Extraction And Prediction Of PM2.5 Spatiotemporal Distribution In Beijing-Tianjin-Hebei Based On Remote Sensing

High-precision continuous PM25 spatiotemporal distribution is an important basis for studying the formation and diffusion mechanism of PM2.5.Aiming at the problems of high-value underestimation and the discontinuity of time and space of traditional PM2.5 inversion methods,this paper studies the time and space continuous PM2.5inversion and prediction in Beijing-Tianjin-Hebei(Jing Jin Ji).A deep neural network(DNN)PM2.5 inversion model based on meteorological factors,gaseous pollutant monitoring data,and AOD is established,which effectively improved the phenomenon of high-value underestimation and generated daily PM2.5 spatial distribution.Based on the spatiotemporal autocorrelation of PM2.5,a spatiotemporal autoregressive model is designed to generate hourly PM2.5 spatial distribution map from 0 to 23 hours,which makes up for the defect of spatiotemporal discontinuity of PM2.5.The prediction model of PM2.5 was built to realize the prediction of regional PM2.5 concentration in the next day.The main research contents are as follows:(1)PM2.5 inversion model considering gaseous pollutants and DNNAiming at the high-value underestimation problems in the process of PM2.5concentration inversion from heavy polluted weather in Jing Jin Ji,a PM2.5 inversion model based on meteorology,gaseous pollutants and AOD is established,and the influence of gaseous pollutants on the inversion accuracy of PM2.5 is analyzed.The experimental results show that after the introduction of gaseous pollutants,the cross-validation R~2increased by 0.12 to 0.87,and the root-mean-square prediction error(RMSE)decreased by 9.72??g/m~3to 27.11?g/m~3,indicating that the prediction accuracy of DNN model was significantly improved.In view of the lack of AOD,the inversion model of PM2.5 based on gas pollution and meteorology is established,which improves the accuracy of the inversion results of PM2.5 in each season,and the spatial distribution of PM2.5 in AOD missing area is more reasonable.(2)Spatiotemporal autoregressive model considering the impact of PM2.5 at monitoring stationsIn view of the lack of hourly regional PM2.5 spatial distribution model,this paper uses daily PM2.5 spatial distribution and site monitoring data to develop a spatiotemporal autoregressive model of PM2.5,generating hourly(0-23)PM2.5 spatial distribution in Jing Jin Ji in 2014,and analyzes the accuracy of the model.The experiment results show that the model cross-validation R~2was 0.82,and the RMSE was 37.37?g/m~3.The hourly spatial distribution prediction results of PM2.5concentration showed that the model in this paper can provide accurate spatial and temporal distribution characteristics for short-term PM2.5 exposure studies.(3)Regional PM2.5 prediction method for the next dayGiven the existing statistical models can only predict the pollutant concentration at the monitoring sites,this study use the PM2.5 spatial distribution data to establish a spatiotemporal autoregressive model,analyze the effects of different variables and simulation functions on the model prediction accuracy,and explores the prediction method of high-precision regional PM2.5 spatial distribution in Jingjinji.Results showed that the spatiotemporal autoregressive model had the best prediction performance when GBDT was used as the simulation function.The R~2,RMSE,index of agreement(IA),and mean absolute error(MAE)of the spatiotemporal autoregressive model were 0.85,27.08?g/m~3,0.96,and 20?g/m~3,respectively,indicating that the model had acceptable prediction accuracy.The spatial distribution prediction results of PM2.5 showed that the+1-day PM2.5 spatial distribution prediction results were in good agreement with the PM2.5 spatial distribution results predicted by AOD to provide accurate spatiotemporal distribution data for air pollution early warning.This study constructed a spatiotemporal continuous PM2.5 inversion model,which is expected to provide data guarantee for the study of PM2.5 formation and diffusion process;on the other hand,it can provide a new way and solution for regional PM2.5prediction,which can provide decision support for correctly guiding public travel and making effective preventive measures for the government.