Thermal infrared remote sensing data downscaling investigations: An overview on current status and perspectives

Land surface temperature (LST) retrieved from moderate resolution or downscaled from coarse thermal infrared (TIR) data is one of key environment parameters. Over the last four decades, most advanced remote sensing sensors/systems can acquire TIR data at a low spatial resolution but high temporal resolution. However, per different application purposes, both high spatial and temporal resolution TIR data are needed. Given that many investigations on downscaling LST (DLST) processes have been done and findings have been reported in the literature, it necessitates to have an updated review on DLST investigations of the status, trends, and challenges and to recommend future directions. An overview is provided on various polar orbits and geostationary orbits' satellite TIR sensors/systems and on scaling factors’ determination and selection techniques/methods suitable for DLST processes. Existing various techniques/methods for DLST processes are presented and assessed, and limitations and future research directions are identified and recommended. In this review, several concluding remarks were made, including (1) most investigations on DLST processes used coarse spatial resolution but high temporal resolution MODIS TIR data; (2) compared to fusion-based method, the kernel-driven processes are the most frequently used thermal downscaling methods; (3) machine-learning methods have demonstrated their excellent performance and robustness in improving DLST accuracy; (4) more advanced spatiotemporal fusion-based methods consider synergic powers by combining a kernel-driven process with a fusion-based process method. The three future research directions for DLST processes are recommended: further reducing uncertainties of DLST results, developing novel DLST models and algorithms, and directly reducing the spatial scaling effect in DLST processes.