Decoding PM_2.5 oxidative potential in Ningbo, China: Key chemicals, sources, and health risks via dual-assay and machine learning

PM_2.5 oxidative potential (OP), a key driver of health risks, was investigated in Ningbo, China, using dual dithiothreitol (DTT) and ascorbic acid (AA) assays combined with machine learning (ML). This approach accounts for the complexity of interactions among key chemical drivers and accurately identifies chemical species and PM_2.5 sources associated with OP - a critical gap in prior studies relying solely on correlation analysis and linear regression. Year-long PM_2.5 samples revealed higher nighttime and summer OP (volume-based OP-DTTv and OP-AAv), linked to aerosol acidity and photochemical aging. Among six ML models, Extremely Randomized Trees (ERT) outperformed others by 9.5–30.7 %, identifying Cu, Fe, V, As, Co, Cd, NO₃^-, Ni, and quinones as primary OP drivers, with synergistic effects for most constituents except antagonistic Fe. Source apportionment attributed OP mainly to vehicular emissions (40 %), marine/sea salt (20 %), and secondary aerosols (16 %). Biomass burning, industry, and road dust contributed minimally. Results emphasize targeting quinones, traffic-related metals (Cu, V), and synergistic metal interactions to mitigate PM_2.5 toxicity in coastal cities. The dual-assay ML framework provides actionable insights for prioritizing OP-driven regulation, particularly in regions blending anthropogenic and marine influences, to reduce oxidative stress-related health burdens.