Verification of INSAR Capability for Disaster Monitoring - A Case Study on Chi-Chi Earthquake in Taiwan
Shoji Takeuchi1, Yuzo Suga1, and Yoshinari Oguro2
1Professor,
2Assistant-professor
Hiroshima Institute of Technology,
2-1-1,Miyake,Saeki-ku,Hiroshima 731-5193,Japan
Tel &Fax :(81)-82-922-5204
E-mail:
sh-take@cc.it-hiroshima.ac.jp
A. J. Chen
Professor
Center for Space and Remote Sensing Research,
National Central University of Taiwan
Chung-Li,Taiwan
Chinatsu Yonezawa
Researcher
Remote Sensing Technology Center of Japan
Roppongi First Bldg.,1-9-9,Roppongi,Minato-ku
Tokyo 106-0032,Japan
Key Words
ERS/SAR, SAR interferometry, Chi-chi Earthquake, Coherence,Land displacement
Abstract
The authors conducted a verification study on the capability of interferometric SAR
(InSAR) technology for monitoring damages by Chi-chi earthquake occurred on Sep.21,1999 in Central Taiwan by using
ERS-2/SAR data received at Center for Space and Remote Sensing Research (CSRSR). The items for verification are detection of damaged urban areas by building collapses, detection of land-slide areas, and extraction of land displacement patterns caused by the earthquake. We obtained positive results for supporting high capability of InSAR for detecting damaged urban areas and for extracting land displacement patterns in flat or semi-flat areas around Taichung city. On the other hand, for detecting landslide areas, InSAR did not work because of poor coherence of interferograms in mountainous regions by repeat-pass
ERS-2/SAR data pairs, while SAR backscattering intensity by ERS-2/SAR was partly available for detecting land slide areas. Above results verified that InSAR by
ERS/SAR is effectively used for disaster monitoring in urban or agricultural areas with flat or semi-flat topography, although InSAR is difficult to be used practically in steep mountainous areas with dense vegetation.
Introduction
On 21st of September in 1999 at 1:47 a.m. local time, a Ms=7.7 earthquake shook the central area of Taiwan. The epicenter was 160 km south-west of Taipei, the capital of Taiwan and near a small town Chi-Chi. This is the largest earthquake on the Taiwan island in the 20th century, which caused 2470 fatalities, 11,305 injuries, 53,551 buildings totally collapsed, and 53,633 half collapsed. The total capital lost is estimated to be US $:11.8 billions. In addition, as the results of the earthquake, wide spread landslides occurred in Central Taiwan.
Center for Space and Remote Sensing Research (CSRSR), which has been receiving SPOT data and SAR data from ERS-2 and RADARSAT operationally, started the intensive reception of SPOT data just after the earthquake, and analyzed these SPOT data to detect land slide areas as early as possible and to monitor their temporal changes (Chen and Wang,2000).In this study, the authors studied the applicability of another data
source,ERS-2/SAR data received at CSRSR, for monitoring damages or environmental changes caused by the earthquake. Currently the interferometric SAR
(InSAR)technology has been one of the important and effective approaches using SAR data for disaster or environmental monitoring. Therefore, we attempted to conduct three kinds of interferometric analyses, the detection of the damaged urban areas using coherence information, the extraction of land displacement patterns using two-pass differential
interferometry, and the detection of landslide areas in mountainous regions using coherence and intensity.
Test Site, Data and Processing
The test site is located at the central area of Taiwan. Figure 1 shows ERS-2/SAR intensity image of the test site acquired on Sep.23,1999.The left-half areas of the image are almost flat or semi-flat areas including some urban areas, the biggest of them is Taichung City located in the upper part of the image. The right-half areas are rather steep mountainous areas, where big geometric distortions of SAR data are recognized due to foreshortening effect brought by ground height and a small incidence angle of
ERS/SAR. The epicenter is indicated by a cross located in the lower-right part of the image.
Four repeat-pass ERS-2/SAR data acquired on Jan. 21,May 6,Sep.23 and Oct.28 in 1999 were used as the test data. For interferometric processing, four data pairs were used as shown in Table 1.The table also shows nominal values of the perpendicular
baseline component for each pair. The second and third data pair (pair-2 and pair-3)includes the earthquake occurrence between the times of observation, while pair-1 was acquired before the earthquake,pair-4 after the earthquake, and both pairs do not include the earthquake occurrence.
Table 1.Data pairs of ERS/SAR for interferometric analysis
(I: Include the earthquake, N: Not include)
|
Data pair
|
Data combination
|
Baseline
(perp.comp.)
|
I or N
|
|
Pair-1
|
Jan.21 -May 6,1999
|
96 m
|
N
|
|
Pair-2
|
May 6 -Sep.23,1999
|
213 m
|
I
|
|
Pair-3
|
Jan.21 -Sep.23,1999
|
309 m
|
I
|
|
Pair-4
|
Sep.23 -Oct.28,1999
|
224 m
|
N
|
These SAR data were processed from signal data to generate multi-look intensity images, coherence images and differential interferograms using 3dSAR processor developed by Vexcel Corporation in U.S.A. The size for multi-look was 2 range pixels by 10 azimuth lines, which resolution was about 40 by 40 meters on the ground. The coherence images were generated by computing the complex correlation coefficient in a small corresponding patch using the two single-look complex
(SLC) images registered each other as follows;
Coherence =
where C1 and C2 are complex values for the first and the second data,*means complex conjugate and E()means the expectation in the corresponding patch.The size of the corresponding patch was 2 pixels by 10 lines, which was the same as the pixel size after multi-look processing.
Fig.1.ERS-2/SAR intensity image (Sep.23,1999).(ŠESA/ERS 1999)
The differential interferograms were generated by a rather complicated procedure,in which orbital fringes and topographic fringes are removed almost perfectly from the initial interferograms generated by computing phase differences between two SLC data which were co-registered each other precisely.The topographic fringes were removed by subtracting the simulated topographic fringes using a digital elevation model
(DEM)with 100 m by 100 m pixel spacing from the real interferograms.
Detection of Damaged Urban Areas Using Coherence Information
Figure 2 shows an example of the overlaid images of two ERS-2/SAR multi-look intensity images acquired on May 6 and Sep.23,with the color assignment of red for the former and cyan for the latter. The test site is the surrounding area of Taichung City. From this image, it is almost difficult to interpret the intensity changes due to the earthquake occurrence in all urban areas located inside the image. By the report for the survey of urban damages, the urban areas in
Dongshi, Puli and Wufeng located inside the image were damaged severely (Kokusai Kogyo
Co.LTD.,1999).Especially in Dongshi and Puli, more than 50 percent of the buildings in the central urban areas were collapsed.
 |
 |
Figure 3 shows the overlaid image of two coherence images, which were obtained from the data pair-1 and pair-2 respectively. Red color was assigned to the coherence by pair-1 and cyan was assigned to the coherence by pair-2.It is clearly indicated in Figure 3 that the urban areas in
Dongshi, Puli and Wufeng are colored as red, which means that the coherence by pair-2 significantly decreased compared with that by pair-1.On the other hand, in Taichung and Fengyuen there hardly seen red colors inside the urban areas. By the report, the central urban areas in these two cities were not damaged seriously by the earthquake, although point-wise damages were seen in some small parts of those sub-urban areas.
From the two figures described above, the effectiveness of the coherence information compared to the intensity information for damage detection in urban areas is definitely clear. We evaluated the changes of intensity and coherence by the normalized difference of two power data and two coherence data respectively. They are defined as follows;
 |
 |
All the power data were normalized by the maximum power value among all the data in the two dates. The left and right graphs in Figure 4 show the results of the evaluation of the changes in the intensity and the coherence respectively in some sample urban areas in Taichung and Nantou Prefecture. The group from Dongshi to Jungliau in the right side (attached *)is the group for severely damaged urban areas and the group from Tsautuen to Shalu in the left side is the group for non-damaged or slightl y damaged urban areas. For Taichung City, the data of three sample areas were averaged. In the left graph in Figure 4,any significant separation is not seen between the damaged urban group and non-damaged urban group by
Power_ND. On the other hand, in the right graph, two groups are clearly separated by
Coherence_ND.
