Long-term ozone formation sensitivity in China: spatiotemporal evolution and machine learning attribution

Abstract. Accurate diagnosis of ozone (O3) formation sensitivity (OFS) is essential for designing effective precursor-control strategies, yet long-term, observation-based, and interpretable national-scale assessments remain limited. Here, we combined OMI satellite observations of tropospheric nitrogen dioxide (NO2) and formaldehyde (HCHO) with ground-based O3 measurements to derive region-specific FNR (HCHO / NO2) threshold ranges from the relationship between FNR and O3 exceedance probability. Based on these thresholds, we characterized the spatiotemporal evolution of OFS across China during the warm season (April–September) from 2005 to 2023 and coupled Random Forest (RF) with SHapley Additive exPlanations (SHAP) to quantify contributions from emission-related and meteorological factors. The results reveal a clear phase reversal in OFS over China. From 2005 to 2012, the national area fraction of nitrogen oxides (NOx)-limited regimes decreased by 7.9 %, while transitional and volatile organic compound (VOC)-limited regimes increased by 5.6 % and 2.3 %, respectively. After 2013, this pattern reversed, and by 2023 NOx-limited regimes expanded to 76.5 % of the polluted area, whereas VOC-limited regimes accounted for only 2.8 %. Regionally, after 2013, both Beijing–Tianjin–Hebei (BTH) and Fenwei Plain (FWP) exhibited transitions between transitional and NOx-limited regimes, while the Yangtze River Delta (YRD) showed a shift toward NOx-limited regimes. Sichuan Basin (SCB) remained predominantly transitional, and Pearl River Delta (PRD) also shifted toward NOx-limited regimes. SHAP analysis shows emission-related variables contributed 50.1 %–69.4 % of total importance, exceeding meteorological factors, which increased after 2013. Overall, OFS evolution is primarily associated with emission changes but increasingly modulated by meteorological conditions under cleaner conditions.