Environ. Sci. Technol. 2002, 36, 2354-2360
Reassessment of the Hydrocarbons in Prince William Sound and the Gulf of Alaska: Identifying the Source Using Partial Least-Squares STEPHEN M. MUDGE* School of Ocean Sciences, University of WalessBangor, Menai Bridge, Anglesey LL59 5EY, U.K.
There has been considerable exchange in the literature regarding the source of the background hydrocarbons in Prince William Sound and the Gulf of Alaska. Page and coworkers suggest oil-based sources while Short and coworkers suggest coal. The multivariate statistical methodology of partial least-squares (PLS) has been used to reassess the percentage contribution of coal, seep oil, shales, and rivers to the hydrocarbon loading in the Gulf of Alaska. Data have kindly been provided by Short (NOAA) and Page (Bowdoin College, for Exxon), and these have been analyzed using selected sites as sources in order to develop signatures. These signatures are based on 40 and 136 compounds, respectively, and include the PAH and terpane/sterane biomarkers in the case of the Exxon data. The principal components describing these sources are then fitted to the data for other sites around the Prince William Sound (PWS) and Gulf of Alaska (GoA) to determine the proportion of the variability described by each source. Using the Exxon data, a mixed source of coal, seep oil, eroding shales, and rivers (1 and 2) sources described ∼13%, 18%, 24%, 26%, and 20%, respectively, of the variance in PWS and GoA data. The rivers 1 signature was very similar to that of coal, while rivers 2 was more similar to the eroding shales. New coal data (Short, unpublished work) also indicated considerable overlap with the Exxon seep oil. With the NOAA data, spatial plots of the explained variance indicate that the prespill background has a wide range of explained fits. There is considerable overlap in the signatures developed from the data, and Coomans’ Plots identify those compounds which are the most diagnostic: for the Exxon seep oil signature, naphthalene and methyl- and dimethylnaphthalene are the best markers, whereas for the NOAA prespill background, Exxon coals and shales are best defined by the larger PAHs such as benzo[ghi]perylene. The evidence suggests mixed sources whose contributions vary significantly across the sampling area.
much work published on the cleanup and potential longterm fate of the oil (1, 2). However, an outstanding issue does exist with regard to the origin of the background hydrocarbons present within Prince William Sound and the Gulf of Alaska (GoA), generally. There are two schools of thought on this matter. Short et al. from the NOAA Auke Bay Laboratory and USGS cite evidence based on the presence of coal on nearby beaches and the extensive coal measures between the Bering River and Ice Bay that these background (prespill) hydrocarbons come from such deposits (3). In contrast, Page and co-workers from Bowdoin College, Arthur D. Little, and the Exxon Corporation believe that they originate from natural oil seeps in the same regions around the eastern shores of the GoA (4, 5), although both groups now recognize the importance of eroding source rocks. The motivation for their work revolves around the likely toxicity of the prespill sediments based on their PAH content. This may relate to the level of fine imposed on Exxon. If the background hydrocarbons consist principally of oil-derived materials, the fine would be less than if coal was the source. Several papers have been published presenting evidence for each hypothesis which have then been challenged in replies. Much of the work revolves around the choice of biomarkers, ratios between selected PAHs, their refractory index, and TOC/mass balance considerations. Recently, Page and coworkers have used some fitting algorithms to quantify the amount of each source present (6). Multivariate statistical techniques have become available in recent years which allow for signatures to be determined from source samples which can then be quantitatively extracted from environmental samples. One such method is Projection to Latent Structures by means of partial leastsquares analysis (PLS). This techniques was developed by Wold et al. (e.g., ref 7) and has evolved into a powerful tool used principally by the pharmaceutical industries in determining quantitative structure activity relationships (QSARs). However, a new use is environmental forensics, and this has been demonstrated by Yunker at al. (8) and Mudge and Seguel (9). In an attempt to resolve the issue of source identification and partitioning outlined previously, data have been obtained from both Page and co-workers (Exxon data) and Short (NOAA data). These data have been chosen because they have been used to develop arguments for their respective points of view. They have also been “validated” as previously published results in peer-reviewed journals. Initially, the data sets have been used to generate signatures only for the oil and coal components in line with the original arguments. The extent to which each of these signature explains the variance seen in the PWS and GoA samples was assessed. An extension to this approach is to use PCA to identify how many different sources there may be and to expand the number of PLS signatures to encompass other possible sources, such as eroding source rocks. These may then be combined to determine the relative contribution from each and also to quantify the overlap between similar signatures.
Introduction
Methods
Following the grounding and loss of oil from the Exxon Valdez tanker in Prince William Sound (PWS), Alaska, there has been
The data provided by Short comprised 1168 samples analyzed for 40 compounds, whereas that of Page had 47 samples with 136 compounds. One of the major chemical differences in the data sets is the range of compounds analyzed; the Exxon data have the traditional terpane/sterane biomarkers
* Corresponding author phone and fax: +44 1248 382879; e-mail:
[email protected]. 2354
9
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 36, NO. 11, 2002
10.1021/es015572d CCC: $22.00
2002 American Chemical Society Published on Web 05/03/2002
TABLE 1. Compounds Used and Their Abbreviations compound
abbrev
compound
abbrev
compound
abbrev
naphthalene C1-naphthalenes C2-naphthalenes C3-naphthalenes C4-naphthalenes acenaphthylene acenaphthene biphenyl fluorene C1-fluorenes C2-fluorenes C3-fluorenes anthracene phenanthrene C1-phenanthrenes-1 C1-phenanthrenes-2 C1-phenanthrenes-3 C1-phenanthrenes-4 C1-phenanthrenes-5 C1-phenanthrenes/ anthracenes C2-phenanthrenes/ anthracenes C3-phenanthrenes/ anthracenes C4-phenanthrenes/ anthracenes dibenzothiophene C1-dibenzothiophenes-1 C1-dibenzothiophenes-2 C1-dibenzothiophenes-3 C1-dibenzothiophenes C2-dibenzothiophenes C3-dibenzothiophenes fluoranthene pyrene C1-fluoranthenes/pyrenes C2-fluoranthenes/pyrenes C3-fluoranthenes/pyrenes benzo[a]anthracene chrysene C1-chrysenes-1 C1-chrysenes-2 C1-chrysenes-3 C1-chrysenes-4 C1-chrysenes C2-chrysenes C3-chrysenes C4-chrysenes
N0 N1 N2 N3 N4 ACE ACL Bph F0 F1 F2 F3 AN P0 P1A P1B P1C P1D P1E P1F P2 P3 P4 D0 D1A D1B D1C D1D D2 D3 FL PY FP1 FP2 FP3 BaA C0 C1A C1B C1C C1D C1E C2 C3 C4
benzo[b]fluoranthene benzo[k]fluoranthene benzo[e]pyrene benzo[a]pyrene perylene indeno[1,2,3,-c,d]pyrene dibenzo[a,h]anthracene benzo[g,h,i]perylene T0-C19 diterpane T1-C20 diterpane T2-C21 diterpane T3-C22 diterpane T4-C23 diterpane T5-C24 diterpane T6-C25 diterpane T6a-C24 tetracyclic terpane T6b-C26 tricyclic(S)terpane T6c-C26 tricyclic(R)terpane T7-C28 tricyclic triterpane T8-C28 tricyclic triterpane T9-C29 tricyclic triterpane T10-C29 tricyclic triterpane T11-trisnorhopane (TS) T12-trisnorhopane (TM) T13-trisnorhopane T13a-C29,C30 bisnorhopane T14-bisnorhopane T14a-C28,C30 bisnorhopane T14b-C29,C25 norhopane T15-norhopane T16-neonorhopane T17-normoretane T18-oleanane T19-hopane T20-moretane T21-homohopane T22-homohopane T22a-gammacerane T23-homohopane T24-homomoretane T25-diploptene T26-bishomohopane T27-bishomohopane T28-bishomomoretane T29-homohopane
BbF BkF BeP BaP Per IP DA BgP T0 T1 T2 T3 T4 T5 T6 T6a T6b T6c T7 T8 T9 T10 T11 T12 T13 T13a T14 T14a T14b T15 T16 T17 T18 T19 T20 T21 T22 T22a T23 T24 T25 T26 T27 T28 T29
T30-trishomohopane T31-trishomohopane T32-tetrakishomohopane T33-tetrakishomohopane T34-pentakishomohopane T35-pentakishomohopane C20-TAS (A1) C21-TAS (A2) C26-TAS (20S) (A3) C26,C27-TAS (A4) C27-TAS (20R) (A5) C28-TAS (20S) (A6) C28-TAS (20R) (A7) S1-pregnane S2-pregnane S4-diacholestane S5-diacholestane S6-diacholestane S7-diacholestane S8-methyldiacholestane S10-methyldiacholestane S11-methyldiacholestane S12-cholestane S14-cholestane (20R) S15-cholestane (20S) S17-cholestane (20R) S18-ethyldiacholestane S19-ethyldiacholestane S20-methylcholestane S22-methylcholestane (20R) S23-methylcholestane (20S) S24-methylcholestane S25-ethylcholestane S26-ethylcholestane (20R) S27-ethylcholestane (20S) S28-ethylcholestane S29-C30 cholestane (R) S30-C30 cholestane(s) D1-diasterane D2-diasterane D3-diasterane D3a-diasterane D4-diasterane D4a-diasterane D5-diasterane D6-diasterane
T30 T31 T32 T33 T34 T35 A1 A2 A3 A4 A5 A6 A7 S1 S2 S4 S5 S6 S7 S8 S10 S11 S12 S14 S15 S17 S18 S19 S20 S22 S23 S24 S25 S26 S27 S28 S29 S30 D1 D2 D3 D3a D4 D4a D5 D6
as well as the PAHs, whereas the NOAA data set only has the PAHs. A full list of the compounds used in both the X and Y blocks and their abbreviations used in the figures can be found in Table 1. In essence, PLS performs principal components analysis (PCA) on a set of data which is defined as the signature (10, 11). This data set, called the X block, ideally will be a pure source sample but can be made up of environmental samples which have a high proportion of a single source of organic matter such as coal deposits or sediments from unoiled areas. Because the data come from the same source, although the concentrations may vary, PCA will generate a principal component 1 (PC1) that explains most of the variance in the data, typically >90%. This projection or vector in n dimensional space, where n is the number of chemical compounds analyzed, can be described by a series of loading factors on each compound; those compounds which have a major impact on PC1 will have high loadings (either positive or negative), whereas those compounds which are relatively unimportant and therefore do not have a major influence on the data will have values close to zero. PC2 is fitted orthogonal to the first component, so there is no overlap between the two. Once the first two PCs have been elucidated, their
projection can be described in terms of the two sets of loadings.
PC1 ) a1[compound1] + b1[compound2] + c1[compound3] ... PC2 ) a2[compound1] + b2[compound2] + c2[compound3] ... where a-c are the individual loadings on PC1 and PC2. These projections, which represent the signature defined in terms of the chemical compounds used, can now be applied to the environmental data (Y block). The amount of variance explained by each X block signature can be quantified. This can be shown graphically either through a scatter plot of the weightings on each sample or as the total variance explained. In this case, the Y block data are also multivariate, having the same chemical composition as the X block signatures (known as PLS2). If the signature is similar to that of the environmental data, a high value for the explained variance is produced. Conversely, if a poor fit is produced, the explained variance is also small. Each signature can be fitted in turn, and all are fitted independently of each VOL. 36, NO. 11, 2002 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
2355
FIGURE 1. (a) Explained variance (R 2) using coal as the signature. (b) Loadings on the first principal component. The first PC in the signature (t[1] explains 93.7% of the variance as compared to 48.3% in the environmental data, u[1]). other. If none of them explain the variation seen in the data, the fits will be small in every case. A fuller treatment of the methodology including the matrix manipulations used can be found in Geladi and Kowalski (10). The advantage of PLS over other methods is the way it takes all compounds and develops a signature based on the internal relationships between each one. In general, the more compounds that are used, the better the specificity of the signature. PCA can be used independently of the PLS technique to determine the number of potential sources that may be present in the Y block. The scores plot from such an analysis will group sites according to their chemical composition; those that co-vary are likely to have the same or a similar source. Inspection of the groupings may provide an insight into the number of sources, although care must be exercised when dealing with mixtures of variable composition. This PCA technique may also be used to explore the source data and to determine the groupings within the possible source rocks and rivers. As with most statistical analyses, the data were first checked for normality. In both cases, the data required log transformations in order to overcome the large number of low concentrations found. The presence of zeros in the data has to be considered; in reality, these represent values less 2356
9
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 36, NO. 11, 2002
than the limit of detection and may be of considerable importance in defining a signature. The data were logtransformed after the addition of small values representing the limit of detection such as 0.1, 0.01, 0.001, and 0.0001. Altering this value made quantitative changes to the absolute fit but did not change the relative fit (i.e., all values were depressed or elevated together), so consistency was more important than the value used, and a value of 0.01 was used in all of the log transformations. The data were also mean-centered to unit variance to enable the compounds to be directly compared with each other. To facilitate the whole statistical process, a PCA and PLS computer package from Umetrics was employed (SIMCAP, version 8.1).
Results Exxon Data. The Y block data were investigated by PCA to provide an estimate of the number of signatures required to explain the variance in the data. Two features were readily apparent in the scores plot: first, two major groups of sites were formed separated mainly on the second principal component, and second, each group was composed of two subgroups. Therefore, in line with the original hypothesis,
FIGURE 2. (a) Explained variance (R 2) using oil as the signature. (b) Loadings on the first principal component (t[1] explains 72.5% of the variance in the signature and 51.1% in the environment, u[1]). two contenders for the major source material were investigated initially; these were represented by coal and coke from the Bering River and seep oils from Katalla, Poul, Johnston, and Munday Creeks. Initially, the coal and coke samples were used to generate a signature, but the coke had a different suite of compounds from the two coals, as might be expected due to the heat treatment process. Therefore, the coke sample was excluded and replaced with a hybrid coal sample (coal3) made from the mean of the other two samples. This was necessary as a minimum of three samples are needed to define the signature. The results of the PLS analysis can be seen in Figure 1. Two diagrams are presented; the first (a) shows the overall degree of variance explained by this signature in the environmental samples. The second (b) demonstrates how similar the loadings on each individual compound are between the X block and Y block data. Key aspects of the figures are that the “coal” signature apparently describes ∼70% of the variance in the environmental samples from the PWS and GoA (Figure 1a). Notable highs (>60% explained variance) also include the Bering River beach sediments (sed1 and sed2), several of the eastern GoA shales (shale3 and shale4) and rivers (river 10 and river 11), and the Duktoth River and Fountain Stream Glacial Flour (rivers 2 and 3, respectively). The loadings diagram (Figure 1b) has nearly all of the PAHs in the positive quadrant and the terpanes in the negative quadrant. The t[1] and u[1] are
TABLE 2. Similarity between Signatures Expressed as a Proportion of the Explained Variance coal coal seep oil shales rivers 1 rivers 2
0.29 0.51 0.61 0.28
seep oil
shales
Rivers 1
Rivers 2
0.43
0.42 0.46
0.73 0.42 0.69
0.37 0.50 0.63 0.72
0.62 0.60 0.53
0.83 0.71
0.70
loadings on the first PC in the X and Y blocks, respectively, and are not particularly well-correlated to each other. PCA analysis of the five seep oils samples (as collected from the environment) indicated a wide variation in their composition. The three that clustered most closely were Katalla, Poul Creek, and Johnston Creek seep oils. These three were selected for the signature block and fitted to the environmental data. The results can be seen in Figure 2. The explained variance (Figure 2a) also indicates ∼70% for the PWS and GoA sites with a wider variation in the fit for the remaining sites. In Figure 2b, a much better relationship can be seen between the loadings on PC1 from the X block, t[1], and those for the Y block, u[1]. As before, most of the PAHs load positively and the terpanes load negatively. Several comVOL. 36, NO. 11, 2002 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
2357
FIGURE 3. Coomans’ Plot comparing compounds contributing to each of the rivers signatures. pounds cluster around the origin (0,0), indicating little overall effect on that PC. With the PAHs, the more alkylated the compound, the greater the effect on the first PC. For instance, the C3 naphthalenes (N3) have the greatest leverage on this component. Many of the nonalkylated parent PAHs (e.g., C0, D0, F0, and N0) are close to the origin and have little influence on the signature. The amount of variance explained by the first PC in the signature is smaller for the oil data (72.5%) as compared to the coal data (93.7%). This indicates a greater degree of variability in the oil signature data. Boehm et al. (5) have stressed the importance of the Tertiary shales along the eastern GoA in providing hydrocarbons to the PWS area. After the aforesaid analyses, all of the shales present in the Exxon data were used as a signature as well as the river sediments. Initial PCA analysis indicated two groups within the shale data; shales2, 6, and 10 had a similar composition that was different to the remaining sites. These three samples were excluded, and PLS was performed on the data. The results indicated (1) a very high explained variance in the PWS and GoA data averaging 88%; (2) a good relationship between the loadings in the X and Y blocks (t[1] vs u[1]), explaining 82.4% and 63.5% of the variance, respectively; (3) strong similarities between Poul Creek Oil and the shales but a relatively poor relationship with the other seep oils; (4) a strong relationship between coal1 and the shales but not for the other coal or coke; and (5) that removing the terpanes did not improve the definition of the signature. Similar PCA analyses with the river data also identified two distinct groups. Rivers 1-3 and 10-12 had similar values in a scores plot and were different from the other rivers. These were used as signatures (Rivers 1 and Rivers 2, respectively). The results indicated (1) a very high explained variance in the PWS and GoA data, averaging 94% and 72% for Rivers 1 and Rivers 2, respectively (the t[1] and u[1] paired values are 89.9% and 65.4% for Rivers 1 and 89.0% and 55.1% for Rivers 2); and (2) that the Bering River sandbar “coaly” particles site has a high explained variance from both signatures. The addition of these extra potential sources increased the total explained variance for the samples to values substantially greater than 1.0. It is possible to rebase the fits to provide percentages from each source, but it may be questionable as to how many sources to include. This would happen if there was too much similarity between the source signatures. In the previous diagrams, the amount of variance explained in the other signature can be seen; compare the five seep oils in Figure 1a and the two coals and the coke in Figure 2a. There is approximately 35% overlap on average. A summary of the overlap can be seen in Table 2 where the source (X block) signature forms the top row and the Y block 2358
9
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 36, NO. 11, 2002
FIGURE 4. Proportion of explained variance according to source in the Exxon data set. samples are as columns. The results are not symmetrical, as fitting is direction dependent. Notably, Rivers 1 is more similar to the coals than Rivers 2, although the shale signature is very similar to both rivers. Ideally, there should be significantly less overlap with values less than 0.10% or 10%. In an attempt to improve the selectivity, three different approaches were investigated: (1) the terpanes and other biomarkers were excluded from the data set (mimicking the NOAA data set); (2) the PAHs were excluded; and (3) the 24 compounds having the greatest influence on the signatures were used. In general, this group contained a mixture of relatively small PAHs and some terpane biomarkers which can be seen at the extremes of the loadings diagrams (Figures 1b and 2b). In general, when the terpane and sterane biomarkers were removed from the data set and only the PAHs and their alkylated homologues were used, the selectivity decreased and the overlap between the different sources increased. This confirms the arguments of Boehm et al. (12) who suggest that, without the biomarkers, it is difficult to ascribe sources in such data. Conversely, when the biomarkers were used on their own, the specificity increased and lowered the total explained variance. Unlike case 1, however, not all of the sites behaved the same. For example, when using the seep oils as the signature, some of the PWS sites had a lower explained variance, whereas the GoA samples all had a greater explained variance. Interestingly, the other two oils had explained variance of almost 1.0, suggesting that the difference between the oil types was based on their PAH composition and not their biomarkers. The Bering River coaly sediment from Pt. Hey (sed1) was specifically investigated using the techniques to improve
FIGURE 5. Spatial plot of the explained variance (prespill background signature). The first PC explained 91.5% of the variance in the signature and 43.6% in the environmental data. specificity. If only the PAHs were used to generate the signature, there was a slightly better fit for the oil signature over the coal signature (∼10% better). However, if only the terpane/sterane biomarkers are used, the oil signature is ∼8x better at explaining the variance, supporting the views of Boehm et al. (12) that this sample is derived from oil. Similar results were also found with the Katalla Beach sediments (sed5 and sed6). The third approach at improving the selectivity utilized 24 compounds, 12 PAHs, and 12 of the biomarkers which had the greatest leverage in the previous models. These included N1-N4 and T19-T22 among others. These compounds were chosen, in part, through examination of the Coomans’ Plot for pairs of models. These plot the distance to the model in x space for two signatures. Compounds which make similar contributions to each signature fall in the region to the lower left (e.g., Figure 3). In this example, the signatures for the two river groups are used; those compounds that are distinctive to the Rivers 1 signature project toward the upper left and those in Rivers 2 toward the lower right. Those in neither model with a poor explained variance appear in the upper right quadrant. Most of the PAHs and terpanes appear in both signatures and are not diagnostic. However, several compounds do separate from the rest suggesting that, for instance, the D1C and S19 and T6a can be used to trace the Rivers 1 material. A similar plot for the Exxon seep oil and coal data has almost all compounds in the common quadrant except naphthalene and methyl- and dimethylnaphthalene, which are characteristic of the oil signature. This might be expected as these compounds are present in oils but, due to their relatively volatile nature, would be present in much lower concentrations in coal and coke. The results of using these selected compounds, however, did not lead to a lower explained total variance for the PWS and GoA sites due to increased signature overlap. The rebased proportions (proportion each signature constitutes of the total explained variance) based on the full data set are presented in Figure 4 for the PWS and GoA samples together with the sediment samples. Sediments 1 and 2 are from the Bering River and the remainder from Katalla Beach. All of the PWS and GoA samples are subtidal. NOAA Data. The NOAA data set is significantly larger in terms of the number of samples, although no terpane/sterane biomarker concentrations were available. It was hoped that similar signature data for coal and oil could be used in these data, but the two possible source terms provided were Exxon Valdez oil (pure source term) and the prespill background which may contain a wide range of possible sources. Therefore, in keeping with the Exxon data, PLS analyses were conducted on the prespill background only, and the results
are presented as a spatial plot as there are too many sites to display as a bar chart. Several samples of different matrices or different depths were collected at the same location, so the individual spots are shown as a rosette about the sampling point. The results using the prespill background as the signature can be seen in Figure 5. A wide range of explained variances were generated with the best fits to the signature in the channel leading toward the southwest and the inner sites toward the east of PWS have lower explained variances.
Discussion These large complex data sets can be used to develop complex fingerprints for sources rather than using relatively simplistic ratios between selected compounds. The problem in this case, and this is may be why there has been a considerable exchange in the scientific literature, is that the signatures for the potential sources of hydrocarbons are quite similar. This may be expected given the origin of the materials, although the postformation processes that lead to coal and oil are quite different. Van Kooten (13) has indicated that several of these coals are thermally immature and show very strong oil-prone characteristics. For instance, 25% of one coal would be converted into oil upon heating. The results suggest, however, that >30% of the signature is common between each source based on these data. New PAH data for coals and shales in the GoA region (Short, unpublished data) were subsequently used to develop a signature with the NOAA data. However, when applied to the Exxon results, better agreement (overlap) was found with the seep oils of Katalla and Poul Creek (∼70%) than the Bering River coal and coke (∼42%). This reinforces the oil-like nature of these coals proposed by Van Kooten (13). The Coomans’ Plots indicate that the small PAHs (N0, N1, and N2) are the best diagnostic compounds in the model for the oil signature and that the large PAHs are good for coal, although, in a comparison with the new coal data, the C4-chrysenes and C2-fluoranthenes became diagnostic of the seep oil. The best fits suggest that ∼13% of the background hydrocarbons in the GoA and PWS come from coal-based and 18% comes from seep oils. The eroding shales contributed 24% and the majority (26% and 20% for 1 and 2, respectively) arises from the rivers of the eastern GoA. Caution must be used when adding potential sources because of the commonality between many of the compounds in both oils and coal. The use of the terpane/sterane biomarkers does improve the selectivity in many cases, although it was not possible to test the NOAA in the same way. Secondary evidence should be used to demonstrate a mechanism and path through which materials can reach the effected zone. It appears from the data that Rivers 1 may well be another form of coal, although VOL. 36, NO. 11, 2002 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
2359
Rivers 2 may well be a distinct independent source although it is more related to the Exxon seep oil signature than the coal signature (see Table 2). The shales are also well-correlated to both of the Rivers groups but less so to the coal and oil. This may be explained by the wide variety of hydrocarbon sources and histories in this region of North America to the extent that the seep oil may well have a coal origin (13). The NOAA data indicate that the prespill background is not a simple single source, and considerable variation can be seen in the wide geographic area. This is consistent with the results from the Exxon data, and new data (Short, unpublished data) has indicated the similarity between shales and coal. Unfortunately, only eight samples from the PWS and GoA area were provided by Exxon (through Page), so no comment can be made on spatial variability. However, the NOAA data indicate that a wide range of fits may be possible if more samples were collected. Boehm et al. (12) have challenged the coal source proposal by Short et al. (3) on the grounds of mass balance. These investigations reported here determine whether the hydrocarbon pattern observed in the PWS and GoA areas can be explained by the signatures developed from the sources. The mass balance considerations are implicit in the calculations as all compounds are considered together and the results are expressed as percentages. It is likely that there will be weathering of the seep oils, and possibly coals, once they enter the environment and that the degree of weathering will alter the signatures as compared to the original source materials. This weathering will be more apparent in the seep oils, which still contain relatively small PAHs with an appreciable volatility. Short and Heintz (14) have developed a model to identify those samples containing Exxon Valdez oil, and a similar multivariate PLS model should be applied to the seep oils and other sources to characterize materials entering PWS. It may be concluded from this reanalysis of existing data that several more things are needed to resolve this issue. This includes the analysis of pure source materials in sufficient numbers to provide a good signature for each. The new data from Short (unpublished data) may help here. Samples that quantify the weathering processes would also help characterize the sources in the PWS area and the inclusion of the aliphatic hydrocarbons may provide a mechanism for doing this and is borne out by the high hydrogen content of several coals in the GoA (13). Some interpretation of the results with reference to their position and sample type should be applied to explain why, at one site, their are a range of possible fits to the signatures. The conclusions can be summarized as follows. (1) Continued addition of sources will push up the total explained variance due to the commonality of the source compounds. The PAHs and the majority of the terpanes provide relatively few compounds that can be used in absolute source identification. Care needs to be exercised in the choice of possible sources and their defining chemical composition. (2) The coal signature from the Exxon data explains ∼13% of the variance in the samples from the GoA and PWS. The seep oil signature explains 18% and eroding shales 24%, and the remaining 46% can be attributed to the rivers of the eastern GoA. However, there is significant overlap between the rivers and the coal/oil signatures; a reductionist view may combine
2360
9
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 36, NO. 11, 2002
Rivers 1 with the coal signature and Rivers 2 with the eroding shales. The Coomans’ Plots of the two signatures indicates the importance of the N0, N1, and N2 in the signature for seep oil. Most of the other compounds are common between each source. (3) Several sites on the eastern shore of GoA have relatively high coal signatures. These include the glacial flour at Fountain Stream, the Bering River sand bar, and some other beach sediments. Some beach sediments near Katalla also have high seep oil signatures, and there is an overlap with most of the shales. The new coal signature from Short (unpublished data) has considerable overlap with the seep oil signature in the Exxon data. The fit indicates a higher degree of commonality than for the Bering River coal data. (4) In the NOAA data, the prespill background provides localized regions of highly explained variance, but a wide range of fits were observed indicating the mixed source nature of the system. The best fits tend to be toward the southwest of PWS. On average, the prespill background explained 47% of the variance in the data.
Acknowledgments The author thanks both David Page and Jeff Short for supplying their data with no preconditions regarding outcomes. The author also thanks them and their colleagues for helpful comments in early drafts of this work. Helpful comments and suggests were also provided by the anonymous referees and the author thanks them as well.
Literature Cited (1) Wells, P. G.; Butler, J. N.; Hughes, J. S. Exxon Valdez Oil Spill: fate and effects in Alaskan waters; ASTM: Ann Arbor, MI, 1995; p 955. (2) Rice, S. D., Spies, R. B., Wolfe, D. A., Wright, B. A., Eds. Proceedings of the Exxon Valdez Oil Spill Symposium; American Fisheries Society Symposium, 18; American Fisheries Society: Bethesda, MD, 1996. (3) Short, J. W.; Kvenvolden, K. A.; Carlson, P. R.; Hostettler, F. D.; Rosenbauer R. J.; Wright, B. A. Environ. Sci. Technol. 1999, 33, 34-42. (4) Page, D. S.; Boehm, P. D.; Douglas, G. S.; Bence, A. E.; Burns, W. A.; Mankiewicz, P. J. Mar. Pollut. Bull. 1999, 38, 247-260. (5) Boehm, P. D.; Page, D. S.; Burns, W. A.; Bence, A. E.; Mankiewicz, P. J.; Brown, J. S. Environ. Sci. Technol. 2001, 35, 471-479. (6) Burns, W. A.; Mankiewicz, P. J.; Bence, A. E.; Page, D. S.; Parker, K. R. Environ. Toxicol. Chem. 1997, 16, 1119-1131. (7) Wold, S.; Ruhe, A.; Wold, H.; Dunn, W. J. SIAM J. Sci. Stat. Comp. 1984, 5, 735-743. (8) Yunker, M. B.; Macdonald, R. W.; Veltkamp, D. J.; Cretney, W. J. Mar. Chem. 1995, 49, 1-50. (9) Mudge, S. M.; Seguel, C. G. Mar. Pollut. Bull. 1999, 38, 10111021. (10) Geladi, P.; Kowalski, B. R. Anal. Chim. Acta 1986, 185, 1-17. (11) Naftz, D. L. J. Chemom. 1996, 10, 309-324. (12) Boehm, P. D.; Douglas, G. S.; Brown, J. S.; Page, D. S.; Bence, A. E.; Burns, W. A.; Mankiewicz, P. J. Environ. Sci. Technol. 2000, 34, 2064-2065. (13) Van Kooten, G. K. Alaska Geological Society and Geophysical Society of Alaska, Science and Technology Conference, 2000; Swenson, R. F., Ed.; 2000. (14) Short, J. W.; Heintz, R. A. Environ. Sci. Technol. 1997, 31, 23752384.
Received for review June 20, 2001. Revised manuscript received March 5, 2002. Accepted March 15, 2002. ES015572D
