Spatiotemporal estimates of daily PM2.5 concentrations based on 1-km resolution MAIAC AOD in the Beijing–Tianjin–Hebei, China

Abstract

The accurate prediction of PM2.5 concentrations is essential for health risk assessment and the formulation of air pollution prevention and control strategies. In this study, we selected a two-stage statistical regression model, which combined the linear mixed effect (LME) and geographically weighted regression (GWR) model, and applied aerosol optical depth (AOD) with 1-km spatial resolution, meteorological variables and land use parameters as predictors, to predict the daily near-ground PM2.5 concentrations (ρ(PM2.5)) from 2013 to 2017 in the Beijing–Tianjin–Hebei (BTH) region. The model was fitted annually, and the coefficient of determination (R2), mean prediction error (MPE), root-mean-squared prediction errors (RMSPE), and relative prediction error (RPE) ranged of the model's cross-validation results were 0.85–0.94, 8.24−17.78 µg/m3, 13.14−29.90 µg/m3 and 21.42−32.71%, respectively. The LME + GWR model significantly outperformed the LME model, but with a slight increase in overfitting. The ρ(PM2.5) in the BTH displayed obvious temporal (decrease per year, highest in winter and lowest in summer) and spatial (high in southern plains and low in northwestern mountain areas) characteristics. During the investigated period, the annual mean ρ(PM2.5) dropped by 35.5%, and the areas with high concentrations shrank significantly. Moreover, a sharply decline occurred from 2015 to 2017 due to the strict implementation of the government's pollution-preventing policies on pollutant emissions. This study can provide a scientific basis and support for the understanding and prevention of air pollution in the BTH region and beyond.