21.10.2025 BOKU-Met Seminar
Biological aerosols have wide-reaching impacts on Earth’s climate, ecosystems, air quality, and
consequently, human health, yet their detection and classification remain challenging. Previous
studies have successfully derived aerosol size and chemical composition from aerosols' intensive
optical properties and inferred aerosol type based on their absorption and scattering Ångström
Exponents (SAE and AAE).
Characteristic regions within the Ångström matrix (AAE vs. SAE scatter plot) have been found
for Mineral dust, black carbon (BC), and brown carbon (BrC) aerosols; however, a similar well-
defined region for bioaerosols has not been established. This thesis investigates if an optical
footprint for bioaerosols can be identified within the Ångström matrix using measurements
conducted at the high-alpine Sonnblick Observatory, and assesses the feasibility of machine
learning classifiers, trained on the aerosol intensive optical properties AAE, SAE, Single
Scattering Albedo (SSA), and backscatter ratio (BSR), for their automatic classification.
The results revealed a consistent bioaerosol optical signature within the Ångström matrix,
characterized by SAE values between 1.2 and 2.1 and AAE values between 1.17 to 1.60. The
footprint was statistically distinct from mineral dust, biomass burning, and volcanic
aerosol events.
The XGBoost machine learning classifier achieved an F1 score of 0.89. Mineral dust and
biomass burning events were correctly classified in most instances, resulting in close to perfect
recall and high precision scores. Bioaerosol events were correctly identified in the majority of
cases, with a high recall of 93.62% but drastically lower precision of 80.37%, which can be
attributed to many “non-event” datapoints during the summer being classified as bioaerosols, since the bioaerosol signature was persistent throughout the summer. This potential in this type of classification, while also showing the need for further improvement in class labelling.