Retrieval And Validation Of Long-term Land Surface Temperature From Satellite Remote Sensing

In recent years,climate change at the regional and global scales has become the focus of attention of different countries,scientific communities,and the public.With a significant increasing trend of the global temperature,the gradual warming trend has led to changes in land surface processes.As an important indicator of climate change,the long-term global land surface temperature(LST)has become an urgent need in the related fields.Compared with traditional in-situ measured LST,regional and global LST retrieved from satellite remote sensing has many advantages in timeliness,spatial coverage,and cost.In the process from the ground surface to the spaceborne sensor,the radiation emitted from the ground surface is affected by the atmosphere,terrain,and view-geometry of the satellite.Nowadays,the most widely used and accurate way for retrieving LST is from thermal infrared(TIR)remote sensing.However,there are bottlenecks in the framework of LST retrieval and validation.First,the global performance of the LST retrieval algorithm needs to be further studied,especially the feasibility under different atmospheric conditions.Second,the scale miss-match between the satellite pixel size and the ground radiometer's field of view(FOV)hampers the validation since the quantification of ground validation's spatial representativeness is challenging.Meanwhile,the near-surface atmosphere also affects the accurate measuring of the in-situ LST,but few studies have been conducted for this point.This dissertation has strived to address the above issues and achieved the following results.(1)An integrated retrieved method,namely RF-SWA,for retrieving LST from TIR remote sensing has been developed.Nowadays,the algorithms for retrieving LST from TIR remote sensing have come to maturity,but the performance and feasibility of the retrieval algorithm need to be further studied,especially under different atmospheric conditions.To this end,the 17 widely used split-window algorithms(SWAs)were trained,tested and sensitivity analyzed under different atmospheric conditions for the long-term NOAA AVHRR series archived data.From the initially selected SWAs,the algorithms with good performance and low sensitivity were determined.Based on the LSTs from the filtered SWAs and the corresponding truth,an integrated retrieved method,namely the RF-SWA,is proposed using the Random Forest regression.The RF-SWA method was evaluated using independent datasets,and the result shows that the accuracy is higher than 1 K,which is better than that of any individual SWA,and it is low sensitivity for the error of the input parameters.Finally,based on the RF-SWA method,the global daily clear-sky LST product for 1981-2000 has been generated from the NOAA AVHRR data.(2)A time-series evaluation model for quantifying the ground site's spatial representativeness was proposed.The in-situ LST is usually used as reference data to validate the satellite retrieved LST.However,there is a scale difference between the ground radiometer's FOV and satellite pixel,and this scale mismatch induces great difficulties for the validation of satellite LST,especially for the heterogeneous surfaces.Therefore,how to quantify the ground site's spatial representativeness and its impact on the validation result are the key issues in satellite LST validation.To address these issues,the ground site's spatial representativeness was defined as the LST difference between the ground radiometer's FOV and the corresponding satellite pixel,and the difference value is the spatial representativeness indicator(SRI).Considering the temporal variation of LST,the evaluation model for the ground site's representativeness was proposed based on the temporal decomposition model of LST,and SRI was temporally extended so that the ground site's spatial representativeness can be quantified in the temporal dimension.Based on the model,the spatial representativeness variation characteristics and influencing factors of 16 ground sites,which are located on different climate zones,on MODIS and AATSR pixel scale were analyzed.According to the definition,the temporal extended SRI was used as the proxy to convert the in-situ LST to satellite pixel scale,and the MODIS and AATSR LST were validated with the in-situ LST after scale conversion.The result shows that the ground site's spatial representation can induce a systematic error of-1.95-5.6 K and a random error of 0.07-3.72 K in the Temperature-based validation.(3)An atmospheric effect correction method for in-situ LST measured by ground radiometer is proposed.For satellite LST validation,radiation-based measurement is one of the most important approaches to obtain in-situ LST.Usually,the in-situ LST can be directly calculated according to the upwelling and downwelling longwave radiation measured by ground radiometers and the land surface emissivity.However,radiation from the near-surface atmosphere accounts for ?30% of that from the whole atmosphere,ignoring the atmospheric effect in the near-surface atmosphere would inevitably lead to uncertainty into the in-situ LST,while few studies have examined the atmospheric effect of near-surface atmosphere.Based on the radiative transfer simulation,the estimation equation of atmospheric parameters in the near-surface atmosphere was constructed.The atmospheric effect was reduced based on the radiative transform equation assisted with the atmospheric parameters estimation equation.The test and application results show that the in-situ LST would be overestimated about 0.2-5 K with a random error of 0.3-1.8K at the height of 1-50 m when the near-surface atmospheric effect is ignored.When the atmospheric effect is considered,the systematic deviation was reduced by about 67%-98%for the generally used approach.It is can be concluded that the proposed atmospheric effect correction method can effectively reduce the systematic deviation of the in-situ LST.(4)The LST retrieved from NOAA AVHRR was validated against the in-situ LST.The key in LST validation is to obtain the ground “truth” at the pixel scale.In general,the satellite retrieved LST was directly compared with the in-situ LST.In this chapter,the generated LST product from 1981-2000 was preliminary validated against the in-situ LST,and the result shows that the accuracy of the product is better than 3 K.Further,the insitu LST at the pixel scale was obtained by conversion from the atmospheric effect corrected in-situ LST with temporal extended SRI,and it was used in LST validation,the validation target LST is retrieved from the NOAA AVHRR data of 2018-2020.The result shows that the accuracy of the LST at most sites is better than 2.5 K based on the variance analysis,and the systematic deviation is 0.55 K at daytime and 1.27 K at nighttime.Further analysis shows that the large overestimation at nighttime may be caused by the overestimation of atmospheric water vapor content used in the retrieval process.This finding provides suggestions for subsequent improvement of the LST product.