Source Apportionment of Polychlorinated Biphenyls in Chicago Air

Mar 11, 2013 - While the overall atmospheric polychlorinated biphenyl (PCB) levels in many urban areas are declining, it is not clear whether this dec...
1 downloads 14 Views 1MB Size
Article pubs.acs.org/est

Source Apportionment of Polychlorinated Biphenyls in Chicago Air from 1996 to 2007 Lisa A. Rodenburg*,† and Qingyu Meng‡ †

Department of Environmental Science, Rutgers, the State University of New Jersey, New Brunswick, New Jersey 08901, United States ‡ School of Public Health, University of Medicine and Dentistry of New Jersey, Piscataway, New Jersey 08854, United States S Supporting Information *

ABSTRACT: While the overall atmospheric polychlorinated biphenyl (PCB) levels in many urban areas are declining, it is not clear whether this decline is due to control strategies or merely due to natural attenuation. To investigate this issue, Positive Matrix Factorization (PMF) was used to identify the dominant sources of gas-phase PCBs in the atmosphere of Chicago, IL using a data set collected from 1996 to 2007 by the Integrated Atmospheric Deposition Network (IADN). Both the older PMF2 software and the newer EPA-sponsored PMF 3.0 software were employed. Both models resolved 5 factors, but they yielded somewhat different results in terms of the congener patterns of the factors and their temporal variation. The PMF2 software resolved factors that better resembled the original Aroclor formulations. While it is possible to apply an exponential decay model to this data set and derive statistically significant rate constants that indicate that ΣPCBs and some of the resolved factors are declining in Chicago air, examining plots of the 365-day moving average concentrations shows that they do not decrease in a fashion consistent with exponential decay. Instead, they display periods of decline as well as periods of increase. Thus an exponential decay model is not appropriate, and long-term time trends identified from this 12-year data set cannot be used to predict the future trends in PCB concentrations in the air of Chicago. Two of the five resolved factors resemble low MW Aroclors, and declined from 1996 to 2007. The other three factors, which represent the majority of the mass in the data set, are either not declining or actually increasing over time. Thus past efforts to eliminate PCBs from the Great Lakes ecosystem have been only marginally effective, if at all. Additional effort is needed to identify and eliminate atmospheric PCB sources in Chicago.



INTRODUCTION Despite the ban on polychlorinated biphenyl (PCB) use and manufacture in the late 1970s, PCB levels are still elevated in urban areas due to sources such as joint sealants, caulks, waste incineration, storage and disposal facilities, Superfund sites, and accidental releases.1−5 Atmospheric deposition can be an important source of PCBs to adjacent water bodies.6−10 For example, atmospheric deposition from the Chicago region is considered to be one of the largest sources of PCBs to Lake Michigan.7,8,11 In response, the Great Lakes Binational Toxics Strategy (GLBTS) was instituted in 1997. The GLBTS called for a 90% reduction, management, and proper disposal of electrical equipment containing PCBs by the year 2006.12 The Integrated Atmospheric Deposition Network (IADN) was established in the early 1990s to characterize how PCBs and other contaminants originating in urban districts impact the nearby Great Lakes. IADN has amassed an extensive data set with over 15 years of PCB measurements allowing the determination of long-term trends. Generally, PCB concentrations measured in both the precipitation and gas phases in Chicago have decreased over the past 10 years,13 but it is not clear whether this decline is a result of actions such as GLBTS or is merely due to natural attenuation. A better understanding © 2013 American Chemical Society

of the sources of atmospheric PCBs would help to determine the cause of the decline. Positive Matrix Factorization has been used successfully in many source apportionment studies, including studies of PCBs in sediment14−16 and water.17,18 Recently, PMF was used for source identification of atmospheric PCBs in the Camden/ Philadelphia area.19,20 These earlier studies used the PMF2 software developed by Paatero and Tapper.21 In part because PMF2 requires purchase of a software license, a new version of PMF developed by the U.S. EPA and freely available on their Web site (http://www.epa.gov/heasd/products/pmf/pmf. html), PMF 3.0, has become a popular choice for factor analysis.22 Both programs conduct a factorization of the data, but each relies on a different algorithm to find the most stable solution. In the present study, both PMF programs were used to identify the types of sources that contribute to the PCB concentrations measured in Chicago. In addition, time trends in Received: Revised: Accepted: Published: 3774

December 10, 2012 March 5, 2013 March 10, 2013 March 11, 2013 dx.doi.org/10.1021/es305024p | Environ. Sci. Technol. 2013, 47, 3774−3780

Environmental Science & Technology

Article

the factors identified through PMF were examined in order to determine whether PCB control strategies are being effective.



EXPERIMENTAL SECTION Data used in the PMF model were provided by Indiana University (IU) for the time period ranging from January 1996 through December 2007. Samples were collected as part of the Integrated Atmospheric Deposition Network (IADN), which was designed to characterize the loadings of persistent organic pollutants originating in the atmosphere into the urban and rural regions of the Great Lakes.23 Sample Collection and Analysis. The IADN monitoring equipment for the Chicago satellite location is set on the roof of the Farr Hall building situated on the Illinois Institute of Technology (IIT) campus. Samples were generally collected at a 12-day frequency for 24 h. The data set consisted of gas-phase concentrations of 62 PCB congeners or coeluting congener groups. Particle-phase PCB concentrations could not be included in the data set because collection of particle phase data ceased in 1996. A summary of sampling procedures is given in the Supporting Information (SI). Full details regarding the collection and analysis of air samples can be found in the quality assurance and quality control documents on the IADN Web site.24 Positive Matrix Factorization Model. Source apportionment was conducted with the PMF model. There are two common algorithms to the PMF model, i.e. the Paatero PMF2 algorithm21 and the EPA PMF 3.0 (http://www.epa.gov/ heasd/products/pmf/pmf.html) which is based on the multilinear engine algorithm. In this work, both algorithms were applied to conduct source apportionment for PCBs. Although the similarities and differences between the two algorithms have been reported in previous studies for PM2.5 source apportionment,25 it is still unknown how comparable the results are across the two algorithms to extract PCB sources for air pollution studies. The PMF model was run using the congener-specific PCB data as reported by the IADN. PCBs 12, 13, and 206 were not included in the final PMF model because they were not well modeled in preliminary PMF runs (see SI for details). Thus the final data matrix consisted of 59 peaks representing 74 PCB congeners in 329 samples, of which 593 data points (3.0%) were below detection limit. Additional details on the PMF modeling are provided in the SI.



Figure 1. Concentration versus time for ΣPCBs in Chicago. Solid line represents the 365-day moving average, which was calculated by averaging all of the data points collected in the preceding 365-day period. Upper panel is the 365-day moving average of average molecular weight (g/mol).

(Venier et al.29 analyzed more recent data showing that concentrations of some PCB congeners began to decrease again around 2007.) This high/low/high concentration pattern can be seen for many of the individual congeners with four or more chlorines. It is difficult to discern between the natural variation in concentration that is primarily driven by temperature, and the long-term time trends. The variation due to temperature is often removed by applying the Clausius−Clapeyron equation, but it is also possible to remove a great deal of variability (due to temperature as well as other seasonal effects) by plotting the 365-day moving average concentration (solid line in Figure 1). When this is done, it is easier to see that ΣPCB concentrations were increasing from 1996 to late 1998, and then went into a long period of decline until about early 2005, when they began to increase again. The 365-day moving average certainly does not exhibit an exponential decay over this whole period, although the period from 1998 to 2005 may be fairly well described as a period of exponential decay. This observation calls into question the common practice13,29−31 of fitting the data to a multiparameter linear regression of natural log of concentration (ln Cgas) versus time (t in Julian days since collection of the first sample) and inverse temperature (1/T in Kelvin) in order to characterize any long-term concentration trends:

RESULTS AND DISCUSSION

Before conducting the PMF analysis, it is important to thoroughly investigate the raw data. Hites et al.26−29 have published several analyses of these data, and have generally concluded that atmospheric PCB concentrations throughout the Great Lakes area are declining, with half-lives ranging from 6 to 13 years.13,30,31 Most recently, Venier et al.29 used several different approaches to model the long-term trends in concentrations of PCBs 18 and 52 in the gas phase in Chicago and found periods of increase as well as periods of decline. Similarly, an examination of the ΣPCB concentrations versus time for Chicago (Figure 1) reveals that PCB concentrations may have declined overall from 1996 to 2007, but this decline does not follow a pseudo-first-order type exponential decay. In fact, it appears that PCB concentrations in Chicago may have been increasing from 1996 to about 1999, then declined from 1999 to about 2004, when they began to increase again.

⎛1⎞ ln Cgas = a0 + a1⎜ ⎟ + a 2t ⎝T ⎠

(1)

where a0, a1, and a2 represent the coefficients for the intercept, temperature, and time parameters, respectively. Although application of this type of equation may yield statistically significant “rate constants”, the recent work of Venier et al.29 demonstrates that there is no reason to believe that the PCBs will continue to exhibit the implied long-term declines. In fact, none of the individual congeners display a well-behaved exponential decline, and some appear to be increasing. Application of eq 1 results in statistically significant values of a2 for 56 out of 62 congeners in the data set, but only 29 of these are negative, i.e. signaling a decline in concentration over time. The other 25 display a significant increase from 1996 to 2007. The concentrations of one congener, PCB 12, are not 3775

dx.doi.org/10.1021/es305024p | Environ. Sci. Technol. 2013, 47, 3774−3780

Environmental Science & Technology

Article

Table 1. Coefficients of Determination (R2) between the Congener Patterns of the PMF2 and PMF 3.0 Factors

significantly dependent on time or on temperature. In general, the low molecular weight congeners are decreasing and the higher molecular weight congeners are increasing (see SI Figure S-1), suggesting that the average molecular weight is increasing. In fact, examination of the 365-day moving average of average molecular weight shows that it has been increasing steadily since at least 2002, and was relatively flat before that (Figure 1). Examination of the raw data therefore indicates that the atmospheric concentrations of PCBs in Chicago are not necessarily declining, and that regardless of whether they are increasing or decreasing in concentration, their behavior cannot be interpreted as a simple exponential decay (or growth). PMF Analysis. The fact that both the PMF2 and PMF 3.0 programs indicated that the correct number of factors was five gives a high degree of confidence in the choice of the number of factors. In the case of PMF2, the five-factor solution was chosen because it was stable: nine runs of the PMF model starting with different seed values gave an average RSD of all points in the G matrix of 0.5%. In contrast, when six factors were requested, the RSD jumped to 36% (see SI Table S-1 of). The five factors (designated 2-A through 2-E) were also interpretable (see below), and they gave a good fit to the data. The coefficient of determination (R2) between the measured and modeled concentrations was good (see SI Table S-2). The congeners that were not well modeled were PCBs 100, 81, and 77. None of these congeners were crucial to the interpretation of the results, and excluding them did not significantly alter the number or composition of the resolved factors. However, it should be noted that congeners 77 and 81 are dioxin-like PCB congeners that are thought to have higher toxicity than the other congeners.32 The PMF 3.0 software includes a bootstrapping routine that makes it easier to identify the correct number of factors. We used the default routine in the PMF 3.0 software, which is to conduct a bootstrapping analysis of the base run with the lowest Q value. When this was done, the five factors mapped onto the correct base factor in at least 97 out of 100 bootstrapping runs (see SI Tables S-3 and S-4). In contrast, when 6 factors were selected, three of them were “smeared” across other factors in the bootstrapping runs such that in some cases the factors correctly mapped to their base factors in as few as 47 of the 100 bootstrap runs. The coefficient of determination (R2) between the measured and modeled concentrations was good (see SI Table S-2). The congener patterns of the resolved factors from PMF2 and PMF 3.0 were compared with one another by plotting their congener patterns against each other and looking for the matching factor with the highest coefficient of determination (R2) (Table 1). The congener patterns of the resolved factors are shown in Figure 2. To avoid confusion, each PMF2 factor was assigned a matching PMF 3.0 factor and the PMF 3.0 factors were then numbered as 3-A through 3-E. (Note that this designation should be approached with caution, however, since differences in the G-matrix for each factor pair are evident; see below). For the A and E factor pairs, the correlation coefficient was greater than 0.9, making identification unequivocal. For others, the correlation was less strong (Table 1). Factors 2-B and 3-B were only strongly correlated when PCB 33 was removed, which caused the correlation coefficient to increase from 0.55 to 0.89. It seems unrealistic that PCB 33 alone should comprise 25% of the PCBs in factor 3-B. PCB 33 was an outlier in most of the plots of Aroclor congener pattern versus the PMF 3.0 resolved congener patterns (see below). Thus

3-A with PCB 33 2-A 0.94 2-B 0.73 2-C 0.15 2-D 0.01 2-E 0.00 without PCB 33 2-A 0.97 2-B 0.72 2-C 0.15 2-D 0.00 2-E 0.00

3-B

3-C

3-D

3-E

0.56 0.55 0.19 0.00 0.01

0.21 0.69 0.78 0.08 0.01

0.00 0.01 0.44 0.97 0.12

0.01 0.00 0.23 0.68 0.80

0.60 0.89 0.46 0.00 0.00

0.32 0.80 0.81 0.08 0.01

0.00 0.02 0.45 0.97 0.12

0.01 0.00 0.23 0.68 0.80

Figure 2. Congener patterns of the resolved factors. Numbers in parentheses refer to the percent of the mass in each data set that this factor represents for the PMF2 and PMF 3.0 models, respectively.

PMF 3.0 seemed to have difficulty in dealing with PCB 33, which is surprising since PCB 33 is a prominent congener that was above detection limit in all samples and comprised about 4.7% of the total PCB mass in the data matrix. The only unusual aspect of PCB 33 is that its temperature dependence is somewhat weaker than the other trichlorobiphenyls (see SI Figure S-1). Something similar seems to be happening with PCB 52, another abundant congener with zero nondetects. It is 3776

dx.doi.org/10.1021/es305024p | Environ. Sci. Technol. 2013, 47, 3774−3780

Environmental Science & Technology

Article

resemble unaltered Aroclor 1242 or vaporized Aroclor 1248. Since the process of vaporization will cause higher MW Aroclors to resemble lighter Aroclors, we interpret this to mean that the B factors represent vaporized Aroclor 1248. The similarity between factor 3-B and vaporized Aroclor 1248 is much weaker, however. Factors 2-E and 3-E somewhat resemble unaltered Aroclor 1254. Factors 2-D and 3-D weakly resemble vaporized Aroclor 1254, and factors 2-C and 3-C weakly resemble Aroclors 1248 and 1242, respectively. Time and Temperature Trends of Resolved Factors. Plots of concentration versus time for the five factors (SI Figure S-2) show that the factors generated by PMF2 display time trends similar to their corresponding PMF 3.0 factors. Only the B and C factors appear to be decreasing steadily over time. The A factors appear largely unchanged over time. In contrast, the D factors increase by about a factor of 3 after January of 2005, and concentrations of the E factors display the high/low/high pattern noted above: they are low and roughly constant from January of 2000 through January of 2005, but are higher by about a factor of 2 before and after that. Thus the temporal trends in the concentrations of the D and E factors cannot be well described by a simple exponential decay model such as eq 1. Fitting the resolved factor concentrations to eq 1 demonstrates that the temperature constants are within the range of those typically observed for gas-phase PCBs,28 but are significantly different between the PMF2 and PMF 3.0 model results for the C and E factors (Figure 3). Although the temperature dependence of atmospheric PCB concentrations has been observed to increase with the MW of the PCB congener, homologue, or formulation,28 the temperature constants do not increase in lock-step with average MW, which generally increases from factors A through E. The

4% of factor 3-E, despite being absent in factor 2-E. In other data sets not described here, we have noted that PMF 3.0 has trouble accurately apportioning species that are particularly abundant in the data matrix. We speculate that this may be due to the “robust” mode of PMF 3.0 being less effective than the robust mode of PMF2. Although both PMF2 and the PMF 3.0 follow the same default definition of outliers (scaled residuals >4), the two programs handle the outliers differently. The PMF2 approach is based on the Huber influence function, wherein outliers are kept but downscaled for minimizing the target Q function.22 In contrast, in PMF 3.0, the Q function is minimized with the identified outliers excluded.33 This difference may explain our observation that PMF 3.0 sometimes cannot converge on a solution when the data matrix contains concentrations that vary over 6 or more orders of magnitude, while PMF2 can (data not shown). As a result, data must sometimes be normalized (i.e., each concentration expressed as a percent of the sum of all concentrations in the sample) for PMF 3.0 analysis, while this is rarely necessary with PMF2. In an attempt to identify the resolved factors, their congener patterns were compared to those of the Aroclors. For this purpose, the Aroclor congener patterns of Frame et al.34 were used, because they used a similar GC column and an ECD. In previous work,19 we accounted for the process of volatilization by multiplying the Aroclor congener patterns by their liquid vapor pressures before comparing them with the resolved factors. In contrast, Basu et al.35 compared the congener patterns measured in Chicago with the unaltered Aroclor congener patterns and concluded that the PCBs in Chicago air resemble a 1:1 mix of Aroclors 1242 and 1254. Thus in order to be consistent with their approach, we also compared the resolved factors with the unaltered Aroclor congener patterns (Table 2). Also, we removed PCB 33 from these correlations, Table 2. Coefficients of Determination (R2) for the Congener Patterns of the Resolved Factors versus Those of the Most Popular Aroclors from Frame et al.34 a unaltered Aroclors

vaporized Aroclors

factor

1242

1248

1254

1260

1242

1248

1254

1260

2-A 2-B 2-C 2-D 2-E 3-A 3-B 3-C 3-D 3-E

0.62 0.56 0.17 0.01 0.01 0.60 0.51 0.33 0.00 0.02

0.04 0.18 0.34 0.08 0.00 0.03 0.18 0.28 0.10 0.02

0.01 0.00 0.06 0.20 0.42 0.00 0.00 0.00 0.13 0.41

0.03 0.03 0.01 0.00 0.12 0.02 0.01 0.03 0.00 0.05

0.63 0.20 0.00 0.03 0.02 0.56 0.15 0.05 0.02 0.03

0.58 0.51 0.16 0.00 0.03 0.53 0.58 0.25 0.00 0.03

0.03 0.06 0.17 0.17 0.00 0.01 0.05 0.11 0.23 0.03

0.34 0.11 0.03 0.00 0.04 0.31 0.09 0.05 0.00 0.02

a

PCB 33 was omitted from these correlations.

because PCB 33 was an outlier in the comparison between factors 2-B and 3-B. The effect of removing PCB 33 from these correlations was minimal for the PMF2 factors. The effects on the correlations between the PMF 3.0 factors and the Aroclors was more pronounced, generally leading to an increase in correlation coefficients. In general, the resemblance between the resolved factors and the Aroclors (unaltered or vaporized) was not strong. The best correlation was between factors 2-A and 3-A versus either unaltered or vaporized Aroclor 1242. The B factors best

Figure 3. Time and temperature dependence of the resolved factors in the air of Chicago 1996−2007 from eq 1. Top panel represents the temperature coefficient (a1 of eq 1). Bottom panel represents the time constant (a2 from eq 1 in 1/d). Error bars represent 95% confidence limits. 3777

dx.doi.org/10.1021/es305024p | Environ. Sci. Technol. 2013, 47, 3774−3780

Environmental Science & Technology

Article

Interpretation. The B factors comprise a large share of the mass in the data set (17−20%) and are declining with half-lives of about 4−7 years. They bear some similarity to low MW Aroclors and display relatively weak temperature dependence. We therefore speculate that the B factors represent primary emissions of Aroclors from transformers and other electrical equipment. This would explain the decline in the concentration of the B factors, since removal of PCBs from electrical equipment has been a major focus of the GLBTS. If our interpretation is correct, it suggests that the GLBTS is having an impact on at least one source of PCBs to the atmosphere of Chicago. However, it is also possible that the decline in the B factors is unrelated to the efforts under GLBTS. It could represent the normal replacement of equipment at the end of its design life, natural attenuation, or other processes. The only other factors that are declining with time are the C factors. They do not resemble any of the Aroclors and display relatively strong temperature dependence. Thus we speculate that these factors represent air−surface exchange, which would be a secondary source of PCBs. These factors constitute about 21−22% of the mass in the data set, suggesting that secondary sources are important in the urban atmospheric PCB burden, in agreement with the results of passive sampling in the Philadelphia metropolitan area.36 These two factors (2-C and 3-C) display the weakest similarity to each other of the five pairs. Since the air−surface exchange congener pattern may change from day to day and with the seasons, this lack of similarity seems reasonable. Likewise, it seems reasonable to assume that natural attenuation would result in a steady, gradual decrease over time that would at least resemble exponential decay. Thus we interpret the gradual decline in the C factors as a form of natural attenuation. The remaining factors, which represent between 56% and 67% of the mass in the data set, are not declining over time. This is troubling and suggests that the efforts of GLBTS and other initiatives are not effective at decreasing the overall atmospheric PCB burden in Chicago, and therefore are not controlling the atmospheric deposition of PCBs to Lake Michigan. The A factors display constant concentrations over time, and comprise 17−20% of the mass of PCBs in the atmosphere. These factors show a relatively strong resemblance to Aroclor 1242, which accounted for about 52% of U.S. production of Aroclors.37 The very similar Aroclor 1016 accounted for another ∼13%, so it is not surprising that this congener pattern is prevalent in the Chicago area. These Aroclors were used in electrical equipment such as transformers and capacitors, but Aroclor 1242 had many other uses, as a heat transfer liquid, hydraulic fluid, or wax extender, as well as in gas transmission turbines, rubbers, and carbonless copy paper.37 The actions under GLBTS do not seem to be affecting the A factors, perhaps suggesting that they are not associated with electrical equipment. The C and D factors are higher in MW, and the D factors somewhat resemble Aroclor 1254. These factors do not display an exponential decay or growth, but instead display periods of high and low concentrations. Concentrations of both factors appear to have increased substantially (by a factor of 2−3) after about January of 2005. These increases appear to be primarily responsible for the increase in average MW observed starting around 2005 (Figure 1). Concentrations of the E factors were also higher and increasing during the period from January 1996 to about January 2000. These patterns suggest releases of PCBs

relatively strong temperature dependence of the C factors, coupled with the fact that they do not resemble any of the Aroclors, suggests that these factors may represent volatilization of PCBs from surfaces such as soil, water, or vegetation, and therefore represent secondary sources of PCBs. For the D and E factors, the time constants are not significantly different between the PMF2 and PMF 3.0 models. However, they are different for the A, B, and C factors. Factor 2-A is increasing, while factor 3-A is decreasing, and both trends are statistically significant at the 95% confidence level. Nevertheless, the time constants are small and the p-values are marginal (p = 0.02 for the PMF2 results, and p = 0.04 for the PMF 3.0 results). Thus it appears that both models indicate little change in the concentrations of the A factors over time. For the B factors, both models suggest that they are decreasing, but 2-B is decreasing faster (t1/2 = 4.3 y) than factor 3-B (t1/2 = 7.2 y). The opposite is true for the C factors. 2-C is decreasing more slowly (t1/2 = 8.3 y) than factor 3-C (t1/2 = 4.1 y). The D factors are increasing with time, and the E factors are unchanged. As we have seen, visual inspection of the concentration versus time plots indicates that any time constants obtained from eq 1 for factors 2-E and 3-E are not very meaningful, since these factors display a high/low/high concentration pattern over time, nothing like an exponential decay (or growth). The time constants are equally meaningless for the D factors, since they increased markedly in about January of 2005 in a manner that cannot be described as exponential growth. In sum, the temporal variability in the A, D, and E factors cannot be described as exponential growth or decay. The remaining two factors are decreasing, with half-lives between 4 and 9 years. These two factors represent air/surface exchange (factor C) and Aroclor 1242 (factor B). It is unclear why other factors that represent Aroclors (factor A/Aroclor 1242 and factor E/Aroclor) are not decreasing. The high/low/high pattern of factor E might be driven by periodic human activity. We speculate that the decreases in factor B might be a result of removal of PCB-containing equipment under the GLBTS, but this is by no means certain. Comparison to Other Studies. Du et al.19 utilized PMF to examine PCB concentrations in the atmosphere (gas plus particle phase) of Camden, NJ. For that study, the PMF model was applied to a smaller data set of 74 sampling events and 52 PCB congeners or coeluting congener groups. Four factors were resolved (designated here as W, X, Y, and Z). The majority of the PCB mass was represented by factors W through Y. Du et al. suggested that factor W was represented by a combination of low molecular weight Aroclors such as 1016, 1242, and 1248. In the present study, factors A and D similarly were well-described as a mixture of low MW Aroclors. The Camden factor X resembled a mixture of Aroclors 1248 and 1254. Camden factors Y and Z did not bear any resemblance to the Aroclors, though factor Z was thought to resemble the average particle phase PCB profile. In Chicago, the B and C factors bore only a weak resemblance to a mixture of Aroclors. It is likely that given a larger data set, Du et al. could have resolved more factors in the PCB profile in Camden. Both studies suggest that the urban atmospheric PCB signal is dominated by low molecular weight Aroclors. Also, both studies suggest that some portion of the atmospheric signal does not resemble the Aroclors, suggesting that it has undergone substantial weathering. 3778

dx.doi.org/10.1021/es305024p | Environ. Sci. Technol. 2013, 47, 3774−3780

Environmental Science & Technology



that may be associated with human activities, such as disturbance of contaminated soil, transfer of contaminated waste, etc. Venier et al.29 note that the increases in PCB concentrations starting in 2005 could be related to demolition and construction activities occurring near the sampling site starting in the summer of 2005. Alternatively, we speculate that increasing PCB concentrations could reflect deterioration of PCB-containing equipment that was installed more than 30 years ago, before the ban on PCBs, and is now reaching the end of its design life and beginning to release PCBs to the environment. Implications. Our comparison of PMF2 and PMF 3.0 has demonstrated that both models give similar results when fed the same data set. Although both models generally tell the same story about PCBs in the atmosphere of Chicago, PMF 3.0 seemed to have trouble handling abundant congeners such as PCB 33 and 52. Subtle differences in the results (both in the F matrix and the G matrix) are apparent. Similar discrepancies have been reported between PMF2 and versions of EPA PMF for PM2.5 source apportionment. Amato et al.38 reported similar source profiles but 20% differences in source contributions for certain factors. Kim and Hopke25 also reported similar sources extracted by both PMF2 and EPA PMF (an earlier version) but large differences in source contributions. The discrepancies between PMF2 and the EPA PMF have been speculated to be due to (1) different algorithms used to solve the PMF model, (2) different non-negativity constraints, and (3) different ways to handle rotational freedom. To this list, we would add differences in the robust mode and handling of outliers. To the best of our knowledge, no systematic simulation studies have been conducted to examine the similarities and differences in PMF results across the two algorithms under various data structures. In the present work, we have the advantage of some a priori knowledge of the composition of the source terms, since the primary sources of PCBs in the environment in the United States are the Aroclors. PMF2 does a better job of generating factors that resemble the original Aroclor formulations, largely due to its better treatment of PCBs 33 and 52. Thus we can say with some certainty that the PMF2 model outperformed EPA PMF 3.0 in analyzing our data set. It is important to consider that if we had not had this a priori knowledge, we would not have known to eliminate PCB 33 from the correlations in Table 2, and consequently we might have misidentified some of the factors. This analysis has demonstrated that not all urban atmospheric PCB sources are declining via well-behaved exponential decay. Neither the raw data nor many of the resolved factors display this kind of decay, and thus these data cannot be used to predict the future concentrations of PCBs that are likely to prevail in Chicago. It appears that past performance is not a guarantee of future returns in Chicago, and perhaps in other urban areas as well. This analysis also demonstrates that the urban atmospheric PCB signal is comprised of several source types which display quite different patterns of temperature dependence and temporal variability, in agreement with our previous study of PCBs in Camden, NJ.19 The fact that only two of the factors display significant decreases over time is worrisome because it indicates that human efforts to remove PCBs from the environment have been at best only marginally successful in decreasing atmospheric concentrations. Additional effort is needed to identify and eliminate atmospheric PCB sources in Chicago.

Article

ASSOCIATED CONTENT

S Supporting Information *

Two figures, four tables, and other supporting details. This information is available free of charge via the Internet at http:// pubs.acs.org.



AUTHOR INFORMATION

Corresponding Author

*Phone: 848-932-5774; fax: 732-932-8644; e-mail: [email protected]. Notes

The authors declare no competing financial interest.



ACKNOWLEDGMENTS We thank Ron Hites and the IADN team for sharing their data with us.



REFERENCES

(1) Hsu, Y. K.; Holsen, T. M.; Hopke, P. K. Comparison of Hybrid Receptor Models to Locate PCB Sources in Chicago. Atmos. Environ. 2003, 37, 545−562. (2) Hsu, Y. K.; Holsen, T. M.; Hopke, P. K. Locating and Quantifying PCB Sources in Chicago: Receptor Modeling and Field Sampling. Environ. Sci. Technol. 2003, 37, 681−690. (3) U.S. EPA. 1990 Emissions Inventory of Section 112(c)(6) Pollutants: Polycyclic Organic Matter (POM), 2,3,7,8-Tetrachlorodibenzo-p-Dioxin (TCDD)/2,3,7,8,-Tetrachlorodibenzofuran (TCDF), Polychlorinated Biphenyl Compounds (PCBs), Hexachlorobenzene, Mercury, and Alkylated Lead; Office of Air Quality Planning and Standards: Research Triangle Park, NC, 1997. (4) Herrick, R. F.; McClean, M. D.; Meeker, J. D.; Baxter, L. K.; Waymouth, G. A. An unrecognized source of PCB contamination in schools and other buildings. Environ. Health Perspect. 2004, 112, 1051−1053. (5) Kohler, M.; Tremp, J.; Zennegg, M.; Seiler, C.; Minder-Kohler, S.; Beck, M.; Lienemann, P.; Wegmann, L.; Schmid, P. Joint Sealants: An overlooked diffuse source of Polychlorinated Biphenyls in Buildings. Environ. Sci. Technol. 2005, 39, 1967−1973. (6) Franz, T. P.; Eisenreich, S. J. Wet deposition of polychlorinated -biphenyls to Green Bay, Lake Michingan. Chemosphere 1993, 26, 1767−1788. (7) Simcik, M. F.; Zhang, H.; Eisenreich, S. J.; Franz, T. P. Urban contamination of the Chicago/coastal Lake Michigan atmosphere by PCBs and PAHs during AEOLOS. Environ. Sci. Technol. 1997, 31, 2141−2147. (8) Offenberg, J. H.; Simcik, M.; Baker, J. E.; Eisenreich, S. J. The impact of urban areas on the deposition of air toxics to adjacent surface waters: A mass budget of PCBs in Lake Michigan in 1994. Aquat. Sci.s 2005, 67, 79−85. (9) Totten, L. A.; Gigliotti, C. L.; VanRy, D. A.; Offenberg, J. H.; Nelson, E. D.; Dachs, J.; Reinfelder, J. R.; Eisenreich, S. J. Atmospheric Concentrations and Deposition of PCBs to the Hudson River Estuary. Environ. Sci. Technol. 2004, 38, 2568−2573. (10) Totten, L. A.; Panangadan, M.; Eisenreich, S. J.; Cavallo, G. J.; Fikslin, T. J. Direct and Indirect Atmospheric Deposition of PCBs to the Delaware River Watershed. Environ. Sci. Technol. 2006, 40, 2171− 2176. (11) Zhang, H.; Eisenreich, S. J.; Franz, T. R.; Baker, J. E.; Offenberg, J. H. Evidence for increased gaseous PCB fluxes to Lake Michigan from Chicago. Environ. Sci. Technol. 1999, 33, 2129−2137. (12) U.S. EPA. Great Lakes Binational Toxics Strategy. http://www. epa.gov/bns/ (accessed July 15, 2010). (13) Sun, P.; Basu, I.; Hites, R. A. Temporal trends of polychlorinated biphenyls in precipitation and air at Chicago. Environ. Sci. Technol. 2006, 40, 1178−1183.

3779

dx.doi.org/10.1021/es305024p | Environ. Sci. Technol. 2013, 47, 3774−3780

Environmental Science & Technology

Article

(14) Bzdusek, P. A.; Christensen, E. R. Comparison of a new variant of PMF with other receptor modelling method using artificial and real sediment PCB data sets. Environmetrics 2006, 17, 387−403. (15) Bzdusek, P. A.; Christensen, E. R.; Lee, C. M.; Pakdeesusuk, U.; Freedman, D. C. PCB Congeners and Dechlorination in Sediments of Lake Hart Well, South Carolina, Determined from Cores Collected in 1987 and 1998. Environ. Sci. Technol. 2006, 40, 109−119. (16) Bzdusek, P. A.; Lu, J.; Christensen, E. R. PCB Congeners and Dechlorination in Sediment of Sheboygan River, Wisconsin, Determined by Matrix Factorization. Environ. Sci. Technol. 2006, 40, 120−129. (17) Du, S.; Belton, T. J.; Rodenburg, L. A. Source apportionment of Polychlorinated Biphenyls in the Tidal Delaware River. Environ. Sci. Technol. 2008, 42, 4044−4051. (18) Rodenburg, L. A.; Du, S.; Xiao, B.; Fennell, D. E. Source Apportionment of Polychlorinated Biphenyls in the New York/New Jersey Harbor. Chemosphere 2011, 83, 792−798. (19) Du, S.; Rodenburg, L. A. Source Identification of Atmospheric PCBs in Philadelphia/Camden Using Positive Matrix Factorization Followed by the Potential Source Contribution Function. Atmos. Environ. 2007, 41, 8596−8608. (20) Du, S.; Wall, S. J.; Cacia, D.; Rodenburg, L. A. Passive air sampling for polychlorinated biphenyls in the Philadelphia metropolitan area. Environ. Sci. Technol. 2009, 43, 1287−1292. (21) Paatero, P.; Tapper, U. Positive Matrix Factorization: A Nonnegative Factor Model with Optimal Utilization of Error Estimates of Data Values. Environmetrics 1994, 5, 111−126. (22) Reff, A.; Eberly, S. I.; Bhave, P. V. Receptor modeling of ambient particulate matter data using positive matrix factorization: Review of existing methods. J. Air Waste Manage. Assoc. 2007, 57 (2), 146−154. (23) U.S.EPA; Environment Canada Integrated Atmospheric Deposition Network. www.msc.ec.gc.ca/iadn/ (accessed July 15, 2010). (24) Environment Canada. Integrated Atmospheric Deposition Network: IADN Resources. http://www.msc-smc.ec.gc.ca/iadn/ resources/resources_e.html (accessed July 15, 2010). (25) Kim, E.; Hopke, P. K. Source identifications of airborne fine particles using positive matrix factorization and U.S. Environmental Protection Agency positive matrix factorization. J. Air Waste Manage. Assoc. 2007, 57 (7), 811−819. (26) Basu, I.; Hafner, W. D.; Mills, W. J.; Hites, R. A. Differences in Atmsopheric Persistent Organic Pollutant Concentrations at Two Locations in Chicago. J. Great Lakes Res. 2004, 30, 310−315. (27) Buehler, S. S.; Basu, I.; Hites, R. A. Causes of Variability in Pesticide and PCB Concentrations in Air near the Great Lakes. Environ. Sci. Technol. 2004, 38, 414−422. (28) Carlson, D. L.; Hites, R. A. Temperature Dependence of Atmospheric PCB Concentrations. Environ. Sci. Technol. 2005, 39, 740−747. (29) Venier, M.; Hung, H.; Tych, W.; Hites, R. A. Temporal Trends of Persistent Organic Pollutants: A Comparison of Different Time Series Models. Environ. Sci. Technol. 2012, 46 (7), 3928−3934. (30) Venier, M.; Hites, R. A. Time Trend Analysis of Atmospheric POPs Concentrations in the Great Lakes Region Since 1990. Environ. Sci. Technol. 2010, 44 (21), 8050−8055. (31) Sun, P.; Basu, I.; Blanchard, P.; Brice, K. A.; Hites, R. A. Temporal and spatial trends of atmospheric polychlorinated biphenyl concentrations near the Great Lakes. Environ. Sci. Technol. 2007, 41 (4), 1131−1136. (32) Van den Berg, M.; Birnbaum, L. S.; Denison, M.; De Vito, M.; Farland, W.; Feeley, M.; Fiedler, H.; Hakansson, H.; Hanberg, A.; Haws, L.; Rose, M.; Safe, S.; Schrenk, D.; Tohyama, C.; Tritscher, A.; Tuomisto, J.; Tysklind, M.; Walker, N.; Peterson, R. E. The 2005 World Health Organization reevaluation of human and mammalian toxic equivalency factors for dioxins and dioxin-like compounds. Toxicol. Sci. 2006, 93, 223−241. (33) Paatero, P. User’s Guide for Positive Matrix Factorization Programs PMF2 and PMF3. Part 1: Tutorial; 2003.

(34) Frame, G. M.; Cochran, J. W.; Boewadt, S. S. Complete PCB congener distributions for 17 Aroclor mixtures determined by 3 HRGC systems optimized for comprehensive, quantitative, congenerspecific analysis. J. High Res.Chromatogr. 1996, 19, 657−668. (35) Basu, I.; Arnold, K. A.; Vanier, M.; Hites, R. A. Partial Pressures of PCB-11 in Air from Several Great Lakes Sites. Environ. Sci. Technol. 2009, 43 (17), 6488−6492. (36) Du, S.; Wall, S. J.; Cacia, D.; Rodenburg, L. A. Passive Air Sampling for Polychlorinated Biphenyls in the Philadelphia, USA Metropolitan Area. Environ. Sci. Technol. 2009, 43, 1287−1292. (37) Brown, J. F. Determination of PCB metabolic, excretion, and accumulation rates for use as indicators of biological response and relative risk. Environ. Sci. Technol. 1994, 28, 2295−2305. (38) Amato, F.; Pandolfi, M.; Escrig, A.; Querol, X.; Alastuey, A.; Pey, J.; Perez, N.; Hopke, P. K. Quantifying road dust resuspension in urban environment by Multilinear Engine: A comparison with PMF2. Atmos. Environ. 2009, 43 (17), 2770−2780.

3780

dx.doi.org/10.1021/es305024p | Environ. Sci. Technol. 2013, 47, 3774−3780