Source Allocation by Least-Squares Hydrocarbon ... - ACS Publications

The constrained least-squares (CLS) source allocation method that fits concentrations meets geologic and chemical constraints better than partial leas...
0 downloads 0 Views 533KB Size
Research Source Allocation by Least-Squares Hydrocarbon Fingerprint Matching W I L L I A M A . B U R N S , * ,† STEPHEN M. MUDGE,‡ A. EDWARD BENCE,§ PAUL D. BOEHM,| JOHN S. BROWN,| DAVID S. PAGE,⊥ AND KEITH R. PARKER# W. A. Burns Consulting Services LLC, Houston, Texas, University of Wales - Bangor, Anglesey, UK, AEB Services, LLC, Friendswood, Texas, Exponent, Maynard, Massachusetts, Bowdoin College, Brunswick, Maine, and Data Analysis Group, Cloverdale, California

There has been much controversy regarding the origins of the natural polycyclic aromatic hydrocarbon (PAH) and chemical biomarker background in Prince William Sound (PWS), Alaska, site of the 1989 Exxon Valdez oil spill. Different authors have attributed the sources to various proportions of coal, natural seep oil, shales, and stream sediments. The different probable bioavailabilities of hydrocarbons from these various sources can affect environmental damage assessments from the spill. This study compares two different approaches to source apportionment with the same data (136 PAHs and biomarkers) and investigate whether increasing the number of coal source samples from one to six increases coal attributions. The constrained least-squares (CLS) source allocation method that fits concentrations meets geologic and chemical constraints better than partial least-squares (PLS) which predicts variance. The field data set was expanded to include coal samples reported by others, and CLS fits confirm earlier findings of low coal contributions to PWS.

Introduction There is general agreement (1-3) that a substantial background of petrogenic hydrocarbons, derived from natural sources, is present in the benthic sediments of Prince William Sound, Alaska (PWS; Figure 1). These background hydrocarbons were present in PWS benthic sediments long before the 1989 Exxon Valdez oil spill in PWS. There is, however, substantial disagreement among investigators about the relative contributions of coal, natural seep oils, and shales to the petrogenic background. Quantifying the contributions that each of these sources makes is important because assumed differences in their hydrocarbon bioavailability potentially affect Exxon Valdez oil spill impact studies. Three groups of investigators have been involved. One group (1, 2) identified the presence of the natural background and tied it to the organic component of eroding rock * Corresponding author phone: (281) 558-1740; fax: (281) 4932181; e-mail: [email protected]. † W. A. Burns Consulting Services LLC. ‡ University of Wales - Bangor. § AEB Services, LLC. | Exponent. ⊥ Bowdoin College. # Data Analysis Group. 10.1021/es0603094 CCC: $33.50 Published on Web 10/03/2006

 2006 American Chemical Society

formations and natural oil seeps located along a ∼300 km section of the Gulf of Alaska (GOA) coast east of PWS. These hydrocarbons are associated with fine particles carried by rivers and streams to the GOA and subsequently transported into PWS by the westward-flowing Alaska Coastal Current (ACC). The quantitative studies (4, 5) by this first group on the sources of the background use a constrained least-squares (CLS) technique to match the PAH and chemical biomarkers of PWS benthic samples with a blend of source samples. The technique initially applied (4) was limited by software constraints (since removed) to 18 sources and 38 analytes. Seep oils, shales, and GOA stream sediments were found to be the dominant contributors to the natural petrogenic hydrocarbon background in PWS. High maturity coal such as that found in the large Bering River Coalfield contributed only about 1% of background. Total organic carbon (TOC) calculated from their CLS source allocation results agrees with measured TOC values (5). They further note that some of the eroding shales are in the oil-generation stage of thermal maturity (i.e., a measure of the duration and temperature of burial over geologic time) and naturally contain extractable hydrocarbons (4). Such materials may be bioavailable (1). A second group of investigators (3) suggests that the Bering River Coalfield might be the major source of the background hydrocarbons, but later argue that the natural petrogenic background comes from a range of unidentified, nonbioavailable, GOA coals (6, 7). A subsequent paper (8) identifies thin, GOA coal beds of low to moderate thermal maturity, as potential sources but does not present quantitative source analysis or TOC balances. They (8) also conclude that PAH distributions in benthic GOA sediments are almost identical, and state that this could mean that the 38 analytes chosen by ref 4 might be inadequate to accurately resolve their 18 relatively similar sources and that there might be a significant risk of multicolinearity with the model in ref 4. These authors (8) argue that the use of 18 sources “increases the risk that impressive fits of the data are spuriously achieved by means of an unrealistic combination of sources.” Finally, they argue (8) that the Bering River coal used by refs 4, 5 does not adequately represent native coals in the source area. Another investigator (9) uses partial least-squares (PLS) to determine source allocation based on benthic data gathered in previous work (4). The PLS model (9) is able to fit 136 analytes from the dataset. This investigator (9) reports different source allocations for the same dataset used in the earlier study (4). The present paper uses an expanded version of the CLS model, one capable of modeling the 136 analytes used in paper (9). We show that there are fundamental differences in the CLS and PLS techniques that lead to the different results on the same data. We also show that the use of more analytes and additional native coal samples provided by Van Kooten(8) do not substantially change earlier findings (4, 5) that coal is a relatively small contributor to the PWS natural petrogenic background. In addition, we also show in equation form how to calculate the important TOC constraint from source allocation results.

Methods Samples. This study determines the hydrocarbon sources of 29 sediment samples (Figure 1) including four PWS benthic VOL. 40, NO. 21, 2006 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

6561

FIGURE 1. Target sample and stream locations in Prince William Sound and the Gulf of Alaska. Stream designations: (A) Copper River; (B) Bering River; (C) Campbell River; (D) Kaliakh River; (E) Duktoth River; (F) North Yakataga River, (G) Yakataga Streams (White River, Poul Creek, Munday Creek, Munday Creek tributary, Johnston Creek, Big River, Little River, Priest River); and (H) Fountain Stream (Malaspina glacial flour runoff). samples, 18 GOA benthic samples, and seven GOA beach or sandbar samples described elsewhere (4, 5). To compare CLS and PLS, we employ CLS source allocation runs that use either 18 or 30 sources; 18 to match the previously published CLS results and 30 to represent a more complete set of possible sources. These sources come from the dataset used by the earlier CLS and PLS studies (4, 5, 9). The 18 sources include sediment from seven streams (Copper, Kaliakh, Duktoth, North Yakataga, Big, Fountain, and Johnston above the seep), five Katalla and Yakataga seep oils, Katalla organic shale, Poul Creek shale, two Yakataga shales, and Queen Vein coal and bat (an organic shale impurity found in coal beds). The 30 sources include the 18 above plus sediments from White, Little, Priest, Campbell rivers and Munday and Munday Creek tributary above the seep. An oil stained rock, two Point Hey shales, natural coke from the Bering River Coalfield, and two pyrogenic samples (wood soots) complete the 30. To determine the effect of additional coal samples on CLS source allocation we use the 18 and 30 source samples above plus five low maturity coal samples and one organic shale sample from the Kosakuts River area (Kulthieth Formation) kindly provided by Van Kooten (8) and analyzed by our laboratory protocols. The Kosakuts is a tributary of the Kaliakh River (Figure 1). Exxon Valdez spill oil was not used as a source because all samples were either from outside of the spill path or from age-dated core slices that predated the spill. Primary sources include the seep oils, shales, coals, bat, coke, and the wood soots. Fireplace soots were sampled to approximate wood soot from forest fires, villages, and campfires, all of which are dominated by 4- to 6-ring PAH. Stream sediments are composite samples that integrate eroded particles and seep oil residues from the many individual sedimentary beds and seeps drained by GOA streams. The ultimate resolution of primary source contributions is probably impossible since it would require that every primary source be sampled. Many upstream primary sources in this area are buried beneath massive glaciers and thus inaccessible. Laboratory Analysis. Samples were analyzed for 53 polycyclic aromatic hydrocarbon (PAH) analytes and 88 6562

9

ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 40, NO. 21, 2006

chemical biomarkers using GC/MS-SIM methods (4). A list of the 136 PAH and biomarker analytes used in this study is given in ref 9. We quantified C19-C35 terpanes (m/z 191), C21-C29 regular steranes (RRR and Rββ) (m/z 217, 218), C27C29 diasteranes (m/z 259), and C20-C28 triaromatic steroids (m/z 231). Saturated biomarkers were quantified using 5β(H)-cholane. PAH and aromatic biomarkers were quantified using d10-phenanthrene. Analyte concentrations were quantified as ng‚g-1 dry weight, and the quantified PAH and biomarker concentrations are used to determine the source allocation. A base set of 134 analytes were measured for each sample, and other analytes were added for some samples during the course of the analysis as the contract laboratory identified additional peaks. To permit comparison between the PLS and CLS methods, we use the same 136 analytes (i.e., the 134 common ones plus two others) used in the PLS paper. Ordinarily, only source diagnostic analytes whose concentrations (or lack thereof) were measured for all samples should be used in source analysis. In this case, the extra analytes made little difference to results. Some analytes utilized in our source allocation runs are below laboratory reporting limits in some samples as it is difficult to put together a set of sufficiently source-diagnostic analytes that does not include such analytes in at least some samples. TOC was measured using the LECO TOC combustion method (5) Constrained Least-Squares. The CLS method determines the sources of the hydrocarbons in a given sample by adjusting the relative contributions of potential sources until S, the sum of squared residuals (the differences between the measured and calculated concentrations of the analytes) in eq 1, is minimized, subject to eqs 2 and 3 (constraints on permissible source fractions). na

S)

∑ i)1

na

(ei)2 )

∑ i)1

ns

∑ (s

[loge (

i,jxj/Aj))

j)1

- loge (di/Asamp)]2 (1)

Here na refers to the number of analytes and ns to the number of potential sources. A is the sum of the analytes selected for this application. The subscript samp refers to the offshore, downstream, or beach sample whose hydrocarbon sources are being determined. si,j is the concentration of the ith analyte

of source j, and di is the concentration of the ith analyte of the sample. xj is the fraction of the sample’s analytes that are contributed by source j. The xj values are constrained by: ns

∑x ) 1 j

(2)

0 e xj e 1

(3)

j)1

and

The earlier study (4) used a CLS solution algorithm (10) that was limited to 18 sources and 38 analytes by software constraints. Those constraints have been eliminated and the present paper uses a version of the algorithm capable of handling much larger numbers of sources and analytes (maximum limit not determined). This algorithm fits (i.e., includes in least-squares calculations) only those analytes detected in the target sample, even though the analytes might be detected in source samples. We treat nondetected analytes in the source samples as zeros, but we recommend that the effect of this approach be checked, particularly if there are zeros in major contributing sources. Partial Least-Squares. The PLS study (9) used the SIMCAP, version 8.1 (or later) computer program from Umetrics and the following procedure. A small constant equal to approximately half of the limit of detection was added to each analyte to permit log transformation of nondetected analytes. A Principal Components Analysis (PCA) was conducted, followed by a PLS study. PCA identified orthogonal data projections (principal components 1 and 2) that explain much of the variance in the data. Five “source types” (oil, coal, shale, and two groups of river sediments) were chosen based on aggregations seen in the PCA scores plots. Source type selection was based on the first two principal components and some individual samples were excluded (9). PLS was used to determine loadings factors for each source type and the amount of variance in target samples explained by each source-type loadings signature. If variance explained by the various source types totals to more than 100%, variance explained is then normalized to 100%. The PLS procedure quantifies the percentage of the variance that can be predicted by each model. To obtain percent contribution, CLS fits relative concentrations of all individual analytes rather than variance and is, therefore, more consistent with material balance constraints on individual analytes. Different sources may explain the same amount of variance but have different degrees of fit to the relative concentrations. Thus, it is quite possible that PLS and CLS yield different results. For comparison with CLS results, PLS results for percent variance predicted are treated in this paper as percent contribution. As employed above (9), the PLS procedure returns results for percent variance predicted by the identified source types, not by individual source samples. Each source signature is complied from a group of samples identified as similar due to their aggregation in the PCA scores plot. By contrast, CLS determines the contributions of individual source samples which can then be summed in various groupings to show contributions by source type. Organic Carbon. Organic carbon offers an independent check on the reasonableness of source allocations in sediments containing little recent organic matter (5). TOC can be calculated from source allocation results and measured

TOCs of contributing sources. We compare calculated and measured TOCs of target samples to help rule out incorrect source allocations. Organic carbon in sediment is carried by recent plant and animal debris and particles that carry petrogenic and pyrogenic hydrocarbons (e.g., kerogen, coal, bitumen, weathered oil particulates, oil sorbed to clay minerals, etc.). In sediments that contain little recent plant and animal debris, such as the glacial runoff dominated sediments of our study area, the organic carbon of a sediment sample is essentially the sum of the organic carbon carried by the contributing petrogenic and pyrogenic sources. Source allocation determines how much hydrocarbon each source contributes to the target sediment sample, and that amount multiplied by the organic-carbon-to-hydrocarbon ratio of that source determines how much organic carbon that source contributes to the target sample. Thus, for such target samples, TOC may be expressed as the sum over all sources of the product of (a), the ratio of source sample’s TOC to that source sample’s total hydrocarbon analytes, times (b), the total hydrocarbon analytes contributed to the target sample by that source: ns

TOCsamp )



ns

[(a)(b)] )

j)1

∑ j)1

na

[(TOCj/

∑ i)1

na

si,j)(xj

∑ d )] i

(4)

i)1

Here TOCsamp is the calculated organic carbon content of the sediment sample, TOCj is the measured organic carbon content of source j, and the other terms are as defined above.

Results and Discussion CLS vs PLS. As shown in Figure 2A, average source allocation for four PWS samples (samples 1-4) is clearly different for CLS and PLS even though the same database is used to generate the source allocation. The PLS model uses 136 analytes and groups sources into five source types before source allocation (9). CLS results are shown for both 38 analytes and 136 analytes. Both CLS cases are for 18 sources, but to permit comparison with PLS, sources are summed into the same five source types used by PLS. PLS found higher proportions in PWS sediments than CLS of hydrocarbons from shale, high maturity coal, and Yakataga stream sediments. Average CLS source allocation results are generally similar (Figure 2A) for the 18-source, 136-analyte case and the 18-source, 38-analyte case reported earlier (4, 5). Figure 2B shows the 30-source, 136-analyte CLS result for high maturity coal contributions to 29 PWS and GOA benthic, beach and sandbar samples. These 29 target samples were collected between PWS on the west and Icy Bay on the east (see Figure 1). Sample 14 consisted of concentrated stranded coal particles from the top of a Bering River sandbar located downstream from the Bering River Coalfield. CLS source analysis confirms that the sample hydrocarbons are dominated by high maturity coal, a finding consistent with the vitrinite reflectance data previously reported for this sample (4). Figure 2C shows CLS results for hydrocarbons from Yakataga streams. In keeping with the small size of these streams, their impact on benthic sediments is generally small and local. In contrast, PLS results (9) for high maturity coal and Yakataga streams (Figure 2D) appear to be almost independent of location. This may be the result of using only five source types or grouping sources whose locations are widely dispersed geographically. The PLS procedure (9) interprets the geographically dispersed rivers as having a similar enough geochemical signature that they can be grouped together for VOL. 40, NO. 21, 2006 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

6563

FIGURE 2. Comparison of CLS and PLS results. (A) CLS and PLS source allocation results for Prince William Sound benthic sediments, (B) CLS results for the geographic distribution of coal, (C) Yakataga stream hydrocarbons, and (D) PLS distributions of coal and Yakataga stream hydrocarbons in benthic and beach sediments in PWS and GOA. Contributions of bat in this paper are added to coal contributions. source allocation purposes; CLS recognizes significant differences in analyte distributions among them and treats them as separate sources. As a further test, we mixed together, mathematically, a hypothetical 50:50 mixture of two known sources, then determined the sources of the resulting mixture composition with CLS and PLS using three possible sources, the two that were actually the sources and a third fairly similar source. PCA, the first step used in the process employed by paper (9), identified the three sources as distinct and different, but PLS incorrectly attributed part of the 50:50 mixture to the non-contributing third source. In contrast, CLS correctly identified the two contributing sources and the amount of their contributions and correctly found that the third source did not contribute. We suspect that the reason PLS failed in this test case was that it determines the variance explained by the different sources rather than the contributions of the different sources. The amount of variance explained does not necessarily equate to the amounts contributed by various sources. Variance is a statistical construct; mixture compositions are governed by material balance, not variance. Geographic Consistency. To further illustrate the geographic consistency of CLS results for this dataset, we show in Figure 3 the calculated contributions to Katalla Bay and Controller Bay sediment samples of three sets of sources: (1) high maturity coal, which comes from the Bering River 6564

9

ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 40, NO. 21, 2006

Coalfield, (2) Point Hey shale, which outcrops on the mainland north of Kanak Island, and (3) the Duktoth and North Yakataga Rivers, streams which lie east and upcurrent of the map area. The contributions of other sources to these samples are not shown. These results are for a 30-source, 136-analyte CLS run. CLS results for sediment samples (5-14) that lie along and near the northern mainland shore and the drainage channel of the Bering River (which runs between Kanak Island and the Point Hey shale area) show coal and Point Hey shale contributions but little hydrocarbons from the Duktoth and North Yakataga Rivers, streams which merge before entering the GOA east of the area shown in Figure 3. CLS results for the southern set of sediment samples (15-19) show contributions of the Duktoth and North Yakataga Rivers, but little high maturity coal. The above findings are consistent with the locations of these sources and the probable currents in the area. PLS results (9) for high maturity coal are shown in Figure 2 for samples 5-8 and 12-14. PLS high maturity coal contributions are about 25% or less, even for sample 14, which consisted largely of stranded coal particles washing down the Bering River. No PLS results are available for the other samples in Figure 3 or separately for Point Hey shale or the Duktoth and North Yakataga Rivers. Van Kooten et al. (8) argue that the PAH profiles of GOA benthic sediments are “almost identical.” The relatively uniform source allocations produced by PLS (9) are consistent with such a view. However, by treating sources individually rather than grouping geographically and chemically disparate sources and by using biomarkers in addition to PAH, CLS is able to resolve fingerprint differences and yield source allocations consistent with the geographic locations of sources. In contrast, Van Kooten et al. (8) present only a few biomarkers and focus mainly on the PAH. Recent analyses with PLS have also shown that, in some instances, less rather than more analytes can give better source discrimination (12) and improvements to the SIMCA-P program allow signatures to be developed on the basis of a single sample. In light of these results, more investigations are required in order to determine the usefulness of PLS in this complex, multi-source environment. In any event, a sufficient number of source diagnostic analytes should be used to permit accurate allocation of potential sources that can exhibit a wide range of combinations of thermal maturity and source matter type. Additional Coal Samples. Figure 4 compares CLS source allocation results and the average error in calculated TOC for individual samples made with and without six low to moderate thermal maturity Kosakuts samples (five coals and one organic shale). The Kosakuts samples increase the amount of the PWS background attributed to coals (3, 6, and 11% for Cases D, E, and F, respectively, including the Bering River source samples) but reduce the amount attributed to high maturity coal. Thus, the earlier finding (4, 5) that high maturity coal contributes little PAH and biomarkers to the PWS background is still valid. Even with the addition of the Kosakuts samples, coal is still a smaller contributor to the background than river sediment, oil, or shale. This is consistent with our earlier published results (4, 5, 9). The TOC error results of Figure 4B suggest that cases A and D, which are based on a subset of analytes that differentiate well between our sources, may be the most accurate of the six cases. The cases involving 136 analytes add large numbers of low concentration biomarkers whose concentrations are below method detection limits. This introduces analytical noise and uncertainty into source allocation results but adds a number of thermal maturity and source matter indicators.

FIGURE 3. CLS results for Katalla Bay and Controller Bay area sediment samples. Each sediment sample insert shows the percentage contributions from Point Hey shale and coal on the left and Duktoth and North Yakataga Rivers on the right.

FIGURE 4. Effect of Kosakuts coal and Kosakuts organic shale samples on average results for four PWS samples. (A) CLS results with and without six Kosakuts coal samples as sources. Case A: 18s, 38a (s ) sources, a ) analytes). Case B: 18s, 136a. Case C: 30s, 136a. Case D: 24s, 38a. Case E: 24s, 136a. Case F. 36s, 136a. (B) Average individual sample error in TOC calculated from CLS results (error ) absolute value of difference between calculated and measured TOC divided by measured TOC; average error of the four PWS samples is shown for the various cases). Van Kooten et al. (8) argue that the use of 18 sources increases the risk that impressive fits can be obtained by using an unrealistic combination of sources. However, CLS results show that many of the 18 sources do not contribute to the PWS samples. The 18 sources were originally chosen to model with one set of sources the subtidal samples

along the entire coast from PWS to Icy Bay (4, 5). Many of the chosen sources contribute to benthic sediments near their GOA origins but not to PWS in appreciable amounts. In case A above, only six of the 18 sources are found by CLS to contribute to PWS sample 1, only the same six to PWS sample 2, only five (out of the same six) to PWS sample 3, and only seven to PWS sample 4 (including five sources in common with PWS samples 1, 2, and 3). Some might argue that only PAH should be considered in this source allocation because many of the biomarker concentrations fall below laboratory reporting limits and thus introduce some uncertainty into source allocation results. A 36-source CLS run using only 52 PAH (no perylene since it is not source diagnostic) found an average of less than 3% coal contribution to the four PWS samples and oil contributions that averaged 25%. Average TOC error averaged 47% for this run, however, with calculated TOC being greater than measured TOC in all cases. Analyte uncertainty and confidence intervals are discussed further in the Supporting Information. Fingerprint Comparisons. Cross plots of sample analyte concentrations illustrate why some source contributions are small while others are large. Figures 5A-5E compare analyte distributions from several source samples to that of PWS sample 3. The Kosakuts samples and the Bering River Coalfield samples are deficient in biomarker concentrations (i.e., the steranes and triterpanes) relative to the PWS sample, as exemplified by Figures 5A and B. Thus, these coal samples cannot be large contributors to PWS. Katalla organic shale, the likely petroleum source rock for Katalla seep oil, provides a better match to the PWS sample than the coals (Figure 5C). Other shale samples along the GOA coast vary from coal-like distributions (low biomarker concentrations than PWS) to weathered oil-like distributions VOL. 40, NO. 21, 2006 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

6565

FIGURE 5. Comparison of analyte concentrations normalized to total analytes. Points represent individual analytes. PWS Sample 3 is compared to: (A) Kosakuts Coal 8, (B) Queen Vein Coal, (C) Katalla Organic Shale (petroleum source rock), (D) Katalla Seep Oil, (E) Malaspina Glacial Flour, (F) calculated CLS match for 30 sources, 136 analytes, (G) calculated CLS match of 18 sources, 38 fit analytes, and (H) concentrations of 98 unfit analytes determined from source allocation of (G).

(greater biomarker concentrations than PWS) reflecting differences in organic matter type, stages of oil generation, oil/bitumen content, and weathering. Katalla seep oil (Figure 5D) shows evidence of surface weathering (decreased PAHs and increased relative concentrations of biomarkers). Figure 5E shows that Malaspina glacial flour, a sample collected in stream runoff from the Malaspina glacier, agrees well with PWS sample 3. Figures 5F-G shows how well calculated CLS fingerprint matches compare to PWS. Figure 5H shows that analytes not used in the 38-analyte fit of Figure 5F agree well with PWS, suggesting that the fit of Figure 5F is fairly robust. Exxon Valdez Oil. Finally, using CLS we tested whether our four PWS samples contained any fresh North Slope oil (NS) or weathered Exxon Valdez oil (EV). We found no NS oil and less than 0.05% EV oil using 38 analytes and 4% NS oil and 0.5% EV oil using 136 analytes. This is consistent with earlier interpretations (1, 2, 10) and the locations and/or the prespill nature of these PWS samples. The apparent amounts of NS and EV oils in the 136-analyte case might result from data scatter in the many analytes of the 136-analyte case 6566

9

ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 40, NO. 21, 2006

whose concentrations were below method detection limits. If so, more accurate results might be obtained using fewer analytes and more source-indicative ones than by using all 136 analytes. Implications for Bioavailability. Oil remnants and soot have low bioavailability (13), and many eroding GOA shales contain extractable hydrocarbons that could be bioavailable. Thus, hydrocarbons in the PWS/GOA natural background are probably somewhat bioavailable to benthic organisms and bottom fish and may need to be taken into account in benthic damage assessment studies. In support of these statements, natural sediment hydrocarbons were found to be the probable cause of low concentrations of bile fluorescent aromatic contaminants (FACs) and liver ethoxyresorufin O-deethylase (EROD) activities in several PWS/GOA fish species (14). Comparable levels of these hydrocarbon exposure indicators were found in fish from spill areas of PWS, non spill areas of PWS, and the GOA far upcurrent of the spill area. The sediments of all three of these areas contained high levels of natural petrogenic hydrocarbons. Lower indicator levels were found in fish from an unoiled control site located far away from the high natural background area (15).

Acknowledgments ExxonMobil provided financial support to authors Bence, Burns, Boehm, Page, Brown, and Parker for the sample collection, chemical analysis, and/or data interpretation associated with this paper.

Supporting Information Available Concentrations of 136 analytes and TOC in target and source samples, analyte variability studies, confidence intervals, and details of sample/analyte run combinations. This material is available free of charge via the Internet at http:// pubs.acs.org.

Literature Cited (1) Page, D. S.; Boehm, P. D.; Douglas, G. S.; Bence, A. E. Identification of hydrocarbon sources in the benthic sediments of Prince William Sound and the Gulf of Alaska following the Exxon Valdez oil spill. In Exxon Valdez Oil Spill: Fate and Effects in Alaskan Waters, ASTM STP 1219; P.G. Wells, J.N. Butler and J.S. Hughes, Eds; ASTM: Philadelphia, PA, 1995; pp. 41-83. (2) Page, D. S.; Boehm, P. D.; Douglas, G. S.; Bence, A. E.; Burns, W. A.; Mankiewicz, P. J. The natural petroleum hydrocarbon background in subtidal sediments of Prince William Sound, Alaska, USA. Environ. Toxicol. Chem. 1996, 15, 1266-1281. (3) Short, J. W.; Heintz, R. A. Identification of Exxon Valdez oil in sediments and tissues from Prince William Sound and the Northwestern Gulf of Alaska based on PAH weathering. Environ. Sci. Technol. 1997, 31, 2375-2384. (4) Boehm, P. D.; Page, D. S.; Burns, W. A.; Bence, A. E.; Mankiewicz, P. J.; Brown, J. S. Resolving the Origin of the Petrogenic Hydrocarbon Background in Prince William Sound, Alaska. Environ. Sci. Technol. 2001, 35, 471-479. (5) Boehm, P. D.; Burns, W. A.; Page, D. S.; Bence, A. E.; Mankiewicz, P. J.; Brown, J. S., Douglas, G. S. Total Organic Carbon, An Important Tool in a Holistic Approach to Hydrocarbon Source Fingerprinting. Environ. Forensics 2002, 3(3/4), 243-250. (6) Short, J. W.; Kvenvolden, K. A.; Carlson, P. R.; Hostettler, F. D.; Rosenbauer, R. J.; Wright, B. A. Natural hydrocarbon background in benthic sediments of Prince William Sound, AlaskasOil vs. Coal. Environ. Sci. Technol. 1999, 33, 34-42.

(7) Hostettler, F. D.; Rosenbauer, R. J.; Kvenvolden, K. A. PAH refractory index as a source discriminant of hydrocarbon input from crude oil and coal in Prince William Sound, Alaska. Org. Geochem. 1999, 30, 873-879. (8) Van Kooten, G. K.; Short, J. W.; Kolak, J. J. Low-maturity Kulthieth Formation coal: A possible source of polycyclic aromatic hydrocarbons in benthic sediment of the Northern Gulf of Alaska. Environ. Forensics 2002, 3(3/4), 227-241. (9) Mudge, S. M. Reassessment of the hydrocarbons in Prince William Sound and the Gulf of Alaska: Identifying source using partial least squares. Environ. Sci. Technol. 2002, 36, 23542360. (10) Burns, W. A.; Mankiewicz, P. J.; Bence, A. E.; Page, D. S.; Parker, K. R. A principal-component and least-squares method for allocating polycyclic aromatic hydrocarbons in sediment to multiple sources. Environ. Toxicol. Chem. 1997, 16, 1119-1131. (11) Wold, S.; Ruhe, A.; Wold, H.; Dunn, W. J. The colinearity problem in linear regression: The partial least squares approach to generalized inverses. SIAM J. Sci. Stat. Comp. 1984, 5, 735-743. (12) Mudge, S. M.; Birch, G. F.; Matthai, C. The Effect of grain size and element concentration in identifying contaminant sources. Environ. Forensics 2003, 4, 305-312 (13) Oil in the Sea III: Inputs, Fates, and Effects; National Academy of Sciences: Washington, D.C., 2003; p 128. (14) Huggett, R. J., Stegeman, J. J.; Page, D. S.; Parker, K. R.; Woodin, B.; Brown, J. S. Biomarkers in fish from Prince William Sound and the Gulf of Alaska, 1999-2000. Environ. Sci. Technol. 2004, 38, 4928-4936. (15) Horn, T.; Varanasi, U.; Stein, J. E.; Sloan, C. A.; Tilbury, K. L.; Chan, S.-L. Assessment of the exposure of subsistence fish to aromatic compounds after the Exxon Valdez oil spill. In Proceedings of the Exxon Valdez Oil Spill Symposium, American Fisheries Symposium 18; Rice, S. D., Spies, R. B., Wolf, D. A., Wright, B. A., Eds.; American Fisheries Society: Bethesda, MD, 1996; p 856.

Received for review February 11, 2006. Revised manuscript received August 30, 2006. Accepted September 5, 2006. ES0603094

VOL. 40, NO. 21, 2006 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

6567