Fu, DS; Song, ZJ; Zhang, XL; Xia, XG; Wang, J; Che, HZ; Wu, HJ; Tang, X; Zhang, JQ; Duan, MZ (2020). Mitigating MODIS AOD non-random sampling error on surface PM2.5 estimates by a combined use of Bayesian Maximum Entropy method and linear mixed-effects model. ATMOSPHERIC POLLUTION RESEARCH, 11(3), 482-490.

PM2.5 estimates solely based on the Moderate Resolution Imaging Spectroradiometer (MODIS) AOD products may lead to a substantial bias because of non-random AOD sampling deficiency in cloudy conditions and swap-gap regions. Furthermore, this non-random sampling issue can be exacerbated in polluted regions where heavy aerosol loadings are likely misclassified into clouds. Here, to mitigate non-random sampling deficiency in MODIS AOD product for surface-level PM2.5 estimates, we have combined Bayesian maximum entropy (BME) method with the Linear Mixed-Effects (LME) model, for the first time, to produce more spatiotemporally complete and precise AOD products and thereafter PM2.5 estimates. This combined BME-LME approach was applied to MODIS and sunphotometer AOD products over the North China Plain. Relative to the standard MODIS AOD product, the integration of MODIS and sunphotometer AOD through BME showed increases in both spatiotemporal completeness (up to 96%) and the quality. The resultant monthly PM2.5 estimates from the BME-LME had a bias of 3.5 mu g m(-3) and a root mean square error (RMSE) of 5.5 mu g m(-3), showing substantial improvement over PM2.5 estimations from original MODIS AOD product (a bias of 84.1 and a RMSE of 112.1 mu g m(-3)). Merging sunphotomter and satellite AOD observations with BME-LME is a prospective method to simultaneously improve AOD and PM2.5 estimates.