Enhancing Differentiation of Oxygenated Organic Aerosol: A Machine Learning Approach to Distinguish Local and Transboundary Pollution

Accurate source apportionment of particulate matter (PM), especially of organic aerosol (OA), is crucial for targeted mitigation efforts. Positive Matrix Factorization (PMF) is powerful in source attribution of primary OA (POA); however, it often struggles to differentiate sources of oxygenated OA (OOA) due to their similar chemical profiles. In this study, a support vector regression machine learning (ML) model was developed to enhance the OOA source apportionment in Dublin from 2016 to 2023. Rolling PMF analysis identified four POA factors and differentiated OOA into less- and more-oxidized (LO-OOA and MO-OOA), highlighting the significant role of the OOA (47–74% of total OA). The ML model further distinguished locally produced OOA (LO-OOA_local and MO-OOA_local) from transboundary transport OOA and exhibited robust performance across different pollution scenarios. The relative importance analysis revealed that LO-OOA_local was more impacted by fossil fuel emissions like hydrocarbon-like OA (20%) and coal (14%), whereas MO-OOA_local was most influenced by LO-OOA (17%), providing insights into their sources and formation mechanisms. During a mixed pollution episode, the results show that despite the significant contribution of transboundary transport, local heating emissions were more critical sources of OA, with local OA accounting for 68% of total OA and reaching 78% during heating hours. These findings highlight the ongoing need to reduce local emissions to achieve cleaner air in Dublin. The ML model’s ability to quantitatively separate local and transboundary OOA offers invaluable insights for future air quality regulations.

Machine learning is applied to distinguish local and transboundary contributions to oxygenated organic aerosol in Dublin, providing quantitative insights into air quality regulations.