The high resolution Terra SAR-X dataset was employed with DIn SAR and persistent scatterer interferometry（PSI） technique for subsidence monitoring in a mountainous area. For DInS AR technique, the generally used SRTM and relief-DEM, which was derived from aerial topographic map, were used to evaluate the influence of external DEM. The results show that SRTM could not fully compensate the complex topography of the research area. The corner reflectors installed during the acquisition of SAR dataset were used to estimate the accuracy of geocoding. The terrain corrected geocoding results based on relief-DEM were much better than using SRTM, with the root mean square error（RMSE） being 6.35 m in X direction and 11.65 m in Y direction（both in UTM projection）, around one pixel of the multilooked intensity image to be geocoded. For PSI technique, the results from time-series analysis of multi-baseline differential interferograms were integrated to restrict only persistent scatterer candidates near the boundary of subsiding area for regression analysis. The results demonstrate that PSI can refine the boundary of subsidence, which could then be used to derive some angular parameters to help people to learn the law of subsidence caused by repeated excavation in this area.
A surface soil moisture model with improved spatial resolution was developed using remotely sensed apparent thermal inertia（ATI）.The model integrates the surface temperature derived from TM/ETM＋ image and the mean surface temperature from MODIS images to improve the spatial resolution of soil temperature difference based on the heat conduction equation,which is necessary to calculate the ATI.Consequently,the spatial resolution of ATI and SMC can be enhanced from 1 km to 120 m（TM） or 60m（ETM＋）.Moreover,the enhanced ATI has a much stronger correlation coefficient（R^2） with SMC（0.789） than the surface reflectance（0.108） or the ATI derived only from MODIS images（0.264）.Based on the regression statistics of the field SMC measurement and enhanced ATI,a linear regression model with an RMS error of 1.90%was found.
