FISHERIES OCEANOGRAPHY Fish. Oceanogr. 14 (Suppl. 1), , 2005

Transcription

1 FISHERIES OCEANOGRAPHY Fish. Oceanogr. 14 (Suppl. 1), , 2005 Do patterns of Steller sea lion (Eumetopias jubatus) diet, population trend and cetacean occurrence reflect oceanographic domains from the Alaska Peninsula to the central Aleutian Islands? ELIZABETH H. SINCLAIR,* SUE E. MOORE, NANCY A. FRIDAY, TONYA K. ZEPPELIN AND JANICE M. WAITE National Marine Mammal Laboratory, NOAA/NMFS Alaska Fisheries Science Centre, 7600 Sand Point Way, NE, Seattle, WA 98115, USA ABSTRACT Shipboard surveys were conducted along the Aleutian Islands in 2001 and 2002 to assess the influence of a suite of biophysical parameters on regional patterns in the distribution of cetaceans and Steller sea lions (SSL; Eumetopias jubatus). Distributions of four large whale species: fin (Balaenoptera physalus), humpback (Megaptera novaeangliae), minke (B. acutorostrata) and sperm (Physeter macrocephalus) aligned with proposed metapopulation breaks in diet and population trend of SSLs. Dall s porpoise (Phocoenoides dalli) and killer whales (Orcinus orca) were widely distributed throughout the study area, and killer whales were particularly prevalent along the north Aleutian Island coastlines between Unimak Pass and Samalga Pass. Biopsies determined that most killer whales (92%) were of the piscivorous (resident) ecotype as opposed to the mammal-eating (transient) ecotype observed in 2002 only. Generalized additive models (GAMs) were used to explore relationships between these multispecies patterns in distribution, oceanographic variables (salinity, temperature, fluorescence and depth) and proximity to six Aleutian passes. The GAMs indicated the best-fit models and most significant correlations as determined by the Akaike function and Cp-statistics were: depth and proximity to the nearest measured pass for SSLs and all cetaceans, respectively; frequencies of herring and salmon in SSL diet with *Correspondence. beth.sinclair@noaa.gov Received 12 December 2003 Revised version accepted 20 July 2005 population trend; fluorescence in the top 50 m with occurrence of humpback, minke, and killer whales; and surface temperature with occurrence of humpback, killer, and sperm whales. Results of the GAM analyses suggest foci for future investigation of relationships between physical variables and interspecific patterns of marine mammal distribution. Key words: biophysical, cetacean, generalized additive models, killer whales, metapopulation, orca, Steller sea lion INTRODUCTION The range of the endangered western stock of the Steller sea lion (SSL; Eumetopias jubatus) extends from the central Gulf of Alaska westward along the Aleutian Islands. A metapopulation composed of at least four (York et al., 1996; Sinclair and Zeppelin, 2002) and as many as six (Call and Loughlin, 2005) subpopulations is inferred by patterns in population trends and diet among adult females (York et al., 1996; Sinclair and Zeppelin, 2000, 2002; Call and Loughlin, 2005), patterns of movement among juveniles (Raum- Suryan et al., 2002; Fadely et al., 2005), and genetic analysis (B. Taylor, personal communication, Southwest Fisheries Centre, La Jolla, CA, USA). Mechanisms forcing these regional patterns have not been defined. The current investigation was prompted by the observation during the research cruises conducted in 2001 and 2002 that regional breaks in the distributions of four large cetacean species: fin (Balaenoptera physalus), humpback (Megaptera novaeangliae), minke (B. acutorostrata) and sperm (Physeter macrocephalus) seemed to correspond roughly with regional breaks in SSL diet and population trends. The central hypothesis investigated here is that biophysical properties of the passes along the Aleutian Islands broadly influence distribution patterns of marine mammals. Data collected during research cruises in 2001 and 2002 were analysed to explore the Ó 2005 Blackwell Publishing Ltd. 223

2 224 E.H. Sinclair et al. potential influences of physical characteristics of the eastern (Unimak, Akutan, and Umnak), transition (Samalga), and central Aleutian passes (Amukta, Seguam, and Tanaga; see Ladd et al., 2005) on metapopulation patterns in diet and decline of SSLs, and on the distribution of five common species of cetaceans observed throughout both study years: humpback whales, minke whales, sperm whales, killer whales (Orcinus orca) and Dall s porpoise (Phocoenoides dalli). Fin whales were commonly seen during both cruises, but were excluded in the final analysis because they fell outside the study area once it was reduced for consistency between survey years. The present and previous studies in this volume (Byrd et al., 2005; Call and Loughlin, 2005; Fadely et al., 2005) examine the influence of bottom up structuring on patterns of distribution among apex and secondary trophic level predators along the Aleutian Islands. Although we do not explicitly examine top down forcing (e.g. predation on SSL by killer whales), toward this goal, and largely because of recent attention on the role of these top predators in the population declines of Aleutian Island sea otters (Enhydra lutris) and pinnipeds (Lord, 2001; Springer et al., 2003), we document the distribution of killer whales by ecotype. Identified ecotypes within the range of the SSL western stock include transient (mammal-eating), resident (fish-eating) and a more recently defined offshore type, which also appears to focus its diet on fish (see Barrett-Lennard et al., 1996; Herman et al., 2004). A more thorough analysis of top down effects based on killer whale abundance estimates and predation rates incorporating data collected during the 2001 and 2002 cruises (for preliminary results see Wade et al., 2003) will be presented in subsequent publications. METHODS SSL diet and population trend Metapopulation boundaries for the western stock of SSLs based on diet and population data collected during and (Sinclair and Zeppelin, 2002), respectively, were reanalysed to include data available through 2001 to update these boundaries to reflect the most recent data available (Fig. 1). The data do not lend themselves to interannual comparison because of temporal patchiness in collection; therefore all data were combined for reanalysis. For modelling purposes, only diet and population data collected on rookeries within the study area of the 2001 and 2002 cruises were included. Diet diversity and diet and population cluster analyses were reconstructed following the approach of Sinclair and Zeppelin (2002). Diet analysis was based on 1262 scats collected during May through September from 14 of the total 16 rookery sites falling within the study area (Table 1). The relative importance of each prey species was determined by its frequency of occurrence in scat samples. Frequency of occurrence is defined as the number of scats containing a particular prey species expressed as a fraction of the total number of scats containing identifiable prey items. Like every sampling method used to determine the diet of marine mammals (Pierce and Boyle, 1991), scats have their own set of particular biases including the limited representation of large prey remains and overrepresentation of small prey parts (Sinclair et al., 2000). In the case of SSLs, we are able to estimate the proportion of animals consuming a particular prey item (frequency of occurrence, the presence or absence of prey species in a scat), but not the individual number of prey consumed, because few prey hard parts survive the digestive process (Zeppelin et al., 2004). Prior studies of SSL diet using scats are comparable with those based on stomach contents (see Sinclair and Zeppelin, 2002 for review) and provide the most recent and longest term data set with which to conduct the present study. At this time of year, scats collected on rookeries represent primarily the diet of adult females. Similarities in prey composition between rookeries were determined by principal component analysis (PCA) and an agglomerative hierarchical cluster analysis (Ludwig and Reynolds, 1988). Principal component analysis was calculated on a correlation matrix using prey species as variables and 14 rookery sites as observations, thereby reducing data to those species accounting for most of the variance. To minimize zeros in the analysis, only prey that occurred in 5% of the scats across all sites were included as variables in the PCA. Cluster analysis was then conducted on the PCA factors using a squared Euclidean distance (Ludwig and Reynolds, 1988) as a measure of similarity between sites, and Ward s (1963) method to compare cluster distances. Diet diversity was calculated using Shannon s index of diversity, H, where p i is the proportion of the ith species in the scat sample (Ludwig and Reynolds, 1988): H ¼ Xn p i ln p i ð1þ i¼1 The SSL population trend values were calculated by regressing the natural logarithm of the

3 Regional distribution patterns of marine mammals 225 (a) (b) Figure 1. Clusters of Steller sea lion rookeries with similar trends in (a) diet during and (b) population during Colours in panel (b) represent general patterns ranging from least (red) to greatest (green) overall declines during those years. non-pup counts over time following Sease and Gudmundson (2002) and York et al. (1996). As with the diet data, the longest time series available was used for analysis of the population data to decrease the effect of patchiness in data collection. Use of natural logarithms of non-pup counts and scats for diet analysis facilitated comparison with previously published results. Population data were available from all 16 rookeries falling within the study area occupied in 2001 and Counts were averaged in years where multiple counts were made. Cetacean distribution Visual surveys for cetaceans were conducted for h each day during both cruises from 4 to 23 June, 2001 and May 19 June 18, In 2001, the area surveyed was from Seward, Alaska, to Seguam Pass (including the Gulf of Alaska south of the Kenai Peninsula, Shelikof Strait, and along the Alaska Peninsula). In 2002, the cetacean survey focused solely on the area from Unimak Pass to Tanaga Pass (see Fig. 1, Hunt and Stabeno, 2005). Surveys were conducted from port and starboard stations on the bridge whenever the ship was in transit. Observers searched for cetaceans from the bow to 90 abeam with the naked eye and 7 50 handheld binoculars. Data were recorded by the starboard observer using WINCRUZ software on a laptop computer interfaced to the ship s global positioning system (GPS). When marine mammals were seen, bearing (in 2001 only), distance

4 226 E.H. Sinclair et al. Table 1. Steller sea lion (SSL) population trends (natural logarithm of counts over ), scat sample sizes, frequencies of prey occurring at 5%, diet diversity, cluster values, and location for rookeries falling within the study area along the Aleutian Islands in 2001 and Prey percentage frequency of occurrence values Diversity (H) Cluster AM CEPH SAL POLL PH PSL PC Scat sample size ( ) SSL trend ( ) Longitude ( W) Latitude ( N) Rookery site Ugamak ) Akun/Billingshead ) Akutan/Cape Morgan ) Bogoslof/Fire Island ) Ogchul ) Adugak ) Yunaska ) Seguam/Saddleridge ) Agligadak )0.18 Amlia/Sviech Harbour ) Kasatochi/North Point ) Adak/Lake Point ) Adak/Cape Yakak )0.05 Gramp Rock ) Tag ) Ulak/Hasgox Point ) These values were used in SSL models. Prey species are abbreviated as follows: AM, Atka mackerel; CEPH, cephalopods; SAL, salmon; POLL, walleye pollock; PH, Pacific herring; PSL, Pacific sand lance; PC, Pacific cod. (reticular), species, group size, course, and speed were recorded. When possible, during encounters with killer whales, the ship was diverted or a small zodiac deployed, to allow close approach for photographic and biopsy sampling. These efforts provided data needed for the subsequent identification of individual whales (see Ford et al., 1994) and their assignment to either mammal-eating (transient) or fish-eating (resident and offshore) ecotypes (see Barrett-Lennard et al., 1996). Survey effort and cetacean sightings along the visual survey trackline were plotted in ARCMAP 8.2 (ESRI, 2002). Cetacean sightings were plotted from Seward to Tanaga Pass to provide an interspecies comparison of distribution patterns. However, sightings data were restricted to the study area extending from Unimak to Tanaga Pass (Fig. 2) for modelling purposes involving biophysical observations. To facilitate modelling cetacean distribution, the study area was post-stratified by constructing a grid in ARCMAP that consisted of km wide cells (Fig. 2). These gridded cells were then split into north and south sections at the midpoint of the nearest island, resulting in 216 cells. The distance of each cellcentre to the centre of the nearest sampled pass was measured. The presence of on-effort trackline and each cetacean species was determined by cell. In total, there were 153 cells with survey effort. Presence or absence of sighted cetaceans was scored in each cell. There were sufficient sightings to model the occurrence of Dall s porpoise, killer, sperm, minke and humpback whales. Passes were not randomly chosen for oceanographic measurements during the two cruises but were selected because they were the geographic end points of proposed SSL subpopulation boundaries described by Sinclair and Zeppelin (2002). Oceanographic characteristics for each measured pass were determined by integrating the field sampling values over the top 50 m (see Ladd et al., 2005, for description of oceanographic collections). For each SSL rookery and each cell with cetaceans, the nearest measured pass was identified and the oceanographic characteristics of that pass were linked to the cell or rookery. This followed the assumption that, if the physical dynamics of the passes themselves were relevant to marine mammals, then the nearest pass would have the greatest influence. When a pass was sampled more than once, average temperature, salinity, fluorescence and maximum depth values were used. Average values were used in 2001 for Akutan Pass, which was sampled on June 14 and 15. Field research was initiated on June 4, 2001, and May 19, 2002, with the result that fluorescence

5 Regional distribution patterns of marine mammals W 170 W 165 W 58 N 54 N 56 N 52 N 54 N Figure 2. A grid overlay of the 2001 and 2002 study area consisting of km wide cells. The centre of each cell was the point from which distance to the centre of the nearest measured pass was made for on-effort cetacean sightings falling within that cell. Tracklines are those from which on-effort sightings were conducted. 50 N 48 N 175 W 170 W 165 W 52 N 50 N values in the passes were an order of magnitude higher at the beginning of the survey in 2002 than 2001, presumably because of overlap with the end of the spring phytoplankton bloom. To avoid potential effects of start date on model results, the first 8 days of the 2002 cruise were removed from the data set. May 27 was selected as the start date for 2002, because it was the first day when fluorescence values fell to a level comparable with the average values for both cruises. This resulted in the loss of the May transects of Akutan and Unimak Passes in However, both passes were sampled in June of that year, and those values were used in analyses. Modeling Generalized additive models (GAMs) were constructed for SSL diet diversity and rookery population trend (with and without prey frequency of occurrence values as explanatory variables), and for the five predominant cetacean species in relation to the oceanographic and geographic aspects of the passes. We chose GAMs over general linearized models (GLMs) because we wanted to assess additive combinations of a large number of variables on multiple marine mammal species. As a non-parametric extension of GLMs, GAMs share the same statistical properties, but the functional relationship between response variable and predictor variable is estimated rather than restricted to a particular form (Hastie and Tibshirani, 1999). The complete model is the sum of all predictor functions,, plus a constant, c: y ¼ c þ X n i¼1 fðxþ ð2þ Estimates of diet diversity ( ) and population trend ( ) were modelled as Gaussian response variables with identity link functions. For cetacean distribution, the presence/absence of each species was modelled as a generalized additive logistic model; a binomial response variable with a logit link function. Continuous explanatory variables were modelled as linear functions, smoothing splines with three equivalent degrees of freedom, or removed from the model. Three equivalent degrees of freedom for smoothing splines were selected to allow non-linear effects and to limit unrealistic tracking of the data. Categorical (non-numerical) explanatory variables were modelled as linear functions or removed from the model. Oceanographic and geographic variables were chosen as explanatory variables based on whether they were sampled during the cruises, likelihood of providing a characteristic signature of the measured pass, and potential influence on productivity and predator distribution. Because of correlations between explanatory variables, six sets of explanatory variables were fitted for each response variable. Each set involved four explanatory variables: (i) mean fluorescence in the nearest measured pass; (ii) maximum depth, mean temperature or mean salinity in the nearest measured pass; (iii) distance to the nearest measured pass, either as an absolute value or

6 228 E.H. Sinclair et al. as a combined distance direction variable with values to the west being negative; and (iv) the direction to the cell/rookery from Samalga Pass. Nitrate was also measured during field studies but was excluded in the GAMs because it was correlated with salinity and temperature and unavailable for Umnak Pass. For SSL population trend, an additional 18 sets were constructed by adding one of the three diet variables (diet cluster, diet diversity, or frequency of occurrence of seven prey species) as explanatory variables to the four oceanographic and geographic variables. For each set of explanatory variables, initial models with all variables were fit using the GAM function in S-PLUS 2000 (MathSoft, 2000). Models were then constructed using step.gam, an automated stepwise regression procedure provided in S-PLUS. Step.gam explores combinations of variables, which include omitting a variable, including a variable as a linear term, and including a variable as a smooth term, and then selects the best-fit models and most significant variables through application of Akaike s information criterion (AIC). The AIC procedure does not provide a list of the combinations of variables tried or the resulting AIC values for those combinations. In order to obtain a quantitative value for the significance of the models and variables selected by the AIC, we applied the S-PLUS ANOVA function using Cp-statistics. The Cp-statistic, which is the Pearson chi-squared version of AIC, was chosen as the models were not necessarily nested. The Cp-statistic with the lowest relative value is considered the most significant and was selected as the overall best-fit model. RESULTS SSL diet and population trend The clustered regional patterns of SSL diet and population trends, updated with 3 more years of data through 2001, were consistent with those of Sinclair and Zeppelin (2002; Fig. 1). From Kodiak Island westward, diet and population trend again formed three closely aligned clusters with the least amount of overlap between cluster boundaries at Samalga Pass. Rookeries in the vicinity of Unimak Pass, the area of greatest long-term population stability (York et al., 1996), remained the area of highest diet diversity when data were extended through Seven prey species occurred at frequencies 5% falling into two clusters within the study area, one on either side of Samalga Pass. Cluster 2 (east of Samalga Pass) contained both the highest and lowest values for prey diversity but averaged higher overall relative to cluster 1 (west of Samalga Pass; Table 1). Cluster 2 also contained the most stable range in population trend counts ( ) compared with cluster 1 ( ). Cetacean distribution Nine species of cetaceans were seen during the cruises (Table 2; Fig. 3). Seven species were observed on both cruises, with the distribution of four fin, humpback, minke, and sperm whales aligning with regional patterns of SSL diet and population trends. For example, fin whale distribution (Fig. 3) extended from Shelikof Strait west to the Shumagin Islands, corresponding to SSL diet cluster 3 (Fig. 1a). Humpback whale distribution was coincident with SSL diet Table 2. Cetacean sightings during the 2001 and 2002 cruises along the Aleutian Islands Sightings (n) Individuals (n) Sightings (n) Individuals (n) Harbour porpoise (Phocoena phocoena) Dall s porpoise (Phocoenoides dalli) Pacific white-sided dolphin (Lagenorhynchus obliquidens) Killer whale (Orcinus orca) Baird s beaked whale (Berardius bairdii) Sperm whale (Physeter macrocephalus) Humpback whale (Megaptera novaeangliae) Minke whale (Balaenoptera acutorostrata) Fin whale (Balaenoptera physalus) The number of sightings refers to the number of times each species was seen during the cruise. The number of individuals represents the total number of individual animals counted during the cruise. The surveys were designed to determine presence/ absence of each cetacean species across the study area, thus sighting frequency was the value used in the GAMs.

7 Regional distribution patterns of marine mammals 229 (a) (b) Figure 3. Distribution of large and small cetaceans observed in the study area during the summer of 2001 and cluster 2 with many sightings at the Shumagin Islands and at Unimak and Akutan Passes, but only single sightings at Umnak, Samalga and Amukta Passes. Conversely, with the exception of a single sighting, killer whales and minke whales were only seen west of Unimak Pass in SSL diet cluster areas 1 and 2, and sperm whales only west of Samalga Pass in the region of SSL diet cluster 1 (Fig. 3). Of note, the distribution of Dall s porpoise did not reflect the longitudinal pattern of the large whales, but was seen across the entire area surveyed. Sightings of the remaining three species (harbor porpoise, Pacific white-sided dolphin and Baird s beaked whale) were too rare to infer distribution pattern (Fig. 3). Killer whales were frequently seen near all passes sampled, and were particularly common along the north Aleutian Island coastlines between Unimak Pass and Samalga Pass (Figs 3 and 4). Killer whale ecotype was determined for 29 of 42 encounters by genetic analyses of biopsied tissue, and by photographic identification of individual whales. Where ecotype was determined, most encounters involved the piscivorous (resident) killer whale ecotype (Fig. 4). No mammal-eating (transient) killer whales were identified during the 2001 cruise and only 8% (39 of 500) of the killer whales identified in 2002 were the mammal-eating (transient) ecotype. The four encounters with transient killer whales occurred in Unimak Pass and north of Akutan Pass. In addition, there was one sighting of the ecotype called offshore, animals that also appear to focus their diet on fish (Herman et al., 2004).

8 230 E.H. Sinclair et al. Figure 4. Distribution of resident (fisheating), transient (mammal-eating), and offshore (probably fish-eating) type killer whales based on biopsy samples collected from animals observed in the study area during the summer of 2001 and Modeling The number of rookeries included in the GAMs ranged from 14 to 16, depending upon data availability (Table 3). Model results for SSL diet diversity and population trend reflected the importance of maximum depth of the nearest measured pass to rookery location. When modelled as a function of pass oceanography and geography, diet diversity was a function of pass depth alone. Diet diversity and pass depth have a linear relationship with diet diversity increasing at shallower depths (Fig. 5a). Population trend was significantly (Cp ¼ ) related to maximum depth and position (east versus west) relative to Samalga Pass when modelled as a function of pass oceanography and geography alone (Table 4; Fig. 5b,c). In this model, which incorporated the 16 rookeries with population trend data, rookery populations near deeper passes and east of Samalga Pass had higher growth rates. Population trend was also modelled with and without (as a model control) diet data as explanatory variables along with the oceanographic and geographic variables for the 15 rookeries containing diet information. The best-fit model was that which included the frequency of occurrence of herring (Clupea pallasii) and salmon (Oncorhynchus sp.) as well as maximum depth. In this model, population trend (while always negative) was positively related (less negative) with increasing depth in the nearest measured pass as well as increased frequencies of occurrence of herring and salmon (Fig. 6; Table 4). The best-fit models for cetacean distribution varied among the five species (Tables 3 and 4), but all showed significant relationships with distance to the nearest measured pass and all except Dall s porpoise showed significant relationships with fluorescence, temperature, or both. The probability of sighting increased closer to the passes for all five cetacean species (Fig. 7a e). Whether the cell was east or west of the nearest pass was a factor for humpback whales and Dall s porpoise (Fig. 7a,b) because the best-fit models for both species used a combined distance direction variable rather than simple distance (Table 4). Both humpback whales and Dall s porpoise had unimodal relationships with distance, which peaked at small (Fig. 7a,b) distances and remained higher west of the passes. Humpback whales were most often observed within passes. Fluorescence in the top 50 m of the pass was consistently selected as a variable related to the distribution of humpback, minke, and killer whales (Tables 3 and 4; Fig. 8a c). The relationship between humpback presence and fluorescence was a unimodal smooth function that peaked at around 0.18 V which corresponds to Unimak Pass (Fig. 8a). Minke and killer whales were linearly related to fluorescence. The probability of sighting minke whales increased near passes with higher fluorescence values, such as Tanaga and Samalga (Fig. 8b), but killer whales were more likely to be sighted near low fluorescence, such as Umnak and Amukta (Fig. 8c). Mean temperature in the top 50 m was correlated to the distribution of humpback, sperm, and killer whales (Tables 3 and 4; Fig. 9a c). Humpbacks were linearly related to mean temperature in the nearest

9 Regional distribution patterns of marine mammals 231 Table 3. Sample size and explanatory variables for the best-fit models selected by Akaike s information criterion (AIC) from among each generalized additive model of Steller sea lion (SSL) diet diversity, SSL population trend, and humpback, minke, sperm, and killer whale and Dall s porpoise occurrence. SSL diet diversity SSL population trend SSL population trend w/out diet w/diet cluster w/diet diversity W FO Humpback whales Minke whales Sperm whales Killer whales Dall s porpoise Sample size /153 23/153 16/153 35/153 88/153 Mean pass fluorescence Linear Smooth Linear Linear Mean pass temperature Linear Smooth Linear Mean pass salinity Maximum pass depth Linear Linear Linear Smooth Distance to nearest pass Smooth Linear Linear Distance to nearest pass Smooth Smooth Smooth Smooth Smooth (west is negative) East/west of Samalga Pass Factor Factor Atka mackerel FO Cephalopod FO Herring FO Linear Pacific cod FO Pollock FO Salmon FO Linear Sand lance FO Diet cluster Factor Diet diversity Shaded boxes indicate variables that were not used in that model. Significant variables are indicated by the type of relationship, which was modelled (factor, linear function, and smooth function). For cetacean occurrence, the sample size numerator indicates the number of cells with sightings and the denominator is the number of cells with effort. FO represents SSL prey frequency of occurrence.

10 232 E.H. Sinclair et al. (a) 0.4 (b) (c) Maximum depth in nearest pass (m) West Direction from Samalga Pass East Figure 5. Generalized additive model function and residual values for (a) Steller sea lion (SSL) diet diversity (diet diversity with oceanographics see Table 3, column 1) in relation to maximum depth in the nearest pass; and (b) the model and residual values for SSL population trend (population trend on 16 rookeries with oceanographics see Table 3, column 2) in relation to maximum depth in the nearest pass, and (c) direction from Samalga Pass. Dashed lines represent upper and lower pointwise twice-standard error curves. Direction to Samalga was a categorical (non-numeric) variable and the width of each function bar represents relative sample size. pass, with higher probability of sighting near warmer (eastern) passes (Fig. 9a). The relationship between sperm whale presence and mean pass temperature was a smooth function which generally increased with decreasing temperature except for a slight decline between 4.4 (Tanaga Pass) and 4.2 C (Seguam Pass; Fig. 9b). The partial relationship between the probability of killer whale sightings and temperature was linear, with increased sightings associated with lower temperatures (Table 3; Fig. 9c). However, when killer whale presence was modelled as a function of temperature alone, this relationship reversed. Exploring bivariate models by pairing temperature with each of the other significant variables in turn revealed that the relationship between killer whales and temperature reversed when direction to Samalga was included, thus illustrating an interaction between these two variables. Dall s porpoise occurrence was most strongly related to the maximum depth in the nearest pass (Table 4; Fig. 10). This relationship was a smooth function with Dall s presence generally increasing with increasing depth except for a slight decline between 125 (Akutan Pass), 166 (Umnak Pass), and 168 m (Unimak Pass; Table 3). That model highlights a relationship between the presence of Dall s porpoise and maximum depth in the pass that, like SSL, is interesting because both species were widely distributed along the extent of the trackline. For the cetacean models, 58% of the 153 cells had Dall s porpoise sightings, but only 10 23% of the cells had sightings of the whale species ultimately limiting the power of the GAMs. Among cetaceans, the direction east or west of Samalga Pass specifically was selected by the AIC only for killer whales. Killer whales had a higher probability of occurrence east of Samalga Pass (Tables 3 and 4; Fig. 11). DISCUSSION The coincident breaks in regional distributions of fauna across the Aleutian Island chain are not limited to marine mammals (Call and Loughlin, 2005; Fadely et al., 2005), but parallel those of copepod (Coyle, 2005), fish (Logerwell et al., 2005), and bird species (Byrd et al., 2005) as well. Concomitant breaks in distributional patterns across species are particularly pronounced at Samalga Pass where physical changes are also marked (Ladd et al., 2005). The large temporal variability between data sets across these studies suggest that the regional physical characteristics of hydrography and current flow influenced by the passes (Ladd et al., 2005) along the eastern and central Aleutian Islands are predictable and comprise a series of eco-boundaries, with the most pronounced break at Samalga Pass. The similarity in spatial patterns demonstrated between species across the trophic scale provides strong inference for physical forcing or bottom up structuring of the marine environment in the central and eastern Aleutian Islands. The results of our exploration with GAMs identified geographic proximity to the nearest measured pass as a significant explanatory variable for SSLs and all cetaceans, further reinforcing that the passes play a role in structuring the nearshore ecosystem. Fluorescence (a measure of phytoplankton standing stock, but assumed to be a proxy measure of primary productivity)

11 Regional distribution patterns of marine mammals 233 Table 4. Generalized additive models (GAMs) constructed for Steller sea lion (SSL) diet diversity and rookery population trend and for the five predominant cetacean species in relation to the oceanographic and geographic aspects of the passes. Model Cp DCp SSL diet diversity variables Depth, fluorescence, distance, Samalga Depth Salinity, fluorescence, distance, Samalga Salinity + distance Temperature, fluorescence, distance, Samalga Distance + Samalga Temperature, fluorescence, distance (west negative), Samalga All variables combined Depth, fluorescence, distance (west negative), Samalga All variables combined Salinity, fluorescence, distance (west negative), Samalga All variables combined SSL population trend (16 rookeries) variables Depth, fluorescence, distance, Samalga Depth + Samalga Depth, fluorescence, distance (west negative), Samalga Depth + Samalga Temperature, fluorescence, distance, Samalga Fluorescence + Samalga Salinity, fluorescence, distance, Samalga Fluorescence + Samalga Temperature, fluorescence, distance (west negative), Samalga Fluorescence Salinity, fluorescence, distance (west negative), Samalga Fluorescence SSL population trend (15 rookeries) variables Depth, fluorescence, distance, Samalga, atka mackerel, Depth + herring + salmon cephalopods, herring, Pacific cod, pollock, salmon, sand lance Depth, fluorescence, distance (west negative), Samalga, atka mackerel, cephalopods, herring, Pacific cod, pollock, salmon, sand lance Depth + herring + salmon Temperature, fluorescence, distance (west negative), Samalga, diet cluster Salinity, fluorescence, distance (west negative), Samalga, diet cluster Temperature, fluorescence, distance (west negative), Samalga, atka mackerel, cephalopods, herring, Pacific cod, pollock, salmon, sand lance Salinity, fluorescence, distance (west negative), Samalga, atka mackerel, cephalopods, herring, Pacific cod, pollock, Fluorescence + distance (west negative) + diet cluster Fluorescence + distance (west negative) + diet cluster Distance (west negative) + cephalopods + salmon Distance (west negative) + cephalopods + salmon salmon, sand lance Temperature, fluorescence, distance (west negative), Samalga Distance (west negative) Depth, fluorescence, distance (west negative), Samalga Distance (west negative) SSL population (15 rookeries) variables Salinity, fluorescence, distance (west negative), Samalga Distance (west negative) Temperature, fluorescence, distance (west negative), Samalga, diet diversity Depth, fluorescence, distance (west negative), Samalga, diet diversity Salinity, fluorescence, distance (west negative), Samalga, diet diversity Depth, fluorescence, distance (west negative), Samalga, diet cluster Distance (west negative) Distance (west negative) Distance (west negative) Depth + distance (west negative) + diet cluster Fluorescence + salmon Temperature, fluorescence, distance, Samalga, atka mackerel, cephalopods, herring, Pacific cod, pollock, salmon, sand lance Salinity, fluorescence, distance, Samalga, atka mackerel, cephalopods, herring, Pacific cod, pollock, salmon, sand lance Temperature, fluorescence, distance, Samalga, diet cluster Temperature + fluorescence + diet cluster Fluorescence + salmon

12 234 E.H. Sinclair et al. Table 4. Continued Model Cp DCp Salinity, fluorescence, distance, Samalga, diet cluster Salinity + fluorescence diet cluster Depth, fluorescence, distance, Samalga, diet cluster Depth + fluorescence diet cluster Depth, fluorescence, distance, Samalga Depth + Samalga Depth, fluorescence, distance, Samalga, diet diversity Depth + Samalga Temperature, fluorescence, distance, Samalga Fluorescence Salinity, fluorescence, distance, Samalga Fluorescence Temperature, fluorescence, distance, Samalga, diet diversity Fluorescence Salinity, fluorescence, distance, Samalga, diet diversity Fluorescence Humpback whale variables Temperature, fluorescence, distance (west negative), Samalga Temperature + fluorescence distance (west negative) Salinity, fluorescence, distance (west negative), Samalga Salinity + fluorescence distance (west negative) Depth, fluorescence, distance (west negative), Samalga Fluorescence + distance (west negative) + Samalga Temperature, fluorescence, distance, Samalga Temperature + fluorescence distance Salinity, fluorescence, distance, Samalga Salinity + fluorescence distance Depth, fluorescence, distance, Samalga Depth + fluorescence distance Minke whale variables Temperature, fluorescence, distance, Samalga Fluorescence + distance Depth, fluorescence, distance, Samalga Fluorescence + distance Salinity, fluorescence, distance, Samalga Fluorescence + distance Temperature, fluorescence, distance (west negative), Samalga Fluorescence + distance (west negative) Depth, fluorescence, distance (west negative), Samalga Fluorescence + distance Salinity, fluorescence, distance (west negative), Samalga (west negative) Fluorescence + distance (west negative) Sperm whale variables Temperature, fluorescence, distance, Samalga Temperature + distance Depth, fluorescence, distance, Samalga Depth + distance Salinity, fluorescence, distance, Samalga Salinity + fluorescence + distance Temperature, fluorescence, distance (west negative), Samalga Temperature + distance (west negative) Depth, fluorescence, distance (west negative), Samalga Depth + distance (west negative) Salinity, fluorescence, distance (west negative), Samalga Salinity + fluorescence + distance (west negative) Killer whale variables Temperature, fluorescence, distance, Samalga Temperature + fluorescence + distance + Samalga Depth, fluorescence, distance, Samalga Depth + fluorescence + distance Salinity, fluorescence, distance, Samalga Salinity + fluorescence + distance + Samalga

13 Regional distribution patterns of marine mammals 235 Table 4. Continued Model Cp DCp Temperature, fluorescence, distance (west negative), Samalga Temperature fluorescence + distance (west negative) + Samalga Depth, fluorescence, distance (west negative), Samalga Depth + fluorescence distance (west negative) Salinity, fluorescence, distance (west negative), Samalga Salinity + fluorescence distance (west negative) + Samalga Dall s porpoise variables Depth, fluorescence, distance (west negative), Samalga Depth + distance (west negative) Temperature, fluorescence, distance, Samalga Distance + Samalga Depth, fluorescence, distance, Samalga Distance + Samalga Salinity, fluorescence, distance, Samalga Distance + Samalga Temperature, fluorescence, distance (west negative), Samalga Distance (west negative) Samalga Salinity, fluorescence, distance (west negative), Samalga Distance (west negative) + Samalga Variables are the initial subsets of explanatory variables tested. Model represents the combinations of variables selected from the variable subsets by step.gam and ordered as best-fit by Akaike s information criterion (AIC). Models in italics represent smooth functions. The category Cp is the value equated by Cp-statistics used to compare the best-fit models that were selected by the AIC, and the DCp is the difference between the smallest (most significant) Cp-value and the Cp-value for each of the other models that were run. and temperature (the variable most representative of boundary zone delineations) were significant explanatory variables for all large cetaceans regardless of whether they were apex predators, such as sperm and killer whales, or secondary level predators, such as minke and humpback whales. Although correlation does not imply causation, we interpret the utility of these independent variables to explain cetacean distribution as being consistent with the concept of bottom up structuring in the Aleutians. Dall s porpoise, consumers of both oceanic (salmon, cephalopods, myctophid fishes) and nearshore (salmon, pollock, herring) prey (Crawford, 1981; Beamish et al., 1999), bridge the trophic level between the two cetacean groups, as reflected in their presence throughout the study area and the lack of significance of specific nearshore hydrographic variables to their distribution in the models. Dall s porpoise was the only species besides SSLs to be strongly associated with maximum depth in the nearest pass. This could be related to similarities in their diets (Crawford, 1981; Beamish et al., 1999; Sinclair and Zeppelin, 2002), particularly with regard to herring and salmon, the two prey types related to population strength in SSL models. The models highlight the complexity of the marine ecosystem. Similarly to most models of this type (Levin, 1992; Forney, 2000; de Young et al., 2004), ours are limited by small sample sizes and large temporal variability. Despite these limitations, the results of the GAMs corroborated findings from the use of other types of models in this volume. The influence of distance to nearest pass on cetacean distributions is consistent with the effect of this distance on population trends of nearshore piscivorous birds (Byrd et al., 2005). This same pattern was detected for SSLs (adult females are also nearshore piscivores), prior to reducing our models to the 16 rookeries falling within the study area. As with large baleen whales in our study, planktivorous and offshore feeding birds were more abundant near the passes (Byrd et al., 2005). The oceanography of the passes themselves probably does not define faunal distributions, but, depending upon size and sill depth, a pass may represent a transition along a gradient of temperatures, salinities, and production (Fig. 12). As transition points along a physical gradient, the passes function as boundary zones where nutrients, zooplankton and upper trophic organisms aggregate, in this way serving as focal areas of bottom up structuring. In our focus on oceanographic features at the passes, we may have overlooked influential factors operating between the passes, a consideration that should be addressed in future work.

14 236 E.H. Sinclair et al. (a) 0.04 (b) (c) Maximum depth in nearest pass (m) Frequency of occurrence of Herring Frequency of occurrence of Salmon Figure 6. Generalized additive model functions and residual values for Steller sea lion population trend (population trend with oceanographics on 15 rookeries, see Table 3, column 3) in relation to (a) maximum depth in the nearest measured pass, and the frequency of occurrence of (b) herring, and (c) salmon (see Table 3, column 6). Dashed lines represent upper and lower pointwise twice-standard error curves. (a) (b) Distance to nearest pass (m) (c) (d) (e) Distance to nearest pass (m) Figure 7. Generalized additive model partial functions and residual values for (a) humpback whale, (b) Dall s porpoise, (c) minke whale, (d) sperm whale, and (e) killer whale occurrence in relation to distance to the nearest pass. For humpback whale and Dall s porpoise, the significant variable combined distance and direction to the nearest measured pass such that the sign indicated the direction (negative is west and positive is east) and the absolute value indicated the distance. Dashed lines represent upper and lower pointwise twice-standard error curves.

15 Regional distribution patterns of marine mammals 237 Figure 8. Generalized additive model partial functions and residual values for (a) humpback, (b) minke, and (c) killer whale occurrence in relation to fluorescence in the top 50 m of the nearest measured pass. Dashed lines represent upper and lower pointwise twice-standard error curves. (a) (b) (c) Mean fluorescence in top 50 m of nearest pass (volts) (a) (b) Figure 9. Generalized additive model partial functions and residual values for (a) humpback, (b) sperm, and (c) killer whale occurrence in relation to temperature in the top 50 m of the nearest measured pass. Dashed lines represent upper and lower pointwise twice-standard error curves. (c) Mean temperature in top 50 m of nearest pass (ºC) Figure 10. Generalized additive model partial function and residual values for Dall s porpoise occurrence in relation to maximum depth in the nearest pass. Dashed lines represent upper and lower pointwise twice-standard error curves Maximum depth in nearest pass (m) Distributions of humpback and sperm whales were clearly attributable to boundaries between neritic and oceanic water signatures (Ladd et al., 2005) and zooplankton prey fields (Coyle, 2005) at Samalga Pass. Humpback distribution was particularly aggregated near Unimak and Akutan passes, where abundance indices of neritic Calanus marshallae and Pseudocalanus spp. were high. Schooling fishes are a mainstay of humpback whale diet, so it is unclear whether the whales were consuming the zooplankton directly, or feeding upon forage fish consumers of the plankton. Conversely, sperm whales generally feed on large, deep

16 238 E.H. Sinclair et al. Figure 11. Generalized additive model partial function and residual values for killer whale occurrence in relation to direction to Samalga Pass. Direction to Samalga was a categorical (non-numerical) variable and the width of each function bar represents the sample size West Inside East Sighting east, west, or inside of the Samalga Pass living, squid and fish in oceanic waters (Rice, 1989) and were only observed west of Samalga Pass, the area most strongly influenced by a limited continental shelf and oceanic signature as indicated by cold temperatures. Of note, minke whale distribution crossed the neritic/oceanic boundary observed at Samalga Pass (Coyle, 2005; Ladd et al., 2005), and an increase in the probability of sighting minke whales was associated with high fluorescence, a factor that may be associated with euphausiids and copepods in minke whale diet (Mordy et al., 2005). Although the cetacean distributions from the Aleutian passes surveys convey only a brief snapshot, we suggest that the observed patterns are representative for each species, at least during early summer because they are similar to results from broad scale cetacean surveys conducted in July and August of 2001 and 2002 (Zerbini et al., 2004). Further, Zerbini et al. (2004) concluded that large whales observed in 2001 and 2002 are residing in traditional, prewhaling summering grounds, lending support to the idea that the summer distributional boundaries may be long-standing and the hydrographic influences of the area predictable. If long-standing, the dynamic regional signatures associated with at least some of the Aleutian passes may have far reaching ecological effects. The long-term diet (12 yr) and population trend (25 yr) data for SSLs provide the best example of the potential extent of these regional signatures in effecting the associations between predator and prey. For example, prey consumed in summer by post-parturient SSLs of the western stock are largely those that show strong, predictable, nearshore migratory movements along the Aleutian Island chain (Sinclair and Zeppelin, 2002; Fadely et al., 2005; Logerwell et al., 2005; McDermott et al., 2005). Among otariid pinnipeds, population gain and decline is most heavily influenced by the reproductive success of these adult Figure 12. Overview of alignment of spatial boundary points in the long-term distribution of Steller sea lion diet and population clusters and short-term observation of cetaceans in this study with geographic and oceanographic boundary points (from Hunt and Stabeno, 2005; Ladd et al., 2005) of Akutan Pass, Samalga Pass, and Amchitka Pass (ANSC).