The accurate identification of sources of air pollution remains challenging due to the chemical complexity of aerosol particles and the number of sources. Single-particle mass spectrometry (SPMS) enables real-time chemical characterization of individual aerosol particles, preserving information that is lost in bulk analyses. However, non-linear physical phenomena in the ionization and consequently high dimensionality of SPMS spectra are a bottleneck for rapid and versatile analysis. In this study, we apply the dimensionality reduction algorithm Uniform Manifold Approximation and Projection (UMAP) to SPMS measurements of airborne particles to visualize, cluster, and differentiate residues, advancing data-driven detection strategies and enabling quick labeling for machine learning approaches. In this study, we focus on analyzing the emissions of ships entering and leaving a port area, with the purpose to identify ships running on high-sulfur fuels like heavy fuel oil (HFO) and low-sulfur marine gas oil (MGO). Measurement data from a field study at the main shipping lane to the port of Rostock, Germany, with up to 30,000 daily measured particles were transformed by UMAP into two-dimensional embeddings preserving the local chemical similarity. Chemical signatures obtained from a previous laboratory study at a research engine were exploited to extract and annotate clusters, and to analyze inter and intra-cluster differences. Time instants of characteristic mass spectra were linked to with wind data from a local weather station and recorded ship positions to identify ships running on specific fuels. This general approach to source apportionment based on data-driven evaluation of single-particle mass spectra is ready to be extended to a large variety of target analytes.     «

The accurate identification of sources of air pollution remains challenging due to the chemical complexity of aerosol particles and the number of sources. Single-particle mass spectrometry (SPMS) enables real-time chemical characterization of individual aerosol particles, preserving information that is lost in bulk analyses. However, non-linear physical phenomena in the ionization and consequently high dimensionality of SPMS spectra are a bottleneck for rapid and versatile analysis. In this stud...     »