Environ. Sci. Technol. 2006, 40, 7811-7819
Source Apportionment of Molecular Markers and Organic Aerosol. 2. Biomass Smoke
ratios of source profiles are critical parameters to consider when evaluating CMB solutions.
A L L E N L R O B I N S O N , * ,† R S U B R A M A N I A N , †,⊥ N E I L M D O N A H U E , ‡ ANNA BERNARDO-BRICKER,§ AND WOLFGANG F ROGGE§ Department of Mechanical Engineering, Carnegie Mellon University, Pittsburgh, Pennsylvania 15213, Department of Chemistry and Department of Chemical Engineering, Carnegie Mellon University, Pittsburgh, Pennsylvania 15213, Department of Civil and Environmental Engineering, Florida International University, Miami, Florida
Emissions from biomass combustion can be an important source of organic aerosol in urban environments (1-4). Organic molecular markers have been used in conjunction with the chemical mass balance (CMB) model to apportion ambient organic aerosol to biomass smoke and other primary sources (1-4). Simoneit (5) reviews molecular markers for emissions from incomplete biomass combustion. In this paper we focus on some of the most commonly used markers: levoglucosan, syringols, and resin acids. Levoglucosan is a general marker for biomass combustion emissions (6). It is a cellulose pyrolysis product; since cellulose is the dominant natural biopolymer, levoglucosan is emitted during the combustion of most biomaterials. Additional molecular markers are used to differentiate between emissions from softwood and hardwood combustion (3, 7). Softwood smoke contains resin acids from gymnosperms (conifers), while emissions from hardwood combustion are enriched in syringols (7, 8). Using CMB with molecular markers to apportion ambient OC to different source classes relies heavily on two implicit assumptions: first, the aggregate emissions from a given source class are well represented by an average source profile; second, the marker-to-OC and marker-to-PM2.5-mass ratios of the emissions are stable and well-known (9). These assumptions are challenged by the highly variable nature of biomass combustion. Currently, more than 35 source profiles with speciated condensed and semi-volatile organics have been published for biomass smoke (10-18). Comparing these profiles reveals large variations in emissions with fuel type and combustion conditions (19). However, most previous CMB analyses of molecular marker data do not consider the effects of source profile variability, even though source profile selection can significantly alter the source contribution estimates (15). This paper is one of a series of papers that examines issues associated with CMB analysis of organic molecular markers using a large dataset collected in Pittsburgh, Pennsylvania (9, 20-22). The goal of this paper is to estimate the contribution of biomass smoke to organic aerosol and fine particle mass using the CMB model, taking into consideration the issues associated with source profile variability. First, ambient concentrations of biomass smoke markers are examined for seasonal patterns and for correlations between different makers. Next, the data are compared to the available source profiles using the approach described in Robinson et al. (9, 23). Scenarios for CMB analysis are then defined, and the paper concludes with a discussion of the CMB results.
Chemical mass balance analysis was performed using a large dataset of molecular marker concentrations to estimate the contribution of biomass smoke to ambient organic carbon (OC) and fine particle mass in Pittsburgh, Pennsylvania. Source profiles were selected based on detailed comparisons between the ambient data and a large number of published profiles. The fall and winter data were analyzed with fireplace and woodstove source profiles, and open burning profiles were used to analyze the spring and summer data. At the upper limit, biomass smoke is estimated to contribute on average 520 ( 140 ng-C m-3 or 14.5% of the ambient OC in the fall, 210 ( 85 ng-C m-3 or 10% of the ambient OC in the winter, and 60 ( 21 ng-C/m-3 or 2% of the ambient OC in the spring and summer. In the fall and winter, there is large day-to-day variability in the amount of OC apportioned to biomass smoke. The levels of biomass smoke in Pittsburgh are much lower than in some other areas of the United States, indicating significant regional variability in the importance of biomass combustion as a source of fine particulate matter. The calculations face two major sources of uncertainty. First, the ambient ratios of levoglucosan, resin acids, and syringhaldehyde concentrations are highly variable implying that numerous sources with distinct source profiles contribute to ambient marker concentrations. Therefore, in contrast to previous CMB analyses, we find that at least three distinct biomass smoke source profiles must be included in the CMB model to explain this variability. Second, the marker-to-OC ratios of available biomass smoke profiles are highly variable. This variability introduces uncertainty of more than a factor of 2 in the amount of ambient OC apportioned to biomass smoke by different statistically acceptable CMB solutions. The marker-to-OC
* Corresponding author phone: (412) 268-3657; fax: (412) 2683348;
[email protected]. † Department of Mechanical Engineering, Carnegie Mellon University. ‡ Departments of Chemistry and Chemical Engineering, Carnegie Mellon University. § Florida International University. ⊥ Current address: Department of Civil and Environmental Engineering, University of Illinois at Urbana-Champaign, Urbana, Illinois 61801. 10.1021/es060782h CCC: $33.50 Published on Web 11/15/2006
2006 American Chemical Society
Introduction
Methods CMB analysis was performed to apportion ambient OC and fine particulate matter in Pittsburgh, Pennsylvania to sources of primary organic aerosol. The analysis uses ambient concentrations of individual organic compounds, PM2.5 elemental carbon, and PM2.5 elemental composition measured on 96 days between July, 2001 and June, 2002 (24). Daily measurements were made in July 2001 and most of January 2002; during other periods 24-hr samples were collected on a 1-in-6 day schedule. Additional details of the dataset are provided in the Supporting Information. VOL. 40, NO. 24, 2006 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
7811
CMB calculates source contribution estimates by fitting ambient concentrations of a specified set of compounds using a linear combination of source profiles. The selection of compounds included in the CMB model is a critical issue; for example, all major sources of each compound must be included in the model and the species should be conserved during transport from source to receptor (25). This work uses the basic set of compounds and source classes developed by Schauer et al. (1, 3). In addition to the biomass-smoke-related markers discussed below, all calculations include four n-alkanes, iso-hentriacontane, anteiso-dotriacontane, four hopanes, four PAHs, two alkanoic acids, palmitoleic acid, cholesterol, titanium, iron, and elemental carbon. Source profiles for eight source classes are included in the model: diesel vehicles, gasoline vehicles, road dust, biomass combustion, cooking, coke production, vegetative detritus, and cigarettes. The calculations were performed using the computer program CMB8 distributed by the U.S. Environmental Protection Agency (USEPA). Additional details on the CMB analysis are contained in the Supporting Information. This paper focuses on using levoglucosan, resin acids, syringaldehyde, and acetosyringone to apportion biomass smoke. CMB analysis is performed using these species with different combinations of biomass smoke source profiles (in addition to a standard set of non-biomass smoke profiles and molecular markers). The specific combinations of source profiles are selected based on comparisons made using scatter and ratio-ratio plots of the ambient molecular marker data with source profiles. More details on the construction, interpretation, and mathematics of ratio-ratio plots are provided in Robinson et al. (12, 15). Biomass smoke also contributes to other species fit by the model such as polycyclic aromatic hydrocarbons, elemental carbon, palmitoleic acid, alkanoic acids, and n-alkanes, but CMB apportions the vast majority of these species to non-biomass-smoke source profiles. CMB analysis requires that compounds fit by the model be chemically stable. Fraser and Lakshmanan (26) argue that levoglucosan is stable during transport and we make that assumption here. We are not aware of research assessing the photochemical stability of syringols. Resin acids are known to interconvert in the atmosphere, with abietic- and pimarictype resin acids being converted to dehydroabietic acid and ultimately to 7-oxodehydroabietic acid (5, 27, 28). For CMB analysis, we assume that resin acids are conserved as a compound class, but that individual resin acids are not conserved. Therefore, we add all of the resin acids together in both the ambient data and the source profiles, and include this sum as a single “species” in the CMB model. The Pittsburgh data are consistent with substantial conversion of unaltered acids to dehydroabietic and 7-oxodehydroabietic acid. The samples were analyzed for 11 different resin acids, but ambient concentrations were dominated by dehydroabietic and 7-oxodehydroabietic acids. These two acids contributed 86 ( 10% (average ( standard deviation) of the total resin acid concentrations, while fresh emissions are often enriched in unaltered acids (11, 12, 17, 18). The ratio of dehydroabietic acid to 7-oxodehydroabietic acid varied seasonally (average wintertime ratio of 4.7 versus 0.7 in the summer). Finally, ambient concentrations of dehydroabietic and 7-oxodehydroabietic acid were inversely correlated (a linear regression yields a slope of -0.93 and R2 of 0.71), consistent with photochemistry.
Results Figure 1 presents a box plot of ambient concentrations of levoglucosan, total resin acids, and syringaldehyde. Results are shown both as a time series, based on grouping the data into 1- or 2-month periods (depending on the number of samples), as well as for other specific groups of samples. The 7812
9
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 40, NO. 24, 2006
FIGURE 1. Box plots constructed from time series of 24-h ambient concentrations of (a) levoglucosan, (b) total resin acids, and (c) syringaldehyde. The boxes stretch from the lower to the upper quartile values; median values are shown as lines across the boxes; the filled diamonds indicate average values. The whiskers indicate the maximum and minimum values; whiskers that intersect upper boundary of the plot indicate that the maximum value is beyond the scale of the graph. The numbers in parentheses in x-axis labels indicate number of samples for specified period. Winter weekend and weekdays are based on the December-01 and the January-02 data. highest levoglucosan and resin acid concentrations were observed in the fall; the peak syringaldehyde concentrations were observed in the winter. During the rest of the year (AprilSeptember), ambient concentrations of the biomass smoke markers were low, with only an occasional modest spike. On average, the levoglucosan concentrations in Pittsburgh appear comparable to those in Houston, TX (29), but are a factor of 5-or-more lower than those in the Southeast U.S. (4). Biomass combustion encompasses a diverse set of sources, the distribution of which varies seasonally. When the weather is warm, forest fires, structural fires, agricultural burns, and other activities involving open burning (land clearing, yard wastes, etc.) are likely the dominant sources of biomass smoke. During colder weather, wood combustion in fireplaces and woodstoves used for space heating may be an important source. Therefore, ambient temperature provides one indicator of the types of biomass sources influencing Pittsburgh. We classified each sampling day as “warm” or “cold” based on the average ambient temperature; days with an average temperature less than 12 °C are defined as cold. We explored other definitions and found that our results are not sensitive to specific boundary between cold and warm days. Thirtyeight percent of the sampled days are classified as cold.
FIGURE 2. Scatter plots of ambient concentrations of biomass smoke markers measured in Pittsburgh. Panels (a) and (b) plot resin acids versus levoglucosan and panels (c) and (d) plot syrignaldehyde against levoglucosan. The lines indicate different published source profiles (10-13, 18). Measurement uncertainty indicated for selected points, uncertainty on other points is comparable. Winters in Pittsburgh are cold with typical average daily temperatures between -5 and 5 °C indicating a consistent demand for space heating. However, the most striking feature of the ambient wintertime data is the significant day-to-day variability in the biomass smoke marker concentrations. For example, wintertime (December through February) levoglucosan levels ranged from 0.3 to 120 ng m-3. On two-thirds of the wintertime sampling days, levoglucosan levels were less than 5 ng m-3. The other biomass smoke markers exhibited similar variability. The large number of cold days with very low levoglucosan levels suggests that wood is not an important fuel for space heating in Pittsburgh nor in upwind areas that contribute significant emissions to the regional air mass. Figure 1 shows that the December and January levoglucosan and syringaldehyde data exhibit a pronounced weekend-weekday pattern. The greatly elevated weekend concentrations are consistent with residential wood burning for pleasuresenjoying a weekend evening around the fire. Alternatively, more wood may also be burned on weekends when people spend a significant number of hours awake at home. Comparing the winter and summer resin acids and syringaldehyde data provides an indication of the type of wood used for space heating. Syringaldehyde concentrations were, on average, seven times higher on cold days (2.8 ng/ m3) than on warm days (0.4 ng/m3), while resin acid concentrations in the winter and summer were more comparable. This indicates that hardwood is burned in the winter for space heating; a survey of wood distributors found that 90% of the fuel wood sold in the Pittsburgh region is hardwood, primarily oak, ash, and maple. In the fall (October and November), the average daily temperature in Pittsburgh varies widely, ranging from 3 to 21 °C on the days on which samples were collected. The peak levoglucosan and resin acid concentrations were observed on cold fall days, but high concentrations of these markers were also observed on warm fall days. High
concentrations on warm days are unlikely to be caused by local wood combustion for space heating; more likely explanations include long-range transport and/or open burning emissions. Although data are not available to assess its contribution on these particular days, regional transport is generally the dominant contributor of fine particle mass and OC in Pittsburgh (30). There are state and local regulations that limit open burning, but one often sees burning of yard waste in rural areas outside of the city, especially in the fall. The high resin acid concentrations in the fall indicate a significant contribution of softwood smoke. Concentrations of biomass smoke markers were consistently low in the spring and summer, indicating minimal influence of summertime wildfire emissions during the study. For example, modestly elevated levels of levoglucosan (3647 ng m-3) were only observed on three of the forty-eight summer samples. On these days, resin acid concentrations (but not syringaldehyde) also spiked, indicating influence of softwood smoke. The importance of softwood smoke on nonfall warm days is also supported by the general increase in resin acid concentrations with increasing levoglucosan shown in Figure 2b (a linear regression yields an R2 of 0.44). In comparison, there is little change in the syringaldehyde concentrations with increasing levoglucosan on non-fall warm days (Figure 2d; R2 ) 0.11). Scatter and ratio-ratio plots for the biomass smoke markers are shown in Figures 2 and 3. These plots are useful for identifying correlations in the ambient concentrations and to compare the ambient data to the published source profiles (9, 23). Figure 2 shows scatter plots of resin acids and syringaldehyde versus levoglucosan. Figure 3 shows ratio-ratio plots of the four biomass smoke markers, with levoglucosan used as the normalizing compound. Figure 3 divides the ambient data into two groups to facilitate the subsequent discussion of scenarios for CMB analysis. Data from the fall and winter are shown in Figure 3a and b, and VOL. 40, NO. 24, 2006 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
7813
FIGURE 3. Ratio-ratio plots of biomass smoke markers: (a) and (b) ambient data on fall and winter days compared to the two space heating mixing scenarios; and (c) and (d) ambient data on non-winter (spring, summer and fall) days compared to open burn scenario. “Softwood” and “Hardwood” are fireplace or woodstove profiles (10-12, 18). “Open burn” includes profiles for prescribed and simulated open burns of hardwood, softwood, and agricultural residues (13-15). Measurement uncertainty indicated for selected points, uncertainty on other points is comparable. data from the non-winter months (spring, summer, and fall) are shown in Figure 3c and d. The organization of the data in a ratio-ratio plot can depend on which compound is used to normalize the data. We examined plots with all possible options, and our conclusions are not sensitive to which species is chosen. Levoglucosan seems like a logical choice since it is present in almost all biomass smoke. Resin acids are a poor choice because they are not present in hardwood smoke, which prevents display of hardwood source profiles. There is only modest correlation among the biomass smoke markers in the Pittsburgh dataset, with linear regressions yielding R2 values between 0.5 and 0.6 (if one excludes the few very high concentration days). The notable exception is that on cold days syringaldehyde and acetosyringone concentrations are strongly correlated, with a linear regression yielding a slope of 4.2, zero intercept, and an R2 of 0.94. The modestly correlated ambient data creates relatively disorganized ratio-ratio plots. For example, Figure 3 indicates that the ratios of the ambient concentrations of the different biomass smoke markers vary by at least 1 order of magnitude. This variability implies that there are significant day-to-day changes in the chemical composition of the biomass smoke influencing Pittsburgh. The greatest variability was observed in the winter, when the syringaldehyde-to-levoglucosan ratios varied by almost 2 orders of magnitude. Although ratios on 7814
9
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 40, NO. 24, 2006
some low concentration days are highly uncertain (e.g., the point in the upper-right-hand corner of Figure 3a with very large error bars), the overall variability is much greater than measurement uncertainty on most days. The poorly organized ratio-ratio plots imply that a diverse set of sources contributes to the biomass smoke marker concentrations in Pittsburgh. Comparison of Ambient Data with Source Profiles. We consider more than 35 different biomass-combustion source profiles that report levoglucosan, syringols, and resin acids data (10-15). The vast majority of these profiles are for soft or hardwood combustion in either residential fireplaces or woodstoves (10-12, 18). More limited data are available for open or prescribed burns (13-15). We assume that hardwood smoke contains no resin acids because common hardwoods do not have resin acid ducts. A few of the published hardwood profiles do report resin acid emissions, which have been attributed to contamination from earlier experiments using softwood (10). Profiles from Rogge et al. (17) and McDonald et al. (16) were not considered because they do not report levoglucosan data. Source profiles appear as lines in the scatter plots presented in Figure 2 and as points in the ratio-ratio plots shown in Figure 3 (9, 23). To reduce clutter, only a few profiles that appear consistent with some fraction of the ambient data are shown in the scatter plots. Comparing source profiles
to the ambient data in Figure 2 illustrates the day-to-day changes in biomass smoke composition. For example, on the winter days with the highest concentrations of syringaldehyde and levoglucosan, the ambient data are most consistent with the Fine et al. (18) Black Cherry profile, while data on lower concentration winter days are more similar to Fine et al. (18) American beech profile. While there is some clustering of source profiles by wood type (hardwood versus softwood) in the ratio-ratio plots shown in Figure 3, the most striking feature is the significant variability across the set of published source profiles. For example, resin-acid-to-levoglucosan ratios of the published softwood profiles span 3 orders of magnitude. Each profile represents the emissions from a single or, at most, a small number of experiments; therefore, the large variability of the published profiles is to be expected given the heterogeneous and poorly controlled nature of wood combustion. Significant variability is observed across the set of source profiles reported by a single study, so the problem is not due to differences in analytical methods used by different research groups. The ratio-ratio plots presented in Figure 3 allow one to easily compare the ambient data to the entire set of source profiles. The majority of the ambient data, including all of the high concentration data, falls within regions of the plot that can be explained by linear mixing of different combinations of the available source profiles. However, the variability in the ambient ratios means that the source profile representing the aggregate emissions from biomass combustion changes from day to day, greatly complicating selection of source profiles for CMB analysis. A subset of the ambient data is inconsistent with any possible combination of the published profiles. For example, ten winter days have syringaldehyde-to-levoglucosan ratios greater than 0.5; these points appear in the upper-righthand corner of the ratio-ratio plot in Figure 3a and to the left of the Fine et al. (18) American beech profile in Figure 2d (the profile with the largest syringaldehyde-to-levoglucosan ratio). The syringaldehyde-to-levoglucosan ratios on these days are greater than any of the published source profiles. One potential explanation is measurement uncertainty. The ambient concentrations on many of these days were low, resulting in highly uncertain marker ratios. However, on some inconsistent days, ambient concentrations of syringaldehyde were above average, and the error bars do not overlap any possible source profile combinations. A likely explanation for these points is that the available source profiles do not capture the actual variability in biomass smoke composition. Another factor may be photochemical aging of the emissions, which maybe especially important in Pittsburgh given the significant contribution of regional transport. CMB Analysis. CMB analysis was performed using different combinations of biomass smoke source profiles, referred to as mixing scenarios, to constrain the potential contribution of biomass smoke to OC in Pittsburgh. Given the size of our ambient dataset, our approach for selecting source profiles differs from most previous CMB studies. Most studies select source profiles based on wood consumption data (1, 4, 31). This approach assumes that fuel type is a major factor for the differences between source profiles. Certainly differences in the emissions of certain markers, for example those used to differentiate between hard and softwood smoke, are directly related with fuel type. However, combustion conditions also play a major role in source profile variability. For example, levoglucosan-to-OC ratios range by almost a factor of 3 across the set of published oak profiles. This is not surprising since levoglucosan emissions depend strongly on combustion temperature (6) and biomass combustion is typically a poorly controlled and highly variable process. Therefore, differences in combustion conditions likely contribute significantly to the variability within the set
of published hardwood or softwood profiles. Thus we deemphasize using specific types of wood as the primary basis for source profile selection and instead rely on our large ambient dataset to guide our decisions. We do avoid profiles for woods not found in the Eastern U.S., but this leaves a large number of potential profiles. To sort through this set we used a combination of ratio-ratio plots and CMB analyses to identify combinations of profiles that best describe the entire ambient dataset. Another important consideration is the number of source profiles included in the model. Most previous CMB analyses have represented the aggregate emission from biomass combustion using one source profile, either based on a single source test (32) or, more commonly, a composite of several profiles (1, 4, 31). A source profile appears as a point on a ratio-ratio plot (9, 23). Given the significant variability of the Pittsburgh data (Figure 3), it is not surprising that CMB models based on a single biomass source profile, even one based on an average of many different profiles, produces an overall solution for the entire dataset with poor statistical quality. A more sophisticated approach is to include separate hardwood and softwood profiles in the CMB model (3). A two-profile approach assumes that, within uncertainty, the ambient data organize along a mixing line in a ratio-ratio plot (9, 23). Several mixing lines are drawn in Figure 3; each line represents all possible linear combinations of a source profile pair. Clearly a two-source mixing line defined by any single pair of source profiles cannot explain the widely varying ratios of the biomass smoke marker concentrations in Pittsburgh. In principle, one could select a different composite profile or a different two-profile combination for each data point; however, this will yield a very large number of CMB solutions with little basis for selecting among them. The approach adopted here is to include three separate biomass smoke source profiles in the CMB model. Three profiles is the minimum number required to define a mixing region large enough to encapsulate the majority of the ambient data. A mixing region defines all possible linear combinations of the three source profiles (9, 23). The benefit of this approach is that a CMB model based on a single set of three biomass source profiles can be used to analyze the entire dataset with good statistical quality. To account for expected seasonal changes in the nature of biomass combustion sources, the winter data (Dec-Mar) are analyzed with different combinations of space heating profiles (woodstove and fireplace), while open burning profiles are used in the analysis of spring and summer data (April-September). We analyze the fall data (October-November) with both sets of the scenarios because of the large number of cold days as well as the possibility for significant contributions from open burning during that period. We perform CMB analysis with two space heating scenarios each involving three fireplace or woodstove profiles. Space Heating No. 1 includes the Schauer et al. (10) pine and oak profiles and the Fine et al. (18) American beech profile. Space Heating No. 2 includes the Fine et al. (11) eastern hemlock, eastern white pine, and red maple profiles. The selection of these two scenarios was motivated in part by organization of the ambient data and source profiles in the ratio-ratio plots shown in Figure 3. The distribution of the ambient data relative to the source profiles in Figure 3 indicates that we must include both soft and hardwood profiles in the model. Space Heating No. 1 includes a single softwood profile and two hardwood profiles whose syringolto-levoglucosan ratios essentially span the range of the published hardwood profiles. Space Heating No. 2 includes a single hardwood profile and two softwood profiles whose resin-acids-to-levoglucosan ratios essentially span the range of the published softwood profiles. Therefore, Space Heating VOL. 40, NO. 24, 2006 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
7815
No. 1 uses the variability in the published hardwood profiles to account for the variability in the ambient dataset while Space Heating No. 2 uses variability in the softwood profiles. Within measurement uncertainty, the mixing lines defined by either space heating scenario encompass the majority of the ambient data. A final motivation for considering these two scenarios is that, among the set of potential three-profile solutions, these two scenarios fall at the upper and lower ends of the range with respect to the amount of ambient OC apportioned to biomass smoke. The reasons for this are discussed below. Figure 3 indicates that the mixing region defined by the Space Heating No. 2 scenario encapsulates a larger fraction of the fall and winter ambient data than the Space Heating No. 1 scenario, especially if one considers the acetosyringone data plotted in Figure 3b. Not unexpectedly, the statistical quality of the overall CMB solution for the Space Heating No. 2 scenario is slightly better than the Space Heating No. 1 scenario. As previously discussed, a number of fall and winter points in the upper-right-hand quadrant of the Figure 3a and b are inconsistent with any of the published profiles. Measurement uncertainty allows CMB to find solutions even for these points outside the mixing region, but with somewhat lower statistical quality. Given the limited number of published profiles, we only consider one open burn scenario: the Hays et al. (13) MHFF and Florida Palmetto and Slash Pine profiles, and the Hays et al. (14) wheat straw profile. Figure 3 indicates that other combinations of three open burning profilessone agricultural burning profile, the MHFF profile (the only published hard wood open burning profile), and one softwood open burning profilesalso describe the non-winter ambient data. These other combinations produce similar CMB results to the open burn scenario discussed here. Figure 3 indicates generally good agreement between the non-winter ambient data and the mixing region defined by the open burn scenario; however, days with acetosyringoneto-levoglucosan ratios less than 0.01 shown in Figure 3d present a challenge. Given the uncertainty in the data, CMB finds statistically acceptable solutions even when acetosyringone is included as a fitting species. Eliminating acetosyringone improves the statistics of the CMB solution but only modestly increases the amount of OC apportioned to biomass smoke. Almost all of the non-winter ambient acetosyringone data can be satisfactorily explained if the Schauer et al. (10) Pine Fireplace profile is added to the model; this improves the fitting statistics but does not appreciably change the amount of OC apportioned to biomass smoke. However, justifying the use of a fireplace profile to analyze summer data seems difficult. The CMB model using the open burn scenario does not converge on the fall day with the highest biomass smoke marker concentrations (240 ng m-3 of levoglucosan). The problem is not the biomass smoke markers but the higher odd n-alkanes (C27, C29, C31, C33) commonly associated with vegetative detritus (33). These compounds are found in appreciable levels in the Hays et al. (13) MHFF and the Hays et al. (14) agricultural residue burning source profiles. CMB models using these profiles significantly over-apportion the higher odd n-alkanes on days with high biomass smoke marker concentrations. CMB calculates a number of statistical parameters used to evaluate the quality of a solution. Watson et al. (25) describes the target ranges for these parameters. Based on these guidelines, all of the scenarios considered yield solutions that are statistically acceptable on almost all days. For example, the average R2 value for both space heating solutions is 0.91 with confidence levels greater than 90% on all days based on the χ2 and number of degrees of freedom. On some days the CMB solutions do not satisfy all of the 7816
9
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 40, NO. 24, 2006
FIGURE 4. Scatter plot of the daily ambient OC apportioned to biomass by the Space Heating No. 1 and Space Heating No. 2 scenarios for the October through March period. Error bars indicate one standard error as calculated by CMB. Solid line is linear regression forced through the origin. performance parameters; for example, the T-STAT for the biomass smoke OC is less than 2. This occurs on low concentration days and when the ambient data fall well outside of the mixing region and have small error bars. Overall, the statistical quality of the CMB solutions is quite similar (especially on days with high biomass smoke marker concentrations); therefore, the performance measures calculated by CMB do not provide a basis for selecting one solution over the other. Additional details on the statistical quality of the solutions are provided in the Supporting Information. Although the statistical quality of the different CMB solutions is comparable, the amount of ambient OC apportioned to biomass smoke by the two space heating scenarios is different. This is illustrated in Figure 4, which presents a scatter plot of the daily ambient OC apportioned to biomass smoke by the two space heating scenarios for the October through March period. The results are highly correlated with a linear regression through the origin calculating a slope of 2.4 (Space Heating No. 2 greater than Space Heating No. 1) and an R2 value of 0.96. On the high concentration days, the bias between the two solutions is significantly greater than the standard errors calculated by CMB. In the fall, when calculations were made with all three scenarios, all three solutions are strongly correlated and the amount of ambient OC apportioned to biomass smoke by the open burning scenario falls between the two space heating scenarios. The fact that all of the different solutions are strongly correlated is not surprising, given that they are all based on fits of the same ambient species. The systematic biases between different solutions are caused by differences in the marker-to-OC ratios of the biomass smoke source profiles. CMB only fits the ambient concentrations of the species included in the model; OC is not included in the model because source profiles do not exist for important contributors to OC such as secondary organic aerosol (SOA). After determining the optimum fit, CMB apportions the ambient OC to each source using the marker-to-OC ratio of the different source profiles. This second step is built into the analysis because the source profiles used in CMB have been normalized by the OC. Figure S.1 (Supporting Information) compares marker-to-OC ratios for a large number of published biomass smoke profiles. Differences in the levoglucosan-to-OC ratios of the source profiles explain much of the bias in the amount of OC apportioned to biomass smoke by the two space heating
FIGURE 5. Time series of ambient OC apportioned to biomass smoke by CMB. Error bars are one standard error as calculated by CMB. scenarios. Space Heating No. 1 includes two Schauer et al. (10) profiles with large levoglucosan-to-OC ratios while the Space Heating No. 2 scenario is based on three profiles with relatively small levoglucosan-to-OC ratios. The arithmetic average levoglucosan-to-OC ratio of the three profiles used in Space Heating No. 1 is 0.19 versus 0.085 for Space Heating No. 2. Therefore, assuming equal weighting of each profile, Space Heating No. 2 will apportion 2.2 times more OC to biomass smoke than Space Heating No. 1 for a given ambient levoglucosan concentrationsessentially the same bias as that shown in Figure 4. Figure 5 shows a time series of the ambient OC apportioned by CMB to biomass smoke. For the October through March period, results are plotted for the maximum space heating scenario (Space Heating No. 2) and for the open burn scenario for the remainder of the study. Therefore, Figure 5 represents our upper-bound estimate. On a study average basis, this combination apportions 160 ( 50 ng-C/ m3 ((standard error) or 5% of the ambient OC to biomass smoke. The peak contribution is in the fall when an average of 520 ( 140 ng-C/m3 or 14.5% of the total OC is apportioned to biomass smoke. In the fall and winter, the variability in the ambient marker concentrations causes significant dayto-day variability in the amount of OC apportioned to biomass smoke. On only 28% of the fall and winter days (and zero days in the other periods) was more than 10% of the ambient OC apportioned to biomass smoke; the maximum daily contribution was 43%. CMB analysis was also performed to estimate the contribution of each source profile to ambient fine particle mass. This was done by repeating the CMB calculations with profiles normalized by PM2.5 mass emission rates as opposed to OC emission rates. Therefore, the difference between the OC and PM2.5 mass results is determined by the PM2.5-massto-OC ratio of the different source profiles. The average PM2.5mass-to-OC ratio of the three profiles used in the Space Heating No. 2 scenario is only 1.17 so it only apportions slightly more ambient PM2.5 mass than OC to biomass smokesfor example, on average it apportions 550 ( 160 ng m-3 of PM2.5 mass versus 520 ( 140 ng-C/m3 of OC during the fall. The average bias between the Space Heating No. 2 and Space Heating No. 1 solutions is only a factor of 1.6 on a PM2.5 mass basis versus a factor of 2.4 on an OC basis (Figure 4). This reduction is due to marker-to-PM2.5 ratios of the source profiles being more similar than the marker-toOC ratios. The unexpectedly small PM2.5-mass-to-OC ratios of many of the biomass smoke profiles raise concerns about the potential effects of organic sampling artifacts on the CMB results. Biomass smoke is expected to have PM2.5-mass-to-
OC ratios of around two, given the relatively polar nature of the organic species (34) and the contribution of elemental carbon and other non-carbonaceous species to the emissions. However, the ratios of all of the profiles used by the Space Heating No. 2 scenario are much smaller than two. Unexpectedly small PM2.5-mass-to-OC ratios are consistent with a large positive sampling artifact caused by the adsorption of organic vapors to quartz filters. These artifacts can be significant when sampling wood smoke emissions (35). Since many published source profiles do not report artifactcorrected OC emissions, we only consider non-corrected data from undenuded quartz filters for both source and ambient samples in our analysis. Correcting for positive sampling artifacts reduces the marker-to-OC ratios, which, in turn, reduces the amount of OC apportioned by the CMB model. For example, the very small PM2.5-mass-to-OC ratio of the Space Heating No. 2 profiles (1.17) indicate that up to 50% of the OC emissions could be positive sampling artifact, correcting for such an artifact would reduce the ambient OC apportioned by CMB to biomass smoke by 50%. Although not used in the CMB calculations, ambient measurement of OC provides a useful reference point for artifacts; in Pittsburgh, sampling artifact create a net positive artifact of 10-20% of the ambient OC measured with a quartz filter (36).
Discussion On most days biomass smoke is only a minor source of ambient OC and fine particulate matter in Pittsburgh. The amount of OC apportioned to biomass smoke in Pittsburgh is comparable to that in Houston, but on a monthly average basis, are a factor of 5-20 less than that in the Southeastern U.S. (4). This indicates that there are large regional differences in the importance of biomass combustion as a source of fine particles, even for locations with cold climates. On only a few peak days were biomass-smoke OC levels in Pittsburgh comparable to the monthly average levels in the wintertime in the Southeast. The peak biomass smoke concentrations in Pittsburgh are also substantially less than those predicted during a wintertime air pollution episode in the central valley in California (3). These differences are ultimately caused by the lower ambient biomass smoke marker concentrations in Pittsburgh, and not the specific source profiles used by the different studies. The amount of ambient OC apportioned to biomass smoke varies by more than a factor of 2, depending on which source profiles are used in the CMB model. Some of this variability is related to sampling artifacts. A critical parameter is the source profile marker-to-OC ratio - we have shown that differences in this ratio underlie the biases in the ambient OC apportioned by CMB using different sets of profiles (Figure 4). In fact, by simply comparing marker-to-OC ratios of different profiles one can gain significant insight into how much OC will be apportioned to different profiles without even running CMBsmore OC will be apportioned to profiles with smaller marker-to-OC ratios. Marker-to-OC ratios of the biomass smoke profiles span more than an order of magnitude (Figure S.1), which can create similar variability in the CMB results. Our approach of including three separate biomass smoke profiles in the CMB model substantially reduces the effects of profile-toprofile variability because each solution is a weighted average of multiple profiles. For the sets of profiles considered here better agreement was observed when comparing CMB solutions on a PM2.5 mass basis. However, source profile marker-to-PM2.5 ratios are also highly variable (Figure S.1); therefore, performing the calculations on a PM2.5 mass basis does not eliminate problems associated with source profile variability. Uncertainty created by source profile variability is a problem for all source classes (9, 20, 21). The key difference VOL. 40, NO. 24, 2006 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
7817
between biomass smoke and other major source classes is the large variability in the ambient biomass smoke molecular marker data. For example, ambient concentrations of molecular markers for motor vehicles and meat cooking create relatively organized scatter and ratio-ratio plots (9, 20, 21). Well-correlated ambient data implies that the aggregate emissions from all of the sources in a given source class have a well-defined source profile. This occurs because atmospheric mixing averages out the source-to-source variability in emissions from a large number of individual sources. The source profiles used in CMB are assumed to represent the aggregate emissions from a source class. In comparison to other source-specific sets of markers, the ambient concentrations of the biomass smoke markers are only modestly correlated (Figure 2) and form relatively disorganized ratioratio plots (Figure 3). This means that the composition (and the aggregate source profile) of the aged biomass smoke influencing Pittsburgh changes substantially from day to day, presumably because of the poorly controlled, highly variable nature of biomass combustion. In this paper we have accounted for the variability in the ambient data by including three separate biomass smoke source profiles in the CMB model. However, even with three profiles, all of the variability in the ambient dataset cannot be accounted for. Significantly, on some days, no possible combination of published source profiles can explain the ambient biomass marker data, pointing to the need for more source characterization, better understanding of the photochemical stability of molecular markers, and improved analytical techniques.
Acknowledgments This research was conducted as part of the Pittsburgh Air Quality Study, which was supported by the U.S. Environmental Protection Agency under contract R82806101 and the U.S. Department of Energy National Energy Technology Laboratory under contract DE-FC26-01NT41017. This research was also supported by the EPA STAR program through the National Center for Environmental Research (NCER) under grant R832162. This paper has not been subject to EPA’s required peer and policy review, and therefore does not necessarily reflect the views of the Agency. No official endorsement should be inferred.
Supporting Information Available Additional details of our analysis. This material is available free of charge via the Internet at http://pubs.acs.org.
Literature Cited (1) Schauer, J. J.; Rogge, W. F.; Hildemann, L. M.; Mazurek, M. A.; Cass, G. R. Source apportionment of airborne particulate matter using organic compounds as tracers. Atmos. Environ. 1996, 30 (22), 3837-3855. (2) Watson, J. G.; Fujita, E. M.; Chow, J. C.; Zielinska, B. Richards, L. W.; Neff, W.; Dietrich, D. Northern Front Range Air Quality Study Final Report and Supplemental Volumes, DRI Document no. 6580-8750-1F2; Desert Research Institute: Reno, NV, 1998. (3) Schauer, J. J.; Cass, G. R. Source apportionment of wintertime gas-phase and particle-phase air pollutants using organic compounds as tracers. Environ. Sci. Technol. 2000, 34 (9), 18211832. (4) Zheng, M.; Cass, G. R.; Schauer, J. J.; Edgerton, E. S. Source apportionment of PM2.5 in the southeastern United States using solvent-extractable organic compounds as tracers. Environ. Sci. Technol. 2002, 36 (11), 2361-2371. (5) Simoneit, B. R. T. Biomass burningsa review of organic tracers for smoke from incomplete combustion. Appl. Geochem. 2002, 17, 129-162. (6) Simoneit, B. R. T.; Schauer, J. J.; Nolte, C. G.; Oros, D. R.; Elias, V. O.; Fraser, M. P.; Rogge, W. F.; Cass, G. R. Levoglucosan, a tracer for cellulose in biomass burning and atmospheric particles. Atmos. Environ. 1999, 33 (2), 173-182. 7818
9
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 40, NO. 24, 2006
(7) Simoneit, B. R. T.; Rogge, W. F.; Mazurek, M. A.; Standley, L. J.; Hildemann, L. M.; Cass, G. R. Lignin pyrolysis products, lignans, and resin acids as specific tracers of plant classes in emissions from biomass combustion. Environ. Sci. Technol. 1993, 27 (12), 2533-2541. (8) Hawthorne, S. B.; Miller, D. J.; Barkley, R. M.; Krieger, M. S. Identification of methoxylated phenols as candidate tracers for atmospheric wood smoke pollution. Environ. Sci. Technol. 1988, 22(10), . (9) Robinson, A. L.; Subramanian, R.; Donahue, N. M.; BernardoBricker, A.; Rogge, W. F. Source apportionment of molecular markers and organic aerosols1. Polycyclic aromatic hydrocarbons and methodology for data visualization. Environ. Sci. Technol. 2006, 40, 7813-7820. (10) Schauer, J. J.; Kleeman, M. J.; Cass, G. R.; Simoneit, B. R. T. Measurement of emissions from air pollution sources. 3. C-1C-29 organic compounds from fireplace combustion of wood. Environ. Sci. Technol. 2001, 35 (9), 1716-1728. (11) Fine, P. M.; Cass, G. R.; Simoneit, B. R. T. Chemical characterization of fine particle emissions from fireplace combustion of woods grown in the northeastern United States. Environ. Sci. Technol. 2001, 35 (13), 2665-2675. (12) Fine, P. M.; Cass, G. R.; Simoneit, B. R. T. Chemical characterization of fine particle emissions from the fireplace combustion of woods grown in the southern United States. Environ. Sci. Technol. 2002, 36 (7), 1442-1451. (13) Hays, M. D.; Geron, C. D.; Linna, K. J.; Smith, N. D.; Schauer, J. J. Speciation of gas-phase and fine particle emissions from burning of foliar fuels. Environ. Sci. Technol. 2002, 36 (11), 22812295. (14) Hays, M. D.; Fine, P. B.; Geron, C. D.; Kleeman, M. J.; Gullett, B. K. Open burning of agricultural biomass: Physical and chemical properties of particle-phase emissions. Atmos. Environ. 2005, 39, 6747-6764. (15) Lee, S.; Baumann, K.; Schauer, J. J.; Sheesley, R. J.; Naeher, L. P.; Meinardi, S.; Blake, D. R.; Edgerton, E. S.; Russell, A. G.; Clements, M. Gaseous and particulate emissions from prescribed burning in Georgia. Environ. Sci. Technol. 2005, 39 (23), 9049-9056. (16) McDonald, R. D.; Zielinska, B.; Fujita, E. M.; Sagebiel, J. C.; Chow, J. C.; Watson, J. G. Fine particle and gaseous emission rates from residential wood combustion. Environ. Sci. Technol. 2000, 34 (11), 2080-2091. (17) Rogge, W. F.; Hildemann, L. M.; Mazurek, M. A.; Cass, G. R.; Simoneit, B. R. T. Sources of fine organic aerosol. 9. Pine, oak and synthetic log combustion in residential fireplaces. Environ. Sci. Technol. 1998, 32 (1), 13-22. (18) Fine, P. M.; Cass, G. R.; Simoneit, B. R. T. Chemical characterization of fine particle emissions from the wood stove combustion of prevalent United States tree species. Environ. Eng. Sci. 2004, 21 (6), 705-721. (19) Fine, P. M.; Cass, G. R.; Simoneit, B. R. T. Organic compounds in biomass smoke from residential wood combustion: Emissions characterization at a continental scale. J. Geophys. Res.: Atmos. 2002, 107, (D21). (20) Robinson, A. L.; Subramanian, R.; Donahue, N. M.; BernardoBricker, A.; Rogge, W. F. Source apportionment of molecular markers and organic aerosols3. Food cooking emissions. Environ. Sci. Technol. 2006, 40, 7830-7837. (21) Subramanian, R.; Donahue, N. M.; Bernardo-Bricker, A.; Rogge, W. F.; Robinson, A. L. Contribution of motor vehicle emissions to organic carbon and fine particle mass in Pittsburgh, Pennsylvania: Effects of varying source profiles and seasonal trends in ambient marker concentrations. Atmos. Environ. 2006, in press. (22) Subramanian, R.; Donahue, N. M.; Bernardo-Bricker, A.; Rogge, W. F.; Robinson, A. L. Sources of ambient aerosol in a regionally dominated, urban location: Results from chemical mass balance modeling with molecular markers applied to Pittsburgh, Pennsylvania. Environ. Sci. Technol. in preparation. (23) Robinson, A. L.; Donahue, N. M.; Rogge, W. F. Photochemical oxidation and changes in molecular composition of organic aerosol in the regional context. J. Geophys. Res: Atmos. 2006, 111, (D03302), doi:10.1029/2005JD006265. (24) Wittig, A. E.; Anderson, N.; Khlystov, A. Y.; Pandis, S. N.; Davidson, C.; Robinson, A. L. Pittsburgh air quality study overview. Atmos. Environ. 2004, 38 (20), 3107-3125. (25) Watson, J. G.; Robinson, N. F.; Fujita, E. M.; Chow, J. C.; Pace, T. G.; Lewis, C.; Coulter, T. CMB8 Applications and validation
(26)
(27) (28)
(29) (30)
(31)
protocol for PM2.5 and VOCs, document no. 1808.2D1; Desert Research Institute: Reno, NV, 1998. Fraser, M. P.; Lakshmanan, K. Using levoglucosan as a molecular marker for the long-range transport of biomass combustion aerosols. Environ. Sci. Technol. 2000, 34 (21), 45604564. Corin, N. S.; Backlund, P. H.; Kulovaara, M. A. M. Photolysis of the resin acid dehydroabietic acid in water. Environ. Sci. Technol. 2000, 34 (11), 2231-2236. Egenberg, I. M.; Holtekjølen, A. K.; Lundanes, E. Characterisation of naturally and artificially weathered pine tar coatings by visual assessment and gas chromatography-mass spectrometry. J. Cultural Heritage 2203, 4, 221-241. Fraser, M. P.; Yue, Z. W.; Tropp, R. J.; Kohl, S. D.; Chow, J. C. Molecular composition of organic fine particulate matter in Houston, TX. Atmos. Environ. 2002, 36 (38), 5751-5758. Tang, W.; Raymond, T.; Wittig, A. E.; Davidson, C. I.; Pandis, S. N.; Robinson, A. L.; Crist, K. Spatial Variations of PM2.5 during the Pittsburgh Air Quality Study. Aerosol Sci. Technol. 2004, 38 (S2), 80-90. Sheesley, R. J.; Schauer, J. J.; Bean, E.; Kenski, D. Trends in secondary organic aerosol at a remote site in Michigan’s upper peninsula. Environ. Sci. Technol. 2004, 38 (24), 64916500.
(32) Fraser, M. P.; Yue, Z. W.; Buzcu, B. Source apportionment of fine particulate matter in Houston, TX, using organic molecular markers. Atmos. Environ. 2003, 37 (15), 2117-2123. (33) Rogge, W. F.; Hildemann, L. M.; Mazurek, M. A.; Cass, G. R.; Simoneit, B. R. T. Sources of fine organic aerosol.4. Particulate abrasion products from leaf surfaces of urban plants. Environ. Sci. Technol. 1993, 27 (13), 2700-2711. (34) Turpin, B. J.; Lim, H. J. Species contributions to PM2.5 mass concentrations: Revisiting common assumptions for estimating organic mass. Aerosol Sci. Technol. 2001, 35 (1), 602-610. (35) Lipsky, E. M.; Robinson, A. L. Effects of dilution on fine particle mass and partitioning of semivolatile organics in diesel exhaust and wood smoke. Environ. Sci. Technol. 2006, 40 (1), 155162. (36) Subramanian, R.; Khlystov, A. Y.; Cabada, J. C.; Robinson, A. L. Positive and negative artifacts in particulate organic carbon measurements with denuded and undenuded sampler configurations. Aerosol Sci. Technol. 2004, 38 (S1), 27-48.
Received for review April 1, 2006. Revised manuscript received September 12, 2006. Accepted October 2, 2006. ES060782H
VOL. 40, NO. 24, 2006 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
7819
