examining the impact of land use/land cover characteristics on flood losses

16
This article was downloaded by: [University of Sussex Library] On: 28 October 2014, At: 10:49 Publisher: Routledge Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK Journal of Environmental Planning and Management Publication details, including instructions for authors and subscription information: http://www.tandfonline.com/loi/cjep20 Examining the impact of land use/land cover characteristics on flood losses Samuel Brody a , Russell Blessing b , Antonia Sebastian c & Philip Bedient c a Departments of Marine Sciences/Urban Planning, Texas A&M University, 200 Seawolf Pkwy, Galveston, TX 77553, USA b Center for Texas Beaches and Shores, Texas A&M University at Galveston, Galveston, TX 77553, USA c Department of Civil and Environmental Engineering, Rice University, Houston, TX USA Published online: 06 Jun 2013. To cite this article: Samuel Brody, Russell Blessing, Antonia Sebastian & Philip Bedient (2014) Examining the impact of land use/land cover characteristics on flood losses, Journal of Environmental Planning and Management, 57:8, 1252-1265, DOI: 10.1080/09640568.2013.802228 To link to this article: http://dx.doi.org/10.1080/09640568.2013.802228 PLEASE SCROLL DOWN FOR ARTICLE Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”) contained in the publications on our platform. However, Taylor & Francis, our agents, and our licensors make no representations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of the Content. Any opinions and views expressed in this publication are the opinions and views of the authors, and are not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied upon and should be independently verified with primary sources of information. Taylor and Francis shall not be liable for any losses, actions, claims, proceedings, demands, costs, expenses, damages, and other liabilities whatsoever or howsoever caused arising directly or indirectly in connection with, in relation to or arising out of the use of the Content. This article may be used for research, teaching, and private study purposes. Any substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any form to anyone is expressly forbidden. Terms &

Upload: philip

Post on 26-Feb-2017

215 views

Category:

Documents


1 download

TRANSCRIPT

Page 1: Examining the impact of land use/land cover characteristics on flood losses

This article was downloaded by: [University of Sussex Library]On: 28 October 2014, At: 10:49Publisher: RoutledgeInforma Ltd Registered in England and Wales Registered Number: 1072954 Registeredoffice: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK

Journal of Environmental Planning andManagementPublication details, including instructions for authors andsubscription information:http://www.tandfonline.com/loi/cjep20

Examining the impact of land use/landcover characteristics on flood lossesSamuel Brodya, Russell Blessingb, Antonia Sebastianc & PhilipBedientc

a Departments of Marine Sciences/Urban Planning, Texas A&MUniversity, 200 Seawolf Pkwy, Galveston, TX 77553, USAb Center for Texas Beaches and Shores, Texas A&M University atGalveston, Galveston, TX 77553, USAc Department of Civil and Environmental Engineering, RiceUniversity, Houston, TX USAPublished online: 06 Jun 2013.

To cite this article: Samuel Brody, Russell Blessing, Antonia Sebastian & Philip Bedient (2014)Examining the impact of land use/land cover characteristics on flood losses, Journal ofEnvironmental Planning and Management, 57:8, 1252-1265, DOI: 10.1080/09640568.2013.802228

To link to this article: http://dx.doi.org/10.1080/09640568.2013.802228

PLEASE SCROLL DOWN FOR ARTICLE

Taylor & Francis makes every effort to ensure the accuracy of all the information (the“Content”) contained in the publications on our platform. However, Taylor & Francis,our agents, and our licensors make no representations or warranties whatsoever as tothe accuracy, completeness, or suitability for any purpose of the Content. Any opinionsand views expressed in this publication are the opinions and views of the authors,and are not the views of or endorsed by Taylor & Francis. The accuracy of the Contentshould not be relied upon and should be independently verified with primary sourcesof information. Taylor and Francis shall not be liable for any losses, actions, claims,proceedings, demands, costs, expenses, damages, and other liabilities whatsoever orhowsoever caused arising directly or indirectly in connection with, in relation to or arisingout of the use of the Content.

This article may be used for research, teaching, and private study purposes. Anysubstantial or systematic reproduction, redistribution, reselling, loan, sub-licensing,systematic supply, or distribution in any form to anyone is expressly forbidden. Terms &

Page 2: Examining the impact of land use/land cover characteristics on flood losses

Conditions of access and use can be found at http://www.tandfonline.com/page/terms-and-conditions

Dow

nloa

ded

by [

Uni

vers

ity o

f Su

ssex

Lib

rary

] at

10:

49 2

8 O

ctob

er 2

014

Page 3: Examining the impact of land use/land cover characteristics on flood losses

Examining the impact of land use/land cover characteristics

on flood losses

Samuel Brodya*, Russell Blessingb, Antonia Sebastianc and Philip Bedientc

aDepartments of Marine Sciences/Urban Planning, Texas A&M University, 200 Seawolf Pkwy,Galveston, TX 77553, USA; bCenter for Texas Beaches and Shores, Texas A&M University

at Galveston, Galveston, TX 77553, USA; cDepartment of Civil and Environmental Engineering,Rice University, Houston, TX, USA

(Received 8 November 2012; final version received 1 May 2013)

Characteristics of the built environment and overall local-level land use patterns areincreasingly being attributed to greater surface runoff, flooding and resultingeconomic losses from flood events. Specific configurations of impervious surfaces andland cover may be as important to determining a community’s flood risk as baselineenvironmental conditions. This study addresses this issue by statistically examiningthe impacts of adjacent land use and land cover (LULC) on flood damage recorded onparcels within a coastal watershed in southeast Texas. We analyse empirical models toidentify the influence of different LULCs surrounding over 7900 properties claiminginsured flood losses from 1999–2009. Results indicate that specific types ofsurrounding LULCs impact observed flood losses and provide guidance on howneighbourhoods can be developed more resiliently over the long term.

Keywords: flood losses; land use; land cover; Texas; resilience

1. Introduction

Flooding and associated flood impacts have long been linked to urbanisation and the

proliferation of impervious surfaces (Anderson 1970; Hall 1984). Numerous researchers

have examined and quantified the impact of urban development, particularly as it relates

to surface runoff, using small experimental plots, hydrological simulations and watershed

level land use changes (Tourbier and Westmacott 1981; Tong 1990; Im, Brannan, and

Mostaghimi 2003). The advent of computer software, such as SWMM, GIS L-THIA,

HEC-HMS, HEC-RAS, etc. has made the prediction of runoff from land use change even

more widespread. In contrast, little observational and empirical research has been

conducted on the role of surrounding land use/land cover (LULC), particularly using the

parcel unit of analysis. Even less work has been done to quantify the amount of property

damage incurred as a result of LULC patterns. Increasingly, changes in the built

environment and local-level land use configurations are attributed to flooding and

resulting economic losses from flood events (Brody et al. 2011). Specific patterns of

impervious surfaces and land cover may contribute to a community’s flood risk as much

as baseline environmental conditions (Brody, Kim, and Gunn 2013). We pose the

following research question: do LULCs affect flood losses on adjacent properties and if

so, which types have the most impact?

Our study addresses this question directly by statistically examining the impacts of

*Corresponding author. Email: [email protected]

� 2013 University of Newcastle upon Tyne

Journal of Environmental Planning and Management, 2014

Vol. 57, No. 8, 1252–1265, http://dx.doi.org/10.1080/09640568.2013.802228

Dow

nloa

ded

by [

Uni

vers

ity o

f Su

ssex

Lib

rary

] at

10:

49 2

8 O

ctob

er 2

014

Page 4: Examining the impact of land use/land cover characteristics on flood losses

adjacent LULC on flood damage recorded on parcels within the Clear Creek watershed in

southeast Texas. Specifically, we analyse spatially-weighted linear regression models to

identify the influence of multiple LULC categories within a 0.5 mile buffer surrounding

over 7900 properties claiming insured flood losses from 1999–2009. Control variables

measuring various environmental, socio-economic and built environmental

characteristics within the watershed statistically isolate the effect of each LULC category

on property damage from floods. Results provide insights for local planners, developers

and homeowners on the degree to which different land use patterns affect flood impacts

and how neighbourhoods can be developed more resiliently over the long term. Our study

is intended to complement and extend previous work on development and flooding by

taking a less conventional approach to predicting the impacts of LULC.

1.1. Land use/land cover and flood impacts

A major factor when considering the relationship between LULC and flooding is the

amount of impervious surfaces incorporated into development patterns. Conversion of

natural landscapes to urban or suburban developments can reduce the functionality of

hydrological systems (Wheater and Evans 2009). Large areas of impervious surface can

reduce rainfall infiltration into the soil and increase surface runoff into nearby streams

and rivers (Paul and Meyer 2001; Gill et al. 2007). For example, Arnold and Gibbons

(1996) asserted that stormwater runoff within a drainage basin will nearly double with

only a 10–20% increase in impervious surfaces. More recently, White and Greer (2006)

found that as impervious surface in the Penasquitos Creek watershed in southern

California expanded from 9 to 37%, total runoff was amplified approximately 200% over

their study period (1973 to 2000). Overall, empirical and simulated studies suggest that

increased surface runoff from impervious surface coverage is an important aspect of

flood risk because it can translate into heightened frequency and severity of flooding in

vulnerable areas.

Impervious surfaces are also implicated in increased peak discharge (Brezonik and

Stadelmann 2002). Under these conditions, the lag time between the centre of

precipitation volume and runoff volume is shortened so that floods peak more rapidly

(Hirsch et al. 1990). When rainfall is unable to infiltrate into the soil, rapid onset of

floods can occur because runoff enters water bodies more quickly (Hsu, Chen, and Chang

2000; Hey 2002). For example, Rose and Peters (2001) found that peak discharge

increases approximately 80% in urban catchments containing greater than 50%

impervious area. Similarly, Burns et al. (2005) noticed a 300% increase for a basin with

an impervious area of only 11% when examining mean peak discharges for 27 storms in

the Croton River Basin in New York.

Finally, the resulting increase in surface runoff and peak discharge from development-

based land uses exacerbate human losses. For example, Brody et al. (2007a) found that an

increase in impervious surfaces coincided with a significant increase in stream flow

exceedances over a 12-year period across 85 coastal watersheds in Texas and Florida. A

subsequent study of 37 coastal counties in Texas found that each square metre of

impervious surface added to the landscape translated into approximately $3602 of added

property damage caused by floods per year from 1997–2001 (Brody et al. 2008).

Adverse impacts from floods do not stem solely from the percentage of impervious

surfaces present within the landscape, but the intensity with which they are clustered.

That is, the pattern and form of development becomes important for predicting flood

losses over time. For example, a low density, sprawling suburban setting with 25%

Journal of Environmental Planning and Management 1253

Dow

nloa

ded

by [

Uni

vers

ity o

f Su

ssex

Lib

rary

] at

10:

49 2

8 O

ctob

er 2

014

Page 5: Examining the impact of land use/land cover characteristics on flood losses

impervious cover may result in a different degree of impact compared to a high-intensity,

clustered urban development pattern with 85% paved surfaces. Indeed, our previous

research showed that, even when controlling for multiple environmental and socio-

economic variables, the type of development pattern within counties/parishes along the

Gulf of Mexico coast significantly affects the degree to which communities report

property damage from floods (Brody et al. 2011). Perhaps unexpectedly, jurisdictions

with large amounts of spatially-focused, high-intensity development patterns claimed, on

average, reduced amounts of property damage. This finding supports the efficacy of well-

defined urban cores associated with smart growth approaches to development. It appears

that older, denser urban development is generally situated away from areas exposed to

flooding (such as the floodplain), so this development pattern can lead to more resilient

local communities over the long term. In addition, high density urban areas may be more

likely to have in place a co-ordinated system of flood mitigation infrastructure that can

appropriately handle large amounts of runoff. In general, urbanised communities have

greater resources and commitment to provide municipal-level drainage and flood

protection infrastructure. Finally, dense urban cores are more likely comprised of multi-

storey buildings where only the ground floors may flood.

In contrast, increasing percentages of low-intensity development greatly exacerbate

property damage from flooding. This result supports the argument that outwardly

expanding, low-density development patterns can compromise hydrological systems and

amplify surface runoff by spreading-out impervious surfaces across a watershed (Brody,

Kim, and Gunn 2013). This form of development is also more likely to infringe upon

flood-prone areas that were previously left untouched.

Another type of LULC that has become an increasingly important factor in predicting

flood impacts is based on the presence (or loss of) naturally-occurring wetlands. A large

amount of simulated and observational work has been conducted on mostly rainfall-based

flooding, where wetlands help to reduce flooding and associated losses because of their

ability to hold, store and slowly release accumulated runoff (Bullock and Acreman,

2003).

Both anecdotal and empirical research suggest that wetlands may reduce or slow

precipitation-based flooding (Mitch and Gosselink 2000; Lewis 2001). Some research

also suggests wetlands reduce the impact of storm surge caused by hurricanes (Costanza

et al. 2008). Empirical research in Texas and Florida indicates the value of naturally

occurring wetlands in reducing the adverse impacts of floods. When controlling for

multiple socio-economic and geophysical contextual characteristics, Brody et al. (2008)

found that the loss of wetlands across 37 coastal counties in Texas from 1997 to 2001

significantly increased the observed amount of property damage from floods. On average,

wetland alteration as measured by the number of approved permits added over $38,000 in

property damage to a jurisdiction per flood. A companion analysis for every county in

Florida showed similar results (Brody et al. 2008). In this case, the alteration of wetlands

increased the average property damage per flood by over $400,000. Given the evidence,

we expect that wetland loss will exacerbate property damage from floods even over

larger study areas, such as the Gulf of Mexico coast.

Other types of LULC may also influence the extent of flood loss reported at the parcel

level. For example, modern agriculture operations are thought to affect surrounding

properties in terms of flood impacts. Soil compaction from ploughs and heavy machinery,

loss of vegetative cover and rapid drainage off properties are all implicated in increasing

surface runoff and lead to potential surrounding flood losses (O’Connell et al. 2007;

Wheater and Evans 2009). In particular, reduction in soil infiltration rates and water

1254 S. Brody et al.

Dow

nloa

ded

by [

Uni

vers

ity o

f Su

ssex

Lib

rary

] at

10:

49 2

8 O

ctob

er 2

014

Page 6: Examining the impact of land use/land cover characteristics on flood losses

storage capacity can significantly increase overland flow into adjacent areas (Carroll et al.

2004). In their long-term study of south-east England, Boardman, Evans, and Ford (2003)

found a significant relationship between the presence of autumn-sown cereal fields and

local ‘muddy floods’. It should be noted that muddy floods stemming from agricultural

land uses for the most part adversely impact off-farm properties.

Forest lands surrounding a property may also play a role in determining the extent of

damage experienced during a flood event. On the one hand, forested areas lack

impervious surfaces that reduce infiltration and expedite runoff into local streams. Trees

and woody plants also help to slow and trap runoff, reducing flood peaks more so than

other types of land cover (McCulloch and Robinson 1993). On the other hand, forest

lands along the coast (particularly in the Gulf of Mexico and similar environments) can

be located in low-lying areas comprised of floodplains. Houses situated adjacent to or

within forested bottom-lands are extremely vulnerable to flooding regardless of LULC

composition.

Finally, open space, such as parks and local preserves may reduce adverse impacts to

houses located on adjacent parcels. Open space land use usually contains much less

impervious surface compared to other development types, such as commercial or high-

density residential. With the presence of ball-fields and playgrounds, drainage is usually

well-maintained and runoff appropriately captured. Finally, developed open space is

most often integrated into the spatial fabric of local neighbourhoods such that their flood-

reducing effects will be more evident. Gill et al. (2007), for example, found that under

the right conditions, green space can significantly reduce surface run-off in Manchester,

England.

One of the major limitations of the research cited above is the spatial scale at which it

has been conducted. Conclusions drawn from both observational and simulated data were

made at a watershed or jurisdictional (city or county/parish) level. Lumped hydrologic

models, such as HEC-HMS (USACE 2000), which are used to explicitly simulate

watershed hydrology, lack the ability to capture small variations in LULC on flooding at

the parcel level, while distributed models, such as SWMM (Huber and Dickinson 1988),

are limited in their application to large watersheds by data availability and computational

power. Recent studies have used hydrologic modelling to examine the impact of Low

Impact Development technologies on peak flow and runoff volumes in small watersheds

or neighbourhoods (Willams and Wise 2006; Deitz and Clausen 2008), but few have

directly examined the impact of adjacent LULC on flooding at the parcel level using

statistical methods. We hope our research outlined below offers an important initial step

in examining flood impacts at the site level.

2. Research methods

We selected as the study area the Clear Creek watershed, which is located 20 miles south

of Houston, Texas. This 197 square mile area is situated adjacent to Galveston Bay on the

Gulf of Mexico coast and encompasses the four counties of Galveston, Brazoria, Harris

and Fort Bend (Figure 1). The watershed is drained primarily by Clear Creek and

associated tributaries. Clear Creek itself is a tidally influenced bayou that terminates as it

enters Clear Lake, which then opens into west Galveston Bay. The Clear Creek

Watershed typifies the Gulf Coast physical environment, with very little topographic

relief, large percentages of floodplain area, wide floodplain boundaries and exposure to

frequent storm events. Over the study period, from 1999 to 2009, both multiple small-

scale and major flooding impacted the Clear Creek Watershed. For example, Tropical

Journal of Environmental Planning and Management 1255

Dow

nloa

ded

by [

Uni

vers

ity o

f Su

ssex

Lib

rary

] at

10:

49 2

8 O

ctob

er 2

014

Page 7: Examining the impact of land use/land cover characteristics on flood losses

Storm Allison in 2001 dropped over 18 inches of rain across the study area, becoming the

costliest tropical storm in US history. Overall, significant increases in residential

development in vulnerable areas over the last decade have resulted in large amounts of

property damage caused by repetitive flooding events.

2.1. Concept measurement

Flood damage, the dependent variable for the study, was measured as the dollar amount

(contents plus building damage) of claims paid per household under the National Flood

Insurance Program (NFIP). The NFIP is administered through FEMA (Federal Emergency

Management Agency) to provide flood insurance to residents and businesses. In

participating communities, residential structures within the 100-year floodplain with a

mortgage are required to purchase flood insurance. While insured loss does not cover all

damage incurred from a flood event, the market penetration of NFIP policies is so high

within our study area that these loss figures more accurately reflect overall flood damage

than in regions where the NFIP is less prevalent. The dollar value of individual insured

claims for both building contents and structures was aggregated at the parcel level and

log-transformed to better approximate a normal distribution (see Table 1). Out of

approximately 31,000 policies purchased, 7942 made claims during the study period. Each

claim was geocoded using the centroid of the parcel and analysed in a Geographic

Information System (GIS).

LULC variables were derived from the NOAA C-CAP dataset using Landsat

Thematic Mapper data at 30 metre resolution from both 2001 and 2006. Because

comparable remote sensing datasets for the study area were available for only these two

years, we tied LULC variables to the nearest claim, temporally. For example, a flood-

damaged parcel in 2002 would be tied to the 2001 dataset, while a loss in 2007 would be

Figure 1. Clear Creek watershed study area.

1256 S. Brody et al.

Dow

nloa

ded

by [

Uni

vers

ity o

f Su

ssex

Lib

rary

] at

10:

49 2

8 O

ctob

er 2

014

Page 8: Examining the impact of land use/land cover characteristics on flood losses

Table1.

Conceptmeasurement.

Measurement

Source

Range

Mean

Standarddeviation

Dependentvariable

Floodloss

Logged

NFIP

claimsforfloodlosses

from

1999–2009

NFIP

0–13.56

7.70

4.24

AbsoluteNFIP

claimsforfloodlosses

from

1999–2009in

USdollars

NFIP

$0–780,798.7

$33,999.61

$53,642.67

Landuse/landcovervariables

Highintensity

development

80%þ

impervioussurfacewithin

0.5milebuffer

NOAA,CoastalChange

&AnalysisProgram

0–47.61

5.32

5.49

Medium

intensity

development

50–79%

impervioussurfacewithin

0.5milebuffer

NOAA,CoastalChange

&AnalysisProgram

0–84.60

26.96

19.82

Lowintensity

development

21–49%

impervioussurface

within

0.5milebuffer

NOAA,CoastalChange

&AnalysisProgram

0–57.01

21.09

10.36

Developed

open

space

Green

space;<20%

impervioussurface

within

0.5milebuffer

NOAA,CoastalChange

&AnalysisProgram

0–68.57

11.53

9.29

Agriculture

Croplandswithin

0.5milebuffer

NOAA,CoastalChange

&AnalysisProgram

0–87.70

3.70

7.89

Forest

Deciduous,evergreen,andmixed

forestcover

within

0.5milebuffer

NOAA,CoastalChange

&AnalysisProgram

0–38.41

7.09

7.56

Grass

>80%

herbaceousvegetation

within

0.5milebuffer

NOAA,CoastalChange

&AnalysisProgram

0–30.94

2.03

2.18

Scrub

<20%

shrubsless

than

16feettall

within

0.5milebuffer

NOAA,CoastalChange

&AnalysisProgram

0–13.55

1.03

1.71

Barren

Earth

material;<20%

vegetation

cover

within

0.5milebuffer

NOAA,CoastalChange

&AnalysisProgram

0–7.73

0.10

0.29

Palustrinewetland

Non-tidalforested,scrub/shrub,

andem

ergentdominated

vegetation

NOAA,CoastalChange

&AnalysisProgram

0–43.39

4.24

6.67

Estuarinewetland

Tidalforested,scrub/shrub,

andem

ergentdominated

vegetation

NOAA,CoastalChange

&AnalysisProgram

0–22.25

0.85

2.67

Environmentalcontrolvariables

Elevation

Feet

NationalElevationDataset

0–19

6.69

4.59

Precipitation

Inches

per

month

PRISM

0.02–21.47

13.62

5.14

Structuralcontrolvariables

Property

value

Assessedtaxvalue

County

taxoffice

0–5.56eþ

07

219,395

1,122,545

Ageofstructure

Yearbuilt

FEMAPolicy

Data

1900–2009

1972.61

88.19

Floodplain

Position

In-outof100-yearfloodplain

FEMAQ3data

0/1

0.54

0.49

Spatiallag

Nearestneighbourspatialautocorrelation

FEMANFIP

data

�0.63-4.05

0.008

0.611

Journal of Environmental Planning and Management 1257

Dow

nloa

ded

by [

Uni

vers

ity o

f Su

ssex

Lib

rary

] at

10:

49 2

8 O

ctob

er 2

014

Page 9: Examining the impact of land use/land cover characteristics on flood losses

attributed to the 2006 data. We extracted and analysed the following previously classified

11 variables: high (contains 80–100% constructed materials), medium (50–79%

constructed materials), and low intensity (21–49% constructed materials) development;

developed open space (less than 20% constructed materials), agriculture, grassland,

palustrine wetland, estuarine wetland, forest, scrub and barren land (for more

information, see www.noaa.gov/landcover).

High intensity development consists primarily of concrete, asphalt and other

constructed materials accounting for 80 to 100% of the total cover for each 30�30 meter

pixel. This variable signifies heavily built-up urban centres and large constructed surfaces

in suburban and rural areas. Medium intensity development consists of constructed

materials covering 50–79% of total area. This variable captures multi and single-family

housing areas, especially in suburban neighbourhoods (neither high nor medium intensity

variables distinguish between single or multi-storey buildings). Low intensity

development includes areas with a mixture of constructed materials and large amounts of

vegetation or other cover. Constructed materials for this variable account for 21–49% of

the total area, and commonly includes single-family housing areas, especially in rural

neighbourhoods. The developed open space variable contains some constructed

materials, but consists mostly of managed grasses or low-lying vegetation planted in

developed areas for recreation, erosion control or aesthetic purposes. This variable

captures public parks, ball fields and other recreational use areas. Constructed surfaces

account for less than 20% of total land cover.

Agriculture was measured by combining cultivated crops intensely managed for

production and pasture areas consisting of grasses planted for livestock grazing or hay

crops. Grassland measures areas dominated (over 80%) by herbaceous vegetation. These

areas are not subject to intensive management such as tilling, but can be utilised for

grazing livestock. We combined the following three land classifications to derive the

forest variable: deciduous, evergreen and mixed forest. Trees in each category are over

16 feet tall and greater than 20% of the total vegetation cover. The scrub land cover

variable contains areas dominated by shrubs less than 5 metres tall, with shrub canopy

typically accounting for greater than 20% of total vegetation. This variable includes tree

shrubs, young trees in an early successional stage or trees stunted by environmental

conditions. Barren land cover was measured as areas of bedrock, mines, gravel pits and

other accumulations of earth material. In general, vegetation for this class accounts for

less than 10% of total cover. Finally, we measured two types of wetlands land cover

types within the study area. First, palustrine wetlands were measured by combining three

remote sensing classes: tidal and non-tidal forested, scrub/shrub and emergent dominated

vegetation. Second, we included in our analysis an estuarine wetland variable comprised

of the same three vegetation types, but only for tidal areas. This variable better captures

wetlands associated with the direct coastline influenced by saltwater. It should be noted

that the study area entirely comprises clay-based soils.

The percentage area for each LULC was calculated within a 0.5 mile buffer around

each parcel incurring flood loss during the study period. We selected a 0.5 mile buffer as

the most reasonable distance surrounding LULC will impact a specific parcel within the

Clear Creek watershed. We based this measurement decision by measuring the average

diameter of developed patches across the study area. Using this procedure, we

determined that a 0.5 mile buffer would best represent the run-off effects of surrounding

LULC. In addition, we took into account the spatial resolution of our data, local

topography and drainage network. We also conducted a sensitivity analysis using buffer

distances of 0.25 and 1 mile, which yielded the same overall results.

1258 S. Brody et al.

Dow

nloa

ded

by [

Uni

vers

ity o

f Su

ssex

Lib

rary

] at

10:

49 2

8 O

ctob

er 2

014

Page 10: Examining the impact of land use/land cover characteristics on flood losses

Several contextual control variables were also measured and analysed in the statistical

models. Under structural characteristics, the mean assessed value (based on local tax

records for 2011, the date closest to the study period when all properties were available)

for each damaged parcel was calculated as a key economic control variable. Multiple

parcels in the county tax record were missing assessment values. In these instances, we

dropped the observation for the dataset. Assessed value was also used in an alternate

specification of the dependent variable (insured loss/assessed value), but did not yield

significantly difference results than including this variable as a separate independent

statistical control. The year in which each structure was built was also incorporated into

the statistical models. The year of construction ranged from 1900 to 2009. As a measure

of risk exposure, we also calculated whether a parcel was in or out of the 100-year

floodplain using FEMA Q3 data. Fifty-five per cent of all claims were located outside of

the floodplain.

We also included several environmental baseline controls. Traditionally, the biggest

predictor of flooding and flood damage is precipitation. We measured this variable based

on the annual rainfall amounts recorded in the PRISM (Parameter-elevation Regressions

on Independent Slopes Model) Climate Group dataset. Monthly precipitation data were

mapped at a scale of 2.5-arcmin (approximately 4 � 4 km grid cells) normal and joined

to the flood damaged parcel with the data time period corresponding to the month of the

claim. Precipitation estimated for each cell is an average over the entire area of that cell.

It is important to note here that the vast majority of flood claims happened during long

duration storm events that accounted for nearly all of the rainfall that occurred during an

entire month. This rainfall pattern is common along the Gulf Coast and will, for the most

part, mask any potential inaccuracies due to the temporally low resolution dataset. That

said, PRISM datasets are considered to be the highest-quality spatial climate data

currently available. Elevation of each property was measured using 1/9 arc-second

(approximately 3 � 3 metre grid cells) LIDAR data from the National Elevation Dataset.

The average elevation is approximately 22 feet, which illustrates the low-lying

topography of the Texas coastal plain.

2.2. Data analysis

We analysed the data using linear regression models to explain the variation in insured

flood losses across the study area. A Global Moran’s I statistic revealed the presence of

significant (p <0.000) spatial-autocorrelation in the dependent variable. To correct for this

bias, we estimated the spatial lag using a nearest-neighbour function. The lag variable was

calculated based on eight nearest neighbours of flood damage parcels. Sensitivity analysis

and model fit suggested eight nearest-neighbours was the most appropriate statistical

threshold. Once created, we included the spatial lag variable in the regression model as a

statistical control. Diagnostics also revealed the presence of heteroskedacticity in the data,

leading us to estimate the model with robust standard errors. No other statistical biases

were found. Both non-standardised and standardised (‘beta’) regression coefficients were

reported for regression models. Beta coefficients are useful to compare the relative strength

of various predictors within a single regression model. Because the beta coefficients are

measured in standard deviations, instead of the original units of the variables, they can be

effectively compared to one another. Essentially, the betas are the coefficients one would

obtain if the outcome and predictor variables were all transformed standard scores before

running the regression. Thus, the coefficient for the jth regressor indicates how many

standard deviations y would change given a 1-standard deviation change in xj.

Journal of Environmental Planning and Management 1259

Dow

nloa

ded

by [

Uni

vers

ity o

f Su

ssex

Lib

rary

] at

10:

49 2

8 O

ctob

er 2

014

Page 11: Examining the impact of land use/land cover characteristics on flood losses

3. Results

Over the 11-year study period, we mapped and analysed across the Clear Creek watershed

9792 NFIP-based flood damage claims totalling over $356 million (in 2009 dollars), most

of which stemmed from structural losses. Over 50% of these losses occurred outside of

the 100-year floodplain. The average loss per claim across the watershed was $36,585

($26,902 for building and $9683 for contents) (for more information on descriptive

statistics, see Brody et al. 2013).

As shown in Table 2, baseline environmental contextual controls behave as expected.

Parcels located at higher elevations (note the average for the study area is only 6.7 feet

above sea level) incurred significantly less property damage (p < 0.001). Elevation

proved to be an especially important variable for mitigating flood loss during surge based

events, such as Hurricane Ike in 2008. Precipitation is by far the most powerful predictor

variable in the model (aside from the spatial lag), where increasing amounts of rainfall

result in increased property damage (p < 0.001). Together, environmental controls

explain approximately 10% of the variation in the dependent variable.

The four structural control variables by comparison almost double the explanatory

power of the statistical model. On average higher assessed property values (more

expensive homes) correspond with less insured damage. This result may seem counter-

intuitive, but more expensive homes are usually located in developments with high

quality drainage systems and their owners are more likely to invest in mitigation

alternatives, such as elevating lower floors, sum pumps, drains, etc. Over the long term,

these modifications can result in a significant reduction in the adverse impacts associated

with floods. The age of the structure does not have a statistically significant impact on

insured flood loss. Newer homes may have better building codes or drainage systems, but

they tend to be located in more flood-prone areas that were originally considered

undesirable. As expected, the position of a structure in relation to the 100-year floodplain

has a major effect on the likelihood of flood damage. Those properties located outside the

floodplain are less exposed to flood risk and therefore report significantly less (p < 0.001)

property damage from flooding events. While over half of the total number of flood

claims is located outside of the floodplain, a significant dollar amount of loss remains

within these boundaries. Finally, the spatial lag control based on the strong presence of

spatial autocorrelation within the dependent variable is significant, where increasing

spatial clustering of claims results in greater amounts of flood loss.

Specific types of land use and land cover surrounding flood-impacted parcels also

affect the amount of insured damage incurred by property owners. Built environment

land uses have a particularly strong effect, indicating the intensity of neighbourhood

development is important when considering resiliency to floods. For example, increasing

percentages of high intensity development lead to greater dollar amounts of flood damage

(p < 0.001). Within the Clear Creek Watershed, these areas consist of large shopping

centres located off major highways. Even with drainage infrastructure, the large amount

of impervious surfaces exacerbates runoff and associated flooding. However, surrounding

low intensity development characterised by sprawling, low-density residential

development patterns has a far greater positive effect on property damage caused by

floods. Based on the standardised betas (standardised coefficients in the regression

model) shown in Table 2, the average impact on flooding is over three times greater

under low intensity development patterns than high intensity. In contrast, neighbourhoods

dominated by medium intensity land use development (50–79% imperviousness) are

significantly less likely to experience large amounts of flood loss (p < 0.001). These

1260 S. Brody et al.

Dow

nloa

ded

by [

Uni

vers

ity o

f Su

ssex

Lib

rary

] at

10:

49 2

8 O

ctob

er 2

014

Page 12: Examining the impact of land use/land cover characteristics on flood losses

Table2.

Regressionmodelsexplaininginsuredfloodlosses,1999–2009.

Model1

Model2

Model3

Variable

bt

beta

seb

tbeta

seb

tbeta

se

Elevation

�0.217

�19.49��

��0

.233

0.0111

�0.0601

�5.20��

��0

.0647

0.0116

�0.0642

�4.5

���

�0.0691

0.0143

Precipitation

0.2523

26.23��

�0.3038

0.0096

0.249

27.42��

�0.2998

0.0091

0.2906

28.73��

�0.3499

0.0101

Assessedvalue

0�2

.13�

�0.0243

00

�1.15

�0.0215

0Yearbuilt

�0.0011

�1.31

�0.0216

0.0008

�0.0009

�1.15

�0.0182

0.0008

Floodplain

�0.8302

�9.54��

��0

.0964

0.087

�0.9407

�10.45��

��0

.1092

0.09

Spatiallag

2.6368

31.71��

�0.3707

0.0831

2.3668

27.65��

�0.3327

0.0856

Highintensity

dev.

0.0346

3.59��

�0.044

0.0096

Medium

intensity

dev.

�0.0118

�4.84��

��0

.056

0.0024

Lowintensity

dev.

0.0569

12.47��

�0.143

0.0046

Open

space

�0.0145

�2.59��

�0.0325

0.0056

Agriculture

�0.002

�0.22

�0.0037

0.0094

Grassland

�0.0736

�2.62��

�0.0386

0.0281

Forest

�0.0294

�0.33

�0.005

0.09

Barren

0.1039

0.63

0.0073

0.1662

Scrub

0.1032

2.28�

0.0423

0.0453

Wetlandpalustrine

�0.0475

�4.73��

��0

.0706

0.0101

Wetlandestuarine

�0.0131

�0.78

�0.0085

0.0169

Constant

5.6332

41.82��

�0.135

7.2997

4.43��

�1.648

5.7974

3.59��

�1.6134

N7942

7942

7942

R-squared

0.0936

0.2341

0.2551

Adjusted

R-squared

0.0911

0.2334

0.2534

Notes:

� p<0.05;��p<0.01;��� p

<0.001.

Journal of Environmental Planning and Management 1261

Dow

nloa

ded

by [

Uni

vers

ity o

f Su

ssex

Lib

rary

] at

10:

49 2

8 O

ctob

er 2

014

Page 13: Examining the impact of land use/land cover characteristics on flood losses

areas usually consist of fairly dense, more recently-built suburban developments, such as

master-planned communities. Finally, the presence of surrounding developed open space,

which has the least amount of impervious cover compared to the other built environment

land uses, significantly reduces the amount of recorded flood damage (p ¼ 0.01). Parks

and recreation areas which comprise mostly green space and vegetation seem to provide

a buffer against negative flood impacts.

In terms of what we consider undeveloped LULC, agricultural operations surrounding

a flooded property have a negative, but statistically insignificant effect on the dependent

variable. Large percentages of adjacent grassland, however, do decrease observed

property damage from floods, where p > 0.01. Herbaceous plants with non-woody stems

may be enough to capture surface run-off without causing ponding. Forest, barren and

scrub land covers, which make up a very small amount of total area, have a negligible

impact on flooding. Finally, naturally-occurring wetlands within the study area seem to

play an important role in influencing the degree of property damage from floods.

Estuarine wetlands located primarily along the coastline do not have a significant effect.

However, freshwater palustrine wetlands surrounding a property significantly attenuate

flood losses (p < 0.001). In fact, this wetland type has the strongest flood damage

reduction effect among all LULCs in the model. This result corroborates previous work

examining the relationship between wetlands and flooding (Brody et al. 2007b).

4. Discussion

Our statistical results indicate that the local configuration of land use plays an important

role in predicting the amount of property damage caused by floods at the parcel level.

Although the overall explanatory power of LULCs is low compared to baseline

environmental and structural characteristics, several categories nevertheless have a

significant impact on the amount of adjacent flood losses.

Specifically, the intensity of development based on the percentage of impervious

surface within a 30-metre grid has the most profound statistical implications when

considering flood risk reduction. A surrounding built environment pattern dominated by

medium intensity development typical of planned suburban communities appears to have

the greatest effect or magnitude on reducing the amount of insured losses incurred across

the watershed. These developments typically have better on-site drainage mechanisms as

well as neighbourhood level stormwater infrastructure, such as retention/detention ponds

with higher capacity to offset the increases in surface runoff. Because this pattern of

development also includes interspersed areas of vegetation (including parks and golf

courses), there is also more opportunity to incorporate green space for flood control. This

argument is supported by our finding that developed open space land use decreases

property damage from floods. Local parks, playing fields and other recreational facilities

consist primarily of green space that can act as a storm buffer to surrounding properties.

Parks also often double as runoff detention areas to hold and slowly release water on to

adjacent parcels. Open space land use can be strategically sited within neighbourhoods

not just to provide recreational opportunities, but also to reduce the adverse impacts of

localised flooding.

In contrast, low intensity, more sprawling development patterns significantly increase

flood losses. In fact, this development category has the strongest overall statistical effect of

all LULCs in our model, where a percentage increase in surrounding low-intensity

development translates into, on average, approximately $1734 in additional property

damage caused by floods. Low-density development patterns can compromise hydrological

1262 S. Brody et al.

Dow

nloa

ded

by [

Uni

vers

ity o

f Su

ssex

Lib

rary

] at

10:

49 2

8 O

ctob

er 2

014

Page 14: Examining the impact of land use/land cover characteristics on flood losses

systems and amplify surface runoff by spreading out impervious surfaces across a

watershed, placing more structures and residents at risk from flooding over a larger area.

Furthermore, unlike well-established urban centres, low-density suburban and ex-urban

communities are less likely to have adequate storm drainage systems and other

infrastructure to accommodate increase increased surface runoff. Drainage in these areas is

often the responsibility of an individual subdivision (land divided into pieces for the

purpose of development), which could simply convey runoff offsite through hardened

channels and exacerbate downstream flood impacts. Finally, sprawling residential

development on the periphery of older urban areas and outside extraterritorial jurisdictions

(ETJs) (government exercised authority outside it normal boundaries) may be more likely

to encroach on floodplains that were originally left as open space or low-impact uses.

Despite the expectation that agricultural land uses will exacerbate flood losses, our

results show a negative, albeit statistically insignificant relationship. Closer investigation

of agricultural operations within the Clear Creek watershed reveals that only a fraction

are cultivated crops involving active tilling and soil compaction. Almost all of the

agricultural acreage is classified as pasture or hay fields which produces far less surface

runoff that may negatively impact adjacent parcels. In general, we find that more passive

agricultural uses can be successfully integrated into a residential setting without causing

flood damage to surrounding properties during a storm event.

Our results also support the notion that more naturally occurring land cover, such as

grass and forestland, help reduce flood-related losses. Once we controlled for elevation,

both these surrounding land covers significantly reduce reported property damage

throughout the study period. Grass and forest cover are effective for capturing and

slowing surface runoff before it reaches residential parcels. Forest cover has a stronger

effect, most likely because it contains more woody materials that help retard and retain

overland flow.

Finally, our results corroborate previous work at larger spatial scales supporting the

flood attenuation properties of naturally occurring wetlands (Highfield and Brody 2006;

Brody et al 2008). In particular, palustrine wetlands consisting of freshwater

environments significantly reduce observed flood losses over a variety of storm types. It

is the second-most powerful LULC variable in our analysis and should be considered an

important flood mitigation tool when considering the location of future development. In

contrast, we did not find any statistical evidence that estuarine wetlands provide flood

damage reduction functions, even for surge-based events. The estuarine wetlands in our

study area comprise salt marsh, which may not suppress surge compared to other

vegetation types, such as mangroves.

5. Conclusion and future research

Overall, our study finds that the patterns with which we develop and use the landscape

have important implications when establishing flood resilient communities. Planners and

decision makers should factor in the local pattern of LULC when designing flood

reduction programmes and be more strategic when deciding where future development

should take place. Local land uses should also be coupled with regional structural

mitigation strategies, such as dikes and levees. Our study focuses on a coastal watershed

in upper Texas, but the findings should translate to other similar coastal settings

nationally and internationally. For example, planners in countries with traditional land

use patterns, such as England, Australia and the Netherlands, can gain insight from our

results on the factors contributing to flood loss.

Journal of Environmental Planning and Management 1263

Dow

nloa

ded

by [

Uni

vers

ity o

f Su

ssex

Lib

rary

] at

10:

49 2

8 O

ctob

er 2

014

Page 15: Examining the impact of land use/land cover characteristics on flood losses

While this study is one of the first to examine flood losses over a large number of

individual parcels, it should be considered only a first step in a more thorough

investigation of the topic. First, our analysis only considers one watershed and

approximately 9700 parcels. Future research should expand the study area to include

multiple watersheds in different coastal settings. This approach would enable researchers

to compare across different watershed settings, not just specific parcels. Second, we only

examine the percentage of specific LULCs within a 0.5 mile buffer around each damaged

property. Additional work should be done to determine the specific mixture of land uses

and their synergistic effects on impacting flood losses. Adjacent development patterns,

including connectivity, may also be important variables to consider. This line of inquiry

will provide more information on how to plan for and develop more flood resilient

communities. Third, future analyses should include additional proximity and site-specific

variables. For example, measuring a property’s distance from the floodplain, coastline

and other physical features will help better explain the relationship between residential

location and flood vulnerability. Fourth, future study should include additional control

variables to further isolate the effects of LULC, if the data are available, including

drainage system characteristics, structural and non-structural mitigation measure, soil

properties, etc. Finally, future research should investigate the type of flooding event (e.g.

precipitation or tidal-based) and its impact on the characteristics of loss over time.

References

Anderson, D. G. 1970. Effects of Urban Development of Floods in Northern Virginia. USGS WaterSupply Paper 2001-C:26. Washington, DC: USGS.

Arnold, C. L., and J. C. Gibbons. 1996. “Impervious Surface Coverage: The Emergence of a KeyEnvironmental Indicator.” Journal of the American Planning Association 62 (2): 243–258.

Boardman, J., R. Evans, and J. Ford. 2003. “Muddy Floods on the South Downs Southern England:Problems and Responses.” Environmental Science and Policy 6 (1): 69–83.

Brezonik, P. L. and T. H. Stadelman. 2002. “Analysis and Predictive Models of Stormwater RunoffVolumes, Loads and Pollutant Concentrations From Watersheds in the Twin CitiesMetropolitan Area, Minnesota, USA.”Water Resources 36 (7): 1743–1757.

Brody, S. D., R. Blessing, A. Sebastian, and P. Bedient. 2013. “Delineating the Reality of FloodRisk and Loss in Southeast, Texas.” Natural Hazards Review 14: 89–97.

Brody, S. D., J. Gunn, W. E. Highfield, and W. G. Peacock. 2011. “Examining the Influence ofDevelopment Patterns on Flood Damages Along the Gulf of Mexico.” Journal of Planning andEducation Research 31 (4): 438–448.

Brody, S. D., W. E. Highfield, H. C. Ryu, and L. Spanel-Weber. 2007a. “Examining theRelationship Between Wetland Alteration and Watershed Flooding in Texas and Florida.”Natural Hazards 40 (2): 413–428.

Brody, S. D., H. J. Kim, and J. Gunn. 2013. “The Effect of Urban Form on Flood Damage.” UrbanStudies 50 (4): 789–806.

Brody, S. D., S. Zahran, W. E. Highfield, H. Grover, and A. Vedlitz. 2008. “Identifying the Impactof the Built Environment on Flood Damage in Texas.” Disasters 32 (1): 1–18.

Brody, S. D., S. Zahran, P. Maghelal, H. Grover, and W. Highfield. 2007b. “The Rising Costs ofFloods: Examining the Impact of Planning and Development Decisions on Property Damage inFlorida.” Journal of the American Planning Association 73 (3): 330–345.

Bullock, A., and M. Acreman. 2003. “The Role of Wetlands in the Hydrological Cycle. “Hydrologyand Earth System Sciences 7 (3): 358–389.

Burns, D., T. Vitvar, J. McDonnell, J. Hassett, J. Duncan, and C. Kendall. 2005. “Effects ofSuburban Development on Runoff Generation in the Croton River Basin, New York, USA.”Journal of Hydrology 311 (1–4): 266–281.

Carroll, Z. L., S. B. Bird, B. A. Emmett, B. Reynolds, and F. L. Sinclair. 2004. “Can treeshelterbelts on agricultural land reduce flood risk?” Soil Use and Management 20 (3): 357–359.

1264 S. Brody et al.

Dow

nloa

ded

by [

Uni

vers

ity o

f Su

ssex

Lib

rary

] at

10:

49 2

8 O

ctob

er 2

014

Page 16: Examining the impact of land use/land cover characteristics on flood losses

Costanza, R., O. P�erez-Maqueo, M. L., Martinez, P. Sutton, S. J. Anderson, and K. Mulder. 2008.“The Value of Coastal Wetlands for Hurricane Protection.” Ambio 37 (4): 241–247.

Deitz, M. E., and J. C. Clausen. 2008. “Stormwater Runoff and Export Changes With Developmentin a Traditional and Low Impact Subdivision.” Journal of Environmental Management 87 (4):560–566.

Gill, S. E., J. F. Handley, A.R. Ennos, and S. Pauleit. 2007. “Adapting Cities for Climate Change:The Role of the Green Infrastructure.” Built Environment 33 (1): 115–133.

Hall, M. J. 1984. Urban Hydrology. London: Elsevier.Hey, D. L., 2002. “Modern Drainage Design: The Pros, the Cons, and the Future.” Hydrological

Science and Technology 18 (14): 89–99.Highfield, W. E. and S. D. Brody. 2006. “The Price of Permits: Measuring the Economic Impacts of

Wetland Development on Flood Damages in Florida.” Natural Hazards Review 7 (3): 23–30.Hirsch, R. M., J. F. Walker, J. C. Day, and R. Kallio. 1990. “The Influence of Main on Hydrological

Systems.” In Surface Water Hydrology, edited by M. G. Wolman and H. C. Riggs, 329–359.Boulder, CO: Geological Society of America, Vol. 0–1.

Hsu, M. H., S. H. Chen, and T. J. Chang. 2000. “Inundation Simulation for Urban Drainage BasinWith Storm Sewer System.” Journal of Hydrology 234 (1–2): 21–37.

Huber, W. and R. E. Dickinson. 1988. Storm Water Management Model, Version 4: User’s Manual.Athens, Georgia: US EPA. EPA/600/3-88/001a.

Im, S., K. M. Brannan, and S. Mostaghimi. 2003. “Simulating Hydrologic and Water QualityImpacts in an Urbanizing Watershed.” Journal of the American Water Resources Association39 (6): 1465–1479.

Lewis, W. M., 2001. Wetlands Explained: Wetland Science, Policy, and Politics in America. NewYork: Oxford University Press.

McCulloch, J. S. G., and M. Robinson. 1993. “History of Forest Hydrology.” Journal of Hydrology150 (2–4): 189–216.

Mitch, W. J., J. G. Gosselink. 2000.Wetlands. 3rd ed. New York: John Wiley & Sons.O’Connell, E., J. Ewen, G. O’Donnell, and P. Quinn. 2007. “Is There a Link Between Agricultural

Land-Use Management and Flooding?” Hydrology and Earth Sciences 11 (1): 96–107.Paul, M. J., and J. L. Meyer. 2001. “Streams in the Urban Landscape.” Annual Review of Ecological

Systems 32: 333–365.Rose, S., and N. Peters. 2001. “Effects of Urbanization on Streamflow in the Atlanta Area (Georgia,

USA): A Comparative Hydrological Approach.” Hydrological Proceedings 15: 1441–1457.Tong, S. T. Y. 1990. “The Hydrologic Effects of Urban Land Use: A Case Study of the Little Miami

River Basin.” Landscape and Urban Planning 19 (1): 99–105.Tourbier, J. T., and R. Westmacott. 1981. Water Resources Protection Technology: A Handbook of

Measures to Protect Water Resources in Land Development. Washington, DC: The Urban LandInstitute.

USACE (United States Army Corps of Engineers). 2000. Hydrologic Modeling System. HEC-HMSTechnical Reference Manual. CPD-74B. Davis, CA: Hydrologic Engineering Center.

Wheater, H., and E. Evans. 2009. “Land Use, Water Management, and Future Flood Risk.” LandUse Policy 26S: S251–S264.

White, M. D., and K. A. Greer. 2006. “The Effects of Watershed Urbanization on the StreamHydrology and Riparian Vegetation of Los Pen

asquitos Creek, California.” Landscape and

Urban Planning 74: 125–138.Williams, E. S., and W. R. Wise. 2006. “Hydrologic Impacts of Alternative Approaches to Storm

Water Management and Land Development.” Journal of the American Water ResourcesAssociation 42 (2): 443–455.

Journal of Environmental Planning and Management 1265

Dow

nloa

ded

by [

Uni

vers

ity o

f Su

ssex

Lib

rary

] at

10:

49 2

8 O

ctob

er 2

014