Keywords: change detection, image rectification, kriging, remote sensing
Abstract: Landuse change detection using remotely sensed images has been widely investigated. Most applications of this type involve either image differencing or image classification using multitemporal images. If multitemporal images are to be used for quantitative analysis based on their radiometric information, as in the case of change detection or landuse classification, geometric rectification and radiometric correction must be performed priori to subsequent image analyses. In particular, geometric rectification has significant effect on the accuracy of landuse change detection in areas of rugged terrain. Remote sensing image rectification is commonly done by applying a polynomial trend-mapping (PTM) model to image coordinates and map coordinates of ground control points. A major drawback of the PTM model is that it does not capture the random characteristics of terrain elevation. In this study a new anisotropic spatial modeling approach of image-to-image registration is applied to identify landslides induced by the ferocious Chi-Chi earthquake. The approach considers residuals of the PTM model as anisotropic random fields, and employs the ordinary kriging method for spatial interpolation of the residual random fields. By means of a cross validation procedure and visual check, we found that high accuracy of image-to-image registration was achieved. Band-ratioing technique was also employed for relative radiometric normalization. From the gray-level histograms of pre- and post-earthquake band-ration images, we determined the areal percentage of landuse changes in the study area. Image differencing was then performed using the pre- and post-earthquake band-ratio image pair. Finally, a gray-level threshold of the band-ratio-difference image is assigned as the value its exceeding probability equals the areal percentage of landuse change. DTM data of the study area was also used to further restrict landslide areas to steep slope areas.
1. Introduction
Change detection has been one of the major applications of remote sensing since 1960s. There are many approaches to change detection including post-classification comparison, temporal image differencing, temporal image ratioing, multifractal analysis, Bayesian probabilistic method, etc. Despite of their differences in change identification algorithms, accurate spatial registration of the various dates of imagery is a requirement for all these methods. There are essentially two different categories of image rectification approaches, the deterministic and the statistical approaches. The deterministic approach relies on data of the flight parameters and the terrain information, and is effective when types of distortion are well characterized (Richards, 1995). The statistical approach, by means of a ground-control-points (GCP) data set, establishes mathematical relationship between image coordinates and their corresponding map coordinates using standard statistical procedures. The most widely used method in this category is the polynomial trend mapping (PTM) technique that employs polynomial regression equations to relate image coordinates and their corresponding map coordinates. Although commonly applied, the PTM technique often yields significant registration errors in mountainous or rugged terrain areas. Cheng et al. (2000) proposed an anisotropic spatial modeling (ASM) approach using ordinary kriging estimation for image rectification. The approach takes into account the spatial variation structure of terrain elevation, and yields zero registration error at GCPs. For the purpose of identifying terrain changes such as landslides induced by the ferocious Chi-Chi earthquake, the ASM approach is adopted in this study.
2. Study Area and Data
The Experimental Forest of National Taiwan University (EF-NTU) locates in central Taiwan and covers a total area of 32,781 hectares. The area is very close to the center of the Chi-Chi earthquake, occurred in September 21, 1999, and suffered severe damages. Two SPOT multispectral images, acquired on 01/04/1999 and 01/10/1999, are used in this study.
