Research On Spatial Upscaling Of In-Situ Land Surface Temperature Over Heterogeneous Surface

Global warming,environmental change,and other issues are getting more and more attention.Therefore,the demand for land surface temperature(LST)has also increased among relevant institutions and researchers.In-situ LST measured from the ground stations and low-altitude platforms plays an important role in the validation of satellite LST in isothermal pixels and the directional radiation of non-isothermal pixels.However,in a non-isothermal three-dimension scene,the in-situ LST is also affected by factors such as the spectral response differences of sensors,the directional radiation difference within the sensor's field-of-view,and the differences in the spatial scales.Therefore,it is hard to compare,validate and apply the multi-source LST.This dissertation has conducted research on the mentioned issues,and achieved the following main results.(1)A method for retrieving LST from the in-situ broadband thermal infrared data has been proposed.The spectral response function segmentation method was employed to improve the radiative transfer algorithm and the single-channel algorithm to adapt retrieving LST from the broadband thermal infrared data.The improved algorithms were applied to 5 widely used sensors in field experiment.Simulation result shows that the influence of segmentation width on LST accuracy depends on the characteristics of the sensor's spectral response function and sensor's height.At 2 m,with a 2.5 ?m segmentation width,the RMSE is lower than 0.5 K for the 5 sensors;at 1100 m,the RMSE is lower than 1.0 K in most cases.With the increase of the segmentation width,the accuracy of all sensors gradually decreases,and the LST differences among the sensors increase significantly.Therefore,the segmentation width should be less than 2.5 ?m to ensure the accuracy of LST.The results show that this method can effectively improve the accuracy and the consistency of multi-source LST.(2)A directional radiation model for sparse vegetation scene has been achieved.In addition,a method to correct the angle effect of the in-situ component temperature has been proposed based on this model.In this study,3D cloud point and simple geometry were used to build 3D scene models respectively.The directional radiative of complex vegetation scenes were simulated with the radiative transfer equation,and then the angle correction of in-situ component temperature was carried out according to the simulation results.The results show that the performance of the two scene models is similar in terms of simulating component temperatures,component fractions,and the directional radiations.When comparing with the DART model,the RMSD between the simulated directional brightness temperatures is less than 2.0 K.When the angle effect of the in-situ component temperature was corrected,the RMSD of the upscaled temperature,which is upscaled from the component temperature under different direction,is reduced by 1.1 to2.1 K.It shows that the angle effect can be reduced.(3)A method to calculate the directional proportion of components from classified data and achieve upscale in-situ component temperature has been proposed.First,the simplified 3-D(S3D)model was established based on the classification image and the average vegetation structure parameters;then the component fractions calculating method was established by combining the Boolean(S3D-B)model.Results show that the difference is up to 0.05 when the component fractions are calculated using the S3D-B model without considering illumination condition,and up to 0.17 when illumination condition is considered.The simulation results show that the upscaling method weighted by the component fractions has a RMSD of 0.7 to 1.4 K.The S3 D component fractions and the corrected in-situ measured component temperature can produce the RMSD of 1.7to 3.8 K,and the S3D-B component fractions can produce the RMSD of 2.6 to 4.5 K.The upscaling results are 0.6 to 5.7 K different from TOA brightness temperatures of Landsat-8 and 2.3 to 4.0 K different from MODIS observations.Due to the influence of various factors such as the characteristics of the ground scene and the observation process and data process of component temperature,the component fractions and the upscaled LST have significant uncertainty in some cases,but the proposed method still show significant potential in the field of LST upscaling.