Machine Learning Exploring the Double High PM2.5 and O3 Pollution in the Southeast of China During 2018–2022

PM2.5 and O3 double-high pollution is a complex process, depending on precursors’ emissions, atmospheric chemical processes and meteorological factors. In this study, 29 national monitoring sites of air quality in Fujian Province from 2018 to 2022 were selected to explore the spatiotemporal distributions of co-occurring PM2.5 and O3 pollution. We built an XGBoost machine learning model to elucidate the main drivers of PM2.5 and O3 pollution levels. The results showed that the days with double-high pollution (DHP) were primarily occurred in April and September in the coastal areas of Southeast China. The analysis results of SHAP (SHapley Additive exPlanations) values suggested that RH, NO2, PM2.5, U10, T and V10 were the main drivers on high O3 concentration. NO2 was the most significant contributor to PM2.5 levels, followed by O3, T, and RH, suggesting the influence of increased atmospheric oxidation capacity. During the DHP period in 2022, due to abnormal climatic conditions, reduced RH and a significant decrease in PM2.5 and NO2 concentrations contributed to the O3 increase, with model-attributed contributions of 26%, 21% and 11% respectively. Based on SHAP values, high level O3 had a relative contribution of 34% to PM2.5 concentrations, which may be associated with enhanced secondary transformation. Meteorological fluctuations between 2018 and 2022 showed the strongest influence on O3 variability, representing 60% of the total SHAP contribution (corresponding to 3.09 μg/m3). Extreme weather such as heatwaves in 2022 contributed to abnormally high O3 concentration, further facilitating the occurrence of DHP. The study is beneficial for further understanding the interaction mechanism between PM2.5 and O3 in relatively clean coastal environment, and provide new insights for future deepening pollution prevention control for alleviating PM2.5 and O3 complex pollution.