Methods For Generating All-Weather Land Surface Temperature With High Temporal Resolution

Land surface temperature(LST)is an essential input for global climate change studies,ecological monitoring and assessment,hydrological process simulation,and surface radiation balance and energy budget modeling.With the increasing global environmental problems and the implementation of the carbon neutrality plan,the importance of LST is rising.Satellite remote sensing technology has become the primary means to obtain LST at the regional or global scale.Thermal infrared(TIR)LST has long been widely used in academia because of its higher accuracy and spatial resolution compared with passive microwave remote sensing.Due to the limitation of wavelength,for LST retrieval,the TIR radiation is not able to penetrate the clouds,resulting in large spatial missing under the clouds.Contrary to this limitation,academics often need to obtain LST with cloud-independent and high temporal resolution as the relevant geological study continues to advance.There are few estimation methods for high temporal resolution all-weather(AW)LST,and the existing methods do not consider its high-frequency characteristics in intra-day and the serious lack of thermal infrared data.In view of this,the following main work is carried out in this dissertation with the main line of obtaining high temporal resolution AW LST based on domestic autonomous data.Firstly,a historical reconstruction method of hourly AW LST was proposed;because of the low timeliness of the historical reconstruction results,a near real-time estimation method of the hourly AW LST was proposed;finally,given the low spatial resolution of the obtained hourly AW LST,an LST downscaling method considering the weight of the descriptors was proposed.First,a method is proposed to reconstruct the hourly AW LST by integrating Chinese reanalysis data and TIR data from Geostationary Satellites(RTG),which can improve the temporal resolution of AW LST based on remote sensing data.The LST is decomposed into a Normal Component(NC)and an Abnormal Component(AC).After estimating and optimizing the AW NC and AC,the two components are summed to obtain the hourly AW LST.Then,taking the Tibetan Plateau(TP),which is a focus of climate change as a case,RTG was applied to the Chinese Fengyun-4A(FY-4A)TIR LST and China Land Surface Data Assimilation System(CLDAS)data.Validation based on the in-situ LST shows that the accuracy of the AW LST is better than the FY-4A LST and CLDAS LST under clearsky,cloudy-sky,and all-weather conditions.The mean RMSEs are 3.02 K for clear-sky conditions,4.02 K for cloudy-sky conditions,and 3.57 K for all-weather conditions.Uncertainty and coarse resolution of the original FY-4A and CLDAS data affect the accuracy of the obtained AW LST.The LST time series comparison results also show that the reconstructed AW LST is consistent with in-situ LST.The reconstructed AW LST also has good image quality and provides reliable spatial patterns.Second,considering the peculiarities of near real-time AW LST(NRT-AW LST)estimation without complete annual and daily scale data as input,this dissertation proposed a Spatial-TEmporal Fusion(STEF)method to fuse reanalysis data and geostationary satellite TIR data.The STEF method can be divided into two parts:temporal fusion and spatial fusion.STEF can fuse remote sensing and reanalysis data in both temporal and spatial dimensions.It is insensitive to the spatial coverage of the remote sensing data at the target time and has better fusion accuracy.STEF is tested in the Tibetan Plateau.Validation results on DOY 215-366 of 2020 indicate that STEF has good accuracy: the RMSEs are 3.19 K for clear-sky conditions,3.87 K for cloudy-sky conditions,and 3.56 K for all-weather conditions,respectively.STEF method can improve the accuracies of FY-4A LST,and RMSEs are significantly reduced.The NRTAW LSTs estimated by STEF have better accuracies than CLDAS LSTs under all-weather conditions.The STEF also exhibited similar results in 2021.We believe that the proposed STEF method can meet the requirements of NRT-AW LST estimation and contribute to improving the timeliness of region monitoring and related parameter estimations.Third,after coupling Random Forest(RF)and Geographically Weighted Regression(GWR),this doctoral dissertation proposed a downscaling method(Weighted GWR,WGWR)considering the weights of LST descriptors.To investigate the performance of WGWR,the 100 m Landsat-8 TIR Sensor(TIRS)and Terra Advanced Spaceborne Thermal Emission and Reflection Radiometer(ASTER)LSTs are aggregated to 1000 m as simulated coarse LSTs.The coarse LSTs are then downscaled to 100 m using WGWR,RF,and GWR.Meanwhile,the original 100-m LSTs are used as validation references.Results indicate that the proposed WGWR outperforms RF and GWR: for RF(GWR),the root mean square error(RMSE)can be reduced by 0.34 K(0.26 K)in Zhangye(ZY)and 0.22 K(0.1 K)in Beijing(BJ).Compared to RF and GWR,WGWR also provides better image quality: the downscaled LST images have neither the obvious "smoothing effect" nor the "boundary effect" and retain the details of the image at high spatial resolution.Validation based on in-situ LST indicates that the downscaled LST based on WGWR has better agreement with the in-situ LST,and the RMSE is reduced by 0.57 K.The WGWR was further applied to the LST estimated by RTG(RTG LST)and TRIMS(Thermal and Reanalysis Integrating Moderate-resolution Spatial-seamless)LST(from0.04° to 0.01°).The RTG downscaling results show that the agreement between RTG LST and the in-situ LST can be significantly improved by downscaling at some sites,and the RMSE can be reduced by 0.12 K after downscaling.The results of before and after the downscaling show that WGWR can recover part of the spatial details at a spatial resolution of 0.01°.The downscaling effect of TRIMS LST is consistent with the RTG LST,but its downscaling effect is more obvious due to using vegetation index data from the same sensor family.It can recover more spatial details,and the RMSE at some sites can be reduced by more than 0.5 K after downscaling.