Research On Land Surface Temperature Downscaling Method Based On GTWAR Model

Land Surface Temperature(LST)is a key parameter in many thermal environment studies.Under current technological conditions,satellite sensors cannot simultaneously provide high spatial and temporal resolution surface temperature data.LST downscaling is a useful way to solve this problem.The downscaling algorithms can effectively improve the spatial and temporal resolution of land surface temperature data.Inspired by the existing downscaling algorithms,a new downscaling model of LST is proposed in this thesis,considering the spatial heterogeneity,spatial autoregressions and temporal characteristics of LST and related data.The surface temperature product MOD11A1 of the Moderate Resolution Imaging Spectroradiometer was downscaled from 1000 m spatial resolution to100 m spatial resolution,and the downscaling results were verified and analyzed.Firstly,in this study,the temporality,spatial heterogeneity and spatial autocorrelation of surface temperature data were comprehensively considered,and a LST downscaling algorithm based on the Geographically Weighted Autoregression(GTWAR)model was proposed.Secondly,an adaptive selection method of scale factor under GTWAR model was proposed to further optimize the downscaling model on the basis of GTWAR downscaling algorithm.Thirdly,Beijing and Zhangye were selected as the study area in this study.The effects of normalized vegetation index(NDVI)and normalized build-up index(NDBI)as explanatory variables and NDVI,NDBI and normalized water index(NDWI)as explanatory variables on the downscaling fitting results were analyzed.Finally,the Landsat 8 land surface temperature retrieved by the radiative transfer equation method was used as the reference data,and the results of the downscaling algorithm based on each model were compared and verified.The conclusions are as follows:1.Compared with the local model downscaling algorithm,the downscaling results of TsHARP based on global model have average effect and block effect in spatial distribution.The results of GWR downscaling algorithm have smoothing effect in some regions,and the results of GTWR and GWAR downscaling algorithms have some shortcomings in the restoration of spatial details and error control.2.The downscaling algorithm of land surface temperature based on the GTWAR model was proposed in this study combines the advantages of the GTWR model and the GWAR model,takes each characteristic of the LST data into more comprehensive consideration,and effectively improves the spatial details and error control of the downscaling results.In the validation results based on Landsat 8 LST,the downscaling algorithm based on the GTWAR model has the smallest root mean square error(Zhangye:1.57?,Beijing: 1.22?)and the mean absolute error(Zhangye: 1.06?,Beijing: 0.85?)in the study area,and the R2 of Zhangye and Beijing study areas reaches 0.90 and 0.88.3.In the selection process of scale factors,it is found that the relationship between scale factors and LST may change with the change of land cover types,and the fitting relationship between NDVI and LST in high water content area is quite different from that in other areas.NDWI is sensitive to water content.After adding this scale factor,some errors caused by NDVI in water area were corrected,and the direct fitting coefficient between explanatory variable and LST was significantly improved.4.Since the direct relationship between scale factors and LST is affected by land cover types,different scale factors were considered to be used for fitting for different land cover types,that is,the optimal scale factor under the surface cover types of each region is adaptively selected for downscaling.The downscaling results before and after the application of the adaptive scaling factor selection method under the GTWAR model were verified and compared.The results show that the error of the downscaling results can be further reduced in the areas with complex land cover types by using the adaptive scale factors selection method.