Daily estimation of ground-level PM2.5 concentrations at 4 km resolution over Beijing-Tianjin-Hebei by fusing MODIS AOD and ground observations

The satellite-borne Moderate Resolution Imaging Spectroradiometer (MODIS) aerosol optical depth (AOD) is widely used to estimate ground-level fine ambient particulate matter (PM2.5) concentrations to evaluate their health effects. The associated estimation accuracy is often reduced by AOD missing values and by insufficiently accounting for the spatio-temporal PM2.5 variations. In this study, we aim to estimate ground-level PM2.5 concentrations at a fine resolution with improved accuracy by fusing fine-scale satellite and ground observations in the populated and polluted Beijing-Tianjin-Hebei (BTH) area of China in 2014. We employed a Bayesian-based statistical downscaler to model the spatio-temporal linear AOD-PM2.5 relationships. We used a 3 km MODIS AOD product, which was resampled to a 4 km resolution in a Lambert conic confortnal projection, to assist comparison and fusion with predictions by atmospheric chemistry models. A two-step method was used to fill the missing AOD values to obtain a full AOD dataset with complete spatial coverage. The downscaler has a good performance in the fitting procedure (R-2 = 0.75) and in the cross Validation procedure (R-2 = 0.58 by random method and R-2 = 0.47 by city-specific method). The number of missing AOD values Was serious and related to elevated PM2.5 concentrations. The gap-filled AOD values corresponded well with our understanding of PM2.5 pollution conditions in BTH. The prediction accuracy of PM2.5 concentrations were improved in terms of their annual and seasonal mean. As a result of its fine spatio-temporal resolution and complete spatial coverage, the daily PM2.5 estimation dataset could provide extensive and insightful benefits to related studies in the BTH area. This may include understanding the formation processes of regional PM2.5 pollution episodes, evaluating daily human exposure, and establishing pollution controlling measures. (C) 2016 Published by Elsevier B.V.