Geography Reference
In-Depth Information
In another study, Schumann and Di Baldassarre (2010)
illustrated how this fuzzy flood map may be a useful
tool for event-specific flood risk mapping. In this case,
the possibility of inundation of each SAR pixel may be
expressed as a flood hazard level and combined with a
pixel based vulnerability index that is associated with
a given land cover or land use class.
The objective of the work by Mason et al. (2012, in
press) outlined here was to build on a number of aspects
of the existing algorithms to develop an automatic near
real-time algorithm for flood detection in urban and rural
areas. The algorithm assumes that high resolution LiDAR
data are available for at least the urban regions in the
scene, so that a SAR simulator may be run in conjunction
with the LiDAR data to generate maps of radar shadow
and layover in urban areas. It is therefore limited to urban
regions of the globe that have been mapped using LiDAR.
However, in the UK most major urban areas in flood-
plains have now been mapped, and the same is true for
many urban areas in other developed countries.
The algorithm first detects flooding in the rural areas.
It is well-known in image processing that an improved
classification can be achieved by segmenting an image
into regions of homogeneity and then classifying them,
rather than by classifying each pixel independently using
a per-pixel classifier. Following Martinis et al. (2009,
2011), the image is segmented into homogeneous regions
using the multi-resolution segmentation algorithm of the
eCognition Developer software (Definiens AG, 2009).
These regions can then be classified on the basis of
their backscatter, texture, shape and contextual features.
Classification is performed by assigning all segmented
regions with mean SAR backscatter less than a threshold
to the 'flood' class. To determine the threshold, training
regions for 'flood' are automatically selected from regions
giving no return in the LiDAR data (water), and for
'non-flood' from un-shadowed areas well above the flood
level. The classification step classifies the majority of the
flooded rural area correctly. The initial segmentation is
refined using a variety of rules e.g. flood regions having
LiDAR mean heights significantly above the local flood
height are reclassified as non-flood.
A simpler region-growing technique is used in the
urban areas, guided by knowledge of the local waterline
heights in adjacent rural areas. A set of seed pixels having
backscatter less than the threshold, and heights less than
or similar to the adjacent rural waterline heights, is
identified. Seed pixels are clustered together provided
that they are close to other seeds. Regions of shadow and
layover are masked out in the processing.
The algorithm was developed using the data set
acquired for the 1-in-150 year flood of the rivers Severn
and Avon at Tewkesbury, UK, in July 2007. This resulted
in substantial flooding of urban and rural areas, about
1500 homes in Tewkesbury being inundated. A 3 m
resolution TerraSAR-X image was acquired just after the
6.4.2 Near real-timeflooddetection inurban
andrural areasusinghighresolution
space-borneSARimages
The vast majority of a flooded area may be rural rather
than urban, but it is very important to detect the
urban flooding because of the increased risks and costs
associated with it. Flood extent can be detected in rural
floods using SARs such as ERS and ASAR, but these have
too low a resolution (25 m) to detect flooded streets in
urban areas. However, a number of SARs with spatial
resolutions as fine as 1 m have recently been launched
that are potentially capable of detecting urban flooding.
They include TerraSAR-X, RADARSAT-2, and the four
COSMO-SkyMed satellites.
As outlined in Section 6.3.1.1, an automatic near real-
time flood detection algorithm using single-polarisation
TerraSAR-X data has been implemented by Martinis
et al. (2009, 2011). This searches for water as regions
of low SAR backscatter using a region-growing iterated
segmentation/classification approach, and is very effective
at detecting rural floods, but would require modification
to work in urban areas containing radar shadow and
layover.
A semi-automatic algorithm for the detection of flood-
water in urban areas using TerraSAR-X has also been
developed by Mason et al. (2010). It uses a SAR simulator
(Speck et al., 2007) in conjunction with LiDAR data to
estimate regions of the image in which water would not
be visible due to shadow or layover caused by buildings
and taller vegetation. Ground will be in radar shadow if
it is hidden from the radar by an adjacent intervening
building. The shadowed area will appear dark, and may be
misclassified as water even if it is dry. In contrast, an area
of flooded ground in front of the wall of a building viewed
in the range direction may be allocated to the same range
bin as the wall, causing layover which generally results
in a strong return, and a possible misclassification of
flooded ground as un-flooded. The algorithm is aimed
at detecting flood extents for validating an urban flood
inundation model in an offline situation, and requires
user interaction at a number of stages.
Search WWH ::
Custom Search