Tracing biological, human, and inorganic sources of coarse aerosols via single-particle fluorescence and optical morphology

Large aerosol particles within the coarse mode affect the environment, climate, and human health in ways that strongly depend on particle type. Although this size range is dominated by mineral dust and sea spray aerosol (SSA), less abundant biological particles can exert disproportionate effects, such as triggering ice formation at comparatively warm temperatures. Accurate, type-resolved characterization of coarse-mode aerosols is therefore critical for understanding their environmental and climatic roles. Here, we present a new laboratory-based reference dataset for common coarse-mode aerosol sources, including pollen, dust, bacteria, and microplastics, based on laboratory measurements of single-particle ultraviolet light-induced fluorescence (UV-LIF) spectroscopy and particle morphology. Comparison with existing datasets reveals source-specific fluorescence signatures, but also demonstrates substantial overlap between biological and non-biological particles, which can lead to misclassification when fluorescence information is used alone.

Building on this dataset, we introduce a new machine-learning classification framework that combines fluorescence and morphological features. The algorithm is trained using laboratory data and evaluated with field observations from Zeppelin Observatory, Svalbard. To improve discrimination of combustion-related particles and to better separate dust from SSA, we apply domain adaptation using in situ measurements. The updated classifier successfully reproduces the previously reported annual bioaerosol cycle, yields higher bioaerosol concentrations than a fluorescence-only method, and maintains similar correlations with established biological and combustion tracers. Our open-source code enables more robust quantification of bioaerosols across a range of environments, allows reassessment of prior observations, and can be further improved as new particle characterization data become available.