An AI-Assisted Positive Matrix Factorization Framework for Exploring Sulfate, Nitrate, and Ammonium Sources

Inorganic aerosols (IAs), sulfate (), nitrate () and ammonium (), constitute a major fraction of PM_2.5 and are strongly influenced by both emission sources and atmospheric processes. In this study, a PMF-artificial intelligence (PMF-AI) framework was developed to identify IA sources and quantify their contributions in Taipei, Taiwan. PM_2.5 samples collected at 6-day intervals from January 2017 to November 2022 were analyzed. Positive Matrix Factorization (PMF) applied to heavy metal data identified five major source categories: sea salt, oil combustion, non-ferrous metallurgy, traffic-related/waste incineration emissions, and coal combustion. These PMF-derived source contributions, together with meteorological variables, were subsequently incorporated into an Automated Machine Learning (AutoML) framework to predict IA concentrations. The developed models showed strong predictive performance, with R² values of 0.94, 0.82, and 0.88 for , , and , respectively. Source attribution results indicated that coal combustion was the dominant contributor to (37%) and (37%), while was primarily influenced by coal combustion (23%) and meteorological variability (33%). Coal combustion and oil combustion also showed important contributions, particularly for and formation. These findings demonstrate that integrating PMF with machine learning can improve understanding of the combined influences of emission sources and meteorological conditions on IAs variability.