Research On Aerosol Optical Depth And PM_2.5 Inversion Based On Remote Sensin

In recent years,air pollution has attracted more and more attention at home and abroad.Aerosol Optical Depth(AOD)and Fine Particulate Matter with a Diameter Smaller than 2.5μm(PM2.5)are important indicators for assessing the quality of the atmospheric environment,The acquisition and quantitative analysis of AOD and PM2.5 concentration with temporal and spatial continuity can provide more detailed data support for the fine management of environmental quality.Compared with traditional ground-based monitoring methods of AOD and PM2.5 concentration,satellite remote sensing can obtain AOD and PM2.5 concentration information with more complete local details and higher spatial and temporal resolution.AOD has a strong correlation with PM2.5,and is often used as an important parameter of PM2.5concentration inversion.However,at present,the spatiotemporal resolution of AOD products at home and abroad is generally low,which is difficult to meet the small regional scale air quality monitoring.In addition,due to the influence of modeling and observation errors,AOD obtained by remote sensing means inevitably has deviation,which is bound to reduce the inversion accuracy of PM2.5 concentration.Therefore,in order to give full play to the advantages of high spatial and temporal resolution of China’s Gaofen-4(GF4)satellite,improve the spatial and temporal resolution of AOD and PM2.5 concentration,and enhance the inversion accuracy of PM2.5 concentration,Guiyang City was selected as the research area in this study,and the reflectivity of GF4 was used as the main data source.Based on the machine learning algorithm,the high spatial resolution AOD inversion model and PM2.5 concentration inversion model were constructed respectively.On this basis,the spatial and temporal distribution characteristics of PM2.5 concentration in Guiyang were analyzed.The main research contents and conclusions of this paper are as follows:(1)Construction of Guiyang AOD inversion model based on GF4 PMS reflectanceUsing GF4 reflectance data and MODIS AOD data from 2019 to 2022,and based on Light GBM(Light Gradient Boosting Machine)machine learning algorithm,we constructed Guiyang AOD inversion model(Light GBM-AOD model).The performance of the model is analyzed experimentally.The results show that: The inversion model of AOD based on Light GBM can describe the change trend of AOD well.The Root Mean Square Error(RMSE)and Mean Absolute Error(MAE)of model fitting are 0.0572 and 0.0424,respectively,and the correlation coefficient R between the model fitting value and MODIS AOD is 0.867.A statistically strong correlation;Using GF4 reflectance data not involved in modeling as input parameters,MAE for AOD prediction in urban and non-urban areas were 0.077 and 0.086,and RMSE for non-urban areas were 0.094 and 0.101,respectively,indicating that the model has good generalization ability and can achieve AOD prediction with high precision and spatial and temporal resolution in both urban and non-urban areas of Guiyang City.(2)Construction of PM2.5 concentration inversion model in Guiyang based on RF(Random Forest)In order to analyze the applicability and accuracy of single-type factor model and multitype factor model for PM2.5 concentration inversion,and explore the feasibility of using GF4 geostationary satellite reflectance data directly for PM2.5 concentration estimation,The ERA5-PM2.5 inversion model based on a single type of meteorological factor,the AOD-PM2.5inversion model based on the AOD value predicted by Light GBM-AOD model and the GF4-PM2.5 inversion model based on the reflectance data of GF4 PMS were constructed,and the accuracy of the three models was compared and analyzed.The results show that all the models constructed by the three methods have high inversion accuracy of PM2.5 concentration,with R greater than 0.92.Compared with the single-type factor modeling,the multi-factor model has a higher inversion accuracy for PM2.5.Compared with ERA5-PM2.5 model and AOD-PM2.5model,the prediction accuracy of RMSE and MAE of GF4-PM2.5 model decreased by0.681μg/m3 and 0.934μg/m3,respectively,and the correlation coefficient between the predicted value of the model and the measured value of the site reached 0.61,showing a strong correlation statistically.The results indicate that the GF4 satellite reflectance data can be directly used for the high precision inversion of PM2.5 concentration.(3)Comparative analysis of seasonal models of PM2.5 concentrationIn order to further explore the applicability of the three methods to the inversion of PM2.5concentration in Guiyang city,the three methods were used to establish a seasonal inversion model of PM2.5 concentration,and compared with the measured PM2.5 values at the station.The results show that the seasonal models constructed by each method have high modeling and prediction accuracy,and the correlation coefficient R is greater than 0.9.Seasonal models of different methods were compared and analyzed.The overall accuracy was ranked from highest to lowest as GF4-PM2.5 model,ERA5-PM2.5 model and AOD-PM2.5 model.(4)Analysis of temporal and spatial variation characteristics of PM2.5 concentrationBased on the GF4-PM2.5 model constructed,the PM2.5 concentration values of four years from 2019 to 2022 were reverse-performed,and the spatial distribution map of PM2.5concentration was drawn by using the inverse distance weight interpolation method.The spatial and temporal distribution characteristics of PM2.5 concentration values in Guiyang from 2019 to 2022 were analyzed by comparing with the measured values at ground stations.The PM2.5concentration in Guiyang showed a spatial distribution feature of "high in the east and low in the west".In terms of time,the variation trend of annual and quarterly mean is decreasing year by year,while the variation trend of monthly mean is U-shaped with low middle and high sides.