Inversion of PM2.5 concentration in Beijing based on satellite remote sensing and meteorological reanalysis data

With the acceleration of economic development and urbanization, urban smog has become more frequent, and the scope of smog has been expanding, showing the characteristics of long time and regionalization. The smog not only reduces the visibility of the atmosphere, but also seriously harms people's physical and mental health. In recent years, human activities have caused increased emissions of aerosols and other pollutants, and many smog weathers have caused great threats to the ecological environment and human health, seriously affecting urbanization. The smog mainly includes two pollutants, PM_2.5 and PM₁₀. The traditional method of monitoring PM_2.5 is to set up a ground monitoring station. The method has accurate results and can obtain real-time high time resolution data. The defect is not available. Continuous wide-area PM_2.5 spatiotemporal distribution data, while satellite remote sensing technology has the advantages of wide coverage area, continuous observation time, accurate data acquisition, etc., which can be used to invert the continuous concentration of PM_2.5 in urban scale space. This paper takes Beijing as the research area, fully considers the effects of aerosol optical thickness and meteorological factors, and uses the 2017 MODIS 3km AOD product and ECMWF-ERA5 meteorological reanalysis data, combined with the PM_2.5 data of the air quality monitoring site, based on randomization. Forest, multiple linear regression, support vector machine and neural network method were used to invert PM_2.5 concentration near the ground. The results show that the random forest has the highest accuracy and best effect, and its mean absolute error (MAE) is 12.08 μg/m³, mean square. The root error　 (RMSE) is 18.42 μg/m³. For the importance analysis of the independent variables, the aerosol optical thickness has the greatest influence on the PM_2.5 inversion model, followed by relative humidity, boundary layer height and 2m temperature. The random forest model was used to invert the PM_2.5 concentration in the non-heating period (April-November), and the spatial and temporal analysis was carried out. The results showed that the monthly mean concentration of PM_2.5 was significantly different; the spatial distribution was higher in the central, southern and eastern regions. The lower characteristics of the north, northeast and southwest; the magnitude of the change is larger in the central and southern parts, and smaller in the northwest and southwest.