New methodology to quantify source contributions to nitrate, sulfate and ammonium using PMF-AI

Identifying inorganic aerosol (IA), the major component of PM_2.5, which can originate from both primary and secondary sources, is challenging without significant tracers for sources. Furthermore, there is a lack of studies, specifically investigating a novel method (i.e., PMF-AI (artificial intelligence)) to discern the contribution of each IA source. However, effectively reducing IA levels requires identifying their sources. In this study, measurements of PM_2.5-associated heavy metals in Kaohsiung, Taiwan, were utilized to (1) identify heavy metal sources using the PMF model, (2) quantify the source contributions to each individual IA in PM_2.5 using the PMF-AI technique and (3) analyze source contributions of each IA with domain knowledge using Shapley additive explanation (SHAP). The current analysis suggested that the PMF-AI model can account for up to 99 % of the total variance in daily IA values. SHAP values indicated that oil combustion and biomass burning/waste incineration/agrochemical applications (the main sources) significantly influenced IA concentrations in Kaohsiung among all characteristics. The novel approach utilized in this study constitutes a crucial initial step in shaping future policies and regulations and is adaptable to regions facing IA-related issues.