall-level forecasting problem from a systems
analysis perspective, with the aim often being to optimise forecasts at one or more
lead times (1 hour ahead, 2 hours ahead, for example), making best use of all real
time data available on current and past rainfall and river conditions. Rainfall inputs
are typically used as received, and are not pre-processed to catchment or grid
scale. Some alternative names for this type of model include Black Box or Metric
models. One of the earliest types of model to be used was the unit hydrograph
approach (Sherman 1932), although this has various drawbacks for real time appli-
cation, and is little used nowadays in flood forecasting applications.
In rainfall runoff forecasting applications, one distinguishing feature of the data
based approach is that models are event-based, in that they are usually only oper-
ated when required during a flood event. By contrast, process based and conceptual
models usually need to save some measure of catchment conditions at the end of
each run, or be able to reconstruct that information by retrieving historical data if
the model has not been in continuous operation, or by receiving an external feed of
observations or other estimates of catchment state.
However, the distinction between these types of model is not always clear cut;
for example, process based models can include conceptual components for some
processes, whilst conceptual models can be applied to individual subcatchments, or
to hydrological response units based on catchment characteristics such as vegeta-
tion, soil, geology and topography (e.g. Beven 2001), thereby providing a form of
semi distributed model.
Various hybrid forms of model are also available; for example, conceptual models
incorporating transfer function components, and data based models representing
slow and fast response timescales for catchment response, which can be interpreted
in terms of groundwater and surface flow pathways, and which are sometimes
called hybrid-metric-conceptual or grey box models. More generally, a widely used
option for conceptual and data based models is to interconnect a number of models
for individual tributaries via flow routing models to represent, to some extent, the
6.2 Rainfall Runoff Models 133
134 6 Rivers
variations in runoff as a storm passes across a catchment (see Section 6.1 and
Chapter 8 for examples of this approach)
There have been several major intercomparison studies of the performance of
different types of rainfall runoff model, including the following two studies:
● European Flood Forecasting Study (EFFS) – this major collaborative research
study into improved flood forecasting techniques for large catchments included
a comparison of the performance of eight process based and conceptual rainfall
runoff models on 14 catchments in Europe (European Flood Forecasting
System 2003).
● World Meteorological Organisation 1992 study – this was the third such inter-
comparison exercise organised by WMO, and involved 14 models from 11
countries (previous studies were in 1974 and 1983). Models were compared
using data for catchments with areas of 104, 1,100 and 2,344 km2, using real
time updating where available, and a range of performance statistics (World
Meteorological Organisation 1992).
However, due to the number of factors to consider, conclusions from this type of
study can sometimes be mixed, and Reed (1984), for example, suggests that a thor-
ough assessment of rainfall runoff methods for flood forecasting might need to
consider:
● At least four distinct approaches
● Perhaps four different model structures in each approach
● Several methods of real time correction
● Perhaps several alternative types of rainfall forecast
● Various objective functions both for model calibration and performance
assessment
● Application to a range of flood forecasting problems
However, despite these difficulties, intercomparisons can provide useful insights into
model performance, and some general conclusions on factors which can influence
the performance of rainfall runoff models in forecasting applications include:
● Calibration – the approach used to calibration, the calibration criteria used, and
more qualitative factors, such as the skill of the modeller. Also, for real time appli-
cations, it is often appropriate to use additional performance measures, based on
threshold crossing and the model performance at different lead times, in addition
to the classical measures used for simulation modelling (see Chapter 5).
● Initial conditions – except in some locations (e.g. some desert mountains), the
runoff generated by a catchment often depends strongly on catchment initial
conditions (soil moisture, reservoir and lake levels, snowcover etc.). The degree
to which a model is able to capture this effect is an important consideration for
rainfall runoff models.
● Real time updating – given the many uncertainties in measuring and forecasting
rainfall, and estimating the resulting flows, updating (or data assimilation) can
be particularly effective for rainfall runoff models, provided that the data quality
is sufficient. It is important to examine the performance of the model both with
and without updating, and perhaps to evaluate performance with a range of
approaches to updating (see Chapter 5).
The following sections discuss these points further, whilst additional background
on rainfall runoff modelling techniques for a range of applications can be found in
Singh (1988, 1989), Beven (2001) and Moore et al. (2006), for example, and in
papers and conference proceedings from the MOPEX and IAHS-PUB research
studies for ungauged catchments which are described later.
6.2.2 Process-Based Models
Process based, distributed or physically based hydrological models have been used
in research and simulation studies for many years (e.g. Vieux 2004) and are gradu-
ally being adopted for real time forecasting applications.
Historically, one of the obstacles to real time use has been that models of this
type are data hungry, in the sense that they require detailed spatial information on
rainfall, temperature, evaporation, land use, catchment state and other factors, and
model run times may be too long. Increasingly, however, these problems are being
overcome through improvements in computing power, and the resolution of weather
radar data and weather forecasting models, and in remote sensing and spatial analysis
techniques for assessing factors such as topography, vegetation, land use, river net-
works etc. In particular, in recent years, for short forecast lead times, the typical
horizontal resolution of the nowcasting and Numerical Weather Prediction models
used for weather forecasting (see Chapter 2) has started to reach grid scales which
are of interest for hydrological modelling, which are often a few kilometres or less,
depending on catchment size.
One distinguishing feature of the process based approach is that often the model
parameters are defined to be within specified ranges depending on soil type, slope,
land use, river channel hydraulic characteristics, and other factors. Values are typi-
cally derived from laboratory or field experiments, or datasets from previous studies
on other catchments, and the initial model calibration consists of choosing the most
appropriate parameter set given the characteristics of the catchment. Parameter
values are then fine tuned to achieve a good match between observed and forecast
flows, either by trial and error, or by using optimisation algorithms for those param-
eters for which values are uncertain. Bayesian and other approaches may also be
used to provide an assessment of model and parameter uncertainty. This approach
contrasts with the conceptual and data based approaches described later in which
model parameters are typically estimated primarily by minimising one or more
measures of model fit.
Many types of process-based model have been proposed, and some typical
features can include one or more of the following components:
6.2 Rainfall Runoff Models 135
136 6 Rivers
● Surface and sub-surface flow paths
● Kinematic Wave and similar approaches to routing water between cells, possibly
also considering channel characteristics such as roughness coefficient and channel
geometry (see Section 6.3)
● Flow paths inferred from Digital Elevation Model datasets
● Multi-layer soil moisture models operating at grid scale and, sometimes, at sub-
grid scale
● Evapotranspiration linked to vegetation cover, Leaf Area Index etc.
● Runoff processes linked to soil and land cover properties, typically obtained
from remote sensing
● The option to include additional processes such as snowmelt, lake/reservoir
influences etc.
● A mixture of storage functions and partial differential equations used to repre-
sent individual processes
Alternative grid configurations can also be used, in which flows are either translated
between grid cells, or routed directly to the catchment outlet (e.g. Moore et al. 2006).
For flood forecasting applications, the choice of model will depend on many of
the factors described already, including the forecasting requirement, catchment
response, forecasting system, real time data availability etc., and the information
available for model set up and calibration.
Some real time flood forecasting applications of process-based models include
the following examples:
● G2G (Bell et al. 2007)
● HL-RMS (Koren et al. 2004)
● LISFLOOD (De Roo et al. 2000; Van Der Knijff et al. 2004)
● MGB-IPH (Collischonn et al. 2007)
● MIKE-SHE (Butts et al. 2005)
● REW (Reggiani and Schellekens 2003)
● TOPKAPI (Liu et al. 2005)
An important consideration in assessing the performance of process based models
is the extent to which parameter values can be specified in advance based on land
surface characteristics (soil, vegetation, topography etc.), and used on ungauged
catchments. Two major international studies on parameter estimation for ungauged
catchments are:
● IAHS PUB – the “Predictions in Ungauged Basins” study – is an International
Association of Hydrological Sciences (IAHS) initiative, from 2003 to 2012,
aimed at uncertainty reduction in hydrological practice (Sivapalan et al. 2006).
● MOPEX – The “Model Parameter Estimation Experiment” – which is an open
collaborative study funded by the NOAA Office of Global Programs (USA),
whose aim is to develop techniques for the a priori estimation of the parameters
used in land surface parameterisation schemes of atmospheric and hydrological
models (Andrssian et al. 2006).
As studies like these progress, the options for forecasting on ungauged catchments
should improve, and this topic is discussed further in Chapters 3 and 8.
6.2.3 Conceptual Models
Conceptual rainfall runoff models have many similarities to process based models,
but operate at a catchment scale, and place less reliance on a physically based
description of response. Typically, models represent the flow of water using a series
of conceptual stores, which fill and empty depending on the rainfall inputs, and lose
water from the system due to evaporation (from open water), evapotranspiration
(from vegetation), and to river flows.
Many types of conceptual model have been proposed, and some typical compo-
nents can include one or more stores of the following types:
● Interception store – in which rainfall falling on a catchment initially enters a
store representing water held by vegetation (forests, grass etc.), with some rep-
resentation of evaporation and evapotranspiration processes
● Surface store – which represents flow through the river network and overland
flows
● Soil store – which represents storage of soil moisture in the catchment
● Subsurface store – which represents recharge to groundwater, and subsequent
outflows to the river network
Individual stores may also be represented in a variety of ways, with options includ-
ing modelling of outflows as a function either of volume stored (e.g. linear or power
law functions), or using stores with fixed volumes which pass water downstream
once they fill and overflow.
For flood forecasting applications, conceptual models are typically calibrated by
trying to achieve a good representation of observed flows across several flood
events, and possibly also for the intervening flow periods, to achieve a long term
water balance. The criteria for optimisation can include factors such as the shape of
the hydrograph, the peak levels reached, mean square error, bias, and various real
time related statistics such as those described in Chapter 5. Some types of models
may have 15–20 or more parameters to consider, and a typical approach is to opti-
mise a few parameters at a time (e.g. for the baseflow component), holding all others
constant, and perhaps placing bounds on the permitted values. The optimisation
scheme may also explicitly, or implicitly, include the parameters for the following
two additional modelling components:
● Catchment averaging component – one consequence of using the conceptual
approach with raingauge data is the need to derive catchment average values for
rainfall, and the parameters of the averaging process can also be viewed as part
of the optimisation process. Chapter 2 discusses a range of approaches to catch-
ment averaging, including Thiessen Polygon and geostatistical methods.
6.2 Rainfall Runoff Models 137
138 6 Rivers
● Real time updating – can be particularly important for conceptual rainfall runoff
models to compensate for uncertainties in rainfall values, initial conditions and
other factors. As noted in Chapter 5, error prediction, state updating and param-
eter updating techniques can all be used although, given the process based
flavour of the approach, parameter updating is less widely used.
Many different types of conceptual model have been developed for flood forecast-
ing and other applications (e.g. World Meteorological Organisation 1994), and the
intercomparison experiments described earlier present several examples. Some
approaches which have been used in flood forecasting applications include:
● HBV model (Bergstrom 1995)
● MIKE11/NAM model (Madsen 2000)
● PDM model (Moore 1985, 2007)
● Sacramento catchment model (Burnash 1995)
● TANK model (Sugarawa 1995)
● URBS model (Malone 1999)
● Xin’anjiang model (Zhao 1992)
Some particular model configuration options worth noting include:
● No subsurface store – for models used primarily for flood forecasting, one option is
to use an event based approach and to omit the subsurface component entirely, on the
basis that runoff during a flood event occurs primarily by surface flow. Instead, a fixed
or variable runoff coefficient is introduced to represent the proportion of rainfall
which goes to surface runoff, such as in the Isolated Event Model, for example.
● Enhanced subsurface stores – additional submodels can be included, or updating can
be based on real time monitoring of groundwater levels, to represent the influence
of groundwater levels and storage on river runoff where this is a significant factor.
● Soil store – use of a probability distribution for soil moisture storage to take account
of the variations in soil storage which occur in a catchment (e.g. Moore 1985).
● Time delays – although the store outflow parameterisations typically provide
some time delay between inflows and outflows, it can be helpful to include
additional lag parameters to assist with fitting the overall model to represent
the time delays in surface and subsurface flow pathways.
● Complicating features – some models include the facility to model reservoirs
(including control rules), irrigation schemes, abstractions and discharges related
to water supply, and other factors which influence river flows.
Given the range of modelling possibilities, an increasingly common approach in
developing conceptual models is to provide a range of options for the types of stores
which are included, and for how they are interconnected and parameterised. This has
led in recent years to the development of modelling toolkits, which allow users to
select from a range of sub-models to represent antecedent conditions, subsurface
flows, runoff response etc., whilst providing an overall framework for model calibra-
tion and optimisation. The model can then be configured in the most appropriate way
for each problem.
As with process based models, attempts have been made to relate the parameters
of conceptual models to catchment characteristics, although results are often
dependent on the type of model. However, this approach can help to reduce the
work required in model calibration, and also opens the way to forecasting flows on
ungauged catchments (i.e. catchments with no river gauging telemetry).
6.2.4 Data-Based Methods
Data-Based methods are used in many technical fields for the real time forecasting
and control of complex systems. Examples can be found in the transportation,
industrial process, aviation and financial forecasting sectors, and include transfer
functions, artificial neural networks, and fuzzy logic techniques.
In flood forecasting applications, the main use is usually for rainfall runoff mod-
elling, although data-based methods have also been used for river flow routing,
estuary forecasting, and coastal surge forecasting as described in later sections and
chapters. Data-based rainfall runoff models can also be used to forecast river levels
directly (i.e. rainfall-level models), although a cautionary note is that this may only
work well if there is a unique relationship between levels and flows at the selected
Forecasting Point (for example, if there are no significant backwater influences).
In the early years, one of the drivers for development was the limited computing
power available at that time, and data-based models generally run more quickly
than a more physically based approach. With current processor speeds this is nowa-
days much less of a consideration, except in the case of ensemble forecasting,
where data based techniques have potential as emulators for models with longer run
times (e.g. process based models).
However, the data based approach also views flood forecasting from more of
a systems perspective, in which the main aim is to develop a model to infer future
conditions at one or more locations (Forecasting Points), making best use of all
of the real time and historical information available at the time of the forecast,
and often providing a measure of uncertainty in the resulting outputs. In particular,
models are often optimised to provide forecasts of the particular parameter of
interest for flood warning, at the lead times which are most useful operationally.
This contrasts with the more physically based approaches, in which the focus for
calibration is often reproducing the shape and timing of the whole hydrograph
across a number of events in simulation (off-line) mode, although both threshold
and lead time based criteria are increasingly being used in model development
(see Chapter 5).
Perhaps the most widely used data-based techniques are the transfer function
and artificial neural network approaches. Transfer functions are one of a wide range
of time series analyses techniques used in a range of industries (Box and Jenkins
1970) and can be combined into networks to form integrated catchment models
(e.g. Beven et al. 2005). Various related autoregressive techniques have also been
considered for forecasting applications, although have been less widely used
6.2 Rainfall Runoff Models 139
140 6 Rivers
(although are used extensively in some of the real time updating approaches
described in Chapter 5).
For flood forecasting applications, a simple transfer function formulation might
relate river flows at time t (Qt) to flows and/or rainfall (P) at previous time steps as
follows:
Qt = a
1Q
t-1 + a
2Q
t-2…a
mQ
t-m + b
0P
t-T + b
1P
t-1-T + b
2P
t-2-T …… + b
nP
t-n-T (6.1)
where ai and b
i are the parameters of the model, and T is a fixed time delay. A model
with m flow parameters and n rainfall parameters is said to be of structure or order
(m, n, T). A random noise component may also be included, although is not shown
here. An important objective in transfer function modelling is often to derive a
model which uses the minimum number of parameters possible (i.e. is parsimoni-
ous). The use of a time delay, whilst not essential, can help in reducing the overall
number of parameters required.
Some potential issues with the basic linear approach represented by Equation
(6.1) are that flows are unconstrained and may be oscillatory or negative, and that
the use of observed (total) rainfall can fail to capture the influence of initial catch-
ment state on runoff.
One way to reduce the risk of oscillatory response is to minimise the number of
parameters used. Alternatively, the mathematical formulation can be redefined to
provide a constrained and stable output; for example, in the Physically Realisable
Transfer Function (PRTF) approach of Han (1991).
Various techniques have also been developed to account for non-linear influ-
ences on river flows such as initial catchment conditions, and these include:
● Effective rainfall inputs – using only a proportion of the total rainfall to drive the
model, with rainfall separation techniques including use of a variable, threshold-
based or fixed runoff coefficient, perhaps related to initial catchment state, or
functions based on real time observations of parameters which may influence
runoff (e.g. air temperature), or continuous soil moisture accounting (e.g. Moore
1982; Lees 2000)
● Flow based initialisation – using the current observed flow as a surrogate for
catchment state (for example, a dry catchment will usually have lower flows
than a wet catchment) (e.g. Young and Tomlin 2000)
● Parallel pathway representation – interpretation of the model in terms of one or
more linear pathways in parallel which represent the different time responses of
key catchment processes, such as surface and groundwater flows (e.g. Beven
2001; Young 2001)
Although some of these techniques may seem to introduce conceptual modelling
ideas, the data-based approach is generally retained, with the model structure simply
being extended to capture different timescales and rainfall distribution effects
which are known to exist in the forecasting problem.
The extent to which these enhancements are included will depend on the nature
of the catchment for which forecasts are required. For example, a single pathway
model may be sufficient for a small, fast response catchment, but a multiple path-
way model might be more appropriate to represent runoff in a large complex catch-
ment with significant groundwater influences. Real time updating techniques may
also be used to help to account for the differences between observed and forecast
flows and can include flow substitution, adaptive gain, transfer function noise,
genetic algorithm, and Kalman filter methods (including extended and ensemble
versions), and are intrinsic to some approaches.
Artificial neural networks also use a parallel pathway approach, although with
many more layers or pathways in the formulation. The building blocks of a network
typically include components such as neurons, summing junctions, and activation
functions. For flood forecasting applications, one aspect of the model development
(often called ‘training’) is to achieve a compromise between including too few
neurons, thereby failing to capture the full range of river flow response, and using
too many neurons, focussing on the noise rather than the underlying river flow sig-
nal (see Section 7.3 for more information).
The potential for rainfall runoff modelling applications has been the subject of
several review studies (e.g. Dawson and Wilby 1999; ASCE 2000), and the tech-
nique has been trialled in various flood forecasting applications (e.g. Solomatine
and Price 2004; Abrahart et al. 2004). One active area of research is into how mod-
els can be formulated to provide reliable predictions of extreme events, beyond
those in the calibration or training dataset.
6.3 River Channel Models
6.3.1 Introduction
Whilst rainfall runoff models represent the translation of rainfall into inflows to a
river network, river channel models represent the flow of water within that network,
and include the following approaches:
● Process based – hydrodynamic models which use one-dimensional (1D), 2D or
3D approximations to the mass and momentum equations for both flows and
river levels (sometimes called physically based models)
● Conceptual models – various hydrological flow routing techniques which can
range from empirical techniques to approximations to the full mass and momen-
tum equations for river flow
● Data-based approaches – which use transfer function, artificial neural network,
and other approaches to represent river flows or levels (sometimes called black
box methods)
All of these approaches have been used successfully in river flow forecasting appli-
cations. Note that level to level, and flow to flow, correlations may also be regarded
as a type of river forecasting model, but for convenience are discussed in Chapter 3,
together with a range of other simpler empirical flood forecasting techniques.
6.3 River Channel Models 141
142 6 Rivers
In considering the most appropriate approach to use, it is useful to consider the
forecasting requirement (see Table 6.1), and the nature of river flows at or near the
Forecasting Points of interest and in the river network upstream. If the main model-
ling requirement is for the likely timing or magnitude of a peak, or exceedance of
a flow threshold, then a wide range of approaches may give satisfactory results.
Here the focus is primarily on river flows, although levels may be estimated if a
suitable stage-discharge relationship is available. However, some complicating fac-
tors which can influence the choice of modelling approach, and may require use of
a more complex type of model (e.g. a hydrodynamic model), include:
● Backwater influences – river levels at a Forecasting Point are influenced by river levels
downstream of that point; for example, from tidal influences, inflows from a tributary,
or operations at a river control structure. Often this leads to a non-unique relationship
between river levels and flows, which simpler models may not be able to capture.
● Spillage – river levels may exceed the height of natural river banks, or of flood
defences. Flows may be permanently lost to the river system, or may re-enter
further downstream, or as river levels drop.
● Natural floodplains – flows onto floodplains will tend to reduce (attenuate) the
magnitude of flow peaks further downstream, and delay the arrival of the peak.
Lakes, wetlands and marshes have a similar influence.
● Embanked river channels – flows may spill and be lost permanently from the river
network unless there is a return route via pumping or gravity drainage. However,
return flows may not occur until river levels have dropped considerably.
● Artificial influences – influences from dams, river control structures, off-line
storage (polders, washlands), pumped or gravity fed flows and other factors may
affect the timing, magnitude and duration of flood peaks.
● Tributary inflows – inflows may contribute to peak flows in the main river channel,
and may have different response characteristics to main channel flows (faster
rate of rise etc.).
The following sections describe the extent to which these various factors can be
represented by the three main types of modelling approach.
6.3.2 Process Based Models
The velocity and depth of flows in a river reach depends upon the inflow of water to the
reach, friction losses, and changes in river slope, width and shape along the channel (e.g.
Chow 1959; Chanson 2004). The unsteady water flow in a natural river is governed by
the principles of conservation of mass and momentum.
Except in certain simplified situations, the resulting equations cannot be solved
analytically, and must normally be solved numerically. Solutions can be obtained
using one-dimensional (1D), two-dimensional (2D) or three-dimensional (3D)
approaches, although 1D and, to a lesser extent, 2D approaches are most widely
used in river flood forecasting applications. Typically, solutions are obtained on a
regular or irregular grid using finite difference approaches, although finite element
and finite volume approaches can be used. One widely used approximation for one
dimensional flows is the assumption of a gradual variation of flow and a hydrostatic
pressure distribution, which results in the well-known Saint-Venant equations.
The friction terms are typically parameterised using an empirically based rough-
ness coefficient (e.g. Mannings n) which may be fixed or vary along the river reach,
and may be estimated from look up tables or a database of values for channels with
similar characteristics (bed type, vegetation etc.). Hydraulic controls on river levels
such as bridges, weirs and river control structures may be represented by sub- models,
typically parameterised using a loss coefficient. Logical rules may be included for
barrier, gate and other operations.
The information required to develop a model typically includes:
● Survey data for the channel dimensions at a sufficient number of locations to
capture the hydraulic characteristics of the river channel, any structures to be
included in the model, and possibly of flood defences (dikes/levees) and the
floodplain
● Estimates or measurements for inflows from the upstream end of the reach, and
any tributary inflows
● Estimates for the friction loss terms
● A well-defined downstream boundary condition; for example, from observed or
forecast levels, or a normal depth or observed stage-discharge relationship
● Information on structure control rules (if applicable)
Hydrodynamic models are particularly well-suited to applications where precise
estimates of river levels are needed, such as forecasting the overtopping of flood
defences, or real time control of structures such as pumps and gates, and in situa-
tions with tidal, confluence, multiple channel and other backwater influences, and
for estimating depths and (possibly) velocities on the floodplain.
‘What if’ scenarios, such as culvert blockages, and defence breaches, can also
be considered. Changes in river bed profiles might also be included by incorporat-
ing more up to date survey data. The physical basis also suggests that model outputs
can be extrapolated to flows and levels outside the range of calibration, provided
that there are no significant changes in river characteristics at these higher levels,
and that stage-discharge relationships remain valid.
For flood forecasting applications, even for a well designed and calibrated model,
some potential issues for real time applications include (e.g. Chen et al. 2005):
● Model run times – models initially developed for off-line flow simulation can
run too slowly to be useful for real time operation
● Stability – numerical problems can arise from the discretisation in time and
space causing models to exhibit unstable or oscillatory behaviour, or even to stop
operating during a model run
● Convergence – failure of the model to achieve the required accuracy within a
specified number of iterations of the solver scheme
● Initial conditions – may need to be saved between model runs, and carefully
specified to avoid stability problems
6.3 River Channel Models 143
144 6 Rivers
● Survey requirements – the required survey data may not be available, or too
expensive to capture given available budgets, or of insufficient resolution
● Ungauged inflows – if there are significant ungauged inflows, then flows may
need to be estimated using techniques for ungauged catchments (see Chapter 8)
In addition, the model may not have initially been developed for estimating high
flows, and have performance, stability, and other problems (e.g. with the high flow
end of stage discharge relationships) under these conditions. These problems are
all potentially solved, although can require considerable expertise to achieve a fast
stable model. The following options are possibilities (Huband and Sene 2005;
Chen et al. 2005):
● Removal of model complexity (nodes) in locations where detail is not required;
for example, away from Forecasting Points, and areas of significant hydraulic
control. This might include replacing hydrodynamic modelling reaches with
simpler flow routing reaches, in which flow is conserved but river levels are not
required, or using so-called sparse hydrodynamic modelling techniques, in
which only key cross section locations are retained, and the details of river
response are not considered.
● Removal or aggregation of hydraulic structures which do not exert a significant
influence on river levels in the areas of interest, and using simpler or alternate
representations for any control structures which cause stability or convergence
problems.
● Checking for problems which can cause poor stability, including poor initial
or boundary conditions, river channels running dry, inappropriate spacing of
river cross sections, surcharging at bridges and other structures, operations at
river control structures, transitions from subcritical to supercritical flow, and
other factors.
● More detailed studies of the accuracy of the high flow end of stage discharge
relationships, possibly using 2D or 3D modeling, commissioning additional sur-
vey data near gauging stations, and additional measurement campaigns (if
practicable).
Figure 6.3 shows an example of how both the spacing and lateral extent of river
cross sections might be tailored to modelling river levels or flows for a Flood
Warning Area, with sparse hydrodynamic modelling elsewhere except in the vicin-
ity of a telemetry site (where an estimate for the high flow rating is required).
Improvements to model stability and convergence, and reductions in complex-
ity, will generally also improve model run times, both through reductions in the
number of calculations required per iteration, and the number of iterations
required.
Another consideration with hydrodynamic models is whether to use real time
updating. The simplest option is to split the model at key river telemetry sites, and
to update the model outputs using an independent error prediction algorithm.
However, this can be undesirable since it removes the ability for forecasts of
downstream influences to feed back to locations upstream of the telemetry site,
and mass and momentum conservation will not necessarily be maintained through-
out the river network.
The main alternative is to attempt to update the model at internal model nodes,
and methods which have been developed include:
● State updating – adjustments to main channel or tributary inflows to distribute
errors in flow volume noted at telemetry sites
● Parameter updating – real time adjustments to model parameters such as the
roughness coefficient to ‘fine tune’ the model performance
This topic is an active area of research since both state and parameter updating can
cause unwanted transient behaviour to propagate within the model domain.
6.3.3 Conceptual Models
Hydrological flow routing models provide a simplified representation of river flows
compared to a hydrodynamic modelling approach, yet perform well in some flood
forecasting applications, and can offer the following advantages:
● Models run quickly, with stability and convergence problems less likely
● River cross section information is not necessarily required
● The techniques can work well on steep river sections, where a hydrodynamic
model might fail
The main restriction is that models work in terms of flows, and do not compute
levels, other than through optional application of a stage-discharge relationship.
There is also typically no representation of backwater influences, and the
approach is less suited to shallow sloping rivers. Some of the more complex river
flow phenomena described in the previous section also cannot easily be
represented.
Flood Warning Area
Gauge
Gauge
INFLOW
Fig. 6.3 Illustration of sparse hydrodynamic modelling techniques (not to scale)
6.3 River Channel Models 145
146 6 Rivers
One of the first methods to be developed was the Muskingum method. The
approach uses the concept of a triangular ‘wedge’ and rectangular ‘prism’ to repre-
sent storage in the river channel and flood wave, and aims to maintain a water balance
between upstream and downstream reaches. More recent developments have usually
adopted a more physically based approach, and include the following techniques:
● Muskingum-Cunge method (Cunge 1969)
● Variable Parameter Muskingum-Cunge method (VPMC) (Price 1977; Tang et al.
1999)
● Kinematic Wave approaches (e.g. Lighthill and Whitham 1955; Jones and
Moore 1980)
Cunge (1969) showed that, with an appropriate choice of length and time steps, the
Muskingum Cunge method provides a good approximation to the convective-
diffusion equation, which in turn is a simplification of the full St Venant equations.
Some studies have shown that, for a simple river channel with a floodplain but
no artificial influences, there is sometimes little to choose between fixed and varia-
ble parameter routing models in terms of predicting peak flows, but that variable
parameter models can perform better on the rising limb of the hydrograph. This can
be important in flood warning applications, where the interest may be in success at
crossing a flood warning threshold.
Although stability problems are unlikely, most types of model still require a
numerical solution scheme, in which the river reach is divided into discrete sec-
tions. For example, Tang et al. (1999) showed that there can be some numerical
issues with more complex types of flow routing models (oscillations, lack of vol-
ume conservation etc.) when used in compound channels, which can be greatly
reduced through choice of an appropriate computational scheme.
Some types of routing model also require wavespeed and attenuation parameters
to be estimated. Typically, wavespeeds increase with increasing discharge until the
river level reaches a value close to bank full, then decrease as water spills onto the
floodplain, but increase again once the floodplain flow exceeds a significant depth.
Both fixed wavespeed, and discharge dependent (variable parameter) values, may
be assumed. Values must either be supplied by the modeller or, in some cases, can
be estimated by the calibration software from typical river cross sections.
As with hydrodynamic models, it may be necessary to represent the inflows
from tributaries in a river reach, and flows into or out of the reach due to abstrac-
tions, floodplain flows, and other factors. Lateral flows of this type are easily
included, and can either be modelled directly, or represented as a proportion of
flows in the main river, weighted by catchment area, mean annual rainfall, or some
other indicator of runoff (see Chapter 8).
6.3.4 Data Based Methods
Data Based methods can also be used for river flow modelling, with the main differ-
ence compared to rainfall runoff models (see Section 6.2.4) being that flows are esti-
mated from telemetered values for a location further upstream, rather than from
rainfall values (e.g. World Meteorological Organisation 1994). For example, a simple
linear transfer function formulation for flow routing might be of the following form:
Qt = a
1Q
t-1 + a
2Q
t-2…a
mQ
t-m + b
1q
t-1-T + b
2q
t-2-T …… + b
nq
t-n-T (6.2)
where Q is the flow at the Forecasting Point, q is the inflow to the reach, ai and b
i
are the parameters of the model, and T is an optional fixed time delay. Some for-
mulations also include a random noise component. Artificial Neural Network
techniques may also be used (see Section 7.3 for more information).
Table 6.4 provides some examples of transfer function and artificial neural net-
work flow routing approaches which have been used or trialled in real time flood
forecasting applications.
Table 6.4 Examples of real time flood forecasting applications of data based flow routing models
Model Full name Example applications References
Transfer function Data based mechanistic River Nith, Scotland Lees et al. (1994)
General Artificial neural
networks
River Necker, Germany Shrestha et al. (2005)
Neural network Artificial neural
networks
Rivers in NE England Kneale et al. (2000)
6.3 River Channel Models 147
One interesting feature of the data based approach is that models can also be formu-
lated in terms of river levels, rather than flows. This avoids the need to have well defined,
accurate stage-discharge relationships for the inflow and outflow locations, although
with the possible difficulty that backwater and other influences may cause errors if they
lead to a non-unique relationship between levels and flows at either location.
Chapter 7Coasts
Coastal flood forecasting models are used to estimate conditions at or near locations
which may be at risk from flooding, such as towns, ports and harbours, and coastal
roads and railways. Forecasts may also be required at structures, such as tidal bar-
riers, to assist with operations to reduce the risk of flooding. Models can range from
simple empirical relationships to complex process-based models combining off-
shore, nearshore, wave overtopping and flood inundation components. Forecasts
may be based primarily on coastal observations, or also make use of the surge, wind
and wave forecasting outputs provided by national meteorological services and
coastal observatories. This chapter describes some of the issues in selecting an
appropriate modelling approach and then discusses a range of process based and
data based techniques for coastal flood forecasting.
7.1 Model Design Issues
Coastal flooding can arise from several factors, including high tides, and the
impacts of storms on conditions at the sea surface, generating surge and wave
action. Heat exchange between the ocean and atmosphere can also cause storms to
intensify, as with hurricanes, typhoons and tropical cyclones, for example.
Tidal effects are generated by the gravitational attraction of the sun and the
moon, usually leading to a twice-daily maximum in tidal levels along the coastline,
with the peak values varying depending on the relative alignment of the sun and
the moon. Maximum and minimum tidal ranges occur twice each month when the
earth, sun and moon are aligned so that their gravitational pull combines (high, or
spring, tides) or are approximately at right angles (low, or neap, tides). The influ-
ence of the moon on tides is roughly twice that of the sun and perturbations in the
orbits of the sun, the moon and the earth lead to additional impacts at a range of
timescales which are significant up to a period of 18.61 years.
Surge is generated by a combination of low atmospheric pressure and wind fric-
tion at the ocean surface; for example from tropical cyclones, hurricanes and
typhoons, or mid-latitude storms. Surges may develop locally, or can propagate
from distant areas, and can lead to an increase or decrease in sea levels, although it
K. Sene, Flood Warning, Forecasting and Emergency Response, 149
© Springer Science + Business Media B.V. 2008
150 7 Coasts
is increased levels which are of concern for coastal flood forecasting. In the deep
ocean, a 100 N m−2 (1 mbar) drop in pressure causes roughly a 0.01 m rise in still
water levels (the so-called inverted barometer effect), although wind related effects
are often considerably more important. Onshore winds can cause water to accumu-
late at the shoreline and this effect can be amplified by reflections in estuaries
(deltas) and local channelling effects. Surges can develop rapidly over timescales
of a few hours or less, although more typically last between a few hours and 2–3
days, depending on the scale and duration of the storm. At the shoreline, surge
effects are higher for a shallow sloping continental shelf or sea bed.
Some notable locations for storm surge impacts from tropical cyclones include
the Gulf of Mexico, the Caribbean and the Bay of Bengal, and the eastern coastlines
of Japan and China. For example, surge heights generated in Hurricane Katrina in
August 2005 were reported to have reached 7–9 m along the Mississippi coastline
(Knabb et al. 2006). Note that storms of this type are called hurricanes in the
Atlantic and Eastern Pacific Oceans, tropical cyclones in the Indian Ocean and
typhoons in the Western Pacific. Typical storm sizes are in the range 100–1,000 km,
with maximum recorded wind speeds of the order of 300 km hour−1.
Waves usually develop as a result of wind action, and can be generated both
locally by wind effects (wind waves), or may propagate from distant storms (swell).
The height and frequency of waves generated by a storm can depend on wind
speeds and on the size and duration of the storm. The eventual wave heights at the
shoreline depend on the distance over which waves develop (often called the fetch),
the slope of the sea-bed, channelling by local topography (bathymetry), and other
local factors, such as reflections at cliffs or reefs. At high latitudes, wave develop-
ment may also be affected by sea ice. Deep sea (or swell) waves are generally not
as high as locally generated waves, but have longer periods, and can potentially
cause more damage and overtopping when they encounter sea defences and other
structures. Tsunami waves may also cause flooding, although are generated by sub-
sea landslides, volcanic activity and earthquakes (see Chapter 8).
Whilst these effects generally arise from distinct processes, the resulting impacts on
sea water levels are not necessarily independent. For example, tide and surge effects
may interact, particularly in shallow water, affecting both the timing and magnitude of
peak levels at the shoreline. Similarly, heavy rainfall may occur inland during depres-
sions and tropical cyclones, leading to high river flows as well as surge, particularly on
small coastal catchments. The likelihood of high river flows and tidal levels coinciding
can depend on many factors, including bathymetry, the storm intensity, track and dura-
tion, and catchment response times (e.g. Environment Agency 2005)
Local factors, such as headlands, breakwaters and estuaries (or deltas) can also
have a strong influence on tides, surge and waves. For example, resonance effects
may amplify the tidal range in estuaries, or cause a second high tide peak on each
cycle. Also, operational problems, such as tidal gates failing to close, or breaches
occurring at sea defences, may lead to flooding whilst gate operations (e.g. at tidal
barriers) can also influence tidal levels.
One example of a location with a wide range of factors is the southwest coastline
of England, which is a peninsula approximately 250 km long, and 100 km across at
its widest point. The northeastern coastline lies within the estuary of the River
Severn, which has one of highest tidal ranges in the world, exceeding 10–15 m in
spring tides. Wave and surge influences are relatively small along this coastal reach,
although can be important if they coincide with high tides. By contrast, on the
opposite side of the peninsula, the tidal range is considerably lower, whilst surge
remains of a similar magnitude, and maximum wave heights are now higher, particularly
when they originate from the open waters of the Atlantic Ocean, and can reach
heights of 5 m or more due to local shoaling and wind driven effects.
Other examples include the wide variations in coastal inundation even over short
distances during the December 2004 Tsunami, depending on coastal topography
and nearshore bathymetry relative to the direction of travel of the wave, and the
differences in the impacts of hurricanes and tropical cyclones in locations like the
Gulf of Mexico, where surge effects dominate, and Hawaii, where wave effects
dominate, and can inundate land to heights of 9 m (e.g. Cheung et al. 2003).
For coastal forecasting applications, the modelling approach which is used
needs to be tailored to the types of flooding mechanisms which are most important
for the locations at which forecasts are required. Some examples of possible loca-
tions for Forecasting Points (sometimes called Coastal Cells or Units) include:
● Flood Warning Areas (to assist with issuing warnings)
● High Risk Locations (ports, harbours, refineries, coastal transport routes etc.)
● Tidal Barriers (to assist with structure operations; see Chapter 8)
● Tide Gauges (for real time evaluation and updating of forecasts)
In addition to the choice of modelling approach, some other factors to consider for
a real time application (see Chapter 5) include:
● Forecasting Requirements – the intended use of the forecasts
● Data Availability – the availability of data and other information in real time
● Forecasting System – the system(s) on which models will be operated
● Performance Targets – for model run time, accuracy, and other considerations
Forecasting system and performance considerations are discussed in Chapter 5,
whilst the main sources of real time coastal data are described in Chapter 2.
A typical coastal forecasting requirement might be to estimate tidal (still water)
levels plus the additional impacts of surge and wave action. Wave overtopping esti-
mates may also be required. Some potential requirements (e.g. Environment
Agency 2004) are to inform decisions related to the:
● Areas most likely to flood
● Defence lengths most likely to breach
● Number of people in danger
● Vulnerability of the endangered people
● Extent of likely damage to property
● Extent of impact of individual failed defences on the number of people in danger
● Extent of impact of individual failed defences on the likely extent of damage to
property
● The most beneficial areas to target emergency resources
7.1 Model Design Issues 151
152 7 Coasts
The complexity of the modelling approach used will depend on the precise require-
ments in each application.
As with many other types of modelling, the design of a coastal flood forecasting
model is often a compromise, and the overall approach may also be constrained by
non-technical factors such as costs and system limitations. For example, the level
of risk (e.g. the number of properties) can be a major factor in deciding on the
complexity of approach. These points are discussed further in Chapter 11 when
considering the economic benefits of flood warning schemes.
Coastal forecasting models can be considered to fall into two main categories:
● Process Based models – which represent to varying degrees the main physical
processes which determine tidal levels, surge, and wave action (sometimes
called physically based models)
● Data Based models – which attempt to capture the main features of the response,
but do not represent physical processes directly (e.g. artificial neural networks)
Chapter 6 includes some additional discussion of the relative merits of these approaches
in the context of rainfall runoff modeling. Threshold based methods are also widely
used in coastal flood warning applications and are described in Chapter 3.
In selecting appropriate coastal flood forecasting techniques, it is also helpful to
introduce the following idealisation of the coastal flooding process (Environment
Agency 2004):
Sources
● Offshore Zone – tides, surges, wave generation and the interaction of waves with
each other
● Nearshore Zone – water levels and shallow water effects such as shoaling, depth
refraction, interaction with currents and depth induced wave breaking
Pathways
● Shoreline Response Zone – surf zone/beach response, wave structure interac-
tion, overtopping, overflowing and breaching
● Flood Inundation Zone – flow of flood water over the flood plain area
The boundaries between these zones are indicative, and will vary between different
forecasting situations.
In many cases the starting point for model development will be an offshore fore-
casting model for the open ocean of the type operated by many national meteorological
services or coastal observatories. Models of this type usually provide forecasts on a
gridded basis for which, depending on model resolution, the closest nodes may be
some distance from the Forecasting Point(s) of interest. Additional models may then
be required to represent the details of nearshore and shoreline coastal processes, and
these sub-models will often be site specific, requiring a separate set of calibration
factors, and possibly alternate types of model, for each location. The overall set of
models form an integrated coastal flood forecasting system, in which each model
output acts as the input to the next model (or models) in the chain.
With modern computing power, it is feasible to operate systems of this type in
real time for large numbers of coastal Forecasting Points, although inevitably
some approximations may be required if model run times are too slow to be of
operational use for flood warning applications. Empirical and data-based tech-
niques offer one approach to improving model run times. The development of
integrated coastal flood forecasting models, combining offshore, nearshore and
shoreline components, is an active area of research with both deterministic (e.g.
Cheung et al. 2003) and probabilistic (e.g. Tozer et al. 2007) methods under
development.
There are also many international initiatives in the area of ocean observation,
modelling and operational response, and Table 7.1 gives several examples, of which
the longest established is the World Meteorological Organisation Tropical Cyclone
Programme (see Box 7.1).
Some active areas for research and development include:
● Ensemble/probabilistic forecasting – assessing the uncertainty in surge and
wave components, including transformations from offshore to nearshore and
wave overtopping, and using this information for improved operational deci-
sion making
● Data assimilation – making use of real time observations from the ocean (buoys,
gauges, ferries, offshore platforms etc.), the atmosphere, and satellites to
improve model initial and boundary conditions
● Performance monitoring – developing improved ways to assess and compare the
outputs from different models
Chapter 5 discusses the general principles of these techniques in more detail, whilst
the following sections include several examples of practical applications in coastal
flood forecasting.
Table 7.1 Some international programmes in coastal modelling and forecasting
Programme Scope or objective Example references
TCP – WMO Tropical
Cyclone Programme
To assist member states in tropical
cyclone observation, forecasting
and response
Holland (2007)
JCOMM – IOC/WMO Joint
Technical Commission for
Oceanography and Marine
Meteorology
Promoting appropriate technical
standards and procedures for a
fully integrated marine observ-
ing, data management and serv-
ices system (open ocean)
JCOMM (2007)
WMO (1998)
GODAE – Global Ocean Data
Assimilation Experiment
Marine observation, data assimila-
tion, forecasting, performance
monitoring
http://www.godae.org/
GOOS – IOC, UNEP, WMO,
ICSU Global Ocean
Observing System
Observations, modelling and analy-
sis of marine and ocean variables
to support operational ocean
services worldwide (coastal
regions) e.g. EUROGOOS and
NOOS in Europe
http://www.ioc-goos.org/
7.1 Model Design Issues 153
154 7 Coasts
Box 7.1 The WMO Tropical Cyclone Programme
The WMO Tropical Cyclone Programme (Fig. 7.1) is part of the World
Weather Watch Applications Department, and aims to encourage and assist
the members of WMO to:
● Provide reliable forecasts of tropical cyclone tracks and intensity, and related
forecasts of strong winds, quantitative forecasts or timely assessments of
heavy rainfall, quantitative forecasts and simulation of storm surges, along
with timely warnings covering all tropical cyclone-prone areas
● Provide forecasts of floods associated with tropical cyclones
● Promote awareness to warnings and carry out activities at the interface
between the warning systems and the users of warnings, including public
information, education and awareness
● Provide the required basic meteorological and hydrological data and advice to
support hazard assessment and risk evaluation of tropical cyclone disasters
● Establish national disaster preparedness and prevention measures
Fig. 7.1 Structure of the Tropical Cyclone Programme (Reproduced from Twenty Years
of Progress and Achievement of the WMO Tropical Cyclone Programme (1980–1999),
courtesy of WMO)
(continued)
Box 7.1 (continued)
The Programme was established in 1980, and works closely with regional
and international disaster relief organisations including the United Nations
Economic and Social Commission for Asia and the Pacific (ESCAP), UN
ISDR (United Nations International Strategy on Disaster Reduction), UNEP
(United Nations Environment Programme), UNDP (United Nations
Development Programme) and the International Federation of Red Cross and
Red Crescent Societies. Activities of the Programme are implemented through
coordination with six Regional Specialised Meteorological Centres (Fig. 7.2)
and five Regional Tropical Cyclone Warning Centres which cover the follow-
ing ocean basins:
● Southwest Indian Ocean
● North Atlantic Ocean, Caribbean Sea and Gulf of Mexico
● Eastern North Pacific
● South Pacific and southeast Indian Ocean
● Bay of Bengal and Arabian Sea
● Western North Pacific and the South China Sea
The main activities within the Programme include sharing of best practice
and training in operational meteorology and hydrological techniques, and
initiatives to encourage member organisations to assess risks from tropical
cyclones, and to establish structural and non-structural measures to reduce
property damage and loss of life to a minimum. The Programme also facili-
tates the transfer of technology between member states, including satellite
reception, weather radar, telecommunication, data processing and monitor-
ing equipment, and atmospheric, surge and hydrological models.
Fig. 7.2 Regional Specialised Meteorological Centres within the WMO Tropical Cyclone
Programme (Reproduced courtesy of WMO)
(continued)
7.1 Model Design Issues 155
156 7 Coasts
7.2 Process-Based Models
7.2.1 Astronomical Tide Prediction
Of the various factors which influence coastal flooding, astronomical tide levels are
in principle the easiest to estimate, although in practice estimates are subject to
many uncertainties. Gravitational forces from the sun and moon produce a twice-
daily tidal cycle, with the highest and lowest values (spring and neap tides) occur-
ring twice each (synodic) month linked to the orbit of the moon around the earth,
with annual extremes linked to the orbit of the earth around the sun. Superimposed
on this pattern are other effects arising from variations in the relative orbits of the
sun, the moon and the earth. The longest period of motion which is usually consid-
ered is an 18.61 year cycle arising from variations in the moon’s orbit.
Tide prediction methods use wave theory in which the twice-daily cycle is combined
with other harmonics arising from the perturbations in the orbits of the sun, moon and
earth. More than 30–50 harmonics can be used, although a point is reached at which the
incremental improvements in accuracy are negated by other influences. For example, the
tidal response can be modified in estuaries, where the reflected component and depth
effects may affect the timing and amplitude of the response considerably, depending on
the shape, size and bed profile of the estuary, in some cases leading to a tidal bore. More
generally, factors which can affect the response (Hicks 2006) include:
● The restrictive depths of the oceans not allowing the generated tidal wave to be
in equilibrium with the rotation of the earth
● Irregular ocean depths over which the waves must travel
● Reflections and interactions of the waves from irregularly shaped continents
● Bottom friction
● Turbulence
● Viscosity of the water
Box 7.1 (continued)
Operational Tropical Cyclone Plans/Manuals have also been prepared by
each region covering topics which include station duties, addresses, telephone
and other communication numbers, communication procedures, terminology,
definitions, procedures, tropical cyclone naming conventions, unit conver-
sions, coordination, analysis requirements, radar and satellite observations
and dissemination, aircraft reconnaissance (where applicable), and wording
of warnings (Holland 2007).
Note: Any text/material regarding TCP/WMO does not imply the expres-
sion/endorsement of any opinion whatsoever on the WMO Secretariat con-
cerning the legal status of any country, territory, city or area or of its authorities,
or concerning the delimitation of its frontiers or boundaries.
Given these uncertainties, a semi-empirical technique called harmonic analysis is
often used, and is the process of identifying site-specific coefficients (amplitude,
phase etc.) from observed tidal records. For maximum accuracy, a record of at least
the length of the longest period (18.61 years) is required. Future values can then be
estimated by recombining the harmonics. Hence, although process-based, in practice
the methods used are strongly data-based in implementation, and purely data-based
techniques, such as artificial neural networks, are also sometimes used (see later).
The accurate determination of gauge datums (or benchmarks) is also an important
component in this type of analysis, with long term monitoring also required to
account for changes in land levels over time.
7.2.2 Surge Forecasting
Surge forecasting, by contrast, is often performed using hydrodynamic models for
the response of the oceans to atmospheric influences. Astronomical tide and wave
estimation components are also often included in the analysis. Ocean models may
also serve a range of purposes other than coastal flood forecasting, including
marine forecasting (e.g. for shipping and oil platform operators), and as part of the
ocean-atmosphere component within Numerical Weather Prediction models.
Models of this type can be global in extent (albeit sometimes at a coarse resolu-
tion), or cover smaller areas in a greater level of detail (e.g. the continental shelf, or
specific coastlines). Global scale models are typically operated by national meteor-
ological services, whilst limited extent versions are often developed for local fore-
casting applications by coastal observatories and other organisations. A widely
used approach in coastal flood forecasting is to take a real time feed of offshore
forecasts from a global or continental scale model and then to develop additional
models or empirical relationships to translate these results to the points of interest,
possibly including more detail on bathymetry and other local factors to improve the
model accuracy. However, in some situations, the offshore forecasts alone may be
sufficient for the application, with no need for additional model development.
Table 7.2 provides some examples of regional or local hydrodynamic models,
and Box 7.2 describes the approach used in the United Kingdom’s Surge Tide
Forecasting Service:
Hydrodynamic models typically use a grid-based approach, in which the grid
extent or domain extends along the entire coastal region or reach under considera-
tion and into the open ocean (e.g. to the edge of the continental shelf). Typical hori-
zontal grid lengths might be of the order 10–100 km in the open ocean, and possibly
of higher resolution (e.g. 0.1–10 km) for specific coastal reaches or features. In
three dimensional (3D) models, the vertical division of layers may be fixed, or may
be varied according to the dominant processes under consideration; for example,
Chassignet et al. (2006) describe an ocean model which uses density tracking layers
in the deep ocean, fixed depth (or pressure) coordinates near the surface within the
mixed layer, and terrain following coordinates in shallow coastal regions.
7.2 Process-Based Models 157
158 7 Coasts
Table 7.2 Examples of operational surge, tide and wave forecasting systems
Country Operator Reference
Australia Bureau of Meteorology, CSIRO Brassington et al. (2005)
Hong Kong Hong Kong Observatory Lam et al. (2005)
The Netherlands Royal Netherlands Meteorological Institute
(KNMI) and National Institute for Coastal
and Marine Management (RIKZ)
Verlaan et al. (2005)
United Kingdom Met Office, Proudman Oceanographic
Laboratory
Flather (2000)
USA National Weather Service (Hydrometeorological
Prediction Centre, National Hurricane
Centre, Ocean Prediction Centre, Honolulu
Weather Forecast Office); NOAA Centre for
Operational Oceanographic Products and
Services
Chassignet et al. (2006)
Jelesnianski et al. (1992),
Berg et al. (2007)
Box 7.2 Storm Tide Forecasting Service, United Kingdom
The Storm Tide Forecasting Service (STFS) is operated jointly by the UK Met
Office and the Proudman Oceanographic Laboratory. The service provides
forecasts of surge magnitudes around the coastline of the UK (e.g. Fig. 7.3) and
has been in operation since 1978 (Flather 2000). Forecasts are provided to a
lead time of 36 hours ahead, with a 6 hour hindcast period to initialise the
model run.
The current model (CS3X) was developed by the Proudman Oceanographic
Laboratory and covers the entire continental shelf from the west of Ireland
and into the North Sea and the Bay of Biscay and English Channel. The
model extent was increased in 2006 from an earlier version (CS3) to better
account for surge generating conditions over the westernmost parts of the
continental shelf (i.e. the Bay of Biscay and the Rockall shelf).
The model uses a two-dimensional finite difference scheme with a 12 km
horizontal resolution. The model is forced with 26 tidal constituents, and
wind and atmospheric pressure fields from the North Atlantic Extended
(NAE) component of the Met Office’s Unified Numerical Weather Prediction
model. The lower boundary is provided by detailed bathymetry for the sea
bed and the coastal margins. Wave forecasts are obtained from a 2D spectral
model, again running on a 12 km grid, and the output from the surge model
is used as an input to this model to allow for wave-current interactions. The
model also allows for drying and inundation of inter-tidal areas.
The model runs four times per day and, for each forecast, two model runs are
used; a tide-only version using standard atmospheric pressure and no wind at
(continued)
Box 7.2 (continued)
the ocean boundary, and a version with the atmospheric forcing included.
The surge is then estimated from the difference between the outputs from
these two model runs. Total water levels at specific monitoring sites (e.g. tide
gauges) are then estimated by adding the surge to the site-specific estimate of
tidal levels based on harmonic analysis (e.g. Fig. 7.4).
Finer mesh models have also been developed for some areas; for example,
to provide better resolution of surge-tide interactions in the strongly tidally
influenced Bristol Channel and Severn Estuary in the southwest of England.
Data assimilation techniques are also being evaluated, including 3D-Var and
Optimal Interpolation methods. A prototype 24 member ensemble surge prod-
uct is also being developed, based on multiple CS3X model runs using ensem-
ble estimates for atmospheric pressure and wind speed and direction from the
Met Office NAE model. In collaboration with the Environment Agency, which
is responsible for issuing coastal flood warnings in England and Wales, a
demonstration project is providing probabilistic coastal forecasts to a coastal
Fig. 7.3 Example of an STFS simulation of progression of a storm surge around the coast
of the UK (National Tidal and Sea Level Facility (NTSLF), Proudman Oceanographic
Laboratory, http://www.pol.ac.uk/ntslf)
(continued)
7.2 Process-Based Models 159
160 7 Coasts
Box 7.2 (continued)
location, combining ensemble and Monte Carlo techniques to account for
uncertainties in surge, wave transformations, beach profiles, sea defence con-
dition, wave overtopping, bathymetry, and other factors.
Fig. 7.4 Surge model residuals and astronomical tidal elevation predictions for Sheerness
for a 2.5–3 m North Sea surge event on 9 November 2007 (National Tidal and Sea Level
Facility (NTSLF), Proudman Oceanographic Laboratory, http://www.pol.ac.uk/ntslf)
Various solution schemes can be used, including finite difference, finite element,
and finite volume methods. Some approaches support use of irregular grids or cur-
vilinear grids, in addition to regular grids. Irregular grids typically require more
complex solution schemes, but have the advantage that the resolution and shape of
the grid can be tailored to situations where more detail is required (e.g. in bays and
estuaries and around islands and headlands).
Models of this type aim to solve approximations to the three dimensional mass,
momentum and energy conservation equations for fluid motion. Forecasting may
be in terms of total water levels (combing surge and tidal influences), or the surge
residual may be calculated separately and then added onto the results from har-
monic analysis at tide gauges (e.g. Horsburgh and Wilson 2007). Some processes
which can be represented include:
● Turbulence, friction, temperature and salinity effects
● Surge generation (from atmospheric pressure and wind shear)
● Energy transfer (from the ocean to the atmosphere, and vice versa)
● Freshwater inflows and river levels at fluvial/tidal boundaries
● Rotational effects (e.g. storm scale effects)
● Wave generation and transformation
● Ice formation, movement and dissipation
In practice, some simplifications must usually be made, since the details of proc-
esses may not be fully understood, and model run times would not be practicable
for a real time forecasting application. Some general types of approximation can
include:
● Distance (depth or width) averaging (e.g. shallow water equations)
● Time averaging (e.g. wave spectral analysis, turbulence effects)
● Exclusion of secondary effects (depending on the application)
● Introduction of statistical or empirical sub-grid models
Temperature and density effects may be approximated also. Another key approxima-
tion is often to reduce the number of dimensions of the model, and Table 7.3 illus-
trates some possible options if using Cartesian coordinates.
Table 7.3 Examples of applications of simplified hydrodynamic models
Form of model x y z Potential coastal forecasting applications
One-dimensional
(1D)
Yes No No Shallow narrow estuaries, offshore-onshore processes
where depth effects and alongshore processes are
insignificant.
Two-dimensional
– vertical (2DV)
Yes No Yes Deep narrow estuaries where influences from the
width of the estuary are insignificant.
Two-dimensional
– horizontal
(2-DH)
Yes Yes No Wide shallow estuaries, bays etc. where the vertical
transport is insignificant. Also, models for the
open ocean and some coastal reaches.
Three-dimensional Yes Yes Yes Detailed three dimensional modelling. Options include
barotropic and baroclinic models.
7.2 Process-Based Models 161
162 7 Coasts
In the table, the x-dimension is from offshore to onshore, the y-dimension is
alongshore, and the z-dimension is for depth (although different conventions may
be used).
The issue of model run times is of course a factor for real time applications, and
run times often increase with increases in the number of processes represented and
the dimensions of the model. Apart from selecting a simpler model, with a coarser
resolution, another solution is to run models off-line for a large number of scenar-
ios, then to derive simpler regressions, transfer functions or transformation matrices
based on these results for real time implementation. Some examples of this
approach are described in Section 7.3.
The wind and pressure conditions at the open water boundary are often taken
from Numerical Weather Prediction models of the atmosphere (see Chapter 2).
Various approaches can be used to represent the wind stress at the ocean surface,
including simple quadratic functions using a drag coefficient, and methods
allowing for the relative velocity of the airflow compared to tidal currents. The
pressure and wind fields derived from these models are usually available on a
regular grid, typically with horizontal dimensions of the order 10–100 km for
regional or global scale models, although with values of 1–10 km becoming
more common for national or local (mesoscale) models. The atmospheric and
ocean models may also be coupled, to allow for interactions between these two
components.
Also, given that large extents of ocean may need to be modelled, it is likely that
only a few grid points in the modelling domain will have real time available for
data assimilation into model outputs data (e.g. tide gauges, wave buoys, offshore
platforms, boat observations, weather stations), so further approximations need to
be made when using real time data. Chapter 5 discusses data assimilation tech-
niques further.
Due to the complexities of modelling the atmospheric component, paramet-
ric inputs are also often used for hurricanes, tropical cyclones and typhoons.
Techniques which have been developed include trajectory tracking, statistical,
and expert system approaches (e.g. Holland 2007), and the resulting wind and
pressure fields can then be used as inputs to surge and wave forecasting mod-
els. However, as the accuracy and resolution of models improves, the direct
outputs from Numerical Weather Prediction models are increasingly being
used in the modelling of tropical cyclones (e.g. Davidson et al. 2005; Sheng
et al. 2005).
Also, since surge estimates, in particular, are sensitive to the predicted track and
locations of landfall, ensemble and probabilistic techniques are also used in hurri-
cane forecasting; for example, using historical error characteristics in speed, radius,
intensity etc., or ensemble forecasts from fully dynamic atmospheric forcing models
(Box 7.3).
Box 7.3 SLOSH model, National Hurricane Centre, USA
The Sea, Lake and Overland Surges from Hurricanes (SLOSH) model
(Jelesnianski et al. 1992) is used by the National Hurricane Centre in Miami
to calculate surge magnitudes from hurricane events. The model is used in
planning mode to calculate maximum surge envelopes for input to the hazard
assessment component of Hurricane Evacuation Studies, and operationally in
the 24-hour period leading up to the estimated time of hurricane landfall.
The model solves the depth averaged, shallow water equations of motion,
typically on a curvilinear, polar grid (Fig. 7.5). The smallest grid length is typi-
cally about 250 m, although can be less, and recent developments have intro-
duced elliptic and hyperbolic grids, allowing a finer resolution near to the
shoreline (Massey et al. 2007). Approximately 40 individual models have been
developed for locations and embayments along the Gulf of Mexico, the Florida
coastline and the eastern seaboard of the USA. A number of SLOSH basins
have been developed for countries such as India, South Korea, and China. The
model resolves the effects of estuaries, bays and structures on surge propagation,
Fig. 7.5 Example of SLOSH surge height output (National Oceanic and Atmospheric
Administration/National Weather Service, http://www.nhc.noaa.gov/)
(continued)
7.2 Process-Based Models 163
164 7 Coasts
Box 7.3 (continued)
and also calculates overland flows, including wetting and drying of cells, and
the influence of barriers such as levees and sub-grid resolution channels.
To operate the model, the main inputs required are values for initial
water levels and for the forecast track of the eye of the hurricane, sea level
barometric pressure in the eye, and the radius of the maximum winds
(which is typically of the order 10–100 km). Input values are required at 6
hourly intervals for a model run period of 72 hours, whilst the main model
outputs are surge heights. In the planning mode, maximum values for each
grid cell are estimated by performing model runs assuming a range of typi-
cal hurricane categories, speeds, and paths, based on historical informa-
tion. The resulting Maximum Envelopes of Water (MEOW) and Maximum
of MEOW (MOM) maps are a valuable planning tool for emergency man-
agers. Up to 15–20,000 model runs can be required to generate each MEOW
map, and the results can be categorised by Saffir-Simpson scale, forward
speed, direction etc. as required.
In operational use, estimates for hurricane speed, track and size are obtained
from the National Hurricane Centre official advisory forecasts. Typically one
Fig. 7.6 Illustration of probabilistic hurricane storm surge product (experimental version)
(National Oceanic and Atmospheric Administration/National Weather Service, http://
www.weather.gov/mdl/psurge/)
(continued)
Box 7.3 (continued)
to two basin models are operated based on the estimated location of landfall,
although sometimes up to five may be needed for hurricanes tracking along the
coast. Analysis of historical events suggests that the accuracy in predicted surge
heights is about 20%, based on 30 years of records, with uncertainties in the
track of the hurricane being one of the main potential sources of error, the other
being post-storm observations. To provide an indication of this uncertainty,
cumulative probabilities and exceedance heights for surge are also published on
the NOAA/NWS Meteorological Development Laboratory website (Fig. 7.6).
These estimates are derived by performing multiple model runs (more than
1,000/hour) on a supercomputer, in which the storm track, speed, radius and
intensity are varied over plausible ranges based on historical errors.
Future gains in forecast accuracy are likely to arise from improved atmos-
pheric modelling of hurricane development, improved bathymetry and topog-
raphy, and the use of coupled offshore-onshore wave transformation models.
To assist with real time interpretation of outputs, an ongoing programme of
gauge improvements aims to increase the network density of tide gauges in
some locations, to harden gauges to better withstand extreme surge and winds,
and to monitor and refine datum values for gauges
Source: (National Hurricane Centre website http://www.nhc.noaa.gov/ and
Dr. Stephen Baig, personal communication)
7.2 Process-Based Models 165
7.2.3 Wave Forecasting
Surge forecasting models often include a wave modelling component, and the surge
and wave models may be coupled to allow the interactions between tidal currents
and waves to be represented.
Waves can occur at many different scales and frequencies, and can be affected
by changes in water depth, wind speed and direction, interactions with other waves,
and other factors, such as coastal features and structures. The main processes which
might be considered are wave generation and wave transformation and Table 7.4
briefly describes these factors (e.g. World Meteorological Organisation 1998;
Environment Agency 2004; Holthuijsen 2007).
The two main approaches to wave modelling are:
● Phase averaging – in which the sea state at any location is considered as a
statistical process resulting from the sum of many individual waves, whose
amplitudes, directions, phase, wavelengths and frequencies are represented
by a wave energy spectrum. Typically the energy balance is expressed in
terms of the energy inputs (from the wind field), transfer (e.g. wave-wave
166 7 Coasts
interactions) and dissipation (e.g. from wave breaking), including representa-
tion of the non-linear transfer of energy between frequencies (e.g. from wave-
wave interactions) and from nearshore effects (e.g. currents, water depth,
bottom friction). Key characteristics such as significant wave height, mean
wave period and mean wave direction can be derived by integration. Both
two-dimensional (excluding direction) and three-dimensional representations
can be used, and grid based (e.g. finite difference) or ray tracing solution
schemes.
● Phase Resolving techniques – in which deterministic hydrodynamic models are
used to model the motion of individual waves, or waves within a given spectral
band (direction and frequency), as they propagate. A spectrum of waves can be
considered by running multiple realisations of inputs through the model. Any or
all of the processes listed in Table 7.4 may be represented, although in some
cases may need to be approximated by empirical relationships.
Table 7.4 Summary of some key wave generation and transformation processes
Category Sub-process Description
Wave generation Wind shear, pressure Generation and growth of waves linked to
wind friction and pressure effects at the sea
surface
Seismic (geotechnical) Wave generation due to sub-sea factors such as
earthquakes and land slides (e.g. Tsunami)
Wave transformation Breaking (offshore) Whitecapping (overturning) of waves due to
wave growth
Breaking (onshore) The overturning of waves in shallow water due
to depth related differences in wave propa-
gation speed
Diffraction The transfer of wave energy along a wave crest
due to interactions with headlands, islands
and coastal structures (breakwaters, harbour
walls etc.)
Refraction The change in direction of wave propagation
and subsequent wave interactions due to
changes and variations in water depth (e.g.
as waves approach the shoreline) or interac-
tions with tidal currents
Reflection The reflection and interaction of waves as they
meet headlands, islands, coastal structures etc.
Set-up The onshore transfer of momentum by waves
leading to increased water levels in the surf
zone
Shoaling The increase in wave height, and decrease in
wave length, as waves propagate into shal-
low water (or encounter currents travelling in
the opposite direction to wave propagation)
Phase averaging techniques are widely used in operational wave and surge fore-
casting models. A series of nested models may be used, at global, regional and
local scales, with each model providing the boundary conditions to the next; for
example, swell waves generated in the global model might be propagated into the
regional model. Due to the smaller spatial extent, local models typically have a
higher grid resolution, and it may also be possible to increase the run frequency
and the number of processes represented whilst retaining a satisfactory run time
for flood warning applications. Some well known examples of the phase averaging
approach which are used by national meteorological and coastal services include
the WAM family of models (Wave Model; Komen et al. 1994) and the
WAVEWATCH III (Tolman 1999) and SWAN models (Simulating Waves
Nearshore; Holthuijsen 2007). Many other types of model are also available
through research projects or commercially.
Phase resolving techniques provide more detailed information on changes in
wave height and direction, particularly in areas of complex bathymetry or interactions
with coastal structures. However, they are more computationally intensive and are
generally not suitable for modelling large regions, or for direct use in real time
modelling, unless wave directions and frequencies fall within a narrow band (as on
some parts of the western coast of the USA, for example). Kirby et al. (2005) and
Shi et al. (2001) describe examples of this approach.
7.2.4 Shoreline Processes
Shoreline models consider the processes of wave run up on a beach and overtopping
of defences and natural features (e.g. dunes). Additional factors, such as breaches,
may also occur, although these are discussed in Chapter 8.
Overtopping (or splash-over) can be expressed in terms of the mean overtopping
discharge or peak volumes. Peak volumes apply to the maximum likely value in a
single wave, whilst discharge values may be calculated over a given period (e.g.
hourly) or number of waves (e.g. 1,000 waves). Volumes are important when considering
the risk from a single wave (e.g. the loading on a building), whilst discharge values
can be used to estimate the likely depth and extent of inundation. Models are usu-
ally calibrated using estimates of overtopping rates from video images, photographs
and site surveys.
The rate or volume of overtopping depends on the wave propagation process at
the shoreline and sea defences, and is highly sensitive to both of these factors. For
off-line design applications, in addition to empirical approaches and laboratory
testing, a phase resolving approach is sometimes used (see previous section), in
which the motion of waves is modelled using a 1D, 2D or 3D hydrodynamic model
describing the conservation of mass and momentum. Atmospheric and free surface
influences may also be included and non-linear shallow water (depth averaging)
approximations may also be made.
7.2 Process-Based Models 167
168 7 Coasts
Models of this type operate by propagating waves of a given height, period and
shape (or significant wave height, frequency and spectral shape) to generate a time
series of wave elevations at the seaward boundary of the defence, from which over-
topping discharges and volumes can be estimated (Environment Agency 2004). The
input data requirements can include information on beach profiles, the geometry
and condition of sea defences (such as those illustrated in Fig. 7.7), and of struc-
tures in the shoreline region (e.g. breakwaters, groynes). Some examples of models
of this type are described by Hubbard and Dodd (2002) and Causon et al. (2000).
There are of course many potential uncertainties in this type of approach,
particularly for the interactions between waves and coastal structures, and
physical model tests may be required to estimate the required calibration factors.
Depth averaged models tend to run faster, so perhaps have the greatest potential
for real time use, but have limitations when vertical fluid motions are important
(e.g. for steep or vertical defences). Ensemble and probabilistic approaches
provide one possible route to assessing the uncertainty in model outputs, whilst
data based techniques of the type described in the next section provide an alter-
native approach.
More generally, in wave overtopping research, the focus is to extend the range
of methods to more complex situations, such as steeper sea walls, different types of
armouring on structures, and a wider choice of structure geometries, as well as
developing artificial neural network and process-based approaches to estimation of
overtopping, and the risk of defence breaches.
Fig. 7.7 Examples of sea defences (Kevin Sene, Springer)
In addition to the uncertainties in the approach, model run times are another
factor which to date has prevented real time applications for operational coastal
forecasting. However, models of this type have been used off-line to calibrate sim-
pler data-based methods for operational use, as described in the next section.
7.3 Data-Based Methods
Data-Based methods aim to capture the main features of the coastal response, but
do not attempt to model physical processes directly. Methods include transfer func-
tions, artificial neural networks, and other artificial intelligence techniques, and are
widely used in other fields, including river flow forecasting (see Chapter 6), industry,
finance and transport. Due to their fast run times, and ability to represent non-linear
processes, models of this type are also candidates for use as emulators for more
complex models which are too time consuming to run in real time, particularly for
ensemble forecasting, as described in Chapter 5.
In coastal forecasting applications, artificial neural networks are perhaps the
most widely used approach, although some other techniques which have been
applied include:
● Bayesian techniques – application of fuzzy Bayesian modelling techniques to
estimating the propagation of surge along the East Coast of the United Kingdom
(Randon et al. 2007)
● Chaos theory – the application of linear and non-linear time series analysis tech-
niques to current and surge forecasting, with applications in the Netherlands, for
example (Solomatine et al. 2001)
● Transfer functions – application of the methods described in Chapter 6 to water
level modelling in tidal zones; for example, for a tidal (estuarine) river reach in
Scotland (Lees et al. 1994)
Although the term data-based is usually used to describe artificial intelligence and
related techniques, it is also convenient to discuss several simpler empirical or ana-
lytical techniques in this section, including transformation matrices and wave over-
topping formulae.
7.3.1 Artificial Neural Networks
Artificial Neural Networks were originally devised as a tool for helping to under-
stand the human brain, but have since developed into a powerful technique for
solving complex non-linear multivariate problems.
A network is usually constructed from individual neurons, whose inputs are
adjusted by weighting factors, and are transformed by a function (a so-called acti-
vation or transfer function) into the output. Networks are often constructed in the
7.3 Data-Based Methods 169
170 7 Coasts
form of an input layer, an output layer, and one or more hidden layers which join
the outputs to the inputs, as shown by the example in Fig. 7.8.
Networks can adapt, or be ‘trained’, by adding or removing neurons, and changing
the strength of the interactions between neurons (in this case, via the weighting fac-
tors). Training is performed with reference to an assessment (or optimisation) function,
often chosen to be the mean square error between input and output values.
Apart from the choice of assessment function, some distinguishing features
between different networks can include the number of neurons and layers, and
the choice of activation functions and training techniques. Given the potentially
large number of configuration options, number of neurons, weighting factors,
activation functions and other variables, much research has been performed on
training techniques, including development of Bayesian methods, genetic algo-
rithms, and stochastic approaches, such as simulated annealing. A compromise
is also needed between having too complex a network (operating slowly, and
over-parameterised), and over simplifying the network (potentially losing use-
ful information). Artificial neural networks have been developed and tested for
a wide range of coastal forecasting and related applications and Table 7.5 shows
a few examples.
In coastal forecasting applications, artificial neural networks can be trained
using a range of inputs (depending on the application), including information on
sea defence geometry and condition, historic databases of wave overtopping rates,
real time measurements of water levels, wind speeds, wind directions, river levels,
Input 1
Input 4
Output
Weight 1Weight 2
Weight 4
Function
InputLayer
HiddenLayer(s)
OutputLayer
Weight 3
Input 2
Input 3
Fig. 7.8 Example of a three layer artificial neural network
significant wave heights and periods, and barometric pressure, and forecasts of
wind speeds, wind directions and atmospheric pressure from Numerical Weather
Prediction models. Models are typically optimised for one or more forecast lead
times, and evaluated against a range of performance statistics, including statistical
and categorical measures, as described in Chapter 5.
7.3.2 Other Techniques
Despite advances in computing power, it can be impractical to operate some types
of process-based coastal forecasting models in real time. This has led to the devel-
opment of a range of techniques which use the outputs from off-line simulations to
guide the development of simpler forecasting tools for real time use. Also, for some
situations (e.g. sea defence breaches), it may be desirable to run multiple scenarios
off-line for later use during a flood event as required.
For example, coastal forecasting models are often run at time intervals linked to the
daily tidal cycle (e.g. every 6 or 12 hours) which, after allowing for data gathering, pre-
processing, post-processing and decision times (see Chapter 5), may give a much
smaller time window for the computational aspect of the model run, including any
ensemble analyses (if used). Also, some types of coastal flood event (e.g. surge) can
develop in shorter timescales, or may have little tidal influence, in which case the poten-
tial lead time (and time available for model runs) may only be 1–2 hours or less.
Various approaches can be used for capturing the essential features of process-
based model runs and/or historical observations, including nomograms, look-up
tables, carpet plots, multiple regressions, and decision support tools (e.g. World
Meteorological Organisation 1998).
One example is the operational coastal flood forecasting system used in parts of
the United Kingdom (e.g. Environment Agency 2004; Hu and Wotherspoon 2007).
Conditions at the shoreline are estimated using transformation matrices derived
from off-line modelling using an offshore-nearshore wave transformation model
and a shoreline wave-overtopping model. For each coastal Forecasting Point (or
Table 7.5 Examples of research and other applications of artificial neural networks to coastal
flood forecasting and related problems
Type Location Reference
Tide level predictions Western Australia Makarynskyy et al. (2004)
Surge forecasting Gulf of Mexico, USA Patrick et al. (2002),
Prouty et al. (2005)
Wave forecasting Portugal Makarynskyy et al. (2005)
Estuary forecasting Texas, USA Tissot et al. (2003),
Long Island, NY, USA Huang and Murray (2003)
Sea wall overtopping United Kingdom Wedge et al. (2005)
7.3 Data-Based Methods 171
172 7 Coasts
cell), the model is run using a range of scenarios for the offshore values of the
directional wave spectra (mean wave direction, wave height, and wave period), and
wind field (speed and direction). Site specific transformation coefficients are
derived for each location.
In real time operation, the resulting matrices are used to transform the offshore
wave and wind forecasts from the UK’s Storm Tide Forecasting Service (see Box
7.2) to the shoreline. As with all modelling systems, the forecast performance needs
to be monitored regularly, and after each major event, and the coefficients updated
if there are reductions in performance, or changes to the physical characteristics at
each Forecasting Point (e.g. erosion following a coastal flood event).
This type of approach has also been used for forecasting the coastal impacts
of tropical cyclones and hurricanes. For example, Holland (2007) and others
describe the use of nomograms for estimating the likely surge impacts of tropi-
cal cyclones based on hydrodynamic model simulations for a range of idealised
coastal basins and hypothetical cyclone conditions. The calculations exclude
small islands and assume a regular unbroken mildly curved coastline with no
inland transfer of water (e.g. steeply rising terrain). Tropical cyclones are
assumed to travel in straight lines at a constant angle of attack to the coast and
with constant parameters (e.g. speed). Surge estimates are derived for a range
of possible tracks and parameters, and basin slopes and coastal depths.
Nomograms based on these simulations provide estimates of surge for a range
of possible input parameters, such as radius of maximum winds and ambient
pressure drop. Correction factors may be included for factors such as the angle
of attack to the coast, and shoaling effects.
Another technique which is described is to prepare an atlas of pre-computed
surges based on the historical characteristics of tropical cyclones which have
affected a coastal basin. Cyclones are categorised by preferred track directions,
intensities, and sizes, perhaps assuming that the speed, pressure and size remain
constant. Values are computed for a range of likely tracks, pressures, sizes and
speeds. This approach can be taken a step further by calculating the Maximum
Envelopes of Water (MEOW) to give an indication of the likely worst-case surge
for a given set of cyclone profiles and conditions. Computer based versions can
provide an interactive tool which, near the time of landfall, can be used to select the
most likely range of scenarios, with the MEOW distribution constructed from a
stored set of values computed earlier. This method is used in the USA for example,
as a complement to other more sophisticated approaches (see Box 7.3).
Restrictions on model run times, and uncertainties in model formulation, have
also led to simpler regression and other approaches being adopted for estimating
wave overtopping. For example, some methods provide an estimate of likely maxi-
mum wave heights at the toe of a structure based on functions incorporating key
descriptors for the characteristics of the structure, and offshore parameters such as
mean significant wave height and period, together with information on the sea bed
slope and the water depth. The information required on structures can include crest
height, geometry for the seaward side, surface roughness etc., and model parame-
ters may be derived from physical model tests or hydrodynamic modelling.
Although these methods were usually developed for design applications (e.g.
to estimate optimum values of sea defence crest levels, slope angle, required
roughness), some of these approaches can also be used operationally to estimate
wave overtopping volumes and discharges at a sea defence, based on offshore or
nearshore forecasts of key wave parameters. For example, Tozer et al. (2007)
describe an application for a rail operator in which empirical overtopping formu-
lae are combined with a wave transformation model and offshore forecasts to
provide forecasts to coastal rail lines with up to a 36 hour lead time, together with
a prototype flood warning service for a coastal development. In the latter case,
four hazard levels were proposed as follows:
● Hazard level 0 – Safe use for all areas behind seawall.
● Hazard level 1 – Promenade immediately behind seawall shut to public.
● Hazard level 2 – Promenade behind seawall is unsafe for staff. Storm gates in
secondary defences should be shut.
● Hazard level 3 – Public to be excluded from areas behind secondary defences.
● Hazard level 4 – Areas behind secondary may be unsafe. All protection devices
should be secured.
The categories correspond to different wave overtopping rates in the range 0.03 l
per second per m of sea wall, to more than 1.0 l per second per m.
7.3 Data-Based Methods 173
Chapter 8Selected Applications
Previous chapters have described the main techniques which are used in river and
coastal flood forecasting, whilst this chapter presents a selection of forecasting
applications. These include integrated catchment forecasting models, and forecast-
ing techniques for flash floods, snowmelt, ice jams, dams and reservoirs, control
structures, urban drainage flooding, and geotechnical risks, such as Tsunami, debris
flows, and dam break. The chapter also includes several examples in fields which
are closely related to flood forecasting, such as the real time control and optimisa-
tion of reservoir and urban drainage systems. Some themes which run throughout
the chapter are the use of ensemble and probabilistic techniques to provide information
on risk and uncertainty, and the use of process-based, conceptual and data-based
modelling approaches.
8.1 Integrated Catchment Models
8.1.1 Introduction
Real time integrated catchment models seek to model whole catchments using a
range of rainfall runoff and river flow routing components. Sub-models for additional
features may also be included as required, such as dams, control structures, and flood
defence systems. Models may be used for a range of applications, including:
● Flood forecasting – forecasting of flood flows at various Forecasting Points in
the catchment
● Water resources – real time modelling of river and reservoir conditions, and of
the impacts of abstractions and discharges for water supply, agriculture and
industry
● Navigation – real time forecasting of water levels and velocities to assist ship-
ping authorities and recreational boat users
● Pollution control – real time forecasting of the passage of pollutants through a
catchment following a spillage, or following a major rainfall event; for example,
linked to bathing water quality at beaches
K. Sene, Flood Warning, Forecasting and Emergency Response, 175
© Springer Science + Business Media B.V. 2008
176 8 Selected Applications
The focus of this discussion is on flood forecasting, although the techniques
described are also relevant to other applications.
In addition to the issues of real time data availability, run time, cost, systems,
and model performance which have been discussed in previous chapters, some
additional factors which need to be considered in designing a real time integrated
catchment model include:
● The key processes which influence river flows and levels in the river network
● The forecasting requirements for floods, water resources, and other
applications
● Whether the model should be optimised for the full flow range, or only parts of
the flow regime (e.g. flood flows)
● Opportunities for simplifying the model to improve run times and stability
The modelling strategy also needs to consider whether a process-based, conceptual
or data-based approach is to be used (or some combination of these approaches),
whether the model will run all year round (in both low and high flow periods), or
just as required (e.g. when there is a flood event), opportunities for real time updat-
ing of outputs, and the calibration criteria to be used (and whether a multi-criteria
approach needs to be considered).
Integrated catchment models could potentially make use of process-based, concep-
tual and data-based rainfall runoff and flow routing models in various combinations,
perhaps also combined with a coastal forecasting model for the lower tidal boundary
condition. Some types of process-based rainfall runoff models may also include flow
routing components, and can be viewed as a type of integrated catchment model.
However, although modern forecasting systems allow a range of model types to
be operated together, the usual approach is to combine models of similar type, reso-
lution and complexity. For example, there may be little advantage in developing a
sophisticated hydrodynamic model if inflows are derived from a correlation of
doubtful accuracy. Also, simpler models can have a role, both as a back up to more
complex approaches, and to extend the functionality of the underlying models; for
example, by relating levels at Forecasting Points to levels at key locations nearby
which are not covered by the main model.
As an example of an integrated catchment approach, Huband and Sene (2005)
describe a real time model developed for the Environment Agency for a catchment
in Eastern England with many complicating influences on flows, particularly in the
lower reaches, including pumped and gravity fed discharges and abstractions,
offline storage reservoirs (e.g. ‘washlands’), tidal influences, and manual and auto-
matic flow control structures. In addition to providing flood forecasts, the model
was developed for a range of other real time applications including:
● Management of raw water transfers
● Management of river support
● Management of drought and license control
● Management of pollution incidents
● Management of navigation and of strong stream warnings
The original version of the model included 55 lumped conceptual rainfall runoff
models, operating from raingauge and weather radar data, and rainfall forecasts,
feeding over 400 km of hydrodynamic network, with almost 500 structures includ-
ing flood storage reservoirs, siphons, pumps and complex gates, bridges, weirs
and culverts, with operations at automatic structures represented by logical rules.
River levels and flows were simulated in real time at 52 Forecasting Points in the
catchment.
8.1.2 Modelling Approach
When developing an integrated catchment model, some options for model develop-
ment include:
● Develop an integrated model specifically for the application of interest, with the
minimum complexity needed to meet the requirement
● Develop a ‘best possible’ model for the catchment, from which other more spe-
cialised models can be derived as required
The latter approach is adopted by the Environment Agency for some catchments
in the UK, for example (Huband and Sene 2005), with the best possible model
called a ‘Parent Model’. In this approach, an overall model is constructed which
can either be used unchanged for a range of applications, or from which simpli-
fied/optimised models can be developed for other applications (such as flood
forecasting), or for studies of specific parts of the catchment. The advantage of
the ‘Parent Model’ approach is that, at any one time, there is only one ‘best
attempt’ model for the catchment, which can form the basis for all ‘Child Models’
used for specific applications. This simplifies the process of maintaining and
improving the model, and makes it easier to document and audit the history of
model development. However, the initial development time and costs can be
higher than in the more classical approach of developing a range of models for
different applications, although overall costs may be lower in the long term. The
usual issues of data availability and type of forecasting system described in
Chapter 5 also need to be considered.
Some possibilities for model configuration include the following options:
● Single model – one model for the whole catchment
● Multiple models – two or more models running in sequence or in parallel for all
or part of the catchment
● Nested models – single or multiple models with more complex models nested
within them for specific forecasting issues
Table 8.1 presents some of the strengths and limitations of these approaches. The
choice of approach will depend on the particular modelling and forecasting issues
for each catchment.
8.1 Integrated Catchment Models 177
178 8 Selected Applications
8.1.3 Ungauged Inflows
Another issue which may require particular consideration is the representation of
ungauged inflows to river reaches. These are flows for sub catchments, or areas
adjacent to river reaches, which are not monitored by real time telemetry, or only
partially monitored (e.g. if a gauge is some distance upstream from the confluence).
Some sources of ungauged inflows can include:
● Tributary inflows
● Surface runoff along a river reach
● Gravity fed abstractions or discharges
● Pumped inflows or outflows
● Spillage or seepage at river banks or flood defences
● Groundwater inflows or recharge
Table 8.1 Some strengths and limitations of various model configuration options for hydrody-
namic and flow routing models (Huband and Sene 2005)
Method Strengths Limitations
Single model Simple to manage future develop-
ment of the model
Possible node limitations for
hydraulic models
No complex decisions to make
about which forecast to use
Run times can be long
No discrepancies between forecasts
for the same location
Model must be built and tested in one
operation
No boundaries between models to
define
Updating must be internal to the
model
Multiple models Models can be run independently
(not requiring the full model run)
Boundary conditions must be defined
explicitly at model joins
Parallel processing of model runs
is possible giving faster overall
run times
A staged delivery of models is
possible
The splits between models allow
independent updating algorithms
to be used
Downstream information may not be
correctly transferred to models
upstream e.g. backwater influ-
ences, tidal effects
An overall catchment water balance
may not be preserved
Greater complexity for Duty Officers
during an event (if the model run
sequence is not automated)
Nested models More detailed model outputs can be
provided in areas of interest
More difficult to maintain and update
models
Nested components can be used
‘on demand’ so do not normally
impact upon overall model run
times
Greater complexity for Duty Officers
during an event
Conflicting forecasts possible at the
same location
The physical representation of the
catchment may no longer apply
(e.g. model reaches may need to
overlap)
An alternative simpler model will be
available as a fall back in case of
model failure
Parallel processing is possible
giving faster run times
The need to represent these contributions will depend on the likely magnitude and
timing of inflows or losses compared to flows in the main river channel. Also,
although, in general, it is desirable to develop a model which performs well
throughout the flow range, in practice the emphasis of development may be on the
areas of particular interest (e.g. high flows, for flood forecasting, or low flows, for
drought forecasting), and the relative importance of the inflow components will
vary between applications.
Figure 8.1 shows an example of the upper reaches of a catchment in which there
are five sub-catchments with telemetry sites (shown shaded), one ungauged sub-
catchment (unit A), and five lateral inflow areas (units B–F). Ideally, the contribu-
tions from these areas require modelling individually, although one simplifying
option might be to model various combinations of units B–F, whilst being careful
not to introduce unwanted transient and other effects from introducing unrealisti-
cally high flows at some locations along the river network.
Note that, even for the gauged catchments, there may be areas downstream of
instruments which also need to be modelled. Some reasons for not installing river
gauges at confluences can include the requirement to position the gauge in a loca-
tion upstream of any backwater influences from the main river, and a range of other
factors, such as lack of a suitable site or telemetry connections. If backwater effects
are a factor then, when a hydrodynamic approach is being used, one option is to
extend the hydrodynamic component of the model into the tributaries affected.
Table 8.2 shows some typical approaches to modelling ungauged and lateral
inflow components.
For tributary inflows, if scaling or correlation techniques are used, the main cri-
teria which determine flood flows are often the catchment area and some measure
of rainfall (for example, mean annual rainfall), although many other factors might
A
C
D E
F
B
Fig. 8.1 Illustration of ungauged and lateral flow modelling issues (circles indicate telemetered
river gauges) based on a catchment in NE England (not to scale)
8.1 Integrated Catchment Models 179
180 8 Selected Applications
also need to be considered, including soil type, mean slope, topography etc.
However, when used operationally, this approach does not allow the influence of
local variations in catchment rainfall to be considered and, if the model is to include
this effect, then normally a rainfall runoff model will be required.
If there are no historical records for the sub-catchment (e.g. from an existing
river gauge without telemetry), then some options for developing a rainfall runoff
model include transferring model parameters from a similar nearby catchment, or
attempting to derive a pseudo inflow record by taking the difference between
measurements upstream and downstream of the location(s) at which flows enter
the main stream, allowing for lag times, attenuation, rating curve errors etc. Both
approaches have their disadvantages and, in practice an iterative, trial and error is
often used to obtain the best calibration of rainfall runoff and flow routing models
in combination.
An integrated catchment model can also include a number of sub-models for
other features which can influence river flows, including dams, reservoirs and con-
trol structures, and some of these components are described in later sections.
Table 8.2 Some possible approaches to modelling ungauged and lateral inflow and outflow
components
Model component Some typical approaches
Tributary inflows (see
Chapter 6 also)
Scaling from a nearby telemetered gauge based on catchment
area, and possibly other parameters (e.g. rainfall, effective
rainfall), possibly including a timing difference
Conceptual rainfall runoff modeling using parameters calibrated
from flow data from another similar catchment or historical
(non-telemetered) values for the same catchment (if available)
Distributed or process based hydrological modeling using param-
eters linked to catchment characteristics
Trial and error as part of the overall calibration, possibly based on
scaling inflows or outflows to the reach
Assume a constant inflow per unit river length
Gravity fed abstractions or
discharges
Typical daily, monthly or seasonal profiles
Lumping values for many locations
Pumped inflows or outflows Typical daily, monthly or seasonal profiles
Lumping values for many locations
Spillage or seepage at river
banks or flood defences
Stage discharge relationship derived from theory or off-line
modelling
Side weir representation
Main channel inflow-outflow relationship
Real time hydrodynamic model
Groundwater inflows or
recharge
Correlations related to river flow, river depth, rainfall and/or bore-
hole levels
Conceptual or process based rainfall runoff models with a subsur-
face component
Data-based techniques (artificial neural networks etc.)
Typical daily, monthly or seasonal profiles
3D numerical groundwater models
8.2 Flash Flood Forecasting
Flash floods can be damaging due to the short time in which they develop, and the
high depths and velocities which may be reached. Flows may also contain signifi-
cant quantities of debris and sediment, potentially causing blockages and further
raising river levels at structures such as bridges, weirs and river control structures.
Chapter 10 discusses some of the issues that can arise in responding to this type of
event, and provides an example for a flash flood which occurred in the UK in 2004,
whilst Chapter 3 discusses a range of rainfall threshold based approaches for early
warning of events such as flash floods.
The definition of a flash flood varies between countries, with a common theme
being that flooding of properties and infrastructure may occur in locations where
there is no recent flooding history, and develop sufficiently rapidly that normal flood
warning dissemination and emergency response procedures do not have time to oper-
ate effectively. Flash flooding is therefore often defined in relation to the locations at
risk and local response procedures, as in the following example (ACTIF 2004):
“A flash flood can be defined as a flood that threatens damage at a critical location in the catchment, where the time for the development of the flood from the upstream catchment is less than the time needed to activate warning, flood defence or mitigation measures downstream of the critical location. Thus with current technology even when the event is forecast, the achievable lead-time is not sufficient to implement preventative measures (e.g. evacuation, erecting of flood barriers).”
Some common features which appear in many definitions of flash floods include
the following items (e.g. World Meteorological Organisation 1982; Meon 2006)
although, in many cases, only some of these factors may apply:
● Tend to be caused by short duration, localised, intense storms (e.g. thunderstorms)
● Develop rapidly in response to rainfall
● May have significant mud/debris content
● Tend to occur on small, steep and/or urban catchments
● May be strongly influenced by catchment antecedent conditions
● Tend to occur in locations with no recent experience of such events
If a particular catchment response time is specified, then the values for events classi-
fied as flash floods vary widely between countries and can range from a few minutes
to a few hours (typically 6 hours e.g. World Meteorological Organisation 1982), with
catchment areas typically of at most a few hundred square kilometres, although some-
times much less than this. Various indicators such as catchment area, topography,
mean slope, soil type, response times and flood risk may also be used to identify flash
flood prone catchments. Fast response events from other sources, such as ice jams
(Section 8.3), urban drainage systems (Section 8.5) and dam breaks, landslides, and
Glacial Lake Outburst Floods (Section 8.6), are sometimes also classified as types of
flash flood.
The main difficulty with forecasting for flash floods is the speed at which they
develop, compared to other types of flood event, and the uncertainty about which
8.2 Flash Flood Forecasting 181
182 8 Selected Applications
particular locations will be affected in a region. The rarity of events may also mean
that there has been no need in the past for river level monitoring in the catchment
to confirm that an event is developing (i.e. the catchment is ungauged). These
factors also have implications for issuing flood warnings, and for providing an
effective response, and this topic is discussed in Chapter 10.
Some techniques used for flash flood forecasting include:
● Rainfall Thresholds using information on observed or forecast rainfall, and pos-
sibly current catchment conditions (see Chapter 3)
● Meteorological Indicators of flash flood generation potential from observations,
historical databases, or Numerical Weather Prediction models (see Chapter 3)
● River Level Thresholds in which warnings are issued using decision criteria
based on increasing river levels or flows (see Chapter 3)
● Rainfall Runoff Models using observed rainfall and, possibly, forecast rainfall as
inputs
As indicated, Chapter 3 describes a range of approaches which can potentially be
used for flash flood forecasting and warning, including rainfall depth-duration,
Flash Flood Guidance and probabilistic rainfall threshold techniques, geopotential,
vorticity, precipitable water and lightning meteorological indicators, and river level
threshold crossing, rate of rise and correlation methods (Box 8.1).
Rainfall Threshold and Meteorological Indicator approaches have the advantages
of providing additional lead time, and that they can be used even if there is no river
telemetry available. However, sometimes there can be considerable uncertainties in
the precise location, timing and magnitude of flooding, and this point needs to be
emphasised when issuing advisories, pre-warnings and other alerts based on these
techniques. Flash flood hazard maps, based on indicators of flash flood risk, such
as steepness, vulnerable locations, public awareness, and soil type, can also provide
guidance on catchments at particular risk, together with flash flood risk catalogues
at a national scale. By contrast, River Level Thresholds have the advantage of being
based on observations of current river conditions, although provide less lead time
than the other two approaches.
If a Rainfall Runoff Modelling approach is used, the techniques used for flash
flood forecasting are similar to those for other types of flooding, and include
process-based, conceptual and data-based models (see Chapter 6). However, one
potential difficulty may sometimes be a lack of historical and real time data for the
calibration and operation of models. Rainfall inputs can be obtained from rain-
gauges, weather radar, satellite, and rainfall forecasts, with a typical problem for
raingauges being that there may be no gauge within or close to the catchment and,
for the other methods, that the resolution may sometimes be too coarse compared
to the scale of the catchment to provide estimates which are representative. For
catchments where soil moisture or snow cover is a factor in flash flood generation,
there will often also be uncertainties about the initial catchment conditions.
If the catchment is gauged, then the model can be calibrated to historical river
flow records, if these are available. Given the rapid rate of rise of many flash floods,
the focus of model calibration and development may be more on success at predict-
ing the crossing of thresholds on the rising limb of the hydrograph, rather than on
accurate prediction throughout the event. Rainfall forecasts may be also used to
extend lead times, although the accuracy of forecasts decreases with increasing lead
time. Real time updating (data assimilation) is also usually desirable to help to
account for differences between observed and forecast flows, provided that the data
quality is sufficient. Modelling (and monitoring) may also be complicated by the
risks of debris and sediment, and in some cases that the river may be dry for part of
the year (e.g. on wadis). In urban areas, a range of extra factors may also need to
be considered, such as the capacity of the drainage system, blockage risks, and the
performance of flood storage and detention areas (see Section 8.5).
For ungauged catchments, there is the additional uncertainty arising from having
no river flow data for calibration, and views differ on whether warnings should be
issued on the basis of model forecast outputs alone in flash flood situations. Much
will depend on confidence in the model itself, and whether verification studies
show that the model outputs are a reliable predictor of flooding. Also, considerable
advances are being made in developing techniques for rainfall runoff modelling in
ungauged catchments, as described in Chapter 6.
In addition to these various techniques, the warning time available can some-
times be increased by reducing the various time delays in the overall flood warning
process (see Chapter 5). Some possibilities include making decision making proce-
dures more efficient, adopting faster approaches to warning dissemination, and
improving the speed of operation of telemetry and forecasting model systems.
Automated linkages between telemetry observations or flood forecasting model
outputs might also be implemented to reduce the time required to issue warnings,
such as automated signs or barriers on roads (although with the potential disadvantages
discussed in Chapter 4). Probabilistic and ensemble techniques (see Chapters 1, 5
and 10) also have the potential to improve operational decision making during the
development of flash flood events, and to assist with developing a more risk-based
approach to issuing flash flood warnings.
Given the destructive nature of some flash floods, and recent developments in
modelling and monitoring techniques, the issue of flash flooding is an active research
area, and Table 8.3 summarises some major research programmes on this topic.
8.2 Flash Flood Forecasting 183
Table 8.3 Some international research and collaboration programmes in flash flood forecasting
Project Location Reference
WMO flash flood guidance sys-
tem
International covering all WMO
regions
World Meteorological
Organisation (2007)
Central America Flash Flood
Guidance (CAFFG) system
Belize, Costa Rica, El Salvador,
Guatemala, Honduras,
Nicaragua, and Panama
Georgakakos (2005)
Sperfslage et al. (2005)
ICIMOD flash flood projects
(capacity building, early warn-
ing, satellite detection etc.)
Afghanistan, Bangladesh,
Bhutan, China, India,
Myanmar, Nepal, Pakistan
Erikkson (2006)
Project URBAS (prediction and
management of urban flash
floods)
Germany Castro et al. (2006);
(see Section 8.5)
Box 8.1 WMO Flash Flood Guidance System
The WMO Flash Flood Guidance System (FFGS) aims to assist National
Meteorological and Hydrological Services to improve their capability in pro-
viding warnings for flash flooding. The project was launched as part of the
WMO Flood Forecasting Initiative following the first International Workshop
on Flash Flood Forecasting held in Costa Rica in March 2006. The project
would be implemented by the Hydrologic Research Centre (HRC) in collab-
oration with NWS and funded by the U.S. Agency for International
Development/Office of Foreign Disaster Assistance (USAID/OFDA).
The aim of the Initiative is to enable access to satellite information which
provides warnings for catchments with areas in the range 100–300 km2 for
lead times of 1–6 hours. The FFGS follows a similar approach to that adopted
for the Central America Flash Flood Guidance system (CAFFG), imple-
mented by the Hydrologic Research Centre (HRC). The CAFFG system has
been operational in seven countries in Central America since 2004 (Belize,
Costa Rica, El Salvador, Guatemala, Honduras, Nicaragua, and Panama),
and builds upon the Flash Flood Guidance approach developed and used by
the National Weather Service in the USA since the 1970s (Fig. 8.2).
Flash Flood Guidance is defined as the amount of rainfall of a given duration
over a small basin needed to create minor flooding (bankful) conditions at the
outlet of the basin. Threshold rainfall values can be estimated by inputting histor-
ical events or typical storm profiles to a catchment rainfall runoff model for a
range of durations and soil moisture conditions. In real time operation, rainfall
observations or forecasts are combined with the outputs from physically-based
soil moisture accounting models to estimate the Flash Flood
Fig. 8.2 Sample graphical system output for Nicaragua from the CAFFG System showing
Flash Flood Guidance (FFG) and Flash Flood Threat (FFT) (reproduced from the WMO
Prospectus for Implementation of a Flash Flood Guidance System, courtesy of WMO)
184 8 Selected Applications
(continued)
8.2 Flash Flood Forecasting 185
Box 8.1 (continued)
Threat, which is the amount of rainfall of a given duration in excess of the
corresponding Flash Flood Guidance value.
A phased approach to implementation is envisaged in which regional centres
are established at selected National Meteorological and Hydrological Services,
equipped with the necessary computer, database, hydrological modelling, dis-
play and product generation facilities. Considerable flexibility will be provided
in choice of hydrological models and rainfall observations and forecast prod-
ucts, including the option to use satellite estimates of precipitation, automati-
cally corrected for bias based on telemetered raingauge data (where available).
The facility will be included for forecasters to make adjustments to values
received from regional centres based on experience since this has been shown to
improve success at forecasting flash floods, and to reduce false alarm rates. Data
and products will be transferred using existing WMO telecommunications sys-
tems and protocols, with appropriate training, verification, capacity building
and documentation all forming important components of the overall project.
Box 8.1 provides more information on the first two of these initiatives. Flash
flooding is also a major driver for research into detection techniques, including the
following methods which are discussed in Chapter 2:
● Satellite based techniques – analysis of cloud type, extent and rainfall generating
potential, and of catchment conditions (snow cover, soil moisture etc.)
● Weather radar – improved signal processing algorithms and hardware, particu-
larly for automated detection of heavy rainfall in mountainous areas
● Microwave techniques – methods for rainfall detection along microwave paths
between locations several kilometers (or more) apart
● Raingauges – low cost devices allowing greater network densities, and a greater
range of options for use in mountainous areas
● River monitoring – small, low cost devices allowing larger numbers to be
installed within a given budget, automated analysis of CCTV images, radar and
ultrasonic methods for remote detection of water levels
8.3 Snow and Ice
8.3.1 Snowmelt Forecasting
In catchments where snowfall accumulates, snowmelt has the capacity to cause
significant flooding, and events can be notable both for their magnitude and long
duration.
Flooding can arise directly from the melting of fallen snow from rainfall or
rising temperatures, or from radiation effects, or enhanced runoff from frozen
snow, or some combination of these mechanisms. Snowmelt events may also raise
river levels so that the capacity to absorb even moderate rainfall events is reduced.
Radiation melt often dominates in high latitude and high mountain areas, whereas
warm, moist turbulent airflows are an important factor in mid latitude, lower lying
regions.
Snowmelt forecasting models can range from simple empirical approaches relat-
ing snowmelt to temperature, through to models which attempt to represent both the
mass and energy balances of the snowpack and melted snow. Real time data inputs
(see Chapter 2) can include measurements of meteorological and hydrological con-
ditions, satellite observations of snow cover, ground based measurements (e.g.
ablation stakes, snow pillows), and forecasts for air temperature, humidity, wind
speed, cloud cover, and rainfall.
Some simple empirical approaches include linear or non-linear regressions
between flow volumes and various indicators for factors which can influence the
rate of snowmelt, such as air temperature and rainfall, the water equivalent depth
of accumulated snow, soil moisture, and the depth of frozen soil (e.g. World
Meteorological Organisation 1994). Another commonly used approach is the tem-
perature index or degree-day method in which the rate of snowmelt is represented
as a function of the difference between mean daily air temperature and a threshold
value above which snowmelt is considered to occur (often zero centigrade).
The function is usually a simple constant factor and is site specific, and different
values may be required for open and forest areas. The term degree-day refers
to an integrated value over a day, although other periods, such as hourly or
monthly values, can also be used. Usually the only real time input required is for
air temperature, although more advanced forms may also include wind speed and
radiation inputs, and may allow for elevation influences, and cloudy and sunny
conditions. However, known shortcomings are that the accuracy decreases as the
time interval chosen is reduced, and that it is difficult to account for spatial varia-
bility in snowmelt due, for example, to topographic influences (Hock 2003).
Conceptual models are also used in which the snow store is separated into snow
and melt components, and a simple water balance is used to keep account of each
component, with empirical or simple physically based equations to relate snowmelt
rates to air temperature. Typically, these models use air temperature relative to a
threshold (e.g. zero centigrade) to help to decide whether to partition precipitation
into rainfall or snow, and when to trigger the snowmelt component of the overall
model. Effects such as the proportion of snow cover in the catchment (areal deple-
tion curves), windspeed, and the influences from altitude and aspect (the direction
in which major hill slopes face) may also be included. One or more stores may be
used to represent the current status of melted snow (e.g. stored within the snow-
pack, lost as runoff), and the catchment divided into elevation zones to better
account for differences in temperature with elevation. Other advances include the
use of data assimilation through state updating based on point measurements by
snow pillow or snow cores (e.g. Bell et al. 2000).
186 8 Selected Applications
Forecasting models have also been developed using process-based rainfall run-
off models of the type described in Chapter 6 (e.g. Koren et al. 1999; Dunn and
Colohan 1999). Factors such as elevation and aspect can be included directly, and
the models can also account for other variations across a catchment, such as in wind
speed, the accumulation of snow due to wind effects (e.g. snow drifts) and varia-
tions in snow density. The most detailed models use a full set of mass and energy
balance equations to describe snowmelt, including real time observations of net
radiation, air temperature, wind speed and humidity (e.g. World Meteorological
Organisation 1994). Terms which are included in the energy balance can include
energy storage in the snow layer, net radiation, latent heat fluxes, the heat flux to
the ground or soil, and advection losses to wind flow over the snowpack, whilst the
water balance may allow for two or more layers to represent snow in various stages
of melting or freezing.
Figure 8.3 illustrates some terms in the overall energy and mass balances for a
melting snow layer; note that the directions shown for energy and water fluxes are
indicative, and in some cases can be in the opposite direction.
The net radiation depends on several factors including time of day, season, forest
cover, and snow surface conditions (which are typically parameterised via albedo
and emissivity). Other less significant factors may also contribute to the energy
balance, such as heat input from rainfall.
8.3 Snow and Ice 187
LatentHeat
SensibleHeat
GroundHeat Flux
StoredEnergy
NetRadiation
Wind
Evaporation
Precipitation
Recharge
ENERGY BALANCE WATER BALANCE
RunoffMeltingSurface Layer
Snow Layer
Fig. 8.3 Some key terms in the energy and water balance for a melting snow layer
188 8 Selected Applications
Historically, one barrier to application of this type of model has been the availa-
bility of real time data on both meteorological conditions and snow depth and
cover. However, improvements in the accuracy and resolution of Numerical
Weather Prediction models, and satellite based observation techniques, have helped
to make the application of this type of model more practicable.
Ensemble techniques can also be used; for example, the Extended Streamflow
Prediction System (ESP) used by the National Weather Service River Forecast
Centres in the United States since the 1970s (see Chapter 5). The method uses an
ensemble technique to create probabilistic river stage forecasts for periods of up to
a few months ahead. The main use is to provide forecasts for the Spring snowmelt,
by using the state variables of models at the time of forecast and up to 40 years of
historical time series data for model inputs (precipitation, temperature, potential
evaporation). The outputs from the model include probabilistic forecasts for peak
flows and volumes at multiple forecast points. Kuchment and Gelfan (2005) describe
a similar technique which has been developed for the Sosna River Basin in Russia,
and incorporates a process based rainfall runoff model which represents snow accu-
mulation and snowmelt, soil freezing, soil moisture, and runoff generation. Long
term deterministic and stochastic (Monte Carlo) estimates for daily air temperature
and precipitation are used as input to the model. Shorter range probabilistic flood
forecasting techniques are also under development and are described in Chapter 5.
Various international comparisons have also been performed of snowmelt models;
for example, in the first phase of the Snow Model Intercomparison Experiment
Project (SNOWMIP), more than 20 models from ten countries were compared,
with case study sites in the USA, Canada, France and Switzerland, and the project
has continued into a second phase (SNOWMIP2; Rutter and Essery 2006).
8.3.2 River Ice Forecasting
River ice can lead to an increased risk of flooding through mechanisms which
include:
● Ice formation – impeding flows in river channels and at bridges and other struc-
tures, reducing conveyance capacity, possibly for periods of weeks or months
● Ice jam – site specific blockages following the break up of ice cover further
upstream, particularly at bridges and flow control structures, causing a local shorter
term risk of flooding, and with the potential for significant structural damage
● Ice break up – flood waves generated by the sudden release of water following
melting or release of an ice jam, layer or dam
Ice formation tends to occur over periods of hours or days whilst ice jams can
develop rapidly, causing backwater influences and significant flooding in periods
of as little as an hour or less (and, for that reason, are sometimes called flash
floods). Ice jams may also act as an additional location for ice formation as floating
ice accumulates and ice forms at the point of constriction.
The presence of ice may also cause uncertainties in the inputs to (and hence
outputs from) flood forecasting models, by affecting the operation of river instru-
mentation and control structures, and causing a change in the stage-discharge
relationship (or rating curves) used to convert levels to flows at gauging stations.
River levels may also rise and fall as ice forms and breaks up, sometimes rapidly in
the case of ice jam movement.
Some techniques for monitoring the formation and break up of ice are described
briefly in Chapter 2. Two main categories of ice can be identified; thermally grown
(or sheet) ice and frazil ice. Thermally grown ice tends to occur in slow moving or
still water from temperature influences whilst frazil ice forms in moving water, and
is a common type of ice formation in rivers. Some particular locations where frazil
ice occurs include turbulent regions such as at river confluences, sudden reductions
in river slope, and downstream of rapids or in turbulent flow related to structures.
Techniques for estimating ice formation can range from simple correlation approaches
to process-based techniques of the type first developed for ocean ice formation (e.g.
World Meteorological Organisation 1994; Snorrason et al. 2000; Kubat et al. 2005).
Ice melt and break up can occur from thermal effects such as increases in water
or air temperature, and direct radiation from sunlight, or from mechanical influences
from the hydraulic loading of river flows, or impacts from floating ice, particularly
during flood flows. For thermal melt from water temperatures, river ice tends to melt
at the upstream edge of the ice cover, and the pattern of break up along a river reach
depends on the mean river slope, and changes in slope along the reach.
Procedures for forecasting ice formation and break up are used in several countries
where ice-related flooding is a problem, although are subject to many uncertainties.
For estimating the rate of formation of ice, one simple approach is to use a tempera-
ture index approach similar to that described earlier for snowmelt modelling (e.g.
World Meteorological Organisation 1994). The rate of ice formation is assumed to be
a function of a degree-day total (cumulative temperature over threshold measure)
calculated since the start of freezing. Similar techniques can also be used for estimat-
ing ice melt, in which melt is assumed to be proportional to a function of the river
discharge and the air temperature above a threshold value (typically zero degrees).
The time required to melt a given reach of ice can then be estimated from the melt
rate and the volume of ice estimated from the average ice thickness and width.
Hydraulic models of the types described in Chapter 6 also provide a way to model
potential locations for ice formation, and the impact of ice jams, and both steady
state and unsteady models might potentially be used (e.g. Blackburn and Hicks
2003). Some formulations also include dam break type models, and sub-models for
ice formation and ice melt. Operationally, the challenge in using these models is to
obtain or estimate up to date values for river ice formation, transport and break up
for input to the models; also, the nature of ice jams and break up can be site specific,
making it difficult to develop general approaches (e.g. White 2003; Hom et al.
2004). However, the timescale required for observations will depend on the nature
of the risk so that occasional (e.g. daily) observations may be sufficient in some
cases, for example using visual, CCTV, webcam or video-camera based observa-
tions. For example, in Alaska, an extensive programme of aerial observations, called
8.3 Snow and Ice 189
190 8 Selected Applications
River Watch, operates when the flood risk from ice is high, making use of voluntary
contributions by air taxi operators, private pilots, and others.
Various statistical and regression techniques have also been developed, some-
times linking into extensive databases of historical records (e.g. Mahabir et al.
2006). Some other current areas of research into data-based methods include multi-
variate statistical methods, artificial intelligence (e.g. neural networks), and Kalman
filtering, in some cases combined with a hydraulic modelling approach (e.g. Daly
2003; Morse and Hicks 2005).
8.4 Control Structures
Control structures can be used in rivers and along coastlines for a variety of pur-
poses including:
● Regulation of river flows
● Controlling water levels for navigation, irrigation etc.
● Protecting against high river or tidal levels
● Hydropower generation
● Diverting flows for flood mitigation
● Storing water for water supply, irrigation etc.
Types of structure can include dams, gates, sluices, weirs, barrages, locks and
siphons. Structures may be uncontrolled, in the sense that they always operate when
certain criteria are met (e.g. levels exceeding a certain threshold), or controlled
either manually or automatically.
For flood forecasting applications, a given structure may influence flows or lev-
els sufficiently that if possible it needs to be included in a forecasting model, and
this section describes some typical types of model for the following applications:
● Dams and reservoirs
● River control structures
● Tidal barriers
The resulting forecasts may also be used as inputs to real time control and decision
support systems and several examples are described in the following sections.
8.4.1 Dams and Reservoirs
Dams and reservoirs can be constructed for a range of applications, including
hydropower generation, water supply, navigation, and flood control. The impounding
structure may be in the main river channel (in-line, or on-stream), or adjacent to the
channel (off-line, or off-stream).
In-line structures typically regulate or control flows using gates, valves or sluices at
low to medium flows, but allow flows to pass freely over spillways during flood events.
Some schemes, such as hydropower plants, may also have emergency drawdown
arrangements. Smaller scale reservoirs built specifically for flood mitigation, such as
flood detention ponds, may have no control at low flows (for example, using a free flow-
ing tunnel as an exit), but start to hold back water as levels rise towards the peak. Flood
flows may be discharged over a spillway built into the front of the dam wall, or using
side channels, tunnels or bellmouth (shaft) spillways (Fig. 8.4). Self-priming siphons
may also included to rapidly draw down water levels if they approach the dam crest.
Off-line reservoirs, sometimes known as washlands, typically have an earth, rock-
filled or concrete wall around the perimeter, possibly with one or more internal barri-
ers with gates or sluices to control which areas of land are flooded. Flows in the main
river channel are diverted at weirs and sluices as required to reduce flows further
downstream, or for irrigation of the enclosed areas. Since floods may only occur
8.4 Control Sructures 191
Fig. 8.4 Examples of overflow and bellmouth reservoir spillways
192 8 Selected Applications
occasionally, the land may be used for farming or recreation at other times. Reservoirs
of this type are sometimes called polders although, for a true polder, the main purpose
is to protect reclaimed land within the boundaries of the polder, rather than to act as
a flood storage area, and sophisticated drainage and pumping systems may be used to
help to avoid flooding from rainfall, seepage, groundwater, rivers and the sea.
Figure 8.5 shows some examples of the potential impacts on flood flows down-
stream of fully regulated and unregulated in-line reservoirs, and an unregulated
off-line reservoir (World Meteorological Organisation 1994).
Reservoirs are usually operated using control rules (steering rules) linked to
reservoir levels, and possibly river levels, river flows, and other parameters (e.g.
rainfall). Rules may be static (e.g. seasonally dependent), or updated dynamically
based on observations and the outputs from forecasting models. Real time forecast-
ing systems can be used to assist with decision making and optimising reservoir
operation throughout the flow range (droughts, floods etc.), particularly when there
are several interconnected reservoirs to consider (multi-reservoir systems). Some
key decisions to take during a flood event include:
● Whether the predefined flood buffer (if any) will be sufficient for the anticipated
event and, if not, how much to draw down levels to protect the dam and locations
further downstream, particularly if multiple flood peaks are anticipated (e.g.
during a succession of storm events)
● Whether levels or flows are likely to reach values which might damage or over-
top the dam wall or operating equipment
● Whether normal operations should be suspended and for how long (e.g. for
water supply)
● Whether to warn or evacuate people downstream of the reservoir if it is likely to
spill or breach
● When, and to what extent, to divert flows into an off-line storage area, particularly
if people need to be warned or evacuated, or further flood peaks are forecast
● The optimum fill and drawdown sequence for a chain of reservoirs in a multi-
reservoir system
Fig. 8.5 Effects of reservoirs on floods (a) regulated storage (b) unregulated on-stream storage (c)
unregulated off-stream storage (Reproduced from the WMO Guide to Hydrological Practices – Data
Acquisition and Processing, Analysis, Forecasting and Other Applications, courtesy of WMO)
Also, there will often be an economic dimension to consider as part of the decision
making process; for example, if water is spilled from a hydropower or water supply
reservoir in advance of a flood event, this may incur an opportunity loss for electric-
ity generation or future water supply. Similarly, for off-line reservoirs, there may be
penalty payments to land users when flows are diverted onto farmland. Various
local arrangements may also be in place; for example, a flood warning authority
may rent the upper part of reservoir storage (a flood buffer) at a fixed annual rate,
but incur penalty payments if additional drawdown is required for flood protection.
Dynamic programming, stochastic simulations, artificial neural networks and other
techniques can be used to assist in the optimisation process (e.g. Bhattacharya et al.
2003; Lobbrecht et al. 2005; Nandalal and Bogardi 2007).
Forecasting models for reservoirs can range from simple correlation and water
balance models through to complex integrated catchment models incorporating rain-
fall runoff, snowmelt, flow routing, water transfer, hydrodynamic, and evaporation
models, including representation of the reservoir control rules. Some key factors to
consider in selecting an appropriate model for the reservoir component include:
● The availability of real time information on reservoir levels for model initialisa-
tion since, if the initial storage is not known, the forecast outflows can be
considerably in error.
● The availability of historical and real time information on reservoir control rules
and gate settings (if relevant), and the extent to which these can be encapsulated
into a model (e.g. logical rules). Also, the influence of pumping, water transfer
and other operations.
● The extent to which the reservoir influences flows downstream during flood events.
For example, if the impact is simply to attenuate and delay flows a simple flow
routing model may be sufficient but, if outflows are controlled, or sensitive to fore-
cast levels (e.g. siphon flows), then a full hydrodynamic model may be required.
● Whether components in the water balance such as open water evaporation, and
controlled releases for water supply, environmental or ecological purposes, are
of sufficient magnitude to require inclusion in the model.
The availability of real time data can be a particular issue for reservoir forecasting,
particularly for the reservoir levels required for model initialisation. Ideally, in addi-
tion to reservoir levels, all key inflows, gate settings and outflows would be monitored
in real time, but this is often not the case unless the monitoring network has been
designed specifically to support a real time control and forecasting application. Also,
for a reservoir with a large surface area, gauging of all major inflows may be imprac-
ticable. However, approximate modelling solutions can often be developed, for exam-
ple, parameterising control rules, using ungauged catchment forecasting techniques,
and other methods. The control rules used in practice may also differ from the pub-
lished control rules, which can require considerable investigation of historical records
and discussions with operators to determine the actual rules which are used.
Probabilistic techniques (see Chapters 5 and 10) are increasingly being developed
to assist in decision making during reservoir operations. This can bring several ben-
efits, including increased transparency in decision making, awareness of the likely
8.4 Control Sructures 193
194 8 Selected Applications
worst case scenarios given model uncertainty, and the ability to estimate probabili-
ties of occurrence for input to optimisation routines using cost-loss or utility or pen-
alty functions (see Chapter 10 for a discussion of these terms). Some examples of
optimisation problems might include the trade-offs between opportunity losses due
to draw down compared to the potential flood damage to locations downstream, the
likely cost of repair work at the dam if damage occurs, impacts at the reservoir (rec-
reational, ecological etc.), or the costs of any penalty payments which are incurred.
Stakeholders might also choose to receive warnings at different risk thresholds,
where risk can be defined as the combination of probability and consequence. For
example, a dam operator might be interested in receiving a flood warning at a much
lower probability or risk level (with a higher number of false alarms) than, say, a
community downstream of the reservoir.
Table 8.4 gives some examples of research and operational studies into the use
of forecasting models and decision support systems for real time reservoir opera-
tion and flood control.
Table 8.4 Examples of real time decision support systems for reservoir control
Location Reservoir uses
Model inputs
and structure Optimisation problem Reference
Lake Como,
Italy
Irrigation, hydro-
power
Stochastic Opportunity losses
from drawdown
versus flood dam-
age downstream
Todini and
Codeluppi
(1998)
Lenne River,
Germany
Multiple reservoirs
for hydro-
power, water
supply
Deterministic Optimisation of a
multiple reservoir
system for flood
control
Göppert et al.
(1998)
The
Netherlands
Multiple polder
systems
Artificial
neural
network
Optimisation of water
level management
Bhattacharya
et al.
(2003)
Powell and
Lois rivers,
Canada
Hydropower Ensemble,
statistical
Hydropower generation Howard (2004,
2007)
Folsom project,
California,
USA
Flood protection
to Sacramento,
hydropower,
water supply,
recreation
Deterministic,
Ensemble
Flood control,
Emergency
Response
Bowles et al.
(2004)
China Reservoir flood
forecasting and
control system
Deterministic Flood control Guo et al.
(2004)
Ebro river
basin, Spain
Multi reservoir
systems
(41 reservoirs)
Deterministic Flood control, water
management
Garcia et al.
(2005)
Paranaiba
river basin,
Brazil
Hydropower, flood
control
Deterministic Hydropower operation
and flood control
Collischonn
et al.
(2007a,b)
Feitsui
Reservoir,
Taiwan
Hydropower, water
supply
Deterministic Flood control in
typhoons
Nandalal and
Bogardi
(2007)
8.4.2 River Control Structures
River control structures can include gates, sluices, weirs, barrages, locks, and siphons. It
is also convenient to discuss pumps in this section. Some typical applications include:
● Diverting flows to off-line storage, water transfer schemes, farms etc.
● Maintaining levels upstream for irrigation or water supply
● Reducing river flows to mitigate flooding downstream
● Micro-hydropower schemes
● River flow gauging
● Control of river levels and flow velocities for navigation
Another application is to protect against high tides although tidal barriers are dis-
cussed in the following section.
For flood forecasting applications, in addition to the usual considerations of
cost, data availability and the functionality of the forecasting system etc., the deci-
sion on which, if any, structures to include in a model will depend on the proposed
application of the model, the locations of Forecasting Points, real time data availa-
bility, information on control rules, and other factors.
In some cases, major simplifications may be possible. For example, for pumps
or gravity fed offtakes, large numbers may be grouped together with a combined
operating rule, perhaps linked to seasonal control rules rather than relying on real
time data. A group of pumps might be considered to divert a fixed flow above a
certain threshold, and zero flow at all other times, or to pump or discharge flows at
a rate which depends on river levels. Also, some structures may be distant from the
required Forecasting Points and have minimal influence on the timing and magni-
tude of levels and flows at those locations, and hence can be omitted entirely. The
decision on which structures to include will vary from case to case.
The most appropriate type of model will depend on the mode of operation of each
structure. In some cases, a simple flow routing or water balance approach may be suffi-
cient to represent the effect of the structure on downstream flows. However, a real time
hydrodynamic model may be required if the structure has complex control rules, can
impound significant volumes of water, or if its operation depends strongly on forecasts
of levels upstream and/or downstream of the structure (e.g. backwater influences). If the
structure is within or downstream of a Flood Warning Area, then a detailed model may
be required for its influence on both upstream and downstream river levels.
The modelling of structures can become complex when there are several struc-
tures interacting and controlling levels or flows further downstream. In this case,
feedback effects can develop, and multiple scenarios or a probabilistic approach
may need to considered to achieve an optimum response as described in the previ-
ous section. Figure 8.6 gives an example of the type of situation which can arise.
In this example, river levels in a flood defence system protecting a town are control-
led by diverting flows to off-line storage reservoirs. The flows are diverted at weirs with
diversion channels, and the decision making process includes estimating how much
flow to divert, and at what times in the flood event. For example, one consideration is
that, if water is diverted too early, the reservoirs may not have the capacity required to
8.4 Control Sructures 195
196 8 Selected Applications
protect against the peak of the event. If multiple peaks are forecast (e.g. due to a suc-
cession of heavy rainfall events), then decisions need to be made on whether to delay
diverting flows to preserve capacity for later events (perhaps incurring some flooding
downstream), and on the extent to which reservoirs can be drained down between peaks.
Of course, in a worst case scenario, the forecasting model needs to alert operators that
the flood defences are likely to be overtopped, and a flood warning issued.
The choice of modelling approach will depend on the availability of real time data,
the control rules at the weirs, and the availability of survey data (if a hydrodynamic
model is required), as well as the usual factors of cost, system environment etc. Also,
the required accuracy at the flood defences may be a major consideration; for exam-
ple, if levels are be controlled during a flood event to within a typical freeboard allow-
ance of 0.1–0.5 m, then a hydraulic model would probably be required.
A possible forecasting approach could include rainfall runoff models for the tribu-
tary and lateral inflows, a hydrodynamic model extending from upstream of the weirs
to the flood defence system, and water balance or hydrodynamic models for the off-
line storage. Some other factors to consider include possible numerical transient
effects which may appear in model outputs if a hydrodynamic model is used (e.g.
when gates operate), whether there are any backwater influences from downstream of
the flood defence system (e.g. tidal influences), and whether to use a probabilistic or
deterministic approach. Also if real time updating (data assimilation) is used at river
gauging stations, this can further complicate development of the real time control
algorithms. Hence, although at first sight this is a simple configuration, this is a poten-
tially complex optimisation problem, which might require some exploratory investi-
gations to develop a solution.
Fig. 8.6 Illustration of a simple real time control problem
Flood Defences- Off-line storage 1
-
Off-line storage 2
Weir 1
Weir 2
8.4 Control Sructures 197
(continued)
Box 8.2 Thames Barrier (Environment Agency)
The Thames Barrier (Fig. 8.7) is situated in East London and protects London
from coastal flooding. The barrier consists of six rising sector gates which,
when lowered, lie in sills flat against the river bed allowing river users (ships,
boats etc.) to pass but, when raised, prevent levels rising on the upstream side
of the barrier due to tidal influences further downstream in the Thames
Estuary. Four simple radial gates are also included in the structure.
When necessary, the gates are typically closed four to six hours before high
tide, requiring the use of coastal forecasts to provide sufficient early warning
to allow the barrier to be closed in time to avoid flooding. The control rules are
based upon observed river levels upstream of London, and forecasts for tide
levels and surge in the tidal reaches of the Thames Estuary. These rules are
based on detailed hydraulic modelling performed at the time that the barrier
was being designed, combined with experience gained since completion of
the barrier.
Coastal forecasts are obtained from the UK’s Storm Tide Forecasting
Service (STFS), which operates models on a 12 km grid for the entire coastline
of the UK and the North Atlantic continental shelf, including the Bay of
Biscay. The model provides hourly surge forecasts up to 36 hours ahead at 6-
hourly intervals. A reduced version of the model, for is also operated locally
Fig. 8.7 Thames Barrier (© Environment Agency copyright and/or database right 2007.
All rights reserved)
198 8 Selected Applications
8.4.3 Tidal Barriers
Tidal barriers and barrages have many similarities to river control structures, with
the added complication of a tidal influence further downstream. In flooding appli-
cations, they are usually used to prevent river or estuary levels rising above flooding
threshold levels due to high tidal levels, including situations when river flows are
also high. Other applications include hydropower generation, and amenity use (e.g.
marinas, harbours etc.).
The operating rules for a tidal barrier typically guide the opening and closing of
the barrier gates according to a range of combinations of upstream river levels (or
flows), and tidal levels downstream. Control rules may be presented in the form of
charts, look up tables, or encapsulated within a computer-based decision support
system, possibly combined with real time forecasting models. Rules are often
developed using a combination of experience and detailed hydrodynamic and other
modelling for a large number of scenarios.
The design of a tidal barrier needs to allow for the possibility of river levels
upstream of the barrier rising to flooding levels during a river flood event whilst
the barrier is closed to provide protection from coastal flooding. The rules may
allow for this accumulated river flow to be released during low tide periods,
even when a major coastal event is in progress. The need for river flow releases
will depend on the magnitude of the storage upstream of the barrier in river
channels and estuaries, compared to the flow volumes likely to accumulate dur-
ing a tidal cycle.
Box 8.2 (continued)
at the Thames Barrier which allows data to be assimilated at the model
boundaries to improve model accuracy. Astronomical tide predictions are
obtained from the Proudman Oceanographic Laboratory (POL).
In the tidally influenced river reaches, a real time hydrodynamic model is
used for forecasting tidal levels, supplemented by rainfall runoff and one-
dimensional hydrodynamic models for forecasts of river flow in reaches
further upstream. These models include representation of major tributary
inflows and gates and barriers along the Thames estuary shoreline, and dis-
charges from key waste water treatment works. An empirical look-up table
approach is also used as a backup, and provides forecast levels at 14 locations
in the estuary.
The performance of forecasts for the barrier is regularly reviewed, together
with horizon scanning of research developments which may assist in the
future; for example, increased use of data assimilation, ensemble surge fore-
casting, and use of artificial intelligence techniques (such as artificial neural
networks).
Although barriers can often be operated on the basis of observed levels alone,
the use of river flow and coastal forecasts can extend the lead time available to
operators, allowing more time for mobilisation of staff, and closure of the barrier
(if the time taken is a constraint). Shipping operators can also be warned of possible
barrier closures (if relevant) and local authorities of any potential for flooding, if
this cannot be entirely mitigated by the barrier.
However, it is often desirable to minimise the number of closures per year, lead-
ing to an interesting optimisation problem; for example, balancing the need to pro-
vide flood protection against the cost of operating the barrier (staffing, power etc.),
the impact on maintenance and whole-life costs, interruptions to shipping and other
river users, and other factors. An ensemble or probabilistic approach may help in
optimising the decision and weighing up the costs and risks involved.
Some examples of tidal barriers include the Maeslant barrier near Rotterdam in
the Netherlands, which provides protection against surge events in the North Sea, the
Thames Barrier in London (Box 8.2) and the St Petersburg Flood Protection Barrier,
which is situated in a low lying area where the Neva River meets the Gulf of Finland.
The St Petersburg barrier is due for completion in around 2009–2010 and a decision
support system using observed data and meteorological and hydrodynamic forecasts
is under development to assist with barrier operations (Villars et al. 2007).
8.5 Urban Drainage
Urban drainage systems typically consist of a network of pipes, open channels and
culverts draining into sewers which carry flood flows to wastewater treatment works
or river or sea outfalls. Combined systems may be used for foul water (sewerage)
and surface runoff, or the flows may be handled by separate networks. Systems may
also include flood detention ponds, storm storage tanks and other types of storage.
Urban flood events tend to be characterised by a rapid response, with complicating
factors from debris and blockages, and restrictions on flood flows at major obsta-
cles on the floodplain (road or rail embankments etc.). The percentage of rainfall
which appears as runoff may be high due to the impermeable nature of some sur-
faces, such as roads, car parks and pedestrian areas.
Urban flooding mechanisms may include surface runoff before water enters the
drainage system (pluvial flooding), outflows from combined sewer systems and
from the drainage network where flows exceed capacity (at manholes, for example),
flows developing along paths of least resistance (e.g. roads), and ponding where
normal drainage paths are blocked. River flooding can also occur in urban areas,
although the response may be complicated by factors such as culverts and bridges,
and may lead to additional drainage related flooding (for example, where high river
levels impede drainage into the river network). River flows typically respond on a
longer timescale than urban runoff, so may occur some time after the local urban
response, although this will depend on the timing and distribution of rainfall in the
catchment both within and upstream of the urban area.
8.5 Urban Drainage 199
200 8 Selected Applications
For planning and design, hydraulic models are widely used and can model
drainage networks in great detail, including pipes, pumps, valves, flood detention
areas, and other features. In principle, models of this type could also be used in real
time to forecast surface flooding such as surcharging, outflows at outfalls, and other
factors, although this is rarely done at present. Modelling can be at district, street
or property level, and may include models for surface runoff into the drainage
network, although identification of drainage routes and contributing areas (drain-
age catchments) can be difficult in urban areas, and sensitive to small changes in
elevation (e.g. at roadside kerbs).
The hydrological inputs to such models usually assume idealised design storms
whereas, for real time use, details of storm speed, intensity and distribution are
required, and would ideally be available at a high resolution, comparable to typical
drainage catchment areas. Storm direction is of particular interest, since the load on
the drainage network can be higher if rainfall moves down the network, as runoff
in the later stages of the event adds to water already in the system from higher
elevation areas.
Options for monitoring rainfall in urban areas include raingauges and weather
radar. For raingauges, the network density required is usually higher than for river
catchment monitoring, and may not be feasible on cost grounds, although dense
networks are operated by some authorities; for example, Harris County, Texas oper-
ates approximately 100 real time raingauges within the boundaries of Houston.
Weather radar can potentially provide more information on the spatial distribution
of rainfall although, as noted in Chapter 2, performance will depend on distance
from the nearest radar, and the accuracy in urban areas may be affected by tall
buildings, masts and other factors. Cheaper, low cost radar, and microwave tech-
niques, are also other possibilities for monitoring rainfall in urban areas.
Information on antecedent conditions may also be required, with runoff charac-
teristics defined individually, or in aggregate, for a wide range of features, including
roofs, roads, gardens, fields, and features in sustainable drainage systems (SUDS),
together with gullies, culverts and other local drainage routes. Geographical
Information Systems combined with Digital Terrain Models offer the possibility of
modelling at this level of detail, although obtaining calibration data and parameters
for all combinations of conditions can be difficult.
The requirements for urban flood forecasting models have many similarities to
river forecasting techniques, but in addition to rainfall runoff and flow routing com-
ponents, there may be a need for models for the drainage network and other local
effects, such as temporary flow paths along roads and other open areas. River and
urban models can also be coupled dynamically, allowing the interactions between
systems to be represented (e.g. of river levels on drainage, or river spill into a drainage
system). However, although modern computing systems have the capability to run
very detailed models, even in real time, in practice there is usually a need to simplify
and conceptualise models to focus on locations and factors which have the most influ-
ence on flooding, perhaps linked to surface water flooding maps.
In addition to providing information to guide warnings of potential flooding,
models can also be used for adaptive or predictive real time control of urban
drainage networks (e.g. Cluckie et al. 1998; Vitasovic 2006). For flood events,
the objective is typically to reduce flooding and pollution incidents by making
use of spare capacity in the drainage network by operating pumps, gates, control
valves, weirs etc., or temporarily diverting water to storage areas. Potential
flooding may therefore be reduced or avoided in flood prone locations, whilst
some wider objectives in normal operation can include reducing pumping costs,
optimising the performance of water treatment plant through providing forecasts
of future water and pollution loads, and decision support if problems arise with
parts of the network.
Schilling (1989) and Schütze et al. (2004) present reviews of approaches to the
real time control for urban drainage systems, whilst Table 8.5 presents some examples
of systems with a flood control component which have been used operationally or
in research studies.
The key components in this type of system include monitoring equipment (and
related telemetry), automated or remotely controlled actuators (on valves, gates
etc.), and a defined real time control strategy or set of rules. Some options for
developing control rules include:
● Heuristic techniques – based on experience and trial and error
● Off-line simulation – to develop a pre-defined set of rules
● Real time predictive control – optimising performance in real time
A range of optimisation criteria might be considered, or multi-criteria approaches
used.
Current research themes in real time control for urban drainage systems have
many similarities to the more general themes in river and coastal flood forecasting
which have been described in earlier chapters, and include:
● Objectives – better definition of control objectives with respect to national and
international legislation, and the roles and responsibilities of organisations with
an interest in, or responsibility for, urban flooding
Table 8.5 Examples of studies into real time control systems incorporating flood control
aspects
Location Key features Reference
Haute-Sûre Reservoir,
Luxembourg
Raingauge and radar inputs to sewer model Henry et al. (2005)
Quebec urban community
RTC, Canada
Multi-objective optimisation to minimise
overflows, maximise treatment plant use,
minimise accumulated volumes etc.
Schutze et al. (2004)
Brays Bayou, Houston,
USA
Weather radar inputs to a fine mesh process-
based rainfall runoff model
Vieux et al. (2005)
URBAS project,
Germany
Improved radar and 2D modelling techniques
for urban flood forecasting with case
studies for 15 German municipalities
Castro et al. (2006)
8.5 Urban Drainage 201
202 8 Selected Applications
● Monitoring – improved techniques, particularly for weather radar in urban
areas, automated fault detection and diagnosis (including blockages), and
devices for monitoring surface runoff (more reliable, lower cost, lower visual
impact etc.)
● Modelling – improved ways of modelling surface runoff in urban areas, includ-
ing higher resolution models, automated generation of flow paths from digital
terrain and urban land use data, and alternative approaches such as artificial
neural networks
● Probabilistic methods – using ensemble rainfall forecasts, stochastically gener-
ated scenarios etc., and risk based approaches to decision making (combining
probability and consequence)
● Control – improved optimisation techniques, allowance for human factors,
improved information gathering and display systems
8.6 Geotechnical Risks
Geotechnics is a general expression for a range of specialist areas including the
study of natural hazards such as earthquakes, landslides, mudflows and avalanches,
and of the performance of structures where they interact with the earth surface.
Table 8.6 illustrates some potential causes of flooding to people and property from
geotechnical risks.
In flood forecasting applications, the extent to which these various types of
event can be modelled will depend on how well the underlying cause can be
monitored and predicted, including its influence on river, reservoir or coastal lev-
els. Interactions between flooding mechanisms may also need to be considered.
Within an overall flood forecasting model, additional sub-models can be devel-
oped for these components, or purpose made models developed specifically for
these types of risk. A common theme in this type of event is that estimation of the
forcing component requires an understanding of geotechnical processes, and that
Table 8.6 Some potential causes of flooding from geotechnical risks
General category Type Examples
Structural risks Dam break
Failure of river or coastal
defences
Glacial lake outburst floods/
jökulhlaups, landslide
dam outburst floods
Can occur to any dam or defence if over-
topped, poorly maintained or designed,
damaged by earthquake or landslide etc.
A risk in some mountain areas e.g. the
Himalaya, Canada, Iceland
Earth movements Tsunami December 2004 Tsunami
Landslides or debris flows
into rivers or reservoirs
Western USA (particularly after wildfires),
Central Asia and the Caucasus
the flooding impact can potentially be extensive and extreme, and may require
hydrodynamic modelling techniques to predict.
Some other types of flood event which might be included in this category
(although this is debatable) include:
● Groundwater flooding – due to raised water levels in aquifers, causing increased
flows at springs and in rivers, and sometimes additional flow routes developing
that are not usually observed (e.g. springs appearing within properties)
● Blockage risks – flooding caused by the accumulation of debris (trees, vegeta-
tion etc.) at structures such as bridges, culverts and trash screens
For groundwater flooding, there are well developed hydrodynamic modelling tech-
niques for simulation of aquifers, and these can in principle be applied in near real
time if sufficient real time input data are available on borehole levels, river flows,
rainfall and snow cover, pumping operations etc. Given the often slow nature of
groundwater flooding, a model run frequency of once per day or less may be suffi-
cient; the problem being to obtain enough up to date data to initialise the model.
Other simpler correlation, data based and conceptual or process-based rainfall run-
off modelling techniques may also be used (see Chapter 6).
For blockage risks, although these can pose a considerable hazard, they are
intrinsically difficult to forecast, unless there are known problem locations at which
debris usually tends to accumulate, or up to date information can be obtained on
blockage locations (e.g. from at site river level instrumentation, or observations by
CCTV or webcam). This information can then be input into a flow routing or
hydrodynamic model, possibly trying various scenarios for the extent of blockage,
and for the discharge coefficient at the structure. Simpler threshold based tech-
niques may also be used (see Chapter 3).
8.6.1 Structural Risks
Structural risks include dam break, defence (levee) breaches or failure, and failure
of naturally occurring moraine in mountain regions. This type of event can cause
extensive and rapid flooding, possibly with depths, velocities and extents beyond
any previous experience. Some examples have included the failures of two major
dams and dozens of smaller dams in Henan Province in China during a typhoon
event in August 1975, in which about 85,000 people died, and millions were dis-
placed, and the failure of the Vaiont Dam in Northern Italy in October 1963, due to
the wave caused by a landslide into the lake, in which about 2,000 people died
(Graham 2000). The levee failures during and following Hurricane Katrina in
August 2005 also inundated large parts of New Orleans.
The mechanisms for dam and defence failures can include:
● Piping/seepage – flow routes developing through or under the structure, causing
progressive erosion and eventual failure
8.6 Geotechnical Risks 203
204 8 Selected Applications
● Overtopping or wave action – eroding the crest of the structure in places, again
with progressive increases in erosion and flows
● Liquefaction – for earth structures, for example
● Earth movements – earthquakes and landslides causing direct damage to
structures
For flood defences (or levees or dikes), failure modes can depend on the materials
used for construction (earth, rockfill, concrete etc.), the hydraulic loading, and the
geometry of the structure. The hydraulic loading can arise from water level differ-
ences across a structure, wave action/overtopping, and flows along the length of the
structure. Many sub categories and secondary influences can also be identified.
As part of contingency arrangements, particularly for dam failure, some organisa-
tions maintain maps of likely flood extents, and emergency plans detailing the actions
which should be taken if a failure occurs. During a flood event, rapid flood mapping
exercises might also be commissioned if there are concerns about the integrity of a
structure. Detailed monitoring can also be performed of the structure using in-situ (e.g.
accelerometers, piezometers) and remote techniques (e.g. webcams, laser scanning).
Many studies have been performed into simulation (off-line) modelling of struc-
tural failures. Models for the breach component can range from simple weir equa-
tions assuming a fixed breach size, through to fully dynamic process-based
simulations in which the hydraulic and soil erosion processes are coupled, and
modelled in two or three dimensions. These types of technique could potentially be
applied in real time for scenario modelling if the location of the breach (or likely
location) is known; however, forecasting likely locations remains a research area at
present. For example, the probabilistic risk assessment techniques developed for
flood risk modelling (see Chapter 1), which combine information on probability of
levels (loading) and defence fragility curves, might be used to provide guidance on
potential breach locations. Multiple breach scenarios are also calculated off-line to
provide one of the sources of information available to emergency managers in the
Netherlands, for example (see Chapter 10).
Model calibration is complicated by the lack of reliable calibration data, due to
the rarity and unpredictability of occurrence of this type of event, although a number
of large-scale experiments have been performed to gain an insight into typical
parameter values (e.g. IMPACT 2005). For debris blockages in rivers from land-
slides, added complications include the irregular, and probably unknown, nature of
the materials making up the barrier. For Glacial Lake Outburst Floods and Landslide
Dam Outburst Floods, the conditions leading up to failure of the debris/moraine/ice
dam can often be recognised, although the timing is difficult to predict, and may be
influenced by volcanic and earthquake activity (e.g. Snorrason et al. 2000).
For dam breaks, in addition to breach models, flood propagation models are
usually also required for flows downstream, and are usually of the types described
in Chapter 6 for river flows. However, some added complications may include
high sediment loads, the need to model structures and obstacles which are
normally above extreme river flood levels, and uncertainty about the appropriate
values to use for some model parameters (e.g. roughness coefficients) and for
likely flow paths.
8.6.2 Earth Movements
Earthquakes, landslides, volcanic and other events can affect river and coastal con-
ditions either indirectly (by damage to structures, for example; see the previous
section), or directly, by generation of flood waves, such as Tsunami in coastal
waters, and by causing surge waves, flow blockages, and an increase in debris and
sediment content in river waters.
For rivers, debris or mud flows often result from heavy rainfall, and are in prin-
ciple predictable using similar techniques to those used for flash flooding. The
‘flows’ consist mainly of water, soil, rocks, slurry and other debris (e.g.
Mirtskhoulava 2000). Operational forecasting systems typically use rainfall depth
duration thresholds based on observed rainfall, and sometimes forecast rainfall, and
these methods have been reported for Puerto Rico, Hong Kong, Taiwan, and loca-
tions in Italy and the USA, for example (e.g. NOAA-USGS 2005). A prototype
Debris Flow Warning System for the western USA is also described using a rainfall
threshold approach combined with predictive debris-flow-volume models, but with
a long term development path towards distributed near real time process based
models combining digital terrain models, rainfall distributions, and models for
rainfall infiltration, slope stability, and channel bed erosion and deposition (e.g.
Rickenmann and Chen 2003).
By contrast, in the oceans, perhaps the main hazard is from the waves generated
by subsea earthquakes, volcanic activity or landslides, which are usually known as
Tsunami. In addition to the catastrophic December 2004 event in the Indian Ocean,
other major Tsunami in recent decades have included the 1993 Hokkaido Tsunami
in Japan, which reached heights of 5–10 m, and the 1992 Flores Tsunami in
Indonesia, with heights of about 4–7 m (Satake 2000).
In the open ocean, Tsunami waves typically travel at speeds of about 500–
1,000 km per hour, depending on ocean depth, with a wave length of about 5–10 km
or more (and sometimes hundreds of kilometres). Here, they pose little threat since
wave magnitudes are small, and typically only of the order of 1 m. The destructive
effect of a Tsunami arises in shallow coastal waters, when raised water levels,
combined with the large volumes of water involved, can cause extensive flooding
inland.
The modelling techniques for Tsunami are essentially those already described in
Chapter 7 for surge propagation. For open ocean propagation, the non-linear
shallow water equations provide a reasonable approximation to wave motion due
to the long wavelength, which is comparable to typical ocean depths. However,
techniques are less well developed for modelling the initial wave formation, and the
run up process at shorelines, and high grid resolutions, and three dimensional mod-
eling techniques, are possibly required to obtain accurate estimates (and remain an
active area of research).
Tsunami monitoring and forecasting is performed both nationally and interna-
tionally; for example, in Japan, and at the Pacific Tsunami Warning Centre in
Hawaii. The December 2004 Tsunami has also led to major efforts to install and
upgrade warning systems. Forecasts rely both on ocean modelling, and detection of
8.6 Geotechnical Risks 205
206 8 Selected Applications
seismic activity. Since the speed of seismic waves is about an order of magnitude
more than for Tsunami waves, seismic detectors can also be used to trigger an
increased state of monitoring and mobilisation, if the conditions appear likely to
have caused an event. For example, in some locations, such as the Pacific Ocean,
the time for a Tsunami to cross the ocean is several hours, so forecasts can be based
on sea level observations as well (e.g. Satake 2000). However, locally generated
events can occur only a few minutes after the initial seismic event so that is the
main indicator of likely flooding.
Part IIIEmergency Response
K. Sene, Flood Warning, Forecasting and Emergency Response, 209
© Springer Science + Business Media B.V. 2008
Chapter 9Preparedness
As with other types of natural hazard, the effectiveness of response to a flood event
can be improved if an emergency plan has already been prepared, so that all partici-
pants understand their roles and responsibilities, including the overall chain of
command. The potential disruption from flooding also needs to be considered,
including the possibility of communication, instrumentation, computer and other
systems failing, and access and evacuation routes being cut by flood water. Risk
assessment techniques are also increasingly used to assess the resilience of response
procedures, together with developments in information technology for the spatial
analysis and visualisation of flood extent relative to properties, infrastructure and
transport routes. This chapter provides an introduction to these issues, and discusses
the general trend towards multi-hazard systems, which share systems and resources
across many types of threat.
9.1 Flood Emergency Planning
9.1.1 General Principles
Flood Emergency Plans describe the actions to take between, during and following
flood events, and typically cover operational procedures, emergency response
assets, contact details for key staff, health and safety issues, procedures for liaison
with the media and the public, and information on safe access and evacuation routes
and shelters. Some guidelines on developing flood emergency plans include US
Army Corps of Engineers (1996), NOAA/NWS (1997) and Emergency Management
Australia (1999) for river flooding, and Holland (2007) for tropical cyclone fore-
casting. Depending on the type of flooding, lead time available, and population
affected, examples of actions which may need to be taken in the run up to and dur-
ing a flood event include (USACE 1996):
● Providing search, rescue, and evacuation services
● Scheduling closure of schools and transportation of students
● Curtailing electric and gas service to prevent fire and explosions
● Establishing traffic controls to facilitate evacuation and prevent inadvertent
travel into hazardous areas
● Dispersing fire and rescue services for continued protection
● Establishing emergency medical services and shelters
● Closing levee openings
● Moving public and private vehicles and equipment from areas subject to
flooding
● Relocating or stacking contents of private structures
● Initiating flood-fighting efforts (e.g. sandbagging etc.)
● Establishing security to prevent looting
For tropical cyclones (e.g. World Meteorological Organisation 2006a; Holland 2007),
at about 36–48 hours from landfall, activities can include fishing boats returning to
home ports, setting up emergency operation centres, preliminary precautions being
taken by residents, and starting to provide 6 hourly updates. Then, within about 24–36
hours from landfall, evacuation of exposed properties begins, businesses and industry
commence shutdown procedures, and updates move to 3 hourly intervals. Most prep-
arations should be complete within about 6–8 hours of landfall, leaving the emer-
gency services to check for any remaining vulnerable people and secure community
lifelines. Warnings continue for about 12 hours after landfall, including for river
flooding from heavy rainfall, and any changes in cyclone strength or track.
Information on roles and responsibilities, and inter-agency coordination, is often
an important consideration in developing flood emergency plans, since a river or
coastal flooding event can cut across administrative and political boundaries, and
may affect more than one country, so many organisations may be involved in the
response. The organisational structure and terminology used varies widely between
countries, but Table 9.1 illustrates in general terms some of the organisations which
may be involved in responding to a flood event, and some typical roles and
responsibilities.
In emergency planning guides, some key points which are often emphasised
include:
● Command – a clear chain of command is needed to avoid duplication of effort
and missing vital actions, and for communicating information to the public and
the media.
● Key Contacts – regular communication between agencies through exercises and
training, and personal contacts, pays dividends under the pressure of a major
event.
● Resilience – plans should be sufficiently adaptable to cope with extreme events
beyond those in recent memory, with contingency plans in case of failure of any
component.
● Vulnerable Groups – special arrangements are often needed for people with
physical disabilities or a medical condition, the elderly (in some cases), and others
who are dependent, such as children.
● Transient Populations – plans need to allow for temporary residents of areas at risk,
including tourists, business travellers, road users, student populations, and others.
210 9 Preparedness
Table 9.1 Organisations which may be involved in responding to a flood event
Type Group/department etc. Typical roles and responsibilities
Environmental,
Scientific
Meteorological Service
Flood Warning Service
(if separate)
Weather forecasting and monitoring; river and
coastal monitoring and forecasting; flood warning
dissemination; and possibly operating flow control
structures, flood defence repairs and reinforcement
etc.
Emergency
Services/Civil
Protection
Police Road closures; evacuation of properties; dissemi-
nation of flood warnings (in some countries); law
enforcement
Social and Health
services
Treatment and evacuation of injured people;
precautionary evacuation of locations such as
nursing homes and hospitals; providing food,
water, clothing, counselling etc.
Fire & Rescue Rescues from flooded properties and elsewhere;
pumping of flooded water; fire fighting
National, State,
Local Authorities
Emergency Planners,
Disaster Managers,
others
Overall coordination of activities; establishing
evacuation centers; arranging distribution of
sandbags; temporary defences; levee repairs and
reinforcement; financial assistance etc.
Highways, Contractors Road closures; temporary works; debris removal
Army, Navy, Air Force,
Coastguard
Rescue operations; provision of specialist
equipment and additional staff resources, boats,
hovercraft, helicopters, trucks, amphibious
vehicles etc.
Communities Local representatives Coordinating community response for floodfighting
and evacuation; obtaining relief funding and
additional resources (people, equipment etc.)
Community members Assisting with evacuation; issuing warnings to
neighbours/friends etc.; temporary measures to
reduce flooding and protect property etc.
Utilities Water Installation of temporary defences at treatment
works; switching supplies between works (if
possible); provision of temporary water supplies
Electricity and Gas Installation of temporary defences at substations,
gas works and power stations; precautionary close
down and rerouting of power supplies; safety
inspections and advice; drawdown of hydropower
reservoirs
Telecommunications Installation of temporary defences at communication
hubs; rerouting of networks if needed
Canal/Port operators,
Navigation Authorities
Operating canal and other gates; reinforcing
defences; assisting ship and boat owners; closing
tidal barriers
Transport
Operators
Rail, Road Inspection or closure of flooded rail lines; provision
of buses, trucks etc. to help in evacuation and
recovery
Airports Closure of flooded runways, diversion of flights
Private Sector Contractors, vehicle
owners etc.
Provision of transport, emergency repair equipment,
additional staff, fuel, food etc.
Voluntary Sector Charities, Relief
Organisations, others
Assisting in own specialist areas (the aged,
the young, animals, food and drink, clothing,
counselling etc.)
Media Television, radio, other Reporting, relaying information and warnings
● Community Engagement – it is important to involve community members (or
their representatives) in developing emergency plans, and ensuring that these are
tailored to their own needs and resources.
● Health and Safety – apart from the risk of drowning, flood waters present a range
of other risks to emergency workers, including waterborne diseases, hazardous
materials (fuel, chemicals, sewage etc.), and electrocution from exposed cables,
so strict procedures, safety equipment and decontamination facilities are ideally
required.
● Continuous Improvement – an ongoing process of validation, testing, review and
updates is important, particularly to account for changes in staff, organisational
structures, suppliers, contractors etc.
Social, political, and cultural differences between countries can lead to a wide
range of approaches to command and control structures, but often national disaster
managers, local authorities, the police or the military will assume the role of overall
coordinators, depending on the scale of the event. In a major event, regional and
national command centres may also be established to ensure that resources, funds
and equipment are made available to flooded areas.
The resources available for flood response can include extra staff and private
sector contractors (e.g. for emergency repairs and debris removal), whilst equip-
ment can include vehicles, medical supplies, earthmovers, pumps, generators,
boats, sandbags, and mobile communications. Procedures also need to be estab-
lished for accepting assistance from third parties. For example, in rapidly flowing
water, swift water rescue skills are needed, and offers of assistance may need to be
refused. The establishment of a national system of response and rescue competen-
cies (sometimes called ‘team typing’), provides one way of ensuring that volunteers
and others have the necessary training, equipment and procedures to avoid becoming
victims themselves
Community engagement can cover both education on actions to take when
receiving warnings, and direct inputs to emergency plans from community mem-
bers or their representatives. Much of the literature on early warning systems (e.g.
Handmer 2002; World Meteorological Organisation 2006b; ISDR 2006; United
Nations 2006) emphasises a community based or people-centred approach, in
which those at risk are engaged in the planning process, helping to decide on the
most effective forms of response, and the best way to present and receive warning
messages. Four key elements to consider (ISDR 2006) include:
● Risk Knowledge – Are the hazards and the vulnerabilities well known? What are
the patterns and trends in these factors? Are risk maps and data widely
available?
● Monitoring & Warning Service – Are the right parameters being monitored? Is
there a sound scientific basis for making forecasts? Can accurate and timely
warnings be generated?
● Dissemination and Communication – Do warnings reach all of those at risk? Are
the risks and warnings understood? Is the warning information clear and
useable?
212 9 Preparedness
● Response Capability – Are response plans up to date and tested? Are local
capacities and knowledge made use of? Are people prepared and ready to react
to warnings?
Cultural issues relating to gender, income, language and other factors also need to
be considered, such as informal developments on floodplains, with plans tailored to
meet the needs of specific groups (e.g. do people have livestock or pets which they
may be reluctant to leave behind). In particular, this may increase the success rate
of evacuating people during a flood, with failure to leave property leading to the
emergency services needing to engage in time consuming and potentially high risk
rescues at later stages in the event. Chapter 4 discusses some of the techniques
which can be used for raising public awareness of flooding, and of the actions to
take on receiving flood warnings, including leaflet campaigns, seminars, meetings,
radio and television appearances, open days, flood fairs, school visits, and educa-
tional films.
A distinction can also be made between plans at individual (or family) level,
community level, and at site specific, organisational, local, regional or national
level. For example, individuals can develop (perhaps with assistance) personal
action plans describing actions to take in the event of flooding, including considera-
tion of locations at risk from flooding, how to protect property, checklists of what
to take if evacuated (food and water, medical supplies, phones, radios, food, flash-
lights etc.), key contacts, information on where to go in an emergency, and the saf-
est escape routes (e.g. FEMA 1997). Other items might include batteries, blankets,
protective gloves, disinfectant, a first aid kit, personal documents, insurance poli-
cies, and switching off gas, water and electricity. In the Netherlands, for example,
residents can view neighbourhood flood emergency response plans and maps on a
website by entering their property location. At community/village level (World
Meteorological Organisation 2006a), plans might include:
● Identifying and maintaining of safe havens, safe areas and temporary shelters
● Putting up signs on routes or alternate routes leading to safe shelters
● Informing the public of the location of safe areas and the shortest routes leading
to them
● Having all important contacts ready: district or provincial and national emer-
gency lines; and having a focal point in the village
● Making arrangements for the damage and needs assessment team and health
team
● Setting up community volunteer teams for a 24-hour flood watch
● Improving or keeping communication channels open to disseminate warnings
● Distributing the information throughout the community
In rural areas, cattle, poultry and other livestock might also be moved to safety and
supplies of firewood, drinking water and animal folder secured. The Stormwatch
programme in the USA is another example at community level, and a community
can be designated as StormReady (FEMA 2005) for weather hazards by completing
the following steps:
9.1 Flood Emergency Planning 213
● Establish a 24-hour warning point and emergency operations center
● Have more than one way to receive severe weather warnings and forecasts and
to alert the public
● Create a system that monitors weather conditions locally
● Promote the importance of public readiness through community seminars
● Develop a formal hazardous weather plan, which includes training severe
weather spotters and holding emergency exercises
Site specific plans may discuss particular hazards and needs at individual locations
(e.g. water treatment works, power stations) whilst, at local authority, regional and
national level, flood emergency plans usually also cover the actions needed in the
aftermath of an event (the recovery phase), including which organisation(s) will
assume responsibility for repairs, debris removal, reuniting families, emergency
funding arrangements, and providing shelter, food, water, medical care, counsel-
ling, support to businesses, and restoration of services if interrupted (power, water,
communications etc.). However, these topics fall outside the scope of this book and
are not considered further, except for the topic of post event analysis of flood warning
performance, which is discussed in Chapter 11.
9.1.2 Risk Assessments
One key stage in developing a flood emergency plan is to assess the flood risk,
and to tailor the response to the level of risk. Flood risk can be expressed in
terms of the probability of flooding and the likely impacts (e.g. the number of
people or properties at risk, or economic value, or the combination of exposure
and vulnerability). Here vulnerability concerns not just the threat of flooding,
but the ability of people to cope with a flood event, including mobility, age,
type of residence, awareness of flooding, and access to transport and flood resil-
ience measures.
Chapter 1 describes techniques for estimating the likely extent of inundation,
which can include compiling historical observations, using local knowledge, and
performing hydrodynamic modelling studies of various levels of complexity. If
available, the resulting inundation maps can be overlain on street maps and satellite
or aerial photographs, perhaps combined with terrain models based on digital eleva-
tion datasets. Additional potential sources of flooding may also need to be consid-
ered, including flood defences which are known to be in poor condition, locations
where debris or ice jams may cause raised water levels, and dam breaches.
In addition to identifying properties at risk, high-risk locations such as hospitals,
schools and nursing homes may also need to be considered, and specific procedures
developed for these locations within the overall plan. For a hospital, for example,
this might include defining the criteria for evacuation, and the additional resources
which health workers would require in moving patients to a place of safety (vehi-
cles, people, equipment etc.). For some situations, there can be a difficult balance
of risks to consider, each requiring a separate risk assessment; for example, for
214 9 Preparedness
nursing homes and hospitals, evacuating patients may cause injury or distress, and
this needs to be balanced against the risks from flooding. Specific assessments may
also be needed for other high-risk locations such as power stations, water treatment
works, prisons, high-rise buildings, underground car parks, major roads, airports
and rail stations, shopping centers and subway systems (e.g. Ishigaki et al. 2006),
industrial plant, and sites with hazardous materials. Evacuation sequences can also
be prepared as shown in the simple example in Fig. 9.1.
The figure shows a hypothetical example of a flood warning area with five zones
or districts defined. An indication is provided of the numbers of properties which
need to be evacuated as river levels rise, and the time available before the onset of
flooding (which is shown by grey shading). Links are provided to a map at each
river or tidal level showing the estimated extent of flooding at that level, the proper-
ties affected, and safe evacuation routes. During a flood event, plans would of
course need to be adapted as the situation unfolds, and real time decision support
systems are increasingly being considered for use in updating emergency plans in
real time as described in Chapter 10.
As noted in Chapters 3 and 5, several factors need to be considered when estimat-
ing the actual time available for emergency actions, including the time delay between
observations being made and being ready to use (polling time), the various time
delays in performing forecasting model runs, the decision times needed by opera-
tional staff, and the time required to issue warnings to all recipients. For example, for
tropical cyclone forecasting, the time delays up to the time of issuing a warning (i.e.
excluding the dissemination time) include the time taken for observational data to
arrive at the Tropical Cyclone Warning Centre, the time taken for data to be processed
and then presented to the forecaster, forecaster analysis, assessment and prediction
time, and the time needed for message preparation (Holland 2007).
Estimates such as these can guide response strategies, and help to gain a better
idea of the trade off between delaying a decision (hoping for better information),
and reducing the time available for emergency actions to be taken, such as evacuat-
ing properties. This point is discussed further in Section 10.3.
Perhaps the most highly developed approach to evacuation planning is that used
in the USA for hurricane evacuations (see Box 9.1). For example, during Hurricane
Katrina in 2005, it is estimated that some one million people left their properties.
Evacuation plans are typically developed as part of Hurricane Evacuation Studies
9.1 Flood Emergency Planning 215
Fig. 9.1 Illustrative example of an evacuation sequence
Box 9.1 Hurricane evacuation procedures
Hurricanes, typhoons and tropical cyclones can cause widespread damage,
and evacuation is one of the main options for reducing risk to coastal popula-
tions. Flood risks arise from storm surge and heavy rainfall, with the added
complication of high winds and the possibility of tornadoes.
However, the decision on whether to issue an evacuation order is a difficult
balance between the need to protect people, and to avoid the economic and
other impacts of false alarms. Most states in the USA follow similar proce-
dures as a hurricane approaches land, although the details and terminology
used can differ (e.g. Smith and Ward 1998; Wolshon et al. 2001, 2005a).
A typical five stage alert might initiate the following actions:
● Level V – represents normal, routine operation.
● Level IV – is a preliminary state of alert which is activated if a tropical
storm is developing. A team is formed to monitor developments and report
to key government officials, emergency responders, and the Federal
Emergency Management Agency (FEMA).
● Level III – is activated if a hurricane strike appears likely in the state, with
actions to clear evacuation routes of obstructions, monitor traffic volumes,
and to liaise with the National Guard and officials in neighbouring states.
● Level II – is the stage at which information on evacuation and shelters is
issued to the public, and a Declaration of Emergency is requested.
● Level I – is the highest state of alert, at which recommendations to issue
the order or recommendation to evacuate are made to local authorities, and
arrangements are made (if required) to evacuate vulnerable people; for
example in nursing homes, hospitals, and prisons.
The timing of events varies but typically the move to Level IV might be up to 1
week ahead of the estimated time of landfall, Level III 3 or more days ahead,
Level II 2–3 days ahead, and Level I within 1 day of landfall. Minimum lead times
required for evacuation can be in the range 9–72 hours ahead of landfall depend-
ing on the category of hurricane and road network and population densities, but
are typically in the range 12–24 hours. Evacuation routes are typically closed
shortly before the hurricane makes landfall so that any residents, media represent-
atives etc. remaining can be directed to local shelters or refuges of last resort.
Evacuation orders can be voluntary, recommended or mandatory. Voluntary
orders are used for offshore workers and others who require long lead times to
move, whilst recommended orders are issued if there is a high probability of a hur-
ricane causing a threat to people in at risk areas. Mandatory orders are only used in
some states. Public awareness campaigns between events play an important role in
improving the effectiveness of evacuation orders when a hurricane occurs, with the
aim to maximise the number of people at risk moving to shelters, and minimise the
number of so-called shadow evacuations, where people move although they may
not be directly at risk.
216 9 Preparedness
(USACE 1995), initiated since the 1980s by the Federal Emergency Management
Agency (FEMA), and available to download from the internet. These studies typi-
cally incorporate a hazard analysis, a vulnerability analysis, an evacuee behav-
ioural analysis, a sheltering analysis, and a transportation analysis, guided by
estimates for maximum inland water levels from the National Weather Service
(see Chapter 7), and including plans to evacuate vulnerable groups.
9.1.3 All-Hazard Approaches
Flood emergency plans are often developed as part of an all-hazard (or multi-hazard)
approach to emergency planning, and the methods used may be formalised in
national legislation. A multi-hazard approach brings economies of scale, sustaina-
bility and efficiency, and opportunities for more frequent use and testing than for
single-hazard systems (e.g. ISDR 2006).
For example, in the United States, the Federal Emergency Management Agency
(FEMA http://www.fema.gov/) uses a standard approach called the National
Incident Management System for incident management for all hazards and across
all levels of government. The six main components are:
● Command and Management – covering Incident Command Systems, Multi-
Agency Coordination Systems, and Public Information Systems
● Preparedness – covering planning, training, exercises, personal qualification and
certification, equipment acquisition and certification, mutual aid and publica-
tions management
● Resource Management – covering standardized mechanisms and establishing
requirements for processes to describe, inventory, mobilise, dispatch, track and
recover resources over the life cycle of an incident
● Communications and Information Management – covering Incident Management
Communications and Information Management
● Supporting Technologies – including voice and data communications systems,
information management systems, and data display systems
● Ongoing Management and Maintenance – covering routine review and continu-
ous refinement over the long term
Emphasis is placed on a flexible, adaptable approach, which can be easily adopted
and understood by a wide range of types and scale of organisation, including gov-
ernmental, private sector and voluntary sector organisations. FEMA also produces
the national flood maps used in flood risk assessments and emergency planning,
typically using hydraulic modelling approaches to assess inundation extent.
Some other examples of generic approaches to incident management include those
developed by the French Organisation de la Réponse de Securité Civile (plan ORSEC:
Direction de la Défense et de la Sécurité Civiles 2006), Emergency Management
Australia (1999), the United Nations/International Strategy for Disaster Reduction
(ISDR 2006) and the Civil Contingencies Secretariat in the UK (see Box 9.2).
9.1 Flood Emergency Planning 217
218 9 Preparedness
(continued)
Box 9.2 United Kingdom – Civil Contingencies Act (2004)
The Civil Contingencies Act defines two classes of responder to emergency
situations which, depending on the situation, can include:
● Category 1 responders – emergency services, local authorities, Health
Bodies, Environment Agency/Scottish Environment Protection Agency
(SEPA)
● Category 2 responders – Health and Safety executive, transport operators
(rail, road, airport, harbour, underground), utilities (electricity, gas, water,
telecommunications), Strategic Health Authorities
Category 1 responders are typically first on the scene of an event, and take
responsibility for managing the crisis, and include the two main organisa-
tions responsible for issuing flood warnings (the Environment Agency and
SEPA). The main civil protection duties for Category 1 responders are as fol-
lows (HM Government 2005):
● Risk assessment
● Business continuity management (BCM)
● Emergency planning
● Maintaining public awareness and arrangements to warn, inform and
advise the public
● Cooperation and information sharing
● Provision of advice and assistance to the commercial sector and voluntary
organisations (Local Authorities only)
Collaboration is also facilitated by a system of local and regional Resilience
Forums, which meet regularly to review emergency plans for a range of haz-
ards, including flooding, where appropriate. At a national level, control is
coordinated by the Civil Contingencies Committee which convenes as
required to deal with national and regional scale emergencies, including river
and coastal flooding. Other mechanisms for collaboration include visits,
seminars, joint projects and exercises, via bilateral and multilateral groups
and forums.
The Act distinguishes between generic plans, covering a range of hazards,
and specific plans, to deal with particular kinds of emergency. Plans can be
prepared for single organisations or across groups of organisations (Multi-
Agency plans). The minimum level of information to be contained in a
specific plan is (HM Government 2005):
● Aim of the plan, including links with the plans of other responders
● Information about the specific hazard or contingency or site for which the
plan has been prepared
● Trigger for activation of the plan, including alert and standby
procedures
● Activation procedures
9.1.4 Validation and Testing of Plans
Plans usually also consider arrangements for validation, testing and regular reviews.
Emergency response exercises are often office-based (e.g. table top exercises), and
may occur in ‘accelerated’ time. Typically, the exercise coordinator will introduce
complications as the event unfolds, including escalating the situation, and introduc-
ing issues needing immediate response. Public relations skills may also be tested,
including simulated questions from the public, media, and politicians. Larger scale
exercises may also include simulated television and radio news broadcasts, and
computer-based simulations of the types of output which would be available in a
real event, such as forecasting model outputs, and weather radar displays.
As an example of the complexity of these exercises, Exercise TRITON held in
the UK in 2004 simulated an extreme flood event in England and Wales, and
involved some 1,000 representatives from more than 60 organisations and agencies,
based at 35 locations, over a period of 3 days. The exercise explored the effective-
ness of interagency coordination, command structures, response assets etc. during
both the response and recovery phases of the event.
In the USA, the Hurricane Pam exercise in 2004 was based around a fictional
Category 3–4 hurricane, with sustained winds of 120 mph, a major storm surge, and
up to 20 in. of rainfall in places (US House of Representatives 2006). The aim was
to help officials develop joint response plans for a catastrophic hurricane in
Louisiana. The exercise was performed over four workshops, and the first workshop,
held in July 2004, involved some 300 emergency officials from 50 parish, state,
Box 9.2 (continued)
● Identification and roles of multi-agency strategic (gold) and tactical (sil-
ver) teams
● Identification of lead responsibilities of different responder organisations
at different stages of the response
● Identification of roles of each responder organisation
● Location of joint operations centre from which emergency will be managed
● Stand-down procedures
● Annex: contact details of key personnel and partner agencies
● Plan validation (exercises) schedule
● Training schedule
In addition to the needs of victims and the safety of emergency workers, particular
emphasis is given to the needs of vulnerable groups, who may require special
assistance during an incident.
Crown copyright material is reproduced with the permission of the Controller
of HMSO and the Queen’s Printer for Scotland
9.1 Flood Emergency Planning 219
220 9 Preparedness
federal, and volunteer organizations over a period of 8 days. The topics considered
included search and rescue operations, temporary shelters, medical care, debris
management, and other factors, including dealing with hazardous materials, power,
water and ice distribution, and drainage of flood waters.
The exercise was performed as part of a wider initiative by the Federal Emergency
Management Agency (FEMA) in conducting catastrophic disaster planning, and was
the first in a series of 25 disaster scenarios to be considered (prioritised on the basis
of risk). The lessons learned from the exercise helped with the response to Hurricane
Katrina during 2005 although, as with any exercise, some additional factors occurred
during that event which were not anticipated, and work was still in progress in imple-
menting some of the lessons learned from the exercise at the time of the event.
9.2 Resilience
9.2.1 Introduction
There is an increasing trend towards viewing flood emergency management as an
exercise in risk management and reduction. Except in certain cases, such as closing
a flood gate, or raising a temporary barrier, flood warnings can only reduce the
adverse impacts of flooding on people and property so, during a flood event, the
task is therefore to minimise the risks as effectively and safely as possible.
One key aspect of risk management is to assess the likelihood of failure, and
some common sources of problems in flood incident response include:
● Equipment – a shortage or breakdown of vehicles, high volume pumps, boats etc.
● People – key staff out of contact, insufficient training, insufficient people
● Power – interruptions to electricity or gas supplies and telecommunications
● Fixed Communications – damage to land lines and telecommunications hubs
● Mobile Communications – interference, damage to towers, poor signal strength
● Drinking Water – flooding of treatment works, contamination by flood waters
● Medical Care – shortages of medical staff and equipment if many casualties
● Fuel – shortages due to garages being flooded or inaccessible, oil supplies being
disrupted
Many of these factors are interdependent, with electricity failures in particular
potentially affecting both fixed and mobile communications, operations at water
treatment works, and hospitals and other key infrastructure. Flooding can also be
exacerbated by problems at river control structures, tidal gates, temporary barriers,
and other control structures and contingency plans need to be in place for emer-
gency repairs or manual operation of structures.
Communications failures can also seriously hamper rescue operations. For example,
during Hurricane Katrina, in August 2005, damage to radio and cell phone
communication towers, telephone switching hubs, power supplies, and other
problems, resulted in the loss of three million telephone lines in the states of
Louisiana, Mississippi and Alabama, 38 emergency phone call centres, some
police and fire department communication networks, up to 2,000 cell phone sites,
and 37 out of 41 radio stations in New Orleans and surrounding areas. Backup
generators for radio networks also failed in places due to flooding, or the difficulty
in refuelling them. Satellite phones, either hand-held or in Mobile Emergency
Response Support vehicles, remained the only reliable form of communication in
some locations. Critical infrastructure affected included the New Orleans Police
Department headquarters, and six of the eight police districts’ buildings (US
House of Representatives 2006).
To give an indication of the range of technical and communication problems
which can occur, Table 9.2 illustrates some issues for a number of flood events in
Table 9.2 Illustration of some of the technical and communications difficulties which can occur
during flood events
Topic Point of failure
Some examples of possible contin-
gency arrangements
Control centres/
emergency
operations centres
Communications, power or access,
or the site itself, may be affected
by flood waters or traffic hold ups
One or more backup locations in areas
away from flooding; permanently
relocate if the flood risk is significant
Rescue centres,
shelters
Sites may be at risk from flooding,
or inaccessible to evacuees and staff
and to vehicles bringing food, water
and clothing
Choose locations for rescue centres
away from flood waters and with
good access even in flooding
conditions
Access routes Access to locations expected to
flood may be affected by flood
waters or the weight of traffic
Pre-position key staff, vehicles, and
equipment, install temporary barriers
(if used) before routes are cut off
Evacuation routes Roads may be blocked by
floodwater or the weight of
traffic; fuel may be inaccessible or
in short supply
Plan a phased evacuation, possibly
assisted by computer modelling
of scenarios. Make arrangements
for access to boats, helicopters,
hovercraft, amphibious vehicles etc.
Utilities Flooding may interrupt power,
water, and gas supplies, including
locations distant from the areas
affected by flooding
Have backup supply arrangements
planned as far as possible, and
alternatives such as bottled water,
clothing etc. (or evacuate areas at risk).
Consider flood proofing or relocation
Warning dissem-
ination systems
Failure of telephones, cell phones,
sirens, blocked access for loud
hailer routes, overloading of
websites and phone based warning
services
Develop alternative warning
dissemination arrangements which do
not rely on mains power or access to
specific locations; perform load testing
for extreme call volumes etc.
Communications
between
organisations
Systems may not be compatible Investigate the frequencies
and systems used by other
organisations, particularly for radio
communications, and agree on inter-
agency communication arrangements.
(continued)
9.2 Resilience 221
222 9 Preparedness
Communication
failures
Flooding of key telecomm-
unications hubs or of the
power supplies to cell phone and
radio networks
Arrange back up communication
methods, and floodproofing or
relocation of key communication
infrastructure
Mobile
communication
problems
Poor signal strength on site
for cell phones and two way radio
Investigate the coverage in flood
prone areas, arrange alternatives and
backups if necessary
Media
communications
Television and radio broadcasts
may fail due to power problems
Have additional warning
dissemination arrangements in place
Widespread
flooding
Regional flooding may also affect
organisations normally relied
upon to assist at a local level
Develop links with organisations
located further afield (bilaterally,
or through national coordination),
with the possibility of international
assistance
Telemetry Key instruments or telemetry links
may fail so that information on
meteorological, river or coastal
conditions is
not available
Decide which alternative instruments
would be used in case of failures,
install additional instruments and
telemetry links as a backup, and raise
electrical equipment above likely
peak water levels
Hazardous material Flood waters may affect sewage
treatment works, and contain
fuel, chemicals, animal carcasses,
human waste and other toxins
Procure protective equipment for staff
(dry suits etc.), and develop standard
decontamination procedures, develop
policies for advising the public, and
media on risks
Equipment available Limited availability of high
volume pumps, communication
equipment, sandbags, temporary
barriers, boats, vehicles etc.
Explore the availability within other
organisations and nationally, with
arrangements in place for requesting
assistance if needed; also, stockpiling
of equipment at strategic locations
Staff welfare Health and Safety issues from
floods, waterborne diseases and
hazardous materials, wind, heavy
rain; potential issues with hours
on duty, friends and family
affected by flooding, access to
home, site or office in a flood
event
Review training, equipment, duty
rotas, with backup arrangements for
staff etc.
Europe and the USA in recent years, based on various sources referred to in this
and other chapters, together with some simple examples of possible contingency
arrangements.
In addition to evacuation of properties, and mobilising staff, some preparatory
actions which can be taken by the emergency services and others if warnings are
received in time include:
Topic Point of failure
Some examples of possible contin-
gency arrangements
Table 9.2 (continued)
● Installation of temporary or demountable barriers around key infrastructure, or
rerouting of power, water, communications links etc.
● Controlled shutdown of industrial processes
● Pre-positioning of emergency response assets in locations likely to be cut off by
flood water
● Stockpiling of food, water, fuel, sandbags and other supplies at strategic
locations
For example, for the hypothetical flood warning scenarios presented in Fig. 1.5,
a flood risk mapping exercise might show that, in a heavy rainfall event, the
main areas at risk include the main town, power station, motorway, caravan
park, and isolated farms. An initial assessment (Fig. 9.2) might suggest a range
of options in the early stages of the event before flooding occurs, including
placing residents of the caravan park on standby to evacuate the site, assessing
the readiness of the power station operator to install temporary defences, or
switch to alternate power supplies, placing a police patrol at the main road
bridge ready to close the road if flooding looks likely, and positioning a number
of search and rescue crews, with mobile communications, food and water sup-
plies, in the town in case access routes are cut. The timing of actions would
depend on the probability of flooding, and levels of risk (which, for simplicity,
are not shown on the figure).
Increasingly, spatial analysis tools are being used for this type of assessment,
and are described in Section 9.3.
Town
Power Station
RailwayDam
Village
Farms Chemical Factory
Town
Town
Town
Major road
Caravan Park
Police patrol
Precautionaryevacuation
Pre-position 2crews
Discuss contingencyplans with dutymanager
Flood risk map outline
Fig. 9.2 Simple illustration of operational response early in a developing river flood event
9.2 Resilience 223
224 9 Preparedness
9.2.2 Analysis Techniques
Many other problems can occur in the chain of detection, forecasting, warning, dissemination
and response, and the analysis of possible causes of failure, and devising alternative
plans, is often called contingency planning or business continuity management.
The initial assessment of risks is usually performed based on experience and les-
sons learned from previous flood events, and many analyses may proceed no further
than this stage. However, more formal techniques are widely used in other sectors,
such as the chemical, aviation and oil and gas industries, and the aim is usually to
consider the following questions (e.g. Federal Aviation Authority 2000):
● What can fail?
● How it can fail?
● How frequently will it fail?
● What are the effects of the failure?
● How important, from a safety viewpoint, are the effects of the failure?
Some widely used methods include:
● Failure Mode, Effects and Criticality (FMECA) Analysis – which systematically
considers the potential modes of failure of each component or system, the likely
impacts, the chances of detecting each type of failure, and the overall risks pre-
sented by each scenario
● Probabilistic Risk Assessments – which combine the probabilities and consequences
of various scenarios to estimate the risk from each combination, and are often per-
formed using a Fault or Event Tree analysis, or Monte Carlo simulations
● Fault or Event Tree Analysis – in which a graphical representation is produced
of the chain of faults or events which can lead to a given outcome, typically
linked by logical AND or OR gates. The probabilities of occurrence for each
individual component in the chain can be combined to estimate the overall prob-
ability for each combination of circumstances
● Scenario Modeling – using computer simulation and numerical modeling tech-
niques to investigate alternative scenarios (which in a flooding context might
include factors such as flood defence breaches, pump failures, failures at river
control structures etc.)
Analyses may start from a single event, and explore all outcomes, or consider a
single outcome, and explore what faults or events could realistically have led to that
problem. These are known as top down or bottom up approaches. Systems may be
probabilistically safe, inherently safe, failsafe, or fault tolerant. For example, a
probabilistically safe system has enough redundancy (computers, instruments etc.)
that a failure is unlikely to cause problems, whilst a fault tolerant system can continue
to operate if problems occur, although not as effectively.
Thus an analysis of flood emergency response would need to consider all likely
causes of failure (‘weak links’) and critical paths, including problems due to equipment,
communications, and human factors, either for individual subsystems, or for the entire
system. The analyses could be used as a simple screening tool, to help to explore poten-
tial sources of failure, or taken further to derive quantitative estimates of risk.
These types of analysis are little used at present in flood incident management,
but are being developed by some authorities. To take a simple example, in probabi-
listic risk assessment the risk presented by a pump failure could be estimated as the
probability of failure per event, multiplied by the number of events in the period
being considered (e.g. a year), and the probability of this leading to property flooding,
to give an overall risk score. Figure 9.3 shows a slightly more complex example for
a tidal sluice gate.
The gate needs to be closed during exceptional tides to protect the village of Pill
in southwest England from flooding, and this example was developed to illustrate
the potential use of Bayesian networks in risk assessments for flood incident
management.
For individual components in the chain of people, equipment and systems the
situation being considered may be reasonably self contained and amenable to analysis
(and these techniques may already be used by suppliers of some forms of emergency
response equipment, for example). However, when considering wider issues, there
are numerous interacting factors, both technical and non-technical, which can affect
Fig. 9.3 Representation of the flood incident management process for a site at Pill near Bristol
(Environment Agency 2006, © Environment Agency copyright and/or database right 2008. All
rights reserved)
9.2 Resilience 225
226 9 Preparedness
how the response to a flood event unfolds. For example, the response and expectations
of individuals may need to be considered; including the extent to which people will
help neighbours and the emergency services, or may take risks trying to rescue animals
or valuable possessions. Also, whether people will agree to leave their homes when
issued with a warning.
For analysis of systems with high levels of uncertainty, and complex interac-
tions, other more qualitative methods are available in which user views on probability
(e.g. ‘quite high’) and consequence (e.g. ‘serious’) are translated into a form which
can be analysed. Techniques include fuzzy set logic and linguistic reasoning in
which scoring and ranking systems are used to quantify risks, based on interviews
and expert opinion (e.g. Environment Agency 2007). Other approaches, such as
Bayesian Networks, Artificial Neural Networks, and Agent Based Modelling,
might also be considered. The focus of the analysis could be at a range of levels,
including at strategic, operational or tactical level, covering multiple organisations
and stakeholders, or covering just individual organisations, or single locations.
9.3 Role of Information Technology
9.3.1 Introduction
Information technology plays an increasingly important role in the management of
flood emergencies, and several examples of the components within a typical inte-
grated flood warning and response system have already been discussed, including:
● Flood Risk Assessments – hydrodynamic modelling to derive quantitative esti-
mates of flood risk, combining probability and consequence (Chapter 1)
● Telemetry Systems – to manage the collection of real time data for river levels,
rainfall, tide levels, and other parameters (Chapter 2)
● Dissemination Systems – automated systems for sending flood warnings to the
public and first responders, by telephone, email and other methods (Chapter 4)
● Flood Forecasting Systems – to operate real time rainfall runoff, flow routing,
hydrodynamic, surge, wave and other forecasting models (Chapter 5)
In the area of emergency response, some additional types of system which are
increasingly being used include:
● Geographical Information Systems – which can display and analyse information
on flood risk, emergency assets etc.
● Decision Support Systems – which can provide guidance on optimum response
strategies when developing emergency response plans, and in real time (see
Chapter 10)
● Simulators – which can be used in training exercises and to help to develop
emergency response plans
For all applications, if systems are integrated into operational procedures, then
some important factors to consider include resilience, staff training, data security,
data quality, metadata, and access to confidential information. Also, for inter-
agency systems, there are many issues to consider concerning funding, terminol-
ogy, standards, and the interoperability of systems. Adoption of formal risk
assessments, and software design and acceptance testing procedures, is essential.
9.3.2 Geographical Information Systems
Geographical Information Systems (GIS) are increasingly used to assist both with
planning for emergencies, and in the recovery phase (e.g. MacFarlane 2005; van
Oosterom et al. 2005). Information can be combined from a wide range of sources
and presented in a consistent and intuitive way. In flooding applications, some
information of interest can include rivers, coastlines, flood hazard maps, census
data, administrative boundaries, roads, streets, topography, critical installations,
industrial hazards, commercial properties, utility data, and other features, as well as
information specific to the emergency response (e.g. control rooms, evacuation
shelters, medical facilities). Searches can also be performed for specific sectors of
the community who may require specific assistance during an event (e.g. people
with medical conditions, or who speak a foreign language).
Sources of information can include local authorities, the emergency services,
central government, and the private sector. To combine datasets from many differ-
ent sources, the location and extent of each feature is needed. For records which are
not already geographically referenced, information on street addresses, post codes,
or road names or may be sufficient to generate the required datasets. The function-
ality available in a modern system typically includes:
● Pan, zoom and overlay of map layers and images (e.g. aerial or satellite
photographs)
● Other presentation options (e.g. transparent layers, three dimensional views,
animations, graphs, reports)
● Overlay, Neighbourhood and Buffering analyses (e.g. all properties which lie
within a flood risk area, or within a given distance of a river)
● Boolean analysis allowing complex searches to be performed (e.g. the operating
bases for all helicopters with the required load capacity within 20 minutes flying
time of a site)
● Network analyses (e.g. for travel times of emergency response vehicles from one
location to another)
● Terrain analysis (e.g. for assessing the line of sight of radio and cell phone com-
munications to and from a flood prone area)
● Data modelling using the outputs from models to generate new datasets (e.g.
flood hazard maps)
9.3 Role of Information Technology 227
228 9 Preparedness
Systems can be operated on a desktop computer, or made available over a corporate
intranet, and increasingly selected results are published on the internet (e.g. the
flood hazard maps in the UK and USA). Web-based GIS can also be used to facilitate
the exchange of information between different organisations involved in emergency
response, although may lack the full functionality of a desktop system.
Most systems can also link to an external database, containing additional
information on specific features, such as key data, images or video clips.
For example, the US FEMA National Incident Management System IRIS data-
base, launched in 2007, provides the facility to store detailed information on
approximately 120 types of emergency response asset, including equipment,
communications, contracts, facilities, responders, services, supplies and teams.
The system allows emergency responders to inventory, categorize type, locate,
request, and track all internal and external resources from a single integrated
system to help facilitate the response of requested resources during an incident
(FEMA 2007).
Geographical Information Systems and Decision Support Systems can also assist
with the response during a flood event and this topic is discussed in Chapter 10.
9.3.3 Visualisation and Simulation
Visualisation and simulation tools are increasingly used to present complex infor-
mation in an intuitive way to non-specialists, and for training exercises and assist-
ing in developing emergency response plans. In flood emergency response
applications, two examples of interest are:
● Flood Maps – three dimensional views and animations of flood extent against a
backdrop of topography, buildings, mountains etc.
● Simulators – virtual reality effects incorporated into systems used for emergency
response exercises, and in decision support systems
This is an active area of research (e.g. Pajorova et al. 2007), and practical applica-
tions are already in use in other emergency response applications; for example, in
training search and rescue teams to deal with major fires and nuclear incidents.
Figure 9.4 shows an example of a virtual reality representation of flooding in a
residential area produced by the Virtual Environmental Planning Project (http://
www.veps3d.org).
River levels are shown in bank and for a major flood event. The flood sce-
nario is hypothetical, and extreme, but illustrates how flood extents can be
viewed in context. The software also allows users to animate the flood
sequence, and to view the scene from different directions, and at different
magnifications.
More generally, search and rescue services are increasingly using simulators
for training exercises or to test emergency response plans, often building on tech-
niques developed for computer games, and these techniques also have potential
for flood response applications. Some typical functionality (e.g. Gyorfi et al.
2006) includes:
● Virtual Reality – the trainer and students can move around and interact with a
virtual representation of the control room or incident, including buildings, res-
cue equipment, vehicles, personnel etc. Simulations can be replayed as part of
the training exercise.
● Multimedia – video links are provided to actual or simulated personnel for inter-
views, consultation etc., simulated television news, and synthetic or actual radio
and cell phone discussions.
● Networking – the facility is available for multiple participants to join the simula-
tion over the internet, and to influence the course of events.
● Artificial Intelligence – is used to encapsulate and represent the behaviour of
other participants at the scene of the event (the public, casualties, emergency
services etc.).
The nature of the simulations can be tailored to national incident protocols and
linked into national standards. For example, the Incident Commander training sim-
ulator software developed for the National Institute of Justice, which is the research
arm of the U.S. Department of Justice, can be used by emergency managers in test-
ing response, and training, to hurricane, flood and other small to medium scale
incidents (up to 500,000 residents). The severe storm recovery component provides
the option to deal with many types of problem, included obstructed roads, casualties,
gas leaks, and downed trees blocking roads.
Fig. 9.4 Virtual reality representation of flooding (Copyright 2007 VEPS)
9.3 Role of Information Technology 229
Chapter 10Response
Flood warnings provide local authorities, the emergency services, the public and
others with time to take actions to reduce the risk from flooding, and information
on the likely extent and locations of flooding. Actions which can be taken before a
flood starts include installation of temporary defences, operation of flow control
structures, protection of personal property, evacuation of people from areas at risk,
and positioning of emergency vehicles and other assets in locations which may
become inaccessible due to flood waters. Increasingly, decision support systems are
also being developed to assist in responding to flood events, and can provide advice
on strategies for evacuating property, casualty management, and emergency repairs
to flood defences. This chapter considers these issues, together with the more gen-
eral topic of dealing with uncertainty in decision making during flood events.
10.1 Flood Event Management
10.1.1 Preparatory Actions
One of the key benefits of flood warnings is that, if time permits, a number of
actions can be taken in advance of flooding to reduce the extent of damage to prop-
erty and the risk to life. The key stages in issuing a flood warning have been
described in earlier chapters and include:
● Detection – of meteorological, river and coastal conditions (Chapter 2)
● Thresholds – identification of conditions likely to cause flooding (Chapter 3)
● Dissemination – issuing of flood warnings (Chapter 4)
● Forecasting – prediction of future river and coastal conditions (Chapters 5–8)
If sufficient time is available, warnings are normally escalated from an initial advi-
sory that flooding is possible, through to a full flood warning.
The receipt of an advisory or warning is usually the trigger to activate a flood
emergency plan (if one exists), and to commence preparatory actions. As described
K. Sene, Flood Warning, Forecasting and Emergency Response, 231
© Springer Science + Business Media B.V. 2008
232 10 Response
in Chapter 9, these can include mobilisation of staff and equipment, establishing a
command centre, and issuing of initial warnings to the media and public. Pre-positioning
of people and equipment can also be started, particularly if it is thought likely that
access routes may be cut off by floodwater. For example, emergency services
increasingly use temporary or mobile command centers (e.g. Fig. 10.1) and these
need to be placed in locations suitable for local communications, and clear of any
likely flooding or access problems.
Various actions can also be taken to reduce or even prevent flooding,
including:
● Emergency works – reinforcement of weak spots in flood defences, and at
locations where existing river or coastal works are underway (and patrols to
inspect defences and other structures), clearance of drains and blocked
watercourses
● Temporary defences – raising temporary or demountable barriers, placing sand-
bags along flood defences and river banks, and at individual properties
● Flow control operations – diversion of river flows, closing (or opening) gates,
emergency draw-down of reservoirs etc.
The time available to prepare depends on the type of flood event. For the case of
flash floods from rainfall, ice jams and landslides, only a few hours at best may be
available. However, for events such as floods in the lower reaches of large rivers, or
storm surge, a day or more of warning may be possible. For tropical cyclones,
typhoons and hurricanes, sometimes up to a week of advance warning can be pro-
vided, although with considerable uncertainty about the location, timing and severity
of the event at that lead time.
Fig. 10.1 Federal Emergency Management Agency (FEMA) Incident Command Centre at a
flood event; Kingfisher, Oklohoma, August 20, 2007 (FEMA/Marvin Nauman)
As noted in Chapters 5 and 9, the actual time available is reduced by various
time delays in the detection, forecasting and warning process, and these delays
should be considered in planning actions as the event develops, and in particular
any ‘points of no return’ beyond which a course of action (e.g. evacuation) becomes
unfeasible. In some types of event, the numbers of people to consider can be large;
for example, during Hurricane Katrina in the USA in 2005, more than one million
people are estimated to have left their properties in advance of landfall and, during
1995 in the Netherlands, a precautionary evacuation of more than 200,000 people
took place in advance of a forecast flood event (although the subsequent flooding
was not as serious as expected). The issue of false alarms also needs to be consid-
ered, which can lead to unnecessary costs (e.g. closing down businesses or indus-
trial plant), and can also incur a risk (e.g. if hospital patients are moved). These
considerations of optimum decision making, and flood warning performance, are
discussed further in Section 10.2 and Chapter 11.
Another factor to consider is the cooperation likely to be received from people
in flood affected areas; for example, will people leave their properties if asked to,
what assistance can people provide in helping neighbours and the emergency serv-
ices, and will people interpret warnings correctly? These questions often cannot be
resolved during the pressure of a flood event, highlighting the important role that
community engagement and awareness has to play in maximising the effectiveness
of emergency response (see Chapter 9). In particular, studies have shown that the
format and wording of messages is crucial and this topic is discussed further in
Chapters 4 and 11.
During the onset of a flood event, several actions can also usefully be taken
(if resources are available) to help with the subsequent post event analysis
including:
● Photography – taking photographs and videos of the flood extent from the
ground, and from helicopters and aircraft (if available)
● River gauging – making spot measurements of high flows to help in calibrating
river monitoring equipment (see Chapter 2)
● Flood inundation sensors – reading staff gauges, noting maximum levels, and
sometimes installing equipment (e.g. maximum level recorders) to record the
flood depths reached (see Chapter 2)
Many hydrological services routinely perform tasks of this type during flood
events, with the staff requirements and procedures written into warning procedures
to ensure that they are not overlooked.
Similarly, decisions to pre-position people and equipment in locations likely
to be cut off by flooding are much easier to take if the requirements have been
identified and agreed in advance. For example, there may be a need to preposi-
tion a high volume pump or fire truck at a location expected to flood severely,
even though minor flooding is already occurring at another location where the
equipment is also needed. If a risk-based assessment has already been per-
formed at the planning stage, then this helps to avoid local conflict when the
decision needs to be taken.
10.1 Flood Event Management 233
234 10 Response
Much of the focus in the run up to the onset of flooding is also on avoiding the
need to rescue people, pets or livestock. Compared to other alternatives, such as
evacuation, or installing temporary barriers, rescues can be risky and time con-
suming, and limit the resources available to respond to other problems as they
arise. For example, in the UK, procedures can require a team of five to six people
to rescue one person (or more) trapped in fast flowing river water, including one
person upstream to spot debris flowing towards the rescue site, two people down-
stream to spot the casualty if they are swept away by water, and two to three peo-
ple on site for the rescue itself (e.g. by boat or rope). Similarly, rescues from
vehicles can be hazardous, and survival rates are much lower than for people
trapped in property, so closing roads early to avoid the situation arising is the
preferred option.
Once flooding has started, and flood warnings have been issued, the role of flood
warning and forecasting starts to assume less importance, although forecasting
models (see Chapters 5–8) can continue to assist throughout the event in answering
questions on the likely magnitude and timing of the peak, flooding extent, and
when waters are likely to drop, and roads can be reopened and people allowed to
return to their properties.
10.1.2 Timelines
One widely used concept in emergency response is that of the event timeline.
This describes the sequence of incidents, emergency calls, responses, actions etc.
which occur during an event. Timelines are used in post event assessments of
response, and may also be available in real time to assist other responders in
understanding the situation. Modern incident management and decision support
systems can automatically log occurrences from a wide range of sources and
organisations, and make this information available to all responders over secure
websites (see Section 10.2).
As an illustration of a timeline, Fig. 10.2 provides a hypothetical example, from
the perspective of a flood warning service, for a short lived flash flooding event
affecting a small town called Newtown during the early hours of the morning. The
roles, staff job titles, actions etc. will vary widely between organisations and spe-
cific flooding events so this example is just for illustration. The sequence would
also continue into post event actions, reporting, assisting residents to return to prop-
erties and assess and repair damage, reopening roads, debriefs etc. Also, this event
proceeds predictably with no problems arising such as flow control structures not
operating as expected, or problems being encountered with contacting key people,
or access to equipment etc.
As another example, Box 10.1 shows the actual timeline for a flash flooding
event, with a focus on the emergency response (adapted from North Cornwall
District Council 2004). The event occurred in the village of Boscastle in South
West England in the summer of 2004.
02:00 The 2 am routine discussion with the duty weather forecaster sug-
gests a general risk of heavy rainfall in the region, although it is
difficult to be specific on locations at this stage
02:45 Flash warning received of a 70% probability of heavy rainfall
over the Newtown catchment in the next 1–2 hours. Discussions
with the duty weather forecaster suggest that a major rainfall
event is possible in the catchment
02:55 Duty officer completes reviews of weather forecast, weather
radar, raingauge, river level and flood forecasting model outputs
(including two model runs for different rainfall scenarios)
03:00 Duty officer issues a Flood Watch advisory message, incident
room opened, duty and operations managers informed by tele-
phone at home
03:15 First heavy rainfall recorded by a raingauge in the catchment
03:20 Duty manager arrives at the incident room
03:25 Briefing of duty manager completed; discussions with the town
emergency planning officer. Operations manager arrives.
03:30 Emergency workforce instructed to deploy to Newtown
04:10 Flood warning threshold levels reached; flood warning formally
issued to the properties at risk, the local authority and police; loud
hailer patrol started
04:15 Emergency workforce completes closing of the two flood gates in
Newtown
04:15 Hydrometry team deployed to Newtown gauge for high flow spot
gaugings
04:30 On-site briefing between local authority and police representa-
tives, and the emergency workforce
04:35 Flood incident declared; police control centre established, desig-
nated evacuation centre opened, social service and voluntary staff
called onto site
04:50 Duty weather forecaster informs flood warning duty officer that
rainfall should stop in the catchment within the next 20–30
minutes
05:10 Completion of evacuation of the 125 properties likely to be
affected; nursing home evacuated
05:40 Area likely to be affected cleared of all people and vehicles; main
access roads closed to the public. Sandbagging of properties com-
pleted where feasible
06:10 Flood waters start to overtop river banks
06:40 Flood levels reach a maximum depth of 1 m; 25 properties
affected; hydrometry team completes highest flow gauging
achieved to date at Newtown gauge
07:30 Flood levels falling rapidly at Newtown gauge; local authority
advised that the flood threat is receeding
Fig. 10.2 Illustration of a flood event timeline up to the time that flood levels start to drop from
peak values
236 10 Response
Box 10.1 Boscastle flood event 2004
The village of Boscastle is on the north coast of Cornwall at the bottom of the
23 km2 catchment area of the rivers Valency and Jordan. On 16 August 2004,
in mid summer, the village experienced its worst flooding on record. The
peak hourly rainfall in the area exceeded 80 mm at one raingauge, with a 24
hour total of about 200 mm. Swift action by local people and the emergency
services meant that no lives were lost and there was only one minor injury.
Approximately 1,000 residents and visitors were affected by the event, with
about 100 people rescued by helicopter from rooftops, cars and trees, 58
properties flooded, and 116 cars swept out to sea. Roads, sewers, bridges and
other infrastructure were badly damaged (Fig. 10.3).
Some key actions in the timeline for the event were:
1215 Reports of heavy rainfall in the upper catchment but none in the
middle or lower reaches
Fig. 10.3 Helicopter rescue in main street of Boscastle (Pam Durrant; text based on
a description by Heulyn Lewis, North Cornwall District Council)
(continued)
10.2 Decision Support Systems
Information is a key requirement during flood events, particularly widespread
events, and computer systems are increasingly being used to assist in guiding the
response. Some particular characteristics of decision making in emergencies
(MacFarlane 2005) include uncertainty, complexity, time pressure, a dynamic event
that is innately unpredictable, information and communication problems (overload,
paucity or ambiguity), and the heightened levels of stress for participants, coupled
with potential personal danger, whilst some general advantages of using Geographical
Information Systems (GIS) during emergency incidents include:
● Support for tasking and briefing
● Producing hard copy maps which remain a key information product for respond-
ers and planners
● Integrating data from multiple sources that may flow in during the course of an
emergency
● Developing a Common Operational Picture for multi-agency staff
● Supporting two way flows of information through mobile GIS
● Assessing likely consequences and supporting forward planning
Box 10.1 (continued)
1230 Heavy rainfall starts in the middle reaches of catchment
1415 Rain still persists but seems to be easing
1500 Heavy persistent rain starts again
1515 River Valency approaching bankfull flow
1530 River starts to spill over bank
1535 First call received by fire and rescue service
1545 Visitor car park in village starts to flood
1546 Call to coastguard from local representative saying that the river
has risen about 2 m in the past hour
1600 1–3 m wall of water sweeps across visitor car park trapping peo-
ple in the visitor centre
1603 Helicopter rescue coordination centre put on ‘standby’
1617 Inshore lifeboat launched
1630 All access roads closed by police
1636 Air ambulance put on standby
1700 Flood approaching peak levels
1712 Major incident declared by emergency services
1723 First helicopter winch rescue completed
1755 Two regional hospitals put on standby
2000 Water levels back within river banks
2100 Helicopters start returning to base
10.2 Decision Support Systems 237
238 10 Response
● Managing assets and resources for current and projected future demands
● Keeping the public and other affected parties informed through internet or
intranet mapping systems
● Establishing one element of an audit trail
● Supporting the transition to recovery with a baseline database that also integrates
a full picture of the emergency itself
Here, mobile-GIS is a term used to describe hand-held Personal Digital Assistant
(PDA) or laptop devices able to display maps, possibly linked to real time data
feeds showing the present locations of key assets. Vehicle and personnel tracking
systems using Global Positioning System (GPS) devices can also be used to relay
information on, for example, the positions of rescue helicopters, and of fire and
police vehicles and other assets. This approach is used by some fire service mobile
command centres, for example.
Systems may be designed specifically for flooding, or form part of a wider all-
hazards approach. This includes the concept of Virtual Emergency Operations
Centres for major events, making extensive use of interactive GIS displays for inci-
dent scenes, video links, emergency response assets (pumps, generators etc.), satel-
lite and other imagery, with automated sharing of information between vehicles,
personnel and command centres (e.g. Prendergast 2007)
Compared to the flood risk mapping applications described in Chapter 2, and the
planning role described in Chapter 9, there are of course a number of issues to con-
sider if relying on computerised systems for information and situational awareness
during an event, including the availability of trained operators, resilience to com-
munications and power failures, and the likely availability of sufficient information
in real time. It may also be useful to develop interfaces to other systems, such as
telemetry systems and flood forecasting models, to provide updates on current and
future estimates for flood extent. Some examples of real time use of GIS in flood
related applications include:
● The Shelter Navigation System – a map based system which allows authorised
staff to monitor the status of designated shelters during a hurricane, including
keeping track of evacuees, and which allows the public to look up the shelters
closest to their home (South Carolina Department of Health and Environmental
Control).
● Surrey Alert – a map based system for sharing information during emergencies
of all types, including flooding, in the county of Surrey in the UK. The system
consists of a secure extranet, accessible only to local authorities, the emergency
services and other responders, and a publicly available website, which includes
news, media and travel updates, and the option to show flood warnings in force
on a map. The extranet component includes an emergency contacts database,
information on key facilities (medical, control centers/emergency operations
centres, rest centres etc.), and an incident log, giving times and descriptions for
actions and decisions taken during the event (http://www.surreyalert.info).
● Flood Forecasting Systems – modern forecasting systems (see Chapter 5) also
increasingly include the facility to map flood inundation in real time, including
running ‘what if’ scenarios (e.g. for defence breaches, and structure operations)
and sometimes the option to generate lists of properties at risk for output to
paper based or automated warning dissemination systems, thereby having many
of the characteristics of a GIS system.
Whilst Geographical Information Systems might be viewed as a simple type of
decision support system, more sophisticated systems are available which provide
guidance on actions to take during an emergency, and these can be used both at the
planning stage and during an event. The level of guidance provided can range from
presentation of information in a range of formats (overlays, visualisation etc.)
through to recommendations on optimum response strategies.
These types of system are used in the oil and gas, chemical and nuclear indus-
tries, for example, and often combine the outputs from sophisticated computer
models (e.g. of gas dispersion) with optimisation algorithms or logical rules (for
example, IF the spill is ACID and the ACID is in the GASEOUS state THEN…..),
and links to external databases on equipment characteristics and histories, person-
nel records, emergency checklists and other information. Recent developments in
this area include the use of probabilistic techniques, and artificial intelligence meth-
ods, such as artificial neural networks, fuzzy logic and genetic algorithms. Internet
based applications are also being investigated, allowing decision support/expert
systems to ‘learn’ and update rules from the vast amount of information available
on the world wide web.
Several systems of this type are also under development for use in flood emer-
gency applications, although their use is not yet widespread. The base data (static
data) can include information on:
● Flood defence locations and geometry
● River control structures
● Topography
● Property and census information
● Locations of vulnerable people who may require assistance
● Critical assets such as power stations, water treatment works, telecommunica-
tion hubs, and associated supply/communication networks
● Transient populations such as at campsites, caravan parks, hotels
● Emergency equipment depots (sandbags, earthmovers etc.)
● Fixed emergency response assets (fire and police stations, command centres etc.)
● Medical facilities (doctors, pharmacies, hospitals etc.)
● Emergency shelters and safe areas for escape
● Road network characteristics (evacuation and access routes, capacities etc.)
● Background mapping (rivers, coastlines etc.)
whilst dynamic data, updated during a flood event, can include:
● Water depths, velocities and flows
● Flood defence condition including the locations and size of breaches
● River control structure settings
● Temporary defences (barriers, sandbags etc.)
10.2 Decision Support Systems 239
240 10 Response
● Locations of vehicles, personnel and other resources (e.g. helicopters, boats)
● Traffic density and flow data
Systems may include data gathering and modelling components, or import this informa-
tion from external sources, such as flood forecasting models, telemetry systems, and
dynamic traffic flow models. Dynamic scenarios have the advantage of making use of
the latest information, with the potential disadvantages of lack of real time information
due to communications failures, model uncertainties, and the risk of models failing to
operate correctly, or outputs being misinterpreted by non-experts. Alternatively, stored
scenarios may be used, and have the advantage that many more options can be consid-
ered than would be feasible during a flood event (e.g. for defence/levee/dike breaches,
time of day or day of week, or traffic routing), possibly also using more sophisticated
modelling approaches (e.g. two-dimensional rather than one-dimensional flood mod-
els). Scenarios can also be reviewed and audited before use in a real event although, as
in many aspects of flood modelling, with the difficulty of a lack of calibration and vali-
dation data for the more extreme events (e.g. mass evacuations, widespread flooding).
Perhaps the most widespread application which has been considered to date has been
for evacuation planning and management. The aim of evacuation modules is to provide
guidance on how to optimise the number of people who leave an area in a given time,
taking account of the likely delays due to weight of traffic, and routes and safe havens
becoming inaccessible due to flood water, together with the access requirements of the
emergency services. For example, in the USA, systems which are available (e.g. Fu
2004; Wolshon et al. 2005) include:
● Hurrevac – a GIS based software package developed on behalf of FEMA by the
US Army Corps of Engineers that has been used since 1988 by government
emergency managers to help with managing major evacuations during hurri-
canes, and includes import of surge scenarios from the National Weather
Service, a shelter module, a traffic and evacuation tool, the facility for what-if
scenarios for hurricane track, winds etc., and the option to display data from
river and tide gauges. The system also allows hypothetical storms to be input for
use in training exercises (http://www.hurrevac.com).
● Evacuation Traffic Information System (ETIS) – a web-based hurricane evacua-
tion travel demand forecast model for which inputs include the category of hur-
ricane, tourist occupancies, and anticipated percentages of people leaving
property and arriving in given locations. Outputs include estimates for traffic
volumes, cross-state flows of traffic, locations for congestion, and the numbers
of vehicles arriving at specific locations (http://www.fhwaetis.com).
More generally, dynamic approaches to traffic management are increasingly being
explored, including use of contraflow systems and intelligent transport systems (e.g.
Wolshon et al. 2005). In other applications, Rodrigues et al. (2006) describe a proto-
type internet based decision support tool called DamAid to assist emergency managers
with evacuation of properties and emergency response in case of dam failure, whilst
Simonovic and Ahmad (2005) describe a prototype computer simulation module for
flood evacuation planning for areas at risk from flooding in the Red River basin in
Canada, combining social, psychological, policy and other factors.
To allow for applications other than evacuation, increasingly a modular approach
is being adopted to development of decision support systems, including modules for
advising on evacuation routes, responses to defence breaches, likely casualties,
search and rescue strategies, management of the recovery phase (e.g. return to prop-
erties), and emergency response (e.g. road closures, placing sandbags, reinforcing
flood defences). For example, Van der Vaat et al. (2007) propose a methodological
framework for flood event management decision support systems consisting of haz-
ard, exposure, vulnerability, consequence and risk modules, with inputs from man-
agement response, external driver and tools modules. Systems may also use
web-based displays of data, with dynamic generation and updating of emergency
response plans, and the facility to view plans at a range of scales (national, local, site
specific e.g. power stations). The relative importance attached to the evacuation
component depends to some extent on the scale and rate of development of a flood
event; for example, in a widespread slowly developing coastal event, the focus may
be on evacuation whilst, in a flash flood, the focus may be on optimum access routes
for the emergency services, search and rescue operations, and management of the
recovery phase.
Table 10.1 summarises several examples of research and operational projects
which have developed (or are developing) decision support tools for flood event
management.
Table 10.1 Examples of research and operational decision support tools for flood event
management
Features or
Project Country applications Reference
ANFAS China, France, Flood forecasting Prastacos
Slovakia, component et al. (2004)
Greece, UK
DAMAID Portugal Dam break Rodrigues
forecasts et al. (2006)
FLIWAS The Netherlands, See Box 10.2 See Box 10.2
Germany,
Ireland
GDH The Netherlands Stored scenarios Flikweert et al. (2007)
NISFCDR China Flood forecasting Huaimin (2005)
component
OSIRIS France, Poland, Scenarios Erlich (2007)
Germany
PACTES France Flood forecasting Costes et al. (2002)
component
PREVIEW Europe-wide All hazards PREVIEW (2007)
system
RAMFLOOD Spain, Germany, Artificial neural http://www.
Greece and network cimne.upc.
three others es/ramflood/
Web GIS China Emergency Zhou et al. (2004)
levee repairs
10.2 Decision Support Systems 241
242 10 Response
Box 10.2 provides a more detailed description for the FLIWAS system, devel-
oped as part of a major collaborative study between several organisations in the
Netherlands, Germany and Ireland.
Box 10.2 Flood Information and Warning System (FLIWAS)
The Flood Information and Warning System (FLIWAS) is an information and
communication system to assist operational personnel, coordinators, information
services and decision makers in flood event management (e.g., Langkamp et al.
2005). The system is one of the outputs from the Interreg IIIB-Project NOAH
and is coordinated by the Foundation for Applied Water Research (STOWA) in
the Netherlands in collaboration with several Dutch, German and Irish organisa-
tions, and project observers from France, Poland, Scotland and Germany.
FLIWAS is internet-based and is designed to collect and process information
to assist in managing actions before, during and after a flood event. Typical
actions might include the activation of special dike watches, reinforcing of
flood defences, communication with other organisations and the public, and
preparatory measures for evacuation. Target users include water managers,
local, regional and national authorities, civil protection units, and the general
public and media. The system has been developed in close consultation with
potential users and builds upon several existing systems in Germany and the
Netherlands, including the High Water Information System (HIS) developed
by the Ministry of Transport, Public Works and Water Management in the
Netherlands (Ritzen 2005).
Real time information can be imported from a wide range of sources, including
observations of water levels and flows, weather radar, satellite and helicopter data,
rainfall forecasts, and the outputs from flood forecasting models. Emergency
Fig. 10.4 Flood defence overtopping (FLIWAS News, August 2007, http://www.
noah-interreg.net/)
(continued)
Box 10.2 (continued)
response plans can be defined and viewed for individual locations and organisa-
tions, and updated dynamically as an event occurs, with all actions taken logged
automatically, and locally applicable information available in the field on palm-
top computers and Personal Digital Assistants (PDAs). Standard situation reports
can also be predefined, and updated during the event, and can be sent automati-
cally to selected persons and organisations. Selected information can also be
made available for viewing on the public access website. The system can also
replay historical events for use in training and flood response exercises.
The baseline information required includes data on the geometry and con-
dition of flood defences, action plans, key contacts within organisations,
flood hazard maps, and information on properties at risk, critical locations
such as hospitals and nursing homes, livestock, transport and diversion routes
(e.g. railways), emergency response assets (e.g. sand depots, machinery, fire
trucks), and sites with dangerous substances.
The system is generic with the first modules to be developed covering gener-
ation and monitoring of calamity plans, resource management (people,
equipment etc.), damage and casualty assessment, and evacuation planning.
Fig. 10.5 Illustration of information flow within FLIWAS (http://www.noah-
interreg.net/)
10.2 Decision Support Systems 243
(continued)
244 10 Response
The issue of how to present and handle uncertainty is also increasingly being
considered, with one related example being the applications for optimisation of
hydropower operations discussed in Chapter 8. More generally, flood related
decision support systems might be linked with information systems managed by
other organisations, such as traffic management, communications, and power
supply systems, and guidance provided on likely consequential effects from fail-
ure of any component (e.g. traffic hold-ups, cell phone failures, power cuts).
10.3 Dealing with Uncertainty
One of the challenges in responding to a flood event is the uncertainty both in cur-
rent conditions, and what will happen next. The need to appraise recipients of the
uncertainty in warnings is also widely emphasised (e.g. World Meteorological
Organisation 1994; Emergency Management Australia 1999; Martini and De Roo
2007). Some typical sources of uncertainty which have been discussed in other
chapters include:
● Flood risk – uncertainty in the locations at risk from flooding (see Chapter 1)
● Detection – uncertainty in observations and forecasts of rainfall and other mete-
orological conditions, river levels, tidal levels, river flows, reservoir levels etc.
(see Chapter 2)
● Flood forecasts – uncertainty in estimates for the likely magnitude, timing and
extent of flooding, particularly for extreme events outside recent experience (see
Chapters 5–8)
For the flood response component, there can be additional uncertainties in the spe-
cific risks to people and property during an event, in how people will respond to
warnings, and in which secondary risks, such as power failures or communication
breakdowns, might occur. Time of day or year is also a factor, with people better
able to cope with flooding on a warm, summer day than in winter, and to respond
to warnings during daytime than in the middle of the night. Factors such as traffic
flows, numbers of properties occupied, and the ability to contact people, will also
vary within and outside normal business hours.
Box 10.2 (continued)
FLIWAS will be completed with an evacuation module to advise on opti-
mum evacuation routes and likely travel times, and to help with difficult
choices such as whether there is enough time for people to leave an area, or
they should be advised to move to shelters of last refuge (e.g. high rise build-
ings). Other modules will be included as required for individual applications,
such as for assessing the impact of defence breach scenarios.
Decision support systems can assist by helping to improve situational awareness
during an event, and providing guidance on optimum decisions, whilst flood fore-
casting models, possibly also including real time inundation mapping, help with
understanding likely future conditions. However, information on uncertainty can
also be useful when deciding how to respond to a flood event and some potential
users of this information include:
● Flood forecasters – in deciding on the accuracy of the forecast, and what mes-
sage to convey to people who will use the forecast
● Expert users – who can use probabilistic information directly for input to their
own decision support systems
● Emergency response organisations – if they have the appropriate skills and train-
ing to interpret information on uncertainty
● The public – who may find some types of probabilistic information useful
For the first user group, flood forecasters, scenario and ‘what if’ modelling tech-
niques have been widely used for many years, whilst probabilistic and ensemble
forecasting techniques for both river and coastal flood forecasting are actively
under development in several organisations, as discussed in Chapter 5.
For expert users, earlier chapters provide several examples of the use of probabilistic
information to assist in flood response, including for hydropower applications (see
Chapter 8), and emergency management for polder areas in the Netherlands (see
Chapter 3). The basis of the techniques used is often to compare the cost of taking action
with the expected losses if no action is taken. The range of outcomes can be summarised
in a contingency table, as illustrated by the simple example shown in Table 10.2.
If the probability of flooding is p then, over the long term, taking mitigating
action is cost effective if the cost of taking action C is less than pL, or p > C/L. The
so-called cost/loss ratio then provides an indication of the forecast probability
above which action should be taken, with the expectation that following this strat-
egy over a number of events will result in total costs being less than total losses.
More complex formulations can also be devised, for example taking account of
partial mitigation of losses, and the costs which would have been incurred anyway
in the absence of the event (e.g. Pierce et al. 2005).
For flood warning applications, costs and losses are also dependent on lead time;
for example, apart from the risk to life, it is usually more expensive to rescue people
from properties than to evacuate them, and property damage is also usually higher
when people have less time to prepare. Probability thresholds might therefore be
defined for different stages in the run up to an event, perhaps linked to different
flood warning stages (flood watch, flood warning etc.), with cost-loss relationships
changing during the event (e.g. Roulin 2007).
Table 10.2 Decision table for cost loss approach example
Flooding No flooding
Mitigating action Cost (C) Cost (C)
No mitigating action Loss (L) No cost
10.3 Dealing with Uncertainty 245
246 10 Response
More generally, a utility or penalty function approach may be taken for situa-
tions where costs or losses cannot easily be defined in monetary terms (e.g. intan-
gible losses) or the relationship between costs and losses is more complex. Utility
(sometimes called Value) is a concept widely used by economists, in which typi-
cally a numerical score or ranking is given to the required outcomes, based on
expert judgement, and possibly weighting of different outcomes (with weights
elicited from multi-criteria analysis). Functions may be defined in such a way that
severe outcomes, such as a dam overtopping, have exceptional scores, so that the
optimum solution is steered away from or prevented from reaching that situation.
The issue of how to present uncertainty information to emergency responders, and
the public, is a new and developing area and one that has received much attention by
meteorological, hydrological and social and behavioural researchers (e.g. National
Research Council 2006; Demeritt et al. 2007; Demuth et al. 2007). Some conclusions
from these types of study can include (e.g. Environment Agency 2007):
● The best way to present information will vary between users, depending on their
interests, technical expertise and roles.
● Given the wealth of information available, several alternative types of presenta-
tion may be useful, focussing on spatial, site specific and temporal information.
● Approaches should be simple and intuitive (at least in the initial stages), although a
demand often arises for more sophisticated approaches as skills and experience
develop.
● Forecast products are best developed as a joint exercise between forecasters and
end-users.
In particular, the importance of consultations with end users is often emphasised (e.g.
using focus groups, pilot studies), as is the value of working with probabilistic fore-
casts operationally to gain an intuitive feel for how to interpret them. In addition,
Collier et al. (2005) note that there is a need to keep a clear distinction between the
needs of hydro-meteorological services and flood emergency operations. In the
former case the interest is in getting the best possible forecast, whereas in the latter
case the interest is in making the best possible decision.
Requirements can range from the wish for a simple yes/no answer on whether
to perform an action, through to the need for a detailed understanding of the risks
involved in taking a decision, and a wish to see all of the information that the fore-
caster has available, including that on uncertainty in forecasts; for example, in the
following situations:
● Mass evacuations – an emergency manager needs to balance the risks of an unneces-
sary evacuation against the risks of failing to evacuate properties if flooding occurs.
There is also the trade off between waiting to make a decision, by which time the
forecast will hopefully be more certain, and the time lost for starting the evacuation.
For hurricane evacuations, the use of cost-loss approaches has been proposed to help
in optimising these time based decisions (e.g. Regnier and Harr 2006).
● Flood defence operations – in the run up to a flood event, emergency managers may
make decisions on the height to which defences need to be raised (e.g. using sand-
bags) to protect property or, in extreme situations, may decide to deliberately
breach parts of the defence network to flood, for example, agricultural land to pro-
tect towns and cities further downstream. The example of the Red River of the
North Flood in 1997 is often quoted as an example of a situation where information
on uncertainty could possibly have changed the course of the event (see Box 10.3)
Box 10.3 Red River of the North Flood, Grand Forks, April 1997
The Red River of the North catchment is located in North Dakota and
Minnesota in the United States and southern Manitoba in Canada. Following
record snowfall in the winter months, the most severe floods since 1826
occurred in April and May 1997 during the spring snowmelt (Glassheim
1997; Krzysztofowicz 2001; National Research Council 2003).
The flood affected the cities of Fargo and Winnipeg, and in particular the
two towns of Grand Forks, North Dakota, and East Grand Forks, Minnesota,
where levees were overtopped and floodwaters reached over 3 miles (5 km)
inland, inundating virtually everything in the twin communities and causing
almost US$4 billion in damages and affecting about 5,000 properties, although
fortunately with no lives lost. Nearly 90% of the area was flooded and three
neighbourhoods were completely destroyed.
Flood predictions were estimated using the seasonal forecasting technique
described in Chapter 8. The first forecast of a major event was issued nearly
2 months before, on February 28, with a peak forecast of just under 49 ft
under average precipitation conditions. This value was subsequently revised
to 50 ft on April 14, then 52 and 54 ft on April 17 and 18. In Grand Forks and
East Grand Forks, a previous major event had reached a peak of just under
49 ft and, based on the available information, city officials decided to prepare
the city for a 52-ft river crest, whilst the actual peak reached exceeded 54 ft
(National Research Council 2003). One comment following the event was
that “If someone had told us that these estimates were not an exact science, or
that other countries predict potential river crest heights in probabilities for
various levels, we may have been better prepared.” (Glassheim 1997;
Krzysztofowicz 2001).
As with many flood events, post event studies showed a range of technical,
communication and organisational issues which were quickly addressed, and
of course the science of flood forecasting has improved significantly since the
time of that event, including major improvements in flood forecasting
techniques.
However, the event was influential in changing views on including informa-
tion on uncertainty in forecasts, and on the interactions between forecasters, the
public and decision makers, with one conclusion (National Research Council
2006) being that unclear communication of uncertain forecast information can
hinder decision-making and have significant negative consequences.
10.3 Dealing with Uncertainty 247
248 10 Response
Public awareness of techniques is also likely to increase as products become more
widely available. For example, some meteorological services make information on
uncertainty freely available on the internet and other media (e.g. television), with
examples including:
● Hurricane strike probability and cone of uncertainty forecasts – used by the
National Weather Service in the USA to show the estimated spread of estimates
in forecasts for hurricane tracks
● Rainfall forecast plumes – the approach used in the Netherlands by some televi-
sion broadcasters of showing the range of estimates derived from Numerical
Weather Prediction models
● Rainfall probabilities – maps for the percentage probability of heavy rainfall
presented in weather forecasts on some meteorological service websites
Qualitative guidance, using terms such as ‘may’, ‘probably’ and ‘likely’, is also
suggested in some applications, provided that messages also include advice on the
actions to take (e.g. Emergency Management Australia 1999), whilst ISDR (2006),
in the context of all-hazard warning systems, give the examples of using phrases
such as “if present conditions continue…” or that “there is an 80% chance
that…”.
In Chapter 5, it was also noted how some recipients of warnings might, given the
option, choose to receive warnings at lower levels of risk, defined as the combina-
tion of probability and consequence, and modern automated flood warning dissemi-
nation systems have the capability to target warnings to individual property owners,
if required. Some examples of situations where targeted warnings might be useful
include (e.g. Environment Agency 2007):
● Local authorities who can close riverside and coastal footpaths to walkers
● Large businesses who can prevent customers parking in areas at risk
● Outdoor event organisers who can reschedule or relocate an event
● Farmers who can move livestock between fields
● Residents in frequent flooding locations who can install flood boards or
sandbags
● Emergency managers who can plan staff rotas and check that equipment is
ready
● Operators of temporary defences who can mobilise staff and equipment
● Hospitals who can reschedule operations and alert staff to the possibility of
flooding
● Utility operators who can invoke contingency plans for flood events
Often, the interest is in receiving a personalised warning in advance of the official
warning or whilst uncertainty still remains high. Techniques from the field of risk
communication and perception can also assist in defining requirements. Some
operational considerations with this approach can also include deciding on who is
best placed to define risk thresholds, and to take risk based decisions, and how this
affects the relative roles and responsibilities between the forecasting and warning
authority, and the recipients of warnings.
Chapter 11Review
Reviews of flood warning systems are often required following major flood events,
and may form part of a programme of continuous improvement, sometimes linked
to performance targets for different aspects of the system. Performance monitoring
should ideally cover all aspects of the system, including detection, forecasting,
dissemination, and response to warnings, together with feedback from users on
satisfaction with the system. The lessons learned from post event assessments, and
recommendations from regular reviews, can then guide future investments, and pro-
vide baseline information for use in economic assessment and prioritisation exercises.
However, improvements need not necessarily require significant investment, and
much can be gained from improving operating procedures, and closer collaboration
between the various participants in the flood warning process, including communi-
ties and their representatives. This chapter discusses these various issues, and
highlights some common themes from earlier chapters on ways of improving flood
warning, forecasting and emergency response systems.
11.1 Performance Monitoring
Performance monitoring usually consists of a process of reviews, recommenda-
tions, implementation of findings, and continuous assessment to check that recom-
mendations are being acted upon, and improvements are being made. Also, flood
warning services increasingly need to demonstrate the benefits that they bring, and
that improvements are being achieved over time.
Routine reviews may be performed against benchmarks or targets, and of areas
which may have changed since the time of the last review (key staff, equipment,
procedures etc.). As noted in Chapter 9, regular tests and exercises can also help to
identify problems, and keep staff trained in use of systems. Reviews can be
performed for individual flood warning schemes, or on a regional, national, organi-
sational, or multi-agency basis.
Many organisations also routinely perform formal reviews of performance after
major flood events, and sometimes near misses, both to answer immediate questions
from the public, media, and politicians, and to identify improvements for the future.
K. Sene, Flood Warning, Forecasting and Emergency Response, 249
© Springer Science + Business Media B.V. 2008
250 11 Review
Two high profile examples were the various reports following Hurricane Katrina in
August 2005 (e.g. US House of Representatives 2006) and the December 2004
Tsunami event (e.g. ISDR 2006).
Ideally, each study should refer back to previous studies, and a record should be
maintained of findings over the years, which can be referred to periodically to
check that issues are not being overlooked, including changes which may influence
flood response and flood warnings (e.g. to flood defences, instrumentation, control
structures etc.). Some topics which are often covered in post event or lessons
learned reports include:
● Flooding causes – analysis of the meteorological and hydrological or coastal
conditions leading to the event, and an assessment of the severity of the event
compared to previous events
● Flooding impacts – assessments of the number of people injured, evacuated or
rescued, the number of properties flooded, and of infrastructure and businesses
affected, including a timeline for the event
● Flood warnings – summaries of the warnings issued, the lead time provided, and
the number of people who received and acted upon those warnings
● Flood forecasting – performance of any flood forecasting models during the
event (and related weather forecasts)
● Flood defence – performance of any flood defence and flow control struc-
tures, a summary of emergency works undertaken, and remaining problem
areas
● Coordination – assessments of the coordination between the various agencies
involved in the event, policy implications, and liaison with the media and the
public
● Recovery – actions taken to make property safe, restore utilities (if applicable),
remove debris, decontaminate sites, return people to their properties etc.
● Systems – a summary of how well detection, communication, telemetry, fore-
casting and other systems performed during the event
● Lessons learned – a summary of key findings and an action plan
The precise topics covered will depend on the scope of the review and organisa-
tional responsibilities. The review may also include interviews with residents, the
emergency services and community representatives, and the findings from site vis-
its to survey flood boundaries, and to inspect structures and flood defences.
Routine assessments may involve workshops and research studies to share best
practice, annual reporting against targets, and independent reviews of performance
both within the organisation, and externally in collaboration with other emergency
response organisations. Some questions on the content and delivery of messages
might include (Emergency Management Australia 1999):
● Did the target audience receive the warnings in time?
● Did they understand the warning message?
● Were their responses appropriate? If not, why not?
● What evidence is there for the answers to these questions?
For example, in England and Wales, surveys of flood warning recipients are com-
missioned both annually, and after major events, to determine how many people
received flood warnings, and what their opinion was of the information provided.
Targets, performance measures, indicators and/or benchmarks are another way
of driving improvements to performance, and are identified as an important com-
ponent in flood warning systems, and disaster management systems generally
(e.g. Handmer et al. 2001; Elliott et al. 2003; Andryszewski et al. 2005; Basher
2006). Some high level targets might be that there should be little or no loss of
life from flood events, and that damages and disruption from flooding should be
minimised.
More specific performance measures may relate to the individual components
in the flood warning process, such as the Accuracy, Reliability and Timeliness of
flood warnings, the number of properties receiving flood warnings, flood forecast-
ing accuracy, the frequency of emergency response exercises, the performance of
flood emergency plans, and the number of health and safety incidents. Chapter 3
presents some examples of statistical and other approaches to assessing the per-
formance of flood warnings, including contingency measures such as the
Probability of Detection, and False Alarm Ratios, whilst Chapter 5 discusses a
range of flood forecasting model performance measures. Various social response
factors may also be considered, related to the ability of people to respond to flood
warnings, their satisfaction with the warning service provided, and awareness of
actions to take.
When considering lead time targets, the values which are selected typically
depend on the type of flooding anticipated, and the detection, forecasting and
response systems which are available. For example, for tropical cyclones, typhoons
and hurricanes, events may become apparent several days in advance, and warnings
may be issued with 24 hours or more of notice. For coastal surge events from wide-
spread, less intense storms, a few hours or more may be possible whilst, for flash
floods, sometimes only a few minutes might be available. However, for the flash
flood example, small numbers of people in a mountain village might only need a
short time to move to the safety of higher ground whilst, for the tropical cyclone,
typhoon and hurricane example, a major evacuation of residents might take a day
or more. The speed and effectiveness of the response will also depend on the effi-
ciency of dissemination systems, the local capabilities of emergency services, and
on public awareness about how to respond to warnings. In estimating timeliness
(i.e. the time between issue and receipt of a warning), the various time delays in the
system also need to be accounted for as described in Chapters 5 and 9.
False alarm rates are another measure where views on acceptable values differ
widely (e.g. Barnes et al. 2007). If events only happen infrequently, then the occa-
sional false alarm may be viewed as beneficial as a way of maintaining public
awareness, and rehearsing and testing emergency response systems (e.g. Emergency
Management Australia 1999). Similarly, if a property floods frequently, the owner
may have well rehearsed procedures to protect the building (e.g. installing flood
boards, moving vehicles), and view the occasional false alarm as a small price to
pay for being able to continue living at that location. The analogy is sometimes
11.1 Performance Monitoring 251
252 11 Review
drawn with other types of warnings, such as fire alarms and bomb alerts, where
there can be a high public tolerance to occasional testing of systems and other false
alarms (provided that the reasons are explained). However, if the action on receiving
a warning is a widespread evacuation, or closing down of critical or expensive
installations (e.g. oil refineries), then a low false alarm rate is often desirable. False
alarms are caused in part by the various uncertainties in the detection, forecasting
and warning process, and this topic is discussed in more detail in Chapter 10.
When defining targets or performance indicators, the usual approach is to
consider each stage in the flood warning, forecasting and emergency response process,
and to devise suitable indicators of performance. Overall performance might then
be assessed using summary tables or graphs, or by combining values using weighting
or multi-criteria approaches. The information obtained can also help in understanding
the areas in which future investment will be most beneficial (see Section 11.2).
However, the choices made should consider the feasibility of collecting the support-
ing information required, the likely effort and costs, and how realistic it will be
maintain the performance monitoring system over a period of years. Also, whether
the best approach is to collect high quality information for a small number of indi-
cators, or more comprehensive but less complete information across a larger
number. Many different indicators could potentially be envisaged; for example,
some possible descriptors considered in one research study included preparedness, fore-
casting, warning and promoting response, other communication, coordination, media
management, equipment provision, environmental damage, economic damage, injuries,
loss of life, victim trauma and reputation (Environment Agency 2007).
The concept of levels of service might also be introduced as a way of monitoring
performance and assisting with the design of flood warning systems. For example,
for England and Wales, the Environment Agency (Andryszewski et al. 2005) uses
this approach to ensure consistency in the implementation of flood warning
schemes, and that schemes are prioritised according to risk, calculated from the
probability of flooding and the number of properties at risk. The approach defines
maximum, intermediate and minimum levels of service provision in the following
areas according to the level of risk:
● Detection and forecasting – requirements for raingauge and weather radar
coverage, river gauge network density, and for false alarm rates and probability
of detection
● Dissemination – recommended methods for providing indirect and direct
warnings, including community level methods (e.g. sirens, loudhailers), and
warnings via the media
● Public communications – recommended actions according to the level of risk,
including direct mailing (at least annually, or up to every 3 years), public notices
on noticeboards and in local media, and other activities, such as flood fairs,
leaflets, newsletters, displays in libraries and newspaper articles
Some studies have also used reliability analysis techniques of the type described
briefly in Chapter 9 to examine the various trade offs between indicators such
as flood warning lead times, the success rate of warnings, and false alarm rates.
For example, Krzysztofowicz et al. (1994) describe a Bayesian approach applied to
two case studies in Pensylvania, and Norouzi et al. (2007) describe application of
a similar approach to assessing the reliability of a flood warning system in Iran.
11.2 Performance Improvements
Post event reviews, and regular performance monitoring, can lead to a range of rec-
ommendations for improvement, which might include:
● Detection – improvements to meteorological, river and coastal observations (e.g.
site locations, types of instrument, network density, resilience, accuracy), and
meteorological forecasts
● Thresholds – revision of threshold levels to reduce false alarm rates, improve
success rates, provide more lead time, provide backup in case of failure etc.
● Dissemination – improvements to systems and procedures to increase the pro-
portion of people at risk receiving warnings, change the wording of messages,
update flood risk assessments, raise public awareness etc.
● Forecasting – improvements to models and systems, such as recalibration of
models, use of more sophisticated models, use of data assimilation, use of
ensemble techniques
● Preparedness – revisions to flood emergency plans, more frequent and detailed
emergency response exercises etc.
● Response – improvements to inter-agency coordination, information and com-
munication systems, liaison with the media etc.
In the integrated or total flood warning system approach, all components need to be
improved if the ultimate aim of minimising risks to people and property is to be
achieved (e.g. Emergency Management Australia 1999; Andryszewski et al. 2005).
More general requirements may also be identified at an organisational or
national level; for example, the need to extend the flood warning service to new
locations, to introduce greater consistency in procedures, and to provide warnings
for additional types of flooding, such as urban flooding, or for fast response
catchments. The decision may also be made to introduce or improve flood warn-
ing targets and improved performance monitoring systems to help to drive future
improvements. International reviews and comparisons may also highlight poten-
tial changes; for example, Parker et al. (1994) used the following 14 criteria to
compare flood forecasting, warning and response systems (FFWRS) between
several European countries: flood warning philosophy, dominance of forecasting
vs warning, application of technology to FFWRS, geographical coverage, laws
relating to FFWRS, content of warning messages to public, methods of dissemi-
nating flood warning, attitudes to freedom of risk/hazard information, public
education about warnings, knowledge of FFWRS effectiveness, dissemination of
lessons learned, performance targets and monitoring, national standards, and
organisational culture.
11.2 Performance Improvements 253
254 11 Review
Opportunities may also exist to widen the scope of the flood warning service
into other types of forecasting and warning, and to share costs with other organisa-
tions or departments with common interests. For example, the requirement for real
time monitoring of rivers and coastal waters is often shared by other groups with
an interest in, or responsibility for, managing water resources, pollution incidents,
navigation etc. Similarly, integrated catchment forecasting models can be used for
forecasting across the full range of flows, including drought forecasting (see
Chapter 8, for example), whilst the requirements for decision support systems, GIS
systems, dissemination systems, and other emergency response equipment are com-
mon to many types of natural and technological hazard.
The range of possible areas for improvement is huge and it is only possible to
discuss a few common themes here. The various guidelines and reviews cited in
previous sections and chapters provide more information on potential ways of
improving aspects of the flood warning, forecasting and emergency response
process.
11.2.1 Detection
Improvements in detection can include monitoring at more locations, using new
techniques and technology, and improving existing instruments and telemetry
systems.
For flood warning applications, optimum network densities depend on the type
of flood response, the level of risk, and other factors, such as the need to provide
information for flood forecasting models, and backup instruments in case of failures.
Chapters 2 and 5–8 discuss some of these issues further. There can also sometimes
be advantages in sharing instrumentation across more than one flood warning
scheme to reduce costs, for example by installing raingauges near to catchment
boundaries if there are flood risk areas in two adjacent catchments. Network densi-
ties might also be linked to required levels of service and flood risk (e.g.
Andryszewski et al. 2005; Sene et al. 2006).
Technological developments (see Chapters 2 and 5) can also offer opportunities
for improvement, and some notable developments in recent years include:
● Acoustic Doppler Current Profilers (ADCP) – making high flow gaugings more
feasible during flood events to assist with the subsequent development of stage
discharge relationships
● Nowcasting – improved techniques for high resolution short term rainfall
forecasts
● Ensemble forecasting – probabilistic estimates of rainfall, river flows, surge and
other variables
● Remote sensing – improvements in accuracy and resolution, and the range of
parameters which can be monitored by satellite, and increased availability
of products
● Forecasting systems – improvements in the functionality and usability of sys-
tems for running forecasting models, and for interfacing to other systems
Some emerging technologies, such as low cost sensor networks, disdrometers for
rainfall measurement, and remote techniques for measuring river velocity (and
hence flow), also show potential (see Chapter 2).
Resilience is also an important factor in flood warning applications, and Chapter
2 highlights the need to ensure that instruments are sited in locations where they
will not be damaged by flood water or debris, and with electronic equipment above
likely flood levels. Backup instruments and telemetry can also be provided at the
same site (e.g. for raingauges and coastal instrumentation) or further upstream (for
river level or flow gauges). Issues of site access during flood events also need to be
considered, and the health and safety of staff.
11.2.2 Thresholds
Flood warning thresholds define the conditions (or criteria) under which flood
warnings are issued (or considered for issue) and are described in Chapter 3.
Although values may be defined based on past experience, and sometimes by
sophisticated computer modelling, post event reviews and performance monitoring
may show the need for improvement. Some indicators of the possible need to revise
thresholds include:
● Missed warnings – problems with not issuing warnings when flooding occurs,
or after the start of flooding, may require a detailed investigation into the likely
causes and areas for improvement (e.g. in instrumentation, threshold levels,
forecasting models)
● False alarms – an unacceptable rate of false alarms may indicate that thresholds
are too conservative or, possibly, that an acceptable success rate is only possible
at the expense of a high false alarm rate
● Insufficient lead time – problems with warnings being issued later than would
ideally be required for an effective emergency response
Some approaches to increasing lead times include the use of new or improved flood
forecasting models, additional thresholds on gauges upstream or distant from the
flood risk area (preferably whilst continuing to maintain the existing thresholds)
and, possibly, adjusting existing thresholds so that they achieve more lead time,
whilst still maintaining an acceptable false alarm rate. All adjustments to thresholds
need to be made with care, and fully tested and documented before implementation,
preferably also consulting with those affected to confirm that the approach is
acceptable. More sophisticated approaches might also be considered, including use
of computer modelling to support the development and testing of thresholds, proba-
bilistic techniques, and other methods (see Chapter 3). Improvements can also aim
to reduce the various time delays in the detection, dissemination and response process
(see Chapter 5) through improved systems, procedures, and training.
11.2 Performance Improvements 255
256 11 Review
11.2.3 Dissemination
The principles for issuing flood warnings are similar to those in many other areas
of risk communication and Chapters 4 and 10 discuss this topic. Some key points
to consider (e.g. Emergency Management Australia 1999; Drabek 2000; Handmer
et al. 2001; Martini and de Roo 2007), beyond the need for warnings to be accurate,
reliable and timely, and reach the intended recipients, include:
● To give people time to prepare and plan for flooding, warnings should be staged
(if there is time), starting from advisory/watch/warning alerts (or similar), before
escalating to a full warning
● Messages should be consistent, clear and concise, and tailored to the audience,
ideally in their own language
● Messages should be specific about the threat; for example in terms of flood tim-
ing, depth, duration etc.
● Messages should also include advice on actions to take to protect people and
property, and distinguish between forecasts and warnings
● Messages should be received from a single, authoritative (and trusted) source,
whilst acknowledging that in practice people may seek information from multi-
ple sources, both formal and informal, before deciding to act
● Multiple means of dissemination should be used in case of failure in any one
route, and to improve the effectiveness of response
Potential recipients (communities, emergency services) should also be actively
involved in the choice of appropriate ways of disseminating information and the
wording of messages. A particular challenge can be deciding on methods for issu-
ing warnings to vulnerable groups, and to transient or mobile populations (e.g.
tourists, vehicles, business travellers). Some other factors which may need to be
considered in message design (Elliot and Stewart 2000) include:
● Degree of flood exposure, severity of impact
● Degree of flood experience
● Financial or emotional ‘stake’ in the flood-prone area (e.g. residents as opposed
to tourists)
● Household structure (e.g. age, health status)
● Language and
● Employment status (e.g. likelihood of being at home during the day)
New technologies are also increasing the range of methods which can be used,
allowing better targeting of warnings, and reducing the time taken to issue warn-
ings. Some recent developments include:
● Computerised multimedia systems for issuing warnings
● Internet based systems combining real time data, flood extent and advice on
actions to take
● Cell phone broadcasting techniques to all subscribers within range of a mast
● Digital radio warnings to drivers of vehicles
● Remotely activated barriers and electronic road signs for roads and footpaths
● Low cost battery free radios and satellite transmission of warnings
As with all components in a flood warning system, dissemination techniques
should be able to continue working under flooding conditions and associated prob-
lems, such as high winds and heavy rainfall, with backup methods in case of failure.
For techniques which require access to potential flooding areas (e.g. loud hailer,
door knocking) access may also be interrupted by flood waters, or not be possible
due to health and safety considerations.
11.2.4 Forecasting
Improvements to flood forecasting can focus on improving existing models, devel-
oping new models, and improving the robustness and speed of operation of models.
Performance monitoring techniques can also be used to guide future improvements
(see Chapter 5).
Changes to existing models can include recalibrating models to account for
recent flood events, or changes to instrumentation, flood defences, river channels,
and coastal conditions, and adding in new functionality. Where new models are to
be developed, the opportunity may also be taken to try out new modelling techniques;
for example, using different types of model, or introducing data assimilation and
probabilistic techniques. Such changes may also be linked to improvements to
forecasting systems.
Opportunities may also arise to combine models across a number of flood warn-
ing schemes; for example, using an integrated catchment approach. A spin-off ben-
efit from this approach is often that the need to consider the catchment as whole
may suggest improvements to instrumentation that would not otherwise be obvious
from examination of records for single gauges.
The choice of models should be appropriate to the level of risk and a range of other
factors, and various guidelines are available to help in deciding on an appropriate
approach (e.g. World Meteorological Organisation 1994; USACE 1996; Environment
Agency 2002, 2004). For example, Tilford et al. (2007) describe a structured approach
to the selection of river forecasting models which considers:
● The physical characteristics of the catchment and river(s)
● The varying levels of data availability and quality
● The cost and time of development
● The technical and economic risks associated with the investment
● Performance targets
The method uses a combination of flowcharts, risk assessment matrices and cost
benefit analyses to decide on an appropriate choice of model. The modelling
options provided include empirical, data-based, conceptual and process-based models
for a range of river modelling problems (e.g. floodplains, reservoirs, structures, snowmelt,
tidal influences). The method also includes consideration of a range of practical
11.2 Performance Improvements 257
258 11 Review
issues, such as how long it might take to install instrumentation, the level of modelling
expertise needed, and the need for additional exploratory investigations. Throughout,
room is left for expert judgement and the method avoids being prescriptive in the
choice of modelling solution.
11.2.5 Preparedness
Techniques for developing flood emergency plans are well established (see Chapter 9),
with issues such as inter-agency collaboration, a clear chain of command, interop-
erability of equipment, team typing and other factors increasingly emphasised,
together with the involvement of communities and their representatives in formulating
plans. The needs of vulnerable groups in particular need to be considered.
Increasingly, an all-hazards approach is being adopted by many organisations,
providing benefits of scale and allowing plans to be rehearsed more frequently.
For plans to remain effective, they need to be regularly rehearsed and reviewed
and updated. The issue of resilience is also important for all aspects of the flood
warning process, and probabilistic and risk based approaches from other sectors
may become more widely used in future in assessing potential points of failure.
Information technology also provides the opportunity for more realistic training
exercises, combining multimedia simulations, and animation of flood extents in com-
puter models of towns and cities. Geographical Information Systems can also help dur-
ing the planning phase in examining how access routes, infrastructure, key facilities and
other factors will be influenced by flood water, and in refining risk assessments.
11.2.6 Response
Previous chapters have discussed the important of clear presentation of flood warnings
(Chapter 4), active engagement by communities (Chapter 9), and the use of a range of
approaches for providing information to people (e.g. ISDR 2006; United Nations
2006a). A clear statement can also help with understanding the objectives of a flood
warning and forecasting system; for example (Defra 2004) that:
“Flood warning is the provision of advance warning of conditions that are likely
to cause flooding to property and a potential risk to life. The main purpose of flood
warning is to save life by allowing people, support and emergency services time to
prepare for flooding. The secondary purpose is to reduce the effects and damage of
flooding. This might include moving property to a safer location such as upstairs or
putting in place temporary measures to prevent floodwater entering properties such
as flood boards or sandbags. In addition flood warning informs operating authorities
who need to take action such as closing floodgates or other control structures in
advance of flooding conditions.”
Although local authorities and emergency services can take many actions to
reduce or mitigate flooding, ultimately, if flooding does occur, then the success of
the warnings issued will also depend on residents understanding the risks from
flooding and taking appropriate actions in time to protect people and property.
Some principles which can improve the success of flood warning systems (Handmer
2001; Betts 2003) include:
● The public’s access to both formal and informal sources of warning information
● The value of ‘shared understanding’ between the public and emergency manag-
ers about the warning message and process
● Inter-organisation cooperation
● The recognition of local needs
An ongoing programme of education, training and capacity building helps to keep
the awareness of flood risk high, and improves the likelihood that people will
respond appropriately in the next flood event. Informal warning systems also have
a valuable role to play during flood events (e.g. Parker 2003) and reinforce and add
credibility to formal warnings. Techniques and expertise from the social sciences,
market research, education, and health care promotion, can all be used to help
improve the effectiveness of campaigns.
Avoidance of risks to people is a primary objective for many flood warning
schemes, and some factors which can lead to an increased risk to loss of life include
(Environment Agency 2003):
● Where flow velocities are high
● Where flood onset is sudden as in flash floods, for example the Linton/Lynmouth
floods in 1952, Big Thompson flood, USA, in 1976 and flash floods in Southeast
China in 1996
● Where flood waters are deep
● Where extensive low lying densely populated areas are affected, as in
Bangladesh, so that escape to high ground is not possible
● Where there is no warning (i.e. where there is less than, say, 60 minutes of
warning)
● Where flood victims have pre-existing health/mobility problems
● Where natural or artificial protective structures fail by overtopping or collapse.
Flood alleviation and other artificial structures themselves involve a risk to life
because of the possibility of failure, for example dam or dike failure
● Where poor flood defence assets lead to breaches or flood wall failure, leading
to high velocities and flood water loadings on people in the way
● Where there is debris in the floodwater that can cause death or injury
● Where the flood duration is long and/or climatic conditions are severe, leading
to death from exposure
● Where there is dam failure
Other causes include building collapse and related circumstances (e.g. mobile
homes, campsites), being swept away, falling down manholes or similar, and being
trapped in buildings or vehicles. For example, Jonkman and Kelman (2005) suggest
that drowning in vehicles seems to be a worse problem in the US than in Europe,
and that significant numbers of flood deaths are attributable to unnecessary risky
11.2 Performance Improvements 259
260 11 Review
behaviour, whilst in Australia almost 10% of flood related deaths result from people
trying to retrieve property or animals. Some estimates (e.g. Henson 2001) suggest
that about half of all flash flood related deaths in the USA occur to people in vehicles.
Risks can be reduced by raising public awareness on the dangers of driving through
flood water, precautionary closing of roads in advance of flooding, and dissemination
techniques aimed specifically at drivers, such as digital radio alerts, and remotely
or locally activated barriers or electronic signs (see Chapter 4). Flood risk assessments
may also need to account for flow velocities since this can be a significant factor in
whether vehicles are swept away by floodwater.
For the particular example of non-residential properties (e.g. shops, businesses,
factories), some criteria for the effectiveness of flood warnings which have been
proposed (Defra 2005) include:
● They have a long lead time (preferably at least 8 hours)
● Management have confidence in the warning and the issuing authority
● The warnings give specific information on the timing and likely level of flooding
● Staff are aware of and trained in the actions to take
● There are enough able bodied staff or contractors available to move equipment
and goods and take mitigating actions
● Equipment and goods are able to be moved (e.g. not too large or heavy)
● There is enough space on upper floors or storage areas, in an alternative location,
or on higher ground, to move equipment and goods to
● Appropriate refrigeration is available for storing perishable foodstuffs, drinks,
pharmaceuticals etc. elsewhere
● Surrounding areas and roads are not flooded to facilitate evacuation and move-
ment of equipment
To help deal with the complexity of a flood event, Decision Support Systems and
Geographical Information Systems can also assist emergency managers in assessing
risks, improving situational awareness, sharing information between people and
organisations, automated logging of actions, and (in some cases) providing guidance
on optimum decisions, such as the requirements for evacuating properties. Hand held
units may also be used by staff on site to view locally relevant information.
Probabilistic and cost loss approaches may also provide one way of optimising deci-
sions to take account of uncertainties in factors such as flooding extent, flooding
impact, and the response of individuals to flooding.
Again, as with all other improvements, resilience needs to be considered from
flooding and associated rainfall and high winds etc., together with staff training,
interoperability of systems, and a range of other factors (see Chapter 9).
11.3 Prioritising Investment
The outcome from regular or post event reviews is often a series of recommenda-
tions, some of which may require investment in new equipment, procedures and
other resources. The requirements to justify expenditure vary widely between
organisations, but there will often be a need to consider where priorities lie, and the
relative costs of different options. Some techniques which can be used for prioritising
investment include:
● Cost benefit analysis – in which costs are compared to the estimated benefits
from improving flood warnings
● Multi-criteria analysis – which uses a range of weighting techniques across a
number of factors to rank the relative merits of proposals
● Risk based approaches – in which investment is targeted to the areas of greatest
risk, taking account of factors such as the frequency of flooding, the number of
people affected, risk to life, risk to critical infrastructure, or the risk of flow con-
trol structures not being operated effectively
Each method has its advantages and limitations. For example, cost benefit analyses
focus mainly on the economic aspects of investment, and it can be difficult to bring
in other more intangible quantities, such as loss of life and the long-term health
costs from people affected by flooding who, with sufficient warning, might have
moved to safety. People may also place more value on saving personal items (docu-
ments, memorablia etc.) and domestic animals and pets than on high cost items.
By contrast, Multi-Criteria Analysis (MCA) methods do not consider the eco-
nomic case in detail, and are more subjective in the way that decisions are
reached, although can easily be combined with economic analyses. Other priori-
ties may also influence the overall decision, such as pressure from local residents
and politicians to improve flood warning schemes, and reputational issues,
related to not having issued a warning before an event, or high false alarm rates.
Risk based approaches are also often incorporated into the other two techniques,
as described later.
The scope of the economic analysis also needs to be defined; for example, flood
warning is increasingly seen as a key aspect of overall flood risk management, or
one component in a multi-hazard approach. Economic analyses may therefore need
to be tied into a wider assessment covering flood defences, development on flood-
plains, catchment management and risks from sources other than flooding. Other
complicating factors can include situations where the costs and/or benefits are
accrued by different organisations, some of which may be outside the flood warning
process, and in trans-national river basins, where several countries may participate
in the flood warning scheme. Sensitivity studies, or probabilistic techniques, may
also be used to help to account for uncertainty in inundation extents, depths, veloci-
ties and impacts.
11.3.1 Cost Benefit Analysis
Cost benefit analysis is widely used in a number of fields, including flood risk
management (e.g. ISDR 2006; World Meteorological Organisation 2007), and has
also been applied to flood warning systems. The cost element is built up from
systematic analysis of the individual (unit) costs of the items which make up the
11.3 Prioritising Investment 261
262 11 Review
system, whilst benefits are typically estimated in terms of the damage avoided due
to issuing flood warnings. Some approaches also attempt to assign values to lives
saved, for example in terms of lifetime earnings, or compensation payments,
although this is a problematic area which is often excluded from the analysis.
The scope of the analysis will depend on the level at which financial decisions
are made, and could be just for an individual department or organisation, or
across a number of organisations. Analyses can also be at the scale of individual
flood warning schemes, regional systems, or at national level. Care is needed to
avoid double counting of benefits when considering multiple schemes and
organisations.
Depending on the objectives of the analysis, the cost element of the analysis can
consist of a wide range of items, including:
● Start up costs – feasibility studies, design studies etc.
● Capital costs – instrumentation, computers, models, dissemination equipment,
communication equipment etc.
● Operating and maintenance costs – for office accommodation, public awareness
campaigns, emergency response, post event studies etc.
● Review costs – for periodic reviews and reporting on performance
Depending on the scope of the analysis, staff costs may just relate to the flood
warning and forecasting service, or extend more widely into the emergency serv-
ices, and local authorities, and businesses. Future investments also need to be con-
verted to a common basis; for example using net present value techniques.
Additional costs, such as those incurred in taking mitigating actions (e.g. temporary
closing of businesses), may also need to be considered.
Estimates for the benefits from flood warnings usually focus on the damage
avoided to property by moving items to safety, and perhaps through preventing
flooding by installing temporary measures in time, such as flood boards, or sandbags.
Some techniques for estimating potential damages include (e.g. World Meteorological
Organisation 2007):
● Historical damages – information on actual damages from previous flood events
(e.g. insurance claims, surveys)
● Unit area method – in which losses are estimated based on the floor area of
properties
● Percentage of property values – in which losses are estimated from property
values (ideally for the building alone and excluding land values)
● Weighted average annual damages – based on the frequency and severity of
flooding
These methods all have various advantages and limitations; for example, informa-
tion on previous floods may be incomplete, or biased by the approach used to
collect data, whilst unit area approaches may be more suitable for commercial
properties than residential properties. The percentage of property value approach
uses data which is widely available, but it may be difficult to separate out land
values from building values. Also, building contents can vary widely, particularly
for commercial properties, whilst the average annual damage method requires long
term reliable information on flooding histories and damages. Various combined and
synthetic approaches are also used.
The benefit from flood warnings are often separated into direct and indirect (or
intangible) losses in a number of areas; for example (USACE 1996):
● Reduced threat to life – barricades, evacuations, rescues, public awareness
● Reduced property loss – removal or elevation of residential and commercial
structure contents and vehicles
● Reduced social disruption – traffic management, emergency services, public
awareness
● Reduced health hazards – evacuations, public information, emergency services
● Reduced disruption of public services – utility shutoffs, emergency services,
supplies, inspection supplies, inspection, public information
● Reduction in inundation – flood fighting, temporary flood damage reduction
measures, technical assistance
Here, flood fighting means taking actions to reduce flooding, such as repairing
breaches, clearing channels of debris, sandbagging etc. USACE (1994) addition-
ally notes the benefits arising from a number of other factors, including tempo-
rary flood proofing of properties, reductions to recovery costs, and temporary
suspension of industrial and other processes. Many of these items can in princi-
ple be estimated, although some with difficulty due to lack of data and other
problems, such as the loss of life issue referred to earlier. Values may also
depend on factors such as weather conditions (temperature, wind chill etc.), the
time of day that the warning is received, and the time elapsed since the last flood
(e.g. World Meteorological Organisation 1973). More detailed discussions of
methods for estimating flood warning benefits can be found in World
Meteorological Organisation (1973), USACE (1994), Carsell et al. (2004), and
Parker et al. (2005).
In flood warning applications, the annual average damage approach is perhaps
the most widely used method. For general classes of property (both residential and
commercial), annual average damage curves can be estimated for a range of flood
depths and, possibly, velocities. These values can be estimated from post event data
across a number of flood events and types/ages of property, and from demographic
characteristics, although there can be considerable uncertainty in the estimates
derived (e.g. Merz et al. 2004). Values can also be probability weighted by integrat-
ing damage-depth and depth-frequency curves.
For example, a common approach is to assess typical depth (stage) damage rela-
tionships for various classes of property, and then to assess the reductions in dam-
age for different flood warning lead times, as illustrated in Fig. 11.1 for a specific
community and lead time.
Studies of this type have shown that, as might be expected, the damage avoided
increases with increasing lead time up to a point of diminishing returns, beyond
which any additional lead time becomes of little benefit in reducing damages. Other
benefits might also be included, such as the damage avoided by operating flow
11.3 Prioritising Investment 263
264 11 Review
control structures, or installing temporary defences (barriers, flood boards, sand-
bags etc.) to protect communities or individual properties.
For example, in England and Wales, the following equation (CNS Scientific and
Engineering Services 1991; Parker et al. 2005; Tilford et al. 2007) forms the basis
of the method used to estimate flood warning benefits from reductions in damage
to residential property and road vehicles:
FDA = R x Pi x P
a x P
c x PFDA
where:
FDA = Flood Damages Avoided
PFDA = Potential Flood Damages Avoided
R = Service Effectiveness
Pi = Probability that the individual will be available to be warned
Pa = Probability that the individual is physically able to respond
Pc = Probability that the individual knows how to respond effectively
The Service Effectiveness is the proportion of properties which were sent a flood
warning whilst the Potential Flood Damages Avoided is calculated from the average
annual damages (AAD) as follows:
PFDA = DR x C x AAD
Here, DR (the Damage Reduction factor) is the proportion of damages which
can realistically be avoided by flood warning (since some damage is unavoidable),
and depends on warning lead time, and C is the Coverage of the flood warning
Fig. 11.1 Example of a community stage-damage relationship (World Meteorological Organisation
1973, 1994) (Reproduced from the WMC Guide to Hydrological Practices - Data Acquisition and
processing Analysis, Forecasting and other Applications, courtesy of WMO)
service (i.e. the proportion of properties which receive a flood warning service).
An alternative version (e.g. Environment Agency 2002; Tilford et al. 2007)
includes an allowance for the costs of mitigating actions by property owners;
for example, when taking time off work to protect properties. Similar tech-
niques can in principle also be used for commercial properties, although some
sites may need to be considered on a case by case basis (e.g. major chemical
works, oil refineries, distribution warehouses etc.).
In addition to providing a basis for estimating flood warning benefits, the vari-
ous factors also provide a focus for improvements to individual components in
the flood warning service. For example, in England and Wales, the 2012/13 target
values are in the range 80–90% for most parameters except for the Damage
Reduction factor (e.g. Parker et al. 2005). Additional factors are used to monitor
progress in the percentage of people making preparations in advance of flooding
(e.g. individual flood plans), and the proportion of people at risk signing up for a
direct flood warning service. Progress is assessed through post event reviews,
independent market surveys of flood warning recipients, and other approaches.
However, the extent to which improvements are possible depends in part on the
nature of the flood risk in individual locations, the scope to take actions to prevent
flooding, how frequently flooding occurs, and socioeconomic and other factors.
Computer simulation tools of the type described in Chapter 10 might also be used
to explore the effectiveness of improvements to individual components, including
social, vulnerability, psychological and policy aspects (e.g. Simonovic and
Ahmad 2005).
Many social and behavioural studies have also been performed into losses from
factors other than damage to property, and on the general effectiveness of flood
warnings, including studies on loss of life (Jonkman and Kelman 2005), public
response to flood warnings (Drabek 2000; Pfister 2002), health impacts (Parker
et al. 2005), and risks to people in vehicles (e.g. Henson 2001). Other examples
include studies on the real time assessment of hurricane losses (Dixon et al. 2006),
and the benefits from flood forecasting for reservoir flood control and short and
long term flood forecasting (National Hydrologic Warning Council 2002).
Various estimates have also been derived for the damage reduction component of
flood warnings, with values typically in the range from a few percent to 30–40% or
more (e.g. ISDR 2006; World Meteorological Organisation 1989; Parker et al. 2005).
However, care is needed in interpreting estimates to see which types of losses have
been included in the analysis, and the various other assumptions made. For example,
some studies have focussed mainly on the monetary benefits to residential property
owners, and much less is known about intangible benefits, and the varying reasons
why some people do not take effective action even after receiving a warning.
11.3.2 Multi Criteria and Risk Based Analysis
Rather than working in monetary terms, multi criteria analyses aim to evaluate a
range of options against criteria or objectives agreed with key stakeholders, and can
11.3 Prioritising Investment 265
266 11 Review
be used as an initial screening tool, before proceeding to a more detailed analysis,
or as perhaps the only way forwards when there are many conflicting objectives.
The technique builds on the experience and knowledge of stakeholders, and can
help to make the decision process more transparent, although inevitably includes an
element of subjectivity.
There are many forms of multi criteria analysis technique, typically involving
assigning weights and scores to a range of options and issues, and then combin-
ing the results to reach an overall conclusion (e.g. World Meteorological
Organisation 2007). Multi criteria techniques may also be combined with cost
benefit techniques; for example, a methodology proposed for evaluating river
and coastal flood defence schemes (Ash et al. 2005; Defra 2005) involves the
following steps:
● Definition of problem, the objectives and identification of all options
● Elimination of unreasonable options
● Structuring the problem (high level screening)
● Qualitative assessment of impacts
● Quantitative assessment of impacts
● Determine the tangible benefits and costs of options (economic analysis)
● Scoring of options
● Weight elicitation, as appropriate
● Comparison of options using expanded decision rules
● Testing the robustness of the choice
● Selecting the preferred option
Impact assessments are performed using a structured approach covering a range
of economic, environmental and social issues which need to be considered.
Numerical or descriptive scores can be used, with scoring by key experts or a
committee, whilst weights reflect the preferences of stakeholders. The compari-
son of options stage combines the outputs from the multi-criteria and cost benefit
analyses.
In another example, Sene et al. (2006) describe a simple scoring approach to
supplement cost benefit analyses for the design for telemetry networks for flood
warning applications, in which the scoring criteria were:
● Risk category for the flood warning area (high, medium, low etc.)
● Cost of implementation
● Cost per property of implementation
● Benefit-cost ratio at a flood warning area level
● Views from consultations (e.g. flooding ‘hotspots’)
More generally, risk categories can be used as a guide to the choice of appropriate
instrumentation, performance criteria, and flood warning dissemination methods
in a new or upgraded flood warning scheme (e.g. Andryszewski et al. 2005).
Glossary
A
Action Table – a table of actions to take as meteorological, river and/or coastal
conditions exceed predefined threshold values
Antecedent Conditions – the state of wetness of a catchment prior to an event or
period of simulation (Beven 2001)
Antecdent Precipitation Index – the weighted summation of past daily precipitation
amounts, used as an index of soil moisture. The weight given each day’s precipita-
tion is usually assumed to be an exponential or reciprocal function of time, with the
most recent precipitation receiving the greatest weight (UNESCO/WMO 2007)
Automated Voice Messaging (AVM) – automated telephone system for issuing
flood warnings
Automatic Weather Station (AWS) – an instrument for automatically measuring climate
data in real time including (typically) wind speed and direction, solar radiation, air tempera-
ture, humidity, and rainfall, and possibly other parameters, such as soil temperature
B
Baseflow – part of the discharge which enters a stream channel mainly from
groundwater, but also from lakes and glaciers during long periods when no precipi-
tation or snowmelt occurs (UNESCO/WMO 2007)
Basin – see Catchment
Black Box Model – a model that relates only an input to a predicted output by a
mathematical function or functions without any attempt to describe the processes
controlling the response of the system (Beven 2001)
Boundary Conditions – constraints and values of variables required to run a model
for a particular flow domain and time period (Beven 2001)
Business Continuity Management – a management process to identify and manage
the hazards or threats which can disrupt the smooth running of an organization or
delivery of a service
267
268 Glossary
C
Calibration – adjustment of the parameters of a model, either on the basis of physical
considerations or by mathematical optimization, so that the agreement between the
observed data and estimated output of the model is as good as possible (see ModelCalibration, UNESCO/WMO 2007)
Cascade Warning – an approach to dissemination of warnings in which warnings
are passed from one person or organization to others, who in turn pass on the warn-
ing following predefined procedures. Includes Call Trees and Telephone Trees
Catchment – drainage area of a stream, river or lake, or area having a common
outlet for its surface runoff (see Basin or Catchment, UNESCO/WMO 2007)
Conceptual Hydrological Model – simplified mathematical representation of
some or all of the processes in the hydrological cycle by a set of hydrological con-
cepts expressed in mathematical notations and linked together in a time and space
sequence corresponding to that occurring in nature (UNESCO/WMO 2007)
Contingency Table – a table usually summarizing the relationship between the
frequencies of occurrence of two or more variables, at the simplest level consisting
of a 2 × 2 matrix
Cost Benefit Analysis – a decision making technique which compares the likely
costs of an action or investment with the expected benefits
Cost Loss Analysis – an analysis technique which compares the cost of taking an
action with the likely losses if that action is not taken, which can include depend-
ence on lead time, the influence of only partial protection against losses, and other
factors
D
Damage Avoidance – the potential financial benefit from providing a flood warn-
ing taking into account the maximum damage which could be avoided and possibly
the costs of property owners acting upon the warning
Data Assimilation – the use of current and recent real time observations of mete-
orological, river and/or coastal conditions to improve a forecast (e.g. a flood
forecast)
Data Collection Platform – automatic measuring device with a radio transmitter
to provide contact via a satellite with a reception station (UNESCO/WMO 2007)
Debris Flow/Mud Flow – flow of water so heavily charged with earth and debris
that the flowing mass is thick or viscous (UNESCO/WMO 2007). A high-density
mud flow with abundant coarse-grained materials such as rocks, tree trunks, etc.
(IDNDR 1992)
Decision Support System – in emergency management, usually a computerized
system for collating and displaying real time information of many types (spatial,
time series, descriptive etc.), and sometimes for advising on optimum decisions
Glossary 269
Degree-Day – algebraic difference, expressed in degrees C, between the mean
temperature of a given day and a reference temperature (usually 0°C). For a given
period (months, years) algebraic sum of the degree-days of the different days of
the period (UNESCO/WMO 2007)
Deltas – see Estuaries
Deterministic Model – a model that with a set of initial and boundary conditions
has only one possible outcome or prediction (Beven 2001)
Dike – see Flood Defence
Dissemination – in flood warning applications, the issuing of warnings by a range
of direct, community based and indirect methods
Distributed Model – a model that predicts values of state variables varying in
space (and normally time) (Beven 2001)
E
Effective Rainfall – that part of rainfall which contributes to runoff. In some
procedures the prompt subsurface runoff is entirely excluded from direct runoff and
then effective rainfall is equal to rainfall excess (UNESCO/WMO 2007)
Ensemble Forecast – a number of alternative realisations of future meteorological,
river or coastal conditions based on alternative values for initial conditions, model
parameter values etc., which reflect the inherent uncertainties in observations and
forecasting models
Estuary – the tidal reaches of a river as it outfalls to the sea, where fresh and sea
water mix. Sometimes called a Delta or River Delta (although this term describes
the sediment deposited by some rivers within the tidal zone)
Evapotranspiration – quantity of water transferred from the soil to the atmosphere
by evaporation and plant transpiration (UNESCO/WMO 2007)
F
False Alarm – in flood warning applications, a warning which is issued but for
which no subsequent flooding occurs. Can also include ‘near misses’
Fetch – area in which ocean, lake and reservoir waves are generated by the wind.
The length of the fetch area is measured in the direction of the wind (UNESCO/
WMO 2007)
Finite Difference – the approximate representation of a time or space differential in
terms of variables separated by discrete increments in time or space (Beven 2001)
Finite Element – the approximate representation of time or space differentials in
terms of integrals of simple interpolation functions involving variables defined at
nodes of an irregular discretization of the flow domain into elements (Beven 2001)
270 Glossary
Flash Flood – flood of short duration with a relatively high peak discharge
(UNESCO/WMO 2007). Alternatively, a flash flood can be defined as a flood that
threatens damage at a critical location in the catchment, where the time for the
development of the flood from the upstream catchment is less than the time needed
to activate warning, flood defence or mitigation measures downstream of the critical
location. Thus with current technology even when the event is forecast, the achievable
lead-time is not sufficient to implement preventative measures (e.g. evacuation,
erecting of flood barriers) (ACTIF 2004)
Flood Defence, Dike or Levee – water-retaining earthwork used to confine stream-
flow within a specified area along the stream or to prevent flooding due to waves
or tides (UNESCO/WMO 2007). Can be constructed from a range of materials,
including concrete, steel and rockfill
Flood Fighting – emergency response operations to reduce or prevent flooding,
including reinforcing flood defences, sandbagging, installation of temporary
defences, and other measures
Flood Forecasting System – a computer system for managing the operation of one
or more flood forecasting models, include automated collection and validation of
real time data, post processing of model outputs, and possibly automated alerting
facilities if thresholds are exceeded
Flood Risk Area – an area at risk from flooding which may or may not have an
existing warning service and whose extent is typically estimated from historical
information, modeling, or other methods
Flood Risk Assessment – an assessment of the likely extent and probability of
flooding at one or more locations
Flood Warning Area – an area defined for use in flood warning procedures, within
which people receive flood warnings
Flow Routing (or Flood Routing) – a technique used to compute the movement
and change of shape of a flood wave moving through a river reach or a reservoir
(UNESCO/WMO 2007)
Forecasting Point – a location at which it is useful to have a forecast of future river
or coastal conditions (e.g. a Flood Warning Area, a river or coastal monitoring site,
a control structure)
Freeboard – vertical distance between the normal maximum level of the surface of
a liquid in a conduit, reservoir, tank, canal, etc., and the top of the sides of the
retaining structure (UNESCO/WMO 2007)
G
Geographical Information System – computer software for the graphical presen-
tation and analysis of spatial datasets and the associated hardware, procedures,
equipment etc.
Glacial Lake Outburst Flood – a flood caused by the sudden release of water from
a lake formed by moraine, ice or similar
Glossary 271
H
Hurricane – see Tropical Cyclone
Hydrodynamic Model – a solution to the equations expressing mass, momentum
and energy conservation of water, sediment, heat and other parameters in a river,
estuary or coastal reach
Hydrograph – graph showing the variation in time of some hydrological data such
as stage, discharge, velocity, sediment load, etc. (UNESCO/WMO 2007)
I
Ice Jam – accumulation of ice at a given location which, in a river, restricts the flow
of water (UNESCO/WMO 2007)
Initial Conditions – values of storage or pressure variables required to initialize a
model at the start of a simulation period (Beven 2001)
Intangible Losses – losses which cannot easily be expressed in economic terms,
including impacts on health, business disruption, stress, impacts on tourism etc.
Isochrone Map – map or chart of a drainage basin in which a series of lines
(isochrones) gives the times of travel of water originating on each isochrone to
reach the outlet of the basin (UNESCO/WMO 2007)
K
Kalman Filter – a time series analysis technique which seeks to provide an improved
forecast of future conditions accounting for differences between previous observations
and forecasts. Also extended Kalman Filter and ensemble Kalman Filter variants
L
Lead Time – warning lead time is the time between receipt of a flood warning and
the time of the onset of flooding; forecast lead time is the maximum lead time at
which forecasts can be provided to an acceptable accuracy
Levee – see Flood Defence
M
Monte Carlo Simulation – simulation involving multiple runs of a model using different
randomly chosen sets of parameter values or boundary conditions (Beven 2001)
Multi-Criteria Analysis (MCA) – a structured decision making technique
widely used for evaluating alternative options where multiple criteria and priori-
ties are involved, perhaps including social, environmental, financial, political and
other factors
272 Glossary
N
Nowcast – a meteorological modelling technique which combines the outputs from
weather radar observations and possibly Numerical Weather Prediction model
outputs to produce short term (typically 0–6 hour ahead) forecasts of rainfall and
other parameters
Numerical Weather Prediction (NWP) – computer modelling technique in which
the atmosphere, oceans and land surface are modelled on a three dimensional grid
to produce forecasts of future conditions based on data assimilated from a wide
range of sources (ground based observations, satellite, ships, aircraft etc.)
O
Objective Function – a measure of how well a simulation fits the available obser-
vations (Beven 2001)
Orographic Precipitation – precipitation caused by the ascent of moist air over
orographic barriers (UNESCO/WMO 2007)
P
Parameter – a constant that must be defined before running a simulation (Beven 2001)
Polder – a mostly low-lying area artificially protected from surrounding water and
within which the water table can be controlled (UNESCO/WMO 2007)
Process Based Model – models which to varying degrees solve the partial differential
equations representing catchment and coastal processes, typically on a gridded
basis, perhaps including empirical or conceptual representations for some components
of the model
Public Switched Telephone Network (PSTN) – the telecommunications equipment
and infrastructure which connects land line telephones
Q
Quantitative Precipitation Forecasts – precipitation (rainfall, snow, hail etc.) fore-
casts typically based on nowcasting or Numerical Weather Prediction techniques
R
Rainfall Runoff Model – a model which converts observed or forecast rainfall into
estimated river flows
Rating Curve –see Stage Discharge relationship
Real Time Updating – see Data Assimilation
River Gauging Station – a measuring location where observations of water level
(river, reservoir) and discharge are made
Glossary 273
Resilience – the capacity of a system, community or society potentially exposed to
hazards to adapt, by resisting or changing in order to reach and maintain an accept-
able level of functioning and structure. This is determined by the degree to which
the social system is capable of organizing itself to increase its capacity for learning
from past disasters for better future protection and to improve risk reduction measures
(UN/ISDR 2004)
S
Saffir/Simpson – five categories indicating the damage potential of tropical
cyclones (Holland et al. 2007)
Set-Up – water forced inshore by breaking waves (Holland et al. 2007)
Situation Report – a brief report that is published and updated periodically
during a relief effort and which outlines the details of the emergency, the needs
generated and the responses undertaken by all donors as they become known
(IDNDR 1992)
Snow Pillow – device filled with antifreeze solution and fitted with a pressure sensor
which indicates the water equivalent of the snow cover (UNESCO/WMO 2007)
Soil Moisture Deficit (SMD) – a state variable used in many hydrological models as an
expression of water storage. SMD is zero when the soil is at field capacity and gets larger
as the soil dries out. It is usually expressed in units of depth of water (Beven 2001)
Stage Discharge Relationship – or Stage Discharge Relation – relation between
stage and discharge at a river cross section and which may be expressed as a curve,
table or equation(s) (UNESCO/WMO 2007)
Stochastic – a model is stochastic if, for a given set of initial and boundary condi-
tions, it may have a range of possible outcomes, often with each outcome associ-
ated with an estimated probability (Beven 2001)
Surge – or Storm Surge –a sudden rise of sea as a result of high winds and low
atmospheric pressure; sometimes called a storm tide, storm wave, or tidal wave.
Generally affects only coastal areas but may intrude some distance inland
(IDNDR 1992)
Swell – smooth, regularly spaced waves that have propagated long distances from
their initial generation region (Holland et al. 2007)
T
Threshold – the meteorological, river or coastal conditions or forecasts which initiate
(or escalate) the flood warning dissemination process. Sometimes called triggers,
criteria, warning levels, alert levels or alarms
Trigger – see Threshold
274 Glossary
Tropical Cyclone – a synoptic-scale to mesoscale low pressure system which
derives its energy primarily from evaporation from the sea in the presence of high
winds and low surface pressure and condensation in convective clouds concentrated
near its center (Holland et al. 2007), usually for maximum sustained surface winds
of 64 knots (33 m s−1) or more (although sometimes defined for winds of 34 knots
or more). The term tropical cyclone is used in the Indian Ocean, hurricane in the
Atlantic and Eastern Pacific Oceans, and typhoon in the Western Pacific
Tsunami – a series of large waves generated by sudden displacement of seawater
(caused by earthquake, volcanic eruption or submarine landslide); capable of prop-
agation over large distances and causing a destructive surge on reaching land
(IDNDR 1992)
Typhoon – see Tropical Cyclone
U
Ungauged Catchment – a catchment or subcatchment in which flows are not
recorded to the extent required for the application (e.g. in real time for flood fore-
casting applications)
V
Vulnerability – the conditions determined by physical, social, economic, political
and environmental factors or processes, which increase the susceptibility of a com-
munity to the impact of hazards
W
Wadi – or Ouedd – channel which is dry except in the rainy season (UNESCO/
WMO 2007)
Watershed – see Catchment
Wave – disturbance in a body of water propagated at a constant or varying speed
(celerity), often of an oscillatory nature, accompanied by the alternate rise and fall
of surface fluid particles (UNESCO/WMO 2007)
Weather Radar – an instrument for detecting cloud and precipitation using micro-
waves typically with wavelengths in the range 3–10 cm
Wind Waves – choppy and chaotic waves generated locally by the wind (Holland
et al. 2007)
References
Chapter 1 Introduction
Alexander D (2002) Principles of Emergency Planning and Management. Oxford University
Press, Oxford
Andryszewski A, Evans K, Haggett C, Mitchell B, Whitfield D, Harrison T (2005) Levels of
Service Approach to Flood Forecasting and Warning. ACTIF international conference on
innovation advances and implementation of flood forecasting technology, 17–19 October
2005, Tromsø, Norway. http://www.actif-ec.net/conference2005/proceedings/index.html.
Accessed 15 January 2008
Basher R (2006) Global early warning systems for natural hazards: systematic and people-centred.
Phil. Trans. R. Soc. A 364: 2167–2182. doi:10.1098/rsta.2006.1819
Beven K J (2008) Environmental Modeling: An Uncertain Future. Routledge, London
Drabek T E (2000) The social factors that constrain human responses to flood warnings. In Parker
D J (Ed.), Floods. Routledge, London
Emergency Management Australia (1999) Guide 5 – Flood Warning. Australian Emergency
Manuals Series, Part III Emergency Management Practice, Volume 3 (2nd edition). http://
www.ema.gov.au/. Accessed 15 January 2008
FEMA (2005) Reducing Damage from Localized Flooding: A Guide for Communities. Chapter
6: Warning and Emergency Services, Report 511-07. http://www.fema.gov/hazard/flood/pubs/
flood-damage.shtm. Accessed 15 January 2008
Fortune D (2006) Floods and Hydroinformatics: New Challenges. 7th International Conference on
Hydroinformatics, HIC 2006, Nice, France, Volume 1, pp. 122–127
Guha-Sapir D, Hargitt D, Hoyois P (2004) Thirty Years of Natural Disasters 1974–2003: The Numbers.
Centre for Research on the Epidemiology of Disasters (CRED) Report. UCL Presse Universitaires
de Louvain. http://www.cred.be/cred1/publicat/online.htm. Accessed 15 January 2008
Handmer J (2002) Flood warning reviews in North America and Europe: statements and silence.
Australian Journal of Emergency Management, 17(3): 17–24. http://www.ema.gov.au/agd/
EMA/emaInternet.nsf/Page/Publications. Accessed 15 January 2008
Henson R (2001) U.S. flash flood warning dissemination via radio and television. In Gruntfest E,
Handmer J (Eds.), Coping with Flash Floods. Kluwer, The Netherlands, pp. 243–252
Holland G (Ed.) (2007) Global Guide to Tropical Cyclone Forecasting. Bureau of Meteorology
Research Centre (Australia) WMO/TC-No. 560, Report No. TCP-31, World Meteorological
Organization; Geneva, Switzerland. http://www.bom.gov.au/bmrc/pubs/tcguide/globa_guide_
intro.htm. Accessed 15 January 2008
Howard C D D (2004) Ensemble Optimization for Hydroelectric Operations. Workshop on the
Hydrological Ensemble Prediction Experiment (HEPEX), ECMWF, 8–10 March. http://
ecmwf.int/newsevents/meetings/workshops/2004/HEPEX/presentations.html. Accessed 15
January 2008
275
276 References
Huaimin G (2005) Flood Forecasting and Management in China. ACTIF International Conference
on Innovation Advances and Implementation of Flood Forecasting Technology, 17–19
October, Tromsø, Norway. http://www.actif-ec.net/conference2005/proceedings/index.html.
Accessed 15 January 2008
ISDR (2006) Developing Early Warning Systems: A Checklist. EWC III Third International
Conference on Early Warning from Concept to Action, 27–29 March, Bonn, Germany. http://
www.unisdr.org/ppew/. Accessed 15 January 2008
Krzysztofowicz R (2001) The case for probabilistic forecasting in hydrology. J. Hydrol.
249: 2–9
Kundzewicz Z W, Jun X (2004) Editorial – towards an improved flood preparedness system in
China. Hydrolog. Sci. J. 49(6): 941–944
Martini F, De Roo A (Eds.) (2007) EXCIFF Guide: Good Practice for Delivering Flood Related
Information to the General Public. European Commission/Joint Research Centre Report
EUR22760EN. http://exciff.jrc.it/downloads/exciff-related-documents/EXCIFF_guide.pdf.
Accessed 15 January 2008
National Research Council (2006) Completing the Forecast: Characterizing and Communicating
Uncertainty for Better Decisions Using Weather and Climate Forecasts. National Academies Press,
Washington, DC. http://www.nap.edu/catalog.php?record_id=11699. Accessed 15 January 2008
Pappenberger F, Beven K J (2006) Ignorance is bliss: or seven reasons not to use uncertainty anal-
ysis. Water Resour. Res. 42(5) doi:10.1029/2005WR004820
Pappenberger F, Beven K, Frodsham K, Romanowicz R, Matgen P (2007) Grasping the unavoid-
able subjectivity in calibration of flood inundation models: a vulnerability weighted approach.
J. Hydrol. 333(2–4): 275–287
Parker D J (Ed.) (2000) Floods. Volumes 1 and 2. Routledge, London/New York
Parker D J (2003) Designing Flood Forecasting, Warning and Response Systems from a Societal
Perspective. Paper Presented at the International Conference on Alpine Meteorology and
Meso-Alpine Programme, 21 May 2003, Brig, Switzerland. http://www.map.meteoswiss.ch/
map-doc/icam2003/Presentation/10.1/Brig_document.pdf. Accessed 15 January 2008
Sayers P B, Hall J W, Meadowcroft I C (2002) Towards Risk-Based Flood Hazard Management
in the UK. Proc. ICE 150: 36–42
Smith K (2004) Environmental Hazards. Assessing Risk and Reducing Disaster (4th edition).
Routledge, London/New York
Smith K, Ward R (1998) Floods: Physical Processes and Human Impacts. Wiley, Chichester, UK
Thieken A H, Kreibich H, Müller M, Merz B (2007) Coping with floods: preparedness, response
and recovery of flood-affected residents in Germany in 2002. Hydrolog. Sci. J. 52(5):
1016–1037
Todini E, Alberoni P, Butts M, Collier C, Khatibi R, Samuels P, Weerts A (2005) ACTIF Best
Practice Paper – Understanding and Reducing Uncertainty in Flood Forecasting. ACTIF
International Conference on Innovation Advances and Implementation of Flood Forecasting
Technology, 17–19 October, Tromsø, Norway. http://www.actif-ec.net/conference2005/
proceedings/index.html. Accessed 15 January 2008
UNESCO (2006) Tsunami teacher: an information and resource toolkit, building global capacity
to respond to and mitigate tsunamis. IOC Manuals and Guides, 47. http://unesdoc.unesco.org/
images/0014/001496/149617e.pdf. Accessed 15 January 2008
United Nations (2006a) Global survey of early warning systems. An Assessment of Capacities,
Gaps and Opportunities Towards Building a Global Early Warning System for All Natural
Hazards. United Nations, Geneva, Switzerland. http://www.unisdr.org/ppew/. Accessed 15
January 2008
United Nations (2006b) Guidelines for reducing flood losses. International Strategy for Disaster
Reduction. http://www.unisdr.org/eng/library/isdr-publication/flood-guidelines/isdr-publication-
floods.htm. Accessed 15 January 2008
U.S. Army Corps of Engineers (USACE) (1996) Hydrologic Aspects of Flood Warning-
Preparedness Programs. Report ETL 1110-2-540, Office of Chief of Engineers, Washington,
References 277
DC. http://www.tpub.com/content/USACETechnicalletters/ETL-1110-2-540/index.htm.
Accessed 15 January 2008
Wade S D, Ramsbottom D, Floyd P, Penning Rowsell E, Surendran S (2005) Risks to People:
Developing New Approaches for Flood Hazard and Vulnerability Mapping. 40th Defra Flood
and Coastal Management Conference, York, England
World Meteorological Organisation (1994) Guide to Hydrological Practices (5th edition). WMO
No. 168, Geneva, Switzerland
World Meteorological Organisation (2006a) Strategy and Action Plan for the Enhancement of
Cooperation Between National Meteorological and Hydrological Services for Improved Flood
Forecasting. WMO, Geneva, Switzerland. http://www.wmo.ch/pages/prog/hwrp/documents/
FFInitiativePlan.pdf. Accessed 15 January 2008
World Meteorological Organisation (2006b) Preventing and Mitigating Natural Disasters. WMO
No. 993, Geneva
World Meteorological Organisation (2006c) Social Aspects and Stakeholder Involvement in
Integrated Flood Management. WMO No. 1008, Geneva, Switzerland
Chapter 2 Detection
Beneviste J, Berry P (2004) Monitoring river and lake levels from space. ESA Bulletin 117, July 2004.
http://www.esa.int/esapub/bulletin/bullet117/chapter5_bul117.pdf. Accessed 15 January 2008
Bowler N E, Pierce C E, Seed A W (2006) STEPS: a probabilistic precipitation forecasting
scheme which merges an extrapolation nowcast with downscaled NWP. Quart. J. R. Meteorol.
Soc.132: 2127–2155
Cluckie I D, Rico-Ramirez M A (2004) Weather radar technology and future developments.
GIS and Remote Sensing in Hydrology, Water Resources and Environment (Proceedings
of ICGRHWE held at the Three Gorges Dam, China, September 2003). IAHS Publications,
289
Collier C G (1996) Applications of Weather Radar Systems: A Guide to Uses of Radar Data in
Meteorology and Hydrology. 2nd edition. Wiley, Chichester
Creutin J D, Obled C (1982) Objective analyses and mapping techniques for rainfall fields: an
objective comparison. Water Resour. Res. 18(2): 413–431
Creutin J D, Muste M, Bradley A A, Kim S C, Kruger A (2003) River gauging using PIV tech-
niques: a proof of concept experiment on the Iowa river. J. Hydrol. 277(3–4): 182–194
Crow W T, Bindlish R, Jackson T J (2004) Improving Flood Forecasting Via the Integration of
Remotely Sensed Soil Moisture Observations Derived from Passive Spaceborne Observations.
ACTIF International Conference on Innovation Advances and Implementation of Flood
Forecasting Technology, 17–19 October 2005, Tromsø, Norway. http://www.actif-ec.net/
conference2005/proceedings/index.html. Accessed 15 January 2008
Environment Agency (2002) Fluvial Flood Forecasting for Flood Warning - Real Time Modelling.
Defra/Environment Agency Flood and Coastal Defence R&D Programme. R&D Technical
Report W5C-013/5/TR
Environment Agency (2004) Best Practice in Coastal Flood Forecasting. Defra/Environment
Agency Flood and Coastal Defence R&D Programme, R&D Technical Report FD2206/TR1
Environment Agency (2007) Probabilistic Flood Forecasting Scoping Study. Joint Defra/
Environment Agency Flood and Coastal Erosion Risk Management R&D Programme, R&D
Technical Report FD2901/TR
Franklin J L, McAdie C J, Lawrence M B (2003) Trends in track forecasting for tropical cyclones
threatening the United States 1970–2001. Bull. Am. Meteorol. Soc. 1197–1203. http://www.
nhc.noaa.gov/pdf/03franklin.pdf. Accessed 15 January 2008
Garcia Moya J A, Callado A, Santos C, Santos Muñoz D, Simarro J (2006) Multi-model Ensemble
for Short Range Predictability. Proceedings of the 1st MAP D-Phase Scientific Meeting, 6–8
278 References
November 2006, Vienna. http://www.univie.ac.at/IMG-Wien/meetings/map_d-phase/proceedings.
pdf. Accessed 15 January 2008
Golding B W (2000) Quantitative precipitation forecasting in the UK. J. Hydrol. 239: 286–305
Goovaerts P (2000) Geostatistical approaches for incorporating elevation into the spatial interpola-
tion of rainfall. J. Hydrol. 228: 113–129
Grimes D I F, Coppola E, Verdecchia M, Visconti G (2003) A neural network approach to real
time rainfall estimation for Africa using satellite data. J. Hydrometeorol. 4: 1119–1133
Grimes D I F, Diop M (2003) Satellite based rainfall estimation for river flow forecasting in
Africa. I: rainfall estimates and hydrological forecasts. Hydolog. Sci. J. 48(4): 567–584
Harrison D, Scovell R, Lewis H, Matthews S (2006) The Development of the EUMETNET
OPERA Radar Data Hub. Proceedings of ERAD 2006. http://www.grahi.upc.es/ERAD2006/
index.php. Accessed 15 January 2008
Hershey R W (1999) Hydrometry: Principles and Practices. Wiley, New York
Holt A R, Collier C G, Upton G J G, Rahimi A R, Cummings R J, Robbins G L (2005) The Use
of Dual-Frequency Microwave Links for Measuring Path-Averaged Rain. Paper for the Inter-
Agency Committee on the Hydrological Applications of Weather Radar
Hughes D, Greenwood P, Blair G, Coulson G, Pappenberger F, Smith P, Beven K (2006) An intel-
ligent and adaptable grid-based flood monitoring and warning system. UK eScience All Hands
Meeting
Intergovernmental Oceanographic Commission (1994) Manual on sea level measurement and
interpretation. Volume II - Emerging Technologies, UNESCO
International Standards Organisation (1996, 1998) Measurement of Liquid Flow in Open
Channels. Part 1: Establishment and operation of a gauging station and Part 2: Determination
of the stage-discharge relation. ISO 1100-1, 1100-2, Geneva
Leijnse H R, Uijlenhoet R, Holleman I (2008) Sources of error in microwave link rainfall estima-
tion. International Symposium on Weather Radar and Hydrology, Grenoble, France
Love T B (2006) An Overview of Satellite-Based Precipitation Estimation for Flash Flood
Modeling and Forecasting. International Workshop on Flash Flood Forecasting, 13–17 March
2006, Costa Rica, CA. http://www.nws.noaa.gov/iao/iao_FFW.php. Accessed 15 January 2008
Massel S R (1996) Ocean Surface Waves; Their Physics and Prediction. World Scientific
Publisher, Singapore/NJ/London/Hong Kong
Meischner P (2005) Weather Radar: Principles and Advanced Applications. Series: Physics of
Earth and Space Environments. Springer, New York.
Moore R J, Jones A E, Jones D A, Black K B, Bell V A (2004) Weather Radar for Flood
Forecasting: Some UK Experiences. Sixth International Symposium on Hydrological
Applications of Weather Radar, Melbourne, Australia, 2–4 February 2004
Nemeth K (2006) Enhanced Precipitation Identifier and New Generation of Present Weather
Sensor. International Workshop on Flash Floods Forecasting, 13–17 March 2006, Costa Rica,
CA. http://www.nws.noaa.gov/iao/iao_FFW.php. Accessed 15 January 2008
NOAA/NWS (1997) Automated Local Flood Warning Systems Handbook, Weather Service
Hydrology Handbook No. 2. National Oceanic and Atmospheric Administration, National
Weather Service, Office of Hydrology, Washington DC. http://www.weather.gov/oh/docs/
alfws-handbook/. Accessed 15 January 2008
Oi H, Opadevi J (2006) Development of Equipment for Community Flash Flood Warning in the
Caribbean and the Central America. International Workshop on Flash Flood Forecasting,
13–17 March 2006, Costa Rica, CA. http://www.nws.noaa.gov/iao/iao_FFW.php. Accessed 15
January 2008
Park J, Oh B, Jeon S, Kim J (2006) The Automatic Discharge Acquisition & Management System
(ADAMS) in Korea. 7th International Conference on Hydroinformatics, HIC 2006, Nice,
France
Pedersen L, Jensen N E, Madsen H (2007) Network Architecture for Small X-Band Weather
Radars - Test Bed for Automatic Intercalibration and Nowcasting. Paper 12B.2, 33rd
Conference on Radar Meteorology, Cairns, Australia. http://ams.confex.com/ams/33Radar/
techprogram/programexpanded_423.htm. Accessed 15 January 2008
References 279
Price C, Yair Y, Mugnai A, Lagouvardos K, Llasat M S, Michaelides S (2007) FLASH: a new EU
project related to Mediterranean flash floods. Geophys. Res. Abstracts 9, 02638. http://www.
copernicus.org/EGU/gra/volume_9.html. Accessed 15 January 2008
Rahimi A R, Holt AR, Upton G J G, Cummings R J (2003) Use of dual-frequency microwave
links for measuring path-averaged rainfall. J. Geophys. Res. 108(D15): 4467
Rebora N, Ferraris L, von Hardenberg J, Provenzale A (2006) Rainfall downscaling and flood
forecasting: a case study in the Mediterranean area. Nat. Hazards Earth Syst. Sci. 6: 611–619
Rohr P C, Husebye S (2005) Flood Forecasting Technology in Norway, Innovation and Advances.
ACTIF International Conference on Innovation Advances and Implementation of Flood
Forecasting Technology, Tromsø, Norway, 17–19 October 2005. http://www.actif-ec.net/
conference2005/proceedings/index.html. Accessed 15 January 2008
Rotach M W, Arpagaus M, Dorninger M, Hegg C, Montani A, Ranzi R (2007) The D-PHASE
Operations Period (DOP). 3rd HEPEX Workshop, 27–29 June 2007, Stresa, Italy
Schaake J, Restrepo P, Seo D J, Demargne J, Limin W, Perica S (2005) Hydrological Ensemble
Prediction: Challenges and Opportunities. ACTIF International conference on innovation
advances and implementation of flood forecasting technology, Tromsø, Norway, 17–19
October 2005. http://www.actif-ec.net/conference2005/proceedings/index.html. Accessed 15
January 2008
Schaefer G L, Paetzold R F (2000) SNOTEL (SNOwpack TELemetry) and SCAN (Soil Climate
Analysis Network). Paper Presented at “Automated Weather Stations for Applications in
Agriculture and Water Resources Management: Current Use and Future Perspectives”, March
2000, Lincoln, NB
Sene K J., Gash J H C, McNeil D (1991) Evaporation from a tropical lake; comparison of theory
with direct measurements. J. Hydrol. 127: 193–217
Sene K J, Jones D A, Meigh J R, Farquharson F A K (1998) Rainfall and flow variations in the
Lesotho Highlands. Int. J. Climatol. 18: 329–345
Sene K J, Ainsworth E J, Graham G (2006) Risk Based Design of Telemetry Networks for Flood
Warning Applications. 41st Defra/Environment Agency Flood and Coastal Management
Conference, July 2005, York, England
Stow D, Bradley S G, Farrington K E, Dirks K N, Gray W R (1998) A Rain Gauge for the
Measurement of Finescale Temporal Variations. J. Atmos. Ocean. Tech. 15(1): 127–135
Strangeways I (2007) Precipitation: Theory, Measurement and Distribution. Cambridge University
Press, Cambridge
Todini E (2001) A Bayesian technique for conditioning radar precipitation estimates to rain-gauge
measurements. Hydrol. Earth Syst. Sci. 5(2): 187–199. http://www.hydrol-earth-syst-sci.
net/5/187/2001/hess-5-187-2001.html. Accessed 15 January 2008
U.S. Army Corps of Engineers USACE (1996) Hydrologic Aspects of Flood Warning-
Preparedness Programs. Report ETL 1110-2-540, Office of Chief of Engineers, Washington,
DC. http://www.tpub.com/content/USACETechnicalletters/ETL-1110-2-540/index.htm.
Accessed 15 January 2008
Wilson J W (2004) Precipitation Nowcasting: Past, Present and Future. Sixth International
Symposium on Hydrological Applications of Weather Radar, Melbourne, Australia. http://
www.bom.gov.au/bmrc/basic/old_events/hawr6/qpf/WILSON_KEYNOTE.pdf. Accessed 15
January 2008
World Meteorological Organization (1980) Manual on Stream Gauging. Volumes I and II,
Operational Hydrology Report No. 13. WMO No. 519, Geneva
World Meteorological Organisation (1994a) An Overview of Selected Techniques for Analysing
Surface Water Data Networks. WMO No. 806, Geneva
World Meteorological Organisation (1994b) Guide to Hydrological Practices. 5th edition. WMO
No. 168, Geneva
World Meteorological Organisation (1998) Guide to Wave Analysis and Forecasting. 2nd edition.
WMO No. 702, Geneva
World Meteorological Organisation (2000) Precipitation Estimation and Forecasting. Operational
Hydrology Report No. 46, Geneva
280 References
World Meteorological Organisation (2003) The Role of Satellites in WMO programmes in the
2010s. WMO Technical Document No. 1177, Geneva
World Meteorological Organisation (2005) WHYCOS Guidelines. WMO/TD No. 1282, Geneva.
http://www.whycos.org/IMG/pdf/TD.pdf. Accessed 15 January 2008
Xu K, Zhang J, Watanabe M, Sun C (2004) Estimating river discharge from very high-resolution
satellite data: a case study in the Yangtze River, China. Hydrol. Process. 18(10): 1927–1939
Zakharova E A, Kouraev A V, Cazenave A, Seyler F (2005) Amazon discharge estimated from
TOPEX/Poseidon altimetry. C.R. Geosci. 338: 188–196. doi:10.1016/j.crte.2005.10.003
Chapter 3 Thresholds
Bernard P (2004) Hydrometeorological Forecasts for Hydropower Generation. Workshop on the
Hydrological Ensemble Prediction Experiment (HEPEX), Reading, March 8–10. http://ecmwf.int/
newsevents/meetings/workshops/2004/HEPEX/presentations.html. Accessed 15 January 2008
Collier C G, Rihan F, Davies F, Robbins G L (2005) On the Challenges Arising from the Use of
High Resolution NWP Models to Forecast Heavy Rain. ACTIF International Conference on
Innovation Advances and Implementation of Flood Forecasting Technology, 17–19 October,
Tromsø, Norway. http://www.actif-ec.net/conference2005/proceedings/index.html. Accessed
15 January 2008
Collier C G, (2007) Flash flood forecasting: what are the limits of predictability? Q.J.R. Meteorol.
Soc. 133: 3–23
Cox P M, Betts R A, Bunton C B, Essery R L H, Rowntree P R, Smith J (1999) The impact of
new land surface physics on the GCM simulation of climate and climate sensitivity. Clim.
Dynam. 15: 183–203
Environment Agency (2002) Fluvial Flood Forecasting for Flood Warning - Real Time Modelling.
Defra/Environment Agency Flood and Coastal Defence R&D Programme. R&D Technical
Report W5C-013/5/TR
Environment Agency (2004a) Best Practice in Coastal Flood Forecasting. Defra/Environment
Agency Flood and Coastal Defence R&D Programme, R&D Technical Report FD2206/TR1
Environment Agency (2004b) Performance Measures for Flood Forecasting: Review and
Recommendations. Defra/Environment Agency Flood and Coastal Defence R&D Programme,
R&D Technical Report W5C-021/2b/TR
Environment Agency (2007) Extreme Event Recognition Phase 2. Joint Defra/Environment
Agency Flood and Coastal Erosion Risk Management R&D Programme, R&D Technical
Report FD2208/TR
Fouchier C, Arnaud P, Lavabre J, Mizzi J P (2007) AIGA: an operational tool for flood warning in
southern France. Principle and performances on Mediterranean flash floods. Geophys. Res. Abstracts
9, 02843. http://www.copernicus.org/EGU/gra/volume_9.html. Accessed 15 January 2008
Georgakakos K P (2005) Modern Operational Flash Flood Warning Systems Based on Flash Flood
Guidance Theory: Performance Evaluation. ACTIF International Conference on Innovation
Advances and Implementation of Flood Forecasting Technology, Tromsø, Norway, 17–19 October.
http://www.actif-ec.net/conference2005/proceedings/index.html. Accessed 15 January 2008
Georgakakos K P (2006) Analytical results for operational flash flood guidance. J. Hydrol. 317:
81–103
Graham G, Johnson T (2007) Simple Triggers for Flood Warning. 42nd Defra Flood and Coastal
Management Conference, July 2007, York, England
Jolliffe I T, Stephenson D B (2003) Forecast Verification: A Practitioner’s Guide in Atmospheric
Science. Wiley, Chichester, UK, pp. 240
Kok C J, Vogelezang D H P, Wichers Schreur B, Holleman I (2008) Description and use of the
automated warning system for the Dutch water boards. Royal Netherlands Meteorological
Institute (KNMI), Netherlands
References 281
Martina M L V, Todini E, Libralon A (2006) A Bayesian decision approach to rainfall thresholds
based flood warning. Hydrol. Earth Syst. Sci. 10: 413–426. http:// www.hydrol-earth-syst-sci.
net/10/413/2006/. Accessed 15 January 2008
National Weather Service (2003) Flash Flood Guidance Improvement Team. Final Report: 6
February, Washington, DC. http://www.nws.noaa.gov/oh/rfcdev/docs/ffgitreport.pdf. Accessed
15 January 2008
NOAA-USGS (2005) NOAA-USGS Debris-Flow Warning System. Final Report: U.S. Geological
Survey Circular 1283. http://pubs.usgs.gov/circ/2005/1283/. Accessed 15 January 2008
Obled C, Datin R (1997) Rainfall Information Requirements for Mediterranean Flood Operational
Forecasts. In Proc. Workshop on River Basin Modelling, Management and Flood Mitigation,
Padua, Italy, 25–26 September
Price C, Yair Y, Mugnai A, Lagouvardos K, Llasat M C, Michaelides S (2007) FLASH: A New
EU Project Using Lightning Data to Study Mediterranean Flash Floods. 13th International
Conference on Atmospheric Electricity, August 13–18, Beijing
Reed S, Schaake J, Zhangl Z (2006) A Statistical-Distributed Hydrologic Model for Flash Flood
Forecasting. International Workshop on Flash Flood Forecasting. 13–17 March, San Jose,
Costa Rica. http://www.nws.noaa.gov/iao/iao_FFW.php. Accessed 15 January 2008
Stanski H R, Wilson L J, Burrows W R (1989) Survey of Common Verification Methods in
Meteorology. World Weather Watch Technical Report No.8, WMO/TD No.358, World
Meteorological Organisation, Geneva, Switzerland
Sweeney T L (1992) Modernized Areal Flash Flood Guidance. NOAA Technical Report NWS
HYDRO 44, Hydrology Laboratory, National Weather Service, NOAA, Silver Spring, MD
Tissot P E, Duff S, Michaud P, Rizzo J, Jeffress G (2005) DNR-TCOON: An Integrated
Observation and Operational System for the Gulf of Mexico. American Meteorological
Society’s (AMS) 85th Annual Meeting, 9–13 January, San Diego, CA. http://ams.confex.com/
ams/Annual2005/techprogram/paper_83842.htm. Accessed 15 January 2008
Torfs P J J F (2004) Black Box Forecasting and Uncertainty. ACTIF Workshop on Identification,
Quantification, Propagation and Communication of Uncertainties in Flood Forecasting. 22–23
November, Delft, The Netherlands. http://www.actif-ec.net/news.htm. Accessed 15 January 2008
U.S. Army Corps of Engineers USACE (1996) Hydrologic Aspects of Flood Warning-
Preparedness Programs. Report ETL 1110–2–540, Office of Chief of Engineers, Washington,
DC. http://www.tpub.com/content/USACETechnicalletters/ETL-1110–2–540/index.htm.
Accessed 15 January 2008
World Meteorological Organisation (1994) Guide to Hydrological Practices (5th edition). WMO
No. 168, Geneva, Switzerland
Chapter 4 Dissemination
Andryszewski A, Evans K, Haggett C, Mitchell B, Whitfield D, Harrison T (2005a) Levels of
Service Approach to Flood Forecasting and Warning. ACTIF International Conference on
Innovation Advances and Implementation of Flood Forecasting Technology. 17–19 October
2005, Tromsø, Norway. http://www.actif-ec.net/conference2005/proceedings/index.html.
Accessed 15 January 2008
Andryszewski A, Evans K, Haggett C, Mitchell B, Whitfield D (2005b) Flood Warning – The
Next Step Change. 40th Defra Flood and Coastal Management Conference, York, England
Emergency Management Australia (1999) Guide 5 – Flood Warning. Australian Emergency
Manuals Series, Part III Emergency Management Practice. Volume 3, 2nd edition. http://www.
ema.gov.au/. Accessed 15 January 2008
Environment Agency (2002) Fluvial Flood Forecasting for Flood Warning - Real Time Modelling.
Defra/Environment Agency Flood and Coastal Defence R&D Programme. R&D Technical
Report W5C-013/5/TR
282 References
Environment Agency (2004) Best Practice in Coastal Flood Forecasting. Defra/Environment
Agency Flood and Coastal Defence R&D Programme, R&D Technical Report FD2206/TR1
Holland G (Ed.) (2007) Global Guide to Tropical Cyclone Forecasting. Bureau of Meteorology
Research Centre (Australia) WMO/TC-No. 560, Report No. TCP-31, World Meteorological
Organization, Geneva, Switzerland. http://www.bom.gov.au/bmrc/pubs/tcguide/globa_guide_
intro.htm. Accessed 15 January 2008
ISDR (2006) Developing Early Warning Systems: A Checklist. EWC III Third International
Conference on Early Warning from Concept to Action, 27–29 March 2006, Bonn, Germany.
http://www.unisdr.org/ppew/. Accessed 15 January 2008
Martini F, De Roo A (Eds.) (2007) EXCIFF Guide: Good Practice for Delivering Flood Related
Information to the General Public. European Commission/Joint Research Centre Report
EUR22760EN. http://exciff.jrc.it/downloads/exciff-related-documents/EXCIFF_guide.pdf.
Accessed 15 January 2008
Morel G, Despres S (2006) ISLAND: Improving the Communication of Information on Flooding
and Other Risks to Local Communities in South East Asia. http://www.isted.com/programmes/
island/anglais/documentation.htm. Accessed 15 January 2008
NOAA/NWS 1997 Automated Local Flood Warning Systems Handbook. Weather Service
Hydrology Handbook No. 2. National Oceanic and Atmospheric Administration, National
Weather Service, Office of Hydrology, Washington DC. http://www.weather.gov/oh/docs/
alfws-handbook/. Accessed 15 January 2008
Oi H, Opavedi J (2006) Development of Equipment for Community Flash Flood Warning in the
Caribbean and the Central America. WMO International Workshop on Flash Flood Forecasting,
13–17 March 2006, Costa Rica, CA. http://www.nws.noaa.gov/iao/iao_FFW.php. Accessed 15
January 2008
Parker D J (2003) Designing Flood Forecasting, Warning and Response Systems from a Societal
Perspective. Paper presented at the International Conference on Alpine Meteorology and
Meso-Alpine Programme, 21 May 2003, Brig, Switzerland. http://www.map.meteoswiss.ch/
map-doc/icam2003/Presentation/10.1/Brig_document.pdf. Accessed 15 January 2008
Sponberg K (2006) RANET Dissemination and Communication of Environmental Information for
Rural and Remote Community Development. WMO International Workshop on Flash Flood
Forecasting, 13–17 March 2006, Costa Rica, CA. http://www.nws.noaa.gov/iao/iao_FFW.php.
Accessed 15 January 2008
Tilford K A, Sene K J, Khatibi R (2007) Flood forecasting model selection: a structured approach.
In Begum S, Stive M J F, Hall J W (Eds.), Flood Risk Management in Europe. Springer,
Dordrecht, The Netherlands
UNESCO (2006) Tsunami Teacher: An Information and Resource Toolkit, Building Global
Capacity to Respond to and Mitigate Tsunamis. IOC Manuals and Guides 47. http://unesdoc.
unesco.org/images/0014/001496/149617e.pdf. Accessed 15 January 2008
U.S. Army Corps of Engineers (USACE) (1996) Hydrologic Aspects of Flood Warning-Preparedness
Programs. Report ETL 1110–2–540, Office of Chief of Engineers, Washington, DC. http://www.tpub.
com/content/USACETechnicalletters/ETL-1110–2–540/index.htm. Accessed 15 January 2008
World Meteorological Organisation (1994) Guide to Hydrological Practices. 5th edition. WMO
No. 168, Geneva
World Meteorological Organisation (2006) Preventing and Mitigating Natural Disasters. WMO
No. 993, Geneva
Chapter 5 General Principles
Anderson M G, Bates P D (Eds.) (2001) Model Validation: Perspectives in Hydrological Science.
Wiley, Chichester, UK
Beven K J (2001) Rainfall Runoff Modelling – The Primer. Wiley, Chichester, UK
References 283
Butts M B, Falk A K, Hartnack J, Madsen H, Klinting A, van Kalken T, Cadman D, Price D (2005)
Ensemble Based Methods for Data Assimilation and Uncertainty Estimation in the
FLOODRELIEF Project. ACTIF International Conference on Innovation Advances and
Implementation of Flood Forecasting Technology, 17–19 October, Tromsø, Norway. http://
www.actif-ec.net/conference2005/proceedings/index.html. Accessed 15 January 2008
Carsell K M, Pingel N D, Ford D T (2004) Quantifying the Benefit of a Flood Warning System.
Natural Hazards Review, ASCE, August 2004: 131–140
Chen Y, Sene K J, Hearn K (2005) Converting Section 105 or SFRM Hydrodynamic River Models
for Real Time Forecasting Applications. 40th Defra Flood and Coastal Defence Conference,
York, England
Cranston M, Werner M, Janssen A, Hollebrandse F, Lardet P, Oxbrow J, Piedra M (2007) Flood
Early Warning System (FEWS) Scotland: An Example of Real Time System and Forecasting
Model Development and Delivery Best Practice. 42nd Defra Flood and Coastal Defence
Conference, York, England
El Serafy G Y, Mynett A E (2004) Comparison of EKF and ENKF in Sobek River: Case Study
Maxau ljssel. HYDROINFORMATICS Vol. 1, Proceedings of the 6th International Conference,
Singapore, 21–24 June 2004
Elliott J, Catchlove R, Sooriyakumaran S, Thompson R (2005) Recent Advances in the
Development of Flood Forecasting and Warning Services in Australia. ACTIF International
Conference on Innovation Advances and Implementation of Flood Forecasting Technology,
Tromso, Norway, 17–19 October2005. http://www.actif-ec.net/conference2005/proceedings/
index.html. Accessed 15 January 2008
Environment Agency (2002) Fluvial Flood Forecasting for Flood Warning - Real Time Modelling.
Defra/Environment Agency Flood and Coastal Defence R&D Programme. R&D Technical
Report W5C-013/5/TR
Environment Agency (2004) Performance Measures for Flood Forecasting: Review and
Recommendations. Defra/Environment Agency Flood and Coastal Defence R&D Programme,
R&D Technical Report W5C-021/2b/TR
Environment Agency (2007) Probabilistic Flood Forecasting Scoping Study. Joint Defra/EA Flood
and Coastal Erosion Risk Management R&D Programme, R&D Technical Report FD2901/TR
Fortune D (2006) Floods and Hydroinformatics: New Challenges. 7th International Conference on
Hydroinformatics, HIC 2006, Nice, France, Volume 1, pp. 122–127
Georgakakos K P, Seo D-J, Gupta H, Schaake J, Butts M B (2004) Towards the characterization of
streamflow simulation uncertainty through multimodel ensembles. J. Hydrol. 298: 222–241
Goniadzki D (2006) Institutional Aspects of the Hydrological Warning System in the Del Plata
Basin. WMO International Workshop on Flash Flood Forecasting, 13–17 March, San José
Costa Rica. http://www.nws.noaa.gov/iao/iao_FFW.php. Accessed 15 January 2008
Goswami M, O’Connor K M, Bhattarai K P, Shamseldin A Y (2005) Assessing the performance of
eight real time updating models and procedures for the Brosna River. Hydrol. Earth Syst. Sci.
9(4): 394–411. http://www.hydrol-earth-syst-sci.net/9/394/2005/. Accessed 15 January 2008
Huaimin G (2005) Flood Forecasting and Management in China. ACTIF International Conference
on Innovation Advances and Implementation of Flood Forecasting Technology, 17–19 October
2005, Tromsø, Norway. http://www.actif-ec.net/conference2005/proceedings/index.html.
Accessed 15 January 2008
Jolliffe I T, Stephenson D B (2003) Forecast Verification. A Practitioner’s Guide in Atmospheric
Science. Wiley, Chichester, UK
Krzysztofowicz R (2001) The case for probabilistic forecasting in hydrology. J. Hydrology
249: 2–9
Laio F, Tamea S (2007) Verification tools for probabilistic forecasts of continuous hydrological
variables. Hydrol. Earth Syst. Sci. 11:1267–1277. http://www.hydrol-earth-syst-sci.
net/11/1267/2007/. Accessed 15 January 2008
Moore R J (1999) Real-time flood forecasting systems: perspectives and prospects. In: R. Casale
and C.Margottini (Eds.), Floods and Landslides: Integrated Risk Assessment, Chapter 11.
Springer-Verlag, Berlin and Heidelberg pp. 147–189
284 References
Pappenberger F, Matgen P, Beven K J, Henry J-B, Pfister L, de Fraipont P (2004) The influence
of rating curve uncertainty on flood inundation predictions. Proc. IMA International
Conference on Flood Risk Assessment, Bath, 7–8 September, England
Pappenberger F, Beven K, Frodsham K, Romanowicz R, Matgen P (2007) Grasping the unavoid-
able subjectivity in calibration of flood inundation models: a vulnerability weighted approach.
J. Hydrol. 333(2–4): 275– 287
Paudyal G N (2002) Forecasting and warning of water-related disasters in a complex hydraulic
setting—the case of Bangladesh. Hydrolog. Sci. J. Special Issue: Towards Integrated Water
Resources Management for Sustainable Development 47(S), August: S5–S18
Rebora N, Ferraris L, von Hardenberg J, Provenzale A (2006) Rainfall downscaling and flood
forecasting: a case study in the Mediterranean area. Nat. Hazards Earth Syst. Sci. 6: 611–619.
Reed D W (1984) A Review of British Flood Forecasting Practice. Institute of Hydrology, Report
No. 90
Richardson (2003) Economic Value and Skill. Chapter 8 in Jolliffe I T, Stephenson DB (Eds.)
(2003) Forecast Verification. A Practitioner’s Guide in Atmospheric Science. Wiley and
Sons Ltd
Røhr P C, Husebye S (2005) Flood Forecasting Technology in Norway, Innovation and
Advances. ACTIF International Conference on Innovation Advances and Implementation of
Flood Forecasting Technology, 17–19 October, Tromsø, Norway. http://www.actif-ec.net/
conference2005/proceedings/index.html. Accessed 15 January 2008
Romanowicz R J,Young P C, Beven K J (2006) Data assimilation and adaptive forecasting of
water levels in the river Severn catchment, United Kingdom. Water Resour. Res. 42(6)
doi:10.1029/2005WR004373
Rotach M W, Arpagaus M, Dorninger M, Hegg C, Montani A, Ranzi R (2007) The D-PHASE
Operations Period (DOP). 3rd HEPEX Workshop, 27–29 June, Stresa, Italy
Roulin E (2007) Skill and relative economic value of medium-range hydrological ensemble pre-
dictions. Hydrol. Earth Syst. Sci., 11, 725–737. http://www.hydrol-earth-syst-sci.
net/11/725/2007/hess-11-725-2007.html. Accessed 15 January 2008
Rungo M, Refsgaard J K, Havno K (1989) The updating procedure in the MIKE 11 modelling
system for real time forecasting. In Cluckie I D, Collier C G (Eds.), Hydrological Applications
of Weather Radar, Ellis Horwood, Chichester, West Sussex, England.
Schaake J, Restrepo P, Seo D J, Demargne J, Wu L, Perica S (2005) Hydrological Ensemble Prediction:
Challenges and Opportunities. ACTIF International Conference on Innovation Advances and
Implementation of Flood Forecasting Technology, 17–19 October, Tromso, Norway. http://www.
actif-ec.net/conference2005/proceedings/index.html. Accessed 15 January 2008
Serban P, Askew A J (1991) Hydrological forecasting and updating procedures. IAHS Publ. No.
201: 357–369
Sperfslage J A, Georgakakos K P, Carpenter T M, Shamir E, Graham N E (2005) Design and
Implementation of a Real-Time Regional Flash Flood Guidance System for Central America
Driven by Adjusted GOES Satellite Rainfall. 21st International Conference on Interactive
Information Processing Systems (IIPS) for Meteorology, Oceanography, and Hydrology. http://
ams.confex.com/ams/Annual2005/techprogram/session_17767.htm. Accessed 15 January 2008
Thielen J, De Roo A, Niemeyer S, Gouweleeuw B, Van der Knijff J, Franchello G (2004)
Development and testing of a Pan-European Flood Alert System (EFAS). Abstracts of 2nd
European Flood Alert System (EFAS) Workshop, Ispra, 10–12 November 2004. http://efas.jrc.
it/workshop2004/. Accessed 15 January 2008
Tozer N, Saulter A, Hawkes P (2007) Coastal Flood Forecasting. 42nd Defra Flood and Coastal
Management Conference, York, England
U.S. Army Corps of Engineers (USACE) (1994) Framework for Estimating National Economic
Development Benefits and Other Beneficial Effects of Flood Warning and Preparedness
Systems. Institute for Water Resources Report IWR-94–3, Alexandria, VA.
Vehviläinen B, Huttunen M, Huttunen I (2005) Hydrological Forecasting and Real time
Monitoring in Finland: The Watershed Simulation and Forecasting System (WSFS). ACTIF
International Conference on Innovation Advances and Implementation of Flood Forecasting
References 285
Technology, 17–19 October, Tromsø, Norway. http://www.actif-ec.net/conference2005/pro-
ceedings/index.html. Accessed 15 January 2008
Verlaan M, Zijderveld A, de Vries H, Kroos J (2005) Operational storm surge forecasting in the
Netherlands: developments in the last decade. Phil. Trans. Royal Society 363(1831): 1441–1453
Weerts A H, El Sarafy G (2005) Particle Filtering and Ensemble Kalman Filtering for Input
Correction in Rainfall Runoff Modeling. ACTIF International Conference on Innovation
Advances and Implementation of Flood Forecasting Technology, 17–19 October, Tromsø,
Norway. http://www.actif-ec.net/conference2005/proceedings/index.html. Accessed 15
January 2008
Werner M, Self K (2005) Measuring Performance in Flood Forecasting: A Question of Model
Reliability or of Forecast Reliability. ACTIF International Conference on Innovation
Advances and Implementation of Flood Forecasting Technology, 17–19 October, Tromsø,
Norway. http://www.actif-ec.net/conference2005/proceedings/index.html. Accessed 15
January 2008
Werner M, van Dijk M (2005) Developing Flood Forecasting Systems – Examples from the UK,
Europe and Pakistan. ACTIF International Conference on Innovation Advances and
Implementation of Flood Forecasting Technology, 17–19 October, Tromsø, Norway. http://
www.actif-ec.net/conference2005/proceedings/index.html. Accessed 15 January 2008
Wöhling Th, Lennartz F, Zappa M (2006) Technical Note: Updating procedure for flood forecast-
ing with conceptual HBV-type models. Hydrol. Earth Syst. Sci., 10:783–788. http://www.
hydrol-earth-syst-sci.net/10/783/2006/hess-10-783-2006.html. Accessed 15 January 2008
World Meteorological Organisation (1994) Guide to Hydrological Practices. Fifth Edition. WMO
No. 168, Geneva
Young P C, Tomlin C M (2000) Data-based Mechanistic Modelling and Adaptive Flow
Forecasting. In Lees M, Walsh P (Eds.), Flood Forecasting: What does Current Research Offer
the Practitioner, BHS Occasional Paper 12. British Hydrological Society, London
Chapter 6 Rivers
Abrahart R J, Kneale P E, See L M (Eds.) (2004) Neural Networks for Hydrological Modelling.
Taylor & Francis, London
Andrssian V, Hall A, Chahinian N, Schaake J (Eds.) (2006) Large Sample Basin Experiments for
Hydrological Model Parameterization: Results of the Model Parameter Experiment-MOPEX,
IAHS Publication, 307
Arduino G, Reggiani P, Todini E (2005) Recent advances in flood forecasting and flood risk
assessment. Hydrol. Earth Syst. Sci. 9(4): 280–284. http://www.hydrol-earth-syst-sci.
net/9/280/2005/hess-9-280-2005.html. Accessed 15 January 2008
ASCE (2000) Artificial neural networks in hydrology II: Hydrologic applications. J. Hydrol. Eng.
5(2): 124–137
Bell V A, Kay A L, Jones R G, Moore R J (2007) Development of a high resolution grid-based
river flow model for use with regional climate model output. Hydrol. Earth Syst. Sci. 11(1):
532–549. http://www.hydrol-earth-syst-sci.net/11/532/2007. Accessed 15 January 2008
Bergstrom S (1995) The HBV model. In: Singh V P (Ed.), Computer Models of Watershed
Hydrology. Water Resources Publications, Littleton, CO
Beven K J (2001) Rainfall Runoff Modelling – The Primer. Wiley, Chichester
Beven K, Romanowicz R, Pappenberger F, Young P, Werner M (2005) The Uncertainty Cascade
in Flood Forecasting. ACTIF International Conference on Innovation Advances and
Implementation of Flood Forecasting Technology, Tromsø, Norway, 17–19 October 2005.
http://www.actif-ec.net/conference2005/proceedings/index.html. Accessed 15 January 2008
Box G E P, Jenkins G M (1970) Time Series Analysis Forecasting and Control. Holden Day,
San Francisco, CA
286 References
Burnash R J C (1995) The NWS river forecast system – catchment modeling. In Singh V P
(Ed.), Computer Models of Watershed Hydrology, Water Resources Publications, Highlands
Ranch, CO
Butts M B, Overgaard J, Viaene P, Dubicki A, Stroñska K, Szalinska W, Lewandowski A,
Olszewski T, Kolerski (2005) Flexible Process-based Hydrological Modelling Framework
for Flood forecasting – MIKE SHE. ACTIF International Conference on Innovation
Advances and Implementation of Flood Forecasting Technology, Tromsø, Norway, 17–19
October 2005. http://www.actif-ec.net/conference2005/proceedings/index.html. Accessed
15 January 2008
Chanson H (2004) Hydraulics of Open Channel Flow: An Introduction. 2nd edition. Butterworth-
Heinemann, Oxford
Chen Y, Sene K J, Hearn K (2005) Converting Section 105 or SFRM Hydrodynamic River Models
for Real Time Forecasting Applications. 40th Defra Flood and Coastal Defence Conference,
York, England
Chow V T (1959) Open Channel Hydraulics. McGraw-Hill, New York
Collischonn W, Allasia D G, Silva B C, Tucci C E M (2007) The MGB-IPH model for large scale
rainfall runoff modeling. Hydrol. Sci. J. 52(5): 878–895
Cunge J A (1969) On the subject of a flood propagation method. J. Hydraul. Res. IAHR 7:
205–230
Dawson C W, Wilby R L (1999) A comparison of artificial neural networks used for flow forecasting.
Hydrol. Earth Syst. Sci. 3: 529–540
De Roo A D P, Wesseling C G, Van Deursen W P A (2000) Physically based river basin modelling
within a GIS: the LISFLOOD model. Hydrol. Process. 14(11–12): 1981–1992
Environment Agency (2002) Fluvial Flood Forecasting for Flood Warning - Real Time Modelling.
Defra/Environment Agency Flood and Coastal Defence R&D Programme. R&D Technical
Report W5C-013/5/TR
European Flood Forecasting System EFFS (2003) Final Report WP8. Deliverable 8.3, WL/Delft
Hydraulics. http://effs.wldelft.nl/. Accessed 15 January 2008
Han D (1991) Weather Radar Information Processing and Real Time Flood Forecasting. Ph.D.
thesis, University of Salford
Huband M, Sene K J (2005) Integrated catchment modelling issues for flow forecasting applica-
tions. Scottish Hydraulics Study Group, Catchment Modelling for Flood Risk Management,
18 March 2005
Jones D A, Moore R J (1980) A simple channel flow routing model for real time use. Proc. Oxford
Symposium Hydrological Forecasting, IAHS Publication, 129
Kneale P E, See L, Kerr P (2000) Developing a Prototype Neural Net Flood Forecasting Model
for Use by the Environment Agency. Proceedings 35th MAFF conference of River and Coastal
Engineers, Keele University, England
Koren V, Reed S, Smith M, Zhang Z, Seo D J (2004) Hydrology Laboratory Research Modeling
System (HL-RMS) of the National Weather Service. J. Hydrology. 291(3/4): 297–318
Lees M, Young P, Ferguson S, Beven K J, Burns J (1994) An adaptive flood warning system for
the River Nith at Dumfries. In White W R, Watts J (Eds.), 2nd International Conference on
River Flood Hydraulics. Wiley, Chichester
Lees M J (2000) Advances in transfer function based flood forecasting. In Lees M, Walsh P (Eds.),
Flood Forecasting: What does Current Research Offer the Practitioner. BHS Occasional Paper
12, British Hydrological Society, London
Lettenmaier D P, Wood E F (1993) Hydrologic forecasting. In Maidment D R (Ed.), Handbook of
Hydrology. Chapter 26, McGraw-Hill, New York
Lighthill M J, Whitham G B (1955) On Kinematic Waves. I – Flood Movement in Long Rivers.
Proc. Roy. Soc. London Series A 229: 281–316
Liu Z, Martina M L V, Todini E (2005) Flood forecasting using a fully distributed model:
application of the TOPKAPI model to the Upper Xixian Catchment. Hydrol. Earth Syst.
Sci. 9(4): 347–364. http://www.hydrol-earth-syst-sci.net/9/issue4.html. Accessed 15
January 2008
References 287
Madsen H (2000) Automatic calibration of a conceptual rainfall-runoff model using multiple
objectives. J. Hydrol. 235(3–4): 276–288
Malone T (1999) Using URBS for Real Time Flood Modelling. Water 99: Joint Congress; 25th
Hydrology & Water Resources Symposium, 2nd International Conference on Water Resources
& Environment Research
Moore R J (1982) Transfer functions, noise predictors and the forecasting of flood events in real
time. In. Singh, V.P. (Ed.), Statistical Analysis of Rainfall and Runoff. Water Resources
Publications, Littleton, Colorado pp. 229–250.
Moore R J (1985) The probability distributed principle and runoff production at point and basin
scales. Hydrol. Sci. J. 30(2): 273–297
Moore R J (2007) The PDM rainfall runoff model. Hydrol. Earth Syst. Sci. 11(1): 483–499
Moore R J, Cole S J, Bell V A, Jones D A (2006) Issues in flood forecasting: ungauged basins,
extreme floods and uncertainty. In: Tchiguirinskaia I, Thein K N N, Hubert P (Eds.), Frontiers
in Flood Research, 8th Kovacs Colloquium, UNESCO, Paris
Price R K (1977) FLOUT – A River Catchment Flood Model, Report No. IT 168. Hydraulics
Research Station, Wallingford
Reed D W (1984) A Review of British Flood Forecasting Practice. Institute of Hydrology, Report
No. 90
Reggiani P, Schellekens J (2003) Modelling of hydrological responses: the representative elemen-
tary watershed approach as an alternative blueprint for watershed modelling. Hydrol. Process.
17: 3785–3789
Rotach M W, Arpagaus M, Dorninger M, Hegg C, Montani A, Ranzi R (2007) The D-PHASE
Operations Period (DOP). 3rd HEPEX Workshop, 27–29 June 2007, Stresa, Italy
Sene K J, Ainsworth E J, Graham G (2006) Risk Based Design of Telemetry Networks for Flood
Warning Applications. 41st Defra/Environment Agency Flood and Coastal Management
Conference, York, England
Sherman L K (1932) Streamflow from rainfall by the unit graph method. Eng. News Record 108:
501–505
Shrestha R R, Theobald S, Nestmann F (2005) Simulation of flood flow in a river system using
artificial neural networks. Hydrol. Earth Syst. Sci. 9(4): 313–321. http://www.hydrol-earth-
syst-sci.net/9/313/2005. Accessed 15 January 2008
Singh V P (1988) Hydrologic Systems Volume I: Rainfall Runoff Modelling. Prentice-Hall,
Englewood Cliffs, NJ
Singh V P (1989) Hydrologic Systems Volume II: Watershed Modelling, Prentice-Hall, Englewood
Cliffs, NJ
Sivapalan M, Wagener T, Uhlenbrook S, Liang X, Lakshmi V, Kumar P, Zehe E, Tachikawa Y
(eds.) (2006) Predictions in Ungauged Basins: Promises and Progress. IAHS Publication, 303,
May 2006
Solomatine D P, Price R K (2004) Innovative Approaches to Flood Forecasting Using Data Driven
and Hybrid Modelling. 6th International Conference on Hydroinformatics, 21–24 June 2004,
World Scientific Publishing Company, Singapore
Sugarawa, M (1995) Tank model. In: Singh, V P (Ed.), Computer Models of Watershed Hydrology.
Water Resources Publications, Littleton, CO, pp. 165–214
Tang X, Knight D W, Samuels P G (1999) Volume conservation in variable parameter Muskingum
Cunge method. J. Hydraulic. Engineering, 610–620
Tilford K A, Sene K J, Khatibi R (2007) Flood Forecasting Model Selection: A Structured
Approach. In Begum S, Stive M J F, Hall J W (Eds.), Flood Risk Management in Europe,
Springer, Netherlands
Van Der Knijff J, De Roo A, Franchello G (2004) The LISFLOOD Model. Abstracts of 2nd
European Flood Alert System (EFAS) Workshop, Ispra, 10–12 November 2004. http://efas.
jrc.it/workshop2004/. Accessed 15 January 2008
Vieux B (2004) Distributed hydrologic modeling for flood forecasting. GIS for Remote Sensing
in Hydrology. Water Resources and Environment. Proceedings IGRHWE, Three Gorges Dam,
China, September 2003. IAHS Publication 289
288 References
World Meteorological Organisation (1992) Simulated real time intercomparison of hydrological
models. Operational Hydrology Report No. 38, Geneva
World Meteorological Organisation (1994) Guide to Hydrological Practices. 5th edition. WMO
No. 168, Geneva
Young P C, Tomlin C M (2000) Data-based Mechanistic Modelling and Adaptive Flow forecast-
ing. In Lees M, Walsh P (Eds.), Flood Forecasting: What does Current Research Offer the
Practitioner, BHS Occasional Paper 12, British Hydrological Society, London
Young P (2001) Data Based Mechanistic modeling and validation of rainfall flow processes. In
Anderson M G, Bates P D (Eds.), Model Validation: Perspectives in Hydrological Science.
John Wiley & Sons Ltd
Zhao R J (1992) The Xin’anjiang model applied in China. J. Hydrology. 135: 371–381
Chapter 7 Coasts
Berg R, Clark J, Roth D, Birchard T (2007) The National Weather Service Unified Surface
Analysis. Paper 6A-4, 22nd Conference on Weather Analysis and Forecasting/18th Conference
on Numerical Weather Prediction, 25–29 June, UT. http://ams.confex.com/ams/22WAF18NWP/
techprogram/programexpanded_414.htm. Accessed 15 January 2008
Brassington G B, Warren G, Smith N, Schiller A, Oke P R (2005) BLUElink Progress on opera-
tional ocean prediction for Australia. Bull. Aust. Meteorol. Oceanogr. Soc. 18: 104–109.
http://www.cmar.csiro.au/bluelink/scientific/index.htm. Accessed 15 January 2008
Causon D, Ingram D, Mingham C, Zang J, Hu K, Zhou J G (2000) Numerically Simulating Seawall
Overtopping. Proceedings ICCE2000 International Conference on Coastal Engineering
Chassignet E P, Hurlburt E, Smedstad O M, Halliwell G R, Wallcraft A J, Metzger E, Lanton B O,
Lozano C, Rao D B, Hogan P J, Srinivasan A (2006) Generalised vertical coordinates for eddy
resolving global and coastal ocean forecasts. Oceanography 19(1): 20–31. http://coaps.fsu.
edu/people/chassignet/allpubs.shtml. Accessed 15 January 2008
Cheung K F, Phadke A C, Wei Y, Rojas R, Douyere Y J M, Martino C D, Houston S H, Liu P L F,
Lynett P J, Dodd N, Liao S, Nakazaki E (2003) Modeling of storm-induced coastal flooding
for emergency management. Ocean Eng 30(2003): 1353–1386
Davidson N E, Entel M B, Hanstrum B N, Smith N R, Warren G R (2005) Enhancements to the
Bureau of Meteorology’s ocean surge forecasting system using operational TC-LAPS. BMRC
Research Report No. 114. http://www.bom.gov.au/bmrc/pubs/researchreports/researchreports.
htm. Accessed 15 January 2008
Environment Agency (2004) Best Practice in Coastal Flood Forecasting. Defra/Environment
Agency Flood and Coastal Defence R&D Programme, R&D Technical Report FD2206/TR1
Environment Agency (2005) Dependence Between Extreme Sea Surge, River Flow and
Precipitation: A Study in South and West Britain. Defra/Environment Agency Flood and
Coastal Defence R&D Programme. R&D Technical Report FD2308/TR3
Flather R A (2000) Existing operational oceanography. Coast Eng 41: 13–40
Hicks S D (2006) Understanding Tides. Centre for Operational Oceanographic Products and
Services, NOAA. http://tidesandcurrents.noaa.gov/publications/Understanding_Tides_by_
Steacy_finalFINAL11_30.pdf. Accessed 15 January 2008
Holland G (Ed.) (2007) Global Guide to Tropical Cyclone Forecasting. Bureau of Meteorology
Research Centre (Australia) WMO/TC-No. 560, Report No. TCP-31, World Meteorological
Organization, Geneva, Switzerland
http://www.bom.gov.au/bmrc/pubs/tcguide/globa_guide_intro.htm. Accessed 15 January 2008
Holthuijsen L H (2007) Waves in Oceanic and Coastal Waters. Cambridge University Press,
Cambridge
Horsburgh K J, Wilson C (2007) Tide-surge interaction and its role in the distribution of surge
residuals in the North Sea. J Geophys Res 112: C08003, doi:10.1029/2006JC004033
References 289
Hu K, Wotherspoon C (2007) Numerical modelling in coastal flood forecasting and warning in
England and Wales. In: Begum S, Stive M J F, Hall J W (Eds.), Flood Risk Management in
Europe. Innovation in Policy and Practice. Springer, The Netherlands
Huang W, Murray C (2003) Application of an Artificial Neural Network to Predict Tidal Currents
in an Inlet. Technical Report ERDC/CHL CHETN-IV-58, US Army Corps of Engineers
Hubbard M E, Dodd N (2002) A 2D numerical model of wave run-up and overtopping. Coast Eng
47: 1–26
JCOMM (2007) Guide to Storm Surge Forecasting (in preparation). IOC/WMO Joint Technical
Commission for Oceanography and Marine Meteorology
Jelesnianski C P, Chen J, Schaffer W A (1992) SLOSH: Sea, Lake and Overland Surges from
Hurricanes. NOAA Technical Report NWS 48, NOAA
Kirby J T, Dalrymple R A, Shi F (2005) Combined Refraction/Diffraction Model REF/DIF 1
Version 3.0 Documentation and User’s Manual. Center for Applied Coastal Research,
Department of Civil and Environmental Engineering, University of Delaware, Newark, DE
Knabb R D, Rhome J R, Brown D P (2006) Tropical Cyclone Report: Hurricane Katrina 23–30 August
2005. National Hurricane Center, 20 December 2005 (updated 10 August 2006 for tropical wave
history, storm surge, tornadoes, surface observations, fatalities, and damage cost estimates)
Komen G J, Cavaleri P, Donelan M, Hasselmann K, Hasselmann S, Janssen P (1994) Dynamics
and Modelling of Ocean Waves. Cambridge University Press, Cambridge
Lam H, Lam Q C C, Chu H C K (2005) Risk Management of Tropical Cyclone-Induced Flooding
in Hong Kong. Workshop on Risk Management Towards Development Goals and Socio-
economic Impact Assessment of Typhoon-related Disaster, Kuala Lumpur, Malaysia, 5–9
September 2005
Lees M, Young P, Ferguson S, Beven K J, Burns J (1994) An adaptive flood warning system for
the River Nith at Dumfries. In: White W R, Watts J (Eds.), 2nd International Conference on
River Flood Hydraulics. Wiley, Chichester, UK
Makarynskyy O, Makarynska D, Kuhn M, Featherstone W E (2004) Predicting sea level varia-
tions with artificial neural networks at Hillarys Boat Harbour, Western Australia. Estuarine,
Coastal and Shelf Science 61(2): 351–360. http://espace.lis.curtin.edu.au/archive/00000025/01/
W291.pdf. Accessed 15 January 2008
Makarynskyy O, Pires-Silva A A, Makarynska D, Ventura-Soares C (2005) Artificial neural net-
works in wave predictions at the west coast of Portugal. Comput Geosci 31(4): 415–424.
http://espace.lis.curtin.edu.au/archive/00000243/01/Comp&Geos04.PDF. Accessed 15 January
2008
Massey W G, Gangai J W, Drei-Horgan E, Slover K J (2007) History of Coastal Inundation
Models. Mar Technol Soc J 41(1): 7–17
Patrick A R, Collins W G, Tissot P E, Drikitis A, Stearns J, Michaud P R (2002) Use of the NCEP
MESOETA Data in a Water Level Predicting Neural Network. Proceedings of the 19th AMS
Conference on Weather Analysis and Forecasting/15th AMS Conference on Numerical
Weather Prediction, San Antonio, TX, pp. 369–372. http://wave.oregonstate.edu/Research/
Publications/Cox/pctdsmc-AMS-02.pdf. Accessed 15 January 2008
Prouty D B, Tissot P E, Anwar A A (2005) Using Artificial Neural Networks for Predicting Storm
Surge Propagation in North Sea and the Thames Estuary. Fourth Conference on Artificial
Intelligence Applications to Environmental Science, San Diego, CA. http://ams.confex.com/
ams/Annual2005/techprogram/programexpanded_264.htm. Accessed 15 January 2008
Randon N, Lawry J, Horsburgh K J, Cluckie I (2007) Fuzzy Bayesian Modelling of Sea-Level
Along the East Coast of Britain. IEEE Transaction on Fuzzy Systems (in press)
Sheng Y P, Paramygin V A, Alymov V, Davis J R (2005) A Real-Time Forecasting System for
Hurricane Induced Storm Surge and Coastal Flooding. 9th International Conference on
Estuarine and Coastal Modeling, Charleston, SC
Shi F, Dalrymple R A, Kirby J T, Chen Q, Kennedy A (2001) A fully nonlinear Boussinesq model
in generalized curvilinear coordinates. Coast Eng 42: 337–358
Solomatine D P, Velickov S, Wüst J C (2001) Predicting water levels and currents in the North Sea using
chaos theory and neural networks. Proceedings of IAHR Congress, Beijing, September 2001
290 References
Tissot P E, Michaud P R, Cox D T (2003) Optimization and Performance of a Neural Network
Model Forecasting Water Levels for the Corpus Christi, Texas, Estuary. 3rd Conference on
Artificial Intelligence Applications to Environmental Science, Long Beach, CA. http://ams.con-
fex.com/ams/annual2003/techprogram/programexpanded_148.htm. Accessed 15 January 2008
Tolman H L (1999) User Manual and System Documentation of WAVEWATCH-III Version 1.18.
NOAA/NWS/NCEP/OMB Technical Note 166, National Weather Service, NOAA, U.S.
Department of Commerce
Tozer N, Saulter A, Hawkes P (2007) Coastal Flood Forecasting. 42nd Defra Flood and Coastal
Management Conference, York, England
Verlaan M, Zijderveld A, Vries H, Kroos J (2005) Operational storm surge forecasting in the
Netherlands: developments in the last decade. Phil. Trans. R. Soc. Lond. 363(1831): 1441–1453
Wedge D, Ingram D, Mingham C, McLean D, Bandar Z (2005) On Neural Network Architectures
and Wave Overtopping. Proc. ICE, Maritime Engineering, 158(MA3), 123–133
World Meteorological Organisation (WMO) (1998) Guide to Wave Analysis and Forecasting.
2nd edition. WMO No. 702, Geneva, Switzerland
Chapter 8 Selected Applications
ACTIF (2004) Some Research Needs for River Flood Forecasting in FP6. Achieving Technological
Innovation in Flood Forecasting. European Commission Project EVK1-CT-2002-80014.
http://www.actif-ec.net/documents/ACTIFResearchNeedsfor%20FP6V1.4.pdf. Accessed 15
January 2008
Bhattacharya B, Lobbrecht A H, Solomatine D P (2003) Neural Networks and Reinforcement
Learning in Control of Water Systems. J. Water Resour. Plann. Manage. 129(6): 458–465
Bell V A, Moore R J, Brown V (2000) Snowmelt forecasting for flood warning in upland Britain.
In Lees M, Walsh P (Eds.), Flood Forecasting: What does Current Research Offer the
Practitioner. BHS Occasional Paper 12, British Hydrological Society, London
Blackburn J, Hicks F (2003) Suitability of dynamic modeling for flood forecasting during ice jam
release surge events. J. Cold Reg. Engrg. 17(1): 18–36
Bowles D S, Mathias J D, Chauhan S S, Countryman J D (2004) Reservoir Release Forecast
Model for Flood Operation of the Folsom Project include Pre-releases. Proceedings of the
2004 USSD Annual Lecture, March 2004, St. Louis, MO. http://uwrl.usu.edu/www/faculty/
DSB/USSD2004.pdf. Accessed 15 January 2008
Castro D, Einfalt T, Hatzfeld F, Wagner A (2006) URBAS: Prediction and Management of Flash
Floods in Urban Areas. International Workshop on Flash Floods in Urban Areas and Risk
Management, Oman. http://www.rcuwm.org.ir/events/workshop/8/articles.html. Accessed 15
January 2008
Cluckie I D, Lane A, Griffith R J (1998) Real Time Control of Urban Drainage Systems (UDS)
Using Weather Radar. Radar Hydrology for Real Time Flood Forecasting. Proceedings of an
advanced study course, University of Bristol, 24 June–3 July 1998, pp. 305–321
Collischonn W, Allasia D G, Silva B C, Tucci C E M (2007a) The MGB-IPH model for large scale
rainfall runoff modeling. Hydrol. Sci. J. 52(5): 878–895
Collischonn W, Tucci C E M, Clarke R T, Chou S C, Guilhon L G, Cataldi M, Allasia D (2007b)
Medium-range reservoir inflow predictions based on quantitative precipitation forecasts.
J. Hydrol. 344: 112–122
Daly S F (2003) A State–Space Model for River Ice Forecasting. Cold Regions Research and
Engineering Laboratory Report ERDC/CRREL TR-03-9, New Hampshire. http://www.
crrel.usace.army.mil/techpub/CRREL_Reports/html_files/Cat_D.html. Accessed 15
January 2008
Dunn S M, Colohan R J E (1999) Developing the snowmelt component of a distributed hydro-
logical model: a stepwise approach based on multi objective analysis. J. Hydrol. 223(1–2):
1–16
References 291
Erikkson M (2006) Flash Floods in the Himalaya – Challenges and Opportunities for Regional
Cooperation. Presentation at the First International Flash Flood Forecasting Workshop, World
Meteorological Organisation, March 13–17 2006, Costa Rica, CA. http://www.nws.noaa.gov/
iao/iao_FFW.php. Accessed 15 January 2008
Garcia R R, Sáez A L, Salete E G, García L L (2005) Reservoir Management in Real Time Flood
Forecasting and Decision Support in the Ebro river basin. ACTIF International Conference on
Innovation Advances and Implementation of Flood Forecasting Technology, 17–19 October
2005, Tromsø, Norway. http://www.actif-ec.net/conference2005/proceedings/index.html.
Accessed 15 January 2008
Georgakakos K P (2005) Modern Operational Flash Flood Warning Systems Based on Flash
Flood Guidance Theory: Performance Evaluation. ACTIF International Conference on
Innovation Advances and Implementation of Flood Forecasting Technology, 17–19 October
2005, Tromsø, Norway. http://www.actif-ec.net/conference2005/proceedings/index.html.
Accessed 15 January 2008
Göppert H, Ihringer J, Plate E J, Morgenschweis G (1998) Flood forecast models for improved
reservoir management in the Lenne River catchment, Germany. Hydrol. Sci. J. 43(2):
215–242
Graham W J (2000) Floods caused by dam failure. In Parker D J (Ed.), Floods. Routledge,
London
Guo S, Zhang H, Chen H, Peng D, Liu P, Pang B (2004) A reservoir flood forecasting and control
system for China. Hydrol. Sci. J. 49(6): 959–972
Henry E, Klepiszewski K, Fiorelli D, Solvi A-M, Weidenhaupt A (2005) Modelling of a Sewage
Network: Contribution to the Management of Pollution Risks at the Haute-Sûre Drinking
Water Reservoir. 10th International Conference on Urban Drainage, Copenhagen, Denmark,
21–26 August 2005
Hock R (2003) Temperature index melt modelling in mountain areas. J. Hydrol. 282(1–4):
104–115
Hom V T, Capone E J, Econopouly T W (2004) Forecasting Ice Affected Rivers at the Northeast
River Forecast Centre. 61st Eastern Snow Conference, Portland, Maine. http://www.easterns-
now.org/proceedings/2004/hom_et_al.pdf. Accessed 15 January 2008
Howard C D D (2004) Ensemble Optimization for Hydroelectric Operations. Workshop on the
Hydrological Ensemble Prediction Experiment (HEPEX), ECMWF, 8–10 March 2004. http://
ecmwf.int/newsevents/meetings/workshops/2004/HEPEX/presentations.html. Accessed 15
January 2008
Howard, C D D (2007) Hydroelectric Operations – Ensemble Optimisation Procedures. Workshop on
the Hydrological Ensemble Prediction Experiment (HEPEX), Stresa, Italy, 27–29 June, 2007
Huband M, Sene K J (2005) Integrated Catchment Modelling Issues for Flow Forecasting
Applications. Scottish Hydraulics Study Group, Catchment Modelling for Flood Risk
Management, 18 March 2005
IMPACT (2005) Investigation of Extreme Flood Processes and Uncertainty. Final Technical
Report. http://www.samui.co.uk/impact-project/general_publications.htm. Accessed 15
January 2008
Koren V, Schaake J, Mitchell K, Duan Q Y, Chen F, Baker J M (1999) A parameterization of
snowpack and frozen ground intended for NCEP weather and climate models. J. Geophys.
Res. 104(D16): 19569–19585
Kubat I, Sayed M, Savage S, Carrieres T (2005) Implementation and Testing of a Thickness
Redistribution Model for Operational Ice Forecasting. Proceedings 18th International
Conference on Port and Ocean Engineering under Arctic Conditions, POAC’05. Volume 2, pp.
781–791, Potsdam. ftp://ftp2.chc.nrc.ca/CRTreports/CIS/POAC_05_Forecasting.pdf. Accessed
15 January 2008
Kuchment L S, Gelfan A N (2005) Long Term Ensemble Forecasting Spring Runoff Volume and
Hydrograph Peak. ACTIF International Conference on Innovation Advances and Implementation
of Flood Forecasting Technology, 17–19 October 2005, Tromsø, Norway. http://www.actif-
ec.net/conference2005/proceedings/index.html. Accessed 15 January 2008
292 References
Lobbrecht A H, Dibike Y B, Solomatine D P (2005) Networks and Fuzzy Systems in Model Based
Control of the Overwaard Polder. J. Water Resour. Plan. Manage. 131(2): 135–145
Mahabir C, Hicks F E, Robichaud C, Fayek A R (2006) Forecasting breakup water levels at Fort
McMurray, Alberta, using multiple linear regression. Can. J. Civ. Eng. 33(9): 1227–1238
Meon G (2006) Past and Present Challenges in Flash Flood Forecasting. WMO International
Workshop on Flash Flood Forecasting, 13–17 March 2006, Costa Rica, CA. http://www.nws.
noaa.gov/iao/iao_FFW.php. Accessed 15 January 2008
Mirtskhoulava Ts E (2000) Mud floods and their prediction in the Caucasus and Central Asia. In
Parker D J (Ed.), Floods. Routledge, London
Morse B, Hicks F (2005) Advances in river ice hydrology 1999–2003. Hydrol. Process. 19(1):
247–263
Nandalal K D W, Bogardi J J (2007) Dynamic Programming Based Operation of Reservoirs
Applicability and Limits. Cambridge University Press, Cambridge
NOAA-USGS (2005) NOAA-USGS Debris-Flow Warning System. Final Report: U.S. Geological
Survey Circular 1283. http://pubs.usgs.gov/circ/2005/1283/. Accessed 15 January 2008
Rickenmann D, Chen C L (2003) Debris-Flow Hazards Mitigation: Mechanics, Prediction, and
Assessment. Millpress, Rotterdam
Rutter N, Essery R (2006) Evaluation of Forest Snow Process Models (SnowMIP2). Poster pre-
sentation, European Geosciences Union, Vienna, April 2006
Satake K (2000) Characteristics, causes and predictability of tsunamis. In Parker D J (Ed.), Floods.
Routledge, London
Schilling W (Ed.) (1989) Real Time Control of Urban Drainage Systems. The State of the Art.
IAWPRC Task Group on Real Time Control of Urban Drainage Systems, London
Schütze M, Campisano A, Colas H, Schilling W, Vanrolleghem P A (2004) Real time control of
urban wastewater systems—where do we stand today? J. Hydrol. 299: 335–348
Snorrason Á, Björnsson H, Jóhannesson H (2000) Causes, characteristics and predictability of
floods in regions with cold climates. In Parker D J (Ed.), Floods. Routledge, London
Sperfslage J A, Georgakakos K P, Carpenter T M, Shamir E, Graham N E (2005) Design and
Implementation of a Real-Time Regional Flash Flood Guidance System for Central America
Driven by Adjusted GOES Satellite Rainfall. 21st International Conference on Interactive
Information Processing Systems (IIPS) for Meteorology, Oceanography, and Hydrology.
http://ams.confex.com/ams/Annual2005/techprogram/session_17767.htm. Accessed 15
January 2008
Todini E, Codeluppi R (1998) Operational Use of Real-Time Flood Forecasting; the Case of Lake
Como. EWC98 International Conference on Early Warning Systems for Natural Disaster
Reduction, Potsdam
Vieux B E, Bedient P E, Mazroi E (2005) Real-Time Urban Runoff Simulation Using Radar
Rainfall and Physics-based Distributed Modeling for Site-Specific Forecasts. 10th International
Conference on Urban Drainage, 21–26 August 2005, Copenhagen/Denmark
Villars M, Gerritsen H, Heynert K, Kleiss B, Koster K (2007) A Demonstrator for Management
of the St Petersburg Flood Protection Barrier. 32nd Congress of the International Association
for Hydraulic Research, Venice, Italy
Vitasovic Z C (2006) Real Time Control of Urban Drainage Networks. U.S. Environmental
Protection Agency (EPA) Report EPA/600/R-06/120, Office of Research and Development,
Washington, DC. http://www.epa.gov/nrmrl/pubs/600r06120/600r06120.pdf. Accessed 15
January 2008
White K D (2003) Review of prediction methods for breakup ice jams. Can. J. Civ. Eng. 30:
89–100
World Meteorological Organisation (1982) Flash Flood Forecasting. Operational Hydrology
Report No. 18, WMO No. 577, Geneva
World Meteorological Organisation (1994) Guide to Hydrological Practices. 5th Edition. WMO
No. 168, Geneva
World Meteorological Organisation (2007) Prospectus for Implementation of a Flash Flood
Guidance System with Global Coverage. Geneva
References 293
Chapter 9 Preparedness
Direction de la Défense et de la Sécurité Civiles (2006) Guide ORSEC departemental methode
generale Organisation de la Réponse de Securité Civile. Tome G.1. http://www.cedre.fr/fr/
lutte/guide-elu/guide-orsec06.pdf. Accessed 15 January 2008
Emergency Management Australia (1999) Guide 4 – Flood Preparedness. Australian Emergency
Manuals Series, Part III Emergency Management Practice. Volume 3 - Guidelines. http://www.
ema.gov.au/. Accessed 15 January 2008
Environment Agency (2007) Risk Assessment for Flood Incident Management: Framework and
Tools. Joint Defra/Environment Agency Flood and Coastal Erosion Risk Management R&D
Programme. R&D Technical Report SC050028/SR1
Federal Aviation Authority (2000) FAA System Safety Handbook. Chapter 9: Analysis
Techniques. FAA, Washington, DC. http://www.faa.gov/library/manuals/aviation/risk_
management/ss_handbook/. Accessed 15 January 2008
FEMA (1997) Your Family Disaster Plan, Your Family Disaster Supplies Kit, Food and
Water in an Emergency, and Emergency Preparedness Checklist. Federal Emergency
Management Agency, Washington, DC. http://www.fema.gov/plan/index.shtm. Accessed
15 January 2008
FEMA (2005) Reducing Damage from Localized Flooding: A Guide for Communities. Chapter
6: Warning and Emergency Services, Report 511–07. http://www.fema.gov/hazard/flood/pubs/
flood-damage.shtm. Accessed 15 January 2008
FEMA (2007) National Incident Management System Incident Resource Inventory System
(NIMS-IRIS). Factsheet NIMS RM03, 202–646–3850
Gyorfi J S, Buhrke E R, Tarlton M A, Lopez J M, Valliath G T (2006) VICC: Virtual Incident
Command Center. Presence 2006
Handmer J (2002) Flood warning reviews in North America and Europe: statements and silence.
Aust. J Emerg. Manage. 17(3): 17–24. http://www.ema.gov.au/agd/EMA/emaInternet.nsf/
Page/Publications. Accessed 15 January 2008
Holland G (Ed.) (2007) Global Guide to Tropical Cyclone Forecasting. Bureau of Meteorology
Research Centre (Australia) WMO/TC-No. 560, Report No. TCP-31, World Meteorological
Organization. Geneva, Switzerland. http://www.bom.gov.au/bmrc/pubs/tcguide/globa_guide_
intro.htm. Accessed 15 January 2008
HM Government (2005) Emergency Preparedness Guidance on Part 1 of the Civil Contingencies Act
2004, Its Associated Regulations and Non-statutory Arrangements. Emergency Planning College
ISDR (2006) Developing Early Warning Systems: A Checklist. EWC III Third International
Conference on Early Warning from Concept to Action, 27–29 March 2006, Bonn, Germany.
UN/ISDR Platform for the Promotion of Early Warning. http://www.unisdr.org/ppew/.
Accessed 15 January 2008
Ishigaki T, Toda K, Baba Y, Nakagawa H, Shimada H (2006) Difficulty of Evacuation from
Underground Space in Urban Flood. 7th International Conference on Hydroinformatics, HIC
2006, Nice, France
MacFarlane R (2005) A Guide to GIS Applications in Integrated Emergency Management.
Emergency Planning College, Cabinet Office
NOAA/NWS (1997) Automated Local Flood Warning Systems Handbook, Weather Service
Hydrology Handbook No. 2. National Oceanic and Atmospheric Administration, National
Weather Service, Office of Hydrology, Washington, DC. http://www.weather.gov/oh/docs/
alfws-handbook/. Accessed 15 January 2008
Pajorova E, Hluchy E, Astalos J (2007) 3D Geovisualization Service for Grid-Oriented
Applications of Natural Disasters. Cracow 07 Grid Workshop, Krakow, Poland
Smith K, Ward R (1998) Floods: Physical Processes and Human Impacts. Wiley, Chichester, UK
United Nations (2006) Guidelines for Reducing Flood Losses. A Contribution to the International
Strategy for Disaster Reduction. United Nations. http://www.un.org/esa/sustdev/publications/
flood_guidelines.pdf. Accessed 15 January 2008
294 References
U.S. Army Corps of Engineers (USACE) (1995) Technical Guidelines for Hurricane Evacuation
Studies. September 1995
U.S. Army Corps of Engineers (USACE) (1996) Hydrologic Aspects of Flood Warning-
Preparedness Programs. Report ETL 1110–2–540. Office of Chief of Engineers, Washington,
DC. http://www.tpub.com/content/USACETechnicalletters/ETL-1110–2–540/index.htm.
Accessed 15 January 2008
U.S. House of Representatives (2006) A Failure of Initiative. Final Report of the Select Bipartisan
Committee to Investigate the Preparation for and Response to Hurricane Katrina. US
Government Printing Office, Washington, DC. http://www.gpoaccess.gov/katrinareport/main-
report.pdf. Accessed 15 January 2008
Van Oosterom P, Zlatanova S, Fendel E M (Eds.) (2005) Geo-information for Disaster
Management. Springer, Berlin
Wolshon B, Urbina E, Levitan M (2001) National Review of Hurricane Evacuation Plans and
Policies. LSU Hurricane Centre, Baton Rouge, LA
Wolshon B, Urbina E, Wilmot C, Levitan M (2005a) Review of policies and practices for hurri-
cane evacuation. I: transportation planning, preparedness, and response. Nat. Hazards Rev.
6(3): 129–142
Wolshon B, Urbina E, Levitan M, Wilmot C (2005b) Review of policies and practices for hurricane
evacuation. II: traffic operations, management, and control. Nat. Hazards Rev. 6(3): 143–161
World Meteorological Organisation (2006a) Preventing and Mitigating Natural Disasters. WMO
No. 993, Geneva
World Meteorological Organisation (2006b) Social Aspects and Stakeholder Involvement in
Integrated Flood Management. WMO No. 1008, Geneva
Chapter 10 Response
Collier C G, Cross R, Khatibi R, Levizzani V, Solheim I, Todini E (2005) ACTIF Best Practice
Paper – The Requirements of Flood Forecasters for the Preparation of Specific Types of
Warnings. ACTIF International Conference on Innovation Advances and Implementation of
Flood Forecasting Technology, 17–19 October 2005, Tromsø, Norway. http://www.actif-
ec.net/conference2005/proceedings/index.html. Accessed 15 January 2008
Costes M, Abadie J P, Ducuing J L, Denier J S (2002) Tangible Results and Progress in Flood
Risks Management with the PACTES Initiative. 34th COSPAR Scientific Assembly, The
Second World Space Congress, 10–19 October 2002, Houston, TX
Demeritt, D, Cloke H, Pappenberger F, Thielen J, Bartholmes J, Ramos M (2007) Ensemble pre-
dictions and perceptions of risk, uncertainty, and error in flood forecasting. Environ. Hazards
7: 115–127
Demuth J., Morss R E, Lazo J K, Stewart A E (2007) Assessing How the US Public Understands
and Uses Weather Forecast Uncertainty Information. http://ams.confex.com/ams/pdfpapers/119820.
pdf. Accessed 15 January 2008
Emergency Management Australia (1999) Guide 5 – Flood Warning. Australian Emergency
Manuals Series, Part III Emergency Management Practice. Volume 3, 2nd edition. http://www.
ema.gov.au/. Accessed 15 January 2008
Environment Agency (2007) Probabilistic Flood Forecasting Scoping Study. Joint Defra/EA Flood
and Coastal Erosion Risk Management R&D Programme, R&D Technical Report FD2901/TR
Erlich M (2007) OSIRIS - an example of citizen-oriented technology development in the area of
dissemination of information on flood risk management. In Begum S, Stive M J F, Hall J W
(Eds.), Flood Risk Management in Europe. Innovation in Policy and Practice, Springer,
Dordrecht, The Netherlands
Flikweert J J, Coremans C, De Gooijer K, Wentholt L (2007) Automation of flood contingency
plans: benefits and implementation experiences. In Begum S, Stive M J F, Hall J W (Eds.),
References 295
Flood Risk Management in Europe. Innovation in Policy and Practice, Springer, Dordrecht,
The Netherlands
Fu H (2004) Development of Dynamic Travel Demand Models for Hurricane Evacuation. Ph.D.
dissertation, Louisiana State University and Agriculture and Mechanical College
Glassheim E (1997) Fear and loathing in North Dakota. Natural Hazards Observer XXI(6): 1–4
Huaimin G (2005) Flood Forecasting and Management in China. ACTIF International Conference
on Innovation Advances and Implementation of Flood Forecasting Technology, 17–19 October
2005, Tromsø, Norway. http://www.actif-ec.net/conference2005/proceedings/index.html.
Accessed 15 January 2008
ISDR (2006) Guidelines for reducing flood losses. International Strategy for Disaster Reduction.
http://www.unisdr.org/eng/library/isdr-publication/flood-guidelines/isdr-publication-floods.
htm. Accessed 15 January 2008
Krzysztofowicz R (2001) The case for probabilistic forecasting in hydrology. J. Hydrol. 249: 2–9
Langkamp E J, Wentholt L R, Pengel B E, Gooijer C, de, Flikweert J J (2005) NOAH, the right
information at the right time at the right place. Proceedings of Third International Flood
Defence Symposium, Nijmegen, The Netherlands
MacFarlane R (2005). A Guide to GIS Applications in Integrated Emergency Management.
Emergency Planning College, Cabinet Office
Martini F, De Roo A (Eds.) (2007) EXCIFF Guide: Good Practice for Delivering Flood Related
Information to the General Public. European Commission/Joint Research Centre report
EUR22760EN. http://exciff.jrc.it/downloads/exciff-related-documents/EXCIFF_guide.pdf.
Accessed 15 January 2008
National Research Council (2003) Communicating Uncertainties in Weather and Climate
Information: A Workshop Summary. National Academies Press, Washington. DC http://www.
nap.edu/catalog.php?record_id = 10597. Accessed 15 January 2008
National Research Council (2006) Completing the Forecast: Characterizing and Communicating
Uncertainty for Better Decisions Using Weather and Climate Forecasts. National Academies Press,
Washington, DC. http://www.nap.edu/catalog.php?record_id = 11699. Accessed 15 January
North Cornwall District Council (2004) Boscastle The Flood 16–08–04
Pierce C, Bowler N, Seed A, Jones D, Moore R (2005) Towards Stochastic Fluvial Flood
Forecasting: Quantification of Uncertainty in Very Short Range QPF’s and Its Propagation
Through Hydrological and Decision Making Models. Second ACTIF Workshop on
Quantification, Reduction and Dissemination of Uncertainty in Flood Forecasting, Delft,
23–24 November 2004. http://www.actif-ec.net/Workshop2/ACTIF_WS2_Session1-cont.
html. Accessed 15 January 2008
Prastacos P, Castaings W, Courtois N, Hluchy L, Houdry P, Tran V (2004) ANFAS: a decision
support system for flood risk assessment. In Prastacos P, Cortes U, de Leon J L D, Murillo M
(Eds.), e-Environment: Progress and Challenges. Instituto Politecnico Nacional, Mexico
Prendergast S (2007) London’s Command Support System. Fire Magazine, July 2007: 26–29
PREVIEW (2007) User Link: Special Issue on Floods. PREVIEW and RISKEOS project newsletter
Issue No. 3, http://www.preview-risk.com and http://www.risk-eos.com. Accessed 15 January 2008
Regnier E, Harr P A (2006) A dynamic decision model applied to hurricane landfall. Weather
Forecast 21(5): 764–780
Ritzen M R (2005) The High Water Information System and Flood Management in the
Netherlands. ACTIF International Conference on Innovation Advances and Implementation of
Flood Forecasting Technology, 17–19 October 2005, Tromsø, Norway. http://www.actif-
ec.net/conference2005/proceedings/index.html. Accessed 15 January 2008
Rodrigues A S, Santos M A, Santos A D, Rocha F (2006) Dam-break flood emergency manage-
ment system. Water Resour. Manage. 16(6): 1573–1650
Roulin E (2007) Skill and relative economic value of medium-range hydrological ensemble pre-
dictions. Hydrol. Earth Syst. Sci. 11: 725–737. http://www.hydrol-earth-syst-sci.
net/11/725/2007/hess-11-725-2007.html. Accessed 15 January 2008
Simonovic S P, Ahmad S (2005) Computer-based model for flood evacuation emergency plan-
ning. Nat. Hazards 34: 25–51
296 References
Van der Vaat M, Erlich M, Lumbroso D, Mens M (2007) Methodology for Decision Support
Systems for Flood Event Management. Floodsite Report T19–07–02. http://www.floodsite.
net/. Accessed 15 January 2008
Wolshon B, Urbina E, Wilmot C, Levitan M (2005) Review of policies and practices for hurricane
evacuation. I: transportation planning, preparedness, and response. Nat. Hazards Rev. 6(3):
129–142
World Meteorological Organisation (1994) Guide to Hydrological Practices. 5th edition. WMO
No. 168, Geneva
Zhou H, He B, Zhu L, Liang G (2004) The Application of Web GIS in Emergency Flood Control
System. IAHS Publication, 289, ICGRHWF Conference, Three Gorges Dam, China,
September 2003
Chapter 11 Review
Andryszewski A, Evans K, Haggett C, Mitchell B, Whitfield D, Harrison T (2005) Levels of
Service Approach to Flood Forecasting and Warning. ACTIF International Conference on
Innovation Advances and Implementation of Flood Forecasting Technology, 17–19 October
2005, Tromsø, Norway. http://www.actif-ec.net/conference2005/proceedings/index.html.
Accessed 15 January 2008
Andryszewski A, Evans K, Haggett C, Mitchell B, Whitfield D (2005) Flood Warning – The Next
Step Change. 40th Defra Flood and Coastal Management conference, York, England
Ash J, Postle M, Fenn T, Crossman M (2005) Evaluating a Multi-Criteria Analysis Methodology
for Application to Flood Management and Coastal Defence Appraisals. 40th Defra/
Environment Agency Flood and Coastal Management Conference, July 2005, York, England
Barnes L R, Gruntfest E C, Hayden M H, Schultz D M, Benight C (2007) False alarms and close
calls: a conceptual model of warning accuracy. Weather Forecast. 22(5): 1140–1147
Basher R (2006) Global early warning systems for natural hazards: systematic and people-centred.
Phil. Trans. R. Soc. A 364: 2167–2182. doi:10.1098/rsta.2006.1819
Betts R (2003) The missing links in community warning systems: findings from two Victorian
community warning system projects. Aust. J. Emerg. Manage. 18(3): 37–45
Carsell K M, Pingel N D, Ford D T (2004) Quantifying the benefit of a flood warning system. Nat.
Hazards Rev., ASCE, August 2004: 131–140
CNS Scientific and Engineering Services (1991) The Benefit Cost of Hydrometric Data – River
Flow Gauging. Report prepared for Department of Environment, Reading
Defra (2004) Making space for water. Developing a new Government strategy for flood and
coastal erosion risk management in England. Flood Warning and Forecasting. One of a series
of background papers and technical documents to accompany the consultation exercise. Flood
Management Division, July 2004
Defra (2005) Evaluating a Multi-Criteria Analysis (MCA) methodology for Application to Flood
Management and Coastal Defence Appraisals. Joint Defra/Environment Agency Flood and
Coastal Erosion Risk Management R&D Programme, R&D Technical Report FD2013
Dixon R, Boissonnade A, Krebs T, O’Shay A (2006) Event Response Activities for the Near-Real
Time Assessment of Financial Losses in Landfalling Hurricanes. AMS 27th Conference on
Hurricanes and Tropical Meteorology. http://ams.confex.com/ams/27Hurricanes/techprogram/
programexpanded_339.htm. Accessed 15 January 2008
Drabek T E (2000) The social factors that constrain human responses to flood warnings. In Parker
D J (Ed.), Floods. Routledge, London
Elliot J F, Stewart B J (2000) Early warning for flood hazards. In Parker D J (Ed.), Floods.
Routledge, London
Elliott J, Handmer J, Keys C, Tarrant M (2003) Improving Flood Warning – Which Way Forward ?
Paper presented at the Australian Disaster Conference, Canberra. http://www.ses.nsw.gov.au/
infopages/2337.html. Accessed 15 January 2008
References 297
Emergency Management Australia (1999) Guide 5 – Flood Warning. Australian Emergency
Manuals Series, Part III Emergency Management Practice. Volume 3, 2nd edition. http://www.
ema.gov.au/. Accessed 15 January 2008
Environment Agency (2002) Fluvial Flood Forecasting for Flood Warning - Real Time Modelling.
Defra/Environment Agency Flood and Coastal Defence R&D Programme. R&D Technical
Report W5C-013/5/TR
Environment Agency (2003) Flood risks to people Phase 1. Defra/Environment Agency Flood and
Coastal Defence R&D Programme, R&D Technical Report FD2317/TR1
Environment Agency (2004) Best Practice in Coastal Flood Forecasting. Defra/Environment
Agency Flood and Coastal Defence R&D Programme, R&D Technical Report
FD2206/TR1
Environment Agency (2007) Risk assessment for flood incident management. Joint Defra/
Environment Agency Flood and Coastal Erosion Risk Management R&D Programme. R&D
Technical Report SC050028/SR1
Handmer J, Henson R, Sneeringer P, Konieczny R, Madej P (2001) Warning systems for flash
floods: research needs, opportunities and trends. In Gruntfest E, Handmer J (Eds.), Coping
with Flash Floods. Kluwer, Dordrecht, The Netherlands, pp. 77–89
Handmer J (2001) Are flood warnings futile? Risk communication in emergencies. Aust. J.
Disaster Trauma Stud. 2000(2), New Zealand
Henson R (2001) U.S. Flash flood warning dissemination via radio and television. In Gruntfest E,
Handmer J (Eds.), Coping with Flash Floods. Kluwer, Dordrecht, The Netherlands, pp.
243–252
ISDR (2006) Lessons for a Safer Future: Drawing on the Experience of the Indian Ocean tsunami
disaster. International Strategy for Disaster Reduction. http://www.unisdr.org/ppew/tsunami/
pdf/lesson-for-a-safer-future.pdf. Accessed 15 January 2008
Jonkman S N, Kelman I (2005) An analysis of the causes and circumstances of flood disaster
deaths. Disasters 29(1): 75–97
Krzysztofowicz R, Kelly K S, Long D (1994) Reliability of flood warning systems. ASCE J.
Water Res. Plan. Manage. 120(6): 906–926
Martini F, De Roo A (Eds.) (2007) EXCIFF Guide: Good Practice for Delivering Flood Related
Information to the General Public. European Commission/Joint Research Centre report
EUR22760EN. http://exciff.jrc.it/downloads/exciff-related-documents/EXCIFF_guide.pdf.
Accessed 15 January 2008
Merz B, Kreibich H, Thieken A, Schmidtke R (2004) Estimation uncertainty of direct monetary
damage to buildings. Nat. Hazards Earth Syst. Sci. 4: 153–163
National Hydrologic Warning Council (2002) Use and Benefits of the National Weather Service
River and Flood Forecasts. http://www.nws.noaa.gov/oh/ahps/AHPS%20Benefits.pdf.
Accessed 15 January 2008
Norouzi A, Ghavasieh A, Attari J (2007). Uncertainty Evaluation of Flood Warning System;
Reliability and Trade-off. 32nd Congress of IAHR, the International Association of Hydraulic
Engineering & Research, July 1–6 2007, Venice, Italy
Parker D J, Fordham M, Torterotot J P (1994) Real time hazard management: flood forecasting,
warning and response. In Penning-Rowsell E, Fordham M (Eds.), Floods Across Europe:
Hazard Assessment, Modeling and Management. Middlesex University Press, UK
Parker D J (2003) Designing Flood Forecasting, Warning and Response Systems from a Societal
Perspective. Paper presented at the International Conference on Alpine Meteorology and
Meso-Alpine Programme, 21 May 2003, Brig, Switzerland. http://www.map.meteoswiss.ch/
map-doc/icam2003/Presentation/10.1/Brig_document.pdf. Accessed 15 January 2008
Parker D, Tunstall S, Wilson T (2005) Socio-Economic Benefits of Flood Forecasting and
Warning. ACTIF International Conference on Innovation Advances and Implementation of
Flood Forecasting Technology, 17–19 October 2005, Tromsø, Norway. http://www.actif-
ec.net/conference2005/proceedings/index.html. Accessed 15 January 2008
Pfister N (2002) Community response to flood warnings: the case of an evacuation from Grafton,
March 2001. Aust. J. Emerg. Manage., Autumn 2002. http://www.ema.gov.au. Accessed 15
January 2008
298 References
Sene K J, Ainsworth E J, Graham G (2006) Risk Based Design of Telemetry Networks for Flood
Warning Applications. 41st Defra/Environment Agency Flood and Coastal Management
Conference, York, England
Simonovic S P, Ahmad S (2005) Computer-based model for flood evacuation emergency plan-
ning. Nat. Hazards 34: 25–51
Tilford K A, Sene K J, Khatibi R (2007) Flood forecasting model selection: a structured approach.
In Begum S, Stive M J F, Hall J W (Eds.), Flood Risk Management in Europe, Springer,
Dordrecht, The Netherlands
United Nations (2006a) Global Survey of Early Warning Systems. An Assessment of Capacities,
Gaps and Opportunities Towards Building a Global Early Warning System for All Natural
Hazards. United Nations, Geneva. http://www.unisdr.org/ppew/. Accessed 15 January 2008
United Nations (2006b) Guidelines for Reducing Flood Losses. International Strategy for Disaster
Reduction. http://www.unisdr.org/eng/library/isdr-publication/flood-guidelines/isdr-publica-
tion-floods.htm. Accessed 15 January 2008
U.S Army Corps of Engineers (USACE) (1994) Framework for Estimating National Economic
Development Benefits and Other Beneficial Effects of Flood Warning and Preparedness
Systems. Institute for Water Resources Report IWR-94–3, Alexandria, Virginia
U.S. Army Corps of Engineers (USACE) (1996) Hydrologic Aspects of Flood Warning-
Preparedness Programs. Report ETL 1110-2-540, Office of Chief of Engineers, Washington,
DC. http://www.tpub.com/content/USACETechnicalletters/ETL-1110-2-540/index.htm.
Accessed 15 January 2008
U.S. House of Representatives (2006) A Failure of Initiative. Final Report of the Select Bipartisan
Committee to Investigate the Preparation for and Response to Hurricane Katrina. US
Government Printing Office, Washington, DC. http://www.gpoaccess.gov/katrinareport/main-
report.pdf. Accessed 15 January 2008
World Meteorological Organisation (1973) Benefit and Cost Analysis of Hydrological Forecasts:
A State of the Art Report. Operational Hydrology Report No. 3, WMO No. 341, Geneva
World Meteorological Organisation (1989) Natural Disaster Reduction: How Meteorological and
Hydrological Services Can Help. WMO No. 722, Geneva
World Meteorological Organisation (1994) Guide to Hydrological Practices. 5th edition. WMO
No. 168, Geneva
World Meteorological Organisation (2007) Economic Aspects of Integrated Flood Management.
WMO/GWP Associated Programme on Flood Management. WMO No. 1010, Geneva
Glossary
ACTIF (2004) Some research needs for river flood forecasting in FP6. Achieving Technological
Innovation in Flood Forecasting. European Commission Project EVK1-CT-2002-80014.
http://www.actif-ec.net/documents/ACTIFResearchNeedsfor%20FP6V1.4.pdf. Accessed 15
January 2008
Beven K J (2001) Rainfall Runoff Modelling: The Primer. Wiley, Chichester, UK
Holland G (Ed.) (2007) Global Guide to Tropical Cyclone Forecasting. Bureau of Meteorology
Research Centre (Australia) WMO/TC-No. 560, Report No. TCP-31, World Meteorological
Organization, Geneva, Switzerland. http://www.bom.gov.au/bmrc/pubs/tcguide/globa_guide_
intro.htm. Accessed 15 January 2008
IDNDR (1992) Internationally agreed glossary of basic terms related to Disaster Management.
IDNDR 1990-2000. DHA-Geneva, December 1992
UNESCO/WMO (2007) International Glossary of Hydrology. International Hydrological
Programme. http://webworld.unesco.org/water/ihp/db/glossary/glu/aglu.htm. Accessed 15
January 2008
UN/ISDR (2004) Terminology: Basic terms of disaster risk reduction (http://www.unisdr.org/eng/library)
Index
AAction tables, 59, 73
ALERT systems, 45,75
Artificial neural networks, 54, 133, 141, 147,
169–171, 190, 194, 226, 239
Astronomical tides, 156, 157
Australia, 1, 82, 88, 102, 120, 158, 171,
217, 260
Automatic Weather Station, 24
BBangladesh, 4, 82, 102, 183
Bayesian Techniques, 54, 135, 169, 170,
225, 253
Brazil, 102, 194
Business Continuity Management, 218, 224
CCanada, 17, 194, 201, 252
Catchment rainfall estimation, 26
Central America, 83, 102, 183–184
China, 4, 32, 102, 194, 241
Coastal Forecasting
astronomical tides, 156–157
data based, 152, 169
hurricanes, 22, 149, 150, 151, 162,
163–165, 172
process based, 152, 156–169
surge, 66, 108, 120, 157–165, 169,
171, 172
transformation matrices, 171
tropical cyclones, see hurricaneswave overtopping, 167–169, 171, 172, 173
waves, 165–167, 171
Communication
public awareness, 79, 82, 213, 251
risk, 9–12, 259
uncertainty, 17, 86, 120–122, 244–248
warning messages, 84–87, 250, 256
Conceptual Models
flow routing, 145–146
rainfall runoff, 131, 137–139
Contingency planning, 72, 220–226
Contingency Table
Cost Loss, 245
forecasting model performance,
111–112, 122
threshold performance, 68
Control Rooms, 77–79
Control Structures
river, 142, 176, 190–196
Thames Barrier, 197–198
tidal barriers, 197–199
Cost Benefit Analysis, 261–265
Cost Loss
polder operations, 55–56
rainfall alarms, 54
reservoir operations, 122, 194
Utility Function, 122, 246
Cyclones. See Tropical cyclones
DDams. See Reservoirs
Data Assimilation
error prediction, 106–107
parameter updating, 107–108
state updating, 107–108, 153, 162, 186
Data Based Models
artificial neural networks, 141, 147,
169–171, 190
coastal, 169–171
flow routing, 146–147
ice forecasting, 190
rainfall runoff, 131, 139–141
transfer function, 140, 147
299
300 Index
Debris flows, 181, 205
Decision Support Systems, 122, 194,
237–244
Delta. See Estuaries
Denmark, 30
Detection
automatic weather station, 24
evaporation, 23
network design, 47–49
performance improvements, 254–255
rainfall, 24–36
remote sensing, 28–33
river levels and flows, 37–42
snow, 27–28, 33
stage-discharge relationship, 39–41
telemetry, 44–47
tidal levels, 37–39
wave monitoring, 42–43
weather radar, 28–32
Dikes. See Flood Defences
Disdrometers, 25
Dissemination
internet, 82–83
multimedia, 82–83
performance improvements, 256–257,
258–260
RANET, 85
techniques, 79–87
transient populations, 81, 83, 256
uncertainty, 17, 86, 120–122, 244–248
warning messages, 84–87, 250, 256
EEconomic Analysis
cost benefit, 261–265
flood warning benefits, 263–264
multi criteria analysis, 261, 265–266
stage-damage relationships, 264
Emergency Management Australia, 2, 4, 88,
209, 217, 250
Ensemble Forecasting
coastal, 120, 159, 164
forecast products, 117–119, 120–122
meteorological, 35, 36
multi-model forecasts, 35, 116
performance measures, 114
rainfall thresholds, 54–56
river models, 120, 128, 183, 193–194
snowmelt, 188
Environment Agency, 64, 76–77, 99–100, 120,
171–172, 197–198, 264–265
Estuaries, 63, 64, 150, 158, 161, 169, 171,
198–199
Evacuation
Decision Support Tools, 240
hurricanes, 216, 233, 240
FFalse alarms, 46, 67, 68, 69, 111, 233,
251–252, 255
Federal Emergency Management Agency
(FEMA), 213, 214, 215–217, 220,
228, 240
Finland, 82, 102
Flash Flood
Boscastle Event, 236–237
definitions, 181
Flash Flood Guidance, 53, 183–185
forecasting, 181–185
thresholds, 53–54., 56, 205
World Meteorological Organisation,
183–185
Flash Flood Guidance, 53, 183–185
Flood Defences
breach, 12, 203–204, 240
emergency works, 232, 241
overtopping, 167–169, 173
temporary barriers, 15
Flood Emergency Plans
General Principles, 71–73, 209–219
operational response, 59, 73
Table Top Exercises, 219–220
Validation and Testing, 73,
219–220
Flood Event Management
Preparatory Actions, 231–234
Flood Forecasting
data availability, 126–128
flow routing, 141–147, 190–199
ice, 188–190
integrated catchment models, 133,
175–180
model calibration, 108–112
model design, 93–97, 123–126,
149–153
performance improvements,
113–114, 257
rainfall runoff models, 132–141
simple forecasting techniques,
61–67
simple triggers, 61–67
surge, 157–165, 171–173
systems, 97–104
ungauged flows, 178–180
urban drainage, 199–202
waves, 165–169, 171–173
Index 301
Flood Risk
causes of flooding, 8–9
Flood Risk Assessment, 9–13, 217
Flood Warning Areas, 73–75
hydraulic modelling, 11–12
transient populations, 12
Flood Warning Procedures
Flood Warning Areas, 73–75
Flow Control, 14, 73, 190–202, 232
Flow Routing Models
conceptual, 145–146
data based, 146–147
process based, 142–145
Forecasting Points, 94–95, 124–126, 151
Forecasting Systems
data hierarchy, 104
France, 54, 82, 217, 241, 242
GGeographical Information Systems, 227–228,
237–239
geotechnical risks, 202–206
Germany, 82, 147, 183, 194, 201, 241, 242
Glacial Lake Outburst Floods, 84, 181,
202, 204
Groundwater flooding, 203
HHong Kong, 158, 205
Hurricanes, 9, 150, 162–165, 172,
215–217, 238
Hydraulic Models
coastal, 157–165
ice forecasting, 189
river modelling, 142–145
urban, 200–201
IIce
forecasting, 188–190
ice jams, 189
stage discharge relationships, 189
India, 85, 183
International Strategy for Disaster Reduction
(ISDR), 88, 212
Ireland, 241–242
Italy, 194
JJapan, 11, 32, 82
KKalman filter, 108, 120, 190
KNMI, 55–56, 158
LLead time
evacuation, 216, 233
flash flooding, 181, 232, 236
flood warning, 67–68, 86
forecast, 95–96, 181
forecasting models, 127, 171
targets, 251
telemetry network design, 48
thresholds, 51, 57, 65–66, 255
time delays, 57, 95–96
tropical cyclones, 216
Lessons Learned Reports, 250
Levees. See Flood Defences
Levels of service, 251–252, 265
Luxembourg, 201
MMesoscale, 34, 162
Meteorburst telemetry, 28, 44, 45
Multi Criteria Analysis, 261, 265–266
NNepal, 84, 183
Netherlands, 35, 55–56, 108, 158. 194, 199,
241, 242–244
Norway, 28, 102
Nowcasting, 35–36
Numerical Weather Prediction, 34–35, 162
PPerformance Monitoring
dissemination systems, 83
flood warning systems, 83, 249–253
forecasting models, 113–114
thresholds, 67–70, 255
Polders, 55–56, 192
Portugal, 171
Preparedness
All-Hazard Approaches, 217
Flood Emergency Plans, 209–219
Resilience, 220–226
Probabilistic
also. See Ensemble Forecasting
flood warnings, 74, 244–248
forecasts, 16–17, 114–122
302 Index
Probabilistic (cont.)Numerical Weather Prediction, 35
rainfall thresholds, 55–56
Risk Assessment, 9–13, 224–225
Process Based Models
coastal, 156–169
flow routing, 142–145
rainfall runoff, 131, 135–137
Proudman Oceanographic Laboratory,
158–160
Public awareness, 79, 82, 213, 251
QQuantitative Precipitation Forecast.
See Numerical Weather Prediction
RRainfall
alarms, 51–56
catchment rainfall estimation, 26–27
depth duration, 52
disdrometers, 25
forecasts, 33–36
microwave attenuation, 33
nowcasting, 35–36
raingauges, 24–25
satellite observations, 32–33
thresholds, 51–56
weather radar, 28–32
Rainfall Runoff Models
conceptual, 137–139
data based, 139–141
process based, 135–137
Rating curve. See stage discharge relationship
Real time updating. See Data Assimilation
Red River, 17, 247
Reservoirs
dam break, 203–204, 240
decision support systems, 193–194
flood forecasting, 190–194
probabilistic forecasting, 122, 193–194
Resilience
control rooms, 78
dissemination systems, 80, 88
flood warning systems, 220–226
forecasting systems, 100–101, 103–104
instrumentation, 38
telemetry networks, 46, 48, 104, 128
thresholds, 58, 62
Risk to life, 4, 10, 259, 262, 265
River Gauging Stations
level monitoring, 37–42
Russia, 32
SSatellite
altimetry, 38
rainfall measurements, 32–33
soil moisture measurement, 32–33
telemetry, 44, 45
wave monitoring, 43
SCADA, 46
SLOSH, 163–165
Snow
degree-day method, 186
monitoring, 27–28
snowmelt forecasting, 185–188
Somalia, 18
Spain, 194
Stage-discharge relationship, 39–41
STOWA, 242
Surge forecasting, 66, 108, 120, 157–165,
169, 171, 172
TTaiwan, 194
Telemetry
meteorburst, 28, 44, 45
networks, 48
radio, 44, 45
satellite, 44, 45
telephone, 44, 45
Thames Barrier, 197–198
Thresholds
alarm handling, 46
meteorological indicators, 54, 182
performance, 67–70, 255
rainfall, 51–56, 182
risk based, 54
river level, 56–61, 182
tidal level, 56–61
Tidal barriers, 197–199
Timeline, 234–237
Transfer function, 120, 139–141, 147
Transient populations, 10, 81, 83, 256
Triggers. See Thresholds
Tropical Cyclone Programme, 154–156
Tropical cyclones, 2, 4, 9, 88, 150, 151,
153–156, 162, 172, 210, 215, 216
Tsunami, 9, 205–206, 250
UUncertainty
Emergency Response, 244–248
forecasting models, 114–122
reservoir forecasting, 193–194
thresholds, 54–56, 58
Index 303
Ungauged catchment, 136, 178–180
United Kingdom, 30, 76–77, 87, 99–100, 114,
120, 158–160, 177, 197–198, 218–219,
252, 264–265
Urban drainage, 199–202
Urban Drainage and Flood Control District, 83
US Army Corps of Engineers, 2, 88, 209,
240, 263
US National Weather Service, 28, 30, 45, 53,
75, 82, 88, 102, 120, 158, 164, 172,
184, 188, 240
USA, 11, 28, 30, 45, 53, 82, 88, 102, 120, 158,
163, 171, 172, 194, 205, 213, 215–217,
240, 260
Utility function, 54, 122, 194, 246
VVirtual Emergency Operations Centres, 238
Visualisation and simulation, 228–229
Vulnerability, 10, 12, 214
WWaves
forecasting, 165–167
monitoring, 42–43
overtopping, 167–169
types, 150, 166
Weather radar, 28–32
World Meteorological Organisation, 5–7, 32,
45, 88, 134, 153–156, 183–185