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

Reducing urban particulate pollution is essential for improving air quality and reducing health risks. Achieving this objective requires accurate identification of the sources contributing to particulate matter (PM2.5), but existing approaches are often limited by large data requirements, technical complexity, and computational burdens. This work introduces a machine learning (ML)-based source apportionment model that leverages multiscale aerosol composition data to achieve near-real-time tracking and accurate quantification of PM2.5 sources. Models developed from two-decade observations in the Pearl River Delta (PRD), China and California, show strong generalization potential, identifying secondary sulfate and vehicle emissions as dominant PM2.5 sources in PRD and vehicle emissions, secondary nitrate, and biomass burning in California. The ML models reveal distinctive trends of pollution sources in two megacities driven by different socio-environmental factors. Shenzhen, China, experienced a significant PM2.5 decline over the past decade due to effective control of anthropogenic sources. In contrast, Los Angeles, United States, exhibited a flattened PM2.5 trend, contributed by intensified wildfire pollution. The findings emphasize potential bottlenecks in further urban emission reductions to meet increasingly stringent PM2.5 standards and demonstrate the ability of ML models to efficiently replicate receptor-model-based source apportionment results, supporting near-real-time source analysis and integration into policy-making.