Space-time mapping of ground-level PM2.5 and NO2 concentrations in heavily polluted northern China during winter using the Bayesian maximum entropy technique with satellite data

The accurate and informative space-time mapping of air pollutants is a crucial component of many human exposure studies. In the present work, space-time maps of daily distributions of PM_2.5 and NO₂ concentrations were generated in the severely polluted northern China region using the Bayesian maximum entropy (BME) method. This method can incorporate hard PM_2.5 and NO₂ data (obtained at ground-level monitoring sites), and various kinds of soft (uncertain) data, including satellite data processed in terms of machine learning techniques, meteorological variables, and geographical predictors. The BME maps of space-time PM_2.5 and NO₂ concentrations over northern China generated during the winter season (when severe haze episodes occur frequently) were realistic and informative. As regards their numerical accuracy, for the space-time PM_2.5 estimates, the tenfold cross-validation R ² and the RMSE were, respectively, 0.86 and 14.37 μg/m³; for the space-time NO₂ estimates, the R ² and RMSE values were, respectively, 0.85 and 6.93 μg/m³. Lastly, it was shown that the BME method performed better than the mainstream spatiotemporal ordinary kriging technique in terms of the higher R ² values of both the predicted PM_2.5 and NO₂ concentration maps.