Unveiling the source contributions of fine and coarse particulate matter using PM-bound metals and PMF-AI modeling

Particle pollution is a critical global concern with significant implications for public health and the environment. Both fine (PM 2.5 ) and coarse (PM 2.5-10 ) particles exhibit diverse compositions and origins, leading to distinct health and environmental consequences. In this study, K-means clustering was employed to differentiate between local, regional, and long-range transport (LRT) sources, showing that LRT significantly increases PM 2.5-10 levels, leading to a more than 1.26-fold rise in its annual mean concentration. Using Positive Matrix Factorization (PMF) model, we identified five local and regional source of PM 2.5 and four in case of PM 2.5-10 . Further, AutoML model explains up to 70 % and 71 % of the daily variance in PM 2.5 and PM 2.5-10 , respectively. The complex relationship of these sources was explained using SHapley Additive ExPlanations (SHAP). Among the five major factors identified, SHAP analysis reveals that oil combustion (24 %), coal burning (18 %), and non-ferrous metal smelting/biomass burning (17 %) are the predominant contributors to PM 2.5 . In contrast, ocean spray (28 %) is identified as a significant source of PM 2.5-10 pollution followed by oil, non-ferrous metal smelting/biomass burning (20 %) and traffic related emission (14 %). This study offers a novel and comprehensive methodology for identifying the distinct sources of fine and coarse particulate matter . It provides valuable insights that can inform future policies and regulations, particularly in regions facing challenges related to PM pollution. • To quantify the concentrations of PM 2.5 and PM 2.5-10 originating from domestic sources. • A machine learning technique with a clustering algorithm and backward trajectory analysis was used. • The models demonstrated high accuracy, with an R 2 of up to 0.71 for PMs. • Oil combustion and sea salt are the major sources of PM 2.5 and PM 2.5-10 , respectively.