A machine learning and box modeling approach to comparing the atmospheric chemistry of high- and low-ozone and PM₂.₅ episodes

High levels of ozone (O₃) and particulate matter (PM) have become significant environmental challenges in urban areas worldwide, particularly in developing countries such as China. A comprehensive understanding of atmospheric chemistry during both polluted and non-polluted periods is essential for effective pollution control strategies. In this study, we analyzed reactive atmospheric gases (VOCs, NOₓ, and O₃) in Nanjing, China, during high and low O₃ and PM₂.₅ episodes. The Framework for 0-D Atmospheric Modeling (F0AM) was employed to simulate OH reactivity, the ROx radical budget, ROx radical chemistry, and O₃-related processes. We found that the contribution of model-derived secondary reactants and unconstrained primary hydrocarbons to total OH reactivity was significantly higher during high ozone episodes (HOE) at 50.5 %, compared to 38.8 % during low ozone episodes (LOE). Aromatic compounds contributed more to OH reactivity during high PM₂.₅ episodes (HPME, 20 %) than during low PM₂.₅ episodes (LPME, 14.5 %). The ROx radical recycling rates were approximately 2.5 to 7.5 times higher than primary production rates during HOE, compared to LOE. Additionally, the RO₂ + NO reaction accounted for a larger share of daily mean O₃ production during HOE (50 %) than during LOE (46 %). Conversely, O₃ photolysis contributed more to daily mean O₃ loss during LOE (37 %) than during HOE (33 %). Finally, a Random Forest model revealed that meteorological and atmospheric chemical conditions together enhanced ozone formation by approximately 63 % (other parts could be explained by the concentrations of precursors). This study sheds light on how various atmospheric conditions and pollutants interact to form secondary pollution, offering valuable insights for more effective pollution control strategies.