Terrain-cover classification by integration of SPOT and ERS-1 SAR images over Taiwan

This paper is aimed at the integration of multispectral high resolution visible SPOT image and ERS-1 C-band SAR image data for classification of terrain-cover. The test site was located at central west of Taiwan, where seven terrain covers, from water body to vegetation, were identified. As the first part of this study, the authors apply a simple model to compensate for the elevation effect in the ERS-1 SAR image using the existing DTM data. The geometrically rectified SAR image is then overlaid with the orthoimage of SPOT. The rms error of this overlay produces approximately one pixel or about 12 m which is acceptable for this study. The second part of the study deals with the land cover classification. In particular, fractal image extracted from SAR data is added. So both spectral and spatial information are used simultaneously for classification. Supervised classification was followed to discriminate the terrain-cover from combined images. Extensive ground truth collection along with available base map were used to aid the evaluation of classification accuracy. It was found that when combined them together better classification accuracy can be obtained.