Quantifying drivers contributing to air quality improvements is crucial for pollution prevention and optimizing local policies. Despite advances in machine learning for air quality analysis, their limited interpretability hinders attribution on global and local scales, vital for informed city management. Our study introduces an innovative framework quantifying socioeconomic and natural impacts on mitigation of particulate matter pollution in 31 Chinese major cities from 2014 to 2021. Two indices, formulated based on the additivity of Shapley additive explanations, are proposed to measure driver contributions globally and locally. Our analysis explores the self-contained and interactive effects of these drivers on particulate levels, pinpointing critical threshold values where these drivers trigger shifts in particulate matter levels. It is revealed that SO2, NOx, and dust emission reductions collectively account for 51.58 % and 51.96 % of PM2.5 and PM10 decreases at the global level. Moreover, our findings unveil a significant heterogeneity in driver contributions to pollutant mitigation across distinct cities, which can be instrumental in crafting location-specific policy recommendations.
Keywords: Chinese major cities; Interpretable machine learning; Natural and socioeconomic drivers; Particulate matter pollution; Quantitative attribution.
Copyright © 2024 Elsevier B.V. All rights reserved.