Optimization of seasonal geographically and temporally weighted regression model for accurate estimation of seasonal PM2.5 concentrations in Beijing-Tianjin-Hebei Region

Abstract

Particulate matter with a diameter of less than 2.5 µm (PM2.5) has a significant impact on air pollution, atmospheric visibility, and human health. The most basic and important step of regional air pollution control is to obtain air pollution data in different seasons from both satellite sensors and ground-level observations. The aim of this paper is to accurately estimate the PM2.5 concentration in the Beijing-Tianjin-Hebei urban area in different seasons by establishing a seasonal geographically and temporally weighted regression (S-GTWR) model that integrates multiple complex factors. Using a greedy algorithm, the model results were optimized by selecting the characteristic variables that contributed to the accuracy of the model in different seasons. The measured and estimated PM2.5 concentrations were compared and the cross-validation results were used as a basis for evaluating the accuracy of the model. The results showed that the accuracy of the S-GTWR model that combined the optimal characteristic variables was higher than that of the geographically weighted regression (GWR) model and the kriging method. The mean prediction error (ME), relative prediction error (RPE), and root mean square error (RMSE) of the S-GTWR model were small, and the coefficient of determination (R2) of the model exceeded 0.86 for each season. The accuracy of the S-GTWR model in estimating the PM2.5 concentration was highest in summer and lowest in winter. In addition, the proposed model can accurately estimate PM2.5 concentrations in areas without monitoring sites. The results can provide a scientific basis for the study of pollution control and PM2.5 exposure in large urban agglomerations.