Space-Time Linear Mixed-Effects (STLME) model for mapping hourly fine particulate loadings in the Beijing-Tianjin-Hebei region, China

Fine particulate matter with aerodynamic diameters <= 2.5 mu m (PM2.5) is a proxy for atmospheric pollution levels and is detrimental to human health. The linear mixed-effects (LME) model has been extensively utilized to map regional and hourly PM2.5 levels in the daytime through geostationary satellite-derived aerosol retrievals. However, further improving the performance of the model is difficult because of its limited attention to the spatial and temporal heterogeneity of different predictors that in turn restricts its application to large-scale regions. Using a space-time LME (STLME) model, this study aims to produce an hourly PM2.5 map for the Beijing-Tianjin-Hebei region on the basis of advanced Himawari-8 image aerosol retrievals. In situ ground PM2.5 observations and meteorological and geographical variables are utilized to obtain hourly estimations of PM2.5 concentrations in 2018. Three 10-fold cross-validation (CV) methods, namely, temporal-, spatial-, and sample-based CV, are employed for validation. The results reflect the advantages of the STLME model over the traditional LME model, including its high determination coefficient of (0.83 versus 0.68), small root-mean-square error of (20.9 mu g/m(3) versus 28.1 mu g/m(3)), and minimal mean absolute error of (13.0 mu g/m(3) versus 18.3 mu g/m(3)). Thus, the STLME model shows great potential for air pollution applications because of its effective mapping of surface PM2.5 levels by coordinating the space-time information of different predictors. (C) 2021 Elsevier Ltd. All rights reserved.

Automated summary

This study aims to produce an hourly PM2.5 map for the Beijing-Tianjin-Hebei region using a space-time LME (STLME) model based on Himawari-8 image aerosol retrievals. The STLME model shows better performance compared to the traditional LME model, with higher determination coefficient, smaller root-mean-square error, and minimal mean absolute error. The model has great potential for air pollution applications due to its effective mapping of surface PM2.5 levels by coordinating the space-time information of different predictors.