A Generalized Additive Model Combining Principal Component Analysis for PM2.5 Concentration Estimation

As an extension of the traditional Land Use Regression (LUR) modelling, the generalized additive model (GAM) was developed in recent years to explore the non-linear relationships between PM2.5 concentrations and the factors impacting it. However, these studies did not consider the loss of information regarding predictor variables. To address this challenge, a generalized additive model combining principal component analysis (PCA–GAM) was proposed to estimate PM2.5 concentrations in this study. The reliability of PCA–GAM for estimating PM2.5 concentrations was tested in the Beijing-Tianjin-Hebei (BTH) region over a one-year period as a case study. The results showed that PCA–GAM outperforms traditional LUR modelling with relatively higher adjusted R2 (0.94) and lower RMSE (4.08 μg/m3). The CV-adjusted R2 (0.92) is high and close to the model-adjusted R2, proving the robustness of the PCA–GAM model. The PCA–GAM model enhances PM2.5 estimate accuracy by improving the usage of the effective predictor variables. Therefore, it can be concluded that PCA–GAM is a promising method for air pollution mapping and could be useful for decision makers taking a series of measures to combat air pollution.