Apportioning Air Toxics Risks using Positive Matrix Factorization

Nov 12, 2009 - Next, Positive Matrix Factorization. (PMF) was performed on high time-resolved measurements of these air toxics, and the additive cance...
1 downloads 0 Views 2MB Size
Download PDF
Environ. Sci. Technol. 2009 43, 9439–9444

Identifying Priority Pollutant Sources: Apportioning Air Toxics Risks using Positive Matrix Factorization JENNIFER M. LOGUE, MITCHELL J. SMALL, AND ALLEN L. ROBINSON* Department of Engineering and Public Policy, Carnegie Mellon University, Pittsburgh Pennsylvania 15213

Received June 8, 2009. Revised manuscript received October 13, 2009. Accepted October 20, 2009.

focus on sources that contribute the most to risk. Emissionsbased modeling studies have been used to associate air toxics concentrations to health risks (6, 7). The goal of this work is to demonstrate a method for apportioning air toxics cancer risks to sources in and around Pittsburgh, Pennsylvania. First, a linear, no-threshold risk model was used to identify the risk drivers from a data set of high time-resolved measurements of air toxics made by Logue et al. (4). The measured concentrations were apportioned using Positive Matrix Factorization (PMF) with special emphasis on identifying the sources of high-risk pollutants. Factors were linked to sources and/or source classes by comparisons with source profiles, factor contribution with wind direction, and temporal patterns. Finally, the contribution of each factor to cancer risks was determined.

Methods Hazardous air pollutants or air toxics are pollutants that are known or suspected to cause serious health effects. This paper presents a methodology to quantify source contributions to air toxics health risks. First, a linear, no-threshold risk model was used to identify gas-phase organic air toxics that contribute significantly to cancer risks. Next, Positive Matrix Factorization (PMF) was performed on high time-resolved measurements of these air toxics, and the additive cancer risks associated with each factor was determined. Finally, the PMF factors were linked to sources and source classes (mobile, nonmobile, secondary/background) using a combination of meteorological data and comparisons with published source profiles. The analysis was performed using data from three sites in Pittsburgh, Pennsylvania: a downtown site near a heavily traveled bus route, a residential site adjacent to a heavily industrialized area, and an urban background site. At all three sites emissions from nonmobile sources were the dominant contributors to the cancer risks from air toxics included in the PMF model, including benzene and other air toxics often associated with mobile source emissions. Emissions from both large industrial sources, such as coke works and chemical facilities, and smaller point sources, such as dry cleaners, contributed significantly to the cancer risks at all sites. This method can provide insight for decision makers to prioritize sources for risk reduction.

Introduction A primary goal of air pollution regulation is to reduce human health risks. The US Environmental Protection Agency currently classifies 187 pollutants as hazardous air pollutants or air toxics. The Government Performance and Results Act set a goal of reducing air toxics emissions by 75% from the 1993 levels (1). Achieving that goal requires an understanding of air toxics exposure, risks, and sources. Many previous studies have used receptor models to investigate sources of volatile organic compounds (VOCs) (2, 3). Most of these studies have focused on precursors for ozone formation, with relatively few considering sources of air toxics (4, 5). Results from these studies are always reported in terms of source contributions to ambient concentrations, whereas a more policy-relevant strategy for air toxics would * Corresponding author e-mail: [email protected]; phone: 412 268 3657; fax: 412 268 3348. 10.1021/es901683j CCC: $40.75

Published on Web 11/12/2009

 2009 American Chemical Society

The two-way factor analytic model PMF2 (8) was used to analyze hourly concentrations of gas-phase organic air toxics measured at 3 sites in and around Pittsburgh, PA (4). The Avalon site was located in a residential neighborhood about 0.8 km northeast of a heavily industrialized area, Neville Island. Neville Island is home to chemical and manufacturing facilities including a large metallurgical coke production plant. The downtown site was located on the corner of Fifth and Liberty Avenues, which are major bus routes. The Carnegie Mellon University (CMU) site is representative of urban background conditions (9). Additional details on these sites can be found in the online Supporting Information (SI). At each site an automated GC/MS/FID (gas chromatograph/mass spectrometer/flame ionization detector) instrument was deployed to measure hourly concentrations of gasphase organic air toxics and other VOCs. Details on the design and operation of the instrument and a description of the high time-resolved data sets can be found in Logue et al. (4) and in the SI. The PMF model was used to apportion compounds that exhibited significant spatial and temporal variability in the Pittsburgh region and therefore thought to be strongly influenced by emissions from local sources (4). Table S.1 in the SI (and later, Figure 2) lists the compounds included in the PMF model, which include most of the gas-phase organic air toxics that drive cancer risks in the Pittsburgh area (10), such as benzene, tetrachloroethene, and chloroform. However, we do not consider a few major risks drivers that appear to be predominately emitted by regional sources, such as carbon tetrachloride and formaldehyde (10). These issues are addressed in more detail in the Discussion. PMF analysis was performed independently on the data sets for each site. The specific set of pollutants included in the model varied slightly from site to site because of modest variations in instrument performance. For example, 1,3butadiene was only resolved at the downtown site because of contamination issues during the other two deployments. It was included in the downtown PMF model because it is an important mobile source air toxic (MSAT) and downtown air toxics concentrations are thought to be strongly influenced by mobile source emissions. PMF requires an estimate of the uncertainty for each data point. Measurement uncertainty and the minimum detection level were determined for each compound based on the instrument calibration. When the data were below the minimum detection limit (MDL), the value and uncertainty in the value was set to half of the MDL. The percentage of data below the MDL is available in the SI. VOL. 43, NO. 24, 2009 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

9439

FIGURE 1. Additive cancer risks calculated using the study-average air toxics concentrations calculated from high time-resolved data. The cancer risks of chlorinated compounds are indicated by solid colors. 1,3-Butadiene was only measured at the downtown site. Two important inputs for PMF are the number of factors and the degree of matrix rotation. These parameters were systematically varied to seek an optimum or most interpretable solution. Interpretability was judged based on how clearly a solution identified unique sources. This was done using the wind directional dependence and temporal patterns of factor contributions and by comparing factor profiles with published source profiles and other emissions data taken from the literature. For example, information on source locations was used to associate factors with strong wind directional dependence with specific point sources. Similarly, comparisons of the chemical composition of PMF factor profiles to actual source profiles measured in source tests were used to help link factors to actual sources. Given these constraints, each monitoring site had a clear optimum PMF solution. Several measures of statistical quality of the PMF solution were also considered (11), but multiple solutions satisfied these statistical tests. A major objective of this work was to investigate the dominant sources of cancer risks for the set of air toxics included in the PMF model. Cancer risk estimates were calculated using a linear, no-threshold model (12) and unit risk estimates (UREs) obtained from the U.S. EPA IRIS database (13) and the California EPA (14). Cancer risks were estimated by multiplying the URE for each pollutant by the fraction of the study-average outdoor concentration of that pollutant attributed by PMF to each factor. The total cancer risk for each factor was then calculated by adding the cancer risks of each pollutant apportioned to that factor. Risk additivity assumes no carcinogenic interaction among compounds (i.e., no synergy or antagonism in the toxicological response) and is mathematically valid so long as the individual compound risks are all small (e.g., below 10-3) (15), as is the case in this application. In the remainder of the paper, the phrases cumulative or additive cancer risks refers specifically to additive risks for the air toxics included in the PMF model for a given site. These pollutants are listed in Table S.1. Unless otherwise specified, the term does not refer to the total cumulative cancer risk for all air toxics or all air pollutants.

Results Figure 1 plots the concentrations of cancer-causing compounds considered by this study and the estimated additive cancer risk for each site. Although the high time-resolved measurements were only made for a 1-2 month period, the risk profiles shown in Figure 1 are similar to those derived from a 2-year data set of 24 h measurements (see Figure S.1). 9440

9

ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 43, NO. 24, 2009

In addition, concentrations of most of the air toxics included in the PMF model do not vary seasonally in Pittsburgh area (see SI). Benzene is an important risk driver at all sites; in fact, benzene concentrations throughout the Pittsburgh area are above the 75th percentile of a large compilation of national data (10). At the Avalon site, benzene was the dominant risk driver, contributing 58% of the additive cancer risk for the air toxics considered in this study. At the other two sites chlorinated compounds were the dominant risk drivers. At the downtown site tetrachloroethene and 1,4-dichlorobenzene contributed 25% and 14% of the additive cancer risk, respectively. At the CMU site chloroform is the largest contributor to the additive cancer risk (29%) followed by benzene (23%). The goal of the PMF analysis was to apportion the cancer risks of the compounds included in the model to sources and source classes. PMF apportioned more than 80% of the additive cancer risks at each site. However, PMF factors do not necessarily correspond to sources. There are many more sources than factors, which means that PMF often groups emissions from multiple sources into a single factor. Examples of this issue are noted below. This can lead to both positive and negative biases in the risks attributed to a specific source class. Results for the optimum PMF solutions for each site are shown in Figure 2. The following three sections describe these solutions, focusing on apportioning the estimated cancer risks. The high risk factors are described in detail. The factors that contribute relatively little risk are described in the SI. Avalon PMF solution. Figure 2a plots the species distribution for the optimum PMF model for the Avalon data set. This is a five-factor solution with a -10 rotation that apportioned 95% of the additive cancer risk. The five factors were: Coke Works Factor (40% of the additive cancer risks), Mobile Source Factor (29% of risks), ALCOSAN Factor (14% of risks), Chemical Companies Factor (8% of risks), and a Secondary/Background Factor (4% of risks). We now discuss the high risk factors from the Avalon PMF model. Coke Works Factor. Figure 2a indicates that one factor contributed 66% of the benzene and about half of the carbon disulfide at the Avalon site. Two pieces of evidence link this factor with emissions from metallurgical coke production. First, Figure 3a indicates that the contribution of this factor was greatly elevated when wind was coming from the Shenango Coke Works on Neville Island, a large source of benzene and carbon disulfide (16). The Shenango Coke Works is located approximately 1 km from the Avalon site at a

FIGURE 2. Factor contributions to pollutant concentrations for the optimum PMF solutions for each site (upper panels). The bottom panel shows the total cancer risk apportioned to each factor. heading of 235°. Second, Figure 3a shows that there is excellent agreement between the Coke Works factor profile and the Shenango source profile in the Allegheny County Point Source Emission Inventory (ACPSEI) (16). There was no weekend/weekday pattern in the contribution of this factor, consistent with an industrial source that operates all of the time. Mobile Sources Factor. Figure 2a indicates that PMF apportioned 53%, 60%, and 25% of toluene, xylenes, and benzene in Avalon, respectively, to a single factor. All of these compounds are emitted by mobile sources (17-19). The distribution of species in this factor is similar to that of published mobile source profiles (Figure 3b). The contribution of this factor was elevated when the wind was coming from the east, the direction of downtown Pittsburgh, where there is substantial traffic. However, some nonmobile source air toxics, such as some chlorinated compounds, were also apportioned to this factor. Therefore, the Mobile Source Factor also contains some influence of emissions from nonmobile sources located in the downtown area. Downtown PMF Solution. Figure 2b plots the species distribution for the optimum PMF model for the downtown data set. This is a five-factor PMF model with a -8 rotation that apportioned 89% of the estimated cancer risk at the downtown site. The five factors were Local Industry Factor (26% of risks), Mobile Sources Factor (25% of risks), Clairton Area Factor (24% of risks), Secondary/Background Factor

(11% of risks), and the Alkyl Benzene Factor (3% of risks). We now discuss the high risk factors from the Downtown PMF model. Local Industry Factor. Figure 2b indicates this factor was dominated by tetrachloroethene. Figure 4a indicates that the contribution of this factor was elevated on weekdays and was wind-directional dependent, with higher contributions with wind coming from the south. Tetrachloroethene is emitted by dry-cleaning facilities (20) and a dry-cleaner was located approximately 200 m to the south of the site. Concentrations of tetrachloroethene were 10 times higher at the downtown site compared to the CMU site and two times higher than at Avalon. Mobile Sources Factor. Figure 2b indicates that one PMF factor was the dominant contributor to 15 air toxics at the downtown site, including many MSATs, such as MTBE, benzene, and 1,3 butadiene. Figure 4b demonstrates that this factor profile is similar to those of published diesel and gasoline vehicle source profiles (18, 19). The downtown site is located on a heavily traveled city bus route and within 1 km of the site there are major highways in three different directions. The factor contribution was about 50% higher on weekdays than on the weekend, consistent with the mobile source activity patterns in the downtown area. PMF also apportion some emissions from nonmobile sources to this factor. For example, this factor contributes substantial amounts to several chlorinated air toxics, such as 1,4-dichlorobenzene, that are not associated with mobile VOL. 43, NO. 24, 2009 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

9441

FIGURE 3. Select (high cancer risk) PMF factor derived from the Avalon data set. Panel (a) compares the Coke Works Factor to source profiles for industrial sources in the Clairton Area taken from the ACPSEI (16); and panel (b) compares the Mobile Sources Factor to motor vehicle source profiles from Schauer et al. (18, 19). Also shown is the average wind directional dependence of each factor. The wind rose in the bottom panel shows the location of major source areas relative to the Avalon site. “CA” refers to the Clairton source area. For the Avalon monitoring site, Pittsburgh and Clairton Source Area are both located in the same direction relative to the site. source emissions. The downtown site has high levels of chlorinated compounds compared to other sites in the Pittsburgh area (10). Clairton Area Factor. PMF apportioned 64%, 38%, and 36% of the toluene, benzene, and hexane, respectively, in the downtown data set to a single factor. The contribution of this factor was very wind-directionally dependent (Figure 4c), with much higher contributions when the wind was from the southeast direction. About 20 km southeast of downtown Pittsburgh is a cluster of very large industrial facilities in the Clairton Area, including the largest coke works in North America and two large chemical plants. Together, these three facilities emit 207 tons of toluene per year (71% of the total estimated large point source toluene emissions in Allegheny County) (16) and 87 tons per year of benzene (84% of total estimated large point source emissions of benzene) (16). Furthermore, Figure 4c indicates there is good agreement between the chemical signature of this factor and a composite source profile for these facilities. Although the chemical signature of this factor is also similar to that of published gasoline vehicle source profiles, there are major highways in three directions of the site. Therefore, the strong winddirectional dependence of this factor rules out gasoline vehicles as a major contributor to this factor (the contribution of gasoline vehicles is represented in the mobile source factor). In addition, the PMF model for the CMU site identifies a very similar factor which also points at the Clairton Area. Therefore, we believe this factor represents the emissions from the very large industrial sources in the Clairton area. 9442

9

ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 43, NO. 24, 2009

FIGURE 4. Select (high cancer risks) factors derived from the Downtown data set. Panel (a) shows weekend/weekday profile and wind-directional dependence of the Local-Industry Factor; panel (b) shows the temporal profile of the Mobile Sources Factor and compares this factor profile to motor vehicle source profiles from Schauer et al. (18, 19); and panel (c) compares the Clairton Area Factor to source profiles for industrial sources in the Clairton Area taken from the ACPSEI (16) and the gasoline vehicle source profile from Schauer et al. (18, 19). Panel (c) also shows wind directional dependence of the Clairton Area Factor. Error bars represent uncertainty in the mean weekend/ weekday concentration. The wind rose in the bottom panel shows the location of major source areas relative to the site. Carnegie Mellon University (CMU) PMF Solution. Figure 2c plots the species distribution for the optimum PMF model for the CMU data set. This is a 6-factor PMF model with a -10 rotation that apportioned 82% of the additive cancer risks at the site. The six factors were the Clairton Area factor (27% of the additive cancer risks), Mobile Sources factor (17% of the risks), and four factors associated with CMU research laboratories (38% of the risks). We now discuss the high risk factors from the CMU PMF model. Clairton Area Factor. Figure 2c indicates that PMF apportioned the majority of the 1,1-dichloroethane, benzene, styrene, and xylenes in the CMU data set to one factor. Figure 5a shows that the contribution of this factor is strongly correlated with wind direction. Its contribution was much higher when the wind was from the southeast, when the site was downwind of the sources in the Clairton area. The comparisons with source profiles taken from the ACPSEI (16) shown in Figure 5a support the conclusion that emissions from the Clairton area are strongly influencing pollutant concentrations at the CMU site. For example, the ratios of toluene, benzene, and styrene in this factor match a composite emission profile for major industrial sources in

FIGURE 5. Select (high cancer risk) factors derived from the Carnegie Mellon University data set. Panel (a) compares the Clairton Area Factor to source profiles for industrial sources in the Clairton Area taken from the ACPSEI (16); and panel (b) compares the Mobile Sources Factor to motor vehicle source profiles from Schauer et al. (18, 19). Also shown is the average wind-directional dependence of the two factors. The wind rose in the bottom panel shows the location of major source areas relative to the site. the Clairton area (16). Acetone and MEK also contribute significantly to this factor. These species are secondary products of VOC oxidation. Therefore, this factor appears to be a combination of emissions from the Clairton area and aged more-regional emissions. Mobile Sources Factor. Figure 2c indicates that PMF apportioned many MSATs at the CMU site to one factor. Figure 5b indicates that the ratios of MSATs and VOCs in this factor are similar to published gasoline vehicle source profiles (17-19, 21). Therefore, this factor appears to be dominated by emissions from gasoline powered vehicles as opposed to diesel vehicle emissions due to the low concentrations of acetone and the higher ratio of toluene to benzene. The contribution of this factor exhibits significant temporal variability but is not wind-directional dependent (Figure 5b). This factor also contains some chlorinated compounds that are not associated with mobile source emissions.

Discussion This paper describes a method to identify priority sources for risk reduction. We have applied the method to apportion cancer risks due to gas-phase organic air toxics. It can also be used to apportioned hazard quotients and other types of additive health end points associated with air toxics and other pollutants. The PMF analysis provides insight into the relative contribution of emissions from different types of sources to cancer risks. Of particular interest are the sources of benzene, which is emitted by both mobile and nonmobile sources including coke ovens, chemical plants, and gasoline powered vehicles (16-20). In 2007, the EPA adopted significant new regulations to reduce benzene emissions from motor vehicles. At the Avalon, downtown, and CMU sites, 66%, 53%, and

53%, respectively, of the benzene was apportioned to emissions from metallurgical coke production. Therefore, in Pittsburgh, emissions from nonmobile sources contribute substantially to benzene exposures. There are a number of important commonalities among the PMF solutions. First, all of the solutions have a strong mobile source factor. At the Avalon and CMU sites this factor is more similar to published gasoline vehicle source profiles, whereas downtown it is more similar to published diesel vehicle source profiles. The stronger diesel influence downtown is likely due to very heavy diesel transit bus traffic on Liberty Avenue (black carbon concentrations at the downtown site were more than three times higher than at other urban sites in Pittsburgh (22)). Second, all of the sites have industrial factors linked to metallurgical coke production (the coke works factor in the Avalon solution and the Clairton Area factor in the other two solutions). In particular the PMF solutions for the CMU and downtown site underscore the large spatial influence that a massive source like the Clairton Coke Works has on air toxics concentrations and health risks in the Pittsburgh region. Given the significant contribution of emissions from industrial sources to ambient benzene concentrations, nonmobile sources contribute more than 69% of the additive cancer risk of the air toxics included in the PMF model at each site. Emissions from both large industrial sources and smaller sources were important. Emissions from metallurgical coke works contribute to cancer risks throughout the county, even at sites such as downtown and CMU which are located more than 9 km from coke works. The PMF results also highlight that small sources can substantially affect risks. For example, emissions from dry cleaning appear to contribute significantly to cancer risks downtown, and emissions from research laboratories are estimated to contribute almost 40% of the risks at the CMU site. No one specific source or source category dominates the cancer risks at all of the sites. Currently, many air toxics assessments rely heavily on inventories and modeling. The PMF results can be used to evaluate these inventories. For example, the Shenango Coke Works and Neville Chemical Company on Neville Island are reported to emit similar quantities of benzene per year, 3.0 versus 3.6 tons per year, respectively (16). However, the PMF results at the Avalon site indicated that the Coke Works Factor contributed 7 times more benzene than the Chemical Plant Factor, suggesting that there are problems with how the emissions from these facilities are represented in inventories. The PMF analysis focused on a subset of the air toxics, specifically gas-phase organics that exhibited significant temporal and spatial variability in the Pittsburgh area. These air toxics are the ones that have significant local sources and therefore can be controlled with targeted, local regulations. Other air toxics that were not included in the PMF model also contribute significantly to cancer risks in the Pittsburgh area (10). First, two regionally distributed air toxics (formaldehyde and carbon tetrachloride) together contribute about the same amount cancer risk as the compounds included in the PMF model (10). This indicates that emissions from regional sources contribute significantly to air toxics risks in Pittsburgh. Second, diesel particulate matter also appears to be an important risk driver in the Pittsburgh area (22). Therefore, conclusions regarding the relative importance of different source classes are sensitive to which compounds are included in the analysis. Finally, the cancer risks posed by all air toxics are much smaller than the overall U.S. lifetime cancer risk of 1 in 3 (23).

Acknowledgments We thank Jason Maranche and Darrell Stern from the Allegheny County Health Department (ACHD) for access to VOL. 43, NO. 24, 2009 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

9443

data and monitoring sites used in this work. This work was funded by the ACHD through the Clean Air fund and by the United States Environmental Protection Agency through the Community Scale Air Toxics Monitoring Program. The statements and conclusions in this paper are those of the authors and not necessarily those of the funding agencies.

Supporting Information Available This information is available free of charge via the Internet at http://pubs.acs.org/.

Literature Cited (1) McCarthy, M. C.; Hafner, H. R.; Montzka, S. A. Background concentrations of 18 air toxics for North America. J. Air Waste Manage. Assoc. 2006, 56 (1), 3–11. (2) Brown, S. G.; Frankel, A.; Hafner, H. R. Source apportionment of VOCs in the Los Angeles area using positive matrix factorization. Atmos. Environ. 2007, 41 (2), 227–237. (3) Choi, Y. J.; Ehrman, S. H. Investigation of sources of volatile organic carbon in the Baltimore area using highly time-resolved measurements. Atmos. Environ. 2004, 38 (5), 775–791. (4) Logue, J. M.; Huff-Hartz, K. E.; Lambe, A. T.; Donahue, N. M.; Robinson, A. L. High time-resolved measurements of organic air toxics in different source regimes. Atmos. Environ., In press, Corrected Proof. (5) Pirrone, N.; Keeler, G. J.; Holsen, T. M. Dry deposition of semivolatile organic compounds to Lake Michigan. Environ. Sci. Technol. 1995, 29 (8), 2123–2132. (6) Loh, M. M.; Levy, J. I.; Spengler, J. D.; Houseman, E. A.; Bennett, D. H. Ranking cancer risks of organic hazardous air pollutants in the United States. Environ. Health Perspect. 2007, 115 (8), 1160–1168. (7) Woodruff, T. J.; et al. Public health implications of 1990 air toxics concentrations across the United States. Environ. Health Perspect. 1998, 106 (5), 245–251. (8) Paatero, P.; Tapper, U. Positive Matrix Factorization - a Nonnegative Factor Model with Optimal Utilization of ErrorEstimates of Data Values. Environmetrics 1994, 5 (2), 111–126. (9) Wittig, A. E.; et al. Pittsburgh air quality study overview. Atmos. Environ. 2004, 38 (20), 3107–3125.

9444

9

ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 43, NO. 24, 2009

(10) Logue, J. M.; Small, M. J.; Stern, D.; Maranche, J.; Robinson, A. L., Spatial variation in ambient air toxics concentrations and health risks between industrial influenced, urban, and rural sites. J. Air Waste Manage. Assoc. 2009, accepted. (11) Paatero, P.; Hopke, P. K.; Begum, B. A.; Biswas, S. K. A graphical diagnostic method for assessing the rotation in factor analytical models of atmospheric pollution. Atmos. Environ. 2005, 39 (1), 193–201. (12) Ramaswami, A.; Milford, J.; Small, M. Integrated Environmental Modeling; John Wiley & Sons, Inc.: Hoboken, NJ, 2005. (13) USEPA Integrated Risk Information System; http://www. epa.gov/ngispgm3/iris/index.html. (14) CALEPA Chronic Reference Exposure Levels; http://www. oehha.org/air/chronic_rels/AllChrels.html. (15) National Research Council, Ed. Complex Mixtures: Methods for in Vivo Toxicity Testing; National Academy Press: Washington, DC, 1998; pp 102-105. (16) ACHD. Point Source Emissions Inventory Report 2004; Allegheny County Health Department: Allegheny County, PA, 2006; p 121. (17) USEPA Mobile Source Air Toxics; http://www.epa.gov/otaq/ toxics.htm. (18) Schauer, J. J.; Kleeman, M. J.; Cass, G. R.; Simoneit, B. R. T. Measurement of emissions from air pollution sources. 2. C-1 through C-30 organic compounds from medium duty diesel trucks. Environ. Sci. Technol. 1999, 33 (10), 1578–1587. (19) Schauer, J. J.; Kleeman, M. J.; Cass, G. R.; Simoneit, B. R. T. Measurement of emissions from air pollution sources. 5. C-1C-32 organic compounds from gasoline-powered motor vehicles. Environ. Sci. Technol. 2002, 36 (6), 1169–1180. (20) NIH Hazardous Substances Data Bank; http://toxnet.nlm. nih.gov/cgi-bin/sis/htmlgen?HSDB. (21) Lough, G. C.; Schauer, J. J.; Lonneman, W. A.; Allen, M. K. Summer and winter nonmethane hydrocarbon emissions from on-road motor vehicles in the Midwestern United States. J. Air Waste Manage. Assoc. 2005, 55 (5), 629–646. (22) Lambe, A. T.; Logue, J. M.; Kreisberg, N. M.; Hering, S. V.; Worton, D. R.; Goldstein, A. H.; Donahue, N. M.; Robinson, A. L. Apportioning black carbon to sources using highly time-resolved ambient measurements of organic molecular markers in Pittsburgh. Atmos. Environ. 2009, 43 (25), 3941–3950. (23) American Cancer Society. Cancer Facts and Figures 2008; Atlanta, GA, 2008; p 72.

ES901683J