A Stepwise Downscaling Method for Generating High-Resolution Land Surface Temperature From AMSR-E Data

A stepwise downscaling method is proposed for generating high-resolution land surface temperature (LST) from advanced microwave scanning radiometer for the Earth observing system (AMSR-E) data to benefit the fusion of thermal infrared and microwave data for high-quality all-weather LST. This method sets a series of intermediate resolution levels between the initial (0.25 degrees) and target (0.01 degrees) resolutions, then downscales AMSR-E LST from one resolution to the next one step at a time, starting from 0.25 degrees and ending with 0.01 degrees. The geographically weighted regression model is adopted in each step to construct the relationship between LST and environmental variables, including normalized differential vegetation index, elevation, and slope. The stepwise method is verified over three regions in China that represent different characteristics of landscape heterogeneity varying from the highest to the lowest: the Yunnan-Guizhou Plateau (YGP), the border of Shanxi Province and Henan Province (BSH), and the central part of Inner Mongolia (CIM). Verified using the emulated AMSR-E LST resampled from reference MODIS LST available in 2010, the results show that the proportions of dates when the stepwise method is better are 100%, 78.1%, and 51.5% in the YGP, BSH, and CIM regions, respectively, which means the stepwise method has an advantage over the direct method in the regions with high heterogeneity. For real AMSR-E LST, the downscaled LST exhibits a similar spatial pattern to that of emulated data but suffers from reduced accuracy and contrast, which is caused by the smooth spatial pattern and low accuracy of the real AMSR-E LST.