Particle Partitioning of Semivolatile Organic

To quantify and minimize the influence of gas/particle (G/P) partitioning on receptor-based source apportionment using particle-phase semivolatile org...
0 downloads 0 Views 1MB Size
Article pubs.acs.org/est

Impact of Gas/Particle Partitioning of Semivolatile Organic Compounds on Source Apportionment with Positive Matrix Factorization Mingjie Xie,*,†,§ Michael P. Hannigan,† and Kelley C. Barsanti‡ †

Department of Mechanical Engineering, College of Engineering and Applied Science, University of Colorado, Boulder, Colorado 80309, United States ‡ Department of Civil and Environmental Engineering, Portland State University, Post Office Box 751, Portland, Oregon 97207, United States S Supporting Information *

ABSTRACT: To quantify and minimize the influence of gas/particle (G/P) partitioning on receptor-based source apportionment using particle-phase semivolatile organic compound (SVOC) data, positive matrix factorization (PMF) coupled with a bootstrap technique was applied to three data sets mainly composed of “measured-total” (measured particle- + gas-phase), “particle-only” (measured particle-phase) and “predicted-total” (measured particle-phase + predicted gas-phase) SVOCs to apportion carbonaceous aerosols. Particle- (PM2.5) and gas-phase SVOCs were collected using quartz fiber filters followed by PUF/XAD-4/PUF adsorbents and measured using gas chromatography− mass spectrometry (GC−MS). Concentrations of gas-phase SVOCs were also predicted from their particle-phase concentrations using absorptive partitioning theory. Five factors were resolved for each data set, and the factor profiles were generally consistent across the three PMF solutions. Using a previous source apportionment study at the same receptor site, those five factors were linked to summertime biogenic emissions (odd n-alkane factor), unburned fossil fuels (light SVOC factor), road dust and/or cooking (n-alkane factor), motor vehicle emissions (PAH factor), and lubricating oil combustion (sterane factor). The “measured-total” solution was least influenced by G/P partitioning and used as reference. Two out of the five factors (odd n-alkane and PAH factors) exhibited consistent contributions for “particle-only” vs “measured-total” and “predicted-total” vs “measured-total” solutions. Factor contributions of light SVOC and n-alkane factors were more consistent for “predicted-total” vs “measured-total” than “particle-only” vs “measured-total” solutions. The remaining factor (sterane factor) underestimated the contribution by around 50% from both “particle-only” and “predicted-total” solutions. The results of this study confirm that when measured gas-phase SVOCs are not available, “predicted-total” SVOCs should be used to decrease the influence of G/P partitioning on receptorbased source apportionment.



INTRODUCTION A number of epidemiological studies indicate associations between carbonaceous components in fine particles with adverse health outcomes.1,2 To provide insights into the sources of fine carbonaceous particles, several factor analysis receptor models (e.g., chemical mass balance (CMB), positive matrix factorization (PMF))3−5 have been applied using particulate compositional data, with the estimated source contributions subsequently being employed in epidemiological studies to improve our understanding of the health effects.6−8 Particle-phase data for semivolatile organic compounds (SVOCs) have been widely used as inputs for receptor models.9,10 The resulting factor profiles are assumed to be constant over the ambient and/or source sampling periods.11 However, in addition to emission source categories (e.g., fossil fuel combustion, biomass burning), the output factors of a receptor model can also reflect influences of atmospheric processing such as photochemical reactions and gas/particle © 2014 American Chemical Society

(G/P) partitioning. Due to the variations in meteorological conditions (e.g., solar irradiance, ambient temperature), the influence of atmospheric processes on certain output factors will change during the sampling period. Therefore, the assumption of constant factor profiles will lead to biased source apportionment for long-term studies. In this work, we focused only on the influence of G/P partitioning. According to G/P partitioning theory developed by Pankow,12,13 the absorptive G/P partitioning coefficient (Kp,OM) of each SVOC can either be measured (eq 1) or calculated from theory (eq 2): Received: Revised: Accepted: Published: 9053

May 5, 2014 July 10, 2014 August 1, 2014 August 1, 2014 dx.doi.org/10.1021/es5022262 | Environ. Sci. Technol. 2014, 48, 9053−9060

Environmental Science & Technology K p,OM =

K p,OM =

Kp fOM

=

F /MOM A

RT 10 MWOMζOMpLο 6

Article

collected every sixth day using a medium volume sampler from August 2012 to July 2013. Two quartz fiber filters were installed in series to evaluate PM2.5 associated SVOCs (top QFF, tQFF) and sampling artifacts due to gas-phase adsorption (backup QFF, bQFF). PUF/XAD/PUF sandwich (PXP) was used to collect gas-phase SVOCs. Low to medium MW n-alkanes and PAHs, levoglucosan and 2-methyltetrols were quantified in both QFF and PXP samples.16,17 Steranes and hopanes were quantified for tQFF samples, and were not observed on bQFF or PXP samples. Fatty acids and methoxyphenols were also quantified for QFF samples, but their concentrations in PXP samples could not be reported due to their high concentrations in field blanks. Statistics for those species only quantified for QFF samples (not previously reported) are given in Table S1 of the Supporting Information (SI). Concentrations of bulk elemental carbon (EC) and organic carbon (OC) were also measured for QFF samples.17 Prediction of Gas-phase SVOCs. Gas-phase SVOCs were predicted as in Xie et al.14 Briefly, Kp,OM for each SVOC on each day was calculated using eq 2. Then gas-phase SVOC concentrations were calculated using eq 1 with calculated Kp,OM and measured particulate SVOC concentrations and OM. For the Kp,OM calculation, T was daily average temperature and a MW OM of 200 g mol−1 was assumed for all samples,18,19 ζOM was assumed to be unity for each SVOC in each sample, and poL values for all SVOCs were predicted using the group contribution methods (GCMs) SPARC (http://archemcalc. com/sparc/test/)20 and SIMPOL.21 The poL value for each SVOC on each sampling day was adjusted by daily average temperature:

(1)

(2)

where particulate organic matter (OM) is assumed to be responsible for the absorptive uptake. The G/P partitioning coefficient (Kp, m3 μg−1) is normalized by the weight fraction of the absorptive OM phase (f OM) in the total PM phase to obtain Kp,OM (eq 1). F and A are the mass concentrations (ng m−3) of each SVOC associated with the particle and gas phase, respectively; MOM (μg m−3) is the mass concentration of the particle-phase OM. In eq 2, R (m3 atm K−1 mol−1) is the ideal gas constant; T (K) is the ambient temperature; MW OM (g mol−1) is the mean molecular weight (MW) of the absorbing OM phase; ζOM is the mole fraction scale activity coefficient of each compound in the absorbing OM phase; p°L (atm) is the vapor pressure of each pure compound. For a specific SVOC in a single OM phase at a fixed relative humidity, G/P partitioning is driven largely by ambient temperature. Xie et al.14 predicted 970 daily gas-phase SVOC concentrations from measured particle-phase SVOC concentrations using absorptive G/P partitioning theory (eqs 1 and 2). The total SVOC data set (measured particle-phase + predicted gasphase) was analyzed using PMF. Unlike the “particle-only” based PMF results,15 the factor contributions were consistent between the full data set and temperature-stratified subdata sets solutions, suggesting the possible elimination of the influence of G/P partitioning on receptor-based source apportionment by using predicted gas-phase SVOCs. To validate the predicted SVOC concentrations, 50 samples (August 2012 − July 2013) of both particle- and gas-phase SVOCs were collected and quantified.16,17 The significant correlations (p < 0.05) between measured and predicted Kp,OM indicated that absorptive partitioning theory could reasonably represent the variation in G/P partitioning of target SVOCs (e.g., docosane, fluoranthene). However, measured Kp,OM values of target SVOCs were generally higher than the corresponding predicted values, which likely resulted from uncertainties in the prediction of p°L, ζOM and MW OM.16 In this work, three data sets were derived from the measured SVOCs in Xie et al.16 and analyzed by PMF. The SVOC concentrations in the first data set were the sum of measured particle- and gas-phase concentrations. The second data set was composed of the same SVOCs with measured particle-phase concentrations only. The third data set was also composed of the same SVOCs, while the concentrations were calculated as the sum of measured particle-phase and predicted gas-phase concentrations using the identical method as Xie et al.14 PMF results of the latter two data sets were compared to those derived from the first data set, so as to quantify the influence of G/P partitioning on particle-only based source apportionment, and to evaluate whether the prediction of gas-phase SVOCs in Xie et al.14 can help minimize this influence.

⎡ ΔH * ⎛ ⎤ 1 1⎞ vap ⎜ pLo = pLo, * exp⎢ − ⎟⎥ T ⎠⎥⎦ ⎢⎣ R ⎝ 298.15

(3)

where po,*L was the vapor pressure of each pure compound at 298.15 K, and ΔH*vap (kJ mol−1) was the enthalpy of vaporization of the liquid at 298.15 K. The po,*L and ΔH*vap values of all SVOCs were provided by Xie et al.14 To compute gas-phase concentrations using eq 1, F values of all SVOCs were obtained from existing measurements,16 and MOM was estimated by multiplying the measured OC concentration by a scaling factor of 1.53.22 The total concentration of each SVOC (S, measured particle-phase + predicted gas-phase, ng m−3) on each sampling day was then obtained by eq 4: S=F+A=

1 + K p,OMMOM K p,OMMOM

F (4)

The uncertainty associated with S calculation was estimated using the root sum of squares method,22 δS =



⎞2 ⎛ ∂S ⎞2 ⎛ ∂S ⎜ ⎟ δF + ⎜ δMOM⎟ ⎝ ∂F ⎠ ⎝ ∂MOM ⎠

(5)

where δS was the propagated uncertainty of S; δF and δMOM were the propagated uncertainties associated with the measurements of particle-phase SVOCs and MOM, respectively. Statistics for predicted total concentrations of selected SVOCs from August 2012 to July2013 are listed in Table S2 of the SI, including the mean, standard deviation and signal-tonoise ratio (S/N = mean concentration/mean uncertainty).

MATERIALS AND METHODS Ambient Measurements. Particle- (PM2.5) and gas-phase SVOCs were collected on the top of a two-story elementary school building in urban Denver. Details of the sampling set up and chemical analysis have been provided in Xie et al.16,17 Briefly, 50 pairs of particle- and gas-phase SVOC samples were 9054

dx.doi.org/10.1021/es5022262 | Environ. Sci. Technol. 2014, 48, 9053−9060

Environmental Science & Technology

Article

PMF Analysis. In this work, PMF223,24 coupled with a stationary block bootstrap technique25 was used as the source apportionment tool, which has been introduced and applied in our previous source apportionment studies.14,15,26,27 PMF resolves factor profiles and contributions from a time series of observations using a weighted least-squares fitting approach. The bootstrap technique generated replicate data sets (N = 1000) from the original data set and each was analyzed with PMF. A bootstrap solution was recorded only when each factor of that solution could be uniquely matched to a base case factor by factor profile, generating a rate of factor matching (%) between bootstrapped factors and base case factors. All resampled measurement days in recorded solutions were tracked to examine the bias and variability in contribution of each factor on each day. The selection of factor number was based on the interpretability of different PMF solutions (4−6 factors) and the factor matching rate of bootstrapped PMF solutions. Physical interpretability of PMF factors is widely used for factor number determination;28−30 a high factor matching rate (at least >50%) indicates uniqueness of base case factors and robustness of PMF solution to input data.27 Input Data Set for PMF. Three data sets were constructed for PMF analysis. The first data set (“measured-total”) consisted of measured total (gas + particle phase) SVOC concentrations, of which the PMF results were not influenced by G/P partitioning and considered as reference. The measured total SVOC concentrations were calculated by adding tQFF, bQFF, and PXP concentrations together. Concentrations of nalkanes and PAHs with MW higher than 394 (octacosane) and 202 (pyrene) g/mol, respectively, associated with bQFF and PXP were neglected during the calculation; these species in bQFF and PXP had relatively low (with more missing value) concentrations or S/N ratios (50%) of bootstrapped PMF solutions. Bulk EC and OC fractions (OC1, 2, 3, 4, and pyrolized carbon) were also included for PMF analysis. The 5 OC fractions represented the carbon measured at four distinct temperature steps (340, 500, 615, and 900 °C) with a pyrolized carbon (PC) adjustment in the first heating cycle of NIOSH 5040 thermal optical transmission (TOT) method.31,32 In this work, OC2, 3, 4 and PC concentrations of the tQFF were added together (OC2-PC) to obtain nonvolatile OC. The OC fractions other than OC1

were not detected on bQFF; the OC1 on bQFF (bOC1) was considered to be a positive sampling artifact due to gas-phase adsorption to QFF media. The concentration of gas-phase OC could not be predicted or measured in this work, so the three data sets had the same bulk carbon species for analysis (EC, OC1, OC2-PC, and bOC1). The statistics of each species included for PMF analysis are listed in Table S2 of the SI.



RESULTS AND DISCUSSION Input Data. As shown in SI Table S2, n-alkanes and PAHs with MW higher than 394 and 202, respectively, and steranes exhibited small or no differences in average concentrations across the three different data sets. Average concentrations of light n-alkanes (MW ≤ 394) in the “measured-total” data set were higher than those in the “particle-only” data set by 10− 60%, suggesting that the particle-phase mass fractions were greater than the gas-phase mass fractions for these species. The average concentrations of the same species in the “predictedtotal” data set were several times higher (except octacosane) than those in the “measured-total” data set, thus the current prediction method led to overestimation of the gas-phase concentrations of light n-alkanes. Light PAHs (fluoranthene and pyrene) had much higher (>2 times) average concentrations in the “measured-total” data set compared to those in the “particle-only” data set, and these species were primarily distributed in gas phase. Similarly to light n-alkanes, gas-phase light PAHs were also overestimated in the “predicted-total” data set. PMF Results for Different Data Sets. General information for PMF simulations of all data sets are given in Table 1.

Table 1. Simulation Statistics for All Data Sets data sets no. of species no. of observations no. of factor no. of bootstrap replicate data sets no. of data sets for which factors were uniquely matched

measuredtotal

particleonly

predictedtotal

26 50 5 1000 551

26 50 5 1000 574

26 50 5 1000 687

Five physically interpretable factors were identified for all the three data sets with factor matching rate all higher than 50% (55.1−68.7%). The 4- and 6-factor solutions were not interpretable with factor matching rates lower than 50%. PMF factor profiles of each data set were normalized by

Fkj* =

Fkj p ∑k = 1 Fkj

(6)

where F kj* is the relative weighting of species j in factor k to all other factors. The normalized factor profiles of the three data sets solutions are mostly similar and shown in Figure 1. Median factor contribution time series (median ± standard deviation) derived from bootstrapped PMF solutions are shown in SI Figures S1−S3 for the three data sets solutions. Here the factor contribution was reconstructed by adding the contribution of each factor to all bulk carbon fractions. The average contributions of each factor to bulk EC and OC fractions are presented in Table 2. The coefficient of variation (CV) was calculated by dividing the standard deviation of factor contribution by median factor contribution of each factor on each day. The median CVs of all factors are listed in Table 2, 9055

dx.doi.org/10.1021/es5022262 | Environ. Sci. Technol. 2014, 48, 9053−9060

Environmental Science & Technology

Article

Figure 1. Median normalized factor profiles derived from PMF bootstrap solutions for “measured-total”, “particle-only”, and “predicted-total” data sets. The whiskers represent one standard deviation.

which reflect the variability in factor contribution due to random sampling error. Comparison to Previous Source Apportionment. In a previous source apportionment study at the same receptor site using a daily sampled 32-month (2003−2005) “particle-only” SVOC data set,15 eight factors were identified and five of them could be uniquely matched with those factors from the current work based on factor profiles. These five factors could be linked with summertime biogenic emissions (odd n-alkane factor), unburned fossil fuel (light SVOC factor), road dust and/or

cooking (n-alkane factor), motor vehicle emissions (PAH factor), and lubricating oil combustion (sterane factor). The remaining three factors obtained in the previous study were inorganic ion (dominated by sulfate and nitrate), winter/ methoxyphenol (with relatively low average contribution, ∼ 2% of the total), and medium alkane/alkanoic acid (with relatively low average contribution, ∼1% of the total) factors. Factor Contribution of the Measured-total Data Set Solution. In this work, the PMF results of the “measured-total” data set were considered as least biased, without influences 9056

dx.doi.org/10.1021/es5022262 | Environ. Sci. Technol. 2014, 48, 9053−9060

Environmental Science & Technology

Article

Table 2. Average Source Apportionment Results for Carbonaceous Components, μg m−3 (%), and Median Coefficient of Variation (CV) of Factor Contributions

a

factor

EC

odd n-alkane light SVOC n-alkane PAH sterane subtotal

0.023 (6.17) 0.043 (11.6) 0.032 (8.62) 0.15 (40.6) 0.12 (33.0) 0.38 (100)

odd n-alkane light SVOC n-alkane PAH sterane subtotal

0.034 (9.48) 0.080 (22.3) 0.016 (4.40) 0.14 (38.7) 0.090 (25.1) 0.36 (100)

odd n-alkane light SVOC n-alkane PAH sterane subtotal observed

0.040 (11.3) 0.049 (13.6) 0.034 (9.56) 0.15 (41.5) 0.086 (24.0) 0.36 (100) 0.40

OC2-PCa

OC1 Measured-Total 0.19 (12.6) 0.35 (23.1) 0.082 (5.53) 0.37 (24.8) 0.51 (34.0) 1.49 (100) Particle-Only 0.21 (13.7) 0.59 (39.4) 0.040 (2.64) 0.37 (24.2) 0.30 (20.1) 1.51 (100) Predicted-Total 0.23 (15.9) 0.40 (27.6) 0.11 (7.42) 0.43 (29.8) 0.28 (19.2) 1.45 (100) 1.58

bOC1b

CVc

0.064 (15.4) 0.089 (21.3) 0.012 (2.84) 0.10 (24.5) 0.15 (36.0) 0.42 (100)

0.52 0.55 0.60 0.29 0.41

0.31 (19.7) 0.57 (36.4) 0.094 (6.04) 0.54 (34.6) 0.051 (3.27) 1.56 (100)

0.077 (18.2) 0.15 (34.3) 0.001 (0.33) 0.085 (20.0) 0.12 (27.2) 0.42 (100)

0.37 0.20 0.80 0.23 0.35

0.36 (23.6) 0.38 (25.2) 0.13 (8.52) 0.60 (39.9) 0.041 (2.72) 1.51 (100) 1.98

0.08 (19.3) 0.12 (28.6) 0.015 (3.45) 0.11 (25.6) 0.099 (23.1) 0.43 (100) 0.47

0.37 0.24 0.45 0.21 0.32

0.30 0.39 0.12 0.50 0.24 1.54

(19.4) (25.1) (7.69) (32.1) (15.8) (100)

Sum of OC2, OC3, OC4, and PC. bOC1 on backup QFF. cCV = standard deviation/median factor contribution.

Figure 2. Linear regressions of factor contributions for “particle-only” vs “measured-total”, and “predicted-total” vs “measured-total” solutions.

9057

dx.doi.org/10.1021/es5022262 | Environ. Sci. Technol. 2014, 48, 9053−9060

Environmental Science & Technology

Article

underestimated those of the n-alkane and sterane factors by 38% and 45%, respectively. As such, the regression slopes for the above three factors deviated from unity (Figure 2b,c,e). The lower r value (0.66) for the light SVOC factor in Figure 2b, as compared to Figure 2a,c,d, was likely caused by the fact that PMF analysis of the “particle-only” data set was influenced by G/P partitioning, but we could not rule out other possibilities like the influence of photochemical processes. The lower r value (0.75) in Figure 2e was likely a byproduct of the low r value in Figure 2b, given the total contribution of all factors were strongly correlated (r = 0.95) between “particle-only” and “measured-total” solutions. If one factor with substantial contribution exhibited low correlation between the two solutions, then at least one other factor should have low correlation. Predicted-Total versus Measured-Total. In Figure 1, the light SVOC factor in the “predicted-total” solution has more light n-alkanes and PAHs than the “measured-total” solution. However, the average contributions of the light SVOC factor to carbonaceous species were comparable between the two solutions (Table 2). In Figure 2, the regression slopes of four factors (not including the sterane factor) are close to unity (1.03−1.15), and the correlations are all significant (r = 0.65 − 0.99, p < 0.05). The contributions of the light SVOC and nalkane factors of the “measured-total” solution were reproduced better by using “predicted-total” data set than using “particleonly” data set. However, the light SVOC factor contribution exhibited lower correlation (r = 0.65) between “predicted-total” and “measured-total” solutions than other factors. This was likely caused by the overestimation of gas-phase light n-alkanes and PAHs using absorptive partitioning theory, and the inability to account for photochemical processes. Similarly to the “particle-only” solution, the “predicted-total” solution underestimated the factor contribution of the sterane factor by 51% on average, which was due to the overestimation (10−24%) in contribution of the other four factors.

from G/P partitioning (“particle-only” data set) or overestimation of gas-phase SVOCs (“predicted-total” data set). The PAH factor had the highest average contribution (40.6%) to EC, followed by the sterane factor (33.0%), consistent with the fact that EC was mainly from motor vehicle emissions. More volatile OC fractions (OC1 and bOC1) were largely contributed by the sterane factor (34.0% and 36.0%), followed by the PAH (24.8% and 24.5%) and light SVOC (23.1% and 21.3%) factors; while the nonvolatile OC fractions (OC2-PC) were apportioned largely to the PAH factor (32.1%), followed by the light SVOC (25.1%) and odd n-alkane (19.4%) factors. The average contribution of the n-alkane factor to total bulk carbon was lower than the other four factors (Table 2). The average sum of factor contributions for each carbonaceous species accounted for 78−98% of the observed concentration, which was also observed for the other two data sets solutions. Levoglucosan and 2-methyltetrols are important organic tracers for biomass burning33 and isoprene derived secondary organic aerosol (SOA),34,35 respectively. Their data were available from Xie et al.17 but not included for PMF analysis due to scarcity of measured 2-methyltetrol values in cold periods, and negligible contributions of biomass burning to carbonaceous aerosol at the same receptor site.15 Furthermore, using the same organic species for PMF analysis in this work as had been used in our previous work15 helped to match the factors identified in both studies. As such, the factors related to biomass burning and SOA were not resolved, and their contributions would be apportioned to other factors. To identify the factors containing contributions from biomass burning and isoprene derived SOA, factor contributions were correlated with concentrations of measured total levoglucosan and 2-methylterols. The correlation coefficients are listed in SI Table S3. Concentrations of measured total levoglucosan were strongly correlated with the contributions of PAH (r = 0.79, p < 0.05) and n-alkane (r = 0.63, p < 0.05) factors, indicating that the contribution of biomass burning was apportioned to these two factors. Concentrations of measured total 2-methyltetrols were reasonably correlated with the contributions of light SVOC (r = 0.52, p < 0.05) and odd n-alkane (r = 0.40, p < 0.05) factors, thus these two factors might also reflect the influence of photochemical reactions. Similar results were also observed for the other two data sets solutions. Comparisons across Different Solutions. To understand the influence of G/P partitioning on receptor-based source apportionment using “particle-only” SVOC data, and to determine to what extent the prediction of gas-phase SVOCs can minimize this influence, PMF solutions of the “particleonly” and “predicted-total” data sets were compared to that of the “measured-total” data set. In addition to the comparison in Figure 1 and Table 2 for factor profiles and average factor contributions, factor contribution time series were also compared using linear regression (with intercepts set as 0) and shown in Figure 2. Particle-Only versus Measured-Total. Compared to the factor profile of the “measured-total” solution, the “particleonly” solution apportioned less light PAHs (fluoranthene and pyrene) and more bulk carbon to the light SVOC factor; while the sterane factor had much higher loadings of light PAHs (Figure 1). In Figure 2, the correlations (r = 0.66−0.99) of factor contribution time series between the “particle-only” and “measured-total” solutions are all significant (p < 0.05). However, the “particle-only” solution overestimated the contribution of the light SVOC factor by 61% (average), and



IMPLICATIONS AND LIMITATIONS In this work, we confirmed that the receptor-based source apportionment using only particle-phase SVOC data could be influenced by G/P partitioning. This effect was reduced by predicting gas-phase SVOC concentrations based on absorptive partitioning theory and adding measured particle-phase to predicted gas-phase concentrations for PMF analysis. The decrease in influence from G/P partitioning will highly depend on the accuracy of gas-phase SVOC prediction. In this work, gas-phase SVOCs were overestimated by up to 10 times, which was caused by the underestimation of Kp,OM. According to Xie et al.,16 the underestimation of Kp,OM could be due to the uncertainty in the prediction of p°L, ζOM, and MW OM (eq 2). Although the current method could not predict gas-phase SVOCs with high accuracy, the source apportionment results derived from the “predicted-total” data set were in better agreement with those from the “measured-total” data set, as compared to those from the “particle-only” data set. Therefore, receptor-based source apportionment using “predicted-total” SVOC data should be performed routinely in future PM source studies when measured gas-phase SVOC data are not available. However, the overestimation of gas-phase SVOC might introduce some uncertainties for PMF analysis. If only high MW (less- or nonvolatile) organic markers were used for source apportionment, then the prediction of their gasphase concentrations was unnecessary. However, the light 9058

dx.doi.org/10.1021/es5022262 | Environ. Sci. Technol. 2014, 48, 9053−9060

Environmental Science & Technology

Article

(5) Paatero, P.; Tapper, U. Positive matrix factorization: A nonnegative factor model with optimal utilization of error estimates of data values. Environmetrics 1994, 5 (2), 111−126. (6) Ito, K.; Christensen, W. F.; Eatough, D. J.; Henry, R. C.; Kim, E.; Laden, F.; Lall, R.; Larson, T. V.; Neas, L.; Hopke, P. K.; Thurston, G. D. PM source apportionment and health effects: 2. An investigation of intermethod variability in associations between source-apportioned fine particle mass and daily mortality in Washington, DC. J. Exposure Sci. Environ. Epidemiol. 2006, 16 (4), 300−310. (7) Lall, R.; Ito, K.; Thurston, G. D. Distributed lag analyses of daily hospital admissions and source-apportioned fine particle air pollution. Environ. Health Perspect. 2011, 119 (4), 455−460. (8) Sarnat, J. A.; Marmur, A.; Klein, M.; Kim, E.; Russell, A. G.; Sarnat, S. E.; Mulholland, J. A.; Hopke, P. K.; Tolbert, P. E. Fine particle sources and cardiorespiratory morbidity: An application of chemical mass balance and factor analytical source-apportionment methods. Environ. Health Perspect. 2008, 116 (4), 459−466. (9) Jaeckels, J. M.; Bae, M. S.; Schauer, J. J. Positive matrix factorization (PMF) analysis of molecular marker measurements to quantify the sources of organic aerosols. Environ. Sci. Technol. 2007, 41 (16), 5763−5769. (10) Shrivastava, M. K.; Subramanian, R.; Rogge, W. F.; Robinson, A. L. Sources of organic aerosol: Positive matrix factorization of molecular marker data and comparison of results from different source apportionment models. Atmos. Environ. 2007, 41 (40), 9353− 9369. (11) Chen, L. W. A.; Watson, J. G.; Chow, J. C.; DuBois, D. W.; Herschberger, L. PM2.5 source apportionment: Reconciling receptor models for U.S. nonurban and urban long-term networks. J. Air Waste Manage 2011, 61 (11), 1204−1217. (12) Pankow, J. F. An absorption model of gas/particle partitioning of organic compounds in the atmosphere. Atmos. Environ. 1994, 28 (2), 185−188. (13) Pankow, J. F. An absorption model of the gas/aerosol partitioning involved in the formation of secondary organic aerosol. Atmos. Environ. 1994, 28 (2), 189−193. (14) Xie, M.; Barsanti, K. C.; Hannigan, M. P.; Dutton, S. J.; Vedal, S. Positive matrix factorization of PM2.5Eliminating the effects of gas/ particle partitioning of semivolatile organic compounds. Atmos. Chem. Phys. 2013, 13 (15), 7381−7393. (15) Xie, M.; Piedrahita, R.; Dutton, S. J.; Milford, J. B.; Hemann, J. G.; Peel, J. L.; Miller, S. L.; Kim, S.-Y.; Vedal, S.; Sheppard, L.; Hannigan, M. P. Positive matrix factorization of a 32-month series of daily PM2.5 speciation data with incorporation of temperature stratification. Atmos. Environ. 2013, 65 (0), 11−20. (16) Xie, M.; Hannigan, M. P.; Barsanti, K. C. Gas/particle partitioning of n-alkanes, PAHs and oxygenated PAHs in urban Denver. Atmos. Environ. 2014, 95 (0), 355−362. (17) Xie, M.; Hannigan, M. P.; Barsanti, K. C. Gas/particle partitioning of 2-methyltetrols and levoglucosan at an urban site in Denver. Environ. Sci. Technol. 2014, 48 (5), 2835−2842. (18) Barsanti, K. C.; Pankow, J. F. Thermodynamics of the formation of atmospheric organic particulate matter by accretion reactionsPart 1: Aldehydes and ketones. Atmos. Environ. 2004, 38 (26), 4371−4382. (19) Williams, B. J.; Goldstein, A. H.; Kreisberg, N. M.; Hering, S. V. In situ measurements of gas/particle-phase transitions for atmospheric semivolatile organic compounds. Proc. Natl. Acad. Sci. U. S. A. 2010, 107 (15), 6676−6681. (20) Hilal, S. H.; Karickhoff, S. W.; Carreira, L. A. A rigorous test for SPARC’ chemical reactivity models: Estimation of more than 4300 ionization pKas. Quant. Struct-Act. Rel. 1995, 14 (4), 348−355. (21) Pankow, J. F.; Asher, W. E. SIMPOL.1: A simple group contribution method for predicting vapor pressures and enthalpies of vaporization of multifunctional organic compounds. Atmos. Chem. Phys. 2008, 8 (10), 2773−2796. (22) Dutton, S. J.; Schauer, J. J.; Vedal, S.; Hannigan, M. P. PM2.5 characterization for time series studies: Pointwise uncertainty estimation and bulk speciation methods applied in Denver. Atmos. Environ. 2009, 43 (5), 1136−1146.

SVOC factor could not be resolved, of which the contribution would be apportioned to other factors. In this work, the contribution of light SVOC factor accounted for a considerable mass fraction of bulk carbon species, including both more volatile (OC1 and bOC1) and nonvolatile carbon (EC and OC2-PC). As such, we recommend the inclusion of light organic markers (e.g., docosane and fluoranthene) for source apportionment. For PMF analysis, the small number of observations might lead to some uncertainty in source apportionment results.36 However, solutions for all data sets were interpretable and agreed well with our previous work. Due to the lack of SOA tracer data, no SOA related factor could be identified, and the contribution of SOA was likely apportioned to the light SVOC and odd n-alkane factors. In the future, SOA tracer data need to be included for the calculation of a SOA contribution. In the current work, “measured-total” SVOC data were used to apportion filter-based bulk carbon into sources/atmospheric processes. The bOC1 observed on bQFF was part of gas-phase OC and used to correct the OC measured on tQFF.17 If total gas-phase OC could be measured or predicted, then the receptor model could be extended to apportion total OC in the atmosphere by including more volatile organic compounds.



ASSOCIATED CONTENT

S Supporting Information *

Additional details on statistics for SVOC data, and factor contributions time series. This material is available free of charge via the Internet at http://pubs.acs.org.



AUTHOR INFORMATION

Corresponding Author

*Phone: +1720 708 0812; fax: +1-303-492-3498; e-mail: [email protected]. Present Address §

Civil and Environmental Engineering, Virginia Tech, Blacksburg, Virginia 24061, United States. Notes

The authors declare no competing financial interest.



ACKNOWLEDGMENTS We acknowledge the contributions of Joshua Hemann for his help in developing source apportionment method, and Palmer Elementary School faculty and staff for their assistance with the sampling site.



REFERENCES

(1) Kim, S.-Y.; Peel, J. L.; Hannigan, M. P.; Dutton, S. J.; Sheppard, L.; Clark, M. L.; Vedal, S. The temporal lag structure of short-term associations of fine particulate matter chemical constituents and cardiovascular and respiratory hospitalizations. Environ. Health Perspect. 2012, 120 (8), DOI: 10.1289/ehp.1104721. (2) Ostro, B.; Feng, W.-Y.; Broadwin, R.; Green, S.; Lipsett, M. The effects of components of fine particulate air pollution on mortality in California: Results from CALFINE. Environ. Health Perspect. 2007, 115 (1), 13−19. (3) Watson, J. G.; Robinson, N. F.; Chow, J. C.; Henry, R. C.; Kim, B. M.; Pace, T. G.; Meyer, E. L.; Nguyen, Q. The USEPA/DRI chemical mass balance receptor model, CMB 7.0. Environ. Software 1990, 5 (1), 38−49. (4) Thurston, G. D.; Spengler, J. D. A quantitative assessment of source contributions to inhalable particulate matter pollution in metropolitan Boston. Atmos. Environ. 1985, 19 (1), 9−25. 9059

dx.doi.org/10.1021/es5022262 | Environ. Sci. Technol. 2014, 48, 9053−9060

Environmental Science & Technology

Article

(23) Paatero, P. User’s Guide for Positive Matrix Factorization Program PMF2 and PMF3, Part 1: Tutorial; University of Helsinki: Helsinki, Finland, 1998. (24) Paatero, P. User’s Guide for Positive Matrix Factorization Program PMF2 and PMF3, Part 2: Reference; University of Helsinki: Helsinki, Finland, 1998. (25) Hemann, J. G.; Brinkman, G. L.; Dutton, S. J.; Hannigan, M. P.; Milford, J. B.; Miller, S. L. Assessing positive matrix factorization model fit: A new method to estimate uncertainty and bias in factor contributions at the measurement time scale. Atmos. Chem. Phys. 2009, 9 (2), 497−513. (26) Xie, M.; Coons, T. L.; Hemann, J. G.; Dutton, S. J.; Milford, J. B.; Peel, J. L.; Miller, S. L.; Kim, S.-Y.; Vedal, S.; Sheppard, L.; Hannigan, M. P. Intra-urban spatial variability and uncertainty assessment of PM2.5 sources based on carbonaceous species. Atmos. Environ. 2012, 60 (0), 305−315. (27) Xie, M.; Hannigan, M. P.; Dutton, S. J.; Milford, J. B.; Hemann, J. G.; Miller, S. L.; Schauer, J. J.; Peel, J. L.; Vedal, S. Positive matrix factorization of PM2.5: Comparison and implications of using different speciation data sets. Environ. Sci. Technol. 2012, 46 (21), 11962− 11970. (28) Kim, E.; Hopke, P. K.; Pinto, J. P.; Wilson, W. E. Spatial variability of fine particle mass, components, and source contributions during the regional air pollution study in St. Louis. Environ. Sci. Technol. 2005, 39 (11), 4172−4179. (29) Hwang, I.; Hopke, P. K.; Pinto, J. P. Source apportionment and spatial distributions of coarse particles during the regional air pollution study. Environ. Sci. Technol. 2008, 42 (10), 3524−3530. (30) Mooibroek, D.; Schaap, M.; Weijers, E. P.; Hoogerbrugge, R. Source apportionment and spatial variability of PM(2.5) using measurements at five sites in the Netherlands. Atmos. Environ. 2011, 45 (25), 4180−4191. (31) NIOSH. Method 5040, Issue 3: Diesel Particulate Matter (as Elemental Carbon). In NIOSH Manual of Analytical Methods (NMAM), 4th ed.; National Institute of Occupational Safety and Health: Cincinnati, OH, 2003. (32) Schauer, J. J.; Mader, B. T.; Deminter, J. T.; Heidemann, G.; Bae, M. S.; Seinfeld, J. H.; Flagan, R. C.; Cary, R. A.; Smith, D.; Huebert, B. J.; Bertram, T.; Howell, S.; Kline, J. T.; Quinn, P.; Bates, T.; Turpin, B.; Lim, H. J.; Yu, J. Z.; Yang, H.; Keywood, M. D. ACEAsia intercomparison of a thermal-optical method for the determination of particle-phase organic and elemental carbon. Environ. Sci. Technol. 2003, 37 (5), 993−1001. (33) 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. (34) Claeys, M.; Graham, B.; Vas, G.; Wang, W.; Vermeylen, R.; Pashynska, V.; Cafmeyer, J.; Guyon, P.; Andreae, M. O.; Artaxo, P.; Maenhaut, W. Formation of secondary organic aerosols through photooxidation of isoprene. Science 2004, 303 (5661), 1173−1176. (35) Surratt, J. D.; Murphy, S. M.; Kroll, J. H.; Ng, N. L.; Hildebrandt, L.; Sorooshian, A.; Szmigielski, R.; Vermeylen, R.; Maenhaut, W.; Claeys, M.; Flagan, R. C.; Seinfeld, J. H. Chemical composition of secondary organic aerosol formed from the photooxidation of isoprene. J. Phys. Chem. A 2006, 110 (31), 9665−9690. (36) Zhang, Y. X.; Sheesley, R. J.; Bae, M. S.; Schauer, J. J. Sensitivity of a molecular marker based positive matrix factorization model to the number of receptor observations. Atmos. Environ. 2009, 43 (32), 4951−4958.

9060

dx.doi.org/10.1021/es5022262 | Environ. Sci. Technol. 2014, 48, 9053−9060