Prediction and Source Contribution Analysis of PM2.5 Using a Combined FLEXPART Model and Bayesian Method over the Beijing-Tianjin-Hebei Region in China

Abstract

Fine particulate matter (PM2.5) has a serious impact on human health. Forecasting PM2.5 levels and analyzing the pollution sources of PM2.5 are of great significance. In this study, the Lagrangian particle dispersion (LPD) model was developed by combining the FLEXPART model and the Bayesian inventory optimization method. The LPD model has the capacity for real-time forecasting and determination of pollution sources of PM2.5, which refers to the contribution ratio and spatial distribution of each type of pollution (industry, power, residential, and transportation). In this study, we applied the LPD model to the Beijing-Tianjin-Hebei (BTH) region to optimize the a priori PM2.5 emission inventory estimates during 15–20 March 2018. The results show that (1) the a priori estimates have a certain degree of overestimation compared with the a posteriori flux of PM2.5 for most areas of BTH; (2) after optimization, the correlation coefficient (R) between the forecasted and observed PM2.5 concentration increased by an average of approximately 10%, the root mean square error (RMSE) decreased by 30%, and the IOA (index of agreement) index increased by 16% at four observation sites (Aotizhongxin_Beijing, Beichenkejiyuanqu_Tianjin, Dahuoquan_Xintai, and Renmingongyuan_Zhangjiakou); and (3) the main sources of pollution at the four sites mainly originated from industrial and residential emissions, while power factory and transportation pollution accounted for only a small proportion. The concentration of PM2.5 forecasts and pollution sources in each type of analysis can be used as corresponding reference information for environmental governance and protection of public health.