Study On PM_2.5 Estimation Method Based On GF-5 DPC Dat

With the accelerating industrialization in China,the atmospheric environment problem is becoming more and more serious,and how to effectively use satellite remote sensing technology to estimate the atmospheric PM_2.5 concentration is of great significance to explore the spatial and temporal variation of PM_2.5.At the present stage,the estimation of PM_2.5 concentration by satellite remote sensing is mostly based on Aerosol Optical Depth（AOD）products inverted by optical satellite sensors,and AOD products are obtained from Top of Atmosphere Reflectance（TOA）inversion of satellites.Directly establishing the relationship between TOA and PM_2.5concentration can reduce the error transfer from AOD inversion.Compared with the traditional optical satellite TOA,the satellite apparent polarization reflectance is more sensitive to atmospheric fine particles,so the satellite polarization technique is more advantageous for near-ground PM_2.5 concentration estimation.Based on this,this paper uses the polarimetric reflectance data from the Directional Polarimetric Camera（DPC）of the Gaofen-5 satellite（GF-5）to carry out a study on the estimation of near-ground PM_2.5 concentrations.The main research works are as follows:（1）A method for direct estimation of near-ground PM_2.5 concentration from the polarimetric reflectance of GF-5 DPC is proposed.Firstly,three polarization band data of DPC are selected,and the Stokes parameters are transformed into polarization reflectance,and then the PM_2.5estimation model dataset is constructed by spatio-temporal matching with ground PM_2.5 stations as the reference,and the link between polarization reflectance and PM_2.5 concentration is established directly.（2）A method for direct PM_2.5 estimation using integrated learning idea is proposed.To realize the direct estimation of PM_2.5 concentration using DPC polarized reflectance,this paper proposes a DPC-PM_2.5 estimation model based on improved Stacking,i.e.,DPC＿PM＿IStack,which utilizes the features of DPC polarized reflectance and has strong generalization capability and high estimation accuracy.For the features of GF-5 DPC polarized reflectance with more dimensions and complex feature space,the model is optimized in terms of constructing a training subset without put-back sampling and assigning weights to the base model,respectively,to further improve the model’s ability to estimate PM_2.5 directly from DPC polarized reflectance.（3）The R²,RMSE and MAE of the DPC＿PM＿IStack integrated model based on the 10-fold cross-validation results are 0.91,8.91μg/m3and 4.18μg/m3,respectively,as analyzed by the experimental results.In the 10-fold cross-validation based on sample time and space,the R² was above 0.8 for 88%of days in 2019 in time,and the R² was greater than 0.7 for 70%of sites in space,which made the integrated model more scalable in time and space.The comparison analysis with the PM_2.5 estimation results based on AOD shows that the estimation of DPC polarized reflectance data in the urban bright surface area is better,and the integrated model improves the spatial coverage of the estimation results while simplifying the estimation process and reducing the error transfer.