International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B7. Istanbul 2004
2. THE STUDY SITE AND THE AVAILABLE DATA
3. METHODS
OD
A flood prone area of the river Alzette, approximately 10 km in 3.1 Flood boundary delineation using radar imagery
length, is selected as test site for this research. The selected
river reach with its surrounding villages has been subject to Due to the specular backscattering on plain water surfaces and
severe flooding in the past. A well documented medium sized the resulting low signal return, the flood mapping through SAR
flood event in January 2003, with an estimated return period of is quite straightforward. However, in the transient shallow
ol.lu 5 years, was used in this study. The available database water zone between the flooded and the non-flooded part of the
413, comprises pre-flood and flood SAR images, continuous floodplain (with protruding vegetation producing increased
ters in
discharge measurements upstream and downstream of the river
reach, surveyed high water marks and GPS control points of the
maximum flood extent. A set of photographs taken during the
flooding event is also available. Acquired during the rising limb
and at the peak discharge respectively, two ERS-2 SAR and
Envisat ASAR scenes cover the flooded area at two distinct
stages of the event (Figure 1). Except for two markedly wide
alluvial plains upstream and downstream of Luxembourg-city
(one of these figures as test site in this study), the Alzette river
flows through a narrow valley system, making it impossible to
use earth observation data to delineate the flood extension
conveniently.
100 |
signal returns), the radar signal only gradually increases, thus
making the accurate delineation of the flood boundary more
difficult. Hence, a rather arbitrary choice of a backscattering
threshold value is needed during the automatic binary
segmentation of flooded/unflooded areas of the radar scene.
Because of the poorly defined flood boundaries it is therefore
highly recommended to use a fuzzy rule based calibration
technique as a general approach to hydraulic modelling with
uncertain data derived from earth observation data. Wind
roughening is also a well known effect that eventually has to be
taken into account. Despite these limitations, the use of SAR
imagery compares favourably with other remote sensing
systems (Biggin and Blyth 1996). In a case study Horritt et al.
search | goi (2001) found that, despite the aforementioned limitations, the
model 2 5 3 3 : SAR segmentation algorithm classified 70 % of the SAR
round s s 8 3 : shoreline within 20 m of the shoreline as derived from aerial
es ina ni S T S £ photographic data.
rg are > | / t
|. The : | Because of the lack of standardised, reproducible methods (De
5. Due £ wl Roo et al., 1999), a rather simplistic threshold approach was
shold A used in this study to obtain the inundation maps from the two
y-rule d available radar scenes. Therefore, profiles of pixel values at
». The | several cross sections of the river floodplain are drawn and
round S | confronted with the GPS control points of the maximum lateral
bining 2: TRY MII a o OU 92 flood extent. Thus, threshold values of radar backscattering are
flood Time (hours) determined and'used to classify the radar image as "flooded"
ul for Figure 1. Upstream hydrograph for the 2003 flood event. The qnd «nen flooded? Texpectively, AS ihi Fw classification
n satellite overpass times are indicated. The high method dogs not provide entirely Sutistyins results ne
water marks correspond to the peak flow. probability classes were defined reflecting the lack of
knowledge about the “real” flood extent in the whole area
almost (Figures 2 3). At each cross section the coordinates of the
asures 21 Pre-processing extension with high, intermediate and low probability of
only | The SAR instrument on board of the ERS-2 satellite is a C band ~~ flooding are defined. Therefore, in a pext section, a fuzzy
error | (5.3 GHz) radar, operating in VV polarization with a spatial performance measure will be defined that at each cross section
model | resolution of 30 m and a pixel size of 12.5 m (ESA, 1992). The reflects the uncertainties in the maximüm flood extension as
à) and | Envisat satellite offers the ASAR instrument that enables to derived from the radar data.
ess [ul | acquire datà with alternate polarisation (AP). The combination
event. | of like- and cross-polarisations provides increased capabilities Ad
ud | for flood mapping (Henry et al., 2003). In this study, the AP gn fl TT RE rem
| SAR data were acquired with a VV and VH polarisation
ii € combination. Among the three available radar scenes, the VH
M polarisation in the AP precision image allowed the best )
tal differentiation of “flooded” and “non-flooded” areas. 3
The radar data are first radiometrically calibrated and then E
ed tlie orthorectified using a T5 m resolution DEM. The internal
n (see geometric coherence is evaluated at 1-2 pixels, i.e. the
i (his horizontal shift between any two images never exceeds this
SOS. value. The calibration procedure used in this study to calculate ;
sad tn the backscattering coefficient is based on Laur et al. (1998). 0 200 LR a 600 800
model Speckle noise is reduced using the Frost filtering with a 5 x 5
moving average filter.
Figure 2. Spectral profile across a river transect
| The resulting classified image (Figure 3) shows several flooded
| areas that are not directly connected to the riverbed. These
| patterns can be explained by surface runoff, groundwater
| Y
| 353
