Advanced Receptor Models for Exploiting Highly Time Resolved Data Acquired in the EPA Supersite Project

Abstract

Receptor models have been widely used in air quality studies to identify pollution sources and estimate their contributions. A common problem for most current receptor models is insufficient consideration of realistic constraints such as can be obtained from emission inventories, chemical composition profiles of the sources, and the physics of plume dispersion. In addition, poor resolving of collinear sources was often found. With the high quality time-, composition-, and size-resolved measurements during the EPA Supersite project, efforts towards resolving nearby industrial sources were made by combinative use of Positive Matrix Factorization (PMF) and the Pseudo-Deterministic Receptor Model (PDRM). The PMF modeling of Baltimore data in September 2001 revealed coal-fired and oil-fired power plants (CFPP and OFPP, respectively) with significant cross contamination, as indicated by the high Se/Ni ratio in the OFPP profile. Nevertheless, the PMF results provided a good estimate of background and the PMF-constrained emission rates well seeded the trajectory-driven PDRM modeling. Using NOx as the tracer gas for χ/Q tuning, ultimately resolved emissions from individual stacks exhibited acceptable tracer ratios and the emission rates of metals generally agreed with the TRI estimates. This approach was later applied to two metal pollution episodes in St. Louis during in November 2001 and March 2002 and met a similar success. As NOx measurements were unavailable at those metal-production facilities, highly-specific tracer metals (i.e., Cd, Zn, and Cu) for the corresponding units were used to tune χ/Qs and their contributions were well resolved with the PMF-seeded PDRM. Opportunistically a PM2.5 excursion during a windless morning in November 2002 allowed the extraction of an in-situ profile of vehicular emissions in Baltimore. The profiles obtained by direct peak observation, windless model linear regression (WMA), PMF, and UNMIX were comparable and the WMA profile showed the best predictions for non-traffic tracers. Besides, an approach to evaluate vehicular emission factors was developed by receptor measurements under windless conditions. Using SVOC tracers, seasonal variations of traffic and other sources including coal burning, heating, biomass burning, and vegetation were investigated by PMF and in particular the November traffic profile was consistent with the WMA profile obtained earlier.