Resaerch On Spatial Constrained Model Of Land Surface Temperature Downscaling Over Urban Region

Land surface temperature(LST)is a key parameter to describe the process of surfaceatmosphere interaction and reflect the surface condition.However,due to the constraints of the imaging conditions of the thermal infrared sensor,there is a contradiction in the spatial and temporal resolution of the remote sensing images.This contradiction prevents the LST data being fully utilized.The LST downscaling algorithm can effectively resolve the contradiction of spatiotemporal resolution,and obtain LST images with higher spatiotemporal resolution.Based on summarizing the existing algorithms of LST downscaling and fully considering the spatial characteristics of LST,this study proposed an LST downscaled model to downscale spatial resolution of the moderate resolution imaging spectroradiometer(MODIS)LST data from 1000 to 100 m,and analyzed the LST downscaling results.Firstly,considering the spatial nonstationary and spatial autocorrelation simultaneously,this study proposed a new algorithm based on the geographically weighted autoregressive(GWAR)model to downscale the LST.The downscaled result of GWAR model was compared with the results of traditional global model and the GWR model.Secondly,considering that there is not a strict linear relationship between LST and factors,we proposed the non-linear geographically weighted(NL-GWR)model and the non-linear geographically weighted autoregressive(NL-GWAR)model to downscale the LST.Thirdly,this study selected the plain city Beijing and plateau city Lanzhou as the study areas and analyzed the diffreent effects of normalized difference vegetation Index(NDVI)and normalized difference Build-up Index(NDBI)as the independent variables,respectively.At last,using the Landsat 8 LST as the reference data,which was used to verify accuracy of the downscaled results,and analysis the downscaling results from different models.The conclusions are as follows:1.Compared with the local LST downscaled models,the results of traditional global LST downscaled model have many boxy artifacts,and the LST downscaling algorithm of the GWR has a smoothing effect.2.The GWAR algorithm,proposed by this thesis improves the smoothing effects of the GWR algorithm,and show more spatial details than the GWR model.The Landsat 8 retrieved LST was used to verify the downscaled LST.Compared with other algorithms,the results of GWAR algorithm has lower root mean square error(Beijing: 1.37?,Lanzhou: 1.76?)and mean absolute error(Beijing: 0.86?,Lanzhou: 1.33?).3.Comparing the NDVI and NDBI as the independent variables,respectively,this thesis concluded that the results of NDVI as the independent variable have the lager error in the areas with the high soil moisture.It is because that the relationship between LST and NDVI is not the linear relationship in the areas with high soil moisture.And the NDVI is affected by the season easily.Nevertheless,using the NDBI as the independent variable can avoid the influence of high soil moisture areas and be not susceptible to seasons.4.In addition,when considering the non-linear relationships between the variables,it is found that the downscaling results of NL-GWR and NL-GWAR models considering the non-linear relationship were better than the linear models.5.In this thesis,we proposed the GWAR based algorithm to downscale the LST,which considered the spatial nonstationary and spatial autocorrelation simultaneously in the study of spatial variables.In short,this thesis provides a new method for the study of LST downscaling and improves the results of LST downscaling which is essential for the study of LST in the future.