Verification of INSAR Capability for Disaster Monitoring - A Case Study on Chi-Chi Earthquake in Taiwan
Shoji Takeuchi¹, Yuzo Suga¹, and Yoshinari Oguro^2
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).(©ESA/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. 

We also attempted to detect the damaged urban areas using SPOT panchromatic images acquired before the earthquake (on Feb.9,1999)and after the earthquake (on Sep.27,1999). However, in the SPOT image, it was quite difficult to interpret the image differences between damaged and non-damaged urban areas. It was also difficult to interpret the changes on spatial patterns because the resolution of SPOT panchromatic data is 10 meters and this resolution is still insufficient for detecting the changes of the shape or the size of buildings by collapsing.

Above results by the intensity and coherence of ERS/SAR verifies that the coherence is superior to the intensity as the parameter for detecting damaged urban areas by building collapse. In addition, these damaged areas are difficult to detect even by comparing two SPOT panchromatic images acquired before and after the earthquake. These facts are considered due to higher sensitivity of the coherence to ground surface changes compared with the intensity of SAR backscatter or reflection of sun light. This high sensitivity is considered to be brought by the fact that the coherence is significantly affected by phase variation and the spatial scale generating phase variation is the order of wavelength of microwave and much smaller than the ground resolution of ERS/SAR or SPOT panchromatic data.

Fig.5.Differential interferograms obtained by ERS/SAR.
(a) Pair-1 (Jan.21 - May 6, 1999).
(b) Pair-2 (May 6 - Sep.23, 1999).
(c) Pair-3 (Jan.21 - Sep.23, 1999).
(d) Phase difference patterns between (b) and (c).

Extraction of Land Displacement by Differential Interferogram
We attempted to generate differential interferograms from the first three pairs indicated in Table 1.Figure 5 (a),(b),and (c)shows all of the obtained differential interferograms. The phase patterns in these interferograms indicate the phase differences due to land displacement in the slant range direction under the assumption that the phase patterns caused by orbit difference (orbital fringes)and topography (topographic fringes)are removed perfectly. For the confirmation of removal of both fringes, the phase patterns (a)by pair-1 is a key result, because this pair does not include the earthquake occurrence and so it is hardly possible to get the phase patterns due to land displacement. In fact, the phase patterns in the interferogram by pair -1 are almost flat, which supports that both fringes are almost completely removed. Another important key is the consistency of the phase patterns between the results from different data combinations. This consistency can be investigated by subtracting the interferogram by pair-2 from that by pair -3.The result is shown in Figure 5 (d).The phase patter ns in the interferogram (d) are also almost f lat and consistent with those by pair-1,which interval corresponds to the difference of the intervals by pair-3 and pair-2.

The investigation described above on all of the obtained intergerograms clearly supports that the differential interferometry by ERS/SAR succeeded to extract the phase patterns relat ed to the land displacement caused by Chi-chi earthquake.owever,the usable phase patterns are only available in flat or semi-flat areas.In mountainous areas,the interferograms are almost noisy and they can not bring any information about land displacement.This result clearly comes from poor coherence by ERS/SAR inter- ferometry in mountainous areas as pointed out in the beginning of this section.Therefore the application of differential interferometry by ERS/SAR is actually limited to the areas where topography is flat or semi-flat and the vegetation cover is relatively less like urban or agricultural areas. 

Fig.4 Results of evaluation of SAR intensity changes (left)and coherence changes (right)due to the earthquake in the sample urban areas in Taichung and Nantou Prefectures (except Changhua).(*denotes severely damaged urban group).


Fig.6.Displacement patterns obtained from the differential interferoframs in Figure 5.
[(a):Pair-1,(b):Pair-2,(c):Pair-3 ]

From the above results, the land displacement patterns due to Chi-chi earthquake are possible to extract from the differential interferograms shown in Figure 5.The amount of displacement in a slant range direction is a half of the wavelength (a half of 5.6 cm for ERS/SAR) for one cycle of phase patterns. The direction of the displacement is toward satellite if the phase values decrease toward satellite, and backward satellite if the phase values increases toward satellite. As all test ERS/SAR data were acquired in a descending orbit, he radar signals were illuminated from the right side of the images. In the interferograms of Figure 5 (b)and (c),as the phase decreases in the right direction, namely toward satellite, the directions of the displacement are all toward satellite if the left-side edge portion (actually the coastal line of the left part of the images)does not move.

Figure 6 (a),(b)and (c)show the displacement patterns obtained from the three differential interferograms in Figure 5 (a),(b)and (c)respectively.As the amount of displacement is only computed in the areas where phase unwrappi ng was succeeded,the displacement patterns were not obtained in almost of mountainous areas where the coherence was very poor and so the displacement fringes could be hardly obtained.In addition,the displacement patterns in Figure 6 were computed as the relative displacement inside the whole areas where phase unwrappi ng was succeeded because the displacement is possible to compute only from the relative changes of phase patterns.

As the displacement values in Figure 6 indicate the displacement in a slant range direction, it is necessary to convert the displacement values to those in a vertical or a horizontal direction, although it is impossible to decide either vertical or horizontal only by the interferograms. The conversion is done by the multiplication of the inverse of cos denotes incidence angle of SAR)for a vertical displacement and sin for a horizontal displacement respectively. As is approximately 23 deg. for ERS/SAR, the multiplication factor is 1.09 for a vertical and 2.56 for a horizontal direction respectively. As the maximum relative displacement value seen in Figure 6 (b)and (c)is about 22 cm in a slant range, the maximum displacement value is about 24 cm for a vertical and about 56 cm for a horizontal direction. As the result of the comparison between the displacement patterns in Figure 6 and those by GPS observation (Nat ional Cheng-Kung Univ.,1999),the displacement patterns by SAR were proved to be fairly compatible to those by GPS in both of direction and amount. 

Detection of Landslide Areas Using Coherence and Intensity
We also attempted to detect landslide areas in mountainous regions caused by the earthquake using the change of coherence and intensity. For the extraction of coherence changes, the coherence images by pair-1 and pair-4 were compared each other. The reason why these two pairs were used was that there might be some possibility for coherence increase in landslide areas after the earthquake because forest vegetation was almost lost due to the landslide. However, the result was that coherence by pair-4 (after the earthquake)was extremely low in mountainous regions and still same as that by pair-1 (before the earthquake).This result indicates some limitation of InSAR by C-band SAR in steep mountainous regions. On the other hand, it was found that the intensity by SAR multi-look images partly decreased in landslide areas after the earthquake. Therefore, some of the landslide areas were possible to be detected using these intensity changes, although the landslide areas in the slopes facing to radar illumination were hardly detected due to a big foreshortening effect in ERS-2/SAR images. This result indicates that backscattering intensity is partly available to detect landslide areas even in mountainous regions. 

Conclusion
As the first result of this study, the coherence information was proved to be an effective parameter to detect damaged urban areas by the earthquake, which has been also verified by another case study for The 1999 Great anshin-Awaji Earthquake (Yonezawa and Takeuchi,1999).The consistency between the two case studies seems quite important and valuable because it clearly enhances that SAR observation in which a good interferometreic condition is maintained is significantly important for disaster monitoring by SAR data.The second result suggested that differential interferometry by ERS/SAR is possibly effective to extract land displacement by an earthquake together with some limitation. In flat or semi-flat areas where main land covers are occupied by urban or agricultural lands, the differential interferogframs obtained by ERS/SAR brought reasonable displacement information caused by the earthquake. The third result suggested an essential limitation of C-band InSAR in steep mountainous regions. The first two successful results clearly enhance the importance of interferometric SAR for disaster or environmental monitoring using space-borne SAR data.

References
Chen A.J.and Chien Ying Wang.2000.Using SPOT I mageri es to Monitor Landslides in Chi-chi earthquake.Pr esented At I nter national Wor kshop 2000 in I T on Gener ation of Advanced Earth Environmental Information.

Kokusai Kogyo Co.,LTD.1999.Record of Great Taiwan Earthquake on September 21,1999 (in Japanese). 

National Cheng-Kung University.1999.Satellite Geoinformatics Research Center,921 Chi-chi earthquake damaged area satellite control points displacement map,http://www.sgrc.ncku.edu.tw,

Yonezawa,C.and S.Takeuchi.1999.Detection of Urban Damage Using Interferometric SAR Decorrelation.Proceeding of IGARSS'99,pp925-927.