Assessing spatial and temporal variability of VOCs and PM-components in outdoor air during the Detroit Exposure and Aerosol Research Study (DEARS)

Exposure models for air pollutants often adjust for effects of the physical environment (e.g., season, urban vs. rural populations) in order to improve exposure and risk predictions. Yet attempts are seldom made to attribute variability in observed outdoor air measurements to specific environmental variables. This research presents a statistical strategy to identify and explain the spatial and temporal components of air pollutant measurement variance using regional predictors and large-scale (with impacts over multiple kilometers of distance) emission source effects. The emission sources considered in this investigation include major highways and industries, and were chosen based on their proximity to monitoring areas designated in the Detroit Exposure and Aerosol Research Study (DEARS). Linear mixed effects models were used to investigate 24-h averaged outdoor residential air measurements of several pollutants, including PM 2.5 mass, PM components (elemental carbon, organic carbon, metals, elements), nitrogen dioxide, and volatile organic compounds (VOCs). Three hierarchal statistical models were utilized to calculate and examine variance component estimates for each analyte before and after adjustment for fixed effects, which included sampling season, day of the week, air concentrations at an ambient (centralized) monitoring site, and the frequency of time a receptor was downwind of specific large-emissions sources. Results indicate that temporal variability accounted for the majority of total measurement variance (90% on average). Adjustments for ambient concentration and sampling season significantly reduced temporal variance estimates for most VOCs and for about half of the PM components (generally with reductions of 24–97%). Major exceptions to this trend were found with metals (Fe, Mn, and Zn), ethyltoluene, and p-dichlorobenzene, where only 4–30% of the temporal variance was explained after the same adjustments. Additional reductions in temporal variance (up to 37%) were observed after adjusting for the large-emission sources and day of the week effects, with the strongest effects observed for PM components, including select metals. Thus, for the Detroit airshed, VOCs appear to have been largely affected by regional factors, whereas PM components were explained by both regional factors and localized large-emissions sources. Examination of the radial directions associated with suspected emission sources generally supported a priori expectations of source–analyte associations (e.g., NO 2 increases from areas of high vehicle traffic). Overall, this investigation presents a statistical multi-pollutant analysis strategy that is useful for simultaneously (1) estimating spatial and temporal variance components of outdoor air pollutant measurements, (2) estimating the effects of regional variables on pollutant levels, and (3) identifying likely emissions sources that may affect outdoor air levels of individual or co-occurring pollutants.