Research On The Generation Of Land Surface Temperature Products With High Spatio-temporal Resolution Based On Multi-source Remote Sensing Data

Land surface temperature(LST)plays a key role in the study of global surface energy balance,hydrological process and climate change.Currently,the land surface temperature products which based on thermal infrared remote sensing can be divided into two categories: one with high spatial resolution and low temporal resolution,represented by polar-orbiting satellites,which can provide detail information and is suitable for small-scale thermal environment research,the other with high temporal resolution and low spatial resolution,such as geostationary satellites,which has short revisit period and can be used for dynamic monitoring of thermal environment.It is difficult for a single thermal infrared sensor to have both high temporal and spatial resolution,which greatly limits the application of surface temperature.As a consequence,the LST downscaling and spatio-temporal fusion methods based on multi-source remote sensing data fusion have become an important means to obtain the land surface temperature products with high spatio-temporal resolution.This study took the middle reaches of Heihe River basin as the research area,and the multisource remote sensing data,such as Himawari-8 AHI,MODIS and Landsat-8 TIRS,were used to retrieve LSTs from Himawari-8 AHI and Landsat-8 TIRS,validate and compare the different methods for LST downscaling over heterogeneous landscape,generate and validate the LST products with high spatio-temporal resolution by spatio-temporal fusion method.In this paper,the overall process of obtaining LST products with high spatio-temporal resolution was achieved from the inversion of the different LST products and the combination of LST downscaling and spatio-temporal fusion,to obtaining 30m/10 minutes high spatio-temporal resolution LST products.To evaluate the LST products with high spatio-temporal resolution,the ground measurements collected from three barren surface sites were collected.The specific research contents and results are as follows:(1)For the inversion of landsat-8 LSTs,ASTER GED dataset was used to derive the bare soil emissivity,and the VCM method was used to estimate the land surface emissivity,which improved the accuracy of landsat-8 LSTs.After that,the split window algorithm was employed to retrieving Himawari-8 LSTs,and the LSTs were verified by the ground measurements.The results showed that the average bias and RMSE of Himawari-8 LSTs at the sites were-1.77 k and 3.74 k,respectively.The Himawari-8 LSTs had high accuracy and can be applied to the subsequent study of LST downscaling and spatio-temporal fusion.(2)For the LST downscaling of Himawari-8 AHI LSTs,TsHARP model,Geographic Weighted Regression model and Random Forest model were introduced,and the LSTs derived from Landsat-8 thermal imageries were used to verify the downscaled LSTs.The results showed that the average RMSEs of three LST downscaling models were 3.76 K,3.56 K,and 3.22 K,respectively.Among them,the Random Forest model had the highest accuracy and stable performance.In the analysis of time series,the accuracy of the downscaled LSTs changed periodically,and the downscaled LSTs performed better in summer and autumn,but not in spring.The error analysis showed that the topographic factors had a greater impact on the LST downscaling.In addition,for different land cover types,the LST downscaling models performed better in bare land,followed by cropland and grassland,and then in impervious surface.Finally,the Random Forest model was selected to downscale the Landsat-8 LSTs from 100 m to 30 m for spatio-temporal fusion.(3)The the BLEnding of Spatiotemporal Temperatures(BLEST)algorithm was used to generate high spatio-temporal resolution LST products.The input data were downscaled Landsat-8 TIRS LSTs,MODIS LSTs,and Himawari-8 AHI LSTs.The spatio-temporal fusion at an annual scale and a diurnal scale were performed,and finally the 30m/10 minutes LSTs were generated.The verification of fusion results showed that the predicted fine-scale LSTs were basically consistent with the daily temperature trend of ground measurements,and the average bias and RMSE at the sites were 1.48 K and 3.19 K,respectively.The predicted LSTs of CJZ station had higher accuracy(Bias ~-0.67 K,RMSE ~ 2.45K),the HZZ and JCHM stations were overestimated.