Explainable Machine Learning diagnosis of Ozone Formation Sensitivity in China: Spatiotemporal Evolution and Driver Attribution

Abstract. Accurate diagnosis of ozone (O₃) formation sensitivity (OFS) is crucial for effective control strategies, but a long-term, observation-based, interpretable assessment disentangling the roles of meteorology and emissions at the national scale is lacking. This study integrates OMI tropospheric columns of nitrogen dioxide (NO₂) and formaldehyde (HCHO) from 2005 to 2023, using the HCHO/NO₂ ratio (FNR) as a proxy to track the spatiotemporal evolution of OFS in China. We develop an explainable machine learning framework coupling Random Forest (RF) and SHapley Additive exPlanations (SHAP) to quantify the contributions of meteorology and emissions at regional scales. Our findings reveal a policy-driven phase reversal in OFS: from 2005 to 2012, rising NO₂ columns shifted much of China from NO_x-limited to VOC-limited or transitional regimes. Post-2013, the Clean Air Actions led to a decline in NO₂ and a modest increase in HCHO, triggering a nationwide return to NO_x-limited conditions, especially in eastern China. Regionally, the Sichuan Basin (SCB) remained NO_x-limited, the Pearl River Delta (PRD) transitioned rapidly to NO_x-limited, and the Beijing-Tianjin-Hebei (BTH), Yangtze River Delta (YRD), and Fenwei Plain (FWP) showed gradual shifts from VOC- to NO_x-limited regimes. SHAP analysis identifies temperature and surface shortwave radiation as dominant meteorological drivers, while emission patterns vary regionally: non-methane volatile organic compounds (NMVOCs) dominate in BTH, NO_x in PRD, and carbon monoxide (CO) amplifies radical cycling in FWP, YRD, and SCB. These results support a “climate-dominated, emission-modulated” framework for OFS restructuring, offering a transferable diagnostic tool for differentiated O₃ control strategies.