Methodology Development For Retrieval Of All-weather Land Surface Temperature At High Spatial Resolution

Land surface temperature (LST) is an important physical parameter in the surface-atmosphere interactions. Timely knowledge of spatio-temporal distribution information of LST at regional and global scales has an important significant on the studies of surface-atmosphere energy balance and ecosystem. Satellite remote sensing is the best mean for high-efficient acquisition of LST at regional and global scales. Using polar-orbiting and geostationary satellite data as data source, this thesis aims to develop methods for retrieval of all-weather LST at high spatial resolution, for MODIS LST downscaling based on geographically weighted regression (GWR), and for cross-satellite comparison of MSG-SEVIRI and Terra/Aqua-MODIS LST products. Several main conclusions are obtained:(1) Combining the advantages of LST retrieval from thermal infrared data and passive microwave data, we built an effective model to merge passive microwave cloudy LST and thermal infrared clear-sky LST, and then developed a method for retrieval of all-weather LST at high spatial resolution. Compared with original MODIS LST, merged LST is spatially continuous and temporally reflects inter-annual variation on LST.(2) We built a stationary regression relationship between LST and normalized difference vegetation index (NDVI) as well as digital elevation model (DEM) using geographically varying coefficients, and developed a method for MODIS LST downscaling based on GWR. Compared with ASTER LST product, the performance of the GWR-based LST downscaling method is better than that of the UniTrad and TsHARP methods proposed by the previous studies. The mean absolute error (MAE) and root mean square error (RMSE) of the GWR-based LST downscaling method are approximately 2.3 K and 3.1 K, respectively.(3) By selecting spatial, temporal, and angular consistent pixels as matched pixel pairs, we analyzed the discrepancies between SEVIRI and MODIS LST products over different seasons, time of day, and land cover types. The results show that a significantly seasonal variability in the daytime LST discrepancies is found. Compared with daytime LST discrepancies, night-time LST discrepancies are less dependent on season. Surface types have significant impacts on the daytime LST biases. Compared with the bias during the day, the bias during nighttime is relatively smaller.