Explainable ensemble machine learning revealing enhanced anthropogenic emissions of particulate nitro-aromatic compounds in eastern China

Abstract. Nitro-aromatic compounds (NACs) are important atmospheric pollutants that impact air quality, atmospheric chemistry, and human health. Understanding the relationship between NACs formation and key environmental driving factors are crucial for mitigating their environmental and health impacts. In this work, we combined an ensemble machine learning (EML) model with the SHapley Additive exPlanation (SHAP) and positive matrix factorization (PMF) model to identify the key driving factors for ambient particulate NACs covering primary emissions, secondary formation, and meteorological conditions based on field observations at urban, rural, and mountain sites in eastern China. The EML model effectively reproduced ambient NACs and recognized that anthropogenic emissions (i.e., coal combustion, traffic emission, and biomass burning) were the most important driving factors, with the total contribution of 49.3 %, while significant influences from meteorology (27.4 %), and secondary formation (23.3 %) were also confirmed. Seasonal variations analysis showed that direct emissions presented positive responses to NACs concentrations in spring, summer, and autumn, while temperature had the largest impact in winter. By evaluating NACs formation and loss under various locations in winter, we found that anthropogenic sources played a dominant role in increasing NACs levels in urban and rural sites, while reduced ambient temperature along with secondary formation from gas-phase oxidation was the main reason for relatively high particulate NACs levels at the mountain site. This work provides a reliable modelling method for understanding the dominant sources and influencing factors for atmospheric NACs and highlights the necessity of strengthening emission sources controls to mitigate organic aerosol pollution.