A new land surface temperature fusion strategy based on cumulative distribution function matching and multiresolution Kalman filtering

摘要

Fusing high-level passive microwave (PMW) LST and thermal infrared (TIR) LST products is a promising means of generating high-quality, all-weather land surface temperature (LST) data. In this paper, we propose a new fusion strategy for generating high-quality, all-weather LST data based on cumulative distribution function (CDF) matching and multiresolution Kalman filtering (MKF). CDF matching was employed to improve the quality of the high-resolution (0.01 degrees) LST data downscaled with the retrieved coarse-resolution (0.1 degrees) PMW LST from the Advanced Microwave Scanning Radiometer 2 (AMSR2) brightness temperature (BT) data set. The MKF approach was employed to produce high-quality, all-weather LST data by fusing the retrieved coarse-resolution AMSR2 LST and high-resolution Moderate Resolution Imaging Spectroradiometer (MODIS) LST data sets filled with CDFimproved LST data. The spatial patterns in the high-resolution and coarse-resolution LST data were consistent after MKF fusion. The accuracy of coarse-resolution LST data was considerably improved. The biases, root mean square errors (RMSEs) and determination coefficients (R-2) were 0.79-1.47 K, 3.0-3.82 K, 0.88-0.96, respectively. The RMSE of the fused coarse-resolution LST data decreased by 1.51 K/1.82 K and 1.81 K/1.39 K in daytime/nighttime in the Tibetan Plateau (TP) and Heihe River Basin (HRB) verification regions compared to the RMSE of the retrieved AMSR2 LST data. The accuracy of high-resolution LST data remained unchanged, but the spatial coverage was complete with exceptions at the orbital gaps. The bias, RMSE and R-2 values of high-resolution LST data were - 0.91- -0.81 K, 3.02-3.32 K, and 0.93-0.97 under all-sky conditions in the TP and HRB verification regions, respectively. Compared to methods in the existing LST fusion studies, the proposed strategy has advantages in terms of LST data quality and validation accuracy. When both coarse-resolution and high-resolution LST data are missing, the spatial coverages of fused all-weather LST data are incomplete. We attempted to fill the vacancies in PMW LST data with the DINEOF method in advance and subsequently obtained spatially complete all-weather LST data.