Flood detection using multitemporal
Radarsat and ERS SAR data
Results and Discussions
Fig. 2 shows part of an ERS SAR image of the test area acquired on June 29, 1997, during the
start of the rainy season. Most of the area has not been inundated. In contrast, a Radarsat SAR
(S4) image of the same area acquired on October 8, 1997 during the peak of the flood is shown
in Fig. 3. The inundated areas (dark areas) are clearly delienated from the non-inundated areas
in the RADARSAT SAR image due to the high incidence angle. The linear criss-crossing
features seen in the image are roads or banks of canals on higher ground.
Fig. 2: ERS image acquired on June 29, 1997 during the
beginning of the raining season.
Fig. 3: Radarsat image (S4) acquired on October 8,
1997 at the peak of the flood. The dark areas are
inundated.
An example of using ERS and RADARSAT SAR images in monitoring the progress of flood is
shown in Fig. 4. This image is a multitemporal colour composite image composed from three
SAR images acquired at three different dates (Red: ERS on Aug 3; Green: ERS on Sep 7; Blue:
RADARSAT S4 on Oct 8, 1997). The dark areas in the image are the areas inundated on all
three dates. The areas in red tones were not inundated on the first date (Aug 3) but inundated in
the Sep and Oct images. The areas in yellow tone were flooded later in October. The cyanish
areas were flooded in Aug, but the water either receded or the water was covered by vegetation
during the later dates.
Fig. 4: Multitemporal colour composite image of the
test area, showing the progress of the flood. (Red: ERS
on Aug 3; Green: ERS on Sep 7; Blue: RADARSAT S4
on Oct 8, 1997).
Fig. 5: Composite of three flood maps corresponding to
the image shown in Fig. 4.
The RADARSAT images show a clear distinction between water and land due to a higher
incidence angle. Therefore, it is easy to detect the flooded area by simply applying a threshold.
In the ERS images, the inundated areas were extracted from the change in the SAR backscatter.
The Jun 29 ERS image is taken as the before-flood reference image. For the later images
acquired during the flood period (July to December 1997), a pixel is classified as an inundated
pixel if its backscatter decreases by 3 dB or more when compared to the corresponding pixel in
the reference image. A flood coverage map can thus be obtained for each ERS or RADARSAT
SAR image. Fig. 5 shows the composite flood map composed from three flood coverage maps
derived from the three SAR images used to compose the colour composite image in Fig. 4. The
colours show the areas flooded at different time. The colours have been assigned so that they
correspond to the colours appearing in the colour composite SAR image of Fig. 4.
Fig. 6 is a multitemporal colour composite image composed from three SAR images acquired at
three different dates (Red: ERS Nov 16; Green: RADARSAT S6 Nov 23; Blue: ERS Dec 21,
1997) during the later part of the flood season. The corresponding flood coverage map is shown
in Fig. 7. In this figure, The dark areas still remained inundated by Dec 21, but the flood has
receded in most areas (coloured blue, cyan and magenta) by Dec 21.
Fig. 6: Multitemporal colour composite image of the
test area, showing the recession of the flood. (Red: ERS
on Nov 16; Green: RADARSAT S6 on Nov 23; Blue:
ERS on Dec 21, 1997).
Fig. 7: Composite of three flood maps corresponding to
the image shown in Fig. 6.
Conclusion
In this study, ERS and RADARSAT SAR images have been used to map the flooded areas in a
part of the Mekong River Delta from Jun to Dec 1997. Flood coverage maps were produced and
the evolution of the flood event could be studied. The results of this study illustrated the
effectiveness of using SAR data for flood monitoring. The main problem anticipated in
operational application of space-borne SAR data is the availability of data for near time
monitoring. ERS has a repeat cycle of 35 days and by itself, does not have sufficient temporal
coverage. RADARSAT is able to image at several incidence angles, and hence has an increased
revisit capability of just a few days. At higher incidence angles, inundated areas are better
discriminated from non-inundated land. Combined use of ERS and RADARSAT data will
provide high temporal revisit capability for flood monitoring.
References
- Lai Anh Khoi & Bui Doan Trong, 1998. Application of Radar Imagery for Flood Monitoring in
the Mekong Delta, Proc. the Euro-Asia Space Week on Cooperation in Space, pp. 219-224
