Identification of Exxon Valdez Oilin Sediments and Tissues fromPrince William Sound and theNorthwestern Gulf of Alaska Basedon a PAH Weathering ModelJ E F F R E Y W . S H O R T * A N D R O N A . H E I N T Z

National Marine Fisheries Service, NOAA, Alaska FisheriesScience Center, Auke Bay Laboratory, 11305 Glacier Highway,Juneau, Alaska 99801-8626

We used a first-order loss-rate kinetic model of polynucleararomatic hydrocarbon (PAH) weathering to evaluate 7767environmental samples collected for the Exxon Valdez oilspill (EVOS) for the presence of spilled oil. The model wasdeveloped from experiments with gravel coated with crudeoil and washed for 6 months. The modeled PAH includedthe 14 most persistent compounds of 31 analyzed by GC/MS. Parameters include loss-rate constants related to theenergy required for PAH to escape from petroleum anda quantitative index of weathering. The model accounts for91% of the temporal variability of modeled PAH concentra-tions. We compared the discrepancies betweenmeasured and model-predicted PAH concentrations ofEVOS samples with a probability distribution of these dis-crepancies derived from the experimental weatheringresults. Only 1541 field samples contained sufficient PAHfor valid application of the model; three-fourths fit themodel at R g 0.01 type I error, 9% fit an alternate modelcharacterized by the absence of weathering, 17% fitneither model, and a few fit both models. The 1164 totalsamples that fit the weathering model account for 86%of the summed PAH concentrations detected in all 7767samples. We conclude that first-order loss-rate kinetics ac-count for the dominant PAH weathering processes in theEVOS and that the rate of weathering is determined mainlyby the ratio of surface area to volume of petroleum inthe environment.

IntroductionIdentification of a petroleum pollution source based onchemical analysis of environmental samples is complicatedby compositional changes (weathering) during environmentalexposure. The effects of processes such as evaporation,dissolution, microbial degradation, and photo-oxidation thatcause weathering seem difficult to predict because they aresensitive to a variety of environmental conditions. Changescaused by weathering have restricted identification methodsto those that are based on persistent analytes in environmentalsamples (1) or that explicitly account for compositionalchanges, e.g., by comparison of analytical results fromenvironmental samples with petroleum samples that havebeen weathered artificially (2).

The above weathering processes imply mainly first-order(FO) loss-rate (LR) kinetics for the disappearance of poly-

nuclear aromatic hydrocarbons (PAH) from petroleum.Evaporation and dissolution involve FOLR kinetics explicitly,with rate constants determined by the enthalpy of vaporiza-tion through the Arrhenius equation. Similar FOLR kineticshave recently been demonstrated for microbial oxidation ofpetroleum PAH (3), where PAH dissolution is promoted bybiosurfactants secreted by microbes (4) that increase theeffective surface area of the petroleum phase. The kineticsthat characterize evaporation, dissolution, and microbialoxidation are thus qualitatively similar, and the distributionpattern of PAH that results from the combined effects of theseprocesses characteristically includes preferential losses of PAHthat have lower molecular weight and contain fewer alkylsubstituents (1, 5-9). In contrast, the kinetics of PAH photo-oxidation may be second order and autocatalytic (10). Photo-oxidation probably begins with photolytic generation ofmainly benzylic free radicals (11), which would result inpreferential losses of PAH that contain more numerous alkylsubstituents. However, at low ambient light fluxes, PAHphoto-oxidation rates may be small as compared with ratesfor evaporation and dissolution.

If the dominant PAH weathering processes includeevaporation, dissolution, and microbial oxidation but notphoto-oxidation, then similar overall FOLR kinetics for PAHresult. Although the absolute concentrations of PAH inpetroleum may depend in a complicated way on theenvironmental history of a sample, the similar FOLR kineticsimpose constraints on relative PAH concentrations. Theseconstraints may provide a basis for identification of a majorpetroleum source if alternative sources can be distinguishedconfidently.

We report here our experimental validation of a FOLRkinetic model to describe the disappearance of PAH fromgravel coated with experimentally-weathered petroleum andour subsequent evaluation of this model as a basis foridentifying petroleum spilled from the T/V Exxon Valdez (EV)in the environment. The EV oil spill (EVOS) provided a uniqueopportunity to evaluate the petroleum identification proce-dure we derived. Photo-oxidation was suppressed by thelow sun angle and persistent cloud cover characteristic of theaffected area. The EVOS area was nearly pristine before thespill (12, 13), and within the impacted area PAHs from naturalsources were mainly confined to subtidal sediments (14-16), so PAH interferences from sources other than the spillwere often minimal. Finally, the number of environmentalsamples collected and analyzed by consistent GC/MS methods(17) was exceptionally large (7767 analyses), so the EVOS maybe considered an especially large-scale field test for our kineticmodel.

Model Description and Parameter EstimationSuppose the rate of loss to the environment of a PAH (denotedas P) dissolved in petroleum follows FOLR kinetics, so that

The time dependence of the LR constant, k(t), derives fromthe variable exposure conditions of the petroleum in theenvironment. Writing k(t) as kf(t) and integrating eq 1 gives

where the value of the integral in eq 2 is indicated by aweathering parameter, w, which summarizes the exposurehistory of the petroleum volume element sampled.

* Corresponding author e-mail address: jshort@abl.afsc.noaa.gov;telephone: (907)789-6065; fax: (907)789-6094.

-d[P]

dt) k(t)[P] (1)

ln([P]0

[P] ) ) k∫0

tf(t) dt ) kw (2)

Environ. Sci. Technol. 1997, 31, 2375-2384

S0013-936X(96)00985-6 This article not subject to U.S. Copyright.VOL. 31, NO. 8, 1997 / ENVIRONMENTAL SCIENCE & TECHNOLOGY 9 2375Published 1997 by the American Chemical Society.

Equation 2 may be applied simultaneously to J PAH ineach of I samples as

where j specifies the PAH and i specifies the sample. Notethat each kj is the same for all samples and that each wi isthe same for all the PAH in a sample. If the initialconcentrations [Pij]0 in an environmental sample are known,the parameter kj and wi may be estimated from measurementsof the [Pij] in the samples (see below).

The initial PAH concentrations [Pij]0 may be estimatedfrom the initial amount and composition of the petroleumin a sample. The initial amount of petroleum introducedinto unit mass of the sampled environmental matrix can bedetermined if the petroleum contains temporally invariantanalytes, denoted here as j+ (assume the analyte set isrestricted to PAH). Denoting the proportion of the jth PAHin the unweathered petroleum as πj, then the concentrationci of unweathered petroleum originally introduced into theith sample is determined as the ratio of [Pij+] ()[Pij+]0) to πj+.The mean of these ratios may be used as an estimator of ci

where nj+ indicates the number of temporally invariant PAHincluded in the sum. Because [Pij]0 ) ciπj, eq 3 may beexpressed as

Note that the quantity indicated by dij may be determinedentirely from measurements of PAH concentrations in theunweathered petroleum and in an environmental sample.

Error minimization of eij ) dij - kjwi by least-squares leadsto

It can be shown that eq 5 implies that the vector k of elementskj may be identified with the dominant eigenvector of thematrix D′D, where the elements of D are dij. Imposition ofthe normalization condition ∑kj

2 ) 1 removes the indeter-minancy implied by eq 3, so that k may be identified with thefirst principal component of D′D, with corresponding eigen-value λ1 ) ∑wi

2. Thus, a principal component analysis ofD′D provides least-square estimates of kj, and the wi maythen be calculated from eq 5 and dij. The ratio of the dominanteigenvalue λ1 to the sum of the eigenvalues indicates theproportion of data variability explained by the model. Also,exp ( [(∑j>1

J λj)/IJ]1/2 is a measure of the root mean square fitof predicted and observed PAH concentrations.

The distribution of ∑jej2 provides a basis for evaluating the

probability that the source of PAH in an environmental sampleis weathered EV oil (EVO). Given k and wi, predicted PAHconcentrations may be calculated for sample i by eq 3a. Theagreement of predicted and measured PAH concentrationsin the ith sample may be expressed as the mean of the squareddifferences of logarithms of measured PAH and predictedPAH or mean square error (MSEi):

The quantity MSEi′ may be calculated similarly for PAH ofuncertain origin in an environmental sample i′ and comparedwith the MSEi distribution developed by the bootstrap methodapplied to PAH measurements of experimentally weatheredEVO samples. An estimate of the probability that the PAHin the environmental sample i′ is consistent with theexperimentally weathered petroleum may then be made onthe basis of this comparison.

If PAH concentrations can be predicted on the basis of analternative PAH source, a similar approach may be used toestimate the probability that PAH in an environmental sampleis consistent with that source, provided the alternative meansquare error distribution is available. If the alternative PAHsource is not subject to weathering, so that relative PAHconcentrations do not change with time, then PAH propor-tions of the sum of the PAH measured provide a basis forprediction. Equation 6 may be used to calculate the alterna-tive mean square error MSEi′, which can be compared withthe MSEi′ distribution calculated by the bootstrap methodfrom known samples containing PAH from only the alternativesource. An estimate of the probability that the PAH in theenvironmental sample is consistent with the alternative PAHsource may then be made on the basis of this comparison.

MethodsPetroleum Sources and Composition. The petroleum spilledfrom the EV was produced from the Alaskan North Slope(ANS) oil fields in 1989, and the petroleum used in theweathering experiment was produced from the same fieldsin 1992 and 1993. The PAH analytes we considered are listedin Table 1, with proportions of these PAH analytes inunweathered petroleum from the cargo of the EV as deter-mined by the hydrocarbon analysis methods used herein (17).Also in Table 1 are the PAH analytes we selected for modeling,based on analyte persistence above method detection limits(MDL; see below) during the experimental weathering period.The proportions of selected PAH in unweathered EVO givenin Table 1 are taken as the πj in developing the weatheringmodel.

Petroleum Weathering Experiments. Petroleum wasweathered by continuously washing petroleum-coated gravelfor 6 months. The petroleum was heated at 70 °C overnightto 80% initial mass to remove volatiles and then sprayed ontotumbling gravel. Four different loadings of petroleum ongravel were prepared in 1992 and three in 1993, when theweathering experiment was repeated. The 1992 loadings were55.2, 622, 3130, and 4510 µg of petroleum/g of gravel andwere 281, 717, and 2450 µg of petroleum/g of gravel in 1993.Each 11-kg preparation of petroleum-coated gravel wasweathered by washing it continuously in a tube withalternating freshwater and 30‰ seawater, switching every 6h, at flow rates of 150 mL/min. Water temperatures rangedfrom 2 to 12 °C. Further details on the methods used todetermine petroleum loadings on gravels, gravel preparations,and weathering apparatus and procedures are presentedelsewhere (18).

Gravel samples were composited for PAH analysis from8-15 replicate preparations of each petroleum loading. Ateach sampling, equal numbers of gravel pieces were removedfrom each replicate tube, mixed, and stored at -20 °C untilanalysis. Gravel samples were collected 5, 62, 90, and 180days after the columns were filled in 1992 and 3, 41, 68, and181 days after the columns were filled in 1993. Duplicatecomposite samples were collected at day 62 in 1992; otherwise,single composite samples were collected. The -20 °C storagedid not affect the recovery of PAH, based on comparison withresults for this petroleum stored at +20 °C.

1

J∑j)1

J [ln ([P]ij

[P]ij)]2

)1

J∑j)1

J

(dij - dij)2 ) MSEi (6)

ln ([Pij]0

[Pij] ) ) kjwi (3)

ci )1

nj+∑j+εJ

[Pij+]

πj+

(4)

dij ) lnciπj

[Pij]) kjwi (3a)

kj )

∑j)1

I

widij

∑i)1

I

wi2

wi )

∑j)1

J

kjdij

∑j)1

J

kj2

(5)

2376 9 ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 31, NO. 8, 1997

The PAH content of samples was determined by a GC/MSmethod. PAH were extracted with dichloromethane andpurified by alumina/silica gel column chromatography fol-lowed by size-exclusion high-performance liquid chroma-tography. Purified PAH were measured by MS operated inthe selected ion monitoring mode. Concentrations of PAHin the dichloromethane extracts were determined by theinternal standard method based on a suite of deuterated-PAH internal standards. Four quality control samples wereanalyzed with each batch of 12 samples, including tworeference samples, a method blank, and a method blankspiked with certified hydrocarbon standards obtained fromthe National Institute of Standards and Technology (NIST).Method detection limits of hydrocarbon analytes weredetermined experimentally (19) and were generally 1 ng/g.At the Auke Bay Laboratory (ABL), the accuracy of this analysisis better than about (15% based on comparison with NISTvalues, and precision expressed as coefficient of variation isabout 25% for the analytes in the weathering model, basedon reference sample results (see data analysis). Additionaldetails of the method have been presented elsewhere (17).

Exxon Valdez Oil Spill Study Area. The EVOS introducedsome 35 500 t of ANS petroleum into Prince William Sound(PWS), which traveled about 750 km southwest along theKenai Peninsula and through Shelikof Strait before dispersinginto the northern Gulf of Alaska, oiling about 1750 km ofshoreline along the way (20). The path followed by the spilledpetroleum conformed with the Alaska Coastal Current (ACC)(Figure 1), which flushes PWS through Hinchinbrook Entranceand exits through Montague Strait (21). Sea-surface tem-peratures of the affected area typically range from near 0 to16 °C annually. The ACC transports sediments burdenedwith PAH from natural sources into PWS (16). The patternof relative PAH concentrations characteristic of sediments

transported by the ACC into PWS has been found consistentlyat intertidal stations near Hinchinbrook Entrance (13, 16)and has often been found subtidally within PWS (13-16). AtConstantine Harbor off Hinchinbrook Entrance, concentra-tions of selected PAH in intertidal sediments have beenconstant at least since 1977 and probably much longer (13).

After the EVOS, 3433 samples of sediments, 2150 samplesof mussels (Mytilus trossulus), and 2184 samples of other

TABLE 1. PAHs Determined in Environmental Samples Collected for EVOS and PAH Proportions in Petroleum Spilled from the T/VExxon Valdez in Prince William Sound on March 24, 1989a

PAHs abbreviation πj(×103) coeff of variation (%) k

naphthalene 0.724 7.412-methylnaphthalene 1.33 6.621-methylnaphthalene 1.02 6.17biphenyl 0.183 12.0C-2 naphthalenes 3.15 19.4acenaphthylene 0.0139 8.99acenaphthene 0.0174 20.0C-3 naphthalenes C3-N 2.35 11.3 0.659 (0.653, 0.706)C-4 naphthalenes C4-N 0.598 23.8 0.148 (0.040, 0.215)fluorene 0.0911 17.7C-1 fluorenes 0.225 27.8C-2 fluorenes C2-F 0.191 34.2 0.118 (0.003, 0.188)C-3 fluorenes C3-F 0.151 50.7 0.082 (0.013, 0.147)dibenzothiophene 0.195 16.3C-1 dibenzothiophenes C1-D 0.417 16.7 0.433 (0.392, 0.462)C-2 dibenzothiophenes C2-D 0.570 19.1 0.188 (0.144, 0.258)C-3 dibenzothiophenes C3-D 0.481 35.4 0.056 (0.019, 0.119)phenanthrene 0.255 13.4anthracene <0.001 19.1C-1 phenanthrenes* C1-P 0.755 20.1 0.512 (0.456, 0.585)C-2 phenanthrenes* C2-P 0.892 12.0 0.126 (0.086, 0.181)C-3 phenanthrenes* C3-P 0.558 15.5 -0.027 (-0.068, 0.011)C-4 phenanthrenes* C4-P 0.166 41.1 -0.024 (0.055, 0.005)fluoranthene 0.00909 18.5pyrene 0.0147 12.6C-1 fluoranthenes/pyrenes 0.0716 16.8benz[a]anthracene <0.001 20.0chrysene C0-C 0.0492 20.5 0.041 (0.015, 0.064)C-1 chrysenes C1-C 0.0802 23.3 -0.051 (-0.068, -0.034)C-2 chrysenes C2-C 0.106 38.4 0.036 (0.007, 0.076)C-3 chrysenes 0.0362 43.2

a Abbreviations are listed for the 14 PAHs that persisted above MDL (see text) during the weathering experiments, and the correspondingproportions by weight (×103) are taken as the πj for the weathering model. Also listed are coefficients of variation for these PAH in reference samplesanalyzed at the Auke Bay Laboratory (N ) 102) and the estimated loss-rate (LR) constants k, with the range of the central 95% of bootstrappedestimates within parentheses (see text). An asterisk (*) indicates that it includes anthracene homologues.

FIGURE 1. Area affected by the Exxon Valdez oil spill of March 1989.Arrows indicate the path of the Alaska Coastal Current.

VOL. 31, NO. 8, 1997 / ENVIRONMENTAL SCIENCE & TECHNOLOGY 9 2377

tissues were collected and analyzed for PAH to support thenatural resource damage assessment efforts of the state andfederal governments. ABL staff collected most of thesesediments and mussels, using dichloromethane-rinsed ap-paratus, and stored them in precleaned glass jars fitted withpolytetrafluoroethane cap liners at -20 °C until analysis.Procedures used to collect the remaining samples were usuallysimilar. The PAH analysis methods used (17) were identicalwith those summarized above for the petroleum weatheringexperiment.

Database Archive. All of the PAH results from fieldsamples together with estimates of model parameters for thisreport and sample collection information are contained inthe EVOS of 1989 State/Federal Trustee Council HydrocarbonDatabase (EVTHD) at the ABL and available from the authors.

Data AnalysisPetroleum Weathering Experiments. We could not applythe weathering model to all the PAH initially present in EVObecause progressively more PAH were below MDLs as thepetroleum weathered during the experiment. We thereforeapplied the model to the most persistent PAH selected fromthe five most prominent PAH-homologue groups in EVO:naphthalenes, fluorenes, dibenzothiophenes, phenanthrenes,and chrysenes. Nine of the selected PAH were present aboveMDL in all 32 samples collected during the petroleumweathering experiments and provided 288 observations ofPAH for the model. Another five PAH were simultaneouslypresent in 26 of the 32 samples, and the 14 PAH simultaneouslypresent in these 26 samples provided 364 observations ofPAH. This combination of 14 PAH in 26 samples providedthe maximum number of simultaneous PAH observationspossible. We therefore applied the weathering model to theJ ) 14 PAH (identified in Table 1) simultaneously present inI ) 26 of the 32 samples collected during the petroleumweathering experiment.

We calculated the initial oil concentration parameter (ci)for each sample based on the modeled chrysene homologues,which are persistent in weathered crude oil (5). The sum ofchrysene, C-1 chrysenes, and C-2 chrysenes concentrationsdid not change significantly with time in 1992 or 1993(repeated measures ANOVA; P > 0.05) and therefore wereused to estimate ci from eq 4. Gravimetric determinationsof petroleum initially applied to the gravel used in theweathering experiments were linearly related to ci (r 2 ) 0.86,P ) 0.02) and were about 60% lower after correcting forvolatility losses. We calculated wi for the 26 samples and kj

for the 14 PAH from the 364 PAH observations obtained fromthe petroleum weathering experiment by principal compo-nent analysis of the matrix D′D, after transformation of PAHobservations to the matrix elements dij of D.

Bootstrapped distributions of the MSEi and kj weresimultaneously constructed by Monte Carlo simulation. Oneof the 26 samples was removed randomly, and a new matrixD* was calculated from the PAH observations of 26 randomselections with replacement from the 25 remaining samples.New LR constants were calculated by finding the eigenvectork* of D*′D*, and the MSEi was calculated for the removedsample from eq 6 using k*. This process was repeated 500times, and the 14 kj* and MSEi were recorded for each iteration.The distributions of the 14 kj* are presented as the range ofthe central 95% of the bootstrap results for each kj (Table 1).The distribution of MSEi is used to estimate the probabilitythat PAH in environmental samples are consistent with PAHfrom weathered ANS petroleum.

Characterization of PAH in Sediments from NaturalSources. We based our model of relative PAH concentrationsthat characterize the natural PAH source on samples collectedfrom Constantine Harbor. The relative PAH concentrationpattern and its error distribution are derived from 15 intertidal

sediment samples collected during six samplings in 1989 and1990. Because total PAH (TPAH, i.e., the sum of the PAHanalyzed) did not vary significantly (ANOVA, P > 0.23) amongthese samples (13), the sample with the median value of TPAHwas selected arbitrarily as representative of the characteristicPAH pattern, and an error distribution for this pattern wasgenerated by comparing the remaining samples with thisrepresentative sample as follows. The same 14 PAHs used inthe weathering model were each converted to proportionsby dividing each PAH concentration for a sample by the TPAHof the sample. We denote these PAH proportions in themedian sample ıj and a different sample i as pıjj, and pij,respectively, and calculate MSEi′ for discrepancies amongthese proportions as

A bootstrapped distribution for MSEi′ was constructed by aMonte Carlo simulation analogous to that used to constructthe MSEi distribution (see above).

Hypothesis Testing. We used the error distributions forthe weathering model and the natural sediment PAH sourceas the basis for distinguishing these sources in environmentalsamples. All the environmental samples with concentrationsof the 14 selected PAH above MDL were consecutively fit toboth models, and MSE and MSE′ were calculated for eachsample by eqs 6 and 7, respectively. The probability that thesource of PAH in the kth sample was consistent with ANSpetroleum, denoted here as Proil(k), was determined bysubtracting from 1 the percentile of values eMSEk in thecumulative frequency distribution of MSEi for the weatheringmodel. The null hypothesis that the PAH pattern in the kthsample was consistent with ANS petroleum was rejected whenProil(k) < R, where R specifies the probability of type I error.Similarly, Prns(k), the probability that the pattern of PAH inthe kth sample was consistent with the natural source, wasderived by comparing MSEk′ to the cumulative frequencydistribution of MSEi′. The null hypothesis that the kth sampleis consistent with the natural sediment pattern was rejectedwhen Prns(k) < R. Note that the two models must beconsidered as separate alternatives rather than simultaneously[as by, e.g., SIMCA (22) methods], because the models are notisolinear, so they cannot be combined into the same principalcomponent matrix.

ResultsWeathering Model Parameters. The eigenvector of kj

calculated as the first principal component of the matrix D′Daccounts for 86% of the total variability in the PAH data fromthe petroleum weathering experiments. The mean unex-plained variability per sample and per analyte is 0.161,indicating that most of the PAH values predicted by theweathering model are within -33% and +49% of observedvalues from the petroleum weathering experiments. Threeanalytes, C-4 naphthalenes and C-2 and C-3 fluorenes,together account for 73% of the magnitude of the secondprincipal component of the D′D matrix, and this componentaccounts for an additional 8% of the total PAH variability.The mean variability of logarithmically transformed data persample and per analyte that is not explained by the first twoprincipal components is 0.069, which implies a mean coef-ficient of variation of 26% for the untransformed data,consistent with analytical precision. The first two compo-nents thus account for all the data variability except analyticalerror, and the first component accounts for 91% [i.e., 86/0.94%] of the explainable data variation. The second com-ponent summarizes the discrepancies between the weatheringmodel and the data after discounting variability due toanalytical error.

1

14∑j)1

J

(ln [pıjj/pij])2 ) MSEi′ (7)

2378 9 ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 31, NO. 8, 1997

The values of kj increase with decreasing alkyl substitutionand with the number of aromatic rings (Table 1). The largestvalues of kj are for C-3 naphthalenes, C-1 dibenzothiphenes,and C-1 phenanthrenes, whereas the smallest values are forC-3 and C-4 phenanthrenes and the three chrysene homo-logues. The proximity to zero of the kj of these latter fiveanalytes indicates that the duration of the weatheringexperiments was insufficient for appreciable weathering lossesof them to occur. Also in Table 1 are the ranges for the mostcentral 95% of the bootstrap estimates of the kj as a measureof dispersion of the estimates. This dispersion is proportion-ally least for those analytes that changed most during theweathering experiments.

The values for wi increase linearly with time during thepetroleum weathering experiments and increase faster atlower petroleum loadings (Figure 2). In the 1992 experiment,these linear trends are significant (P < 0.027) for all but thelowest petroleum loading (P < 0.09). Values for wi rangeabove 7 for the most heavily weathered, lowest petroleum-loaded gravel at 85 days exposure but do not exceed 3.5 forthe most heavily loaded gravel at 175 days. The weatheringrate, dwi/dt, increases linearly with decreasing petroleum

loadings (r 2 ) 0.98; P < 0.01) in the 1992 weatheringexperiments. Similar trends occur in the 1993 experiment,but the more limited 1993 data preclude a meaningfulstatistical summary.

The MSEi distribution derived from the Monte Carlosimulation is strongly leptokurtic, with a median of 0.145.The 99th percentile occurs at MSEi ) 0.98, which is usedbelow to evaluate samples from the EVOS. A comparison ofobserved and predicted PAH proportions that correspondwith the median MSEi for a weathered (wi ) 3.95) exampleis depicted in Figure 3B, where the PAH proportions of theunweathered EVO are also presented for further comparison(Figure 3A).

Characterization of PAH in Sediments from ConstantineHarbor. The PAH of Constantine Harbor intertidal sedimentsare proportionally lower in naphthalenes and diben-zothiophenes and higher in phenanthrenes and chrysenes ascompared with EVO (Figure 3C). The distribution of MSEi′derived from Monte Carlo simulation has a median of 0.056and is less strongly leptokurtic than the MSEi distribution ofweathered EVO. This median is equivalent to a coefficientof variation of 24% for untransformed data, consistent with

FIGURE 2. Regression relations of weathering parameters (w) and time (days) at four loadings of petroleum on gravel used in the petroleumweathering experiments.

FIGURE 3. (A) Normalized PAH proportions of unweathered EVO. (B) Predicted and observed normalized PAH proportions of weatheredEVO for wi ) 3.95 and the median MSEi of 0.145. (C) Normalized PAH proportions of sediments from Constantine Harbor, where thin verticalbars indicate the range of the central 95% of results from the bootstrap distribution about the median indicated by the thick vertical bars.Normalized PAH proportions sum to unity.

VOL. 31, NO. 8, 1997 / ENVIRONMENTAL SCIENCE & TECHNOLOGY 9 2379

analytical precision. The 99th percentile occurs at MSEi′ )0.34, which is used below to evaluate samples from the EVOS.

Classification of Sediment Samples from the EVOS. Theresults of our classification procedure indicate that althoughEVO did not contaminate most of the EVOS sediment samplescollected, EVO was the source of most of the PAH detected.Concentrations of the 14 PAH included in the weatheringmodel are above MDL in 996 of the 3433 sediment samplesanalyzed for the EVOS. Of these 996 samples, 618 have MSEk

<0.98 and MSEk′ >0.34, which we accepted as consistentwith weathered EVO (Figure 4). The PAH concentrationsdetected in these 618 samples account for more than 86% ofthe sum of detected PAH concentrations in sediments.

The sediment samples we classified as contaminated byEVO may also contain PAH from other contamination sources.The median MSEk of the 618 EVO-contaminated samples was0.34, or more than twice the median MSEi derived from theMonte Carlo simulation. The larger value of the median MSEk

may be caused by PAH from other sources that alter therelative PAH proportions and consequently fit the weatheringmodel less well. This is most evident for samples that containrelatively low total PAH concentrations. Of the 618 EVO-contaminated samples identified, the median MSEk for the255 of these samples that have total PAH concentrations lessthan 750 ng/g (dry weight) is 0.47, compared to a median of0.25 for the remaining 363 samples. The distribution of theMSEk values for these 363 samples is strongly leptokurtic andsimilar to the MSEi distribution.

Most of the EVO-contaminated sediments we identifiedwere collected from the intertidal and shallow subtidal withinthe EVOS impact area (Figure 1). Epibenthic surface depthwas reported for 546 of these samples, of which 93% werecollected above 20 m subtidal depth within the EVOS area.Another 5.3% were collected from subtidal depths below 20m within the EVOS area, and 1.7% were collected outside theEVOS area.

Another 110 evaluated samples had MSEk >0.98 and MSEk′<0.34, which we accepted as consistent with PAHs derivedfrom the natural PAH source (Figure 4). The PAH concen-trations detected in these samples account for less than 0.06%of the sum of detected PAH concentrations in sediments.The median MSEk′ of these 110 samples was 0.17, or morethan three times the median of the MSEi′ distribution. As

with the EVO-contaminated sediments, this larger value ofthe median MSEk′ may be caused by PAH from other sourcesthat alter the relative PAH proportions of environmentalsediment samples and thereby fit the Constantine Harborpattern less well.

Over 80% of the sediments we identified as contaminatedby PAH from the natural source were collected from subtidaldepths below 20 m within the EVOS impact area (Figure 1).Most of the remainder of these sediments were collected fromthe Constantine Harbor intertidal, east of the EVOS impactarea.

Thirty of the remaining sediment samples evaluated fitboth the weathering model and the Constantine Harbor PAHpattern, and 238 fit neither. The 238 samples that fit neitherPAH pattern include 41 samples of sediment trap filtratesthat were contaminated during sample collection (23) andfive samples of EVO that were so diluted for analysis that C-1dibenzothiophenes were detected just above MDL but wellbelow concentrations predicted by the weathering model,which caused the poor fit. The PAH in the remaining samplesof this category together with the 30 samples that fit bothpatterns account for 0.33% of the total in sediments and mayinclude mixtures of PAH from natural sources, the EVOS, andother, unknown sources.

Most of the 2437 sediment samples that could not be fitto the weathering model contained TPAH concentrations <90ng/g. The absence of chrysenes in 689 of these samples withdetectable PAH indicates that diesel oil accounted for no morethan 0.5% of the sum of detected PAH concentrations ofunclassified samples. Only 1.6% of these summed concen-trations were in samples collected from outside the spill zoneor from depths below 20 m, whereas >90% were in 35 samplescollected from the spill zone above 20 m epibenthic depth.These 35 samples were visibly contaminated with oil but hadsample masses <100 mg, suggesting over-dilution thatresulted in concentration estimates for some of the 14modeled PAH below sample mass-adjusted MDLs. Samplesmissing any of the PAH with the three largest kj wereconsidered too weathered to fit to the model, and theseaccount for 1.4% of summed PAH concentrations of unclas-sified samples, of which 99% were collected in the spill zone.Another 70 samples, collected in the spill zone and knownto be contaminated by diesel exhaust during sampling,

FIGURE 4. Source classification of PAH in environmental samples as a proportion of samples (solid bars) and as a proportion of summedPAH concentrations (open bars) for sediments, mussels, and other tissues. The numbers of samples are listed above the solid bars indicatingproportions of samples. Sources include EVO, petroleum spilled from the T/V Exxon Valdez; Constantine Harbor, representation of the naturalsediment PAH source; neither, mixtures or other unknown sources; both, samples that are ambiguously classified; and not considered,samples with PAH below MDL. For source classification criteria, see text.

2380 9 ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 31, NO. 8, 1997

account for 2.5% of the summed concentrations of theunclassified samples.

Classification of Mussel Samples from the EVOS. As withsediments, the results of our classification procedure indicatethat although EVO did not contaminate most of the musselsamples collected, EVO was the source of most of the PAHdetected. Concentrations of the PAH included in theweathering model are above MDL in 452 of the 2150 musselsamples analyzed for the EVOS, of which 435 met the EVOclassification criteria (Figure 4). The sum of all the PAHconcentrations detected in these 435 samples is >84% of thePAH concentrations in mussels. The MSEk distribution ofthese samples is strongly leptokurtic, with a median of0.25sabout 75% greater than the median of the MSEi

distribution. In contrast with sediment samples, the medianMSEk for sets of EVO-contaminated mussels varies little,regardless of the minimum PAH content of the sample set.

The smaller median MSEk for mussels as compared withsediments suggests that mussels were less subject to PAHcontamination from sources other than the EVOS. Inparticular, the smallest MSEk′ of mussels is 0.57, indicatingthe absence of the Constantine Harbor sediment-PAH patternin mussels. Also, no mussel simultaneously fit the weatheringmodel and the Constantine Harbor pattern. However, 17mussel samples fit neither PAH pattern and account for 0.95%of the PAH concentrations in mussels (Figure 4).

All but three of the EVO-contaminated mussel sampleswe identified were collected from within the EVOS impactarea: two had ambiguous location information reported, andone was reported as collected from eastern PWS.

Most of the 1698 other mussels that could not be evaluatedcontained low PAH concentrations. The median TPAHconcentration in these unclassified samples was <135 ng/g,and these unclassified samples account for 15% of the sumof detected PAH concentrations in all the mussel samplesanalyzed (Figure 4). Only eight samples had all of the modeledPAH except for the chrysenes, indicating that diesel oilaccounts for <2.0% of the summed PAH concentrations ofunclassified samples. More than 95% of the summed PAHconcentrations observed in the unclassified samples werecollected from the spill zone, of which 22% were too weatheredto fit the model.

Classification of Other Tissue Samples from the EVOS.Samples of other tissues are classified as EVO contaminatedby the weathering model much less frequently than sedimentsand mussels. Concentrations of the PAH in the weatheringmodel are above MDL in 93 of the 2184 other tissue samplesanalyzed for the EVOS, of which 80 met the EVO classificationcriteria (Figure 4). The PAH concentrations detected in these80 samples account for 34% of the PAH concentrations in

other tissues. Most of these 80 samples were from externalsurfaces of oiled animals. Another 12 samples did not fiteither model and account for 42% of detected PAH, most(>90%) of which is due to three stomach content samplesfrom bald eagles. One sample fit both models. The remaining23% of the detected PAH is distributed among the 2091samples of other tissues that could not be evaluated, usuallyat concentrations near MDLs. Nearly half of the summedPAH concentrations of these unclassified samples were insamples that contained all of the modeled PAH except thechrysenes, suggesting diesel oil contamination, of which 85%is from 96 samples collected by a single project.

Time-Dependence of the Weathering Parameter in EVO-Contaminated Sediments and Mussels. The weatheringparameter wi is weakly correlated with the sample collectiondate of EVO-contaminated sediment or mussel samples (r 2

) 0.045, P < 0.001; Figure 5). Values of wi ranged from nearzero to >7 during each of the 6 years following the EVOS,which together with the small proportion of variation in wi

explained by the sample collection dates indicate that theeffect of time on weathering varied considerably.

DiscussionAssessment of First-Order Weathering Kinetics for Experi-mentally Weathered EVO. The principal component analysisof the petroleum weathering experiments indicates threefactors that determine the observed PAH variability. Thesethree factors are FOLR kinetics, analytical error, and remainingvariability summarized by the second principal component,which we denote as process error. Because performance ofany model is constrained by the analytical error, that errormust be estimated in order to evaluate model performance.We accept the results of laboratory analysis of referencesamples as the basis for the estimation of precision becausethese samples include matrix effects, and the large numberof repetitive analyses records variability over a period of years.Because the squared coefficient of variation for PAH inreference sample results is equivalent to the variance oflogarithmically transformed PAH results, these variationcoefficients may be used to calculate the approximateanalytical error variance expected. On this basis, the mean-ingful principal components are limited to the first two: FOLRkinetics and process error.

Comparison of the eigenvalues of the first two principalcomponents shows that the process error component is atmost a minor perturbation of the FOLR process. Thecomposition of the process error component indicatessystematic experimental errors rather than incorrect speci-fication of the weathering model. The three largest PAHconstituents of the process error component, C-4 naphtha-

FIGURE 5. Weathering parameter wi for EVO-contaminated sediments and mussels versus sample collection time t (days) after the EVOS.

VOL. 31, NO. 8, 1997 / ENVIRONMENTAL SCIENCE & TECHNOLOGY 9 2381

lenes and C-2 and C-3 fluorenes, could be due to unknownanalytical interferences or to composition differences betweenEVO and ANS petroleum. The composition constants πj inour weathering model are derived from EVO, but the ANSpetroleum we used was produced 3 years after the EVOS andmay have somewhat different PAH composition due tovariable contributions from different ANS oil fields (24). Theweathering model accounts for 94% of the PAH variabilitywhen the composition constants πj are derived from thecomposition of the ANS petroleum initially applied to thepetroleum weathering gravels, and the remaining variabilityis due to analytical error. We conclude that PAH for theweathering experiments follows FOLR kinetics, within thelimitations imposed by analytical and systematic errors.

PAH vaporization from petroleum may be considered asan endothermic chemical reaction that involves breaking thecohesive bonds between PAH solutes and the petroleumsolvent. The physical rate-limiting step (RLS) implied by FOLRkinetics is the energy required to overcome the attractive vander Waals forces between the petroleum phase and departingPAH molecules that constitute these bonds. This energyrequirement is approximately equal to the enthalpy ofvaporization, which is proportional to molecular surface area.The Arrhenius rate equation gives the relation among rateconstant k, activation energy Ea, and temperature as k ) Aexp -(Ea/RT). The linearity of a plot of ln k versus Ea derivedfrom observations of aqueous dissolution rates of PAH frompetroleum is evidence of a similar RLS for PAH vaporizationand aqueous dissolution. An Arrhenius plot of ln kj versusestimates of total molecular surface area (TSA), used here asa surrogate for enthalpies of vaporization (which are notavailable for PAH vaporization from petroleum), is ap-proximately linear (r 2 ) 0.75, P < 0.005; Figure 6). Thislinearity corroborates initial separation of PAH from thepetroleum phase as the RLS, regardless of the nature of thephase receiving the PAH lost from the petroleum. Thisexplains why the weathering model performs equally wellwith EVO in subtidal sediments, where the receiving phaseis aqueous, and with intertidal sediments and mussels, wherethe receiving phase may at times be the atmosphere.

An Arrhenius plot of logarithms of FOLR constants reportedfor a petroleum-weathering field experiment conducted athigher temperatures (Table 3 in ref 3) is also linear (r 2 ) 0.90,P < 0.001; Figure 6) but has a significantly (P < 0.001) lessslope, which implies less variability with TSA among theserate constants as compared with ours. The linearity cor-

roborates the proposed RLS, but the smaller slope is due toexpected temperature effects on the rate constants impliedby the Arrhenius rate equation. Differences among rateconstants increase with decreasing temperature when Ea isindependent of temperature, and these differences areexacerbated in this case by incipient crystallization of PAHin petroleum at temperatures near 0 °C, which would increasevaporization enthalpies of larger PAH as compared withwarmer temperatures. These differences indicate that therelative LR constants presented herein should not be appliedto appreciably different (i.e., J10 °C) thermal environmentswithout correction for these temperature effects, for whichaccurate data on enthalpies of vaporization of PAHs frompetroleum as a function of temperature would be helpful.

Weathering Model and PAH Source Identification. Ourweathering model is related to currently accepted protocolsfor oil-spill source identification based on PAH analysis (25,26). The protocols currently adopted in the United Statesand in Europe compare normalized PAH results for samplesand suspected sources, and patterns that match withinconstraints imposed by analytical precision and by weatheringeffects are accepted as evidence implicating the suspectedsource. The constraints imposed by weathering effectsinclude decreasing trends in pattern discrepancies withincreasing PAH boiling points (26) or with increasing alkylsubstitution within homologous PAH series (25). Thosenormalized PAHs identified as unaffected by weathering canbe included in multivariate statistical comparisons withcorresponding results from suspected oil-spill sources toevaluate whether discrepancies that remain among theseanalytes can be ascribed to analytical precision. The iden-tification procedure thus employs two criteria: patterns ofnormalized PAH that match within the constraints of analyti-cal precision for analytes that are not affected by weatheringand patterns of PAH weathering losses that conform withspecified trends.

Our weathering model may be regarded as an alternativeformulation of the above protocols. It provides an explicitmathematical specification of the PAH weathering-loss trends,and the PAHs included in the matching procedure areextended to weathered PAH. The weathering model reducesto a simple comparison of relative PAH concentrations in asample i and in a suspected source oil as wi approaches zero.As wi increases, progressively more PAH are significantlyaffected by weathering, depending on comparison of theproduct kjwi and analytical precision for the jth PAH expressed

FIGURE 6. Arrhenius plot of logarithms of rate-loss constants (k) from (A) the petroleum weathering experiment and from (B) an independentfield experiment (3) versus total molecular surface area (TSA) for selected PAH. Estimates of TSA are nm2 derived from data presentedin ref 33. Both sets of rate-loss constants are normalized so that ∑kj

2 ) 1.

2382 9 ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 31, NO. 8, 1997

as a coefficient of variation. The European protocol (26)makes greater use of the information produced by thechemical analysis, because all the resolvable isomer peaksare considered individually, in contrast to summation ofisomer peaks for each homologue reported, as was done here.Although this is a substantial advantage of the Europeanprotocol, our model could be adapted to such protocols asa possible refinement.

An advantage of the weathering model is its provision fora precise definition of weathering, sensu Wang and Fingas(27). The weathering parameter wi defines weathering byindexing the relative abundance of a set of PAH with knownLR constants, so that comparisons among samples can beunambiguously controlled for weathering, which leads to adistribution for MSEi that is independent of weathering state.Given a distribution for MSEi derived from laboratoryobservations, source identifications can be evaluated byestimates of the probability of committing type I error. Inour model, a type I error is an erroneous rejection of the nullhypothesis that the pattern of PAH in a sample is consistentwith EVO.

Successful application of our weathering model requirescareful consideration of the analyte set selected, the detectionlimit definition chosen, and the effects of these choices onthe scaling of wi. Detection limit stipulation is critical becausethe value calculated for the fit of the model to the data (i.e.,MSEi) will increase dramatically if contributions from analyteswell below detection limits are included. The choice ofanalytes and detection limits thus determines the range ofweathering states covered by the model. The model mayconsequently fail to apply to samples that contain very highconcentrations of petroleum if it is very weathered, becausethe dilution necessary for valid analysis of the most abundantanalytes may cause the most weathered analytes to fall belowthe detection limits applied.

The choice of analytes also affects the ability of the modelto distinguish among initial stages of petroleum weathering,because most of the information regarding wi is containedin the measurements of the more rapidly lost analytes.Consequently, a model that includes rapidly lost analytes(e.g., naphthalene) will distinguish among earlier weatheringstages better than one that does not, but the latter will beapplicable over a broader range of weathering states, becauseits constituent analytes are more persistent. Also, theweathering states that correspond to a particular value of wi

will not be the same for these two models due to thenormalization condition ∑kj

2 ) 1. This condition causes thevalue of wi to be a function of the analytes included in themodel, so wi values based on different analyte sets cannot becompared directly.

The ratio of surface area to volume of petroleum in asample is an important factor affecting the relation of wi andtime. The decrease of dwi/dt as the film thickness ofpetroleum applied to the gravel substrate increases isconsistent with the RLS for PAH loss from petroleum discussedabove. However, the film thicknesses in our experimentsranged from about 1 to 100 µm, and evaporation rates of theless volatile components of thicker petroleum films increaseslowly with increasing relative surface area (28, 29), probablydue to increasing resistance to mass transfer within thickerpetroleum films. These results imply that a variety ofweathering states may be observed shortly following an oilspill, depending on the surface area to volume ratio of thepetroleum sampled, hence the weak correlation of wi andtime (Figure 5).

Assessment of First-Order Weathering Kinetics forPetroleum Spilled from the T/V Exxon Valdez. The ap-plicability of the laboratory-derived weathering model to fieldresults from an oil spill may be assessed by comparing theerror distributions derived from applying the weatheringmodel to laboratory results versus field results and by

comparing the geographic distribution of samples identifiedas contaminated by EVO with the geographic boundaries ofthe area contaminated by the EVOS.

The error distributions that result from applying theweathering model to field samples of mussels and sedimentsconfirm that the dominant weathering processes of this oilspill followed FOLR kinetics. The median MSEk of 0.25 forEVO-contaminated field mussels means that most of the PAHconcentrations predicted by the weathering model fall within-40% to +65% of observed concentrations. This indicatesthat the weathering model is almost as successful at predictingPAH concentrations in field mussels as it is at predicting PAHconcentrations in experimentally weathered petroleum. Thegreater disparity between observed and predicted PAHs inmussels is probably due to the combined effects of smallinterferences from other hydrocarbon sources in the envi-ronment or introduced during sample collection or storage,small PAH composition differences between the petroleumused for the weathering experiment and the petroleum spilled,and composition differences induced by differences in thethermal histories of the spilled petroleum and the petroleumused for the weathering experiments. These effects maycollectively be regarded as small perturbations as comparedwith the much larger PAH concentration changes that resultfrom FO weathering processes.

The similarly leptokurtic distributions of the MSEk valuesof mussels and the MSEis of experimentally weatheredpetroleum, and the fact that the distribution for musselsincludes nearly all the modeled mussels, corroborate thesimilarity of the underlying weathering processes for EVO.The 17 mussels that were modeled as not consistent withEVO at the 1% type I error rate are probably the result oftruncation of the MSEi distribution after the median value isincreased to 0.25.

The similarity of the error distributions derived frommussels and from the more contaminated sediments indicatesthat the weathering model applies equally well for thesematrices. The median MSEk for mussels is almost identicalwith the median for sediments identified as EVO-contami-nated at total PAH concentrations >750 ng/g, and bothdistributions are similarly leptokurtic. Thus, sediment samplesthat are sufficiently contaminated by EVO so that otherhydrocarbon sources are negligible in comparison displaypatterns of relative PAH concentrations consistent with FOLRkinetics. That the same model produces similar results forsuch disparate environmental matrices further validates thekinetics of the underlying weathering processes assumed bythe model. Following the EVOS, PAH losses due to weatheringfollowed FOLR kinetics regardless of the great variability ofenvironmental conditions among intertidal mussels, intertidalsediments, subtidal sediments, and EVO-contaminated sedi-ments in transport from the intertidal to the subtidal, atgeographic locations of very different aspects. This generalityderives from the simple notion that the relative rate of PAHloss from petroleum is determined by the energy required forPAH molecules to escape from petroleum.

The geographic distribution of the samples we identifiedas contaminated by EVO is generally consistent with thetrajectory of the spilled petroleum, indicating an absence ofspurious identifications generated by the weathering model.This is supported by the low frequencies of mussel or sedimentsamples identified as EVO contaminated that were collectedoutside the trajectory of the spilled petroleum or that werecollected at sediment epibenthic depths below about 20 m.Also, no mussel or sediment sample collected just beforelandfall of the spilled EVO was identified as EVO contami-nated.

The opportunity provided by the EVOS was particularlyappropriate for assessment of the weathering model becausethe most heavily contaminated compartments of the affectedarea were among those least affected by PAH inputs from

VOL. 31, NO. 8, 1997 / ENVIRONMENTAL SCIENCE & TECHNOLOGY 9 2383

alternative sources. Before the EVOS, the PAHs characteristicof EVO were rarely detected in mussels outside Port Valdezin PWS (12, 13). Sediment PAH concentrations derived fromnatural sources decrease with progressively shallowerepibenthic depths, so that total PAH concentrations fromthese sources rarely exceed 100 ng/g in intertidal sedimentsand 200 ng/g in subtidal sediments to 20 m depth in the spillarea (14, 16). Weathered EVO was most prevalent in thesetwo environmental compartments: mussels and sedimentsat less than 20 m epibenthic depth. As a result, PAHconcentration patterns characteristic of weathered EVO areubiquitous in mussels that contain sufficient PAH to beevaluated by the weathering model and common in sedimentsthat have total PAH concentrations J750 ng/g.

The absence of PAH from natural sediment PAH sourcesin mussels together with the observation that the PAH patternthat characterizes these sources does not weather place strongconstraints on the nature of these sources. These sourceshave been identified with natural petroleum seeps along thesouthern coast of Alaska at Katalla and elsewhere (16), butthis identification is not obviously consistent with the absenceof weathering and with the absence of these PAH in mussels,which implies that these PAH are sequestered in such a waythat biological availability is precluded. The absence ofweathering and of biological availability is most clearly evidentat Constantine Harbor, where concentrations of PAH mostsusceptible to weathering have not changed in intertidalsediments during a 15-year monitoring period and were rarelydetected in adjacent mussels simultaneously collected (12,13). Strong association of PAH with sediments has beenproposed (30), but minerals capable of PAH adsorption havenot been identified in these sediments. The pattern of PAHconcentrations of unburned coal is difficult to distinguishfrom petroleum in sediments (31), so coal eroded from theBering River coalfield north of Katalla (32) remains analternative possibility. The source of these PAH, therefore,remains to be established definitively.

AcknowledgmentsThe research described in this paper was supported by theExxon Valdez Oil Spill Trustee Council. However, the findingsand conclusions presented by the authors are their own anddo not necessarily reflect the views or position of the TrusteeCouncil.

Literature Cited(1) Wang, Z.; Fingas, M.; Sergy, G. Environ. Sci. Technol. 1994, 28,

1733-1746.(2) Killeen, T. J.; Eastwood, D.; Hendrick, M. S. Talanta 1981, 28,

1-6.(3) Venosa, A. D.; Suidan, M. T.; Wrenn, B. A.; Strohmeier, K. L.;

Haines, J. R.; Eberhart, B. L.; King, D.; Holder, E. Environ. Sci.Technol. 1996, 30, 1764-1775.

(4) Singer, M. E.; Finnerty, W. R. In Petroleum Microbiology; Atlas,R. M., Ed.; Macmillan Publishing Co.: New York, 1984.

(5) Roques, D. E.; Overton, E. B.; Henry, C. B. J. Environ. Qual. 1994,23, 851-855.

(6) Michel, J.; Hayes, M. O. In Proceedings of the 1993 InternationalOil Spill Conference; American Petroleum Institute: Washington,DC, 1993.

(7) Kennicutt, M. C., II. Oil Chem. Pollut. 1988, 4, 89-112.(8) Boehm, P. D.; Steinhauer, M. S.; Green, D. R.; Fowler, B.;

Humphrey, B.; Fiest, D. L.; Cretney, W. J. Arctic 1987, 40 (Suppl.1), 133-148.

(9) Humphrey, B.; Boehm, P. D.; Hamilton, M. C.; Norstrom, R. J.Arctic 1987, 40 (Suppl. 1), 149-161.

(10) Majewski, J.; O’Brien, J. Environ. Lett. 1974, 7 (2), 145-161.(11) Hansen, H. P. Rapp. P.-V. Reun. Cons. Int. Explor. Mer 1977, 171,

101-106.

(12) Karinen, J. F.; Babcock, M. M.; Brown, D. W.; MacLeod, W. D.,Jr.; Ramos, L. S.; Short, J. W. Hydrocarbons in Intertidal Sedimentsand Mussels from Prince William Sound, Alaska, 1977-1980:Characterization and Probable Sources; U.S. Department ofCommerce, Auke Bay Lab: Juneau, AK, 1993; NOAA TechnicalMemorandum NMFS-AFSC-9.

(13) Short, J. W.; Babcock, M. M. Am. Fish. Soc. Symp. 1996, 18, 149-166.

(14) O’Clair, C. E.; Short, J. W.; Rice, S. D. Am. Fish. Soc. Symp. 1996,18, 61-93.

(15) Short, J. W.; Sale, D. M.; Gibeaut, J. C. Am. Fish. Soc. Symp. 1996,18, 40-60.

(16) Page, D. S.; Boehm, P. D.; Douglas, G. S.; Bence, A. E. In ExxonValdez Oil Spill: Fate and Effects in Alaskan Waters; SpecialTechnical Publication 1219; Wells, P. G., Butler, J. N., Hughes,J. S., Eds.; American Society for Testing and Materials: Phila-delphia, PA, 1995; Special Technical Publication 1219; pp 41-83.

(17) Short, J. W.; Jackson, T. J.; Larsen, M. L.; Wade, T. L. Am. Fish.Soc. Symp. 1996, 18, 140-148.

(18) Marty, G. D.; Short, J. W.; Dambach, D. M.; Willits, N. H.; Heintz,R. A.; Rice, S. D.; Stegeman, J. J.; Hinton, D. E. Can. J. Zool. 1997,75, 989-1007.

(19) Glaser, J. A.; Forest, D. L.; McKee, G. D.; Quave, S. A.; Budde, W.L. Environ. Sci. Technol. 1981, 15, 1426-1435.

(20) Wolfe, D. A.; Hameedi, M. J.; Galt, J. A.; Watabayashi, G.; Short,J.; O’Claire, C.; Rice, S.; Michel, J.; Payne, J. R.; Braddock, J.;Hanna, S.; Sale, D. Environ. Sci. Technol. 1994, 28, 561A-568A.

(21) Niebauer, H. J.; Royer, T. C.; Weingartner, T. J. J. Geophys. Res.1994, 99, 14,113-14,126.

(22) Wold, S.; Sjostrom, M. In Chemometrics: Theory and Application;Kowalski, B. R., Ed.; ACS Symposium Series 52; AmericanChemical Society: Washington, DC, 1977.

(23) Sale, D. M.; Gibeaut, J. C.; Short, J. W. Nearshore Transport ofHydrocarbons and Sediments Following the Exxon Valdez OilSpill, Exxon Valdez Oil Spill State/Federal Natural ResourceDamage Assessment Final Report (Subtidal Study Number 3B);Alaska Department of Environmental Conservation: Juneau, AK,1995.

(24) Bence, A. E.; Burns, W. A. In Exxon Valdez Oil Spill: Fate andEffects in Alaskan Waters; Wells, P. G., Butler, J. N., Hughes, J.S., Eds.; American Society for Testing and Materials: Philadel-phia, PA, 1995; Special Technical Publication 1219; pp 84-140.

(25) ASTM. In Annual Book of ASTM Standards, Section 11: Waterand Technology; 11.02 Water (II); ASTM: Philadelphia, 1996; pp835-845.

(26) NORDTEST. Nordtest Method; NT Chem 001, Ed. 2; NOR-DTEST: Espoo, Finland, 1991.

(27) Wang, Z.; Fingas, M. Proceedings of the 17th Arctic & MarineOilspill Program Technical Seminar; Environmental ProtectionService: Ottawa, 1994; pp 133-171.

(28) Fingas, M. Proceedings of the 17th Arctic & Marine OilspillProgram Technical Seminar; Environmental Protection Service:Ottawa, 1994; pp 189-212.

(29) Fingas, M. Proceedings of the 18th Arctic & Marine OilspillProgram Technical Seminar; Environmental Protection Service:Ottawa, 1995; pp 43-60.

(30) Page, D. S.; Boehm, P. D.; Gilfillan, E. S.; Bence, A. E.; Burns, W.A.; Mankiewicz, P. J. Proceedings of the 19th Arctic & MarineOilspill Program Technical Seminar; Environmental ProtectionService: Ottawa, 1996; pp 1195-1209.

(31) Tripp, B. W.; Farrington, J. W.; Teal, J. M. Mar. Pollut. Bull. 1981,12, 122-126.

(32) Anonymous. Alaska’s High-Rank Coals; Information Circular 33;Alaska Department of Natural Resources, Division of Geologicaland Geophysical Surveys: Fairbanks, AK, 1993.

(33) Pearlman, R. S.; Yalkowsky, S. H.; Banerjee, S. J. Phys. Chem. Ref.Data 1984, 13, 555-562.

Received for review November 25, 1996. Revised manuscriptreceived March 19, 1997. Accepted March 25, 1997.X

ES960985D

X Abstract published in Advance ACS Abstracts, May 15, 1997.

2384 9 ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 31, NO. 8, 1997