SEARCH

SEARCH BY CITATION

Keywords:

  • Polycyclic aromatic hydrocarbons;
  • Oil spill;
  • Exxon Valdez;
  • Principal-component analysis;
  • Fingerprinting

Abstract

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. MATERIALS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. CONCLUSIONS
  8. Acknowledgements
  9. REFERENCES

A method was developed to allocate polycyclic aromatic hydrocarbons (PAHs) in sediment samples to the PAH sources from which they came. The method uses principal-component analysis to identify possible sources and a least-squares model to find the source mix that gives the best fit of 36 PAH analytes in each sample. The method identified 18 possible PAH sources in a large set of field data collected in Prince William Sound, Alaska, USA, after the 1989 Exxon Valdez oil spill, including diesel oil, diesel soot, spilled crude oil in various weathering states, natural background, creosote, and combustion products from human activities and forest fires. Spill oil was generally found to be a small increment of the natural background in subtidal sediments, whereas combustion products were often the predominant sources for subtidal PAHs near sites of past or present human activity. The method appears to be applicable to other situations, including other spills.


INTRODUCTION

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. MATERIALS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. CONCLUSIONS
  8. Acknowledgements
  9. REFERENCES

The Exxon Valdez spill

On March 24, 1989, the Exxon Valdez oil tanker ran aground on Bligh Reef (Fig. 1) in Prince William Sound, Alaska, USA, spilling about 258,000 barrels of Alaska North Slope (ANS) crude oil [1]. Some of the spilled oil was later driven onto intertidal shorelines by winds and currents. A portion of the oil that mixed with intertidal sediments was washed offshore and deposited subtidally.

Contract research organizations (Battelle Ocean Sciences, Duxbury, MA, USA; Arthur D. Little, Cambridge, MA, USA; Dames and Moore, Seattle, WA, USA) and university scientists engaged by Exxon collected thousands of sediment samples from the spill area. These samples were found to contain polycyclic aromatic hydrocarbons (PAHs) from spill oil, diesel oil, combustion products, creosote, products of natural biological processes (diagenesis), and oil from natural oil seeps [2,3]. A similar set of PAH sources was reported in subtidal sediment samples collected for state and federal governments in the Exxon Valdez spill zone [4] and in biological samples reported in the federal government database PWSOIL as well as in tissue analyses reported to the ad hoc Oil Spill Health Task Force formed after the spill by state and federal agencies and Exxon [5]. Others [6] report finding isolated patches of Monterey (California, USA) tar on Prince William Sound shores, a material imported to Alaska prior to the development of the North Slope oil fields. Identifying PAH sources and determining their relative contributions is necessary to properly interpret the fate of the spilled oil and its biological effects.

Source identification and allocation

The techniques used for source identification and allocation of hydrocarbons in mixtures have been developed in the environmental sciences and the petroleum industry largely over the last 20 years [2, 3, 7-15]. P.D. Boehm et al. (personal communication) have recently reviewed the history of this activity. The major approaches used in source identification include pattern recognition, source-specific diagnostic ratios of PAH analytes, and principal-component analysis (PCA) [16–18].

A good first pass at source identification and allocation can be achieved with visual pattern recognition and source-specific diagnostic ratios. Page et al. [2] used these techniques to identify and allocate four major types of PAH sources in the sediments of Prince William Sound following the Exxon Valdez oil spill: biological PAHs, combustion product PAHs, natural petrogenic background PAHs derived from oil seeps outside of the sound, and petroleum PAHs from ANS sources (i.e., either spill oil or diesel oil refined from ANS oil). Page et al. [2] found ANS petroleum to be generally a small part of the total PAH in subtidal sediments and Page et al. [2] and Burns and Bence [19] noted that the low concentrations of alkylated chrysenes in diesel oils could be used to distinguish diesel oils from Exxon Valdez oil, which has higher concentrations of alkylated chrysenes. A more detailed presentation of the source-ratio allocation model can be found in Page et al. [3], who used it to determine the natural petrogenic background in Prince William Sound and Gulf of Alaska deep subtidal sediment samples.

The precision of Page et al.'s [2] approach can be assessed by using a more detailed method, one that uses PCA to help identify possible contributing sources and a computer-based least-squares procedure to match the whole PAH profile of individual samples (the entire suite of PAH analytes rather than just a few analytes). Principal-component analysis is particularly useful when a large number of variables or composition analytes must be considered [16–18]. Short and Heintz [20] developed a PCA approach that used a spill-oil weathering model to determine the presence of Exxon Valdez oil in a set of subtidal sediment samples from Prince William Sound. However, their application of PCA did not include the chrysenes that are needed to distinguish ANS diesel oil from Exxon Valdez oil.

thumbnail image

Figure Fig. 1.. Map of Prince William Sound, Alaska, USA, showing the path of the Exxon Valdez oil spill.

Download figure to PowerPoint

MATERIALS AND METHODS

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. MATERIALS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. CONCLUSIONS
  8. Acknowledgements
  9. REFERENCES

The present study identified sources that contribute PAHs to hundreds of Prince William Sound sediments. This was done by broadening the scope of PCA to include the 36 PAH analytes (including the chrysenes) common to the sediment data sets reported by Page et al. [2,3]. Sediment PAHs are allocated to the contributing sources by using a computer model that finds the best least-squares match of the PAH profile of the sample by adjusting the mix of possible contributing sources found by PCA. In this model, samples of different weathering states of spill oil were used as sources, rather than only one representative weathered oil sample, as was used by Page et al. [2,3] in their study.

Sediment samples

The intertidal and subtidal sediment PAH data used in this study are from previously reported analyses of sediment samples collected from 1989 to 1993 from Prince William Sound after the 1989 Exxon Valdez oil spill, most of which have been described elsewhere [2,3,21,22]. Most are surface samples (0-2 cm) collected from areas that had been oiled by the spill, but subtidal cores (including prespill sediment) and samples from intertidal pits are also included.

Analytical chemistry methods

The analytical chemistry methods used on all samples are described by Sauer and Boehm [23], Douglas et al. [24,25], Page et al. [2,3], and U.S. Environmental Protection Agency [26]. Basically, sediment samples were spiked with appropriate surrogates (ortho-terphenyl, naphthalene-d8, fluorene-d10, and chrysene-d12), dried with sodium sulfate, and extracted using dichloromethane (modified EPA Method 3550 [25]). Extracts were filtered through glass wool and passed through an alumina column for cleanup (EPA Method 3611 [25]). The extractant was concentrated by Kuderna-Danish evaporation [25,26]. The PAH analysis was performed by using gas chromatography with mass spectrometry operating in the selected ion mode using a modified version of EPA Method 8270 [2,23,25]. A high-resolution capillary column was used (J&W fusedsilica DB5 column, 30 m long, 0.25-mm i.d., and 0.25-μm film thickness [J&W Scientific, Folsom, CA, USA]). Sample size ranged from 10 to 100 g dry weight. Example ion chromatograms are given in Sauer and Boehm [23] and Douglas et al. [25]. Extensive cleanup, large sample sizes, and low preinjection extraction volumes resulted in low detection limits. Detection limits for individual analytes were typically about 1 ng/g dry sediment with 10- to 15-g samples in 1989, but 0.1 ng/g dry sediment levels were achieved with 50 to 100 g dry sediment in 1991.

Thirty-six target analytes included the parent and C-1, C-2, C-3, and C-4 homologues of naphthalene, phenanthrene, and chrysene; the parent and C-1, C-2, and C-3 homologues of fluorene and dibenzothiophene; the C-1 homologue of fluoranthene/pyrene; and anthracene, acenaphthene, acenaphthylene, pyrene, fluoranthene, benz[a]anthracene, benzo[b]fluoranthene, benzo[k]fluoranthene (BkF), benzo[a]pyrene, indeno [1,2,3-cd]pyrene, dibenz[a,h]anthracene, and benzo[ghi]perylene. Three other analytes (biphenyl, perylene, and benzo[e] pyrene) were also analyzed for some data sets, but were not used in the PCA and least-squares analysis because they were not common to all data sets. Isomers of C-1, C-2, C-3, or C-4 homologues were grouped together by number of carbons in side groups and PAH family and treated as individual analytes. For example, the concentrations of the two C-1 isomers of naphthalene were combined and treated as one analyte called “C1-naphthalenes.” Similarly, the isomers of C-2 naphthalene were combined and treated as one analyte, “C2-naphthalenes,” and so on.

Good data quality is a prerequisite for accurate source allocation. Data quality objectives included matrix spike recoveries between 40 and 120% for 16 EPA priority pollutant PAHs. Only one PAH analyte could be below its minimum percentage recovery. Chrysene and benzo[a]pyrene had to have at least 10% recovery and the other compounds had to have at least 20% recovery. Surrogate recovery had to be between 40 and 120%. One surrogate could exceed its maximum limit.

Sources of PAHs

Thirty materials that contained PAHs were evaluated as sources of the PAHs in sediment samples from Prince William Sound. These materials were chosen because they were known to occur in the local area or were potential contaminants during sample collection and handling. The petrogenic materials evaluated included spill oil in various weathering states, locally available diesel oils, Monterey (California) tar, oil-seep stream sediment, and the natural prespill petrogenic background [2,3]. Combustion products from human habitation (e.g., ash, soot, and smoke from wood-burning fireplaces to represent the fuel used at area cabins, mines, villages, canneries, etc.), forest fires (which occur occasionally in the area), and atmospheric fallout from remote combustion sources (NIST standard reference material 1649 [National Institute of Standards and Technology, Gaithersburg, MD, USA]) were also evaluated, as was creosote, a high-temperature hydrocarbon product used to protect docks and pilings in Alaska. Possible sample-handling contaminants included cigarette smoke, diesel oil and diesel soot from sampling ships, and aviation fuels (planes often transported samples from the sound). Two samples collected in government studies were evaluated as possible sources: an intertidal sample of soot collected near cabin ruins at Barnes Cove in Drier Bay and a Sleepy Bay sediment trap sample containing combustion products [27,28]. Biologically produced perylene was not included as a source in this study or used in matching PAH fingerprints because it was not measured in all samples.

Not all of the thirty possible sources evaluated above contributed to Prince William Sound sediment. Principal-component analysis was used to compare the 30 sources with hundreds of sediment samples from the sound. From this comparison, 18 sources were chosen as likely contributors. These 18 sources were used in least-squares calculations to allocate each sediment PAH to the sources from which it came.

Principal-component analysis helps to ensure that all major PAH sources have been identified. If PCA plots show clusters of samples or trends in samples that are not associated with identified sources, then additional work is needed to determine other possible sources. Practical limitations on computation time with the least-squares model (i.e., 1 h per sample on a 66-mHz computer) limited the number of sources to the 18 that looked most promising (i.e., those that closely encompassed the data or matched data clusters or trends) in PCA runs and were most prevalent in the samples.

Principal component analysis

Possible sources of the PAHs in sediment can be identified with the aid of PCA [16–18] using PAH compositions of sediment samples as input (either all PAH analytes or selected ones; we used essentially the whole PAH profile of 36 analytes). The 36 analytes were the columns of the PCA data matrix and 954 sediment samples were the rows of the PCA data matrix. We input this data matrix to a commercially available PCA program (JMP, SAS Institute, Cary, NC, USA). All possible linear combinations of analytes are addressed by the PCA. We did not focus on specific key ratios. For each sediment sample, PCA calculates a number of “principal components” (i.e., linear combinations of the analytes that together explain the variance in the data). The first principal component explains the most variance, the second principal component explains the next highest variance, and so on. For each sample, PCA calculates a “score” for each principal component. Plots of the scores of the sediment samples for one principal component versus the scores of the sediment samples for another principal component reveal clusters and trends in the data that shed light about their sources.

In PCA, individual samples are not classified by any criteria other than sample composition, in our case 36 analytes. Principal components explain the highest possible variance among samples and so represent, better than any other linear combination of analytes, the general difference among samples. The similarity in linear coefficients and resulting principal component scores among samples depends directly on the similarity in analyte compositions. When most of the variation is explained by the first two components, a two-dimensional scattergram (component 1 versus component 2) shows the essential features of multidimensional (here, 36-dimensional) scatter. When plotted by principal component score, samples with similar analyte compositions (i.e., scores) will be closer than those with dissimilar compositions. Plots of principal component scores and an analysis of known sources can be used to identify and explain groupings.

Other multivariate analysis techniques could be used instead of PCA if they can identify the major contributing sources and explain data groupings. The partial least squares technique [29], for example, has been applied to determining the constituents in complex mixtures characterized by gas chromatography data [30].

Least-squares model

A computerized least-squares iterative matching procedure was developed to find the best fit of 36 PAH analytes in a sample to 18 possible sources. For each sample, the model initially equalizes the contribution of total PAH from each source. Then the amount of total PAH from each source is varied using an iterative procedure that starts from the initial equalized contribution and systematically changes source contributions (the xj below) by small steps until the following expression is minimized:

  • equation image

In this expression, i refers to the PAH analytes (1-36), j refers to the PAH sources (1-18), si,j is the concentration of the ith analyte of source j, xj is the fraction of the sample's total PAH that is due to source j, TPAHj is the total PAH of source j, di is the concentration of the ith analyte of the sample, and TPAHsamp is the total PAH of the sample. The si,j and TPAHj are listed in Table 1 for 18 possible sources. The computational procedure used to minimize the expression recognizes that no xj can fall below 0 or rise above 1 and that the sum of all xj must equal 1. The procedure fits only the analytes that are detected in the analysis; analytes reported as nondetects are skipped. The least-squares algorithm always converged to the same results from different start values.

For each sample, the least-squares model calculates the contributions (i.e., the xj) of each of the 18 sources and the goodness of fit. The contributions to a sample from any major source type (such as spill oil, diesel oil and diesel soot, natural background, or combustion products and creosote) can be determined by summing the contributions of appropriate sources and multiplying by the total PAH (TPAH) of the sample. In other words, if four different combustion products are used in tie model, their contributions to a sample can be summed together and multiplied by the sample TPAH to yield the combustion product TPAH contribution to the sample. Similarly, the contribution of any individual source to any analyte in a sample can be determined by multiplying the xj of the source by the normalized or standardized concentration of that analyte in that source (i.e., the concentration of the analyte in the source divided by the TPAH of the source) and then multiplying the result by the TPAH of the sample.

Table Table 1.. Polycyclic aromatic hydrocarbon (PAH) analyte concentrations for 18 sources
  Source
  ABCDEFGHIJKLMNOPQR
AnalyteaSymbolBackground (avg. prespill from cores) (ng/g)Background (offshore Yaka-taga) (ng/g)Background (offshore Port Chalmers) (ng/g)Wood soot (mixed oak and pine) (ng/g)Habitation soot (Barnes Cove) (ng/g)Atmospheric dust (NIST SRM 1649) (ng/g)Sediment trap (combustion product) (ng/g)Creosote (piling in Crab Bay) (mg/kg)Alaska diesel oil (sampling ship Glorita) (mg/kg)Lightly weathered diesel oil (Crab Bay) (ng/g)Diesel soot (sampling ship Glorita) (mg/kg)Fresh Exxon Valdez oil (mg/kg)Very lightly weathered Exxon Valdez oil (ng/g)Lightly weathered Exxon Valdez oil (ng/g)Moderately weathered Exxon Valdez oil (ng/g)Heavily weathered Exxon Valdez oil (ng/g)Very heavily weathered Exxon Valdez oil (ng/g)Extremely weathered Exxon Valdez oil (ng/g)
  1. a Isomers of C-1, C-2, C-3, or C-4 homologues were grouped together by number of carbons in side groups and PAH family and treated as individual analytes. For example, the concentrations of the two C-1 isomers of naphthalene were combined and treated as one analyte called “C1-naphthalenes.”

NaphthaleneN11677.841604.09410453,21074500.062243000.93.60
C1-NaphthalenesN1382027.84661.54460811,5802,490202.01,4006301,100131.800
C2-NaphthalenesN28335413.7613.27520421,6105,3101505.01,7801,90014,000857.700
C3-NaphthalenesN3813529.81103.48570681,3107,43043010.71,4102,20041,0003401300
C4-NaphthalenesN4411553.921700370385874,78040018.56961,40036,0009004800
AcenaphthyleneAcl0.3001300.991707349000.002400000
AcenaphtheneAce2.540130.38140624,700000.02000000
FluoreneF18291.76851.061701298,140449211.59312095001.200
C1-FluorenesF132571.96811.43120501,4301,3501105.12243408,000140500
C2-FluorenesF246913.926201.322601301,4501,78023013.036659021,0009409600
C3-FluorenesF343681.961,9001.411,4001136,6701,42015025.339468027,0001,60025000
AnthraceneAN281.182402.095801621,1805901.000031000
PhenanthreneP551117.841,2007.493,0001,04077,0001,3307716.02623403,900666.19.69.9
C1-PhenanthrenesP1942377.848409.651,40029810,3002,08017028.357291019,0004703500
C2-PhenanthrenesP2862365.886108.671,1002203,4381,57013031.57221,20039,0001,800230270
C3-PhenanthrenesP3541591.961,1008.646902289996816012.257688036,0002,7005101100
C4-PhenanthrenesP4431281.962,5005.971,400613,760182120.044649021,0001,900330820
DibenzothiopheneD721000.71190374,540966536.92172802,700342.600
C1-DibenzothiophenesD113460.9801.56230311,1441,95014013.544975015,0003902300
C2-DibenzothiophenesD212361.373101.69340806361,71014025.76351,20036,0001,900290310
C3-DibenzothiophenesD37210.986807.253601122429287720.05791,10041,0003,200580650
FluorantheneFL10803,70019.85,0001,47044,500007.6200267.300
PyrenePY15121.964,60018.94,0001,17027,2002706.0101339035000
C1-Fluoranthenes/pyrenesFP141801.964,000131,5004087,4006103.1821203,4003501403722
Benz[a]anthraceneBaA6904,300151,3004693,210000.029.300000
ChryseneC192205,10015.92,30075511,900202.7461103,0005301905536
C1-ChrysenesC136641.182,60010.96703641,740201.9892106,4001,1002608848
C2-ChrysenesC243941.181,5005.12410154745000.01382609,5001,00027020084
C3-ChrysenesC328681.372,6001.127101210000.01152208,0001,400240280110
C4-ChrysenesC4183601,3000390710000.001406,10095011027085
Benzo[b]fluorantheneBbF1561.1810,00015.26,0001,1404,050002.463.5390029139.4
Benzo[k]fluorantheneBkF0.5102,30014.396002,940000.601600000
Benzo[a]pyreneBaP3406,80019.91,9004921,170000.00227701402.57.54.4
Indeno[1,2,3-cd]pyreneID4105,20011.33,0004361,340001.01000000
Dibenz[a,h]anthraceneDA2.5106804.12460196270000.010005.300
Benzo[ghi]peryleneBgh13404,50012.23,9005031,120000.025.31503721258.8
Total PAH (dry weight)TPAH1,0232,7909270,05624946,38010,784241,86037,3032,37026111,93916,206400,75022,0773,7051,304418

RESULTS

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. MATERIALS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. CONCLUSIONS
  8. Acknowledgements
  9. REFERENCES

Principal component analysis and the least-squares model were used to evaluate 30 potential sources of PAHs in sediments collected after the Exxon Valdez spill. Eighteen of them were identified as possible sources of PAHs in subtidal sediments. Noise or imprecision in analyte values was simulated with the least-squares model to indicate the accuracy of source allocation possible with this approach. Then, the least-squares model was used to determine the contributions of each of the 18 sources in five types of subtidal sediments: nearshore sediments next to oiled shores (i.e., 2 and 3-m water depths); embayment sediments from oiled and unoiled embayments, including samples near sites of past human habitation; offshore sediments 100 to 1,000 m from shore in 8- to 55-m water depths; deep subtidal sediments out to a water depth of 763 m; and prespill sediments from cores.

For simplicity, the contributions of the 18 sources have been combined into four groups according to source type: combustion products and creosote, Exxon Valdez oil, diesel oil and diesel soot, and natural petrogenic background. Contributions of the seven different weathering states used to model Exxon Valdez oil were added together to yield the total Exxon Valdez contribution to the sample. Of course, the Exxon Valdez oil spill may not be the sole origin of ANS oil in the area, so some of the oil attributed to Exxon Valdez may be from other sources. In addition, some least-squares runs were made without diesel oil or diesel soot (i.e., 15 source runs), to avoid possible problems with highly correlated sources and because little diesel oil or diesel soot as found in the subtidal sediments examined in this study.

To put the results in perspective, the PAH levels in samples are compared in the examples below to the levels that define the onset of toxic effects in sediment. This level has been estimated from field and laboratory toxicity data as 4,000 ng/g total PAH [31] and has been confirmed in sediment toxicity tests from the Exxon Valdez spill [21]. The 4,000-ng/g total PAH level is the 10th percentile (low) on a distribution curve of observed lethal and sublethal effects and has been termed the effects range—low (ER-L) [31].

Sources of PAHs

More than 900 sediment samples were used in a PCA computer run to help identify possible sources of the PAHs in the sediments. Analyte concentrations were normalized by dividing by total PAH to eliminate concentration level as one of the principal components. All of the samples had total PAH concentrations greater than 50 ng/g. These samples were chosen because they were thought to contain PAHs from the major sources in relatively unmixed form, making sample identification easier. Four sets of intertidal PAH data were used to characterize weathered spill oil: the 1989 and 1991 intertidal studies [21], a 1993 intertidal study (P.D. Boehm, personal communication), and a 1990 bioremediation study [22]. Three data sets of subtidal data [2] were used to indicate sources in subtidal sediment: a 1989 through 1990 offshore sediment study, where samples were generally taken from 100 to 1,000 m from shore; a 1991 deep subtidal study that included prespill sediment from the bottoms of age-dated cores and samples taken near sites of past or present human habitation; and a 1991 study of two embayments, one oiled and the other unoiled. The PAH analyses of possible sources were also included in the PCA run to help identify the sources contributing to various clusters and trends in the data. Nearshore subtidal samples were not included in this PCA run because they were thought to contain mixtures of spill oil, combustion products, and natural petrogenic background, sources more clearly indicated in the other data. Once the PAH distributions of the sources were established by PCA, these same PAH distributions were used in the least-squares model to allocate PAHs in nearshore samples to the appropriate sources.

thumbnail image

Figure Fig. 2.. Principal components 1 and 2 for sediment data from the Exxon Valdez oil spill. The polycyclic aromatic hydrocarbon sources A through R are listed in Table 1.

Download figure to PowerPoint

The first two principal components explained 54% of the variance in the data, and a plot of sample scores for these two principal components (Fig. 2) was sufficient to identify the major contributing sources. An additional 9% of the variance is explained by the third principal component. Plots of the first and third principal-component scores and the second and third principal-component scores (not shown) did not suggest additional sources; rather, the sources identified in Figure 2 were confirmed. These additional PCA plots also suggest which of the combustion product sources are most similar to the subtidal sediments near human habitation sites (habitation soot and sediment trap), which was borne out by least-squares runs.

Figure 2 indicates that many subtidal samples have compositions similar to the natural background (sources A, B, and C, which have much lower ratios of alkyl dibenzothiophenes to alkyl phenanthrenes than do Exxon Valdez oil or ANS diesel oil and more alkyl homologues relative to parent PAHs than do combustion products). A group of subtidal samples extends from the natural background region of the plot to the combustion product sources (sources D, E, F, and G) in the lower right-hand portion of the figure. These samples contain increasing proportions of combustion product PAHs as they near the combustion product sources on the figure. Similarly, some subtidal samples extend from the natural background region toward the more weathered parts (sources O, P, Q, and R) of the intertidal spill-oil weathering trend, and a couple of samples extend toward creosote (source H).

Sources I and J reflect fresh Alaska diesel oil and slightly weathered Alaska diesel oil from the shores of a boat harbor at Crab Bay, respectively. Source K is diesel soot collected from a ship used in collecting sediment samples.

The entire intertidal spill-oil weathering trend extends from source 3L (fresh) to source 3R (weathered). Generally, the freshest, least-weathered spill oils were found in 1989 and the more weathered spill oils much later. Levels of PAHs in intertidal sediment also generally declined with time.

Some investigators [20] use spill-oil weathering models to predict the composition of weathered oil. The dashed line in Figure 2 shows how Exxon Valdez oil would weather by water-washing because of repeated tidal exposure. This line was calculated by using the octanol/water partition coefficients reported by Neff and Burns [32], assuming equilibrium was achieved between the oil and a fixed amount of water per tidal cycle [33]. In this calculation, 600 to 700 g of water (an amount picked from pore-water chemistry results) was used per gram of oil for each tidal cycle. A new oil composition was calculated after each tidal cycle using octanol/water partition coefficients, and the process was repeated for hundreds or thousands of times to deplete the oil. Each tidal cycle drained away the lighter, more water-soluble parts of the oil, leaving progressively heavier, less water-soluble oil behind. Actual intertidal samples followed the water-washing curve initially but began to bend toward less water-soluble pyrogenic sources as the petrogenic components weathered. All of the intertidal data sets showed this behavior.

The 18 sources (sources A through R of Fig. 2) chosen for use in the least-squares model include spill oil in various weathering states, diesel oil and diesel soot, the natural petrogenic background in Prince William Sound sediments, creosote, and several combustion product sources. More than one source of each type was used to account for variability in laboratory analysis and uncertainty over which sources were actually contributing to the sediment. These 18 sources were chosen on the basis of their locations (shown on Fig. 2) and how well they seemed to match the samples; the other possible sources investigated were farther away from the sediment samples in PCA plots and/or contributed little to most sediments. The PAH distributions for the 18 sources are shown in Figure 3. Each fingerprint in Figure 3 is a plot of the concentration of 36 individual PAH analytes, which appear in the order listed in Table 1.

An examination of Figure 3 L, M, N, O, P, Q, and R shows what happens to the PAH composition of spill oil during weathering. The lighter, more water-soluble components of each analyte family disappear first. With time, eventually only the least water-soluble components are left (Fig. 3R), in this case the chrysenes (which are disappearing by the same mechanism, only more slowly).

The effect of noise

Sediment PAH data contain noise or scatter introduced by the analytical instruments, the procedure, the laboratory technician, or nature. This noise level can be measured by the coefficient of variation of replicate samples, which is the standard deviation of a set of replicates divided by their mean. If the coefficient of variation is low, data scatter is small and the least-squares method should approach the correct solution. If the coefficient of variance is high, results can vary considerably from the correct solution.

To evaluate this effect quantitatively, normally distributed random noise (either positive or negative) was generated and added to each of the 36 PAH analytes of a hypothetical mixture of known sources. The least-squares model was then applied to this artificially noisy data and the best fit obtained. This process was repeated 25 times to determine the effect caused by a given level of random noise.

Noise caused the least-squares model to find best fits that scattered around the true mixture. This is displayed in triangular diagrams of source mix (Fig. 4) for two levels of noise. These diagrams plot the relative amounts of three sets of sources, designated by name at the corners of the diagrams. The amount that the multiple solutions differ from the true mixture depends on the level of noise added to the analytes. Noise can increase or decrease apparent source contributions relative to true values. Noise can even cause PAH to be attributed to a source that is not present. Other combinations of sources used to generate the “true” solution sometimes resulted in noise-study results that skewed to one side or another of the triangular diagram.

Most of the studies we investigated showed noise levels falling between the two levels indicated on Figure 4, based on three replicates collected for each sample and using C-3 naphthalene, C-1 phenanthrene, and C-2 phenanthrene (typically the largest three analytes in a sample) to calculate noise levels. (For studies without replicates, a rough estimate of noise level was obtained using repeat analyses of Exxon Valdez oil or standard sediment samples.) Noise levels were comparable to those shown on Figure 4A (i.e., about five percentage points) when large sediment samples were analyzed (i.e., 50-100 g dry weight). Concentration levels below about 5 to 20 percentage points are probably not significant for data sets derived from smaller samples (10-15 g dry weight) (Fig. 4B).

Figure 5 illustrates the noisiness of a set of 15- to 25-g dry weight offshore sediment samples collected in 1990 [2]. The calculated contributions of four types of PAHs are shown for replicate samples. Pieces of soot sometimes appeared in samples (see Eshamy Bay samples in Fig. 5), as did small tar balls, making calculated noise levels in replicate field samples higher than actual noise levels. Small concentrations of Exxon Valdez oil sometimes appeared in samples from unoiled areas; this is thought to result from noise in the data or possible ANS petroleum from other sources (boating, etc.), rather than from oil from the spill. The diesel oil and diesel soot in these Figure 5 samples are also within the range caused by noise in the data. In addition to noise in measurement, the natural variation of PAH distributions in sources and contributions from unknown sources are also sources of error.

Example whole-profile match

The type of match obtained between measured and calculated PAH profiles is illustrated in Figure 6 for a 1991 embayment sample taken at a depth of 75 m in the mouth of the Bay of Isles (Fig. 1) [2]. No diesel oil, diesel soot, or creosote was found in this sample. Figure 6B shows that combustion products contributed most heavily to the four- to six-ring PAHs. Natural petrogenic background made up the bulk of the PAHs in the sample, whereas the Exxon Valdez contribution was fairly weathered, with chrysenes being the largest spill-oil contributors. Analytes that were not detected in the measured fingerprint, such as BkF in Figure 6A, were not used in determining the least squares fit, but their estimated concentrations are shown in Figure 6B. Finally, the total PAH levels in this sample were well below the 4,000 ng/g toxicity threshold level [31].

Nearshore samples

Two large groups of nearshore sediment samples were collected next to oiled shores [2,21]. In 1989, 85 spill-path samples were collected from a water depth of 2 m at subjectively chosen sites; 82% of the sites were heavily oiled, 12% were moderately oiled, and 6% were lightly oiled. In 1990, 124 samples were collected from a water depth of 3 m at stratified random sampling sites (different sites than those sampled in 1989) and 24 samples were taken from a water depth of 3 m at the sites sampled in 1989.

thumbnail image

Figure Fig. 3.. Polycyclic aromatic hydrocarbon (PAH) fingerprints of 18 sources used in the least-squares model. Each fingerprint plots the concentration of 36 PAH analytes in the order listed in Table 1.

Download figure to PowerPoint

thumbnail image

Figure Fig. 4.. Scatter in least squares results for two levels of random noise imposed on analytes (σ/x on three analytes with the highest concentrations).

Download figure to PowerPoint

thumbnail image

Figure Fig. 5.. Replicate offshore subtidal sediment samples (1990).

Download figure to PowerPoint

thumbnail image

Figure Fig. 6.. Best least-squares match of measured fingerprint (1991 Bay of Isles sample).

Download figure to PowerPoint

thumbnail image

Figure Fig. 7.. Average polycyclic aromatic hydrocarbon concentrations in 1989 and 1990 nearshore samples.

Download figure to PowerPoint

The least-squares model shows a large difference between the arithmetic average concentration of Exxon Valdez oil and the geometric average concentration in the 85 nearshore spillpath sediments collected in 1989 (Fig. 7) because a few samples contained large concentrations of Exxon Valdez oil, probably from small tar balls. Seven 2-m samples out of the 85 samples had total PAH levels exceeding Long and Morgan's level of 4,000 ng/g [31]. These seven samples ranged up to 10,872 ng/g total PAH, five of them had Exxon Valdez oil concentrations exceeding the ER-L sediment toxicity threshold of 4,000 ng/g total PAH [31], and one of the seven samples contained 90% (w/w) combustion products. A number of samples showed little PAHs other than the natural petrogenic background or combustion products. The geometric average of 92 ng/g Exxon Valdez oil reflects the more typical sample.

The 1990 samples contain much less Exxon Valdez oil (Fig. 7). The geometric average concentration of Exxon Valdez oil was 8 ng/g; none of these 3-m samples exceeded 4,000 ng/g. Forty-three percent of the samples came from sites that had been heavily oiled in 1989; another 22% came from sites moderately that were oiled in 1989, and 35% were from sites that were lightly oiled in 1989. For comparison, 70% of the oiled shoreline in Prince William Sound was lightly oiled in 1989, 12% was moderately oiled, and 18% was heavily oiled [34]. Background and combustion product sources are higher in the 1989 samples because several 1989 locations were near human habitation sites and sedimentary areas at the heads of bays.

The least-squares method generally found more Exxon Valdez oil than the source-ratio method used by Page et al. [2]. For example, the arithmetic average Exxon Valdez oil concentration in the 1990 stratified random samples was 78 ± 118 ng/g (standard deviation) by the least-squares method compared to 50 ± 134 ng/g by the method of Page et al. [2].

Oiled embayment

Much of the shoreline in the Bay of Isles was heavily oiled in 1989. The 1991 [2,3] embayment study collected 40 stratified random samples (8- to 124-m water depth) to characterize the condition of the bottom of the bay. The 1991 deep subtidal study [2,3] also collected four samples from the bay at two deepwater locations.

These samples show that the fraction of total PAH derived from Exxon Valdez oil was lowest at the mouth of the bay and highest in the arms (Fig. 8). The least-squares model found that the highest amount of Exxon Valdez oil in one of these samples was 246 ng/g, well below the 4,000 ng/g toxicity threshold reported by Long and Morgan [31], although a couple of shallow nearshore locations (from a separate study), which likely contained small tar balls, did exceed 4,000 ng/g [2]. In the two shallow arms near the head of the bay where the natural petrogenic background was the smallest, more than half of the PAHs came originally from the Exxon Valdez spill (arithmetic average concentrations are shown in the inset). Interestingly, the 30% Exxon Valdez contour is generally consistent with the depth contours of the bay (not shown). Where the bay is shallower, the fraction derived from the spill is higher. The portion of total PAH coming from the natural petrogenic background increases with water depth and clay content in the sediment and is highest in the mouth of the bay. The inclusion of high perylene concentrations (up to 480 ng/g in the arms) in PAH totals would lower the proportion of total PAH resulting from the spill.

thumbnail image

Figure Fig. 8.. Polycyclic aromatic hydrocarbons from Exxon Valdez oil in Bay of Isles subtidal sediments.

Download figure to PowerPoint

thumbnail image

Figure Fig. 9.. Polycyclic aromatic hydrocarbons from combustion products and creosote in Drier Bay subtidal sediments.

Download figure to PowerPoint

An 18-source run found an average of 8% diesel oil or diesel soot in the Bay of Isles sediment. This is not believed to be real diesel oil or diesel soot but is thought to result from noise in the data and the fact that the diesel soot PAH profile was similar to weathering Exxon Valdez oil in most respects except for the low level of chrysenes in the diesel oil and diesel soot profiles. Normally the difference in chrysene concentrations can distinguish diesel soot from Exxon Valdez oil, but there are samples in which Exxon Valdez naphthalenes and fluorenes are similar enough to those in diesel soot to overwhelm the differences in chrysenes in least squares calculations. To ensure that levels of Exxon Valdez oil were not understated, the results of a 15-source run (the 18 sources minus Alaska diesel oil, lightly weathered diesel oil, and diesel soot) are plotted in Figure 8; the 15-source run attributed slightly more PAHs to Exxon Valdez oil than did the 18-source run.

Page et al. [2] reported the arithmetic averages of Exxon Valdez oil concentrations as 49 ± 52, 57 ± 21, and 52 ± 19 ng/g for three depth ranges in this bay (10-50 m, 50-100 m, and 100-150 m, respectively) out of total PAH concentrations of 225, 488, and 622 ng/g (including perylene). In contrast, the least-squares method found the arithmetic average of Exxon Valdez oil levels to be 54 ± 62, 74 ± 22, and 57 ± 25 ng/g for the same three depth ranges.

Lightly oiled embayment

Drier Bay is a west-facing bay (Fig. 1) that received only very light amounts of Exxon Valdez oil on isolated patches of its shoreline [34]. In 1991, the embayment study collected 42 subtidal sediment samples from Drier Bay and the deep subtidal study collected one [3].

The PAHs from combustion products, creosote, and the natural petrogenic background dominate the subtidal sediments of Drier Bay. Levels of total PAH in this relatively unoiled bay are as high as those in the oiled Bay of Isles. Two areas high in combustion products and creosote appear in Drier Bay (Fig. 9), one at the northeast end of the bay at Port Audrey and the other at Barnes Cove on the south shore. These locations match two areas of former human habitation in the bay described by Lethcoe and Lethcoe [35]. Port Audrey, at the northeast end of the bay, is the former site of a fish cannery, various cabins, dock facilities, and mine accesses. Creosotecoated pilings still exist at Port Audrey. Barnes Cove is the site of cabin ruins. Figure 9 illustrates the 18-source least squares model results for Drier Bay.

Page et al. [3] found the arithmetic average of overall pyrogenic concentrations to be 147 ± 235, 66 ± 45, and 69 ± 39 ng/g for three depth ranges of Drier Bay (10-50 m, 50-100 m, and 100-150 m, respectively). However, the least-squares method found average arithmetic pyrogenic concentrations (i.e., combustion products plus creosote) of 237 ± 386, 89 ± 75, and 82 ± 47 ng/g for the same three depth ranges. In addition, the least-squares model found that the arithmetic average of Exxon Valdez oil concentrations was less than 4 ng/g in Drier Bay, well within the noise level of the data.

Offshore sediments

In 1989 and 1990, subtidal sediments were collected from 100 to 1,000 m offshore in water depths ranging from 8 to 55 m. The arithmetic averages of Exxon Valdez oil concentrations from 1989 spill-path stations were 24 ± 46 ng/g according to the least-squares method versus 29 ± 33 ng/g reported in Page et al. [2] out of a total PAH level of 389 ± 269 ng/g. In 1990, the arithmetic average Exxon Valdez oil concentrations for a different set of spill-path stations were 89 ± 106 ng/g according to the least-squares method, as opposed to 38 ± 48 ng/g reported by Page et al. [2] out of a total PAH level of 511 ± 293 ng/g.

thumbnail image

Figure Fig. 10.. Contributions of polycyclic aromatic hydrocarbon sources as a function of water depth for samples from the 1991 deepwater study.

Download figure to PowerPoint

thumbnail image

Figure Fig. 11.. Concentration of combustion products versus depth in core for two age-dated cores. The deposition dates shown for core slices are estimated from 210Pb dating.

Download figure to PowerPoint

Deep subtidal sediments

The 1991 deep subtidal program [2,3] sampled depths in Prince William Sound ranging from 18 to 763 m and from locations well away from shore to embayments near sites of human habitation. Figure 10 shows the contributions of four source types as a function of water depth for surface sediments (0-2 cm) collected in this study. The highest PAH contributions came from combustion product and creosote sources (Fig. 10A) in embayments near sites of past or present human habitation. Only very low levels of combustion product or creosote PAHs were found in deep water away from land. Note that PAHs from the natural petrogenic background increase with increasing water depth (Fig. 10B). Bence and Douglas [36] concluded that these PAHs are associated with fine-grain sediment that settles more easily in deep water. Thus the major contributors to these samples were natural petrogenic background or combustion sources, not spill oil (Fig. 10C); Exxon Valdez oil concentrations in deep samples are generally less than 3 to 5% of TPAH [3] and are probably the result of noise and scatter in the analyte data rather than the presence of oil. Finally, there was very little diesel oil and diesel soot in these sediments (Fig. 10D). The contributions of all four groups of sources were below the 4,000 ng/g total PAH ER-L level [31].

Core samples

In 1991, 10-cm-diameter cores of sediment were collected from various locations in the spill path as part of the deep subtidal study [2,3]. Ten slices from each of five cores were age-dated using 210Pb. In addition, several 2-cm slices (i.e., slices from the top, middle, and bottom of the age-dated cores) were analyzed for PAHs. The 210Pb data indicate deposition rates of 2 to 6.5 mm/year. At these rates, a 2-cm slice represents about 3 to 10 years of deposition. Alternatively, the age of the sediment at the bottom of a 32-cm core could be up to up to 160 years old.

Least-squares model results for combustion products and creosote in two of the age-dated cores are depicted in Figure 11 as a function of depth in the core. One core came from a water depth of 187 m in Latouche Passage (Fig. 1), within several kilometers of historic mining and fish-processing activities that date back to 1899 and within several kilometers of new Chenega Village, which was established at Crab Bay after the 1964 earthquake. The other core was taken from a water depth of 212 m north of Smith Island (Fig. 1), many kilometers away from significant human activity.

The core from Latouche Passage shows a gradual and continual buildup of combustion products that began around the turn of the century, consistent with the development of the first mines nearby on Latouche Island in 1899 and fish processing nearby on Bettles Island (1 km from the core site) in 1912 [37]. The Latouche mine area had a population of more than 500 in 1920. Other fish-processing and lumber-mill activities at nearby Sawmill Bay, Crab Bay, and Latouche Island followed [37], expanding the population of the area and contributing to the buildup of combustion products.

Interestingly, the least-squares model found that from 5 to 10% of total PAH in the prespill portion of the Latouche Passage core was petroleum (i.e., crude oil, refined products, or tar) other than the petrogenic natural background. These low levels are most likely California oil from the Monterey Formation, which was imported to Alaska before the development of the Trans-Alaska Pipeline in the 1970s. In fact, old tank ruins at some human habitation sites in Prince William Sound, including some in the Latouche Passage area, still contain this Monterey oil, probably dating from the 1920s to 1940s. Monterey oil gives a better least-squares fit to the petroleum in the bottom of the core than did Exxon Valdez oil, but the low concentration levels and the similarities of parts of the PAH distributions of these two sources rule out a firm conclusion.

In contrast, the Smith Island core shows little buildup in the combustion product signal over the same time period and no petroleum in prespill sediment (other than the petrogenic natural background). Similar results were found for other Prince William Sound cores that were collected far from sites of human activity in the sound.

DISCUSSION

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. MATERIALS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. CONCLUSIONS
  8. Acknowledgements
  9. REFERENCES

This method versus simpler approaches

The PCA and least squares method used in this study confirm the general conclusions reached by Page et al. [2], who used pattern recognition and diagnostic source ratios, namely that the Exxon Valdez contribution of PAHs in concentration is small relative to the natural petrogenic background, is highest near oiled shores, and is negligible in deepwater sediment away from shore; the background contribution increases with water depth; and few samples exceed the 4,000 ng/g toxic threshold suggested by Long and Morgan [31].

However, although the overall conclusions of this study are similar to those reported by Page et al. [2,3], the new approach offers a more robust assessment (i.e., one based on more information) of the contributions of spill oil and combustion products to subtidal sediment than do source-ratio approaches. First, higher levels of spill oil were found with the new approach than with source ratios, from about 20% (w/w) to sometimes as high as 100% (w/w). The underlying reason for the difference is that subtidal oil is more weathered than the single representative weathered-oil sample used by Page et al. [2] in their source-ratio approach. This means that the source-ratio model used by Page et al. [2,3] underrepresents the amount of ANS-derived chrysenes actually present. By virtue of using seven different weathering states of Exxon Valdez oil as sources, the least-squares model is better able to match the degree of weathering in any given sample.

Similarly, the least-squares method found higher levels of combustion products than the allocation method used by Page et al. [2], who calculated the combustion product (pyrogenic) contribution to subtidal sediments as the sum of the three to six-ring parent PAHs. However, combustion product sources also contain a certain fraction of alkylated PAHs, which amounted to 25 to 30% (w/w) in combustion product sources in the present study. Thus, the correct combustion product contribution is higher than that calculated by Page et al. [2]. However, this is not a significant problem for most sediment samples, because the combustion product contribution is generally small except near human habitation sites.

Allocating sources with similar fingerprints

The least-squares method is capable of allocating the contributions of sources that have fairly similar fingerprints, but difficulties may be encountered. For example, the method was generally able to separate Exxon Valdez oil (an ANS crude oil) from diesel oil or diesel soot that had an ANS crude oil origin, but low levels of diesel oil or diesel soot were sometimes allocated for samples that should have been Exxon Valdez oil only. Similar difficulties were occasionally encountered allocating Exxon Valdez and Monterey oils when only one might have been present. For these reasons, we did not include Monterey oil among the 18 sources routinely used in the least-squares model, and we would run 15-source least squares models without diesel oil or diesel soot to confirm levels of Exxon Valdez oil.

The misallocation of similar sources arises from the fact that parts of the PAH fingerprints are correlated or similar (in the cases above, the napthalenes, fluorenes, and phenanthrenes). These similar parts can, on occasion, dominate in the calculation of least squares, overshadowing the parts of the fingerprints that are different. Generally, the three major types of sources (combustion products, natural background, and Exxon Valdez oil) had sufficiently different PAH distributions that few problems were encountered in allocation by major types.

When the PAH distributions of sources are highly correlated, additional analytes or other techniques [29] may help identify contributing sources. For example, differences in the distributions of the normal alkanes (C10-C36) can distinguish between relatively unweathered refined products like diesel (which lacks the heaviest normal alkanes) and relatively unweathered crude oil (which has the heaviest normal alkanes).

Forest-fire fallout

The PAH distribution of the sediment trap source (Fig. 3G) bears a striking resemblance to that of the atmospheric dust source (Fig. 3F). Sale et al. [28], who took the sediment trap sample, concluded that it and many others like it collected at the same time were diesel soot. The abundance of chrysenes and unsubstituted four- to six-ring PAHs make this interpretation unlikely. An alternative explanation is that this material is fallout from a May 1991 forest fire on the nearby Kenai Peninsula, the largest forest fire there in 22 years [38]. This material started showing up in sediment trap samples all over western Prince William Sound during the June to September 1991 sediment trap collection period, and it continued to appear during the September 1991 to March 1992 collection period. The similarity of the sediment trap material to wood soot and the timing of its appearance argue for a forest-fire origin. The PAHs in the atmospheric dust sample may have had a similar origin, although this sample was collected near Washington, DC, USA. This product of open wood burning also shows up in subtidal sediment samples taken in 1989 through 1991 near human habitation sites or camping sites. For example, this material makes up 40% of subtidal PAHs near a popular camping site in West Twin Bay on Perry Island [35]. Sometimes a piece of this combustion material, perhaps a cinder, showed up at high concentrations in one sample but not in replicate samples taken at the same time from the same place.

The appearance of this burnt-wood material in western Prince William Sound during the summer of 1991 may be the reason that the 1991 Bay of Isles samples appeared to have a higher combustion product content than samples from previous years (12.5% in 1991 compared to 3-4% in 1989-1990). Samples taken in 1991 in deeper water (200-763 m) away from land showed little of this material; it may have been swept away by the Alaska Coastal Current, which flows through Prince William Sound, before the material could be deposited on the seafloor. Others [4] have noted that forest fires on the Kenai Peninsula were potential sources of pyrogenic PAHs for Prince William Sound.

Identifying sources

Identifying sources and their compositions is a trial-and-error procedure. It is difficult to know if all contributing sources have been identified or if the exact compositions of the sources have been determined. Consideration of several potential sources of each type (i.e., combustion products, petroleums, contaminants) proved useful in our study. Principal component analysis indicated which of the sources might be contributing to the sediment, and the least squares model indicated which of these possible contributors best matched the sediment PAH data.

The square of the difference between sample PAH profiles and the best-fit profiles determined by the least-squares model suggested that natural background sources and Exxon Valdez oil sources had been modeled well; the squares of the differences were relatively independent of the proportion of total PAH coming from these sources. However, the square of the difference increased as the proportion of combustion product sources in samples increased, suggesting that the exact composition of this source material had not been identified although the general characteristics (e.g., large proportions of four- to six-ring PAHs) were similar. The sediment PAH data suggested that a range of combustion product sources actually contributed to sediment, perhaps reflecting different combustion temperatures, different materials undergoing combustion, or different combustion end products (e.g., ash, soot, or smoke, all of which differ in composition). This range of conditions, materials, and end products is believed to have caused the increasing squares of differences as the combustion product contribution increased.

Four different laboratories produced the PAH analyses used for the sources. Two of the same laboratories took the sediment data used in this study. The use of different laboratories introduces error into the allocation process, but the error is not very large in this case, based on runs made with alternate sources.

Data quality

Noise studies suggest that data quality (i.e., low noise levels or reproducibility of analyte concentrations) is critical for accurate allocation of PAHs to sources. This is particularly true if PAHs from the sources are present at low concentrations or percentages. Low noise levels can be achieved by using large sample sizes (50-150 g dry weight), small extract preinjection volumes (0.25-0.5 ml), the same instrument and technician on all samples, and consistent peak identification and construction of integration baselines [2; G.S. Douglas et al., personal communication]. Smaller sample sizes (10-15 g dry weight) provide less accurate data for allocation but do provide adequate data for determining whether PAH levels in sediment may be toxic (e.g., above the ER-L toxic-effect threshold of 4,000 ng/g total PAH [31]).

Very weathered samples

The PCA plot (Fig. 2) shows that the most weathered intertidal sediments bend away from the calculated water-wash curve. This bend appears in all data sets at roughly the same degree of weathering and often at relatively low PAH levels (sometimes, however, at levels as high as 10,000 ng/g dry weight). The bend reflects changes in sample PAH composition as determined by the analytical method. The reason for the bend is unclear. Sample compositions could appear to change because of decreasing signal-to-noise ratios as PAHs decline. Alternatively, bacteria could be changing their PAH food source as their favorite PAH food disappears. Kropp et al. [39] have shown that oil-consuming heterotrophic bacteria require naphthalene as a cosubstitute in order to use benzothiophenes as a carbon source. Most of the naphthalenes disappear close to the bend; if a switch in bacteria food source were to occur here, it would change the relative rates of disappearance of various PAHs, resulting in a change in the weathering trend. A model using both water washing and bacterial degradation might to a better job of matching the degradation of intertidal oil. Another possible explanation is that the bend might be occurring at the PAH levels where bulk oil disappears and the remaining oil is tied up in or on particles in the sediment and is thus not as readily bioavailable.

If weathered-oil samples from the water-wash curve had been used instead of the sources actually used, slightly lower amounts of Exxon Valdez oil would have been calculated for sediment samples. These effects demonstrate the key role PCA, using a full suite of PAH analytes, plays in identifying hydrocarbon sources in an actual field situation.

CONCLUSIONS

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. MATERIALS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. CONCLUSIONS
  8. Acknowledgements
  9. REFERENCES

Analysis of sediment samples using the least squares model shows that Exxon Valdez oil was generally a small contributor to subtidal sediments. This oil was highest in shallow sediments immediately adjacent to oiled shores, but even there geometric averages show that the spill-oil content of sediments was relatively low. Farther offshore, spill oil reaching the sea-floor was more weathered and at lower concentrations than in nearshore sediments. There were little spill-oil PAHs in deep-water sediments away from oiled shores. The PAHs from the natural background increased with increasing water depth and combustion product PAHs were highest near sites of past or present human habitation. Few samples exceeded the 4,000 ng/g total PAH toxic-effect threshold [31]. These results are all consistent with the findings of Page et al. [2], who used a simple, although less robust, source-ratio mixing model to allocate PAHs to sources.

Diesel oil, diesel soot, creosote, and urban atmospheric dust were generally not found at appreciable levels in subtidal sediments, although creosote was occasionally found in sediments near pilings. Sediment cores showed that concentrations of combustion products increased when nearby human activity began, but these compounds were not found in appreciable amounts at sites far from shore. Fallout from a nearby large forest fire may have been a small contributor to 1991 embayment samples but not to deepwater sediments swept by the Alaska Coastal Current. Forest-fire fallout dominated sediment trap samples from the same time period and is somewhat similar in composition to urban atmospheric dust.

Data quality is important for reducing error in source allocation, and it can be increased by increasing sample sizes and decreasing preinjection extract volumes. Use of 50- to 100-g dry weight samples and 0.25-ml preinjection volumes proved practical for doing this [2; G.S. Douglas et al., personal communication].

The PCA and least-squares modeling procedure can be generalized to allocate PAHs or other hydrocarbons in other situations, including other spills. For each application, the hydrocarbon sources applicable to it would have to be identified.

Acknowledgements

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. MATERIALS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. CONCLUSIONS
  8. Acknowledgements
  9. REFERENCES

This study was funded by Exxon Company USA. We would like to thank Alice Sullivan for her many hours of assistance on the project and Gregg Douglas for his help on chemical analysis methods.

REFERENCES

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. MATERIALS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. CONCLUSIONS
  8. Acknowledgements
  9. REFERENCES
  • 1
    Maki, A. W. 1991. The Exxon Valdez oil spill: Initial environmental impact assessment. Environ. Sci. Technol. 25: 2429.
  • 2
    Page, D. S., P. D. Boehm, G. S. Douglas and A. E. Bence. 1995. Identification of hydrocarbon sources in the benthic sediments of Prince William Sound and the Gulf of Alaska following the Exxon Valdez oil spill. In P. G.Wells, J. N.Butler, and J. S.Hughes, eds., Exxon Valdez Oil Spill: Fate and Effects in Alaskan Waters. STP 1219. American Society for Testing and Materials, Philadelphia, PA, pp. 4183.
  • 3
    Page, D. S., P. D. Boehm, G. S. Douglas, A. E. Bence, W. A. Burns and P. J. Mankiewicz. 1996. The natural petroleum hydrocarbon background in subtidal sediments of Prince William Sound, Alaska. Environ. Toxicol. Chem. 15: 12661281.
  • 4
    O'Clair, C. E., J. W. Short and S. D. Rice. 1996. Petroleum hydrocarbon-induced injury to subtidal marine sediment resources, Exxon Valdez. In Oil spill, state/federal natural resource damage assessment. Final Report. National Oceanic and Atmospheric Administration, Juneau, AK, USA, pp. 182.
  • 5
    Bence, A. E. and W. A. Burns. 1995. Fingerprinting hydrocarbons in the biological resources of the Exxon Valdez spill area. In P. G.Wells, J. N.Butler, and J. S.Hughes, eds., Exxon Valdez Oil Spill: Fate and Effects in Alaskan Waters. STP 1219. American Society for Testing and Materials, Philadelphia, PA, pp. 84140.
  • 6
    Kvenvolden, K. A., F. D. Hostettler, J. B. Rapp and P. R. Carlson. 1993. Hydrocarbon in oil residues on beaches of islands of Prince William Sound, Alaska. Mar. Pollut. Bull. 26: 2429.
  • 7
    Bentz, A. P. 1976. Oil spill identification. Anal. Chem. 48: 454472.
  • 8
    U.S. Coast Guard. 1977. Oil spill identification system. National Technical Information System. NTIS ADA003803 (CG-D-41–75). Final Report. U.S. Coast Guard, Groton, CT.
  • 9
    Overton, E. B., J. A. McFall, S. W. Mascarella, C. F. Steele, S. A. Antoine, I. R. Politzer and J. L. Laseter. 1981. Identification of petroleum residue sources after a fire and oil spill. Proceedings, 1981 Oil Spill Conference, Atlanta, GA, USA, March 2–5. American Petroleum Institute, Washington, DC, pp. 541546.
  • 10
    Boehm, P. D., D. L. Fiest and A. Elskus. 1981. Comparative weathering patterns of hydrocarbons from the Amoco Cadiz oil spill observed at a variety of coastal environments. Proceedings, International Volume on the Amoco Cadiz: Fate and Effects of the Oil Spill, Brest, France, November 19–22, 1979. Centre National pour l'Exploration des Oceans, Brest, France, pp. 159173.
  • 11
    Volkman, J. K., R. Alexander, R. I. Kagi and J. Rullkotter. 1983. GC-MS characterization of C27 and C28 triterpanes in sediment and petroleum. Geochim. Cosmochim. Acta 47: 10331040.
  • 12
    Boehm, P. D. and J. W. Farrington. 1984. Aspects of the polycyclic aromatic hydrocarbon geochemistry of recent sediments in the Georges Bank region. Environ. Sci. Technol. 18: 840845.
  • 13
    Radke, M. 1987. Organic geochemistry of aromatic hydrocarbons. In J.Brooks, and D.Welte, eds., Advances in Petroleum Geochemistry, Vol. 2. Academic, Orlando, FL, USA, pp. 141205.
  • 14
    Page, D. S., J. C. Foster, P. M. Fickett and E. S. Gilfillan. 1988. Identification of petroleum sources in an area impacted by the Amoco Cadiz oil spill. Mar. Pollut. Bull. 19: 107115.
  • 15
    Peters, K. E. and J. M. Moldowan. 1993. The Biomarker Guide: Interpreting Molecular Fossils in Petroleum and Ancient Sediments. Prentice Hall, Englewood, NJ, USA.
  • 16
    Morrison, D. F. 1976. Multivariate Statistical Methods. McGraw-Hill, New York, NY, USA.
  • 17
    Dillon, W. R. and M. Goldstein. 1984. Multivariate Analysis: Methods and Applications. John Wiley & Sons, New York, NY, USA.
  • 18
    Krazanowski, W. J. and F. H. C. Marriot. 1994. Multivariate Analysis. Part 1. Distributions, Ordination, and Inference. John Wiley & Sons, New York, NY, USA.
  • 19
    Burns, W. A. and A. E. Bence. 1993. Fingerprinting polycyclic aromatic hydrocarbons (PAH) in subtidal sediments. Abstracts, 14th Annual Meeting, Society of Environmental Toxicology and Chemistry, November 14–18, Houston, TX, USA, p. 16.
  • 20
    Short, J. W. and R. A. Heintz. 1993. Qualitative and quantitative determination of Exxon Valdez crude oil in sediment samples using principal component analysis of hydrocarbon data. Abstracts, Exxon Valdez Oil Spill Symposium, February 2–5, Oil Spill Public Information Center, Anchorage, AK, USA, p. 63.
  • 21
    Boehm, P. D., D. S. Page, E. S. Gilfillan, W. A. Stubblefield and E. J. Harner. 1995. Shoreline ecology program for Prince William Sound, Alaska, following the Exxon Valdez oil spill: Part 2-Chemistry and toxicology. In P. G.Wells, J. N.Butler, and J. S.Hughes, eds., Exxon Valdez Oil Spill: Fate and Effects in Alaskan Waters. STP 1219. American Society for Testing and Materials, Philadelphia, PA, pp. 347397.
  • 22
    Bragg, J. R., R. C. Prince, E. J. Harner and R. M. Atlas. 1993. Bioremediation effectiveness following the Exxon Valdez oil spill. Proceedings, 1993 Oil Spill Conference, Tampa, FL, March 29-April 1. American Petroleum Institute, Washington, DC, pp. 435447.
  • 23
    Sauer, T. C. and P. D. Boehm. 1991. The use of defensible analytical chemical measurements for oil spill natural resource damage assessment. Proceedings, 1991 Oil Spill Conference, San Diego, CA, March 4–7. American Petroleum Institute, Washington, DC, pp. 363369.
  • 24
    Douglas, G. S., K. J. McCarthy, D. T. Dahlen, J. A. Seavy, W. G. Steinhauer, R. C. Prince and D. L. Elmendorf. 1992. The use of hydrocarbon analyses for environmental assessment and remediation. J. Soil Contam. 1: 197216.
  • 25
    Douglas, G. S., R. C. Prince, E. L. Butler and W. G. Steinhauer. 1994. The use of internal chemical indicators in petroleum and refined products to evaluate the extent of biodegradation. In R. E.Hinchee, B. C.Alleman, R. E.Hoeppel, and R. N.Miller, eds., Hydrocarbon Bioremediation. Lewis, Ann Arbor, MI, USA, pp. 219236.
  • 26
    U.S. Environmental Protection Agency. 1986. Test methods for evaluating solid waste: Physical/chemical methods, 3rd ed. SW-846. Washington, DC.
  • 27
    Manen, C. A., E. Robinson-Wilson, S. Korn and R. L. Britten. 1993. Management of Natural Resource Damage Assessment samples and analytical data. Program and Abstracts, Exxon Valdez Oil Spill Symposium, February 2–5, The Oil Spill Public Information Center, Anchorage, AK, USA, pp. 320321.
  • 28
    Sale, D. M., J. C. Gibeaut and J. W. Short. 1995. Nearshore transport of hydrocarbons and sediments following the Exxon Valdez oil spill. In Exxon Valdez oil spill, state/federal natural resource damage assessment. Final Report. National Oceanic and Atmospheric Administration, Juneau, AK, USA, pp. 149.
  • 29
    Wold, S., A. Ruhe, H. Wold and W. J. Dunn III. 1984. The collinearity problem in linear regression. The partial least squares approach to generalized inverses. SIAM J. Sci. Stat. Comp. 5: 735743.
  • 30
    Dunn, W. J. III, D. L. Stalling, T. R. Schwartz, J. W. Hogan, J. D. Petty, E. Johansson and S. Wold. 1984. Pattern recognition for classification and determination of polychlorinated biphenyls in environmental samples. Anal. Chem. 56: 13081313.
  • 31
    Long, E. R. and L. G. Morgan. 1990. The potential for biological effects of sediment-sorbed contaminants tested in the National Status and Trends Program. NOAA Technical Memorandum NOS OMA 52. National Oceanic and Atmospheric Administration, Rockville, MD, USA.
  • 32
    Neff, J. M. and W. A. Burns. 1996. Estimation of polycyclic aromatic hydrocarbon concentration in the water column based on tissue residues in mussels and salmon: An equilibrium partitioning approach. Environ. Toxicol. Chem. 15: 22202253.
  • 33
    Neff, J. M. and C. Henry. 1990. Pore water chemistry. In Excavation and rock washing treatment technology: Net environmental benefit analysis. Net Environmental Benefit Report. National Oceanic and Atmospheric Administration, Seattle, WA, USA, p. 66.
  • 34
    Neff, J. M., E. H. Owens, S. W. Stoker and D. M. McCormick. 1995. Shoreline oiling conditions in Prince William Sound following the Exxon Valdez oil spill. In P. G.Wells, J. N.Butler, and J. S.Hughes, eds., Exxon Valdez Oil Spill: Fate and Effects in Alaskan Waters. STP 1219. American Society for Testing and Materials, Philadelphia, PA, pp. 312346.
  • 35
    Lethcoe, J. and N. Lethcoe. 1989. Cruising Guide to Prince William Sound, Vol. 1. Prince William Sound Books, Valdez, AK, USA.
  • 36
    Bence, A. E. and G. S. Douglas. 1993. The natural petroleum hydrocarbon background in the benthic sediments of Prince William Sound, Alaska. Abstracts, Geological Society of America, October 25–28, Boston, MA, p. A151.
  • 37
    Lethcoe, J. and N. Lethcoe. 1994. History of Prince William Sound. Prince William Sound Books, Valdez, AK, USA.
  • 38
    Medred, C. 1991. After the fire. Land recovers quickly after large wildfire. Anchorage Daily News, Anchorage, AK, USA, June 2.
  • 39
    Kropp, K. G., J. A. Goncalves, J. T. Andersson and P. M. Fedorak. 1994. Bacterial transformations of benzothiophene and methylbenzothiophenes. Environ. Sci. Technol. 28: 13481356.