We develop the ortho-rectification model to correct the airborne SAR slant-range imagery. This model include estimation of flight parameters and ortho-rectification procedure. We estimated the flight parameters with the ground control points, both the slant range spacing and azimuth spacing are known. We rectified the SAR imagery with a given digital elevation model and the estimated flight parameters. Here we have considered the effects of layover and shadow, we checked the variation of slant range and look angle along the direction of slant range. The tested data was taken over a mountainous area. The elevation within this area are about 350 m. We rectified the airborne SAR imagery with the digital elevation model data and 30 control points. The root mean square errors of slant range for ground control points and check points is about 6m. The resulting ground error is about 10m (1.5 pixels of slant range imagery) even in the mountainous area. The Paper is concluded by remarking on how to produce ortho SAR images which exhibit the phenomena of layover and shadow.

This study applies the technique and theory of satellite microwave remote sensing to determine the sea surface wind speed. The seas surrounding Taiwan will be an object of study. The microwave measurements (brightness temperature) were collected by the passive microwave instrument SSM/I (Special Sensor Microwave/Imager). The linear algorithm developed by Schluessel and Luthardt (1991) and the iterative algorithm developed by Wentz (1992) were used to estimate sea surface wind speed of local seas. The goals of this study are expected to supply a way to obtain sea-weather measurements that different from the traditional technique for Taiwan.

Two fractal-based compression procedures are investigated and compared with the conventional JPEG procedure for remotely sensed images. All three algorithms are applied to two panchromatic subscenes from SPOT. The performances are evaluated with the indices including entropy, correlation coefficient, root mean square error, pixel value differences, and signal noise ratio, which are computed from the differences between the compressed and the original images. The results indicate that fractal compression methods provide higher potential compression ratio. In general, when the compression ratio is larger than 50, fractal-based approaches have better performance than JPEG. Regarding feature analysis, edges of large and homogeneous areas can be recognized easier..

The overall classification accuracy of both Landsat MSS and TM imagery data were more than 80 percent in this study. These results are acceptable for a large landscape area. It also showed that remote sensing techniques has an important potential value in landscape change monitoring. Since the landscape development type of urban and rural area is very different, they should be analyzed separately to extract a reasonable and correct information. The overall accuracy of the classification of tree-cover area and cultivated land and bared land is about 80 percent, and the predicted tree-cover area is very accurate with 95 percent accuracy. The detected accuracy of crown density is more than 65 percent. The change detection accuracy of two thematic images is almost equal to the multiplication of their overall classi-fication accuracy. It implied that an error expansion effect was existed. It is the except case for detecting the change of crown density growth. Whose accuracy was more than 71 percent. The beginning time of ecological dynamic balance of classes' development predicted by MSS detected data and surveyed data is almost same. All the predicted devel­opment trends can retain the real cases. The social-economic variables, inverse population density, industrial structure ratio, and road density, have a very significant and strong cause-effect relationship on the land­scape developments. All regression models developed by these factors can explain more than 90 percent variance of area percentage of every landscape class. The predicted effi­ciencies of these ridge regression models were very stable and could retain on the level of models. This encourages us to combine the multispectral scanning imagery data and ridge regression models to monitoring and predicting the landscape developments in the future.