Machine learning-based retrieval of aerosol size and hygroscopicity using horizontal scanning LiDAR and PM data

Hygroscopic growth of aerosols significantly affects radiative forcing and visibility, yet remains challenging due to the interplay among size, composition, and humidity. This study integrates in-situ PM and lidar data to address discrepancies between dry mass concentrations and wet optical measurements. Using machine learning inversion (XGBoost, R 2 = 0.98), dry-state size distributions were retrieved from PM data. Mie theory was applied to derive the dry extinction coefficient, and the lidar-based wet extinction coefficient yielded the hygroscopic growth. Aerosol types were classified using Random Forest (accuracy 83.4%), revealing dominance of coarse hygroscopic aerosols in this coastal urban region. Optical response varied with hygroscopicity: wet extinction coefficient increased with RH for hydrophilic types but remained low for hydrophobic aerosols. Notably, clean conditions occasionally showed a sharp increase in wet extinction coefficient despite low PM, highlighting limitations of mass-only assessments. This approach suggests improved aerosol characterization algorithms considering size and hygroscopicity, supporting advanced air quality and climate modeling studies.