Linking Particle Number Size Distribution and Optical Properties to Winter Aerosol Sources in Interior Alaska: An XGBoost Classification Approach

Wintertime air quality in Interior Alaska is severely degraded by residential heating emissions and persistent temperature inversions. Quantifying the relative influence of biomass-burning versus fossil-fuel combustion remains challenging because aerosol microphysics, mixing state, and meteorology are tightly coupled. Here, we present observations from four consecutive winter campaigns in North Pole, Alaska, featuring high-resolution aerosol particle number size distribution (PNSD) measured using a Scanning Electrical Mobility Spectrometer (SEMS), collocated with an Aethalometer and a Total Carbon Analyzer. We use the Absorption Ångström Exponent (AAE) as an optical proxy to distinguish combustion regimes separating fossil-fuel-leaning conditions (AAE < 1.6) from biomass-burning-leaning conditions (AAE > 1.6) and examine how PNSD structure and size-optics relationships vary across winters. Across all campaigns, particle populations exhibit a robust U-shaped diurnal pattern in geometric mean diameter (Dg), with larger, growth- and aging-dominated particles at night and smaller modes near midday, consistent with enhanced daytime atmospheric mixing and dilution. Combined analyses show that AAE increases with Dg, consistent with growth toward larger, more optically absorptive particles during biomass-burning-influenced periods. To operationalize source-regime identification, we trained an XGBoost classifier using PNSD-derived size metrics and shape features, meteorological variables, and temporal features to predict binary AAE regimes, validated with a nested leave-one-campaign-out design that tests generalization to an entirely unseen winter. The model achieved strong discrimination across held-out campaigns (pooled ROC-AUC ≈ 0.78; PR-AUC ≈ 0.90), with performance varying by campaign in a pattern consistent with the diminishing prevalence of the fossil-fuel regime in later winters, enabling scalable probabilistic labeling of winter pollution episodes for exposure assessment and mitigation planning. Overall, these results demonstrate an integrated measurement and analysis framework that links size-resolved aerosol microphysics and optical properties to characterize dominant wintertime aerosol regimes in sub-Arctic environments.