Source apportionment of Baltimore aerosol from combined size distribution and chemical composition data

Several multivariate data analysis methods have been applied to a combination of particle size and composition measurements made at the Baltimore Supersite. Partial least squares (PLS) was used to investigate the relationship (linearity) between number concentrations and the measured PM2.5 mass concentrations of chemical species. The data were obtained at the Ponca Street site and consisted of six days’ measurements: 6, 7, 8, 18, 19 July, and 21 August 2002. The PLS analysis showed that the covariance between the data could be explained by 10 latent variables (LVs), but only the first four of these were sufficient to establish the linear relationship between the two data sets. More LVs could not make the model better. The four LVs were found to better explain the covariance between the large sized particles and the chemical species. A bilinear receptor model, PMF2, was then used to simultaneously analyze the size distribution and chemical composition data sets. The resolved sources were identified using information from number and mass contributions from each source (source profiles) as well as meteorological data. Twelve sources were identified: oil-fired power plant emissions, secondary nitrate I, local gasoline traffic, coal-fired power plant, secondary nitrate II, secondary sulfate, diesel emissions/bus maintenance, Quebec wildfire episode, nucleation, incinerator, airborne soil/road-way dust, and steel plant emissions. Local sources were mostly characterized by bi-modal number distributions. Regional sources were characterized by transport mode particles (0.2– ).