Environ. Sci. Technol. 2002, 36, 1671-1675
Discrimination of Aerial Deposition Sources of Polychlorinated Dibenzo-p-Dioxin and Polychlorinated Dibenzofuran Downwind from a Pulp Mill near Ketchikan, Alaska DANIEL C. PEEK,† MATTHEW K. BUTCHER,‡ W A L T E R J . S H I E L D S , * ,† LISA J. YOST,† AND JOHN A. MALOY§ Exponent, 15375 SE 30th Place, Suite 250, Bellevue, Washington 98007, ENTRIX, Inc., 2701 First Avenue, Suite 240, Seattle, Washington 98121, and Philip Services Corporation, 955 Powell Avenue SW, Renton, Washington 98055
Drinking water is supplied by individual roof-catchment systems for homes and businesses near a dissolving sulfite pulp mill (now closed) located just north of Ketchikan in southeast Alaska. This study was conducted to determine if polychlorinated dibenzo-p-dioxins and polychlorinated dibenzofurans (PCDDs/Fs) found in the sediments of the roof-catchment cisterns resulted from historical deposition of stack emissions from the pulp mill’s multi-fuel power boilers. Fly ash from the power boilers had maximum total PCDD/F concentrations of 3.08 × 1053.10 × 106 ng/kg, which resulted from combustion of bleach plant wastewater sludge and saltwater-soaked wood waste. Cistern sediments had maximum total PCDD/F concentrations of 7.71 × 104 ng/kg. Potential sources of PCDDs/Fs in the cistern sediments were considered to be automobile exhaust, heating oil combustion, and private trash burning as well as pulp mill boiler emissions. Discriminant analysis was used to analyze differences between profiles of tetra through octa homologue classes of PCDDs/ Fs (defined as proportional contributions to total concentration) from different source terms. Homologue profiles of potential sources from Ketchikan included in this analysis were fly ash collected from the mill’s power boilers and soils collected from background areas (areas with similar PCDD/F sources as the residences [e.g., auto exhaust and burn barrels] near the mill but beyond the zone of aerial deposition of emissions from the mill). Profiles for emissions from automobile exhaust, fertilizers, oil heating, residential trash burning, and residential wood heating were also included in the source “training” data set (for the discriminant analysis) using data from published literature. The classification rules developed from the discriminant analysis were applied to the following test media sampled at Ketchikan: roof-catchment cistern sediments and * Corresponding author phone: (425)643-9803; fax: (425)643-9827; e-mail:
[email protected]. † Exponent, Bellevue, WA. ‡ ENTRIX, Inc. § Philip Services Corp. 10.1021/es011273c CCC: $22.00 Published on Web 03/07/2002
2002 American Chemical Society
soils collected from areas in the vicinity of the mill’s power boilers (i.e., nearby residential or commercial [developed] areas, on the mill property, and nearby forestlands). The homologue profiles of cistern sediment and nearby developed area soil samples were similar to background soils, whereas the profiles for the forestland soil samples (influenced by emissions from the mill but not other anthropogenic sources) closely matched the fly ash pattern. The homologue profiles of the emission sources from published data were more similar to one another than either background soils or fly ash. Soil samples from the mill property were classified as members of all source groups. On the basis of these analyses, the composition of PCDDs/Fs detected in the cistern sediments is typical of Ketchikan background conditions and not reflective of mill emissions.
Introduction Ketchikan Pulp Company (KPC) operated its mill on Ward Cove north of Ketchikan, AK, from its construction in 1954 until shutdown in 1997 for the production of high-quality dissolving pulp. To supply power for mill operations, hog fuel (primarily wood waste from saltwater-soaked logs), dewatered wastewater treatment plant sludge, and fuel oil were burned in two power boilers (operating at a rate of 54.4 t of steam/h) located at the mill. The maximum total polychlorinated dibenzo-p-dioxin and polychlorinated dibenzofuran (PCDD/F) concentration in sludge was 9660 ng/kg (dry wt) (1), whereas the maximum total concentration in fly ash was 3.10 × 106 ng/kg. PCDDs/Fs were apparently synthesized de novo during combustion of sludge that contained chlorinated effluent from the pulp bleaching operations and combustion of hog fuel from logs that had been stored in rafts in saltwater. In a previous study, aerial deposition modeling with sitespecific data was used to delineate nearby areas of maximum historical deposition of particles emitted from the power boilers. Maximum nearby deposition of total PCDDs/Fs was predicted to be less than 5000 µg/m2 over the 43-year life of the mill. The predicted deposition was consistent with the PCDD/F mass in the surface soils that was estimated from measured soil concentrations in the area (approximately 872-2.17 × 104 ng/kg) (2). Because of a lack of suitable groundwater and surface water sources, drinking water for domestic purposes in the area around the former mill has been almost exclusively derived from rainwater captured with roof-catchment systems and stored in cisterns (3). PCDDs/Fs have been detected in sediment from cisterns near the mill at total concentrations of 4750-7.71 × 104 ng/kg (1, 4). The objective of this study was to determine whether historical stack emissions from the pulp mill’s power boilers were an important source of PCDDs/Fs to drinking water cisterns.
Materials and Methods Field Sampling and Laboratory Analysis. Twenty-three pulp mill boiler fly ash and bottom ash, 42 on-site soil, 10 nearby developed area (commercial and/or residential) soil, 12 nearby forest soil, 14 background soil, and 4 drinking water cistern sediment samples were collected. Background soil samples were collected from residential areas similar to the residences near the mill but beyond the zone of aerial VOL. 36, NO. 8, 2002 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
1671
FIGURE 1. Soil and cistern sampling stations near the KPC pulp mill. deposition of emissions from the mill. In addition, results of analysis of 67 samples from auto emissions, fertilizers, oil heating emissions, residential trash burning, and wood burning gathered from the literature were included in the data set. Prior to shutdown of the pulp mill in March 1998, samples of power boiler fly ash and bottom ash were collected. Fly ash was collected from both the mechanical cyclone separators (Bresloves), which captured the heavy fraction of fly ash, and the electrostatic precipitators (ESPs), which captured the lighter fraction. Five samples (one per day) of power boiler Breslove fly ash, power boiler ESP fly ash, and power boiler bottom ash were collected (1). In addition, data from two other sampling events were included in the data set: (i) three samples of power boiler stack emissions (after installation of the ESPs) collected during stack testing during 19951997 (5, 6) and (ii) four fly ash samples collected from the mill’s wood waste and ash disposal landfill and one fly ash sample from the on-site fly ash silo in 1997 (4). Local background composite surface soil samples were collected shortly after shutdown of the mill from six residential and seven forested areas in the Ketchikan area that were outside of the influence of aerial deposition from the mill (1). The locations selected for collection of background samples had been, to the extent possible, subject to similar PCDD/F sources that the mill area and adjacent soils would have been subject to in the absence of the mill. Specifically, these sources included highway traffic and residences and/ or small businesses using wood heat (wood stoves or fireplaces) and burn barrels for trash incineration. Fortyseven soil samples were collected from various areas of the mill site itself (1). Previous modeling (2) indicated that the maximum deposition of aerial emissions would have been to the north and northwest of the power boilers. Surface soil samples were collected off-site from both developed (commercial/residential) areas (10 samples) and forested areas (12 samples), and sediment samples (4 samples) were collected from drinking water cisterns in the area that may have been affected by aerial deposition from the mill’s boiler emissions in the area (Figure 1) (1, 4). All samples were analyzed for PCDDs/Fs by Maxim Technologies, Inc., of St. Paul, MN, in accordance with a modified version of U.S. Environmental Protection Agency Method 1613 (7). A complete description of the method can be found in the Supporting Information. A quality assurance 1672
9
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 36, NO. 8, 2002
review was conducted of the sample data and associated quality control information. All data were acceptable. In addition to the results of the PCDD/F analysis of samples collected from the Ketchikan area, data from the literature were included in the training data set for the discriminant analysis. These data represented various sources of PCDDs/Fs and included 10 automobile exhaust samples (8, 9), 20 samples of soot from oil-fired central heating chimneys (10), 8 fertilizer samples (11), 16 results from open burning of household wastes in barrels (12), and 13 soot or emissions results from wood burning (10, 13). The complete data set can be found in the Supporting Information. Data Analysis. Discriminant analysis (14) was used to classify the cistern sediment and nearby soil samples based on their profiles of tetra through octa homologue classes of PCDDs/Fs. This method derives a set of linear functions that translate the original variables into new variables that emphasize the differences between observations from different categories. The analysis is based on a training data set in which the category membership of each observation is known a priori. A set of linear discriminant functions defines regions that demarcate those categories and allows new observations to be classified by them. The source term training data set was constructed from data reported in the literature, data from background soil collected in the Ketchikan area, and data from fly ash associated with the pulp mill. These training data were placed into the following groups: automobile exhaust, fertilizer, emissions (i.e., soot) from oil heating, burning of household wastes, household wood burning, soil samples representing areas not influenced by aerial emissions from the KPC facility (i.e., background), and ash from the KPC power boilers. For each source term sample, homologue data were transformed into proportions of the total of tetra- through octachlorinated PCDDs/Fs. Proportions were used to remove the influence of differences in concentrations between samples and especially between different sources. Analyses were conducted with SYSTAT Version 8.0 (15). An important measure of the performance of the discriminant functions is how accurately they classify data from the training data set. That is, when the discriminant functions are applied to the training data set, are the data correctly classified by category? Are data from some categories classified incorrectly more frequently than others? A test classification table was constructed from our training data set using the jackknife method. In this technique, a set of discriminant functions is generated using the subset of all observations in the training data set except the one being classified, and these functions are then applied to the missing observation; the process is repeated for each observation. Jackknifed classification tables are more conservative than conventional tables in estimating the rate of classification error because the observation in question has no influence over the discriminant functions used to classify it (16).
Results and Discussion Source Discrimination. The more obvious differences between the sources for each homologue class are illustrated in Figure 2. In these bar graphs, the first seven bars represent the proportions for the source terms or training data set (auto, fertilizer, oil heat, trash burning, wood burning, background, and KPC power boiler ash [referred to as KPC fly ash]), and the last four bars represent soil or sediment samples for which the PCDD/F source(s) is(are) not known (cistern, nearby developed, KPC on-site, and nearby forest). KPC fly ash samples tended to have higher proportions of lower chlorinated PCDD homologue groups (i.e., TCDD, PeCDD, and HxCDD) than samples from other sources or background samples. None of the other source groups so
TABLE 1. Classification Results of the Training Data Seta
auto fertilizer oil heat trash burning wood burning background KPC fly ash total b
auto
fertilizer
oil heat
trash burning
wood burning
background
KPCb fly ash
total
% correct
3 0 0 0 0 1 5 9
0 7 2 1 1 0 1 12
0 0 12 0 3 0 0 15
3 0 0 15 4 0 0 22
1 1 6 0 5 0 0 13
2 0 0 0 0 13 0 15
1 0 0 0 0 0 17 18
10 8 20 16 13 14 23 104
30 88 60 94 38 93 74 69
a Observations from each source (rows) are classified by discriminant functions (columns). A jackknifed data set was applied for each datum. KPC, Ketchikan Pulp Company.
FIGURE 3. Canonical variable scores of the training data set. The first two canonical variables are shown; the centroid of each source is labeled and surrounded by a (1 standard deviation confidence ellipsoid.
FIGURE 2. Proportional contributions of PCDD/F homologues for each PCDD/F source. The height of the bar represents the mean proportion by source, and the error bar represents (1 standard deviation. clearly distinguish themselves within any single homologue group. When the discriminant functions were applied to the training data set, 69% of all observations were correctly classified into their original group (Table 1). Training data
for auto emissions had the worst score, with only 30% of the observations correctly assigned; however, only one observation was misclassified as KPC fly ash. One of the 14 background samples was misclassified, and six of the 23 KPC fly ash samples were incorrectly classified (as auto or fertilizer). When the source data are plotted in the first two canonical variable scores, KPC fly ash and the other groups are clearly separated from one another (Figure 3). The homologue profiles of the emission sources from published data were more similar to one another than either background or KPC fly ash. The differences in discrimination between source terms are apparent in the matrix of F values comparing sources (Table 2). F values were smaller for pairs of group means between sources other than fly ash and background, indicating that these sources obtained from the literature could be less clearly distinguished from each other. Classification of Test Media. As with KPC fly ash samples, nearby (aerial deposition area) forest soil samples had higher proportions of lower chlorinated PCDD homologue groups (TCDD, PeCDD, and HxCDD) than other sample types (Figure 2). In addition, although not as prominent as the proportions of the TCDD, PeCDD, and HxCDD groups, the overall pattern of homologue distribution is similar for KPC fly ash and forest soil samples while the pattern of homologue distribution for cistern samples (as well as KPC on-site and nearby developed area soils) is unlike that of KPC fly ash (Figure 4). Results of this graphical comparison of patterns indicate that while VOL. 36, NO. 8, 2002 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
1673
TABLE 2. Discriminant Analysis Results: Matrix of F Statistics between Centroids of PCDD/F Sourcesa
auto fertilizer oil heat trash burning wood burning background KPC fly ash a
auto
fertilizer
oil heat
trash burning
wood burning
background
KPC fly ash
0 3.4 8.4 8.7 10.0 15.2 12.5
0 2.3 2.8 2.4 22.5 10.5
0 5.7 2.4 34.8 27.2
0 2.7 43.2 33.1
0 39.9 27.9
0 35.3
0
Critical F value: F0.001(9,89) ) 3.48. Larger F values indicate pairs of groups whose centroids are farther apart.
TABLE 3. Source Predictions for the Test Media, Based on Discriminant Analysis Resultsa sample type
auto
fertilizer
oil heat
background
KPCb fly ash
total
cistern sediment nearby developed area soils KPC on-site soils nearby forest soils
0 0
0 0
0 0
4 10
0 0
4 10
3 0
0 0
1 0
36 2
2 10
42 12
a No sample types were classified into the trash burning or wood burning categories. b KPC, Ketchikan Pulp Company
FIGURE 4. Canonical variable scores of the site data set. The first two canonical variables are shown. The centroids of the training data sets are shown as filled circles and labeled. The site test data are categorized by symbol, and sample RW-32 is labeled. aerial emissions from the mill (KPC fly ash) are an important source of PCDDs/Fs to forested (i.e., undeveloped) areas within the maximum aerial depositional area, other sources are more important contributors to the PCDDs/Fs found in developed areas. The discriminant functions developed from the training data set were applied to sample data from roof-catchment cistern sediments, nearby developed area soils, KPC on-site soils, and nearby forest soils. Ten of the 12 samples collected from forest soils downwind from the aerial emissions of the KPC power boilers were classified by the discriminant functions as KPC fly ash (Table 3). The two forest area samples (ADFS-C8 and -C12) that were classified as background had the highest (5440 ng/kg) and lowest (145 ng/kg) total PCDD/F concentrations of the forest area soil samples and were located the closest to and farthest from the KPC power boilers, respectively (Figure 1). In addition, both samples had the highest proportions of OCDD and lowest proportions of TCDDs of the forest soil samples (data not shown here; as stated, complete data set can be found in the Supporting Information). Sample ADFS-C8, which had a total PCDD/F concentration well above the maximum background concentration of 1770 ng/kg, was collected from a forested area 1674
9
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 36, NO. 8, 2002
that not only was close to the mill but also was within 50 m of a backyard burn barrel. Thus, the PCDD/F composition of this sample was probably influenced by other sources in addition to aerial emissions from the KPC mill. Sample ADFSC12 was most likely collected from outside, or at least on the margin, of the forested area affected by aerial emissions from the mill. Thirty-six of the 42 soil samples collected from on-site areas were classified as background (Table 3). The classification of six of the on-site samples as auto, oil heat, or KPC fly ash most likely represents sitewide variations in the potential sources of PCDDs/Fs to these samples (e.g., KPC fly ash and petroleum combustion from vehicles and railroad traffic). All of the cistern sediments samples were classified as background as were all soil samples from the nearby developed areas (Table 3). Although all cistern samples were classified as background by the discriminant analysis, one sample (RW-32) is clearly dissimilar to the other three cistern samples (Figure 4). Sample RW-32 had a total PCDD/F concentration (4.09 × 104 ng/kg) within the range of the other cistern samples (4750-7.71 × 104 ng/kg); however, this sample had detectable levels of polychlorinated biphenyls (PCBs) (160 µg/kg), polycyclic aromatic hydrocarbons (PAHs) (5660 µg/kg), phenols (4050 µg/kg), and phthalates (1980 µg/kg) that were not found in soil samples collected from the same residence nor in other cistern sediment samples. Concentrations of PAHs and phenols in fly ash samples from the mill were much lower than those in the RW-32 sample (PCBs and phthalates were not detected in fly ash samples). Thus, the source of the PCDDs/Fs found in cistern sediment sample RW-32 is probably the result of the application of some material containing organic contaminants to the roof itself (e.g., tar or asphalt shingles) or the use of materials containing these contaminants in the construction of the roof or catchment system and not from such other potential sources such as auto exhaust, trash burning, or KPC fly ash. The results of this study indicate that the composition of PCDDs/Fs detected in the cistern sediments is typical of soils from residences outside the area of aerial emissions from the pulp mill and is not significantly influenced by mill power boiler emissions. The PCDDs/Fs detected in the nearby forest
soils are apparently related to historical emissions of fly ash from the mill’s power boilers.
Supporting Information Available Analytical methodology that Maxim Technologies, Inc., used to analyze ash, soil, and sediment samples collected from the Ketchikan area for PCDDs/Fs (Appendix A). Data set used in the discriminant analysis, includes both data for samples collected from the Ketchikan area and data from the literature (Appendix B). This material is available free of charge via the Internet at http://pubs.acs.org.
Nomenclature Polychlorinated Dibenzo-p-dioxins TCDD
tetrachlorodibenzo-p-dioxin
PeCDD
pentachlorodibenzo-p-dioxin
HxCDD
hexachlorodibenzo-p-dioxin
HpCDD
heptachlorodibenzo-p-dioxin
OCDD
octachlorodibenzo-p-dioxin
Polychlorinated Dibenzofurans TCDF
tetrachlorodibenzofuran
PeCDF
pentachlorodibenzofuran
HxCDF
hexachlorodibenzofuran
HpCDF
heptachlorodibenzofuran
OCDF
octachlorodibenzofuran
Literature Cited (1) Remedial Investigation, Ketchikan Pulp Company Site, Vols. I-IV; Prepared for Ketchikan Pulp Company, Ketchikan, AK; Exponent: Bellevue, WA, 1998. (2) Shields, W.; Maloy, J. A.; Yost, L.; Peek, D. In Organohalogen Compounds, Vol. 41, Dioxin 99, 19th International Symposium on Halogenated Environmental Organic Pollutants and POPs, Venice, Italy, September 12-17, 1999; Mocarelli, P., Ed.;
University of Milano-Bicocca: 1999; pp 455-458. (3) Atlas of the Ketchikan Region; Martinson, C., Kuklok, D., Eds.; Ketchikan Gateway Borough Planning Department: Ketchikan, AK, 1977. (4) Final Ketchikan Pulp Company Expanded Site Investigation Report; Prepared for U.S. Environmental Protection Agency, Seattle, WA; Ecology and Environment, Inc.: Anchorage, AK, 1998. (5) Source Emission Evaluations; Prepared for Ketchikan Pulp Company, Ketchikan, AK; Am Test-Air Quality, LLC: Preston, WA, 1995-1996. (6) Source Emission Evaluations; Prepared for Ketchikan Pulp Company, Ketchikan, AK; Am Test-Air Quality, LLC: Preston, WA, 1997. (7) Method 1613: Tetra- through Octa-Chlorinated Dioxins and Furans by Isotope Dilution HRGC/HRMS; EPA821B94005; U.S. Environmental Protection Agency, Office of Water, Engineering and Analysis Division: Washington, DC, 1994. (8) Marklund, S.; Anderson, R.; Tysklind, M.; Rappe, C.; Egeback, K. E.; Bjorkman, E.; Grigoriadis, V. Chemosphere 1990, 20, 553561. (9) Oehme, M.; Larssen, S.; Brevik, E. M. Chemosphere 1991, 23, 1699-1708. (10) Thoma, H. Chemosphere 1988, 17, 1369-1379. (11) Screening Survey for Metals and Dioxins in Fertilizer Products and Soils in Washington State; Final Report 99-309; Washington State Department of Ecology: 1999. (12) Lemieux, P. M.; Gullet, B. K.; Lutes, C. C.; Winterrowd, C. K.; Winters, D. L. Parameters Influencing Emissions of PCDDs/Fs from Open Burning of Household Waste in Barrels. Presented at AWMA/Environment Canada Specialty Conference on Recent Advances in the Science and Management of Air Toxics, Banff, Alberta, Canada, April 2000. (13) MOE Toxic Chemical Emissions Inventory for Ontario and Eastern North America; Johnson, N. D., Schult, M. T., Eds.; Final Report P92-T61-5429/OG; Ontario Ministry of the Environment, Air Resources Branch: Rexdale, Ontario, 1992. (14) Morrison, D. F. Multivariate Statistical Methods, 2nd ed.; Arthur, A. A., Young, L. A., Eds.; McGraw-Hill: New York, 1976. (15) SYSTAT 8.0 Statistics; SPSS Inc.: Chicago, IL, 1998. (16) Manly, B. F. J. Multivariate Statistical Methods: A Primer; Chapman and Hall: New York, 1986.
Received for review September 10, 2001. Revised manuscript received January 9, 2002. Accepted January 15, 2002. ES011273C
VOL. 36, NO. 8, 2002 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
1675
