Mass Concentration Estimation Of Atmospheric Particulate Matter Based On MODIS Aerosol And WRF Model

With the continuous expansion of cities and the rapid development of industry,air pollution occurs frequently in all parts of China,causing great harm to social development and people’s health.Among them,the air pollution mainly composed of inhalable particulate matter(PM₁₀,PM_2.5)occurs frequently,causing great harm to human health,which has gradually been widely concerned.Therefore,the monitoring and prediction of inhalable particulate matter has important value,and can provide a theoretical basis for the treatment of regional air pollution in China.Satellite remote sensing is the main method for monitoring air pollution,which has the advantages of high regional coverage,good real-time performance and low cost,and can effectively make up for the shortage of less and uneven distribution of ground monitoring stations.This paper takes the main urban area of Nanjing as the research area,the mass concentration data of atmospheric particles(PM₁₀,PM_2.5)and MODIS aerosol data were obtained,the annual and seasonal variations of AOD（aerosol optical depth）,FMF（fine-mode fraction）and PM₁₀,PM_2.5concentrations in the study area were analyzed.Use WRF mode to obtain near-ground and high-altitude conditions（including near-ground temperature and boundary layer height,near-ground and high-altitude wind speed,relative humidity）,and perform Pearson correlation analysis and significance of each factor with PM₁₀and PM_2.5mass concentrations test.And then establish multiple linear regression models and random forest models to estimate and verify the mass concentration of PM₁₀and PM_2.5,and select the best estimation modelThe main research contents and results of this article are as follows:（1）By obtaining the data of particulate matter and optical properties of aerosols,it can be found that the mass concentration of PM₁₀in the main urban area of Nanjing tends to be higher in spring and winter,and lower in summer,while that of PM_2.5is higher in winter and lower in other seasons.AOD presents a seasonal downward trend from spring,summer,autumn and winter,while FMF is higher in summer and winter,and lower in spring and autumn.（2）After fine-model correction of AOD by FMF,the correlation between AOD and PM_2.5in each season is improved.In spring,under the action of northerly wind,the study area is more affected by northern sand and dust.In other seasons,the atmosphere is mainly dust-type aerosols and the mass concentration of PM₁₀is high.After fine model correction,the correlation between AOD and PM_2.5has raised significantly,so FMF can reduce the optical thickness contributed by atmospheric coarse particles Errors in the estimation of fine particulate matter PM_2.5.（3）Through the WRF model to obtain the near-ground and high-air phenomenon factors and PM₁₀,PM_2.5concentration for correlation analysis,the results show that the near-ground and high-air phenomenon factors in each season have a certain impact on the concentration of particulate matter,in which wind speed and boundary layer height is mainly negatively correlated.The effects of temperature and relative humidity vary from season to season,indicating that in addition to near-surface meteorological factors,high-altitude weather factors are also one of the factors affecting the mass concentration of particulate matter.（4）Through establishing single linear regression model of AOD-particulate matter,multiple linear regression model of AOD-particulate matter with the introduction of near surface meteorological factors,multiple linear regression model and random forest model of AOD-particulate matter with the introduction of near surface and high altitude meteorological factors,the estimation results of each model are verified.It is found that with the continuous introduction of near surface and high altitude meteorological factors,the correlation of the models is improved.The coefficient and error are better,and the estimation effect of random forest model is better than that of multiple linear regression model.