Prediction of Ambient PM2.5 Chemical Components in Southern California Using Machine Learning

Fine particulate matter (PM2.5, particulate matter with an aerodynamic diameter ≤2.5 μm) poses major public health and environmental risks, but understanding the toxicity of its components remains limited due to sparse chemical speciation data. This study applies machine learning (XGBoost) to predict key PM2.5 components (organic carbon, elemental carbon, nitrate, sulfate, ammonium, and metals) using readily available predictors, including total PM2.5 mass concentrations, meteorological data, trace gas measurements, and records of exceptional events (e.g., wildfires, fireworks). Drawing on a decade of observations from two Southern California monitoring sites (Los Angeles and Rubidoux), the models demonstrate strong predictive performance for key components, including nitrate, ammonium, and organic carbon. Total PM2.5 concentration, relative humidity, and boundary layer height emerged as important predictors. This approach has potential applications in informing satellite remote sensing missions, improving chemical transport models, and providing cost-effective estimates of PM2.5 chemical components during sampling gaps and in regions lacking frequent monitoring. Further research is needed to extend applicability across varied geographic and climatic settings.