Optimization-Based Source Apportionment of PM_2.5 Incorporating Gas-to-Particle Ratios

Abstract

A modified approach to PM2.5 source apportionment is developed, using source indicative SO2/PM2.5, CO/PM2.5, and NOx/PM2.5 ratios as constraints, in addition to the commonly used particulate-phase source profiles. Additional information from using gas-to-particle ratios assists in reducing collinearity between source profiles, a problem that often limits the source-identification capabilities and accuracy of traditional receptor models. This is especially true in the absence of speciated organic carbon measurements. In the approach presented here, the solution is based on a global optimization mechanism, minimizing the weighted error between apportioned and ambient levels of PM2.5 components, while introducing constraints on calculated source contributions that ensure that the ambient gas-phase pollutants (SO2, CO, and NOy) are reasonable. This technique was applied to a 25-month dataset of daily PM2.5 measurements (total mass and composition) at the Atlanta Jefferson Street SEARCH site. Results indicate that this technique was able to split the contributions of mobile sources (gasoline and diesel vehicles) more accurately than particulate-phase source apportionment methods. Furthermore, this technique was able to better quantify the direct contribution (primary PM2.5) of coal-fired power plants to ambient PM2.5 levels.

Multipollutant modeling issues in a study of ambient air quality and emergency department visits in Atlanta

“…While source apportionment introduces an additional layer of uncertainty into epidemiologic analyses and results in a loss of specificity, it is complementary to more traditional modeling approaches. In the Atlanta study, PM 2.5 source apportionment work corroborates the impression provided by the single-pollutant models that CVD visits are related to vehicular emissions (both diesel and gasoline) (Kim et al, 2003(Kim et al, , 2004Marmur et al, 2005Marmur et al, , 2006Sarnat et al, 2006). Reports on these and a number of other analyses of the Atlanta data, including spatial subanalyses and assessment of measurement error impacts, are forthcoming.…”

Effects of Instrument Precision and Spatial Variability on the Assessment of the Temporal Variation of Ambient Air Pollution in Atlanta, Georgia

“…Source apportionment modeling suggests that ϳ50% of OC mass at JS is secondary organic aerosol, although there is high uncertainty associated with this estimate. 36 The EC and OC semivariogram plots also suggest that error in the ambient estimate results from differences between the TOR method (SEARCH data) and the TOT method (ASACA data). EC and OC semivariograms from the SEARCH site pair (JS and Yorkville, hollow triangles at 60 km in the EC and OC plots in Figure 3) are lower than from other pairs.…”

“…Source apportionment modeling indicates that 60 -70% of PM 2.5 mass at JS is secondary. 35,36 Because concentrations of these pollutants are driven more by regional meteorology and chemistry than by specific sources, spatial variation for the secondary pollutants accounts for a smaller fraction of temporal variation than for the primary pollutants. Although values of O 3 were lower at the South Dekalb monitoring site because of titration from local NO sources, as described later, the O 3 correlation with other monitoring sites is strong.…”

“…O 3 is a secondary pollutant, and, as stated previously, 60 -70% of PM 2.5 mass is estimated to be secondary. 35,36 Secondary pollutants are less affected by emission sources than primary pollutants. The authors do note, however, that O 3 minima are observed in directions where NO x peaks occur.…”

Seasonal Study of Primary and Secondary Sources of Carbonaceous Species in PM10 from Five Northern Chinese Cities

“…Besides the above mentioned methods, the CMB model, a useful receptor model (Srivastava et al, 2009), has also been applied to estimate the contributions of primary source categories and the contribution of SOC to TOC (Marmur et al, 2005;. The CMB model needs datasets from the source profile (Lee et al, 2008), but there is no actual profile for SOC.…”