Estimation of monthly 1 km resolution PM2.5 concentrations using a random forest model over 2+26 cities, China

Using satellite observations to estimate PM2.5 concentrations is feasible for monitoring air pollution, which can make up for the deficiencies of sparse ground monitoring stations and short time monitoring data. A Random Forest model (denoted as RF), incorporating the latest aerosol optical depth product (MCD19A2), the meteorological data of European Centre for Medium Range Weather Forecasts (ECMWF) and measured PM2.5 concentrations variables, was constructed to estimate PM2.5. The RF model performs significantly well with a coefficient of determination (R-2) of 0.88, a root-mean-square error (RMSE) of 11.94 mu g/m(3), and a low BIAS of 0.3 mu g/m(3). Based on the derived 0.01 degrees x 0.01 degrees PM2.5 distribution, it indicated that the trend of PM2.5 concentrations of the Beijing-Tianjin-Hebei region and its surrounding areas (2 + 26 cities) decreased with obvious spatiotemporal variations from 2002 to 2018. There were two inflection points around 2007 and 2013, benefiting from emission control in China. PM2.5 pollution is worst in winter. Meanwhile, monthly PM2.5 concentrations displayed a U-shaped pattern. This study exhibited long-term spatiotemporal variation characteristics of PM2.5 concentrations and there was a reference significance to the prevention of air pollution in this region.

Automated summary

Using satellite observations to estimate PM2.5 concentrations is feasible for monitoring air pollution, with a Random Forest model performing well in studying the long-term spatiotemporal variations of PM2.5 concentrations in the Beijing-Tianjin-Hebei region and its surrounding areas. The study found a decreasing trend in PM2.5 concentrations from 2002 to 2018, with the worst pollution occurring in winter and a U-shaped pattern in monthly concentrations.