Land surface temperatur is one of thr important parameters of geogas interaction and energy exhange.In order to obtain high spatial resolution land surface temperature data. The research improved a method of downscaling thermal infrared remote image, and verified it with Shanghai Landsat 8OLI/TIRS image as the data source. The Normalized Difference Vegetation Index (NDVI) was decomposed into low frequency layer and edge layer and detail layers, in which the edge layer and the detail layer are scaled up to the thermal infrared data, and compared with the classical thermal infrared downscaling method DisTrad algorithm and TsHARP algorithm, the simulated surface temperature (LST) ( 270m) LST downscaling (90m) as a downscaled data source.The results show that: 1) All three downscaling methods preserve the spatial characteristics of the original land surface temperature, but the DisTrad algorithm and the TsHARP algorithm increase the detailed information that does not exist in original land surface temperature data.; 2)The improved three-layers decomposition model has a root mean square error of 0.9130k,which is 0.9373k and 0.8320k higher than the DisTrad method and the TsHARP method.