Multi Source Satellite Remote Sensing Retrieval And Application Of Evapotranspiration And Its Key Parameters

Evapotranspiration includes water vapor exchange from soil to vegetation and then to atmosphere,and involves the complex interaction process between soil-vegetation-atmosphere system.Its accurate retrieval has always been an important and difficult problem in the field of satellite remote sensing.Based on the three products of most widely used Moderate-resolution Imaging Spectroradiometer(Terra/MODIS),the Advanced Geostaionary Radiation Imager(FY-4a/AGRI)of China's new generation geostationary meteorological satellite and Enhanced Thematic Mapper(Landsat7/ETM+)with high spatial resolution,this paper aims to improve the remote sensing retrieval accuracy of evapotranspiration and realize the application of evapotranspiration in geoscience related fields.Firstly,the accuracy of existing remote sensing products is verified by using the measured data,and the retrieval algorithms of intermediate parameters such as net radiation,land surface temperature and emissivity are improved to lay a foundation for evapotranspiration retrieval accuracy improving.Secondly,empirical model,the characteristic space method and surface energy balance system(SEBS)are used for evapotranspiration retrieval,and the applicability,retrieval accuracy and improved methods of different models are compared and analyzed.Finally,based on evapotranspiration products,the evapotranspiration stress index is used to monitor the summer drought in Northwest China in 2021,and the applicability of index in drought monitoring is discussed.Results show that:(1)As a key factor of evapotranspiration retrieval,high-precision land surface temperature product is the basis of evapotranspiration retrieval.As the most direct influencing factor of land surface temperature retrieval,emissivity has significant temporal and spatial differences and diurnal variation characteristics.The emissivity model considering land surface conditions,vegetation conditions and atmospheric environment can improve the retrieval accuracy of land surface temperature,and the root mean square errors(RMSEs)of land surface temperature estimated by MODIS and AGRI remote sensing data are reduced by 1?and 0.2?respectively.Based on the emissivity model given in this study,the land surface temperature retrieval algorithm is optimized by the artificial intelligence particle swarm optimization(PSO)algorithm for FY-4A/AGRI data.After optimization,the RMSE between the estimated land surface temperature and the measured value is reduced from 6.3?to 3.5?.(2)Net radiation,as the driving factor of energy exchange in the earth-atmosphere system,is an indispensable parameter in the process of evapotranspiration retrieval.The mean absolute percent error(MAPE)between the instantaneous radiation value estimated by the existing model developed for MODIS data and the measured value at the time of satellite transit is smaller than 15%.However,due to the design differences between different detectors and the different observation attitudes between geostationary satellite and polar orbiting satellite,it is inappropriate to apply the MODIS radiation retrieval model to AGRI data directly.The RMSE between the retrieval net radiation and the measured value is 88 W·m^-2,and MAPE is 37%.In order to improve the accuracy of geostationary satellite radiation retrieval,a simple model for geostationary satellite radiation retrieval is developed with the PSO method.The RMSE of net short-wave radiation in the study area is reduced from 98 W·m^-2 to 64W·m^-2,the RMSE of net long-wave radiation is reduced from 94 W·m^-2 to 48 W·m^-2,the RMSE of net radiation is reduced to 67W·m^-2,and the MAPE is reduced to 27%.(3)Comparing the empirical model,characteristic space method and SEBS single-layer model,it is found that the retrieval results of SEBS which after local optimization are most consistent with the measured values,the RMSE is reduced from142 W·m^-2 to 79 W·m^-2,and the MAPE is reduced to 19%,but the single-layer model needs the assistance of ground observation data.The retrieval results of the widely used characteristic space method are inferior to the SEBS model,and the establishment of characteristic space by using the perpendicular drought index which closely related to soil moisture and vegetation index can improve the retrieval accuracy of evapotranspiration.The RMSE between the retrieval results and the measured values is 92 W·m^-2,and the MAPE is 25%.However,the characteristic space method itself has certain application limitations,and the higher spatial resolution of remote sensing data,the more careful pixel information depicted by remote sensing data,and the higher accuracy of underlying surface evapotranspiration retrieved by characteristic space method.The deviation between the evapotranspiration estimated by the empirical model and the measured value is the largest,and the MAPE is about 35%,but there is a good correlation between the estimated value and the measured value.The empirical model can conveniently and quickly realize the evapotranspiration retrieval on a large-scale,especially in the areas lacking observational data.The drought monitoring index based on evapotranspiration is suitable for the regions and periods where soil water is the dominant factor of evapotranspiration.(4)The results of evapotranspiration stress index for summer drought monitoring in Northwest China in 2021 are consist with MCI index.There is no doubt that it is difficult to monitor the drought accurately on a large sale for soil hydrothermal characteristics and vegetation drought resistance.The accurate monitoring of large-scale drought needs to be further studied.