Research On Satellite Remote Sensing Retrieval Method For Land Surface-atmospheric Variables

Changes in ecosystems are closely related to the development of human society.One of the important means to record and analyze these changes on a large regional or global scale is to use satellite observation data.Global remote sensing satellites have obtained continuous earth observation data for several decades.Researchers have also developed various algorithms to retrieve remote sensing variables from these observation data and generate advanced products,such as MODerate Resolution Imaging Spectroradiometer(MODIS),Global Land Surface Sat-ellite(GLASS),etc.However,these variable products have the following problems.First,the physical consistency among variable products is weak and variables are discontinuous.Tradi-tional methods develop inversion algorithms for a single remote sensing variable based on dif-ferent physical assumptions,models and input data,so that variable products from the same sensor have large differences in characterizing the physical characteristics of the surface.Due to cloud contamination,when there is no effective reflectivity data,these algorithms can not successfully estimate the effective variable value,resulting in product discontinuity.Second,the commonly used satellite remote sensing data have different spatial resolutions.Researchers usually focus on the algorithms for estimating the remote sensing variables at a single scale,while few studies on algorithms for estimating remote sensing variables with multi-spatial res-olution were developed.The generation of multi-spatial resolution remote sensing variables can represent the surface characteristics at different scales,meet the needs of different industries for remote sensing data,and provide multi-resolution data input for ecosystem modeling.In view of the above problems,this paper focused on the consistent estimation method of remote sensing variables and the consistent estimation method of remote sensing variables with multi-spatial resolution.The main research contents and conclusions of the paper include the following aspects:(1)This paper developed the effective optimization method to simultaneously estimate multiple physically consistent land surface and atmospheric variables with a spatial resolution of 250 meters from the time series the top of atmosphere(TOA)reflectance of the FY-3B The Medium-Resolution Spectral Imager(MERSI).The MODIS cloud detection method was ap-plied to the MERSI TOA reflectance to determine the clear observations.Then a physics-based BRDF correction was adopt to reduce the topographic effects using the slope angle and aspect angle of the slope.Finally,a coupled land surface-atmosphere radiative transfer(RT)model and the Shuffled Complex Evolution(SCE)optimization algorithm were used to estimate a suit of variables,including the leaf area index(LAI),the aerosol optical depth(AOD),the cloud optical thickness(COT),the cloud effective particle radius(CER),the land surface reflectance,the shortwave and visible albedo,the incident shortwave radiation(ISR),the incident photo-synthetically active radiation(PAR),the fraction of absorbed PAR(FAPAR),the surface broad-band emissivity(BBE),and the TOA shortwave albedo.The validation results show that the estimated LAI,PAR,FAPAR,shortwave albedo,and ISR are consistent with the field measure-ments,with coefficients of determination(R~2)of 0.692,0.783,0.788,0.559,and 0.833,and root mean square errors(RMSEs)of 0.427,56.681 W/m2,0.092,0.084,and 115.305 W/m2,respectively.The estimated variable values are consistent with GLASS and MODIS products.In addition,this method can be applied to other remote sensing satellites.(2)This paper developed a machine learning method to replace the optimization method and efficiently estimate seven global surface and atmospheric variables from the Visible Infra-red Imaging Radiometer Suite(VIIRS)TOA data:LAI,FAPAR,surface albedo,surface albedo,ISR,PAR and TOA albedo.First,the optimization method was used to generate the training data set from the VIIRS TOA reflectance data,and then the random forest(RF)model was trained to connect the VIIRS TOA reflectance and the seven variables.Finally,the RF model was used to generate the global estimate from the VIIRS TOA reflectance in 2013.Validation results at 54 field sites showed that this method could accurately estimate LAI,FAPAR,shortwave albedo,ISR,and PAR,with:R~2 values of 0.867,0.728,0.762,0.881,and 0.845,RMSE values of 0.629,0.089,0.058,102.127,and 50.125,BIAS values of-0.048,0.033,-0.002,2.135,and-12.077.The estimated global results effectively capture the seasonal dy-namics of vegetation and are highly consistent with the official GLASS and VIIRS products.(3)This paper developed a depth learning method to simultaneously estimate multiple surface variables with seven spatial resolutions from the TOA reflectance data of seven satellite sensors.The estimated land variables include LAI,FAPAR,shortwave albedo,visible albedo,and land surface reflectance.Seven sensors that acquire data at different spatial resolutions,include VIIRS、MODIS、FY-3B、MERSI、CBERS-04、Landsat 8、GF-1 and Sentinel-2A/B.This method first downscaled the VIIRS TOA reflectance to six different spatial resolutions by inputting six kinds of satellite observations using the Multi-scale and multi-depth convolutional neural network(MSDCNN).Then,the multi-resolution VIIRS TOA reflectances were input to the RF model developed by(2)to estimate the multi-resolution surface variables.Validation results using ground measured data showed that the estimated multi-resolution variables have high accuracy,with RMSE of 0.361–0.617(LAI),0.023–0.120(FAPAR),0.013–0.026(snow free shortwave albedo).