Flood detection using multitemporal
Radarsat and ERS SAR data
Ping Chen, Soo Chin Liew and Hock Lim
Centre for Remote Imaging, Sensing and Processing,
National University of Singapore,
Lower Kent Ridge Road, Singapore 119260
Tel: (65) 8748029, Fax: (65) 7757717,
E-mail: crschenp@nus.edu.sg
Keywords: flood, ERS SAR, RADARSAT SAR, multi-temporal images
Abstract
A study using combined RADARSAT and ERS SAR images in flood detection is
described in this paper. The availability of SAR data from the RADARSAT and ERS satellites
offers an opportunity for continuous observation of flood events. This makes it possible to
monitor the progress of flood in near real time and to produce accurate, rapid and cost effective
flooding maps. The objective of this study is to monitor flood by using multitemporal and
multiplatform SAR data. Eight RADARSAT (standard mode S6, S4 and S1), five ERS PRI
SAR and five SPOT multispectral optical images were used in this study. The images were
acquired over the Mekong delta in Vietnam from June 1997 to Dec 1997. The SAR images were
first filtered, co-registered and calibrated. Then the inundated areas were extracted from the
SAR images by using thresholding methods. The majority filter was used for post processing of
the flooding maps. The results were compared and evaluated with SPOT multispectral images
serving as the "ground truth". The characteristics of RADARSAT and ERS data for flood
detection and the determination of threshold value are discussed. In this paper, we present the
time series of flooding maps in 1997. The result of this study verified the effectiveness of the
SAR data for flood monitoring.
Introduction
Although the Mekong Delta’s area occupies only 5% of the entire watershed of the Mekong
River, it collects all the water from the Mekong River, one of the big river systems in Asia.
Floods are almost annual events in the Delta. The floods usually occur during the rainy season
from May to December and culminate around October and November. Remote sensing provides
an effective means of obtaining a synoptic view of areas affected by floods for monitoring the
progress of a flooding event. Multitemporal optical images such as the SPOT images would be
able to detect flooded areas. However, the use of optical imagery is often hampered by the
presence of cloud covers during the flood seasons. Synthetic Aperture Radar (SAR) imagery
acquired by the RADARSAT and ERS satellites can be used for flood monitoring due to the
cloud penetrating capability of SAR. Therefore, the use of SAR imagery acquired at appropriate
time would greatly help in the compilation and timely updating of flooding maps.
A flood map of the Mekong Delta during a single date in 1996 was composed using a
RADARSAT Scan SAR imagery (Lai, 1998). In this study, we monitor the flood of 1997 in the
Mekong Delta by using multitemporal and multiplatform SAR data with high time and spatial
resolution. As the RADARSAT and ERS SAR have different characteristics, it is necessary to
evaluate the performance of these data for flood detection.
Study Area and Available Data
The study area is an area about 100 km by 91 km located in the Mekong River Delta. The
location of the study area is shown in Figure 1. The data set consists of eight RADARSAT
(standard mode S1, S4 and S6) images, five descending mode ERS-2 PRI images and five
SPOT multispectral images which were acquired from June 1997 to Dec 1997. The acquisition
dates of imagery used for this study are listed in Table 1.
The SAR images of the RADARSAT and ERS were used for the flood detection, while the
SPOT images were used to evaluate the results. As the SPOT data were acquired in almost same
period as that of SAR data it offers a high degree of reliability for checking the results.
Fig. 1: Location map of the Study Area
Table 1: The data used in this study
| Date | ERS | RADARSAT | SPOT |
| Jun 29 | Ö | | |
| Jul 26 | | S6 | |
| Jul 29 | | S1 | |
| Aug 3 | Ö | | |
| Sep 6 | | | Ö |
| Sep 7 | Ö | | |
| Sep 28 | | S1 | |
| Oct 8 | | S4 | |
Nov 16 | Ö | | |
| Nov 17 | | | Ö |
| Nov 23 | | S6 | Ö |
| Nov 26 | | S1 | Ö |
| Dec 10 | | S4 | |
| Dec 19 | | | Ö |
| Dec 20 | | S1 | |
| Dec 21 | Ö | | |
Data Processing Method
The calibrated SAR Precision Image (PRI) products of ERS and RADARSAT were used in this
study. The images were acquired and processed at the ground station of the Centre for Remote
Imaging, Sensing and Processing (CRISP), Singapore. Each image was first low-pass filtered
using a 5 × 5 averaging window and then downsampled to a pixel size of 50m × 50m. All the
images were co-registered and calibrated. An edge-preserving speckle removal filter based on
the adaptive Wiener filter for multiplicative noise was applied afterwards.
The SAR backscattering coefficient (
s ° ) in dB was computed from the RADARSAT and ERS
SAR images using the equation
where DN is the digital pixel number, K is a calibration constant and
q is the incidence angle.
For RADARSAT scenes processed at CRISP, the calibration constant K has the same value for
all pixels within the same image.
The inundated areas were extracted using the threshold method, and the threshold conditions
were set as follows:
-
For RADARSAT images,s °I£
TR where,s°I is backscattering coefficient of the
image and TR is the threshold value.
- For ERS images, s°d/s°b£
T AND s°b£TE where,s
°d and s°b are the SAR
backscattering coefficient in dB for the data during flood and before flood
respectively.
-
The values of TR and TE were determined from the mean of the selected AOI (area of
interested) or the histograms of the data. The images were post processed with 3 by 3 majority
filter after flood detection.
