Study On Methods For Downscaling Satellite Land Surface Temperature

The scale mismatch between different satellite remote sensing images has brought many difficulties to the scientific research. At present, many scholars have put forward a lot of down scale method for land surface temperature(LST).According to the relationships between the LST and the physical parameters, this paper uses benchmark model of DisTrad and TsHARP algorithm and improved model of Land cover classification and moving window to achieve downscaling of ASTER and AMSR2 products. And Combine with the thermal airborne Spectrographic imager(TASI) aviation data to simulate the different scales of LST data and land surface parameters by gradient upward, through the analysis of the change on the statistical relationship between the LST and the related parameters under different scales, this paper quantifys the error caused by the scale effect.The results show that in the study on downscaling from Aster LST, the downscaling method based on moving window model achieves better results In the early stage of crop planting and after harvest; while downscaling method based on land cover classification achieves better results in crop growth stage. These two methods have a great help to improve the quality of ASTER LST products, and the obtained LST products provide good temperature data for estimating the evapotranspiration of small scale farmland. In the study on downscaling from AMSR2 LST, a multi-factor model based on DisTrad is improved, and a good result is obtained by using many kinds of surface parameters to describe LST.The influence of scale effect on the downscaling results is larger in Initial and the early planting of vegetation, it becomes smaller in the stage of vegetation growth, and it begins to increase after the harvest. The scale effect is small in the vegetation coverage area. It shows that the scale effect can effectively reduce the error caused by the LST downscaling results when the land cover is complicated, while under uniform surface, different scales has little effect on downscaling results. In the global model and land cover classification model, the errors caused by the scale effect are in the range of 0.5K~1.5K, and the error of moving window model is in the range of-3K~3K.