Sample Sizes and Predator Fork Lengths

Pacific Herring was the most thoroughly sampled of the seven predator species, with a total of 511 stomachs being analyzed. Pacific Herring also exhibited the greatest size range (5–258 mm FL) (Table 1). The sample sizes for Coho Salmon and Chinook Salmon were much lower (n = 60 and 81 stomachs, respectively) than those of the other Pacific salmon species (n > 200) but these two species had the largest mean fork lengths of the Pacific salmon sampled (Table 1). Unlike Eulachon and Pacific Herring, all Pacific salmon sampled had fork lengths of 29 mm or more. There was large variability in both sample size and predator size amongmonths for all seven species (Table 1). There were significant differences in mean fork length between species (ANOVA; F = 73.645, df = 6, 1,539, P < 0.001), months (F = 55.768, df = 2, 1,539, P < 0.001), and year (F = 6.901, df = 2, 1,539, P < 0.001). A post hoc Tukey's HSD test indicated that all species differed significantly in mean fork length (P < 0.015) with the exception of Pacific Herring and Chum Salmon (P = 0.996) and Chinook and Coho Salmon (P = 0.12).

Prey Groups and Accumulation Curves

Using all 21 broad taxonomic prey groups, our species accumulation curves for the stomach contents of Pink Salmon, Sockeye Salmon, Chum Salmon, and Pacific Herring reached asymptotes (defined as a slope of less than 0.5 between the last two points); those for Coho Salmon, Chinook Salmon, and Eulachon did not reach asymptotes (Figure 2).

Prey Abundance in Predator Diets

The diets of Sockeye, Chum, and Pink Salmon were primarily comprised of calanoid copepods, with average percentages ranging from 42% to 63% of stomach contents in the different species (Figure 3A). These species also fed on a diverse mixture of other prey, particularly insects, fish, amphipods, decapods, and larvaceans (Figure 3A). The diets of Coho Salmon and Chinook Salmon were primarily composed of insects, fish, and decapods, while calanoid copepods comprised smaller average percentages of individual stomach contents (11% and 20% in Coho Salmon and Chinook Salmon, respectively). Calanoid copepods made up the largest proportion of individual Pacific Herring and Eulachon stomach contents, accounting for 52% and 73%, respectively, on average. Cladocerans (99% of which were marine) provided the next largest contribution for these two predator species.

TABLE 1.

Number of stomachs sampled by month and year for each predator species as well as summary statistics for fork length and prey richness in the stomachs of fish sampled in the Fraser River plume, May–August 1966–1968. Prey richness is defined as the number of prey species groups and was calculated both as the total across all individuals sampled for a given predator species and the range of richness within individual stomachs for that species.

Frequency of Occurrence

Crustaceans were the most important prey assemblage by FO, with calanoid copepods, amphipods, decapods, and cladocerans occurring in 66, 23, 15, and 13% of all fish sampled, respectively (Table 2). Insects and fish were the next most important groups, occurring in 23% and 21% of all stomachs, respectively. Calanoid copepods had the highest FOs in all the predator species except Chinook and Coho Salmon, for which insects and fish were higher. Fish, insects, and decapods had high FOs in all the Pacific salmon species, as did euphausiids in Coho Salmon. The only prey group other than calanoid copepods that had a relatively high FO in Pacific Herring and Eulachon diets was cladocerans.

FIGURE 2.

Species accumulation curves (based on the average of 10,000 permutations of fish number) for each of the seven predator species.

Fish Prey

A total of 21% of the stomachs analyzed contained fish prey (Table 2). The most important fish prey taxa by FO were Eulachon, Pacific Herring, Pacific Sand Lance, and rockfish (Table 3). In the stomachs containing fish, Eulachon were the most frequently found species in Pacific Herring (47%) and Sockeye Salmon (38%) (Table 3). Pacific Herring occurred most frequently in Chum Salmon (30%) and Pink Salmon (56%) stomachs. Pacific Sand Lances, on the other hand, were the most frequently occurring fish prey in Coho and Chinook Salmon (47% and 39%, respectively). Based on their abundance in predators' stomachs rather than their FO, Eulachon made up the largest proportion of total fish consumed by Chum Salmon (54%), Pink Salmon (47%), Chinook Salmon (55%), and Pacific Herring (70%) (Figure 3B). Eulachon, the most abundant fish caught in the trawls, also comprised a large proportion of fish prey in Sockeye Salmon stomachs (26%), though rockfish constituted the largest proportion for this predator (39%). Pacific Herring made up a substantial proportion of the fish prey in Chum (15%), Pink (45%), Coho (30%), and Chinook Salmon (13%) stomachs. Pacific Sand Lances, however, were the most consumed fish prey of Coho Salmon (45%).

FIGURE 3.

(A) Average proportion (by abundance) of each prey group in the stomachs of predator species and (B) average abundance of each fish prey species relative to the total number of fish prey consumed by predator species. The numbers at the tops of the bars are the sample sizes for the different predator species.

TABLE 2.

Frequency of occurrence of prey groups in the stomachs of individual predator species and all predators pooled. The sample sizes for predators are indicated in parentheses. The “other” category includes cumaceans, isopods, pycnogonids, and mysids.

TABLE 3.

Frequency of occurrence of fish prey species in the stomachs of individual predator species that contained fish and all predators pooled. The sample sizes for predators are indicated in parentheses.

Month and Year Effects

An ANOSIMrevealed high similarity (R = -0.0055, P = 0.86) in stomach contents among years but less similarity across months (R = 0.18, P = 0.001); thus years were combined for analysis. Calanoid copepods, fish, and insects all had relatively high FOs (>0.15) over all of the months sampled. The FO of calanoid copepods, however, declined from 0.91 in April and May to 0.42 in July. Cladocerans and decapods had higher FOs in June and July than in April and May (see Table S.1 in the supplement to this article). A DLI analysis revealed that calanoid copepods were the only strong indicator species differentiating stomachs sampled in different months (DLI value = 0.44), with cladocerans being a moderately important indicator for diets in June (0.14) and decapod crustaceans for diets in July (0.15).

Predator Species Effects: Jaccard Dissimilarity

In terms of monthly averages, the similarity of diets was high among all species except Eulachon (Table 4; for a breakdown by month, see Table S.2). For example, Chum, Pink, and Sockeye Salmon had Jaccard similarity indices ranging from 0.68 to 0.82. The diets of Chinook and Coho Salmon were also similar to each other (0.72). Pacific Herring diets were most similar to those of Chum, Pink, and Sockeye Salmon (0.62–0.68).When the Jaccard similarity indices were calculated using data pooled over all months, however, the diets of all species became highly similar (Table 4). The analysis comparing pooled predator species' diets resulted in no important indicator prey species.

Multivariate Regression Tree Analysis

The most parsimonious tree had five terminal leaves and explained only 13.1% of the variation in diets. The cross-validated relative error was 0.877, indicating that predator species, predator length, and month were generally poor predicators of diet variability (Figure 4). Predator length (<36.5 mm or ≥⃒36.5 mm) was the primary variable separating the diets of individual fish, explaining 6.20% of the diet variation. Approximately 60% of Pacific Herring and Pink Salmon and 76% of Eulachon stomachs were classified into the smaller length class. Of the other Pacific salmon, only 16% of Chum Salmon were sorted into this group. The split, therefore, generally separated the smaller Pink Salmon, Pacific Herring, and Eulachon from the larger Sockeye, Chum, Coho, and Chinook Salmon. Month then separated both predator length clusters, explaining 1.64% of the total variation in the stomach contents of the smaller fish and 3.08% of that of the larger fish. For the smaller fish, stomach contents in April and May clustered separately from those observed in June and July. Conversely, for larger fish, July diets were clustered separately from those of fish caught in April, May, and June. Within the cluster of larger fish caught in July, a final split (explaining 2.22% of diet variation), separated the five Pacific salmon species from Pacific Herring and Eulachon.

Although all prey species except the “other” category were statistically significant indicators for the terminal leaves of the MRT (P <0.05 fromapermutation test with 10,000 permutations), all DLI values for these leaves were relatively small, ranging from only 0.051 to 0.29. The strong indicator prey species (DLI values ≥⃒0.2) delineating predator diets within the five terminal leaves were calanoid copepods, insects, and decapods. Cladocerans, fish, and ostracods were also relatively important indicator species, with DLI values ≥⃒0.15. Calanoid copepods were the primary indicator species for the smaller predators, with an indicator value of 0.45, while fish prey (DLI = 0.33) were the primary indicator for the cluster comprising larger predators. Amphipods (DLI = 0.28), decapods (0.21), and insects (0.28) were also important indicators of this larger predator group.Within both the small-(0.56) and large-predator (0.54) clusters, calanoid copepods served as an indicator of diets observed earlier in the spring. Cladocerans, on the other hand, were indicative of summer (June and July) diets for the smaller predators (0.34), while decapod prey were indicators for July diets within the larger-predator size cluster (0.28). Within this final cluster, comprised of larger predators that were caught in July, insects (0.47) and fish (0.25) were the primary indicator species of Pacific salmon diets, while decapods (0.27) and calanoid copepods (0.40) were the primary indicators of Pacific Herring and Eulachon diets. Ostracods (0.25) and amphipods (0.22) were also indicative of Pacific Herring and Eulachon diets in July (Figure 4).

TABLE 4.

Jaccard similarity indices between the stomach contents of predator species calculated (1) by month (April–May, June, and July) and then averaged (monthly average) and (2) for all months combined (pooled over all months).

The results were qualitatively similar and explained a similar percentage of the variation in diets (19%) when the data were analyzed with trawl taken into consideration (Figure S.1). The relative importance of predator species increased (explaining 6.9% of the total variation) and the importance of month declined (explaining 4.2% of the variation), allowing for some species-specific diet shifts according to season to be seen. The overall patterns and conclusions remain unchanged, as length still determined the first split and the same species were determined to be indicators (Figure S.1). When we used the same explanatory variables to run an MRT on the first four axes obtained from an analysis of principal coordinates (in which the variation attributable to trawl was partialled out), predator diets also separated in a similar way based on predator species and length (Figure S.2). The diets of Pacific Herring and Eulachon were still distinguished from those of Pacific salmon, although length became more important than month in separating them.

Constrained Analysis of Principal Coordinates

Constrained variation comprised only 9.72% of the total variation, while month and trawl (which were partialled out) explained 31.78%. The variance explained by predator species and predator length (pseudo-F = 31.683, df = 7, 1,335, P = 0.001) was significant based on a permutation test with 1,000 permutations.

The first canonical axis explained 73% of the constrained variation but only 10.5% of the total variation in predator diets. This axis, in particular, separated Pacific Herring and Eulachon, which had mean negative scores, from the Pacific salmon species, which had mean positive scores (Figure 5A). Prey that loaded positively on this axis were fish, decapods, amphipods, ctenophores, insects, chaetognaths, and euphausiids. Calanoid copepods, cladocerans, molluscs, eggs, and barnacles loaded negatively on this axis (Figure 5B).

The second canonical axis explained 19% of the constrained variation, or 2.7% of the total variation. This axis partly separated Chum, Pink, and Sockeye Salmon, which had mean positive scores, from the larger Coho and Chinook Salmon, which had mean negative scores. Insects loaded strongly and positively onto the second axis, along with cladocerans and larvaceans. Fish, decapods, euphausiids, amphipods, calanoid copepods, and ostracods loaded negatively on this axis (Figure 5B). Phytoplankton loaded somewhat strongly onto both axes but were rare in the diets. Predator size was associated with both axes: larger fish were associated with fish, decapod, amphipod, and euphausiid prey, which loaded positively on the first canonical axis and negatively on the second.

FIGURE 4.

Multivariate regression tree showing the discriminating explanatory variables and important indicator species (DLI > 0.15) in the stomach contents of juvenile Pacific salmon, Pacific Herring, and Eulachon. The percentages at the nodes are the percentages of the total variation explained by the different splits. Sample sizes (n) are also given for the different groups produced by the splits. Only the one or two most important indicator species are illustrated here. See text for more details. (Images courtesy of J. Holden.)

Predator Species and Length

The MRT and CAP both identified the diets of the five Pacific salmon species studied as being somewhat distinct from those of Pacific Herring and Eulachon. The greatest similarity occurred between the diets of Chum, Pink, and Sockeye Salmon (Table 4), an overlap frequently reported in the literature (Auburn and Ignell 2000; Kaeriyama et al. 2004; Zavolokin and Efimkin 2007). Such overlap suggests the potential for competition under food-limited conditions, such as might occur owing to a mismatch in timing between out-migrating Pacific salmon and the spring phytoplankton bloom (Chittenden et al. 2010). Year, however, did not have much of an effect in the ANOSIM despite fork length differences between years and the inclusion of 2 years of high (1966 and 1968) and 1 year of low (1967) juvenile Pink Salmon abundance. This suggests that a higher abundance of juvenile Pink Salmon did not significantly alter the diets of the other juvenile fish that we studied. The diets of Chum, Pink, and Sockeye Salmon were primarily comprised of calanoid copepods (Figure 3A; Table 2) and thus could be particularly susceptible to declining abundances of calanoid copepods in the Strait of Georgia (Mackas et al. 2013). Chum Salmon, however, consumed higher frequencies of ctenophores than did the other Pacific salmon. The importance of gelatinous zooplankton in the diet of Chum Salmon has been noted previously in adults across different regions (Brodeur 1990; Welch and Parsons 1993), but not early in marine life. The ability to exploit this food source may become increasingly advantageous as anthropogenic gelatinous zooplankton (e.g., Aurelia aurita) blooms become more prominent (Purcell et al. 2007).

Despite both species' failing to reach an asymptote in the accumulation curves (Figure 2), we observed high similarity between Chinook Salmon and Coho Salmon diets (Table 4), a pairing common in other studies (Kaeriyama et al. 2004; Bollens et al. 2010; Cook and Sturdevant 2013). Chinook Salmon diets, however, contained somewhat more insects and calanoid copepods than those of Coho Salmon. Based on a comparison of contemporary diets in Puget Sound with those in the 1970s, Duffy et al. (2010) also identified insects as key, high-quality prey for Chinook Salmon, speculating that insects have become increasingly important as the prevalence of fish prey has decreased. Our results indicate that both prey types played an important role in the Strait of Georgia as far back as the 1960s.

The diets of Pacific Herring and Eulachon overlapped with those of the Pacific salmon species but were still distinct. The first axis of the CAP, for instance, divided most of the Pacific salmon from the Pacific Herring and Eulachon, largely based on the lack of insects in the stomachs of the latter two species (Figure 5A, B). While similar to the diets of Sockeye, Chum, and Pink Salmon, Pacific Herring diets were more distinct from those of Coho and Chinook Salmon (Table 4). Eulachon had the most unique diet among the predator taxa examined (although our diet characterization may have been incomplete: see Figure 2). While the MRT and CAP tended to group Eulachon with Pacific Herring, a preference for primarily calanoid copepods and cladocerans resulted in its specialized diet. Yang et al. (2006) also identified Eulachon as a specialist in the Gulf of Alaska. Such specialization, combined with recent declines in copepod abundance within the Strait of Georgia (Mackas et al. 2013), may have contributed to Eulachon population declines and the poor recovery of this species (Clavel et al. 2011; Gallagher et al. 2015).

Predator size also distinguished diets. The second axis of the CAP separated the insectivorous diets of smaller Pacific salmon from the piscivorous diets of larger Pacific salmon. This separation likely reflects both ontogenetic diet shifts as well as diet differences among species, particularly between the larger Chinook and Coho Salmon and the smaller Chum, Pink, and Sockeye Salmon. The MRT identified predator length as the most important variable distinguishing the diets of the predators, dividing them into clusters of smaller and larger fish (node at 36.5 mm FL). This node essentially divides the Pacific Herring, Eulachon, and smaller Pink Salmon from the larger Sockeye, Chum, Coho, and Chinook Salmon. Cladocerans, molluscs, and eggs distinguish Pacific Herring and Eulachon diets from those of the Pacific salmon, although some of the differences may have arisen from the higher proportion of small Pacific Herring and Eulachon in the data set compared with the juvenile Pacific salmon. Consistent with this hypothesis, cladocerans, molluscs, and eggs were not important indicator species for the Pacific Herring and Eulachon cluster within that of the larger predators identified by the MRT.

The wider range of predator sizes among the Pacific Herring that we sampled, and the consequently high variability in their stomach contents, likely facilitated both the similarities and differences noted between Pacific Herring and the other six predator species in our analyses. The lower end of the size range of Pacific Herring, for instance, overlapped with that of Eulachon, while larger Pacific Herring were more comparable in length to the larger Pacific salmon. When smaller (<30-mm FL) Pacific Herring and Eulachon were removed, the conclusions from the CAP did not change substantially (Figure S.3). The MRT did change somewhat, with different variables being responsible for the splits, but similar species groupings were observed overall (Figure S.4). The lack of a strong influence by the small fish on the CAP, in combination with the similarity in the resulting clusters obtained from both MRTs, indicates that the differences observed were not solely related to differences in size.

Sampling Month

The Strait of Georgia is characterized by strong seasonality in primary productivity, particularly in the late winter and early summer (Mackas et al. 2013). During early summer, large, nutritious calanoid copepods play an important role in the diets of juvenile Pacific salmon (Bi et al. 2011). Consistent with these seasonal productivity differences, we found significant differences in stomach contents between sampling months. Based on their FO, calanoid copepods weremore common in spring than in summer diets, while cladocerans and decapods had greater FOs later in the summer. Furthermore, sampling month defined a node within both size clusters of the MRT. Not surprisingly, calanoid copepods were indicative of diets observed earlier in the spring for both size clusters. Seasonal changes in diet have been reported in many other studies (Brodeur et al. 2007; Duffy et al. 2010) and suggest either ontogenetic shifts associated with changes in size and morphology or simply shifts in prey abundance (Nunn et al. 2012). Given the generalist diet observed in the Pacific salmon from our study and the dynamics of the Strait of Georgia, the diet shifts observed here were likely a result of changes in prey availability. For example, our data set includes high numbers of calanoid copepods during the spring, which coincides with the peak abundance of calanoid copepods in the Strait of Georgia in that season (El-Sabaawi et al. 2010). Neocalanus plumchrus, the dominant calanoid copepod in this region (El-Sabaawi et al. 2010), occurs at high densities in the Fraser River plume during the spring before completing its life cycle and migrating into deeper waters in late May and June (Mackas et al. 2013), presumably beyond the prey field of epipelagic-foraging juvenile fish. Still, ontogenetic shifts may be partially responsible for the changes in diet, given the length differences that we observed in predator species among months.

Fish Prey

This data set provides a unique record of fish prey in Pacific salmon, Pacific Herring, and Eulachon diets prior to recent declines in Eulachon populations (COSEWIC 2011) and the rebuilding of Pacific Herring stocks (DFO 2005). The role of Pacific Herring and Eulachon in the diets of predators was of particular interest because they were the most important prey in this study. Historical records showing the high prevalence of Eulachon in Pacific salmon diets are noteworthy because Eulachon are higher in lipids than are other common salmon prey (Anthony et al. 2000). The high consumption of Eulachon by the Pacific salmon in our data set—in contrast to their low importance in contemporary studies (Duffy et al. 2010; Beamish et al. 2012)—suggests that the loss of this energy-rich food has caused shifts in the feeding ecology of Pacific salmon that may have implications for their growth and survival rates.

A particularly important limitation to our study is that predator size, species, and month of capture were all confounded with one another, which makes disentangling the effects of each factor difficult. The differences in species and month, for example, could correspond to differences in length, given that nearly all predator species differed significantly in mean fork length, as did fish sampled in the early spring from those sampled in the summer. Furthermore, the small sample sizes for Chinook Salmon, Eulachon, and Coho Salmon and their failure to reach an asymptote in the species accumulation curves suggest that the sampling was not sufficient to fully describe the diets of these species.

Another caveat to consider in interpreting our results pertains to our metric for quantifying diet. A variety of metrics, each with its own advantages and drawbacks, have been used to describe the diets of fish, including percent by number, weight, and volume as well as frequency of occurrence (Chipps and Garvey 2007). Most studies of the diets of juvenile salmon use percent by volume or weight (e.g., Brodeur et al. 2007), but due to the nature of our data we were limited to using abundance or frequency of occurrence. Baker et al. (2014) argued that frequency of occurrence might be preferable to other metrics in many cases because of the unquantifiable and potentially significant error associated with the aforementioned methods of analyzing stomach content data. Given that our findings suggested groupings in the diet overlap of juvenile Pacific salmon species similar to those reported in the literature, we do not believe that our use of FO data greatly influenced our results.

Additionally, the MRT did not consider the lack of independence for fish caught in the same trawl, which could affect the relative importance of our explanatory variables in the MRT. However, the conclusions did not change drastically when diets were aggregated at the trawl level. Furthermore, when the variation due to trawl was first removed by an analysis of principal coordinates, the MRT run on the resulting axes still distinguished fish diets by species and predator length, as had the original MRT. This once again suggests that predator length and species are important factors influencing diets, with Pacific Herring and Eulachon having diets that are distinct from those of Pacific salmon. Month may be a less important factor and is confounded with trawl; this may be the reason why it did not define any clustering in predator diets once the variation due to trawl was removed. We ultimately used the results of the original MRT to present the full range of variation in the data because its interpretation is more intuitive while being consistent with the overall results of analyses that account for the nonindependence of fish from the same trawl. Furthermore, because the goals of the MRT were exploratory, we did not draw conclusions from a statistical test based on a probability distribution, for which the independence of the samples would be more important (De'ath 2002).

Hypotheses and Future Work

The diets of the Eulachon, Pacific Herring, and five species of Pacific salmon that we studied indicate that the historical food web in the Fraser River plume had strong interconnections. Such overlap could have had interesting implications for Pacific salmon growth and survival in light of the physical and biological changes that have occurred in the Strait of Georgia ecosystem since the 1960s. Increasing sea surface temperature in the strait, for example, has been associated with increased feeding rates among Pacific salmon (Daly and Brodeur 2015), potentially reflecting increased energy demands. But less availability of prey items high in energy content (such as euphausiids, copepods, and amphipods), coupled with a rise in gelatinous zooplankton (Davis et al. 1998; Li et al. 2013), compounds the pressures on Pacific salmon to meet their energy demands. A comparison of contemporary diets with those in the historical data could be useful for discerning how these changes have influenced fish foraging. Another major change in the Strait of Georgia that could have influenced the diets of these fish is the decline of the Fraser River population of Eulachon. Eulachon were the most abundant fish species caught in this study and the most abundant fish prey found in predator stomachs. The loss of such a high-quality prey item could have had consequences for the trophic interactions of the remaining predator species. We postulate that the larger role of Pacific Herring in contemporary Pacific salmon diets is a response to the declines in Eulachon availability.

Comparing historical diet overlap and prey availability with contemporary competition in the Strait of Georgia in light of the biotic and abiotic changes to this important rearing ground could be an important avenue for future studies. Of particular interest is the relative impact of these changes on juvenile Pacific salmon species, given that Chum Salmon readily utilize gelatinous zooplankton, a specialization that may be advantageous under contemporary conditions. The availability of baseline historical records in the Strait of Georgia makes these types of comparisons possible and may provide insight into ecosystem-scale changes. Such comprehensive data sets, spanning long time intervals, can facilitate ecosystem-based science and help inform management responses to long-term fluctuations in populations.