Evaluating drivers of PM_2.5 air pollution at urban scales using interpretable machine learning

Area of Science:

• Environmental Science
• Data Science
• Urban Planning

Background:

• Urban fine particulate matter (PM2.5) pollution hinders China's Sustainable Development Goals.
• Identifying PM2.5 drivers is crucial for effective pollution reduction strategies.

Purpose of the Study:

• Develop and validate a machine learning model to analyze urban PM2.5 concentrations.
• Identify key socioeconomic and industrial drivers of PM2.5 pollution.
• Propose city-specific strategies for PM2.5 reduction.

Main Methods:

• Utilized a combined CatBoost and Tree-Structured Parzen Estimator (TPE) machine learning model.
• Employed SHapley Additive exPlanations (SHAP) to determine factor importance.
• Analyzed PM2.5 data from 297 Chinese cities between 2000 and 2021.

Main Results:

• The model achieved high accuracy (R² = 96.44%) in predicting urban PM2.5.
• Socioeconomic factors and industrial activities were identified as primary drivers.
• Nitrogen oxide emissions, technology investment, population density, and energy consumption were significant contributors in 2000.

Conclusions:

• Targeted strategies, focusing on population density and mining development, are recommended for future pollution control.
• The framework provides a robust tool for evaluating pollution factors and tailoring reduction strategies.
• Significant improvements in air quality were observed due to reduced nitrogen oxide emissions.