Spatial Downscaling of MODIS Land Surface Temperatures Using Geographically Weighted Regression: Case Study in Northern China

Land surface temperatures (LSTs) at high spatial resolution are crucial for hydrological, meteorological, and ecological studies. Downscaling LSTs from coarse resolution to finer resolution is an alternative way to obtain LSTs at high spatial resolution. In this paper, we proposed a new algorithm based on geographically weighted regression (GWR) to downscale Moderate Resolution Imaging Spectroradiometer LST data from 990 to 90 m. Unlike previous LST downscaling algorithms, this algorithm built the nonstationary relationship between LST and other environmental factors (including the normalized difference vegetation index and a digital elevation model) using geographically varying regression coefficients. The uncertainty in this algorithm was evaluated with a sensitivity analysis. The results show that the total uncertainty in this algorithm is less than 2 K. The performance of the GWR-based algorithm was assessed using concurrent ASTER LST data as a reference LST data set. Moreover, this algorithm was compared against the TsHARP algorithm, which was widely used for LST downscaling. The results indicate that the GWR-based algorithm outperforms the TsHARP algorithm in terms of statistical results. The root mean square error (mean absolute error) value decreases from 3.6 K (2.7 K) for the TsHARP algorithm to 3.1 K (2.3 K) for the GWR-based algorithm.