Estimating ground-level PM_2.5 concentrations in Beijing, China using aerosol optical depth and parameters of the...

The aerosol optical depth (AOD) is widely used to predict surface PM_2.5 (particles with an aerodynamic diameter smaller than 2.5μm) concentrations using regression methods, as a way to supplement observations from sparse ground PM_2.5 monitoring networks. Several meteorological parameters, such as surface humidity, temperature and height of planetary boundary layer (HPBL), are usually combined with AOD to improve the accuracy of the regression model in predicting PM_2.5 concentrations. In this paper, we investigate the role of the temperature inversion layer in the prediction of PM_2.5 concentrations by the optimal subset regression method. The result indicates that the optimal subset regression model with the parameters of the depth and temperature difference of the inversion can significantly improve the accuracy of the predictions of surface PM_2.5 concentrations, compared with the original regression model with one factor of AOD. The determination coefficient (R²) increases from 0.51 to 0.63, and the Root Mean Square Error (RMSE) decreases from 40.38 to 35.45μg/m³. The optimal subset regressions were also built for each season. The temperature difference of the inversion is introduced into the autumn and winter optimal subset regression, and the depth of the inversion is introduced into the spring optimal subset regression. The contribution of the inversion parameters to the regression model is affected by the different type of the temperature inversion layer present in each season.