PM2.5 estimation and its relationship with NO2 and SO2 in China from 2016 to 2020

The distribution of surface particulate matter (PM2.5) exhibits significant spatiotemporal patterns, especially with a spatial clustering effect. Therefore, resolving spatial characteristics is essential in the modeling process. This study proposes a space-time varying coefficients (STVC-LG) model by considering spatial effects from the perspective of eigenvector spatial filtering and incorporating spatial effects and temporal features into the LightGBM model to estimate ground PM2.5. A site-based cross-validation result of the estimated daily 1 km PM2.5 concentration in China from 2016 to 2020 shows that the proposed model offers high accuracy, an R-2 of 0.88, and an RMSE of 12.21 mu g/m(3) (17% and 32% enhanced compared to the original LightGBM model). It uses a series of spatial eigenvectors to describe different spatial patterns. It then constructs spatial interaction terms by combining the eigenvectors and influencing variables, allowing the values of the variable to vary spatially with the eigenvectors. The obtained PM2.5, with a decreasing trend, had a clear distributional consistency with NO2.5 and SO2.5 according to the co-occurrence distribution maps, particularly in winter. In areas where high concentrations co-occur, both NO2.5 and SO2 concentrations are Granger-causing PM2.5 concentrations from a statistical perspective.