Unraveling the Multifactor Predictors of Urban Ozone in China: A Machine Learning Framework for Synergistic Pollution and Greenhouse Gas Mitigation

Abstract As particulate matter pollution declines across Chinese cities, ozone (O3) has emerged as a predominant air quality challenge. This study develops an interpretable, city-scale machine learning (ML) framework to characterize long-term O3 trends, quantify the relative contributions of multifactor predictors, and evaluate synergistic control strategies across China. Integrating feature selection, ensemble multiple ML models, Bayesian optimization, and SHapley Additive exPlanations (SHAP) analysis, we revealed significant spatiotemporal heterogeneity in O3 formation regimes. Northwest China (NWC), the Beijing-Tianjin-Hebei urban agglomeration (BTH), and the Yangtze River Delta (YRD) were identified as persistent high-concentration regions. Quantitative attribution indicates that meteorological factors dominate O3 variability (explaining ∼55%), followed by air pollutants (∼35%), while greenhouse gases (GHGs) and vegetation jointly contribute approximately 10%. This underscores a meteorology-dominated yet multifactor-coupled formation mechanism. Scenario simulations demonstrate that coordinated regulation of policy-relevant factors can substantially mitigate O3 concentrations. Specifically, optimized strategies yield regional average reductions of 12.36 and 10.90 μg/m3 in NWC and Northern Xinjiang (NXJ), 7.20 and 6.92 μg/m3 in BTH and the Triangle of Central China (TCC), and 4.22–7.24 μg/m3 in other major regions. These findings call for integrated strategies that move beyond single-pollutant control toward region-specific, multifactor governance frameworks integrating pollutant control, GHG mitigation, and ecological regulation. Such an approach is critical for advancing O3 management in synergy with China’s broader goals of pollution reduction and GHG mitigation.