In this paper, four European Radar Satellite (ERS-2) images acquired on January 21, May 6, September 23, and October 28, 1999, respectively, in mid-Taiwan are used to test the quality of a digital displacement model (DDM) determined by three pass method, a technology of differential interferometric synthetic aperture radar (D-InSAR). The terrain surface has an abrupt change during the event of Chi-Chi earthquake. Some basic theoretic aspects are firstly analyzed. It verifies that D-InSAR can provide a DDM with an accuracy of a few millimeters, if InSAR can create a digital elevation model (DEM) with an accuracy of a few meters. In general, the DDM show a vertical displacement vector field similar to the vertical displacement map determined by GPS. Inside the DDM, there are some GPS check points. They show that the mean difference of the displacement values estimated by satellite D-InSAR and the ones determined by GPS is about 6cm. They indicate that three-pass method seems to have a potential for detecting height surface deformation in plain area in Taiwan, if useful ERS data are processed properly. On the other hand, D-InSAR might fail to determine an accurate DDM in fluctuant areas with dense and high vegetation since the coherence is low there.

An ideal vegetation index should depend upon vegetation only and independent to soil type, soil humidity and solar zenith angle. Information of vegetation growth may therefore be deduced from such a vegetation index. In this paper, linear mixture model is used to study the soil effect to different vegetation indices. It indicates that NDVI is greatly affected by soil effect than that of SAVI, similar to the finding by Huete (1988). SAVI can also tolerate maximum noise up to 29.0% after 1000 times of noise test. Results demonstrate that SAVI is more sensitive to vegetation growth than NDVI under the conditions of higher soil interference.

The focus of this investigation is the geometric correction for the ROCSAT-2 high resolution satel­lite images. We correct satellite images using digital surface model (DSM) that includes building and terrain information. Unlike frame photography, where all pixels in the image are exposed simultaneously, each line of a ROCSAT-2 image is collected in a pushbroom fashion at a different instant of time. Due to unavailability of test data, we need to simulate a high resolution satellite image which considers that the terrain and the building may be hidden each other test. First of all, we developed a satellite-orbit-model for ROCSAT-2, including coordinate-transformation-model, satellite-fly-model, simulation-model and on-board-data-model, and then using a DSM to simu­late a high resolution satellite image by Bottom-up approach. In addition, the bases of the simulated image are orthorectified aerial images to minimize differences between the simulate image and the true image. The gray values for the walls of each building are assumed homogeneous. Once imaging simulation is completed, we use a satellite-orbit-model of ROCSAT-2 and DSM to orthorectify the simulated ROCSAT-2 satellite's images by Bottom-up method.

In this paper we describe and study the SAR (=synthetic aperture radar) processing steps adopted in our preliminary tests for generating a digital elevation model (DEM) in Taiwan by means of satellite SAR interferometry. We used two modules FOCUS and PHASE of the commercial software system "Vexcel 3dSAR Processing System" and ERS Tandem mode data to generate a DEM by the interferometric syn­thetic aperture radar (InSAR) approach. The generated DEM is compared with the high precision reference DEM determined by aerial photogrammetry with an average flight height of about 1600m in the same area of about 42km x 25km in Taipei. Altogether 84 ground control points are adopted. They show a mean and RMS (=root mean square) value of height differences of -0.11m and 4.71m, respectively. Moreover, there are altogether 93850 common height points in an area of about 12km x 10km. They indicate a RMS value of height differences of 23.73 meter. About 86% of all these points has the absolute height difference |dH| <20m. Comparing the distribution of these points with the aerial ortho image shows the correlation between the height difference and the land cover type. The absolute height difference is larger in the area of dense buildings, about 5-20 meter. In the near bare area of no buildings and less vegetation, the absolute height difference is smaller and less than 5 meter.

The Alishan natural reserve area, abbreviated as ANRA, locates in the middle-southern part of Taiwan. It has been choosing to be the study site for implementing serial remote sensing researches that focus on the long-term development monitoring of forest reserve area. This study is an essential one of the serial researches. It aims to mapping the distribution of vegetation types to supporting the research needs of remote sensing. After review the relative literatures and onsite checking survey, we made a photo interpretation key for mapping the forest types of ANRA. A digital elevation model of ANRA was also produced in this study by the means of digital photogrammetry for applying to analyze the terrain of the study site. The results show that the mean of the elevation and slope of ANRA is 2,110 meters and 23.4 degrees respectively, and there are about 50% of the site area whose slope is greater than the slope mean of the site. The terrain relief of ANRA goes up from the lower northwest to the higher southeast and covers with many folds around the whole site. The dominant species of ANRA are Chamaechyparis spp. Tsuga chinensis, Pinus spp., Cryptomeria japonica and Taiwania cryptomerioides, and other hardwood species. All of these species are mainly distributed in mixed types and rarely existed in pure types. There are 16 forest types that could be interpreted from the aerial photographs. And among all of the forest types, the HP (Hinoki-Pinus), HPH (Hinoki-Pinus-Hardwood), HTP (Hinoki-Tsuga-Pinus), and HTPH (Hinoki-Tsuga-Pinus-Hardwood) are the superior ones and they inhabit totally more than 92% of the site area. The richness of forest types (RFT) of the ANRA is significantly different at varying situations of the elevation, aspect and slope factors.