Machine-Learning Source Apportionment of Particulate Pollution Aids Urban Emission Regulations.

Area of Science:

• Environmental Science
• Atmospheric Chemistry
• Data Science

Background:

• Urban particulate pollution (PM2.5) poses significant health risks, necessitating accurate source identification for effective mitigation.
• Current PM2.5 source apportionment methods are often data-intensive, complex, and computationally demanding.
• Developing efficient and accurate models is crucial for improving urban air quality and public health.

Purpose of the Study:

• To introduce a novel machine learning (ML)-based source apportionment model for near-real-time PM2.5 tracking and quantification.
• To assess the model's generalization potential using data from the Pearl River Delta (PRD) and California over two decades.
• To analyze distinct PM2.5 source trends in Shenzhen and Los Angeles and their underlying socio-environmental drivers.

Main Methods:

• Leveraged multiscale aerosol composition data with a machine learning approach.
• Developed and validated source apportionment models using two decades of observational data from PRD and California.
• Employed ML models to replicate receptor-model-based source apportionment results for efficient analysis.

Main Results:

• Identified dominant PM2.5 sources: secondary sulfate and vehicle emissions in PRD; vehicle emissions, secondary nitrate, and biomass burning in California.
• Observed a significant PM2.5 decline in Shenzhen due to anthropogenic source control.
• Noted a flattened PM2.5 trend in Los Angeles, influenced by increased wildfire pollution.

Conclusions:

• ML models offer an efficient alternative to traditional methods for near-real-time PM2.5 source apportionment.
• Distinct urban pollution trends highlight the impact of local socio-environmental factors and policy interventions.
• Findings support data-driven policy-making for meeting stringent air quality standards and mitigating health risks.