Source-specific control leverage for reducing oxidative potential of PM_2.5 revealed by explainable machine learning

Oxidative potential (OP) of fine particulate matter (PM_2.5) has been increasingly recognized as a health-relevant metric beyond mass concentration. However, the mechanisms governing OP and their implications for source-oriented emission control remain difficult to resolve due to nonlinear interactions among chemical components and emission sources. In this study, year-long observations of PM_2.5 chemical composition and OP in central Taiwan were analyzed to investigate the nonlinear drivers of OP variability and their implications for mitigation strategies. Explainable machine learning approaches were integrated with regime-level response mapping to characterize OP responses to variations in chemical species and source contributions. Chemical-species-based analyses reveal nonlinear dependencies and interaction effects involving transition metals, water-soluble organic carbon, and aerosol liquid water, which cannot be captured by linear models. Extending this framework to a source-resolved perspective, positive matrix factorization was used to identify major emission sources, and their regime-dependent influences on OP were quantified through interaction analysis. The results demonstrate that source combinations can amplify or suppress OP under different pollution regimes. Building on these patterns, a source-resolved OP mitigation leverage framework was developed to evaluate the marginal effectiveness and feasibility of reducing OP through source control. Coal and heavy oil combustion and industrial emissions exhibit consistently high OP mitigation leverage, followed by secondary nitrate and secondary sulfate, whereas traffic emission and dust show relatively limited mitigation leverage. This study provides a system-level framework linking OP response dynamics to source-resolved mitigation leverage, offering quantitative insights for toxicity-oriented air quality management beyond conventional PM_2.5 mass control.