Drivers of biogenic secondary organic aerosols in Eastern China: Evidence from machine learning and high-resolution...

Biogenic secondary organic aerosol (BSOA) represents a critical component of PM_2.5, and exerts a significant impact on air quality and climate change through radiative forcing and cloud condensation nuclei activity. However, BSOA formation mechanisms remain poorly understood due to complex interactions between biogenic precursors and atmospheric conditions. Conventional filter-based studies lack the high temporal resolution to capture the dynamic atmospheric processes. Here, we employed a thermal desorption aerosol gas chromatography to achieve high-temporal-resolution (2-hour) measurements of BSOA tracers in Zibo, Eastern China, during summer and winter seasons. A novel framework integrating principal component analysis and non-negative matrix factorization with machine learning models (extreme gradient boosting and random forest) was developed to identify the drivers of BSOA formation. Our results quantified the key drivers of BSOA formation: for monoterpene-derived SOA, secondary inorganic aerosols and biogenic emissions were the top contributors, with mean absolute SHAP values of 3.2 and 6.1 ng m^-3, respectively. For isoprene-derived SOA, biogenic emissions, and gas-phase oxidation, nighttime production was significant. The contribution of biomass burning to sesquiterpene-derived SOA was much higher than its contribution to isoprene-derived SOA or monoterpene-derived SOA. This study demonstrates the power of ML by leveraging high-resolution measurements to elucidate the anthropogenic enhancement of BSOA, providing valuable insights for targeted PM_2.5 mitigation strategies in urbanizing regions.