AAAR 34th Annual Conference
October 12 - October 16, 2015
Hyatt Regency
Minneapolis, Minnesota, USA
Abstract View
Evaluation of PM2.5 Source Apportionment Methods using Spectral Analysis
SIVARAMAN BALACHANDRAN, Heather Holmes, James Mulholland, Armistead G. Russell, Georgia Institute of Technology
Abstract Number: 520 Working Group: Source Apportionment
Abstract Three receptor sites that measure fine particulate matter (PM2.5) composition are evaluated for variation of source apportionment (SA) results. Variation from spatio-temporal differences and from use of different SA methods are evaluated. Four SA methods are evaluated: a chemical mass balance with gas constraints (CMB-GC) method using three sets of source profiles and positive matrix factorization (PMF). Source profiles used in CMB-GC include measurement-based source profiles (MBSPs), and two sets of profiles from ensemble-averaging multiple models using a standard and a Bayesian technique. SA is conducted at two urban sites in Atlanta, GA: the Jefferson St. (JST) SEARCH site and the South Dekalb (SDK) CSN site, and, for a rural SEARCH site in Yorkville, GA (YRK). Source impacts from the four SA methods at three sites are compared for temporal trends using spectral analysis using the Lomb-Scargle Periodogram Method (LSPM). Results across sites and methods are used to evaluate spatial and method-specific variation, respectively. All SA methods, species and source profiles/factors show a strong power spectra peak at one year. Gasoline vehicle impacts using CMB-GC at JST and both CMB-GC and PMF at SDK have statistically significant peaks (α =0.05) for the frequency associated with one week. CMB-GC and PMF at JST and SDK have peaks (α =0.05) for the frequency associated with one week for diesel vehicle impacts. The greatest variability across methods and locations was seen with biomass burning profiles/factors, especially with PMF factors and source profiles derived using the Bayesian technique. Across the three sites, the variability in OC to EC ratios in biomass burning profiles corroborate power spectra analysis that emissions from biomass burning are more spatially variable than other sources.