Introduction

Oil weathering processes, including evaporation, dissolution, and microbial oxidation, eventually remove even recalcitrant constituents in situ (1), while physical dispersion mediated by wave impacts accelerates these processes. But once oil from accidental spills percolates into stable, porous beaches, transformation through weathering in anoxic sediments is slowed (1). In sediments protected from physical dispersion, weathering rates are also considerably reduced. As a result, oil buried in anoxic sediments may retain toxic components such as polycyclic aromatic hydrocarbons (PAH) on time scales of years to decades (2-5). In contrast, our study of lingering oil from the 1989 Exxon Valdez spill demonstrates oil preservation in aerated sediments for more than a decade by a different mechanism. Evidence presented here suggests preservation can be a result of the viscous, often emulsified state of the oil when deposited.

The Exxon Valdez spill deposited ~20 000 metric tons of oil on beaches, with ~85% of this inside Prince William Sound (PWS) (6). Some was incorporated into porous sediments, occasionally to depths of 1 m or more (7). High annual rates of subsurface oil loss from beaches within PWS (~68% yr^-1 from 1991 to 1992; ref 7) initially implied that remaining subsurface oil would soon be reduced to negligible amounts (8). However, by 2001, ~100 tons of oil remained on beaches within PWS, mostly buried within the uppermost half-meter (9, 10), implying loss rates declined within PWS to ~20% yr^-1 during the period 1992-2001 (9). An unknown additional amount of oil remained elsewhere along the northern Gulf of Alaska (GOA) coastline (Figure 1) (11, 12). These trends imply ever slower loss rates for subsurface oil still present.

	Figure 1 Locations of beaches sampled in Prince William Sound and the Gulf of Alaska. Curved lines indicate spill progress in 1989.
	Figure 2 Estimated annualized loss rates of subsurface oil during the periods indicated.

Declining oil loss rates arise mainly from geomorphological variation among oiled beaches. Within the spill-affected region, a surface layer of cobbles to boulders often protects underlying finer-grained sediments from physical dispersion, thus providing the stability needed for oil persistence (11-13). Variation in substrate-conferred stability accounts for much of the variation in oil persistence at different sites, with the most vulnerable oil dispersed more readily. By 2001, subsurface oil presumably remained in especially well-protected sediments where future dispersion depends on abnormal disturbance events, and suggested loss rates even slower than 20% yr^-1. To test this, we evaluated spatial changes in the area contaminated with subsurface oil and associated oil mass from 2001 to 2005 on 10 beaches in PWS. We also analyzed the composition of oil remaining at sites inside PWS and outside along the coast of the Gulf of Alaska, to determine the extent of weathering in subsurface oil deposits.

Materials and Methods

Ten beaches were randomly selected from 42 PWS beaches that had subsurface oil in 2001 (Figure 1). At each beach, the upper intertidal was partitioned into contiguous rectangles bounded above and below by tidal elevation strata, and two quadrats were randomly located within each rectangle. To optimize the number of beaches and sampling intensity, larger rectangles were used in 2005 resulting in lower sampling density of quadrats (0.96 quadrats m^-1 of alongshore distance in 2001 compared with 0.3 in 2005). In the field, oil was detected visually or by smell, sampled, and the Exxon Valdez source confirmed by analysis of alicyclic biomarkers using identification criteria given in ref 14. Our estimates of oiled beach area derive from oil encounter rates, estimated as the ratio of oiled (k) and total (N) randomly located 0.25 m² quadrats excavated to ~0.5 m depth, within identical beach segments and tidal elevation strata both years. The area containing subsurface oil on each beach was estimated as the sum of the rectangle areas containing oiled quadrats, where only half the rectangle area was included if only one of the two quadrats contained oil (see ref 9 for details). To facilitate estimation of associated oil masses, we also classified the intensity of oiling as heavy [HOR], medium [MOR], and light [LOR] oil residues based on criteria given in ref 15. The products of oiled beach area and the mean amount of oil per unit area for each oil class were summed across oil classes to estimate the mass of oil remaining on each beach. One beach (Site 3, Table 1) had only a trace of oil in 2001 but light residues were found in 2005.

We used a Bayesian approach (16) to distinguish the loss rate of oil from sampled beaches, as measured by the disappearance of oiled quadrats with time, from the associ ated background of inherent sampling variability (i.e., sampling error). If the unknown proportions of a beach area that remained oiled in sampling years 2001 and 2005 (a span of 4 years) are denoted by p₁ and p₂, respectively, then p₂ p₁ and p₂ = p₁ × ⁴, with being the annualized survival rate of oil at the beach, i.e., = (p₂/p₁)^1/4. If N quadrats were sampled from the beach at a particular time when the unknown proportion oiled was p, then the observed number of oiled quadrats on the beach (k) is assumed to be a binomial random variable, i.e., k ~ binomial (p, N). If the numbers of quadrats sampled in 2001 and 2005 are denoted by N₁ and N₂, respectively, and the corresponding numbers of sampled quadrats that were oiled were k₁ and k₂, then the observed annualized survival rate is = [(k₂/N₂)/(k₁/N₁)]^1/4. This observed survival rate is subject to considerable sampling error when the portion of the beach included in the sample is fairly small as for the present study. Further, this observed survival rate is undefined when k₁ is zero and provides little information on the annualized loss rate (1 - ) when k₂ is zero. To better estimate the annualized survival rates among the beaches sampled, the data from all beaches are analyzed simultaneously under a single overarching model. The random changes in observed oiled proportions at the beaches (a subscript i is added to identify each particular beach) are modeled as normally distributed following logit transforma tion, specifically the model structure is logit(p_1i) = log [p_1i/(1 - p_1i)] = _i, logit(p_2i) = _i - , and _i is normally distributed with mean d and precision parameter (=1/²). Use of _i² ensures that the model is constrained to losses of oiled beach area only.

The Bayesian model was fitted to the data in Table 1 using Gibbs sampling (17), which provided estimates for the frequency distribution of the annualized change _i, from which the mean _i and the value corresponding to the 10th percentile of the distribution were estimated. This approach allowed for variation in oiled proportions among beaches within and between years. Uninformative priors were used for the unknowns, specifically d ~ normal(0, 10 ^{- 6}), ~ gamma(10 ^{- 3}, 10 ^{- 3}), _i ~ normal(0, 10 ^{- 5}), i = 1,...,10

Figure 3 Proportion of total PAH (upper panel) and total n-alkanes (lower panel) remaining in samples collected from Prince William Sound and elsewhere in the Gulf of Alaska from 1989 to 2005.

A burn-in of 50 000 samples was discarded, and the next 30 000 samples were used to describe the posterior distribu tion of the unknowns.

Comparison of oil compositional changes provides a more sensitive indication of oil loss rates, reflecting in situ weathering processes acting to remove oil components. Biodegradation typically acts first on n-alkanes, followed by branched and alicyclic alkanes and PAH (18). To assess the effects of these processes, we analyzed 43 samples of subsurface oil collected at intervals from seven locations along the GOA outside PWS from 1989 through 2005, and compared these with 38 samples collected from inside PWS in 2001 (Figure 1). We estimated the proportions of total n-alkanes (n-alk) and total PAH (PAH) remaining by comparing concentrations normalized to the chrysenes (the most persistent PAH readily detectable among those analyzed) in the samples with normalized concentrations in oil collected from the sea surface 11 days following the spill. The ratio of these normalized concentrations provides an indication of hydrocarbon persistence relative to the chrysenes, discount ing evaporative losses, which were nearly complete by 11 days (19). Normalization is necessary to account for variable inclusion of sediments and water in the oil samples.

Figure 4 Aliphatic chromatograms of Exxon Valdez oil collected from (A) the sea surface 11-day following the incident, (B) subsurface sediment in Prince William Sound, Alaska, 12 years later, and (C) subsurface sediments from Cape Douglas in the Gulf of Alaska, 16 years later. Symbols: C no., n-alkane with no. of carbon atoms; d-C no., deuterated n-alkane standard; DCH, dodecycylohexane internal standard; Pr, pristane; Ph, phytane.

Figure 5 Total ion current chromatograms of PAH in Exxon Valdez oil collected from (A) the sea surface 11-day following the incident, (B) subsurface sediment in Prince William Sound, Alaska, 12 years later, and (C) subsurface sediments from Cape Douglas in the Gulf of Alaska, 16 years later. Symbols: d-no., deuterated PAH; HMB, hexamethylbenzene internal standard.

Normal alkane analytes included C₁₀ through C₃₄ analyzed by gas chromatography with a flame ionization detector, and PAH included 2- to 4-ring un- and alkyl-substituted homologues analyzed by gas chromatography with mass selective detector (GCMS; see ref 20 for analysis details). Alicyclic biomarkers were determined by GCMS following ref 14.

Results and Discussion

Direct comparison of results from 2001 and 2005 indicates a slight apparent increase of oiled beach area and associated oil masses (Table 1). Mean oil encounter probabilities increased from 0.049 in 2001 to 0.063 in 2005. The estimated cumulative oiled beach area and oil mass similarly increased from 950 to 1260 m² and from 931 to 938 kg, respectively, increasing nearly as often as decreasing at individual beaches (Table 1). The percent of oiled quadrats classified as HOR, MOR, LOR was 6, 41, and 53% in 2001, compared with 7, 20, and 73% in 2005, a change that was not statistically significant (² = 2.09, P = 0.35, df = 2).

The apparent increases in oiled beach areas and masses between 2001 and 2005 reflect sampling error because no crude oil had been added to these beaches, and the high viscosity, high adhesion, and low concentration of the oil in these sediments preclude substantial diffusive spreading (10, 21). For these reasons, we assumed that the actual proportion of beach area that is oiled, p, is constrained to decline monotonically with time.

Our analysis shows that the most likely rate of decline of oiled beach area within PWS from 2001 to 2005 is 3-4% yr^-1 (1 - ; Table 1). The probability that the actual decline rates exceed 10% per year is less than about 10% (1 - , Table 1). These rates are considerably slower than those noted prior to 2001 (Figure 2), confirming that physical dispersion rates have since slowed.

Degradation rates of PAH and n-alkanes were comparable at the GOA sites, but the rates of n-alkane degradation at the PWS sites were generally higher. The n-alk at the GOA sites often contained ~50% or more of the burden in the 11-day oil, whereas it was usually below 10% at the PWS sites (Figure 3). However, well-preserved n-alkanes and PAH persisted for over a decade at some sites both inside and outside PWS (Figures 4 and 5), and oil in samples containing these alkanes remains semi-liquid and highly adhesive on contact with dry surfaces. The PAH at the GOA sites were even more persistent; with one exception (Cape Gull), subsurface oil retained from ~25-75% of the PAH through 1999 that were present in the 11-day oil, and were still ~25-50% by 2005 (Figure 3). Somewhat lower PAH were evident in the 2001 PWS samples, which ranged from ~10-50%. Overall, these results show remarkable preservation of n-alkanes at some sites, and imply PAH degradation rates on the order of ~3-10% per year at most sites.

Although anoxia may account for long-term persistence of biodegradable oil components elsewhere (2, 3), it is unlikely to be a factor at our study sites. Intertidal sediments of PWS and the GOA usually contain <5% total organic carbon (22), and sedimentary oil refuges armored by boulders are stable, porous and well-flushed by ~5 m tides and wave exposure, especially in the GOA (11, 12, 23). Low concentrations of nitrate in seawater during warm seasons (24) may also retard biodegradation. Extensive degradation at some sites (e.g., Cape Gull, Figure 3) suggests that cool temperatures are a minor factor, because ocean surface temperatures are highly inter-correlated across the spill-affected region (25), yet extensive degradation occurred at some sites.

More importantly, emulsification leading to "mousse" formation inhibited weathering processes directly (11, 26), and caused the viscosity of the oil to increase substantially as it was transported along the GOA coast. Within PWS during the early stages of the spill, less viscous oil readily percolated into the generally finer-grained beaches (10). As the spill progressed along the GOA, the often greater porosity of more exposed beaches allowed the increasingly viscous oil to percolate into them as well (11, 12). Inherent with the trapped, viscous oil parcels, low ratios of surface area to volume provided comparatively little habitat for oil-degrading microbes. Low oil surface to volume ratios may also have played a role in the persistence of n-alkanes following oil spills elsewhere (2, 27).

Our results indicate that the remaining subsurface oil may persist with little change for decades, even in sediments that are not anoxic. Such persistence can pose a contact hazard to intertidally foraging sea otters, sea ducks, and shorebirds (10, 28), create a chronic source of low-level contamination (29), discourage subsistence in a region where use is heavy, and degrade the wilderness character of protected lands (11, 12).

Acknowledgment

This study was supported in part by the Exxon Valdez Oil Spill Trustee Council and the Prince William Sound Regional Citizens' Advisory Council, but the findings and conclusions are those of the authors and do not necessarily reflect the views or positions of these organizations.