Space-Time Estimations and Mapping of PM2.5 Fine Particulates Based on Multi-source Data

Abstract

PM2.5 pollution in China has become an extreme environmental and social problem and has generated widespread public concern. We estimate ground-level PM2.5 from satellite-derived aerosol optical depth (AOD), topography data, meteorological data, and pollutant emissions using a new technique, Bayesian maximum entropy (BME) combined with geographically weighted regression (GWR), to evaluate the spatial and temporal characteristics of PM2.5 exposure in an eastern region of China in winter. The overall 10-fold cross-validation R2 is 0.92, and the root mean squared prediction error (RMSE) is 8.32 μg·m-3. The mean prediction error (MPE) of the predicted monthly PM2.5 is -0.042 μg·m-3, the mean absolute prediction error (MAE) is 4.60 μg·m-3. Compared with the results of the Geographically Weighted Regression model-GWR (R2=0.71, RMSE=15.68 μg·m-3, MPE=-0.095 μg·m-3, MAE=11.14 μg·m-3), the prediction by the BME were greatly improved. In this location, the high PM2.5concentration area is mainly concentrated in North China, the Yangtze River Delta, and Sichuan Basin. The low concentration area is mainly concentrated in the south of China, including the Pearl River Delta and southwest of Yunnan. Temporally, there is migration trend from the coastal areas inland, and PM2.5 pollution is most serious in December 2015 and January 2016. It is relatively low in November 2015 and February 2016.