Research On All-weather Reconstruction Of Near-surface Temperature Derived From Remote Sensin

Near-surface T_a is an essential parameter to measure thermal environments,and the study of its spatiotemporal variability is significant for various research,including climate change,urban heat islands,forest fires,and more.Thermal infrared remote sensing provides near-surface T_a information at relatively high spatial and temporal resolutions.Unfortunately,thermal infrared remote sensing sensors can only obtain land surface information under clear sky conditions,resulting in data gaps over cloud-covered areas.These data gaps reduce the spatial and temporal integrity of remotely sensed T_a,thus highly restricting the applicability of TIR-derived T_a.However.to date,only a few studies have explored the production of seamless remotely sensed T_a products,and all of them estimated T_a from the gap-filled LST without introducing the spatial and temporal information from remotely sensed T_a and other auxiliary data.This study estimates the daily minimum,average and maximum T_a of clear-sky pixels from remote sensing data over the Yangtze River Delta,China,from 2016 to 2020.The estimation results show the random forest model incorporating daytime Terra/MODIS LSTs and other environmental variables performs the best,with the overall R~2,bias,and MAE values for the minimum,average,and maximum T_a ranging from 0.97～0.99,-0.01°C～0.01°C,and0.85°C～1.17°C,respectively.Based on the remotely sensed T_a under clear sky conditions,five gap-filling methods,including spatial,temporal,spatiotemporal,and two multisource fusion-based gap-filling methods,are then applied to fill the data gaps in the remotely sensed T_adata.The performances of these methods are validated against the station-observed T_a and assessed from multiple aspects,such as filling percentages and sensitivity to cloud conditions.The validation results indicate that the multisource fusion-based gap-filling methods are superior to the other methods and that the temporal Fourier analysis(TFA)method provides the highest accuracies,with R~2ranging from 0.93 to 0.96 and MAEs ranging from 1.56°C to 2.39°C.Moreover,the TFA method also shows good robustness,producing desirable results under various cloud cover and terrain conditions.Finally,this study applies the TFA method to create an all-weather daily minimum,average,and maximum air temperature data over the Yangtze River Delta from 2016to 2020,providing a valuable reference for selecting the appropriate method to develop seamless remotely sensed T_a products,as well as a base data for various related fields.