Source apportionment of PM 2.5 in the southeastern United States using receptor and emissions-based models: Conceptual differences and implications for time-series health studies

Abstract

Elevated levels of fine particulate matter (PM 2.5) have been associated with adverse effects on human health, but whether specific components of PM 2.5 are responsible for specific health effects is still under investigation. A complementary approach to examining species-specific associations is to assess associations between health outcomes and sources contributing to PM 2.5. This approach could help target and regulate the sources that contribute most to adverse health effects. Various techniques have been developed to quantify source impacts on air quality, allowing examination of their health impacts. We compare two conceptually different approaches to source apportionment (SA): a receptor model and an emissions-based air-quality model. Daily source impacts for July 2001 and January 2002 at four sites in the southeastern US were calculated using CMB-LGO, an extended chemical mass balance receptor model incorporating the Lipschitz global optimizer, and EPA's Models-3 emissions-based air-quality modeling system (MM5–SMOKE–community multiscale air-quality (CMAQ)). The receptor model captured more of the temporal variation in source impacts at a specific receptor site compared to the emissions-based model. Driven by data at a single site, receptor models may have some significant shortcomings with respect to spatial representativeness (unless a reduced study area is used or data from multiple sites are available). SA results from emissions-based models, such as CMAQ, may be more spatially representative as they represent an average grid-cell value. Limitations in the ability to model daily fluctuations in emissions, however, lead to results being driven mainly by regional meteorological trends, likely underestimating the true daily variations in local source impacts. Using results from either approach in a health study would likely introduce an attenuation of the observed association, due to limited spatial representativeness in receptor modeling results and to limited temporal representativeness in emissions-based models results.