The Denver Aerosol Sources and Health (DASH) study: Overview and early findings S. Vedal a,*, M.P. Hannigan b, S.J. Dutton c, S.L. Miller b, J.B. Milford b, N. Rabinovitch d, S.-Y. Kim a, L. Sheppard a a Department of Environmental and Occupational Health Sciences, University of Washington School of Public Health and Community Medicine, Seattle, WA 98105, USA b Department of Mechanical Engineering, College of Engineering and Applied Science, University of Colorado, Boulder, CO, USA c Department of Civil, Environmental and Architectural Engineering, College of Engineering and Applied Science, University of Colorado, Boulder, CO, USA d National Jewish Medical and Research Center, Denver, CO, USA a r t i c l e i n f o Article history: Received 2 July 2008 Received in revised form 8 November 2008 Accepted 6 December 2008 Keywords: Air pollution Particulate matter Health effects Mortality Source apportionment a b s t r a c t Improved understanding of the sources of air pollution that are most harmful could aid in developing more effective measures for protecting human health. The Denver Aerosol Sources and Health (DASH) study was designed to identify the sources of ambient fine particulate matter (PM2.5) that are most responsible for the adverse health effects of short-term exposure to PM2.5. Daily 24-h PM2.5 sampling began in July 2002 at a residential monitoring site in Denver, Colorado, using both Teflon and quartz filter samplers. Sampling is planned to continue through 2008. Chemical speciation is being carried out for mass, inorganic ionic compounds (sulfate, nitrate and ammonium), and carbonaceous components, including elemental carbon, organic carbon, temperature-resolved organic carbon fractions and a large array of organic compounds. In addition, water-soluble metals were measured daily for 12 months in 2003. A receptor-based source apportionment approach utilizing positive matrix factorization (PMF) will be used to identify PM2.5 source contributions for each 24-h period. Based on a preliminary assessment using synthetic data, the proposed source apportionment should be able to identify many important sources on a daily basis, including secondary ammonium nitrate and ammonium sulfate, diesel vehicle exhaust, road dust, wood combustion and vegetative debris. Meat cooking, gasoline vehicle exhaust and natural gas combustion were more challenging for PMF to accurately identify due to high detection limits for certain organic molecular marker compounds. Measurements of these compounds are being improved and supplemented with additional organic molecular marker compounds. The health study will investigate associations between daily source contributions and an array of health endpoints, including daily mortality and hospitalizations and measures of asthma control in asthmatic children. Findings from the DASH study, in addition to being of interest to policymakers, by identifying harmful PM2.5 sources may provide insights into mechanisms of PM effect. �� 2008 Elsevier Ltd. All rights reserved. 1. Introduction Findings from observational epidemiology studies on the adverse health effects of ambient particulate matter (PM) and other pollutants indicate effects of short-term concentration changes on mortality and morbidity (Dominici et al., 2003 Pope and Dockery, 2006). As a surrogate for exposure, these studies have primarily relied on outdoor PM mass concentrations measured at one or more monitoring sites distributed over large urban areas. An important finding in recent observational studies is that there appears to be heterogeneity between urban areas in the magnitude of the observed acute effect of particulate air pollution (Samet et al., 2000 Dominici et al., 2003). The causes of this heterogeneity are not known. The PM contribution from various sources, and hence the chemical composition of PM, varies by location (US-EPA, 2004). If PM toxicity is dependent on its chemical composition, variability of PM source contributions could be one explanation for the heterogeneity of effects. Evalu- ating associations between source contributions and health effects may thus provide insight into the heterogeneity between urban areas. The National Research Council has stressed the need to understand the comparative toxicity of PM from different sources * Corresponding author. Tel.: ��1 206 616 8285 fax: ��1 206 685 4696. E-mail address: svedal@u.washington.edu (S. Vedal). Contents lists available at ScienceDirect Atmospheric Environment journal homepage: www.elsevier.com/locate/atmosenv 1352-2310/$ ��� see front matter �� 2008 Elsevier Ltd. All rights reserved. doi:10.1016/j.atmosenv.2008.12.017 Atmospheric Environment 43 (2009) 1666���1673

and recommended the use of source apportionment methods (NRC, 2004). To date, a limited number of air pollution health studies have used source apportionment to investigate mortality effects (Laden et al., 2000 Mar et al., 2000, 2006 Tsai et al., 2000 Ito et al., 2006), as well as effects on hospitalizations (Anderson et al., 2007) and emergency room visits (Schreuder et al., 2006 Sarnat et al., 2008). The Denver Aerosol Sources and Health (DASH) study is designed to investigate effects of exposure to sources of fine particulate matter (PM2.5) on an array of health endpoints by incorporating detailed daily PM2.5 chemical specia- tion data and state-of-the-art source apportionment into a health study. Several options are available to identify the role of time-varying pollutant sources in the context of a health study. Concentrations of specific source markers have been used for this purpose (Laden et al., 2000), but valid markers for many sources of interest are not available. Deterministic air quality models can estimate contribu- tions of sources at specific locations at 4 km or finer grid resolution (Held et al., 2005), but their ability to reproduce short-term varia- tions is limited, due in part to the lack of temporally resolved emissions��� information (Marmur et al., 2006). Receptor-based source apportionment approaches use PM component concentra- tions measured at a monitoring site to estimate site-specific source contributions to PM concentrations. Receptor models have been used to help identify PM sources in numerous geographical regions (Maykut et al., 2003 Hopke et al., 2003 Schauer et al., 1996 Long et al., 2005 Lewis et al., 2003). These models either use a chemical mass balance (CMB) approach, which requires detailed information about the chemical composition of sources in a region, or use multivariate factor analysis methods, for example, positive matrix factorization (PMF) or UNMIX, to identify sources based on the temporal patterns among chemical components measured at the receptor(s). These multivariate approaches have the advantage of not requiring prior information on static source profiles, although source profile information is still needed to identify factors extracted from multivariate methods and unambiguous factor identification is not always possible. An important objective of the DASH study is to evaluate the utility of PMF for identifying short- term pollution source contributions to the ambient PM2.5 in a large urban area. This paper provides an overview and introduction to the DASH study, focusing on the PM2.5 characterization methods used and the source apportionment techniques being employed. Interim 4.5- year PM2.5 species concentrations are also presented. In addition, model validation efforts using synthetic data are presented and discussed. Model uncertainty will be explored using a novel method based on a bootstrap approach and neural networks (Hemann et al., 2008). Companion papers describe in more detail the PM2.5 monitoring and chemical analysis protocols, as well as the methodology for point-wise uncertainty estimation (Dutton et al., 2009a). Subsequent papers will detail results of PM2.5 species trends, source apportionment and epidemiological analyses. 2. Methods 2.1. PM2.5 collection and characterization 2.1.1. Core receptor site monitoring A residential site on the roof of Palmer Elementary School located 5.3 km east of downtown Denver was selected as the receptor site for detailed PM2.5 chemical speciation (Fig.1). This site was selected for its central location in a large residential commu- nity and for its distance from point and mobile sources. The school is 5.2 km from the closest large highway and at least 0.6 km from the closest commuter street. Collection of daily 24-h PM2.5 samples from this site has been ongoing since July 1, 2002. PM2.5 samples were collected daily from midnight to midnight on both Teflon and quartz fiber filters with sampling equipment that has been used extensively in air quality studies in the past (Lough et al., 2006 Christoforou et al., 2000). The sampling protocol and the bulk chemical analysis methods briefly described below are presented in more detail in Dutton et al. (2009a). A list of all species being analyzed is contained in Table S1 in the online supplementary data. 2.1.2. Bulk chemical analyses Bulk chemical speciation of all samples includes the quantifi- cation of PM2.5 mass, inorganic ionic compounds (sulfate, nitrate, and ammonium), elemental carbon (EC) and organic carbon (OC). Gravitational mass measurements are made using a microbalance housed in a temperature and humidity controlled chamber with filter weighing methods patterned after those approved by the EPA for PM2.5 mass determination (US-EPA, 1998). Inorganic ionic compounds (sulfate, nitrate and ammonium) are quantified using an ion chromatograph following traditional methodology for airborne particulate matter (Mueller et al., 1978). Bulk EC and OC concentrations are determined by NIOSH Method 5040 a 1.5 cm2 punch is taken from each quartz fiber filter and sent to the Wis- consin State Laboratory of Hygiene for analysis. 2.1.3. Organic compound chemical analysis Methods for the detailed organic chemical analysis of quartz fiber filter samples using solvent extraction and gas chromatog- raphy/mass spectrometry (GC/MS) have been developed previously (Mazurek et al., 1987 Fraser et al., 1996 Schauer et al., 1998 Hannigan et al., 1998) and similar methods were used in this study. High volume injection (50 ml) was incorporated by way of a programmable temperature vaporization inlet, thereby improving detection limits for trace level organic compounds (Crimmins and Baker, 2006). Detailed description of the organic speciation methods is included in Dutton et al. (2009b). Fig. 1. Shown is a map of the Denver metropolitan area, including the five counties included in the DASH study and the locations of the residential monitoring site (star) and the STN site in Commerce City (circle). S. Vedal et al. / Atmospheric Environment 43 (2009) 1666���1673 1667