Improving Flood Modelling and Visualisation using Remote ...

Improving Flood Modelling and

Visualisation using Remote Sensing

David Mason1, Paul Bates2, Johanna Dall’Amico1, Matt Horritt3,

Jeff Neal2, Guy Schumann2, Rainer Speck4.

1Environmental Systems Science Centre, University of Reading, UK2School of Geographical Sciences, University of Bristol, UK

3Halcrow Group Ltd.4DLR Oberpfaffenhofen, Wessling, Germany

Predictions of flood extent

Used -

– for maintaining flood defences

– for emergency flood relief management

– for risk assessment

Object is to improve flood models by using remotely sensed

data (SAR, LiDAR) to validate and parameterise the

models.

ERS-1 SAR image of 1992 Thames flood, with flood

extent (waterline) from snake superimposed (green).

Laser Altimetry - LiDAR

Scanning pulsed laser

Horizontal resolution ~1m

Vertical height accuracy 10-15cm

Sawtooth pattern of heights

Uses of LiDAR for model parameterisation

• Providing an accurate DTM of the floodplain.

• Estimating vegetation heights from which floodplain friction can be derived.

• Generating unstructured grids incorporating buildings and taller vegetation.

SAR image west of Oxford with snake superimposed

Improving SAR flood extent using LiDAR

LiDAR image west of Oxford, with relative SAR snake

heights superimposed

Height differences between pairs of corresponding points on the

snake and aerial photo waterlines versus distance downstream

-500

0

500

1000

1500

2000

0 2000 4000 6000 8000 10000 12000

Downstream distance (m)

Heig

ht

dif

fere

nces (

cm

)

Snake conditioned on SAR

and LiDAR, superimposed

on SAR

Snake conditioned on SAR

and LiDAR, superimposed

on LiDAR

Paired height differences versus distance downstream for

snake conditioned on (a) SAR, (b) SAR and LiDAR.

-500

0

500

1000

1500

2000

0 2000 4000 6000 8000 10000 12000

Downstream distance (m)

Heig

ht

dif

fere

nces (

cm

)

-500

0

500

1000

1500

2000

0 2000 4000 6000 8000 10000 12000

Downstream distance (m)

Heig

ht

dif

fere

nces (

cm

)

(a) SAR (b) SAR and LiDAR

Improved performance measure for model validation

(a) Traditional areal method

F = A / (A + B + C)

where A = area correctly predicted as wet by the model

B = area predicted as wet that is actually dry

C = area predicted as dry that is actually wet

F = 1 when observed and predicted flood extents coincide exactly,

0 when no overlap

(b) Mean height difference method

Select pairs of corresponding points on model and SAR waterlines in

areas of low slope. Use a paired t-test to estimate the probability P(t>|t0|)

that their mean height difference is not significantly non-zero (if it is,

model run is non-behavioural).

Flow modelling

• Modelled the 1992 Thames flood

using LISFLOOD-FP (12km

reach, 50m grid size).

• Flood extent sensitive to channel

friction but insensitive to

floodplain friction.

76

78

80

82

84

86

88

90

92

94

0 20 40 60 80 100 120

Time (hours)

Flo

w (

cu

.m/s

)

ERS-1

Hydrograph at

upstream end of reach

Comparison of performance measures for snake

conditioned on SAR and LiDAR.

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

0 0.02 0.04 0.06 0.08 0.1

Channel friction (Manning's n)

Pe

rfo

rma

nc

e m

ea

su

re

P(t>|t0|)

F measure

Obtaining validation data of urban flood extent

• In urban areas flooding impacts most severe

• Observations of urban flood extents are needed for

validation of modelled flood extents

• In rural areas, 2D models validated using ERS/ASAR – but

these have too low a resolution for urban areas

• But several high resolution SARs recently launched

(TerraSAR-X, RADARSAT-2, ALOS PALSAR and COSMO-

SkyMed)

TerraSAR-X image of the lower Severn flood of July 2007, with

DLR flood extent (blue) overlain (© DLR 2007). The rectangle

covers Tewkesbury.

N

2km

TerraSAR-X image of Tewkesbury flooding on 25th July 2007 showing

urban areas (3m resolution, dark areas are water).

ASAR image of 26th July 2007 (25m resolution).

Aerial photo mosaic of Tewkesbury flooding on 24 July 2007.

LiDAR DSM of Tewkesbury (2m resolution).

Layover (AB) and shadow (CD) in a flooded street between

adjacent buildings.

h1h2

A N B Y C D

O

θ

TerraSAR-X

M

R

DLR SAR End-to-End simulator (SETES )

Regions unseen by TerraSAR-X in LiDAR DSM due to shadow

(satellite looking West).

Regions unseen by TerraSAR-X in LiDAR DSM due to layover

(satellite looking West).

Regions unseen by TerraSAR-X in LiDAR DSM due to combined

shadow and layover (satellite looking West).

TerraSAR-X image of Tewkesbury with flood extent (blue)

predicted by snake superimposed (shadow/layover masked out).

B

A

Flowchart of method of flood detection in urban areas.

Delineation of flood in rural areas using

snake algorithm applied to SAR and

LiDAR data

Supervised classification of urban flood

seed regions

Determination of spatially-varying

urban water height threshold

Seed region growing

Correspondence between TerraSAR-X and aerial photo flood extents

in main urban areas of Tewkesbury, superimposed on LiDAR (yellow =

wet in SAR and aerial photo, red = wet in SAR only, green = wet in aerial photo only).

Near real-time flood extent for flood relief

management

• The Pitt Report concluded that some decision-making was

hampered by lack of information about the flood.

• Need for near real-time visualisation tools to enable the

emergency services to react to and manage fast-moving events,

and target their limited resources at the highest priority areas.

• Useful if near real-time visualisation of the flood extent could

be made available overlayed on map data in a simple GIS.

International Charter for Space and Major Disasters

• Charter has been set up to provide exactly this type of data.

• EA invoked the Charter for the first time in the UK in June

2007.

• EA unable to use the ERS-2 SAR image supplied because not

geometrically corrected.

• TerraSAR-X images can be registered to single pixel accuracy

rapidly and automatically.

TerraSAR-X image of the Cockermouth flood (21/11/09)

Flood extent produced by the DLR algorithm overlayed on

TerraSAR-X image of the Severn flood of July 2007 (© DLR 2007).

Combined algorithm for rural and urban flooding

• DLR algorithm will not work in urban areas due to radar shadow/layover.

• Combine algorithm for rural flooding with that for urban flooding.

• Need to automate a number of steps in urban algorithm.

• Resulting algorithm could only be used where urban areas mapped with

LiDAR.

• Radar shadow/layover calculation done in parallel with processing

TerraSAR-X data.

Possible multi-scale visualisation of flood extent

Rural flood extent (water = blue,

rectangle = Tewksbury)

Urban flood extent (water = yellow)