Characteristics of Satellite SAR Images in the Damaged Areas Due to the Hyogoken-Nanbu Earthquake
Hisashi Aoki, Masashi Matsuoka and Fumio Yamazaki
Earthquake Disaster Mitigation Research Center, Riken
2465-1 Miliyama, Miki, Hyogo 673-0433, Japan.
Tel: (81)0794-83-6632 Fax : (81)-794-83-6695
E-mail: aoki@miki.riken.go.jp
Abstract
It is expected that clouds and smoke do not greatly affect the application of SAR in identifying the extent of damage after an earthquake. Earth observation satellites installed with SAR observed the Kobe area before and after the 1995 Hyogoken-Nanbu, japan earthquake. The relationship between the earthquake damage and the SAR images taken before and after the earthquake was investigated for the purpose of extracting earchquake damage distribution by satellite SAR images.
Introduction
SAR (Synthetic Aperture Radar) observations can be performed night and day. This feature is useful for an effective assessment of damage at an early stage following a disaster when a field survey for a large area is difficult. For gathering damage information due to natural disasters, several methods exist such as field survey, aerial photographs and television, and satellite optical sensors and SAR. A remote sensing method that uses SAR is expected to provide two-dimensional damage assessment on a large regional scale. Detection of crustal deformation by SAR interferogram (Ozawa et al., 1997) and detection of damaged built-up areas by the earthquake using SAR intensity images (Yonezawa and Takeuchi, 1998) have been reported. However, quantitative evaluations on SAR images are still few.
In this study, we attempt quantitative evaluation on the characteristic of satellite SAR image of the damage area due to the Hyogoken-Nanbu Earthquake. Data used is the satellite SAR images taken before and after the earthquake and the building damage data obtained by a comprehensive field survey. Earth observation satellites installed with SAR observed the Hanshin-Awaji area four months after the earthquake. The field survey data were digitized by building Research Institute (BRI), Ministry of Construction using a geographic information system (GIS). Quantitative assessment on the accuracy of SAR images is carried out using these two data sets on building damage.
Satellite SAR Images
The satellite SAR images used the following analysis are those taken from the ERS-1 on October 12, 1994 (before the earthquake) and May 23, 1995 (after the earthquake, shown in Fig 1). The main characteristics of ERS -1 are listed in Table 1. In each recorded image, one pixel is equivalent to a ground distance of 12.5 m. The image that was obtained in the nearest day after the earthquake is that of four months after. Because of this time difference, a direct comparison between the SAR images and the field survey data is not very accurate. However, we think that basic evaluation is still possible on the potential of the satellite SAR images to detect the changes in the earth's surface of urban areas.
| Altitude | 785km |
| Frequency | 5.3 GHz |
| Wavelength | C- band |
| Depression | 23°at midrange |
| Swath width | 100 km |
| Spatial resolution | 30m |
| polarization | v v |
Table 1 Characteristics of ERS -1
Fig.1 ERS-1/SAR image on 1995/05/23
Building Damage Data by Field Survey
Earthquake damage data used in this study were those on building damage in Kobe and the surrounding areas due to the Hyogoken-Nanbu Earthquake. A compreshensive building damage survey was conducted few weeks after the earthquake and the results of the survey were compiled by the Architectural Institute of Japan (AIJ) and the City Planning Institute of Japan (CPIJ). The compiled data set was digitized by BRI using GIS. In this GIS map, building damage levels were classified into five categories: damage by fire, severe structural damage, moderate damage, slight damage and no damage. The data set contains the total number and the number of damaged buildings per each block of city district. The damage ratios evaluated for each block were used as the reference data.
Analytical Procedure and Satellite SAR Image Data Processing
The procedure of this study is shown in Fig. 2. Each step of the flow is as follows:
Fig.2 The flow of this study
(1) Geometric correction
A geometrically corrected image is generated through a process which uses control points on the ground having high correlation before and after the earthquake. The correlation coefficient is calculated by dividing the covariance of the two images by the dispersion of each image (Remote Sensing Society of Japan, 1992), and is expressed as
In which a and b indicate the backscattering coefficients of pixels before and after the earthquake.
(2) Transforamtion to backscattering
A pixel value of satellite SR images is transformed into the backscattering coefficient. Which is the basic physical value of the quantitative comparison of images at two time instants. The backscattering coefficient is calculated by the following equation (national Space Development Agency of Japan, 1996).
so = 20 log
10 X - 65.3 (2)
(3) Correlation image
The correlation coefficient is used as a characteristic value of satellite SAR images at two different time instants. The correlation image is generated using a pixel value of the average correlation coefficient within a 5 x 5 pixel window. The correlation coefficient is calculated by equation (1)
(4) Difference image
The difference of the backscattering coefficients is also considered as a characteristic value. This difference image is generated using a pixel value of the average difference of backscattering coefficients before and after the earhquake within a 5 x 5 pixel window.
