Cluster analysis of single particle mass spectra measured at Flushing, NY

Abstract ART-2a and a density based cluster method, density based spatial clustering of application with noise (DBSCAN), have been used for classification of the single particle mass spectra measured at New York City. Using too large of a vigilance factor in ART-2a leads to many similar clusters with overlap, and thus a low vigilance factor was used in this study. The DBSCAN method can identify clusters with complex shapes and various sizes, and representative spectra are chosen to identify different particle types within each cluster. The cluster structure of the single particle mass spectra were examined by DBSCAN. Both methods found that the major clusters were sea salt and anthropogenic combustion emissions. The continua in sulfate, potassium and OC particles were found by DBSCAN and a large cluster was formed, while ART-2a broke it into several small clusters without finding this continuum. A detailed discussion of the cluster analysis results including representative mass spectra, size distributions and temporal behavior will be provided.