Unraveling Biological Particles in High-Altitude Terrain with Machine Learning: Implications for Ice Cloud Formation

Biological particles can influence cloud formation and hydrological cycles by acting as efficient ice-nucleating particles (INPs). Their low abundance and complex composition make precise identification difficult, particularly in mountain locations where variable terrain and complex wind patterns create spatiotemporal heterogeneity in particle transport, abundance, and composition. This study integrates single-particle morphological and chemical analysis, and supervised machine learning (SML) to evaluate the abundance and composition of biological particles during the Surface Atmosphere Integrated Field Laboratory campaign on Mt. Crested Butte in Colorado. Single-particle analysis of 124,913 particles combined with a trained SML method showed that biological particles constitute ∼9.5% of total particles, with 38.4% internally mixed with dust and other inorganic constituents. The SML method identified a larger fraction of biological particles than traditional rule-based methods, especially for smaller particles (<4.0 μm). Elemental analysis showed that 3.4% of SML-classified biological particles contained substantial sodium (Na > 1 to 23%) and thus could be misclassified as salt by traditional methods. Furthermore, INP measurements showed heat treatment reduced INPs by 51–88% at −15 °C, coinciding with higher biological particle abundance on those days. These findings confirm the presence of heat-sensitive biological INPs and highlight the episodic nature of biological particle emissions.