HR: 1340h
AN: A33B-1177    [Abstracts]
TI: Characterization of Positive Matrix Factorization of Aerosol Mass Spectrometer Data With Real and Synthetic Data
AU: Canagaratna, M
EM: mrcana@aerodyne.com
AF: Aerodyne Research, Inc., 45 Manning Rd., Billerica, MA 01821, United States
AU: * Ulbrich, I M
EM: ulbrich@colorado.edu
AF: Dept. of Chemistry and Biochemistry and CIRES, University of Colorado, Boulder, CO 80309, United States
AU: Zhang, Q
EM: qz@asrc.cestm.albany.edu
AF: 3Atmospheric Sciences Research Center, State University of New York, Albany, NY 12203, United States
AU: Worsnop, D R
EM: worsnop@aerodyne.com
AF: Aerodyne Research, Inc., 45 Manning Rd., Billerica, MA 01821, United States
AU: Jimenez, J L
EM: jose.jimenez@colorado.edu
AF: Aerodyne Research, Inc., 45 Manning Rd., Billerica, MA 01821, United States
AB: Mass spectrometric measurements of ambient aerosols yield organic spectra that are a mix of nucleated particles, freshly emitted particles from many sources, and particles which have undergone some amount of processing (condensation, oxidative reaction, cloud processing, etc.). Further understanding of the important sources and processes for organic aerosols requires deconvolution of the organic fraction of ambient aerosols. A well-known source apportionment technique, Positive Matrix Factorization (PMF), has been applied to Aerodyne aerosol mass spectrometer (Q-AMS) datasets acquired in Pittsburgh (2002). Extensive sensitivity analyses of the Pittsburgh case are performed with synthetic datasets to characterize the behaviour of PMF with AMS datasets PMF is a least-squares fitting method for source apportionment commonly applied to speciated aerosol datasets (Paatero, Chemomet. Intell. Lab. Sys. 1997, 37, 23-35). The organic portion of the Pittsburgh dataset (Zhang et al. ( ES&T 2005, 39, 4938-4952) was analyzed with PMF. The factors from the two- and three-factor solutions are used directly and modified to create synthetic datasets to explore the ability of PMF to retrieve factors with strongly (weakly) correlated mass spectra or (and) time series, to retrieve factors with a small fraction of the mass, and the effect of rotations of the solution on these retrievals. When PMF solves for more factors than were used in the input, major components are "split" into multiple factors and/or "mixed" to create new factors that have high correlation with actual AMS database spectra from ambient studies, chamber studies, and pure compounds (http://cires.colorado.edu/jimenez-group/AMSsd/spectra.html).
DE: 0305 Aerosols and particles (0345, 4801, 4906)
DE: 0325 Evolution of the atmosphere (1610, 8125)
DE: 0345 Pollution: urban and regional (0305, 0478, 4251)
DE: 0365 Troposphere: composition and chemistry
DE: 0394 Instruments and techniques
SC: Atmospheric Sciences [A]
MN: 2007 Fall Meeting