Sea otter population dynamics and the ExxonValdez oil spill: disentangling the confounding effects

Authors


Summary

  • 1 Oil that spilled after the grounding of the Exxon Valdez in 1989 killed large numbers of sea otters in western Prince William Sound, Alaska, USA. However, our boat-based counts of sea otters during 1990–96 at oiled sites were as high or higher than boat-based counts in the same area in the early 1980s.
  • 2 Another study reported a significant decline in sea otter numbers after the spill, but our reanalysis of that data produced results very similar to ours. Counts of otters were higher than pre-spill counts in the oiled area; the only detectable decline was in the northern part of the sound, outside the area of oiling.
  • 3 We suggest that otter numbers in the western sound may have been increasing during the late 1980s, masking the loss due to the spill. Direct evidence for such an increase is lacking because no counts were conducted during this period. However, for several years after the spill pup production was higher than normal, which, if characteristic of the period immediately pre-spill, could have spurred a population increase.
  • 4Heightened pup production may have been caused by increased food supplies: after the spill, otters obtained more and larger clams per dive and spent less time feeding per day than in the early 1980s.
  • 5 We postulate that in the early 1980s clams were still recovering from the uplift caused by the 1964 earthquake, which resulted in massive clam mortality and habitat change in the western sound. Lingering effects of previous catastrophic events, like the earthquake and even 19th-century fur harvests, have hampered attempts to assess the impacts of the oil spill on sea otter population dynamics. The effects of uncontrolled and unreplicated environmental incidents, even major disasters, may be difficult to assess because of confounding factors.

Introduction

The once common belief that populations steadily rise to a stable equilibrium, or carrying capacity, and then remain at a more or less constant size, has been challenged repeatedly. Andrewartha & Birch (1954) showed that stochastic events can depress populations such that they never attain a stable equilibrium. Their examples dealt mainly with insects, whose populations tend to be especially variable, and some of their cases may have represented anomalies rather than generalities; gradually their ideas fell into disfavour. More recently, the concept that populations may be periodically and significantly perturbed has re-emerged (Sousa 1984), prompting greater consideration of such events in population dynamics (Lewin 1983; Mangel & Tier 1993, 1994; Brook & Kikkawa 1998; Root 1998). Evidence of sudden, severe (i.e. catastrophic) change is now available for many invertebrates (Dungan, Miller & Thomson 1982; Lessios, Robertson & Cubit 1984; Fiori & Cazzaniga 1999), small vertebrates (Blaustein, Wake & Sousa 1994; Valone, Brown & Jacobi 1995; Everson et al. 1999), large mammals (Young 1994) and even whole ecosystems (Botkin 1990; Lockwood & Lockwood 1993; Hughes 1994). Examples of natural environmental disturbances that can severely disrupt animal populations include fires (Singer et al. 1989), volcanic eruptions (Franklin et al. 1985), storms (Spiller, Losos & Schoener 1998) and the El Niño Southern Oscillation (Bodkin & Jameson 1991; Pounds & Crump 1994; Wright et al. 1999). While often calamitous, these events provide opportunities to investigate the dynamics of natural systems.

Human-caused environmental disruptions demand particular scrutiny because of their legal, political and economic ramifications. Site-specific population trends, background noise, and lingering effects of past perturbations can obscure, compound or supplant the impacts of human disturbances (Osenberg et al. 1994; Thrush, Pridmore & Hewitt 1994; Underwood 1994). Consequently, population changes often cannot be ascribed unequivocally to anthropogenic causes (Pechmann et al. 1991; Pascual & Adkison 1994; Dayton et al. 1998; Hofmann & Powell 1998). Moreover, due to confounding factors, populations subjected to even the most blatant human-caused environmental disasters may respond contrary to expectation (Baker et al. 1996).

Our study examined the demographic consequences of a major oil spill on a population of sea otters Enhydra lutris L., a species known to be especially sensitive to contamination of its pelage (Geraci & Williams 1990). The spill occurred in March 1989, when, after filling with Prudhoe Bay crude oil, the tanker Exxon Valdez ran aground in Valdez Arm, Prince William Sound (PWS), Alaska, USA (Fig. 1). Approximately 41 million litres of oil leaked into the water, much of which floated southward and was washed ashore on islands in the western sound.

Figure 1.

Study areas (shaded) in western Prince William Sound, Alaska, USA. Population dynamics in the sound varied by geographical area. We divided the sound into three regions, as shown: western (oiled), northern (un-oiled area with deep fjords), and eastern (un-oiled area with shallow bays and the best food supply).

Studies of the effects of this spill have spanned, quite literally, plankton to whales. Despite extraordinary funding for post-spill studies (Kaiser 1999), investigators often faced great difficulty in distinguishing spill-related effects from background noise (Wiens & Parker 1995; Hilborn 1996; Shaw & Bader 1996). In many cases, different investigators studying the same species reached disparate conclusions (Wheelwright 1994; Wells, Butler & Hughes 1995).

Sea otters became a target of media attention (Batten 1990) and litigation (Estes 1991) after the spill, as a conspicuously large number died or were debilitated by the oil. Our study focused on counts of live otters that remained. We hypothesized that their numbers would be significantly reduced and that recovery to pre-spill levels could be delayed by long-term depression of their reproduction or survival caused by the oiling of habitat and prey. Unexpectedly, in the 2 years following the spill, we counted as many or more otters in some of the most heavily oiled areas of PWS as were observed 5 years before the spill (Johnson & Garshelis 1995). These high counts seemed inconsistent with studies that estimated losses of > 2500 otters in PWS, or about 40% of the population (Garrott, Eberhardt & Burn 1993; DeGange, Doroff & Monson 1994), and warned of prolonged injurious effects (Ballachey, Bodkin & DeGange 1994).

In this paper we try to reconcile these contradictory assessments. We examined counts conducted since our previous report (Johnson & Garshelis 1995) plus new data on pup production, food habits and activity time budgets. We also reanalysed count data from another study whose conclusions differed markedly from our study. We attempted to resolve the discrepancies and unexpected results by teasing out potentially confounding factors.

Methods

Study area

PWS, in south-central Alaska, contains the most northerly population of sea otters. They were extirpated from all but the south-western corner of the sound during fur harvests in the 1800s. Protection in the early 1900s enabled the population to grow and eventually re-occupy the entire area. The few records from the early 1900s indicate that sea otters were present at Montague Island in the south-western part of the sound by the 1930s (Williams 1938) and neighbouring Green Island by the 1950s (Lensink 1962). However, the northern and eastern parts of the sound were not occupied until the 1970s (Garshelis, Garshelis & Kimker 1986; Johnson & Garshelis 1995); thus, during the 1970s and 1980s population dynamics and ecological relationships in the northern and eastern sound differed from those in the longer-occupied western sound.

Habitats in the sound also differ by geographical area. The western sound is characterized by large islands, the northern sound by deep glacial fjords, and the eastern sound by shallower bays and the best food resources (Garshelis, Garshelis & Kimker 1986). Oil from the Exxon Valdez covered much of the western sound, but did not impact shorelines of the northern or eastern sound. Based on these differences in habitat and oiling, we divided the sound into three geographical regions (Fig. 1).

We initially chose study sites in both the western and eastern sound where otters had been counted prior to the spill. Locations in the un-oiled eastern sound were used mainly as reference sites to test our methodology (Johnson & Garshelis 1995). This paper presents data from five study sites in the western sound: Applegate Rock, Knight Island, Green Island, Naked Island and Montague Island (Fig. 1). Otters were counted along the entire shoreline of each main island and accompanying smaller islands (although counts were incomplete in some years, as noted), except Montague Island, where counts were made only in Stockdale Harbor and Port Chalmers, a portion of the island that had a high density of otters.

Study sites differed in the amounts of oil to which they were exposed, depending on their location and orientation. We quantified the extent of oiling at each site based on the proportional length of shoreline oiled and the width of the band of oil on each stretch of shoreline during the summer of 1989 (for exact methodology see Johnson & Garshelis 1995). On a scale of 0–400 (400 representing heavy oil on 100% of the shoreline, 200 representing heavy oil on half the shoreline or light oil on the entire shoreline), oil indices at our study sites were as follows: 225 at Applegate Rock, 136 at Knight Island, 131 at Green Island, 20 at Naked Island, and 0 at Montague Island (i.e. Stockdale Harbor and Port Chalmers were un-oiled).

Observations of foraging otters were conducted only at Green and Knight Islands and observations of activity were made only at Green Island. All foraging and activity data at Green Island were obtained in and around Gibbon Anchorage, a partially enclosed breeding and pup-rearing area on the north-west side of the island (Fig. 1), some of which had been heavily oiled. At Knight Island foraging observations were made at several oiled and un-oiled locations.

Surveys of abundance and pup production

This study was designed as a pre-spill vs. post-spill comparison, so methods followed those of pre-spill observers. The most recent pre-spill data on otter abundance were from (i) Johnson (1987) and A. M. Johnson & A. R. DeGange (unpublished data), who conducted several boat-based counts of sea otters at Green Island, Applegate Rock and Montague Island (i.e. Stockdale Harbor and Port Chalmers) during 1977–85, and (ii) Irons, Nysewander & Trapp (1988), who conducted a single boat-based count in all of PWS during the summers of 1984 and 1985. These investigators differentiated pups accompanied by their mother from independent otters, enabling calculation of a pup ratio (pups/total otters other than pups), which is an indicator of pup production.

Both groups of pre-spill observers surveyed from boats, but they used different survey techniques. Johnson (1987) attempted to count all otters at a site, whereas Irons, Nysewander & Trapp (1988) travelled 100 m offshore and counted otters as well as seabirds, primarily within the 100 m-wide swath on either side of their boat; beyond these boundaries they tallied otters that they saw, but they did not deviate from their route to count groups further offshore. We used Johnson’s (1987) counting technique at Green Island, Applegate Rock and Montague Island, because he made more counts at these sites than Irons, Nysewander & Trapp (1988) and because his surveys of each site encompassed a larger area. Although the actual area surveyed by Johnson (1987) was not specified in his report, from discussions with him and participation in some of his surveys, we set an offshore limit of 1500 m. Johnson (1987; A.M. Johnson, unpublished data) also made separate counts of Gibbon Anchorage, which were not included in his Green Island counts. His boundaries of this area were employed in our study. Johnson (1987) did not survey Knight Island or Naked Island, so at these sites we followed the methodology of Irons, Nysewander & Trapp (1988), except we counted only otters (not birds).

Counts were made during the summers (May–August) of 1990–96, except 1992. During 1990, counts were made at Green Island, Applegate Rock and Montague Island. In 1991 these same sites plus Knight Island were surveyed. During 1993, counts were conducted at Naked Island and portions of Green and Knight Islands, and in 1994 counts were made at Green Island, Applegate Rock and portions of Knight Island. Only Gibbon Anchorage was counted in 1995. In 1996 complete counts were made at all sites.

Surveys were started in the morning (generally 07.00–08.00) and completed by mid–late afternoon. Surveys were postponed, terminated or not included in the analysis if weather conditions interfered significantly with counting (heavy rain, fog, wave height > 0·6 m, or Beaufort Wind Scale ≥ 3). Each of the study sites was surveyed in a single day, except Knight Island, which took 7–13 consecutive days.

The types of boats used during our surveys differed from those of pre-spill observers. Johnson (1987) stood in a 5–6-m open boat without an elevated platform. Our counts were made from 13–21-m boats with 3-m high viewing platforms. Because of the better visibility afforded by the elevated platform, we maintained a course about 500 m offshore, whereas the smaller lower boat used by Johnson (1987) necessitated a more zigzag course to cover the same area. In shallow areas that were difficult to navigate with the larger boat, we counted from a 4·5-m inflatable raft in which we could not stand up. In 1996, this raft was used to make one complete survey of Green Island to compare against a count of the same area from the larger boat conducted the next day.

At Knight and Naked Islands, Irons, Nysewander & Trapp (1988) used 5–10-m cabin cruisers, with an observer standing on each side. In these areas we followed the same course as Irons, Nysewander & Trapp (1988) (100 m offshore) but used a 4·5-m inflatable raft in which we could not stand. We estimated visually the 100-m wide survey area, but periodically recalibrated our distance estimate with a measured rope. All survey craft travelled at 6–19 km h−1 (3–10 knots), but stopped as necessary to count groups of otters.

We used two, or infrequently three, observers, who scanned for otters with 8–10 × binoculars. Pups were distinguished from independent otters by the following criteria: small size and close association with a larger otter; being carried, fed or groomed by a larger otter; the inability to swim or dive; and (or) swimming with a larger otter but making short, shallow, dives, often with the tail breaking the water surface.

The location of each otter or group of otters was plotted on maps. For some comparisons at Knight and Naked Islands, we used counts of otters within certain segments of shoreline, as defined by Pitcher (1975). Shoreline areas were measured using Atlas GIS software (Strategic Mapping Inc., Santa Clara, CA) on digitized US Geological Survey maps (1 : 63360) or aerial photographs (Aeromap, Anchorage, Alaska, AK).

Observations of activity

Activity data were obtained by scan-sampling (Altmann 1974; Estes, Jameson & Rhode 1982) from shore during daylight hours (05.00–23.00), May–August 1990 and July–August 1991. We used the same model 50–80 × telescope, the same procedure, and the same viewing locations near Gibbon Anchorage as Johnson (1987), who collected activity data during 1974–84. Scans were conducted in a slow methodical manner, in an effort to categorize the activity of all otters in the viewing area, including those diving for food; a single scan across a viewing area required about 30 min. Otters with pups were tallied separately from lone otters, but pups were not included in the activity data.

Activities were categorized as feeding, resting, swimming, grooming or interacting. Otters that were not feeding or resting were later pooled into a single category (other activities) for comparison with data collected by Johnson (1987). The numbers of otters in each activity category were summed for 2-hr time blocks, and the proportion calculated from the total number observed in each time block. Pre-spill data from April had been grouped with May–June (and could not be separated), so comparisons were made between the April–June pre-spill data and our 1990 data from May–June. No pre-spill data were collected in July, so data from this month were excluded from the analyses presented here. Pre-spill data for August had been grouped with September–October, so comparisons were made between pre-spill August–October and post-spill August 1990 and August 1991.

Observations of foraging

Foraging observations were conducted in or near Gibbon Anchorage during May–August 1990 and July–August 1991. Otters were observed using the same techniques and from many of the same viewing locations as used by pre-spill observers, Johnson (1987) and Garshelis (1983). Foraging data were also obtained at Knight Island (July–August 1991), where no pre-spill foraging observations had been conducted. Observations were made opportunistically from the shore when a feeding otter could be observed well enough to identify prey items through a 50–80 × telescope (prey were consumed while the otter was floating on its back on the water surface). A foraging otter selected for observation was watched for 30 min, or until it stopped feeding or moved out of view. These periods of observation were called sampling sessions.

The sex of foraging otters was ascertained from the presence/absence of a penile ridge or abdominal nipples. Three age classes were differentiated by pelage colour (Garshelis 1984) and behaviour: dependant pup, subadult (dark head) and adult. We timed each foraging dive, marked the location on a map, and after completion of the sampling session measured the water depth where the otter had been diving. We counted the number of prey obtained on each foraging dive and identified each to the lowest taxon possible, but for analysis we grouped them into broader categories: clams, mussels, crabs, worm-like organisms, urchins, sea stars, octopi and kelp. Three size classes of prey were distinguished, which were gauged relative to the width of an otter’s paw: small (< 6 cm; smaller than the width of a paw), medium (6–12 cm) and large (> 12 cm).

Analytical procedures

Analysis of variance ( anova) and regression (JMP, v. 2.0.4, SAS Institute, Cary, NC, and SPSS, v. 4.0, SPSS Inc., Chicago, IL) were used to test hypotheses related to changes in abundance (counts or densities), pup production, activity budgets and foraging success (proportion of dives within each sampling session in which food was obtained). Non-parametric analyses were used for tests of prey size and quantity, and for comparisons of our counts within shoreline segments against Irons, Nysewander & Trapp’s (1988) unreplicated count. Statistical comparisons with Irons, Nysewander & Trapp’s (1988) count remained the same whether or not otters seen outside the 200-m wide transect were included.

Several years of survey data (Green, 5 years; Applegate, 4 years; Montague, 3 years) collected by Johnson (1987) were pooled into a single pre-spill baseline and compared with our counts for individual post-spill years. We included only complete pre-spill counts made when the pre-spill investigator rated survey conditions (weather and sea state) as good or excellent.

anova models (with complete counts at each site as sample units) were constructed first in a preliminary model with main effects (e.g. site, month and year) and meaningful two–way interaction terms (e.g. site-by-year), and then in a final model with only significant (P < 0·2) terms. Contrasts were used to make pairwise comparisons within a significant main effect (e.g. between 2 years at all sites) or within a two–way interaction term (e.g. between two years at a single site). One-way anovas for individual sites were used to test differences among years other than pre-spill and 1990 because all sites were not sampled in the same months or years. anova models of activity budgets and foraging success were executed in the same stepwise manner. Otter densities were log-transformed to approximate normal distributions. Pup ratios, activity budgets and foraging success were transformed using the arcsine of the square root. The null hypothesis for some tests was that otter numbers after the spill would be less than or equal to those before (rather than a null hypothesis of no difference); for these cases one-tailed statistical tests (specifically noted as such in the text) were used. Power of tests was assessed using 95% confidence intervals (Steidl, Hayes & Schauber 1997). Confidence intervals were calculated by dividing root mean square error by the square root of the sample size of each site-by-year combination; means and confidence intervals were then back-transformed to the original scale for portrayal in figures.

Results

Abundance

Counts (converted to densities) of independent sea otters (i.e. otters other than dependant pups) conducted over a 7-year period after the spill were consistently equal to or higher than counts made before the spill (1977–85) for three of four oiled sites and one un-oiled site in western PWS (Fig. 2). Sources of variation in counts were evaluated at Green Island, Applegate Rock and Montague Island, the three sites that were surveyed during several different months (April–August) both pre-spill and post-spill [anova model with site, year (pre-spill, 1990, 1991 and 1996) and month as main effects, and a site-by-year interaction term]. Counts varied among years differently at each site (site-by-year interaction, F = 9·55, d.f. = 6, 65, P≤ 0·001) but counts did not vary among months (F = 1·35, d.f. = 4, 65, P = 0·26).

Figure 2.

Mean density and 95% confidence intervals of independent (non-pup) sea otters (number counted per surveyed area) in Prince William Sound, Alaska, USA, before (Johnson 1987 and unpublished data; Irons, Nysewander & Trapp 1988) and after (this study) the Exxon Valdez oil spill. Data were log-transformed for analysis, and then back-transformed for presentation here. Values above bars indicate the number of surveys. Areas are ordered, from left to right, by increasing extent of shoreline oiling in 1989 (surveyed portion of Montague was un-oiled; Applegate Rock was heavily oiled).

To assess potential effects of oiling, pre-spill vs. post-spill comparisons were made at each site individually. At un-oiled Montague Island, counts of independent otters in 1990, 1991 and 1996 were each significantly higher than pre-spill (one-way anova, F = 11·37, d.f. = 3, 18, P < 0·001; contrasts: pre-spill vs. 1990, F = 33·72, d.f. = 1, 18, P < 0·001; 1991, F = 15·40, d.f. = 1, 18, P < 0·001; 1996, F = 5·13, d.f. = 1, 18, P = 0·04; Fig. 2). Similarly, both counts made at lightly oiled Naked Island during 1993 (63 and 64 independent otters) were > 50% higher than the single pre-spill count (40 independents), but the difference was not significant (one-tailed Wilcoxon signed-rank test comparing counts at each of seven shoreline segments, Z = 0·94, d.f. = 6, P = 0·17). In 1996 the count at Naked Island and two neighbouring islands (Storey and Peak) was 30% higher than pre-spill but, again, not significantly so (Z = 0·47, d.f. = 8, P = 0·32).

At Green Island, which was heavily oiled, no significant differences or apparent trends were observed among pre-spill counts and 4 years of post-spill counts (one-way anova, F = 0·78, d.f. = 4, 25, P = 0·54; Fig. 2). Our 1996 count made from an inflatable raft (145 independents) was similar to a count made from the larger survey boat the next day (130 independents), indicating that our counts were not biased by using a larger boat than was used in pre-spill surveys of this site.

At heavily oiled Knight Island, all three post-spill counts in 1991 (266, 301 and 305 independent otters) exceeded the single pre-spill count from 1984 (199 independents; one-tailed Wilcoxon signed-rank test among 16 shoreline segments, Z = 1·96, d.f. = 15, P = 0·02). One incomplete survey of Knight Island was conducted in 1993; although the total count for the surveyed area was higher (225 independents) than the pre-spill count in the same area (179 independents), the difference was insignificant when compared by shoreline segment (one-tailed Wilcoxon signed-rank test, 8 of 14 shoreline segments higher in 1993, Z = 1·07, d.f. = 13, P = 0·14). In 1994, our count for 10 shoreline segments on Knight Island (190 independents) was about the same as the pre-spill count in these segments (182 independents), but we probably undercounted in 1994 because some segments were only partially surveyed. In 1996, a survey of all 16 Knight Island segments was completed and more otters were counted than were counted pre-spill (Z = 2·58, d.f. = 15, P = 0·005). The 1996 count included two loose groups of otters without pups in areas where, in past surveys, pups normally occurred and otter numbers were considerably lower. The boundaries of these groups were vague, but it appeared that one group comprised about 30 and the other about 40 otters. It is possible that these 70 otters were transient or immigrant males, as they seemed to have suddenly appeared (some of which were captured and tagged) the previous year (J. Bodkin, personal communication). If these animals came from another part of the sound, they should be excluded from inter-year comparisons of local abundance. Without these animals, the 1996 Knight Island count (322) was still the highest of all counts made at this site, and significantly greater than pre-spill (Z = 2·07, d.f. = 15, P = 0·02).

Pre-spill vs. post-spill comparisons at Applegate Rock, the most-heavily oiled site, were obfuscated by two anomalously high counts in May 1977. These two counts, made on consecutive days, were the first counts at this site and were more than twice as high (138 and 133 independent otters) as any subsequent pre-spill count (44–64; Johnson 1987). It appeared that a large group of otters occupied this area for a short time and then left, as 1 week later the count declined by > 50%. Because we used the means of pre-spill counts to represent the expected number of otters present just prior to the spill, these abnormally high early counts were excluded. Counts made 1 year after the spill were less than the pre-spill mean (one-way anova, F = 4·74, d.f. = 4, 24, P = 0·006; contrast, F = 10·82, d.f. = 1, 24, P = 0·003). Subsequent counts in 1991, 1994 and 1996 were not significantly different from pre-spill (individual contrasts, F < 1·84, d.f. = 1, 24, P > 0·19), although the 1994 tally (74 independents) was higher than all previous counts, except the two abnormal counts from 1977.

A total of 201 counts has been made in Gibbon Anchorage, Green Island, during 14 different years over a 20-year span, making this data set the most comprehensive of any site in PWS. Johnson’s (1987) counts were made throughout the year, but our counts of this area were conducted only during May–August, so pre-spill vs. post-spill comparisons were limited to these 4 months. Numbers of otters varied among years ( anova, F = 8·94, d.f. = 11, 94, P < 0·001) and among months (F = 8·98, d.f. = 3, 94, P < 0·001). We attempted to reduce the monthly variation by restricting the data set to July and August (when most of the post-spill counts were made), but both year and month effects remained significant ( anovas: year, F = 12·97, d.f. = 4, 23, P < 0·001; month, F = 13·80, d.f. = 1, 23, P = 0·001). However, monthly variation did not differ by year during 1990–96 (year-by-month interaction, F = 1·53, d.f. = 4, 23, P = 0·23). A time-of-day effect, previously recognized by Garshelis & Garshelis (1984), was also apparent, involving the departure of several otters from Gibbon Anchorage during the morning feeding period. Surveys of this anchorage done by Johnson (1987) and in this study tended to occur in early morning, before otters left, but some variation in the data probably was due to differences in the exact time of day that the counts were made. Nevertheless, the general pattern that emerged was a decline in the number of independent otters from 1977 to 1984, followed by an increase during post-spill years (Fig. 3).

Figure 3.

Numbers of independent (non-pup) sea otters counted in Gibbon Anchorage, Green Island (Prince William Sound, Alaska, USA) during May–August 1978–84 (prior to the Exxon Valdez oil spill; A.M. Johnson, unpublished data) and 1990–96 (after the spill). The drawn curve approximately connects yearly means. The portion of curve during the span of years with no counts (1985–89) shows the hypothesized rise in population during the late 1980s and decline following the spill in 1989.

None of our five larger survey sites were counted regularly enough to ascertain a population trend. However, for all survey areas combined, which comprises a large portion of the western sound (Fig. 1), our 1996 total count was 42% higher than in 1984–85 (Table 1). This increase over the 12-year period cannot be converted to an average yearly rate of increase because it includes the loss (of unknown magnitude) in the year of the spill. After the spill, however, independent otters increased by 13% over the 5-year period from 1991 to 1996, an average annual increase of 2·5%.

Table 1. Total counts (means for n > 1 survey) of sea otters before and after the Exxon Valdez oil spill at five sites in western Prince William Sound, Alaska, USA. Pre-spill counts (1984–85) were made by A.M. Johnson & A.R. DeGange (unpublished data) at Green Island, Applegate Rock and Montague Island (Stockdale Harbor and Port Chalmers), and by Irons, Nysewander & Trapp (1988) at the other two sites. Post-spill counts, made during this study, are shown only for the 2 years that we completely surveyed at least four of the five sites
 1984–8519911996
SiteIndependentsPups n IndependentsPups n IndependentsPups n
Green Island178 292184 706180 661
Applegate Rock 44  41 44 258 45 181
Montague Island155 631205 967201 821
Knight Island group199 59129113233921221
Naked Island group 75 201  –  –0107 451
Total         
 Excluding Naked Island576155 724323 818288 
 Including Naked Island651175   –  – 925333 

Pup production

At both oiled and un-oiled sites, post-spill pup ratios were often higher than pre-spill (Fig. 4). Comparing pup ratios from pre-spill to 1990 (the only post-spill year with May–August data) at Montague Island, Green Island and Applegate Rock, significant differences were attributable to month ( anova, May–June vs. July–August, F = 19·63, d.f. = 1, 46, P < 0·001) and site (F = 4·13, d.f. = 2, 46, P = 0·02), and yearly variation differed by site (year-by-site interaction, F = 6·73, d.f. = 2, 46, P = 0·003). Pup ratios did not differ significantly between pre-spill and 1990 at un-oiled Montague (contrast, F = 1·93, d.f. = 1, 46, P = 0·17) or at heavily oiled Green Island (contrast, F = 0·22, d.f. = 1, 46, P = 0·64), but the 1990 pup ratio was significantly higher than pre-spill at the most heavily oiled site, Applegate Rock (contrast, F = 14·77, d.f. = 1, 46, P < 0·001). Pup ratios appeared to increase in 1991 at all three of these sites (Fig. 4), although differences between pre-spill and 1991 (for July–August, the only months in common between these years) were not statistically detectable due to low power of the tests (one-way anovas: Applegate, F = 0·90, d.f. = 4, 15, P = 0·49; Montague, F = 6·75, d.f. = 3, 11, P = 0·008, contrast, F = 1·26, d.f. = 1, 11, P = 0·28; Green, F = 8·94, d.f. = 5, 14, P < 0·001, contrast, F = 1·54, d.f. = 1, 14, P = 0·23). By 1994 at Applegate (contrast, F = 0·10, d.f. = 1, 15, P = 0·76) and 1996 at Montague (contrast, F < 0·01, d.f. = 1, 11, P = 0·98), pup ratios were near pre-spill levels. However, at Green Island the increase in pupping continued: the 1993 pup ratio was the highest ever observed at this site (0·51 pups per independent; pre-spill vs. 1993 contrast, F = 13·99, d.f. = 1, 14, P = 0·002). Pupping remained high at Green Island in 1994 (contrast, F = 4·73, d.f. = 1, 14, P = 0·05) but in 1996 returned to near the pre-spill mean (contrast, F = 0·36, d.f. = 1, 14, P = 0·56).

Figure 4.

Mean pup ratios and 95% confidence intervals in Prince William Sound, Alaska, USA, observed before (Johnson 1987 and unpublished data; Irons, Nysewander & Trapp 1988) and after (this study) the Exxon Valdez oil spill. Data were subjected to an arcsine square root transformation for analysis, and then back-transformed for presentation here. (a) Comparison of pup ratios between pre-spill and 1990, the only years in which May–August were sampled. (b) Comparison of pup ratios among years in July and August, the only months sampled after 1990. Values above bars indicate the number of surveys. Within each panel, areas are ordered, from left to right, by increasing extent of shoreline oiling in 1989.

Pup ratios at heavily oiled Knight Island followed the same yearly trend as at Green Island (Fig. 4). Pup ratios were higher on all three surveys in 1991 (0·40, 0·46 and 0·50), a partial survey in 1993 (0·59) and a partial survey in 1994 (0·47), than on a single complete survey in 1984 (0·30) (Irons, Nysewander & Trapp 1988). The 1996 pup ratio was about the same as pre-spill, although if the two presumed male groups, totalling 70 otters, were subtracted from the count (as discussed above), the 1996 ratio was higher (Fig. 4). A pairwise test of the pre-spill pup ratios for individual shoreline segments against the means of the 1991 pup ratios was insignificant (one-tailed Wilcoxon signed-rank test, Z = 0·44, d.f. = 11, P = 0·33); however, the power to detect a difference was low because four of 16 segments that had otters with pups in 1991 were excluded from the analysis because no otters were present there in 1984 (i.e. these segments had incalculable pre-spill pup ratios). To circumvent this problem, we compared the number of pups observed in each segment, which, compared with pre-spill pup numbers, was higher in both 1991 (one-tailed Wilcoxon signed-rank test, Z = 2·30, d.f. = 15, P = 0·01) and 1993 (Z = 1·83, d.f. = 13, P = 0·03) but not in 1994 (Z = 1·12, d.f. = 9, P = 0·13) or 1996 (Z = 1·33, d.f. = 15, P = 0·09).

Pup production at lightly oiled Naked Island was also high in 1993 but, unlike the other areas we surveyed, was highest in 1996 (Fig. 4). As at Knight Island, some shoreline segments contained no otters, resulting in incalculable pup ratios, so we used pup counts in each segment for comparisons among years. About twice as many pups were counted on our two surveys of Naked Island in 1993 (27 and 28 pups, one-tailed Wilcoxon signed-rank test, Z = 1·69, d.f. = 6, P = 0·05) and our single survey in 1996 (30 pups) as were counted there during the pre-spill survey (15 pups). The neighbouring islands of Storey and Peak had three times as many pups in 1996 (15) than in the pre-spill survey (5), but because of tied counts between pre-spill and 1996 in three of the nine shoreline segments of the Naked Island group the difference was not significant (Z = 1·15, d.f. = 8, P = 0·12).

Activity patterns and time budgets

Sea otters spent less time feeding ( anova, F = 18·13, d.f. = 2, 12, P < 0·001, contrasts, F ≥ 18·59, d.f. = 1, 12, P≤ 0·001) and more time resting ( anova, F = 12·12, d.f. = 2, 12, P = 0·001, contrasts, F≥ 9·97, d.f. = 1, 12, P≤ 0·008) at Green Island in both 1990 and 1991 than they did before the spill (Fig. 5). The hourly feeding and resting patterns, however, did not differ between years ( anovas, hour-by-year interactions, F ≤ 0·55, d.f. = 7, 6, P > 0·77) or months (hour-by-month interactions, F ≤ 1·41, d.f. = 7, 6, P > 0·34), and monthly variation (more feeding and less resting in August–October than April–June, anovas, F≥ 30·37, d.f. = 1, 6, P≤ 0·002) did not differ between years ( anovas, month-by-year interactions, F ≤ 0·43, d.f. = 2, 12, P > 0·54). The hourly variation in feeding and resting behaviour was significant ( anovas, F≥ 2·87, d.f. = 7, 21, P≤ 0·03) in models with only main factors (year, hour and month). Other activities (swimming, grooming and interacting, combined) increased from pre-spill to 1990 ( anova, F = 21·71, d.f. = 2, 12, P < 0·001, contrast, F = 32·99, d.f. = 1, 12, P < 0·001) but did not differ between pre-spill and 1991 (contrast, F = 0·006, d.f. = 1, 12, P = 0·94) or between hours ( anova, F = 1·64, d.f. = 7, 21, P = 0·18). Other activities increased from April–June to August–October ( anova, F = 8·65, d.f. = 1, 6, P = 0·03).

Figure 5.

Proportion of sea otters observed feeding, resting and involved in other activities (i.e. swimming, grooming or interacting) during scan samples conducted in and around Gibbon Anchorage, Green Island (Prince William Sound, Alaska, USA) before (Johnson 1987) and after (this study) the Exxon Valdez oil spill. Stars represent no data.

Foraging success and diet

Foraging success in and around Gibbon Anchorage was similar in 1980–81, 1990 and 1991 ( anova, F = 1·23, d.f. = 2, 112, P = 0·30). Foraging success differed by sex–age class (adult males, solitary adult females, adult females with pups and subadults; F = 3·35, d.f. = 3, 112, P = 0·02) and by duration of the sampling session (10–19 min, 20–29 min and 30 min; F = 3·10, d.f. = 2, 112, P = 0·05), but these factors did not differ by year (interactions with year: sex–age, F = 1·07, d.f. = 6, 112, P = 0·38; duration, F = 1·82, d.f. = 4, 112, P = 0·13). Adult otters were more successful than subadults, and adult males were more successful than solitary adult females but not more successful than adult females with pups (Table 2). Adult females with pups were more successful than those without. Otters that were observed longer had higher success rates. Although we attempted to observe for 30 min, sampling sessions were sometimes cut short if an otter stopped feeding or moved to another feeding site out of view, and these situations tended to occur more frequently when an otter’s foraging success was low.

Table 2. Foraging success of sea otters before and after the Exxon Valdez oil spill at Green Island, and at oiled and un-oiled parts of Knight Island, Prince William Sound, Alaska, USA. Pre-spill data are from D. L. Garshelis (unpublished data), 1980–81. Means were calculated among sampling sessions (n); success within a sampling session was calculated as the percentage of dives in which an otter retrieved a food item
 Green IslandKnight Island 1991
 Pre-spill19901991 Un-oiledOiled
 Mean (%) n Mean (%) n Mean (%) n Mean (%) n Mean (%) n
Adult males9514971887 7934 766
Females with pups96 591188711818 923
Solitary adult females8816891685 67121001
Subadults82 783 58711 –0  –0

Clams were the primary prey of sea otters in the Gibbon Anchorage area. Clams were retrieved on 58% (calculated from Johnson 1987) to 77% (D. L. Garshelis, unpublished data) of successful foraging dives observed before the spill, 62% of 691 successful dives observed in 1990, and 66% of 502 dives in 1991. When otters obtained clams, they surfaced with a greater number per dive in 1990 (χ2 = 45·05, d.f. = 3, P < 0·001) and 1991 (χ2 = 42·85, d.f. = 3, P < 0·001) than during pre-spill years (Fig. 6). Also, the clams obtained were significantly larger in 1990 (χ2 = 14·45, d.f. = 2, P < 0·001) and 1991 (χ2 = 13·14, d.f. = 2, P = 0·001) than before the spill. The depth that otters obtained these clams, and the amount of time that they were submerged foraging, differed by year (one-way anovas: depth, F = 24·18, d.f. = 2, 225, P < 0·001; time, F = 20·22, d.f. = 2, 215, P < 0·001); in 1990 depths (mean = 7·9 m, SE = 0·53, n = 62) and dive times (mean = 61 s, SE = 3·5, n = 62) for otters foraging on clams did not differ from pre-spill (mean = 6·9 m, SE = 0·30, n = 95; mean = 67 s, SE = 2·2, n = 93, Dunnett’s T3 multiple comparisons, P≥ 0·27 for both), but in 1991 dives were shallower (mean = 4·1 m, SE = 0·37, n = 71) and shorter (mean = 42 s, SE = 2·9, n = 63, Dunnett’s T3 multiple comparisons, P < 0·001 for both).

Figure 6.

Frequency distributions of the number and size of clams obtained by foraging sea otters (percentages based on dives where they obtained ≥ 1 clam) in and around Gibbon Anchorage, Green Island (Prince William Sound, Alaska, USA) during 1980–81 (D.L. Garshelis, unpublished data) vs. 1990 and 1991 (this study).

Crabs were the second most frequent item in the diet of sea otters during 1980–81 (14%; D. L. Garshelis, unpublished data) and 1990 (12%), but were third (8%) behind mussels Mytilus trossulus Gould in 1976–84 (Johnson 1987) and in 1991 (9%). Otters obtained fewer (χ2 = 8·59, d.f. = 2, P = 0·01) but larger (χ2 = 49·40, d.f. = 2, P < 0·001) crabs per dive in 1990 than during pre-spill. In 1991 they continued to catch larger crabs than pre-spill (χ2 = 20·58, d.f. = 2, P < 0·001) but the number per dive was not different from pre-spill (χ2 = 1·96, d.f. = 2, P = 0·38). We were unable to evaluate differences in size and number of mussels obtained by otters before and after the spill because the mussels were all small and thus difficult to count.

At Knight Island, foraging success varied with shoreline oiling ( anova, oiled vs. un-oiled, F = 7·86, d.f. = 1, 16, P = 0·01) and duration of sampling sessions (F = 4·45, d.f. = 2, 16, P = 0·03) but not sex–age class (F = 1·50, d.f. = 2, 16, P = 0·25; subadults had to be excluded from this analysis due to an insufficient sample). However, there was a significant interaction between sex–age-specific success rates and shoreline oiling (F = 5·62, d.f. = 2, 16, P = 0·01). Foraging success was higher at oiled than at un-oiled sites for solitary adult females (contrast, F = 7·51, d.f. = 1, 16, P = 0·01) and adult females with pups (contrast, F = 5·36, d.f. = 1, 16, P = 0·03), but was not different for adult males (contrast, F = 0·37, d.f. = 1, 16, P = 0·55; Table 2).

Clams were second (28% of successful dives) to mussels (37%) in the sea otter diet at Knight Island. For dives where clams were caught, neither size (χ2 = 2·28, d.f. = 2, P = 0·32) nor number (χ2 = 0·38, d.f. = 2, P = 0·83) differed between oiled and un-oiled sites (Fig. 7). Crabs were not obtained frequently enough to test for differences between oiled and un-oiled sites.

Figure 7.

Frequency distributions of the number and size of clams obtained by foraging sea otters (percentages based on dives where they obtained ≥ 1 clam) near oiled (in 1989) vs. un-oiled shorelines at Knight Island (Prince William Sound, Alaska, USA) 2 years after the Exxon Valdez oil spill.

Examination of conflicting count data

Our high counts in the western sound appeared to be in conflict with those of Burn (1994), who also conducted boat-based counts of sea otters after the spill. Burn (1994) used Irons, Nysewander & Trapp’s (1988) survey method, paralleling the shoreline and counting within the 100-m wide strip on each side of the boat, but he sampled < 30% of the nearshore area. Burn (1994) also sampled some offshore zones, which were used to estimate total abundance and thereby track population change after the spill; however, as Irons, Nysewander & Trapp (1988) had not sampled offshore areas, pre-spill vs. post-spill comparisons were limited to nearshore transects.

Burn (1994) reported a 35% decline in otter density between the 1984–85 pre-spill count and 1989, the year of the spill, within the region defined as the oil-affected part of PWS; conversely, he observed a 14% increase in otter density within the un-oiled northern and eastern parts of the sound. However, from 1989 to 1990, density declined substantially in both oiled and un-oiled zones. The next year this downward trend reversed in the oiled zone, but continued in the un-oiled zone, resulting in an overall decline of 33% in un-oiled areas between pre-spill and 1991. Burn (1994) proffered that this decline in nearshore density in the un-oiled zone was due to a shift in otter distribution from nearshore to offshore, whereas he presumed that the decline in the oiled area was attributable to the spill. These conclusions have been widely cited (Ballachey, Bodkin & DeGange 1994; Monson & Ballachey 1994; Loughlin, Ballachey & Wright 1996) and even used as the basis for a spill-related mortality estimate (Garrott, Eberhardt & Burn 1993).

The apparent discrepancy between Burn’s (1994) results and ours might suggest a difference in counting technique or accuracy. However, separate counts made by us and by Burn during the same time period (July–August 1991) on the same sample of shoreline segments at Knight Island were nearly identical (mean of our three counts = 136, Burn’s single count = 133). Thus, inter-observer variability appears not to be the reason for our differing results.

We compared Burn’s (1994) counts (1989–91 and 1993, provided as an electronic database) with those of Irons, Nysewander & Trapp (1988) for matching shoreline segments in oiled areas of western PWS and did not observe the post-spill decline that he had reported (Table 3). A decline became evident only when counts from the northern mainland, outside the area actually oiled, were considered as part of the oiled zone. Burn (1994) included a 5-km wide ‘buffer strip’ along the oiled zone in case the effects of the spill extended beyond the actual margin of the oil. Some portions of this buffer strip along the northern mainland showed sizeable declines in otter numbers between pre-spill and post-spill counts.

Table 3. Summer nearshore counts of sea otters in oiled regions of western Prince William Sound, Alaska, USA, before and after the Exxon Valdez spill. The single pre-spill count (1984) was made by Irons, Nysewander & Trapp (1988). Post-spill counts were made by D. M. Burn (unpublished data; 1989 and 1990 data are the mean of three counts, 1991 and 1993 are single counts)
Region*19841989199019911993
  1. All counts were conducted using the same methodology, but whereas the entire shoreline was counted pre-spill, < 30% was sampled post-spill; thus, for comparative purposes, pre-spill counts from the same portions of shoreline are shown.

  2. *Regional boundaries defined by Irons, Nysewander & Trapp (1988).

Green Island  8 25 23 27 36
Naked Island 21 35 16 17 12
Knight Island 70 69 55110101
Port Bainbridge126150 89 62126
Port Nellie Juan 32 42 27 24 39
Passage Canal  2  2  1  2  5
Total259323211242319

The geographical pattern of decline in the region near the buffer strip, however, suggests that it was not oil-related. The most conspicuous decline occurred in a small segment of coast along Axel Lind Island (Fig. 1), where 97 otters were counted in 1984 but, on average, only two were seen during 1989–90; the low numbers in this transect alone accounted for nearly all of the 35% decline reported by Burn (1994) (without this transect the decline in the oiled area plus buffer strip = 4%). Other transects in the same general area along the northern mainland, but outside the buffer strip, also showed significant declines. For example, a transect in Unakwik Inlet (Fig. 1) declined from 36 otters in 1984 to an average of two in 1989–90, a group of transects in Harriman Fiord declined from 107 to an average of seven, and counts in Barry Arm dropped from 135 to 23. These transects were all well beyond (> 20 km) the outer edge of the oil, suggesting a regional non-oil related population decline or shift in distribution (to other areas, or from nearshore to further offshore).

Notably, the highest post-spill count in an offshore transect was also in this area along the northern mainland. Omitting just this single high count from the offshore data set eliminated the apparent rise in offshore numbers in the un-oiled zone. That is, just as a dramatic decline in the single nearshore transect at Axel Lind Island was largely responsible for the perception of an overall decline in the oiled zone, a high count in a single offshore transect was the basis for the impression that the decline in the un-oiled zone was due to a shift offshore. Because the offshore sampling was sparse (encompassing only about 2% of the offshore area) and counts were low (82% of transects = 0), it was impossible to determine whether this one particularly high offshore count was an anomaly or indicative of an actual offshore shift.

When the data were partitioned into regions in which environmental conditions affecting population changes were fairly uniform (Fig. 1), Burn’s (1994) results appeared quite different from indicated in his report. Nearshore abundance in the eastern and western portions of the sound increased 40% and 12%, respectively, from 1984 to 1989, whereas the northern mainland, none of which was oiled, declined by 42%. Whether this decline in the northern sound was attributable to movement further offshore, movement out of sampled transects, or a real decline in total numbers, possibly due to emigration into the area that was subsequently oiled, cannot be ascertained. If otters did emigrate from this area to the oiled area, they would have to have done so before the spill, rather than in response to diminished density afterwards. The relatively low nearshore counts in the northern sound and high counts in the oiled area were evident during Burn’s (1994) first survey in June 1989, just 2 months after the spill, when the spill zone was congested with boats involved in clean-up and rescue efforts (Carpenter, Dragnich & Smith 1991). It seems improbable that within this time frame and under these conditions, large numbers of otters could have or would have moved > 20 km to take up residency in the spill zone.

Discussion

We believe there is strong evidence, both from our study and that of Burn (1994), to conclude that sea otter abundance in the oil-affected part of PWS was as high or higher during the 7-year period after the spill as it was during the early mid-1980s. This result is surprising.

Although Garshelis (1997) and Garshelis & Estes (1997) found that other investigators (Garrott, Eberhardt & Burn 1993; DeGange, Doroff & Monson 1994) probably overestimated sea otter mortality due to the spill, the loss was nonetheless substantial (probably 600–1000 otters). It would seem that a decline of this magnitude, constituting 15–25% of the resident population, should have been detectable from boat surveys. Previously we considered various hypotheses to explain the failure to detect this loss (Johnson & Garshelis 1995), the most probable being that spill-related mortality was masked by a population increase in the late 1980s, when no counts were made. At the time of our earlier report, the only support for this proposed late-1980s population growth was an increase in otter numbers at Montague Island, an un-oiled site in the western sound, and high pup production during 1991, which if characteristic of the period immediately preceding the spill could have spurred such an increase.

Subsequent data, presented here, indicated that high pup production (in both oiled and un-oiled areas) continued during 1993 and 1994, but returned to near pre-spill levels at four of our five survey sites in 1996 (Fig. 4). Significant year-to-year variation in sea otter pup production, as observed in the past (Johnson 1987; Johnson & Garshelis 1995), may be related to year-specific weather conditions. However, the number of exceedingly good years of pup production post-spill strongly suggests the presence of a more pervasive environmental factor that elevated the pupping baseline above what it was in the late 1970s and early 1980s. The relatively young age of first reproduction discovered among the reproductive tracts of otters killed during the spill (Bodkin, Mulcahy & Lensink 1993), compared with ages of primiparity observed earlier (Garshelis, Johnson & Garshelis 1984; Jameson & Johnson 1993), is evidence that pupping rates were increasing by the late 1980s.

We postulate that this heightened pup production was triggered by an enhanced food supply (Fig. 6). Although the diet in the early 1990s remained unchanged from that of the early 1980s, with clams being the dominant component (60–70%; Doroff & Bodkin 1994; Johnson & Garshelis 1995), our results indicate that otters obtained more and larger clams per dive during the early 1990s. Moreover, in 1991 these clams were obtained on shorter dives in shallower water than in the early 1980s, enabling otters potentially to make more foraging dives and hence obtain more food per unit time. Foraging success did not change between the early 1980s and the early 1990s, but foraging success tends to be related more to dietary composition than to food abundance (Estes, Jameson & Johnson 1981; Garshelis 1983; Doroff & DeGange 1994).

Our time budget data provide further evidence of more plentiful food during the 1990s. We observed less feeding during our daytime scan samples than had been observed using the same technique at the same observation sites during the late 1970s and early 1980s. The premise that time budgets of otters reflect food availability is well supported by other studies (Estes, Jameson & Rhode 1982; Garshelis, Garshelis & Kimker 1986; Gelatt 1996). Collectively, the time budget and foraging data are a compelling indication of an increased food supply in the area around Gibbon Anchorage, Green Island. Moreover, our observation that post-spill pup production was high throughout western PWS (Fig. 4) suggests that this increased food supply was not a local phenomenon.

The scenario that we propose is that, in response to an increasing food supply before the spill, pup production increased, resulting in a population increase. The rate of population increase probably varied with year-to-year changes in pup production, as witnessed during our study, as well as yearly variation in juvenile survival, as observed in two post-spill telemetry studies (Rotterman & Monnett 1991; B. Ballachey & A. Doroff, personal communication). We suspect that the population surged during years with good pup production and survival, but remained stationary or declined following poorer years. Thus, an overall increasing trend in otter numbers might not be obvious from a short series of counts. Additionally, in our study, incomplete data (missing years and partial surveys) probably obscured trends in otter abundance at each particular site (Fig. 2). However, the combined count for all surveyed sites (Table 1) and the trend data from Gibbon Anchorage, the only individual site with a large number of surveys (Fig. 3), provide persuasive evidence of an overall increasing population.

Research from the late 1970s and early 1980s led Johnson (1987) to conclude that otters around Green Island were at carrying capacity. We concur with this interpretation, and suggest that otters remained at carrying capacity through the late 1980s. This is not a contradiction to our belief that otter numbers were increasing. Populations at carrying capacity are not necessarily stationary; they can remain food-limited but track changes in food abundance. We suggest that food supplies were constant or even declining during the early 1980s, but then increased during the mid-1980s; consequently, otter numbers followed this same trajectory.

The cause of this increased size and abundance of otter prey is unknown, but we theorize that it may represent long-term recovery from the massive earthquake of 1964, which was centred in the northern part of the sound (near Unakwik Inlet; Fig. 1). This earthquake was the second largest ever measured (moment magnitude 9·2); the last seismic event of this magnitude in this area occurred at least 800 years ago (Plafker 1990). The 1964 earthquake uplifted the western sound, including Knight Island (1–2 m), Green Island (Gibbon Anchorage = 2·5 m) and Montague Island (Port Chalmers = 3 m), whereas the northern part of the sound, including Harriman Fiord, Barry Arm and Unakwik Inlet, subsided (0·6–2 m) (Baxter 1971; Hanna 1971). Significant immediate clam mortality occurred, averaging 30–40% in areas that rose 1·4 m (the average uplift for the sound) and > 80% in places like Gibbon Anchorage that rose more (Baxter 1971).

Prior to the earthquake, butter clams Saxidomus giganteus DeShayes were probably the most common prey of sea otters at Green Island. Butter clam shells, imbedded in uplifted sediments in the posture in which they died, are still evident in many places in the sound (D. L. Garshelis & C. B. Johnson, personal observations). Beds of these stranded shells indicate that at the time of the earthquake many dense patches of medium–large butter clams occurred at Green Island (Estes & VanBlaricom 1985). During the 1970s (Calkins 1978), 1980s (Johnson 1987; D. L. Garshelis, unpublished data) and early 1990s (Doroff & Bodkin 1994; this study), this species remained common in the diet of otters in the western sound; however, even though sea otters tend to select the largest available clams (Kvitek et al. 1992), the ones they obtained in Gibbon Anchorage in the 1980s and 1990s (Fig. 6) were smaller than the ‘earthquake fossil’ clams present there (haphazard sampling; mean = 8·0 cm, range = 6·6–9·0 cm, n = 35). Butter clams are the major food of sea otters in soft-bottom habitats in parts of Alaska that were further from the earthquake (Kvitek & Oliver 1992; Doroff & DeGange 1994); where otters have not yet had a major impact on this prey, shell sizes are equivalent (mean ≈ 8 cm) (Kvitek & Oliver 1992; Kvitek et al. 1992) to those that apparently were abundant at Green Island at the time of the earthquake.

Shells from the littleneck Protothaca staminea Conrad, another common (but smaller) clam consumed by sea otters, also occur in dense earthquake-stranded patches at Green Island (Estes & VanBlaricom 1985). Surveys made at a site in the eastern sound 10 years after the earthquake indicated that littleneck densities had recovered to only about 40% of what they had been before (Paul, Paul & Feder 1976).

Besides the enormous outright mortality of clams caused by the earthquake, habitat alteration negatively affected their recovery. The lower vertical distribution for species like S. giganteus and P. staminea is limited by a layer of soft decaying organic silty mud (possibly oxygen deficient), which reduces survival of larvae. When the sound was uplifted by the earthquake, the vertical zone that had been preferred clam habitat was exposed above water level, and the new preferred zone was covered by the silty mud, which impeded clam recruitment (Baxter 1971). Paul, Paul & Feder (1976) indicated that clam recruitment remained poor through 1971 in an area of the eastern sound that was uplifted by about 1 m. We know of no data on persistent effects of the earthquake in western PWS, but we presume that clams there fared even worse, because the uplift was greater than in other parts of the sound.

The initial loss of clams following the earthquake would have severely depressed the carrying capacity of the area for sea otters. At the time, the otter population in the sound was growing and expanding, as it recovered from over-exploitation during the fur trade. Counts at Green Island rose from 42 in 1959 to 116 in 1964 (7 months after the earthquake) and continued to rise through the early 1970s (Pitcher 1975). We suspect that otters were not food-limited at this time. Had it not been for the earthquake, otter numbers probably would have climbed even more quickly and ultimately would have attained higher levels. At Green Island, otters apparently reached carrying capacity by the late 1970s (Estes, Jameson & Johnson 1981; Johnson 1987), and by the early 1980s counts in Gibbon Anchorage showed a decline (Fig. 3).

We believe that the availability of clams limited otter numbers at Green Island during the early 1980s. Medium–large sized clams (i.e. ≥ 6 cm by our definition) that survived the earthquake would have been reduced by otter predation and other mortality, and replacement of this size class would have been minimal, due to the gap in recruitment following the earthquake and the slow growth of clams (Paul & Feder 1973, 1976; Feder & Paul 1974; Paul, Paul & Feder 1976). Clams recruited during the mid-1970s, when habitat conditions in the western sound may have normalized, would still have been small in the early 1980s, and moreover, densities of clams big enough to constitute sea otter prey (> 3 cm) would have been low due to the long period of poor recruitment. These conditions would have improved by the mid–late 1980s as clams born in the early 1980s became sufficiently large to be preyed upon by otters, and some clams born in the 1970s reached the larger size class (Fig. 8). Because the edible biomass of clams increases by roughly the cube of their shell length (Feder & Paul 1973, 1974), increased availability of larger clams would have provided substantially more nourishment for otters. Ten-year-old clams of the species most frequently consumed by otters at Green Island [S. giganteus, P. staminea and Mya truncata L. (a clam intermediate in size between S. giganteus and P. staminea)] contain > 10 times more meat than 5-year-old clams (calculated from data presented by Feder & Paul 1973, 1974; Paul & Feder 1976; Kvitek et al. 1992).

Figure 8.

Hypothesized trajectory of the sea otter population in western Prince William Sound, Alaska, USA, in response to changing biomass of prey-sized clams and the 1989 Exxon Valdez oil spill. Trend in otter numbers was surmised from data presented by Pitcher (1975), Johnson (1987) and this study. The portion of the curve between 1978 and 1996 approximately follows the count data from Gibbon Anchorage (Fig. 3). Clam biomass was significantly reduced (more than shown on graph) from the uplift caused by the 1964 earthquake in Prince William Sound (Baxter 1971) and recruitment was depressed for several years afterwards (Paul, Paul & Feder 1976). The slow growth rate of clams caused a lag between the normalization of clam recruitment in the mid-1970s and increased biomass of prey-sized clams for otters during the 1980s (data from this study).

We found that 19% of the clams consumed by otters at Green Island in the early 1990s were 6–12 cm in length, compared with only 11% in this size category during the early 1980s, and whereas otters in this area were never seen preying upon > 12-cm clams in the early 1980s, 1% of the clams retrieved in the early 1990s were this large (Fig. 6). In 1984, A. M. Johnson (personal communication) found, for the first time since beginning his sea otter study in the mid-1970s, a large accumulation of otter-cracked butter clam shells (mean = 6·2 cm) on an intertidal beach in Gibbon Anchorage. We observed several accumulations of medium-sized butter clams that had been cracked open and consumed by otters in Gibbon Anchorage during our study. No data on clam sizes from the early 1980s were available for other sites in the western sound, but our data from the early 1990s at Knight Island indicate that clams obtained by otters there were at least as large as at Green Island (Fig. 7). We hypothesize that the increased availability of larger clams (and also crabs) promoted renewed growth of otter numbers in the western sound during the mid-1980s (Fig. 8).

Improving food conditions and growing otter numbers in the western sound seemed not to have been mirrored in the northern sound. Clam population dynamics there differed radically because this area subsided rather than uplifted during the earthquake. Unlike the uplifted area, subsided habitats appeared to be immediately suitable for deposition and survival of larval clams, and the subsidence appeared not to increase mortality of living clams (Baxter 1971). At the time of the earthquake, otters had not yet reoccupied the northern sound (Pitcher 1975). After they did, in about 1973, numbers probably grew quickly, as they took advantage of unexploited food resources. This pattern was observed after otters reoccupied the last stretches of eastern PWS in the late 1970s and early 1980s (Garshelis & Garshelis 1984; Simon-Jackson 1986; Monnett & Rotterman 1989). However, by the late 1980s otters had apparently reduced their preferred prey within shallower habitats along the edges of the deep fjords of the northern sound and were subsisting mainly on mussels, a low-quality food (Anthony 1995). This may have prompted the decline or movement further offshore that was reported by Burn (1994) and attributed to the oil spill.

It appears that at least three catastrophic events, the 19th-century fur trade, the 1964 earthquake, and 25 years afterwards, the Exxon Valdez oil spill, profoundly shaped the population dynamics of sea otters in PWS. Also, in the early 1990s, killer whales Orcinus orca L. started attacking and killing sea otters in western PWS (Hatfield et al. 1998), adding yet another confounding variable to their demography (Estes 1999; Garshelis & Johnson 1999). Distinguishing population effects of the oil spill from this bewildering array of background noise was far more complex than sea otter investigators, including ourselves, initially recognized.

Acknowledgements

We greatly appreciate the assistance provided by B. Cooper, R. Day, T. DeLong, B. Lance, T. Mabee, M. MacDonald, S. Murphy, A. Mut, J. Schauer, A. Wildman and A. Zusi-Cobb. We thank B. Ballachey, D. Burn, D. Irons and A. Johnson for provision of unpublished data and discussion of ideas, K. Parker and J. Harner for reviewing our analytical procedures, and J. Bodkin and A. Johnson for providing comments on an earlier draft of the manuscript. This study was supported by Exxon USA, although the results and interpretations reflect the opinions of the authors, and not necessarily those of Exxon.

Received 5 October 1998; revision received 3 May 2000

Ancillary