Evaluating drivers of PM2.5 air pollution at urban scales using interpretable machine learning.

Abstract

Reducing urban fine particulate matter (PM_2.5) concentrations is essential for China to achieve the Sustainable Development Goals (SDGs). Identifying the key drivers of PM_2.5 will enable the development of targeted strategies to reduce PM_2.5 levels. This study introduces a machine-learning model that combines CatBoost and the Tree-Structured Parzen Estimator (TPE) to analyze PM_2.5 concentration across 297 cities between 2000 and 2021. SHapley Additive exPlanations (SHAP) were employed to identify the primary factors influencing urban PM_2.5 concentrations. The study revealed that the proposed model has high accuracy in predicting urban PM_2.5 concentrations, achieving a coefficient of determination (R²) score of 96.44%. Socioeconomic and industrial activity are key drivers of PM_2.5 concentrations. This study not only quantifies the primary factors exacerbating or alleviating pollution for each city or province during the 2000-2021 period but also evaluates the influence of operational factors such as technological and public financial expenditures. In 2000, the main contributors to pollution in four heavily polluted cities included substantial nitrogen oxide emissions, inadequate technology investments, and excessive population density and liquefied gas consumption. Due to the rapid reduction in nitrogen oxide emissions, pollution levels in these cities have improved substantially. In the future, the most effective strategies for pollution reduction in these cities will focus on controlling population density and slowing down mining development. The proposed framework serves as a robust evaluation tool and can propose tailored strategies to control PM_2.5 concentrations effectively in each city.