| With the rapid development of China's industrialization and urbanization,the emissions of airborne particulates and trace gases have also increased significantly,and the problem of air pollution has become increasingly serious.At the same time,the accompanying epidemic diseases and respiratory diseases have seriously endangered humans.Physical and mental health,therefore,it is of great significance to obtain the spatial distribution of near-ground pollutant concentrations with high temporal and spatial resolution.With the continuous development of the Internet and information technology,data volume has grown exponentially.Information explosion is an important feature of today's era,and emerging machine learning algorithms can extract features quickly and with high accuracy,which has been applied to medical,face recognition,speech recognition,machine translation and many other fields.At the same time,various remote sensing satellites related to atmospheric monitoring have also been launched in recent years,monitoring the changes of various components in the atmosphere at all times and all days.which provide us with massive amounts of data.Therefore,this paper applies machine learning algorithms,combined high spatial and temporal resolution NO2 data provided by OMI and TROPOMI,NDVI data provided by MODIS,meteorological data provided by EMCWF,emission inventory data provided by MEIC,and other types of data.The near-surface O3concentration data observed at the site are verified by spatiotemporal matching and training fit,so as to predict the near-ground O3 concentration distribution with high spatial-temporal resolution and perform population weighted exposure assessment.The main research conclusions are as follows:?1?There are many types of regression prediction models.In order to select a high-precision and good-performance regression model,this paper chooses the most popular regression models in recent years.Geographically and Temporally Weighted Regression?GTWR?,random forest?RF?,and gradient boosted regression tree?GBRT?are used for In the cross-validation experiments of the same data set,the determination coefficients R2 are:0.80,0.84,0.91,and the root mean square errors are:14.75??g/m3?,13.12??g/m3?,9.67??g/m3?The prediction accuracy of GBRT is the best,so this paper adopts it as the regression model of the high-temporal-resolution remote sensing product data to participate in the model comparison test.?2?The feature variables are selected for the data set after the spatiotemporal matching,so that the model was more streamlined and efficient,and the Missing group,OMI-NO2 group,and TROPOMI-NO2 group are compared for near-surface O3concentration prediction experiments.The results show that the cross-validation of the TROPOMI-NO2 group has a decision coefficient R2 of 0.92 and a root mean square error RMSE of 12.52??g/m3?,which are better than those of the other two groups of experiments.In terms of the predicted O3 spatial distribution map,The TROPOMI-NO2 group's prediction result has a large number of data samples,no significant banding in the spatial distribution,smoothness and continuity,the spatial distribution details are more obvious,and the underestimation of the prediction has been significantly improved;the actual observations with the ground In terms of value comparison,the national average annual distribution of the TROPOMI-NO2 group has the best consistency with the ground observations,and the concentration is closest to the true ground value.In the comparative tests of Jiangsu,Sichuan,Guangdong,and Jilin,the TROPOMI-NO2 group is the closest to the ground The true value,and OMI-NO2 group and Missing group in the O3 median area?Sichuan,Guangdong?have a certain deviation.?3?Based on the spatial distribution of near-surface O3 concentration predicted by the model,population-weighted exposure assessments of provinces in China are performed.The five provinces with the highest population-weighted O3 concentration values are:Henan?109.44?g/m3?,Hebei?108.65?g/m3?,and Shaanxi.?107.28?g/m3?,Shandong?106.46?g/m3?,Anhui?103.40?g/m3?,the lowest five provinces:Hainan?55.16?g/m3?,Liaoning?74.76?g/m3?,and Guizhou?75.55?g/m3?,Fujian?76.43?g/m3?,Guangxi?77.03?g/m3?,the five provinces with the longest duration of population-weighted O3 exposure are:Henan?203 day?,Shandong?200 day?,Shaanxi?196 day?,Ningxia?195 day?,Hebei?193 day?,an average of 54%of the days in a year are in a state of non-compliance with O3concentration,the shortest five provinces:Hainan?41 day?,Guizhou?76 day?,Heilongjiang?85 day?,Guangxi?85 day?,and Yunnan?89 day?.About 21%of the days in a year were in a state where the O3 concentration was not up to standard. |
PDF Full Text Request