Application of mobile sampling to investigate spatial variation in fine particle composition

Long-term exposure to particulate matter (PM) is a major contributor to air pollution related deaths. Evidence indicates that metals play an important role in harming human health due to their redox potential. We conducted a mobile sampling campaign in 2013 summer and winter in Pittsburgh, PA to characterize spatial variation in PM2.5 mass and composition. Thirty-six sites were chosen based on three stratification variables: traffic density, proximity to point sources, and elevation. We collected filters in three time sessions (morning, afternoon, and overnight) in each season. X-ray fluorescence (XRF) was used to analyze concentrations of 26 elements: Na, Mg, Al, Si, S, Cl, K, Ca, Ti, V, Cr, Mn, Fe, Co, Ni, Cu, Zn, As, Se, Br, Rb, Sr, Zr, Cd, Sb, and Pb. Trace elements had a broad range of concentrations from 0 to 300 ng/m3. Comparison of data from mobile sampling filters with stationary monitors suggested that the mobile sampling strategy did not lead to a biased dataset. We developed Land Use Regression (LUR) models to describe spatial variation of PM2.5, Si, S, Cl, K, Ca, Ti, Cr, Fe, Cu, and Zn. Using ArcGIS-10.3 (ESRI, Redlands, CA), we extracted different independent variables related to traffic influence, land-use type, and facility emissions based on the National Emission Inventory (NEI). To validate LUR models, we used regression diagnostics such as leave-one-out cross validation (LOOCV), mean studentized prediction residual (MSPR), and root mean square of studentized residuals (RMS). The number of predictors in final LUR models ranged from 1 to 6. Models had an average R2 of 0.57 (SD = 0.16). Traffic related variables explained the most variability with an average R2 contribution of 0.20 (SD = 0.20). Overall, these results demonstrated significant intra-urban spatial variability of fine particle composition.