Geography Reference
In-Depth Information
ERS-SAR (12.5 m)
1
0.8
0.6
ENVISAT-ASAR (75 m)
0.4
0.2
0
(b)
(a)
Figure 6.7
River Dee, UK in December 2006: A) SAR imagery of medium resolution (ERS-SAR, top row) and low resolution
(ENVISAT-ASAR, lower row); B) uncertain flood inundation map generated by combining ten flood extent maps derived from the
two imagery applying five different image processing techniques (Schumann et al., 2009a).
(Figure 6.7). This is quite a unique set of SAR images
as in terms of the inundation process, the true flood
extent at both acquisitions was the same. Given that all
other significant acquisition parameters (e.g. frequency,
polarisation and incidence angle) were the same for both
images all apparent difference and therefore uncertainty
in flood extent mapping was attributed to differences in
spatial resolution and flood area extraction algorithms
(Schumann and Di Baldassarre, 2010). Schumann et al.
(2009a) processed these two images to derive a vari-
ety of flood extent maps by using five different image
processing procedures: visual interpretation, histogram
threshold, active contour modelling, image texture vari-
ance and Euclidean distance. This resulted in ten different
flood extent maps, for which significant differences were
observed. Fusing these different flood maps using equal
weighting, Schumann et al. (2009a) produced a fuzzy map
(Figure 6.7) that expresses for each pixel the possibility
of being inundated according to the multi-algorithm
ensemble generated.
An investigation into the value of such a map for
flood model evaluation revealed that accounting for the
uncertainty in extracting flooded area increases the infor-
mation content and leads to a more identifiable flood
model parameter set than any of the conventional binary
flood maps. Di Baldassarre et al. (2009b) presented a
technique to produce an uncertain flood inundation
map from LISFLOOD-FP (Bates and De Roo, 2000)
model simulations conditioned on the fuzzy SAR flood
map. This calibration exercise clearly demonstrated the
necessity to move from traditional, deterministic binary
(wet/dry) maps to fuzzy or probabilistic maps of flood
extent. Furthermore, it was argued that accounting for
uncertainties in flooded area observations considerably
increases confidence in flood model results and ultimately
flood inundation predictions.
Search WWH ::
Custom Search