Use of synthetic data to evaluate positive matrix factorization as a source apportionment tool for PM2.5 exposure data

Positive matrix factorization (PMF) was applied to synthetic datasets that simulate personal exposures to airborne PM2.5 from 12 sources. Three different filter analysis scenarios using different analytical chemistry techniques were considered. The full suite scenario quantified elemental carbon, organic carbon, inorganic ions, trace elements, and trace organic species including carboxylic acids and organic compounds with -OH functionality. A second scenario excluded trace elements and a third assumed that derivatization steps to quantify polar organic compounds were not performed. Similar errors in source apportionment were seen with all three scenarios. In most cases, PMF failed to separate out factors corresponding to road dust and vegetative debris, two sources that made relatively uniform contributions to the synthetic exposures. Factors representing wood smoke, natural gas combustion, and meat cooking sources were difficult to identify due to a lack of unique tracers with concentrations reliably above the detection limits assumed in the study. Factors representing cigarette smoke, candle smoke, gasoline exhaust, and secondary aerosols were comparatively easy to identify. When contributions from a pair of sources, such as diesel and gasoline exhaust, were highly correlated in the synthetic datasets, a single factor corresponding to both sources was usually found.