MAIAC-based long-term spatiotemporal trends of PM2.5 in Beijing, China

Satellite-driven statistical models have been proven to be able to provide spatially resolved PM2.5 estimates worldwide. The North China Plain has been suffering from severe PM2.5 pollution in recent years. An accurate assessment of the spatiotemporal characteristics of PM2.5 levels in this region is crucial to design effective air pollution control policy. Our objective is to estimate daily PM2.5 concentrations at 1 km spatial resolution from 2004 to 2014 in Beijing and its surrounding areas using the Multi-angle implementation of atmospheric correction (MAIAC) aerosol optical depth (AOD). A high-performance three-stage model was developed with AOD, meteorological, demographic and land use variables as predictors, which includes a custom-designed PM2.5 gap-filling method. The 11-year average annual coverage increased from 177 days to 279 days and annual PM2.5 prediction error decreased from 14.1 mu g/m(3) to 8.3 mu g/m(3) after gap-filling techniques were applied. Results show that the 11-year overall mean of predicted PM2.5 was 67.1 mu g/m(3) in our study domain. The cross-validation R-2 value of our model is 0.82 in 2013 and 0.79 in 2014. In addition, the models predicted historical PM2.5 concentrations with relatively high accuracy at the seasonal and annual levels (R-2 ranged from 0.78 to 0.86). Our long-term PM2.5 prediction filled the gaps left by ground monitors, which would be beneficial to PM2.5 related epidemiological studies in Beijing. (c) 2017 Elsevier B.V. All rights reserved.