applications of nwp and radar-based qpf techniques for flash flood and landslip warnings in hong...

Applications of NWP and Radar-Applications of NWP and Radar-based QPF Techniques for Flash based QPF Techniques for Flash

Flood and Landslip Warnings in Hong Flood and Landslip Warnings in Hong KongKong

International Workshop on Flash Flood ForecastingInternational Workshop on Flash Flood ForecastingSan Jose, Costa RicaSan Jose, Costa Rica 13 - 17 March 200613 - 17 March 2006

Edwin S.T. Lai Edwin S.T. Lai Hong Kong ObservatoryHong Kong Observatory

134A Nathan Road, Kowloon, Hong Kong134A Nathan Road, Kowloon, Hong Kong

Presentation OutlinePresentation Outline

• Rain-related Hazards in Hong KongRain-related Hazards in Hong Kong

• QPF in support of Warning SystemsQPF in support of Warning Systems

• Symbiotic Evolution of Forecast and Symbiotic Evolution of Forecast and Warning SystemsWarning Systems

Mechanisms leading to Mechanisms leading to Heavy rain over southern Heavy rain over southern ChinaChina• Rainstorms over southern China display Rainstorms over southern China display

full range of spatial and temporal full range of spatial and temporal characteristics characteristics – Monsoon trough – warm moist air from south Monsoon trough – warm moist air from south

merging with cool dry air from the north merging with cool dry air from the north (spring, summer, autumn)(spring, summer, autumn)

– Convergence between Indochina southwest Convergence between Indochina southwest monsoon and Pacific ridge (summer)monsoon and Pacific ridge (summer)

– Tropical cyclone nearbyTropical cyclone nearby– Land-sea coastal effectsLand-sea coastal effects

Computer Drainage Models and FlComputer Drainage Models and Flood Hazard Mapsood Hazard Maps

Sheung ShuiSheung Shui (Northern New Terri (Northern New Territories)tories), 2001, 2001

Wing Kei Tsuen, 2001Wing Kei Tsuen, 2001

Collection of Hydrological data By Collection of Hydrological data By DSD Wireless Flow Gauging StatioDSD Wireless Flow Gauging Stationn

Alert Level

Sensor

Warning to Local

Villagers

Telemetry

• Warnings will be issued when floodwater Warnings will be issued when floodwater reaches a predetermined alert level. reaches a predetermined alert level.

• Warning signals are disseminated automatically Warning signals are disseminated automatically through flood sirens or via automatic telephone through flood sirens or via automatic telephone calls to village representatives. calls to village representatives.

Village Flood Pumping Scheme Village Flood Pumping Scheme –– Conceptual LayoutConceptual Layout

Shenzhen River Regulation Shenzhen River Regulation ProjectProject

Tsuen Wan, 2001Tsuen Wan, 2001

Mongkok, 1997Mongkok, 1997

Drainage TunnelsDrainage Tunnels

下游市區DownstreamUrban Area

中上游半山Upstream

Urban Area

填海區Reclaimed Area

Intercepting tunnel

Tai Hang Tung Stormwater Storage Tai Hang Tung Stormwater Storage TankTank

Capacity = Capacity = 100,000 cu. m100,000 cu. m

Rainstorm WarningsRainstorm Warnings

• floodingflooding in some low-lying and poorly in some low-lying and poorly drained areasdrained areas

• Key Key GovernmentGovernment departments and departments and major major transporttransport and and utilityutility operators operators are put on are put on alertalert. .

• publicpublic will be given clear will be given clear adviceadvice on on the appropriate actions to take.the appropriate actions to take.

Special Announcement Special Announcement on Flooding in the on Flooding in the Northern New TerritoriesNorthern New Territories • criterioncriterion for issuance is based on heavy for issuance is based on heavy rainrain

having fallen in the northern New Territories. having fallen in the northern New Territories.

• Drainage Services DepartmentDrainage Services Department– clear blocked drains and watercourses in the clear blocked drains and watercourses in the

northern NTnorthern NT

• emergency response departmentsemergency response departments – evacuation and rescue evacuation and rescue – emergency treatmentemergency treatment– transport of casualties to hospitalstransport of casualties to hospitals

Hong Kong Observatory’s Hong Kong Observatory’s Nowcasting System - SWIRLSNowcasting System - SWIRLS

• SShort-range (0 - 3 hort-range (0 - 3 hours)hours)

• WWarning ofarning of• IIntense ( > 30 mm per ntense ( > 30 mm per

hour) hour) • RRainstorms inainstorms in• LLocalized ( 10 - 100 km)ocalized ( 10 - 100 km)• SSystemsystems

• Operational since 1998, Operational since 1998, originally for rainstorm, originally for rainstorm, but has evolved into a but has evolved into a multi-function systemmulti-function system

2/12

22

2

2

12

1

2121

- )( - )(

)()(1

- )()( R

kk

k k k

ZNkZZNkZ

kZkZN

kZkZ

TREC (Pattern Matching)TREC (Pattern Matching)

Searching radius

Pixel matrix

Maximum correlatedpixel matrix

TREC vector

T T – 6min

Searching radius

0.5, 1, 1.5, 2, … 5 km CAPPI

64, 128, 256 km range

where Z1 and Z2 are the reflectivity at T+0 and T+6min respectively

Searching radiusSearching radius ：： 55 ，， 1010 ，， 20 20 km km ((64, 128 and 256 km CAPPI64, 128 and 256 km CAPPI))Pixel matrix sizePixel matrix size ：： 5, 10, 20 km 5, 10, 20 km ((64, 128 and 256 km CAPPI64, 128 and 256 km CAPPI))Max resolvableMax resolvable ：： 50, 100, 200 km/h50, 100, 200 km/h ( (64, 128 and 256 km CAPPI64, 128 and 256 km CAPPI))ResolutionResolution ：： 1.3, 2.6, 5.2 km/h1.3, 2.6, 5.2 km/h ( (64, 128 and 256 km CAPPI64, 128 and 256 km CAPPI))

TREC fields in Weather TREC fields in Weather SystemsSystems

GEO + DSD + HKO rain gaugesGEO + DSD + HKO rain gauges

Total ~ 140, update frequency 5 minsMean distance x, y ~ 1.5 km

HKO rain gauge

HKO AWS

GEO rain gauge

DSD rain gauge

Direct Matching – Direct Matching – Dynamic Dynamic Z-RZ-R relation relation Z = aRZ = aRbb

An adaptive Z-R relation (updated every 5 min.)

dBZ = 10*log(a) + 10*b*log(R)

1 km radar reflectivityZ: 00, 06, 12, … min

140 rain gaugesR: 05, 10, 15, … min

20

25

30

35

40

45

50

55

60

5 7 9 11 13 15 17 19 21

dBG

dBZ

Searching radius

Adjustable data assimilation window

140

1

2)(mini

ii dBRdBG bdBRadBZ log10

Modified Semi-Lagrangian Modified Semi-Lagrangian Advection SchemeAdvection Scheme

• Robert scheme (3 iterations)Robert scheme (3 iterations)

• Bi-cubic Lagrangian interpolationBi-cubic Lagrangian interpolation

• Flux limiter (local max, min constraint)Flux limiter (local max, min constraint)

• One-way nestedOne-way nested– resolution 1.1km -> 0.5kmresolution 1.1km -> 0.5km

• Non-conservative, to allow:Non-conservative, to allow:– growth/decay (intensity profile easily to be added)growth/decay (intensity profile easily to be added)– echo entering, departing the domainecho entering, departing the domain

Forecast reflectivityForecast reflectivity

TREC wind Forecast reflectivityUp to 6 hr (6-min interval)

FCT hour 0hr 1hr 2hr 3hr 4hr 5hr 6hr

Max. ratio 1.00 0.98 0.97 0.96 0.96 0.96 0.94

Extrapolation of pure rotationExtrapolation of pure rotation

FCT hour 0hr 1hr 2hr 3hr 4hr 5hr 6hr

Max. ratio 1.00 0.98 0.98 0.97 0.97 0.97 0.96

Quantitative Precipitation ForecastQuantitative Precipitation Forecast

4 steps:4 steps:• Storm Tracking (radarStorm Tracking (radar reflectivityreflectivity))

– TREC winds (Tracking Radar Echoes by Correlations) TREC winds (Tracking Radar Echoes by Correlations) • Adaptive Adaptive Z-RZ-R relation relation

– Dynamically calibrate radar reflectivity using rain gauge data Dynamically calibrate radar reflectivity using rain gauge data • EExtrapolationxtrapolation

– linear: upwind advection or semi-Lagrangian advectionlinear: upwind advection or semi-Lagrangian advection• IntegrationIntegration

– Forecast 1, 2, up to 3 hr accumulated rainfallForecast 1, 2, up to 3 hr accumulated rainfall

TREC QPF (1 – 3 hours)TREC QPF (1 – 3 hours)

SWIRLS rainfall forecast for a rainstormSWIRLS rainfall forecast for a rainstorm

SWIRLS 1-hr forecast Actual rainfall (rain-gauge)

TREC winds

Actual rainfall (rain gauge) SWIRLS 1hr rainfall forecast

SWIRLS rainfall forecast for a TyphoonSWIRLS rainfall forecast for a Typhoon

20:00 HKT, 2 Sep 2003

20:12 HKT, 2 Sep 2003

SWIRLS Landslip Warning Forecast

GEO Studies → factors contributing to landslip : 24-hr accumulated rainfall vulnerable area

Based on 1982-1997 data

• Each grid has 4Each grid has 4 slope slope types and percentagestypes and percentages ：：– Soil-cut (Soil-cut ( 切削泥斜坡切削泥斜坡））– Fill (Fill ( 填土坡填土坡））– Rock-cut (Rock-cut ( 切削石斜坡切削石斜坡））– Wall (Wall ( 擋擋土墙土墙））

RF (min)RF (min) RF (max)RF (max) weightweight

00 0.10.1 00

0.50.5 5050 0.0000039750.000003975

5050 100100 0.0000107270.000010727

100100 150150 0.000028950.00002895

150150 200200 0.0000781290.000078129

200200 250250 0.0002108940.000210894

250250 300300 0.0005690250.000569025

300300 20002000 0.001535650.00153565

SWIRLS Landslip ForecastSWIRLS Landslip Forecast

• If fIf forecastorecast >= 15 landslip to occur -> issue Landslip Warning >= 15 landslip to occur -> issue Landslip Warning

• 24 accumulated rainfall = 24 accumulated rainfall = past 21 hr accumulated rainfall (actual) + future 3-hr rainfall past 21 hr accumulated rainfall (actual) + future 3-hr rainfall (forecast)(forecast)

= 21hr (actual) + 3hr (forecast)

SWIRLS QPF

No. of landslip occurrence as function of 24 hr accumulated rainfall and total vulnerable area

SWIRLS 3hr accumulated rainfall SWIRLS 3hr accumulated rainfall forecastforecast

Past 21hr rain gauge accumulated Past 21hr rain gauge accumulated rainfallrainfall

5 May 2003 – prolonged rainfall 5 May 2003 – prolonged rainfall brought by a low pressure troughbrought by a low pressure trough

Landslip Warning Criteria reached = 8:40 HKTSWIRLS forecast warning reached = 5:42 HKTLead time = 2:58 hr

Verification of SWIRLS Verification of SWIRLS WarningsWarnings

HazardsHazards

DisastersDisasters

HazardsHazards

DisastersDisasters

PoliticalPoliticalBacklashBacklash

PoliticalPoliticalBacklashBacklash

IdentificationIdentificationofof

StakeholdersStakeholders

IdentificationIdentificationofof

StakeholdersStakeholders

NMSNMS

Major Players(e.g. hydrologists)

Major Players(e.g. hydrologists)

Key Users(e.g. Disaster Managers)

Key Users(e.g. Disaster Managers)

Peripheral UsersPeripheral Users

Action PlanAction PlanAction PlanAction Plan

ResponseActions

ResponseActions

WarningSystem

WarningSystem

?

ForecastSystem

Development

ForecastSystem

Development

NMS Collaborationwith Major Players

and Key Users

NMS Collaborationwith Major Players

and Key Users

Something cannot be done and is not done.Disaster struck and something has to be done.Something was crudely done and remains crudely done.Until the next disaster struck when all were undone………

A Good Warning SystemA Good Warning System

does not end with the warning bulletins leaving the does not end with the warning bulletins leaving the forecast officeforecast office

offers a platform with agreed terminology for ready offers a platform with agreed terminology for ready exchange of information on expected scenariosexchange of information on expected scenarios

manages expectations of stakeholders and users, manages expectations of stakeholders and users, “downscaling” problems into something more “downscaling” problems into something more manageablemanageable

prompts prompt actions from decision-makers, prompts prompt actions from decision-makers, vulnerable groups and those at riskvulnerable groups and those at risk

provides incentive for further development of provides incentive for further development of forecasting toolsforecasting tools

From Action to Information EndFrom Action to Information End

Consolidate stakeholders’ requirementsConsolidate stakeholders’ requirements Design warning system that can address the Design warning system that can address the

key issueskey issues Assess technical feasibility and develop Assess technical feasibility and develop

necessary technical support for forecastingnecessary technical support for forecasting If technically not feasible, go back to If technically not feasible, go back to

stakeholders and re-work the process until a stakeholders and re-work the process until a level of balance and synergy is reached level of balance and synergy is reached between forecasting and warning systemsbetween forecasting and warning systems

Three Most Popular DemandsThree Most Popular Demands

Extended QPFExtended QPF Forecast of “Warnings”Forecast of “Warnings” Location-specific/time-specific QPFLocation-specific/time-specific QPF

RAPIDSRAPIDS(Rainstorm Analysis and Prediction Integrated Data-processing System)(Rainstorm Analysis and Prediction Integrated Data-processing System)

SWIRLS Amber Verification SWIRLS Amber Verification (May)(May)

• CorrectCorrect (73%) (73%)Amber: Amber: 5/11 (46%)5/11 (46%)No amber:No amber:3/11 (27%)3/11 (27%)

• WrongWrong (27%) (27%)FAR:FAR:3/11 (27%)3/11 (27%)NAP:NAP:0/11 (0%)0/11 (0%)

• Lead timeLead time::35 minutes35 minutes

SWIRLS Amber Verification SWIRLS Amber Verification (Jun)(Jun)• CorrectCorrect (55%) (55%)

Amber: Amber: 4/18 (22%)4/18 (22%)No amber:No amber:6/18 (33%)6/18 (33%)



SWIRLS Amber Verification SWIRLS Amber Verification (Jul)(Jul)• CorrectCorrect (86%) (86%)

Amber: Amber: 3/7 (43%)3/7 (43%)No amber:No amber:3/7 (43%)3/7 (43%)



SWIRLS Amber Verification SWIRLS Amber Verification (Aug)(Aug)

• CorrectCorrect (57%) (57%)Amber: Amber: 3/7 (43%)3/7 (43%)No amber:No amber:1/7 (14%)1/7 (14%)



7 types of rainstorm7 types of rainstorm

• Land-sea breeze/Weak Northerly Storms (LS)Land-sea breeze/Weak Northerly Storms (LS)• Quasi-stationary Southwesterly Rainbands (QU)Quasi-stationary Southwesterly Rainbands (QU)• Southesterly Rainbands (SE)Southesterly Rainbands (SE)• Squall-lines/Bow Echoes (SQ)Squall-lines/Bow Echoes (SQ)• Supercells (SU)Supercells (SU)• Tropical Cyclone Rainbands (TC)Tropical Cyclone Rainbands (TC)• X-type Rainbands (X)X-type Rainbands (X)

7 Rainstorm Types7 Rainstorm Types

LS

QU

SE

SQ

SU TC X

Data sourceData source

• 2001 – 2005 all rainstorms either triggered Amber, Red,2001 – 2005 all rainstorms either triggered Amber, Red, Black, Landslip or Flooding Warning Black, Landslip or Flooding Warning

• A total of 73 rainstorm casesA total of 73 rainstorm cases• 800 hours of radar images800 hours of radar images• 8000 TREC images; 8000 GTrack images8000 TREC images; 8000 GTrack images

Distribution (Occurrence)Distribution (Occurrence)7 rainstorm types during 2001-7 rainstorm types during 2001-2005)2005)

0

5

10

15

20

25

30

35

40

LS QU SE SQ SU TC X

Occ

urre

nce

• QU – 43%QU – 43%

• X – 18%X – 18%

• LS – 14%LS – 14%

• SQ – 10%SQ – 10%

• TC – 8%TC – 8%

• SE – 6%SE – 6%

• SU – 3%SU – 3%

GTrack Cell Features AnalysiGTrack Cell Features Analysiss

ellipse

AdbZI

ab

aEP

aba

abA

)/(tan

),2/(4

/

1

22

max dBZ <--> max rainfallave dBZ <--> ave rainfallmax50% rainfallmax25% rainfall (upper quartile)

x

ye.g. 128 km,3km CAPPI, 39dBZ threshold

a (major axis)

b (minor axis)

(orientation)

V

(TREC speed and direction)

Area

Eccentricity

Perimeter

Orientation

TotalIntensity

Sample Feature MatrixSample Feature MatrixTable 1 - Frequency distribution of various radar parameters extracted from the 3-km CAPPI 128-km range 39-dBZ threshold TREC/GTrack analyses for

different rainstorm types. On each column, scales on the x-axes are the same. Scales on the y-axes are 0 to 100%.

Parameter

(unit)

x-axis

scale

Type of

Rainstorm

TREC

Speed

(km/hr)

0 100

TREC

Direction

(bearing)

0 360

Eccentricity

0 1.0

Orientation

(bearing)

0 180

Area

(km2)

0 1000

APR

(km)

0 20

Upper-quartile

rainfall

(mm/hr)

20 200

Total

Intensity

(km2-dBZ)

104 105

LS

QU

SE

SQ

SU

Implications on Warning Implications on Warning StrategiesStrategies

ConclusionsConclusions

• Not All Rainstorms Are Created Equal!Not All Rainstorms Are Created Equal!– Different rainstorm carries different impactDifferent rainstorm carries different impact– Different rainstorm shows different feature matrixDifferent rainstorm shows different feature matrix

• Rainstorm databank for more systematic Rainstorm databank for more systematic studiesstudies

• Classification process can be automated for Classification process can be automated for objective analysisobjective analysis– Feature space (matrices)Feature space (matrices)– Cluster analysisCluster analysis– Artificial Neural NetworkArtificial Neural Network

• Conceptual Framework for Warning DecisionConceptual Framework for Warning Decision



– At this moment in time, we may consider such At this moment in time, we may consider such demands to be unrealistic. But then we re-wind demands to be unrealistic. But then we re-wind back to ten years ago when we thought it would back to ten years ago when we thought it would be impossible to do what we are routinely doing be impossible to do what we are routinely doing now, and we think again and say: “Maybe we now, and we think again and say: “Maybe we shall see …….”shall see …….”

Emerging Technology for Emerging Technology for FFFFFF

• Meteorologists’ angleMeteorologists’ angle– QPE (?)QPE (?)– QPF (??)QPF (??)– EPS (???)EPS (???)

• Hydrologists’ angle (????)Hydrologists’ angle (????)• IT angleIT angle

– Computer processing powerComputer processing power– Visualization and displayVisualization and display– IT methodology and AI algorithmsIT methodology and AI algorithms– Telecommunication and enhanced connectivityTelecommunication and enhanced connectivity


• Meteorologists’ angleMeteorologists’ angle– QPE (?)QPE (?)– QPF (??)QPF (??)– EPS (???)EPS (???)

• Hydrologists’ angle (????)Hydrologists’ angle (????)• IT angleIT angle

– Computer processing power Computer processing power – Visualization and displayVisualization and display– IT methodology and AI algorithmsIT methodology and AI algorithms– Telecommunication and enhanced connectivityTelecommunication and enhanced connectivity


• Meteorologists and hydrologists Meteorologists and hydrologists joining forces to attain 60-min joining forces to attain 60-min reliable warning lead time.reliable warning lead time.


• Meteorologists and hydrologists Meteorologists and hydrologists joining forces to attain 60-min joining forces to attain 60-min reliable warning lead time.reliable warning lead time.

• But only good IT utilization and But only good IT utilization and applications can make that 60 applications can make that 60 minutes productive and meaningful!minutes productive and meaningful!

Thank YouThank You

applications of nwp and radar-based qpf techniques for flash flood and landslip warnings in hong...

Documents

hong kongqpf

warning systemsmechanisms

warning signals

flood sirens

governmentrain alert

predetermined alert

heavy rain

range of spatial