Typical classification schemes for large data sets of single-particle mass spectra involve statistical or neural network analysis. In this work, a new approach is evaluated in which particle spectra are pre-selected on the basis of an above threshold signal intensity at a specified m/z (mass to charge ratio). This provides a simple way to identify candidate particles that may contain the specific chemical component associated with that m/z. Once selected, the candidate particle spectra are then classified by the fast adaptive resonance algorithm, ART 2-a, to confirm the presence of the targeted component in the particle and to study the intra-particle associations with other chemical components. This approach is used to characterize metals in a 75,000 particle data set obtained in Baltimore, Maryland. Particles containing a specific metal are identified and then used to determine the size distribution, number concentration, time/wind dependencies and intra-particle correlations with other metals. Four representative elements are considered in this study: vanadium, iron, arsenic and lead. Number concentrations of ambient particles containing these elements can exceed 10,000particlescm-3 at the measurement site. Vanadium, a primary marker for fuel oil combustion, is observed from all wind directions during this time period. Iron and lead are observed from the east-northeast. Most particles from this direction that contain iron also contain lead and most particles that contain lead also contain iron, suggesting a common emission source for the two. Arsenic and lead are observed from the south-southeast. Particles from this direction contain either arsenic or lead but rarely both, suggesting different sources for each element.
- Ambient aerosol
- Chemical composition
- Particle number concentration
- Real-time single-particle mass spectrometry
ASJC Scopus subject areas
- Atmospheric Science
- Environmental Science(all)