Seasonal variability in condition and spatial distribution of Chinook Salmon: Implications for ecosystem‐based management

Abstract

INTRODUCTION

Chinook Salmon Oncorhynchus tshawytscha sustain socially and economically valuable fisheries, are a key prey item for threatened resident killer whales Orcinus orca (Ford et al. 1998), and provide marine‐derived nutrient subsidies to diverse terrestrial ecosystems (Healey 1991). Unlike most Pacific salmon, Chinook Salmon mature at a wide range of ages and, depending on the population, individuals may rear either in continental shelf ecosystems or offshore (Weitkamp 2010; Shelton et al. 2019; Chinook Technical Committee [CTC] 2022). Although declines in Chinook Salmon abundance are widespread and thought to be linked to marine conditions (Kilduff et al. 2015; Welch et al. 2021), the abundance of some populations is stable or increasing (Atlas et al. 2023). For example, the spawner abundance of the Fraser River summer 4.1 (i.e., primarily age 4 at maturity and enter the ocean as subyearlings) stock management unit in 2023 was the largest on record (CTC 2024). Ultimately, differences in marine life history and distribution moderate interactions among Chinook Salmon, prey, predators, and fisheries, contributing to divergent dynamics among populations.

In the case of Chinook Salmon from Oregon, Washington, and British Columbia, coded wire tag recoveries suggest at least three general marine distributions: (1) distant offshore migrants, which are rarely encountered in marine fisheries; (2) far north migrants, which are common in Alaskan fisheries; and (3) residents, which are observed in fisheries along the west coast of Vancouver Island (WCVI), British Columbia, but are rarely observed in Alaskan fisheries (CTC 2022). However, marine species are often patchily distributed within their spatial range because foraging behaviors are driven by conditions that change at local scales (Sydeman et al. 2006). Given the importance of bottom‐up processes for growth and survival, there is increasing recognition that identifying patterns of habitat use is key to the effective management of marine species (Cox et al. 2018). Although fisheries‐dependent samples provide valuable estimates of stock‐specific Chinook Salmon abundance, these data are typically collected at relatively coarse spatial and temporal scales, with management areas spanning hundreds of kilometers and catches binned at monthly or seasonal intervals (Weitkamp 2010; Shelton et al. 2019; Freshwater et al. 2021). Fine‐scale variation in Chinook Salmon biological traits and spatial distributions can impact several dimensions of fisheries management. Many Chinook Salmon fisheries are mixed‐stock, including both abundant populations that can be sustainably harvested and less productive populations for which harvest impacts should be minimized. Divergent patterns of habitat use or migratory behavior that moderate a population's exposure to specific fisheries can be used to tailor interventions, such as time–area closures (Beacham et al. 2008; Dann et al. 2013; Welch et al. 2014). Similarly, differences between hatchery‐ and wild‐spawned individuals can inform the design of spatially restricted mark‐selective fisheries that are intended to harvest hatchery fish (Hoffmann and Pattillo 2010). Current Chinook Salmon management measures in southern British Columbia are targeted at areas as small as several square kilometers (Fisheries and Oceans Canada [DFO] 2024), resulting in a growing need to identify fine‐scale variation in Chinook Salmon habitat use and its impacts on marine ecology.

More broadly, successful ecosystem‐based management requires a nuanced understanding of interactions between predators and prey, which ultimately are moderated by fine‐scale spatiotemporal overlap. Pairing Chinook Salmon size and distribution data may improve our understanding of how bottom‐up drivers impact salmon productivity because juveniles become increasingly piscivorous as they grow and migrate away from nearshore areas (Hertz et al. 2016). Changes in habitat use coincident with changes in trophic position provide detailed information on the prey populations that are likely to drive Chinook Salmon survival (Duffy et al. 2010). Similarly, spatial variation in size or condition will influence interactions between Chinook Salmon and their own predators. A heterogeneous landscape of prey quality may result in predators, such as resident killer whales (Ford et al. 1998), favoring particular areas. Ultimately, interactions among condition, distribution, and stock identity can be used to quantify the relative value of different Chinook Salmon populations as prey to at‐risk predators.

A particularly diverse suite of Chinook Salmon populations is seasonally abundant along WCVI. Similar to the Salish Sea (Chamberlin et al. 2011), WCVI provides year‐round rearing habitat for Chinook Salmon stocks that do not migrate beyond the northern California Current (Weitkamp 2010; Freshwater et al. 2021). Additionally, it falls along the migratory corridor for Columbia River, Washington and Oregon coastal, and California populations migrating south; Fraser River, Puget Sound, and east coast Vancouver Island populations migrating through Juan de Fuca Strait; and local populations returning to spawn (Weitkamp 2010; Shelton et al. 2019; Freshwater et al. 2021). Finally, many, but not all, of the stocks that are present are supplemented by hatcheries in the United States or Canada. As a result, there is the potential for the condition and distribution of Chinook Salmon to vary at fine scales within this region along at least three ecological dimensions: age‐class, stock identity, and rearing history (i.e., wild or hatchery).

Here, we use data collected from a multiyear, fisheries‐independent troll survey to quantify variation in biological traits and habitat use (at scales of <50 km) of immature and adult Chinook Salmon. First, we tested the hypothesis that stocks, as well as hatchery‐ and wild‐spawned individuals, would differ in two metrics of condition: size and lipid content. Since condition is likely to be influenced by ontogeny, we also estimated the effect of sampling date and maturation stage (i.e., whether an individual was likely to spawn in a given year). Second, we used spatiotemporal models to determine whether Chinook Salmon distributions were correlated with multiple spatial (bottom depth, slope, distance to shore, and location) and temporal (diel, tidal, and seasonal cycles) variables rather than uniformly distributed within the study area. Third, we used spatiotemporal models to test the hypothesis that Chinook Salmon distributions vary among size‐classes (a proxy for age), among stocks, and between hatchery‐ and wild‐spawned individuals. Finally, we discuss the implications of variability in condition and habitat use in the context of ecosystem‐based management frameworks that are intended to rebuild at‐risk Chinook Salmon populations and increase prey availability for resident killer whales.

METHODS

Study area

Our study area was located along the southwest coast of Vancouver Island, British Columbia, in the North Pacific Ocean. The majority of sampling occurred in nearshore areas outside of Barkley Sound, as well as Amphitrite and La Perouse banks (study area size was approximately 1500 km²; Figure 1). The region's biological oceanography is influenced by a mix of wind‐induced upwelling on the outer shelf, persistent transport of nutrient‐rich water seaward from Juan de Fuca Strait, and topographically enhanced upwelling transported alongshore from northern portions of Vancouver Island (DFO 1990). As a result, the region is biologically productive and historically sustained fisheries for Pacific Herring Clupea pallasii, Pacific Hake Merluccius productus, and Sablefish Anoplopoma fimbria as well as Pacific salmon (DFO 1990). Additionally, the region includes critical habitat for resident killer whales (DFO 2017).

FIGURE 1

Set locations along the southwestern coast of Vancouver Island, British Columbia, with sampling years represented by different colors. Inset shows the study area relative to Vancouver Island, with southern resident killer whale critical habitat on La Perouse Bank (location of the study area) shown in red. Bathymetry of the study area is represented by 50‐m contour lines, with depths greater than 400 m (only present along the western border) excluded. The projection is Universal Transverse Mercator zone 10, and shapefiles for coastlines are from the rnaturalearth package (South 2017).

Open in new tabDownload slide

Sampling

Fisheries and Oceans Canada began a study in southwest Vancouver Island in 2019 to identify Chinook Salmon habitat use; the study was part of a larger effort to quantify spatiotemporal patterns of prey availability for endangered southern resident killer whales. Between April and September of 2019–2023, a commercial troller (FV Nawanie) was used to conduct fishing, sampling, and tagging operations. The primary objective of the field program was to deploy acoustic tags on Chinook Salmon to describe patterns of habitat use throughout the migratory corridor (Freshwater et al. 2024a; C. Freshwater, unpublished data), and the majority of effort was concentrated on Amphitrite and La Perouse banks to ensure that a sufficient number of tags were deployed. However, other habitats were sampled regularly throughout the study and we did not alter sampling efforts based on encounter rates (i.e., the vessel trolled continuously so that a given location was only fished once per tidal cycle). Model predictions were constrained to areas with spatial characteristics (location, depth, and slope) similar to those where effort occurred (Supplement S1 available in the online version of this article). We also conducted a simulation exercise to evaluate how a random sampling design would have impacted our results (Supplement S1).

To nonlethally sample and enumerate Chinook Salmon, we used standard commercial troll gear fished during approximately 30‐min sets (n = 1089). Specifically, we deployed two or four lines (one or two per vessel side), with four leaders per line and one hooked lure per leader. All hooks were size 4/0, and lures were a mix of spoons (metal lures imitating forage fish; top and third leaders) or flashers with hoochies (soft plastic lures imitating forage fish or cephalopods; second and fourth leaders). The depth of the bottom lure varied with bottom bathymetry, ranging from about 15.24 to 60.96 m (~50–200 ft). When a strike was observed, the line was retrieved and catch was enumerated by species. If no strikes were observed, the gear was retrieved every 15 min. All Chinook Salmon that were visually estimated to be smaller than 55 cm fork length were immediately released, as were all other species. In the case of Chinook Salmon that were estimated to be greater than 55 cm fork length, we landed the fish, removed the hooks, and immediately transferred the individuals to a vinyl trough with flow‐through seawater. Once the fish were in the trough, we measured fork length and girth with a measuring tape, noted the presence or absence of an adipose fin (removed prior to release by hatcheries), estimated the energy density by using a microwave oscillator (additional details below), and removed a tissue sample (primary: adipose, caudal, or pelvic fin clip; secondary: scales) for genetic stock identification. If an individual met the necessary criteria, it was tagged with an acoustic transmitter (Innovasea Model V13; 69 kHz) for separate studies on habitat use and survival, and all fish (tagged and untagged) were immediately released. Note that no results associated with tagging data are reported here.

We estimated tissue lipid concentration using a Distell Model 692 Fish Fatmeter (Distell Inc.), which contains a microwave oscillator that emits a low‐powered wave (frequency = 2 GHz; power = 2 mW) that interacts with water in somatic tissues. The fatmeter's sensor was placed along the dorsal surface at two positions just above the lateral line (Crossin and Hinch 2005). We averaged the two readings to generate a single index and converted this value to an estimate of whole‐body lipid content (percent wet weight; Lerner and Hunt 2023).

An experienced scale reader estimated ages visually from scale samples by enumerating freshwater and marine annuli. Resorbed scales and scales with uncertain assignments were excluded. Scales were only collected from legal‐sized Chinook Salmon, and ages were only estimated from scales collected in 2019, 2020, and 2021 (n = 523 individuals with readable scales). Chinook Salmon body size covaries with age, and age data were not available for all individuals; therefore, statistical analyses exploring ontogenetic effects focused on size rather than age.

Ecological covariates

Stock identity was inferred using single‐nucleotide polymorphisms (Beacham et al. 2018), and we assumed that stock identity was known without error when individual assignment probabilities exceeded 80% (>99% of samples). Given the large number of Chinook Salmon populations present in the study area, we pooled genetic stock assignments to create stock aggregates composed of populations spawning in the same geographic region and with similar life histories (i.e., freshwater out‐migration age and adult run timing; Table S1 available in Supplement S3 in the online version of this article). Hereafter, stocks with individuals predominantly migrating to the ocean in the year of emergence are referred to as subyearling stocks, while stocks with individuals that frequently migrate after overwintering as juveniles in freshwater are referred to as yearling stocks.

We used a multistep process to infer whether an individual was of hatchery origin. First, all individuals with a missing adipose fin were identified as hatchery fish. Second, all individuals with a genotype that matched parentage‐based tagging associated with Canadian hatcheries, as determined during genetic stock identification, were identified as hatchery fish (Beacham et al. 2018). Third, since nearly all hatchery‐origin fish produced in Washington and Oregon are marked prior to release (Anderson et al. 2020; Washington Department of Fish and Wildlife 2021; Oregon Department of Fish and Wildlife 2022), individuals assigned to stocks within these regions were assumed to be wild if they had intact adipose fins. Fourth, individuals with intact adipose fins that belonged to British Columbia (and did not have a parentage‐based tag), Oregon coastal, or California stocks or for which stock identity could not be determined were assigned to a third category, “unknown.”

Chinook Salmon mature at multiple ages and did not yet exhibit secondary characteristics associated with sexual maturity when encountered in this survey. However, we were able to assign maturity stage for a subset of individuals by using acoustic and passive integrated transponder tag detections to identify when fish entered freshwater (Freshwater et al. 2024a). We used these data to parameterize a Bayesian binomial regression model, which predicted maturation stage (i.e., whether an individual would mature during the year in which it was captured or would spend at least one additional winter at sea) as a function of stock identity, body size, and capture date. The model had relatively high classification accuracy (>87% with a 50% classification threshold; additional details are provided by Freshwater et al. 2024a). We used the fitted model to generate predictions for the full catch data set presented here, and an individual was assigned a status of “mature” if the estimated 90th percentile posterior interval for maturation probability did not include values below 50%.

We downloaded bathymetric data from the National Oceanic and Atmospheric Administration's (NOAA) 3‐arc‐second (British Columbia; NOAA National Geophysical Data Center 2007a) or 1/3‐arc‐second (coastal Washington; NOAA National Geophysical Data Center 2007b) digital elevation models. We calculated mean bottom depth and mean slope within 1‐ × 1‐km grid cells.

We included temporal covariates to test whether catch rates differed at weekly and subdaily scales. We accounted for tidal cycles by calculating the time to the next or previous slack tide, whichever was closer, for each set using hourly tide heights published by the Government of Canada for tidal station 8595 near Ucluelet, British Columbia (https://www.isdm‐gdsi.gc.ca/). We also accounted for changes in the magnitude of tides due to shifting lunar cycles by calculating the difference between the maximum and minimum tide heights observed each day. Finally, we included time since sunrise to account for diel cycles.

Modeling framework

We used generalized linear models (GLMs) with spatial and spatiotemporal Gaussian Markov random fields to identify size‐ and stock‐specific patterns of habitat use within the study region (additional details below). We included spatial and temporal covariates as fixed effects, and we accounted for residual spatial and spatiotemporal variability using Gaussian Markov random fields. We fitted a multilevel or hierarchical model, which included random intercepts for year, to generate average, population‐level predictions while accounting for interannual variability in abundance. Spatiotemporal models often result in more precise predictions relative to nonspatial models (e.g., Thorson and Ward 2013; Stock et al. 2019).

Our hypothesis was that spatial distributions would vary among size‐classes (a proxy for age), among stocks, and between hatchery and wild individuals. Unfortunately, it was not possible to estimate unique spatiotemporal distributions along each dimension simultaneously due to the large number of unique parameters and because size, stock identity, and rearing history are often confounded. Instead, we fitted multiple spatiotemporal models, each of which was focused on one ecological dimension. In the first model, individuals were assigned to one of four size‐bins based on fork length: sublegal (<55 cm), small (55–65 cm), medium (65–75 cm), and large (>75 cm). The second model used categories based on genetic stock assignment (Table S1). The stock model excluded east coast Vancouver Island fall‐run subyearling and Fraser River yearling populations due to convergence issues with rarely encountered populations. The third model estimated differences in distribution among hatchery fish, wild fish, and individuals of unknown rearing history. Finally, as a supplementary analysis we fitted a fourth model that estimated stock and hatchery/wild effects simultaneously to a subset of individuals belonging to three stock groups (lower Columbia River, Puget Sound, and Fraser River fall) that were relatively abundant and that included substantial numbers of both hatchery and wild individuals. Except for the size‐class model, we included only medium and large individuals (i.e., >65 cm fork length) to minimize the confounding effect of size on comparisons and because sublegal individuals were not sampled for genetic stock identity or the presence of an adipose fin. For the purposes of model fitting, each size‐class, stock, or rearing history type was considered a categorical group g in subsequent equations. We did not fit a maturity stage model since it was strongly correlated with body size in a previous study (Freshwater et al. 2024a).

We accounted for latent differences in spatial distribution as well as monthly variation in spatial distribution that differed among groups (i.e., group‐specific spatiotemporal random fields). We included a Gaussian Markov random field constrained by a Matérn covariance function, with ω denoting a vector of ω_j across space. We also modeled temporal changes in distribution using group‐specific random fields by month m, which followed a random walk (⁠$ϵg,m$ and δ_j,g,m). Given that sampling effort in April was low and concentrated at the end of the month, we pooled April and May samples when estimating spatiotemporal random fields. Spatial and spatiotemporal random fields were fitted with a shared marginal standard deviation among categories (covariance matrices Σ_ω and Σ_ϵ). Spatial and spatiotemporal random fields had separate κ parameters defining the Matérn decorrelation rate with distance. Spatial correlations were modeled as anisotropic (i.e., directionally dependent; Haskard 2007; Thorson et al. 2015).

We modeled spatial random fields using a triangulated mesh (Lindgren et al. 2011) with 175 vertices. We then used the estimated values at each knot and bilinear interpolation to approximate a continuous field that represented both spatial and spatiotemporal effects (Lindgren et al. 2011).

We used conditional predictions to evaluate the relative impact of individual fixed effects while holding other covariates constant at reference values. We explored spatial patterns in each categorical group's catch rate by projecting model predictions onto a 1‐ × 1‐km grid (Universal Transverse Mercator zone 10 projection) that was approximately bounded by our sampling locations and constrained to depths no greater than 225 m. We fitted spatiotemporal GLMs using the sdmTMB package (Anderson et al. 2022), which combines automatic differentiation and the Laplace approximation via the TMB package (Kristensen et al. 2016) with input matrices for the stochastic partial differentiation equation approximation to Gaussian random fields from the INLA (Integrated Nested Laplace Approximation) package (Rue et al. 2009) and minimizes the negative marginal log likelihood of the model with the nonlinear minimizer nlminb() in R (R Core Team 2021). We fitted both hierarchical GAMs and spatiotemporal GLMs in R version 4.3.2 (R Core Team 2021).

We confirmed that the nonlinear optimizer had converged by checking that the Hessian matrix was positive definite and that the maximum absolute gradient for all fixed effects was less than 0.005. We calculated randomized quantile residuals with fixed effects that were set at their maximum likelihood estimates and random effects that were sampled with Markov chain–Monte Carlo (sensu Rufener et al. 2021) to evaluate residual diagnostics independent of the Laplace approximation (Figures S3–S5 available in Supplement S2 in the online version of this article).

RESULTS

Stock composition, age composition, and condition

We captured 3162 Chinook Salmon between 2019 and 2023. Among these fish, 1638 legal‐sized individuals were biologically sampled and 1605 individuals were successfully assigned to stock based on genetic assignments. Stock diversity was greatest during midsummer (Figure 2). Stock composition differed between hatchery and wild groups; the former was dominated by Puget Sound and lower Columbia River individuals, while the latter had relatively more upriver Columbia and Fraser River individuals (Figure 2). Similarly, the relative proportion of individuals that were predicted to be mature in the year of capture varied among stocks (Figures 2 and S2).

FIGURE 2

Chinook Salmon stock composition estimates by month (top panel), rearing history (bottom left panel), and maturity stage (bottom right panel). Note that only legal‐sized individuals (i.e., >55 cm) were sampled and assigned to a stock, rearing history, or maturation stage. Stocks included west coast Vancouver Island (WCVI), east coast Vancouver Island (ECVI), Fraser River yearling (Fraser Year.), Fraser River summer 4.1 (Fraser 4.1), Fraser River fall (Fraser Fall), Puget Sound, lower Columbia River (Low Col.), upriver Columbia River (Up Col.), Washington and Oregon coastal (WA_OR), California (Cali), and unidentified (NA).

Open in new tabDownload slide

Based on age estimates from scales, we sampled both yearling and subyearling life history types (i.e., freshwater age 1 and age 0) as well as four ocean age‐classes. We did not age sublegal individuals, but given their size as well as the presence of both yearling and subyearling stocks in the area, sublegal individuals were likely a mix of ocean age 0 (i.e., entered the ocean during the year in which they were encountered) and ocean age 1. The majority of legal‐sized fish were ocean age 2, whereas ocean age‐4 individuals were particularly rare (Figure S1). Ocean age covaried with size‐class; large individuals were predominantly ocean age 3, while small and medium individuals were predominantly ocean age 2 (Figure S1).

Fork length and lipid content consistently varied among stocks and maturity stages, with Fraser River yearling stocks having the largest mean fork length and the greatest lipid content (Figure S2). Furthermore, differences between stocks weakened but were still present after we accounted for sampling day, year, rearing history, and maturation stage (Figure 3).

FIGURE 3

Predicted stock‐specific fork length (cm; top panel) and lipid content (% wet weight; bottom panel) of Chinook Salmon. Predictions are marginal effects that fix rearing history and capture date to reference values. Points represent mean estimates, and whiskers represent 95% confidence intervals. Stock abbreviations are defined in Figure 2.

Open in new tabDownload slide

Besides variation among stocks, fork length and lipid content also showed strong seasonal effects. Both fork length and lipid content increased nonlinearly over the summer (Figure 4). In the early summer, immature fish, which were predicted to remain at sea for an additional year, had markedly lower lipid content than mature fish, which were predicted to spawn in the year of capture; however, by the end of summer, immature individuals had a slightly greater mean lipid content than mature fish after we accounted for size and stock identity (Figure 4). Lipid content also increased with fork length, but this relationship weakened at the largest observed body sizes, likely due to variation among stocks and capture dates (Figure S6). After other covariates were accounted for, hatchery fish were smaller than unknown fish and wild fish, although these effects were modest (Figure S7).

FIGURE 4

Model‐predicted seasonal changes in mean fork length (cm; left panel) and lipid content (% wet weight; right panel) of mature (dark lines) and immature (light lines) Chinook Salmon. Predictions represent conditional effects, with other covariates fixed at mean or reference values. Ribbons represent 95% confidence intervals.

Open in new tabDownload slide

Variation in abundance

Abundance was negatively correlated with bottom depth and positively correlated with bottom slope in each model (Figures 5 and S8). In the size model, abundance was greatest immediately after sunrise and when tide changes were small. The effect of slack water changed from negative to positive as the size of daily tide changes increased. Tide effects, however, were relatively modest and uncertain for this model (Figure 5). Parameter estimates differed slightly among models but were directionally similar (Figure S8).

FIGURE 5

Predicted effects from the size‐based model of bottom depth, bottom slope, hours from slack, and hours from sunrise on the scaled abundance of Chinook Salmon. Predictions for hours from slack include high‐magnitude (dotted line) and low‐magnitude (dashed line) daily tidal changes. Each covariate's predictions are conditional, with all other covariates fixed at reference values, and are scaled so that 1.0 represents the maximum observed abundance. Lines represent mean estimates and ribbons represent 95% confidence intervals.

Open in new tabDownload slide

Abundance was inversely related to size, with sublegal Chinook Salmon being an order of magnitude more abundant than large individuals (Figure 6). Size‐classes also showed distinct seasonal patterns; the abundance of small and sublegal‐sized Chinook Salmon increased from spring to summer, while the medium and large size‐classes peaked in abundance during July (Figure 6). Abundance also varied among stocks within the medium and large size‐classes. Lower Columbia River and Puget Sound fish were approximately twice as abundant as the next most common stocks (Fraser River fall and upriver Columbia River; Figure 7). Fraser River summer 4.1, California, WCVI, and Washington/Oregon coastal stocks were rarer still, while too few east coast Vancouver Island and Fraser River yearling stocks were encountered for reliable model estimates. Stocks also showed distinct seasonal patterns. Fraser River summer 4.1 and Washington/Oregon coastal stocks peaked in late summer, while lower Columbia River and particularly Puget Sound stocks had relatively stable abundance from early May to early September (Figure 7). Hatchery fish were nearly twice as abundant as wild fish and showed a more gradual seasonal increase in abundance (Figure S9).

FIGURE 6

Estimated mean catch rates (left panel) and weekly trends in predicted scaled abundance (right panels) by Chinook Salmon size‐class. Mean catch rates reflect size‐specific intercepts assuming fixed effort (individuals captured with two lines, four leaders each, fished for 500 m). Weekly trends are conditional predictions (other covariates are fixed at reference values) and scaled within a size‐class so that 1.0 represents the maximum observed abundance. Points and lines represent mean estimates, and whiskers and ribbons represent 95% confidence intervals.

Open in new tabDownload slide

FIGURE 7

Estimated catch rates (left panel) and weekly trends in predicted scaled abundance (right panels) by Chinook Salmon stock. Mean catch rates reflect stock‐specific intercepts assuming fixed effort (individuals captured with two lines, four leaders each, fished for 500 m). Weekly trends are conditional predictions (other covariates are fixed at reference values) and scaled within a stock so that 1.0 represents the maximum observed abundance. Points and lines represent mean estimates, and whiskers and ribbons represent 95% confidence intervals. Stock abbreviations are defined in Figure 2.

Open in new tabDownload slide

Variation in spatiotemporal distribution

Chinook Salmon catch distributions varied among size‐classes (Figure S10), stocks (Figure S11), and rearing history types (Figure S12). By fitting catch data to a spatiotemporal model, we were able to separate sampling month effects on group‐specific spatial distributions (via the random fields $ϵg,m$⁠) from sampling week effects on mean abundance, which provided inference on how spatial distributions evolved seasonally. We found much stronger evidence of monthly variability in spatial distributions among size‐classes than among stocks or rearing history types. For example, medium and large individuals became relatively more common in nearshore areas through July, August, and September (Figure S13). The distribution of small individuals was relatively stable in the southern half of the study area, while the distribution of sublegal fish expanded over the course of the summer (Figure S13). Conversely, spatial distributions within a stock (Figure S14) or rearing history (Figure S15) were relatively stable among months.

We incorporated the additive effects of spatial variables (e.g., bathymetry), weekly changes in abundance, spatial random fields (ω), and group‐ and season‐specific spatiotemporal random fields to generate integrated predictions of how Chinook Salmon abundance varied spatially during the summer months. All three models indicated that Chinook Salmon were generally concentrated along bathymetric features, especially inshore of drop‐offs, and were relatively rare in deeper habitats (Figures 8–10). We found strong evidence that spatial distributions differed among size‐classes. For example, sublegal fish were concentrated in the eastern and southern portion of the survey domain off the mouth of Barkley Sound (Figure 8). Medium and large individuals were concentrated offshore and in the northern nearshore portion of the study domain, but they became more widespread as the summer progressed (Figure 8).

FIGURE 8

Predicted size‐class‐specific spatial distributions of Chinook Salmon from late spring to early fall. Abundance is scaled within a size‐class, with yellow cells representing the maximum prediction and purple cells representing the minimum prediction. Predictions account for spatial fixed effects, size‐class‐specific weekly fixed effects, the static spatial random field shared among size‐classes (ω), and size‐class‐specific spatiotemporal random fields that varied between months (⁠$ϵg,m$⁠). Time to slack, time to sunrise, and tide magnitude effects are set to mean values.

Open in new tabDownload slide

We also found differences in distribution among medium and large individuals within the eight most abundant stocks. Individuals belonging to the WCVI stock were concentrated in the northern portion of the domain, nearest to shore (Figure 9). Fraser River summer 4.1, Puget Sound, and lower Columbia River stocks had relatively broad distributions (Figure 9). Fraser River fall, upriver Columbia River, Washington/Oregon coastal, and California stocks were most abundant offshore, though their specific distributions differed from one another (Figure 9). Although the seasonal abundance of hatchery fish, wild fish, and individuals of unknown rearing history differed due to the effect of sampling week, all three rearing history types had similar spatial distributions (Figure 10). Furthermore, a supplementary analysis focusing on a subset of stock groups indicated that hatchery and wild individuals within a stock had similar distributions (Figure S16).

FIGURE 9

Predicted stock‐specific spatial distributions of Chinook Salmon from late spring to early fall. Abundance is scaled within a stock, with yellow cells representing the maximum prediction and purple cells representing the minimum prediction. Predictions account for spatial fixed effects, stock‐specific weekly fixed effects, the static spatial random field shared among stocks (ω), and stock‐specific spatiotemporal random fields that varied between months (⁠$ϵg,m$⁠). Time to slack, time to sunrise, and tide magnitude effects are set to mean values. Stock–month combinations with highly uncertain predictions (standard deviation > 2) are not shown. Stock abbreviations are defined in Figure 2.

Open in new tabDownload slide

FIGURE 10

Predicted rearing history‐specific spatial distributions of Chinook Salmon from late spring to early fall. Abundance is scaled within a rearing history type, with yellow cells representing the maximum prediction and purple cells representing the minimum prediction. Predictions account for spatial fixed effects, origin‐specific weekly fixed effects, the static spatial random field shared among rearing histories (ω), and rearing history‐specific spatiotemporal random fields that varied between months (⁠$ϵg,m$⁠).

Open in new tabDownload slide

Model uncertainty was generally more strongly related to time than to space. Overall uncertainty was greatest early and late in the year; for ecological groups with relatively small sample sizes; and in deeper or offshore locations, where relatively less sampling effort occurred (Figures S17–S19).

DISCUSSION

Chinook Salmon body size and lipid content varied seasonally and with maturation stage in our study area. Nevertheless, we found evidence of stock‐specific differences in size and lipid content even after accounting for seasonal changes in condition. Conversely, hatchery and wild fish within a given stock had only small differences in size and negligible differences in lipid content. Chinook Salmon catch rates, used as a proxy for abundance, were strongly influenced by depth and slope, with hot spots concentrated in nearshore and bank habitats having complex bathymetries. Abundance varied at multiple temporal scales, but the influence of sunlight and tidal cycles was modest compared to strong seasonal changes in abundance. Moreover, seasonal abundance and spatial distributions varied among size‐classes (a proxy for age) and among stocks but not between hatchery and wild individuals.

Variation in condition

We found that the maturation stage of Chinook Salmon was strongly correlated with individual condition. During early summer, mature individuals had a higher lipid content relative to body size than immature individuals that were predicted to remain at sea for an additional year. However, because immature individuals showed rapid nonlinear increases in lipid content, by late summer the mean lipid content of immature fish was approximately the same as that of mature fish. Notably, increases in the lipid content of immature individuals co‐occurred with a modest reduction in growth rates (as indexed by changes in mean fork length). Together, these patterns suggest that immature age‐classes shift their energy allocation from somatic growth to lipid storage prior to overwintering at sea, consistent with a similar hypothesized transition prior to juveniles' first winter at sea (Beamish et al. 2008).

Chinook Salmon belonging to populations that begin freshwater migrations early in the year and spawn in interior locations have a higher lipid content than individuals belonging to fall‐run populations that spawn in coastal systems (O'Neill et al. 2014; Lerner and Hunt 2023), likely due to more challenging migrations (Quinn 2018). However, previous comparisons have used samples collected in terminal marine areas or shortly after freshwater entry, resulting in stock‐specific energy density estimates that were confounded with seasonal differences in migration timing. By using nonterminal samples collected over multiple months, we determined that among‐stock differences in size and lipid content are widespread and present weeks prior to freshwater entry.

Shared drivers of abundance and distribution

The predicted abundance of Chinook Salmon was greatest in shallow areas and along steep drop‐offs near bank edges. Within our study area, there were two general hot spots of abundance—a smaller northern, nearshore zone and a larger zone on La Perouse Bank between two areas of deep water. Chinook Salmon are likely concentrated in these areas to maximize foraging opportunities. In continental shelf systems, interactions among tidal cycles, currents, and physical structures create local upwelling that enhances primary and secondary productivity (Bakun 1996; Genin 2004; Reese et al. 2011). Along southwest Vancouver Island, Chinook Salmon are likely targeting Pacific Herring, which are particularly abundant inshore of the continental shelf break (Godefroid et al. 2019). Notably, similar high‐relief areas are also where Chinook Salmon diving behaviors (Freshwater et al. 2024a) and southern resident killer whale foraging behaviors (Thornton et al. 2022) are locally concentrated.

Catchability, especially when using hook‐and‐line gear, is influenced by the condition and behavior of individual fish as well as dynamic environmental conditions (Stoner 2004; Lennox et al. 2017). After spatial and seasonal variables were accounted for, Chinook Salmon catch rates were greatest shortly after sunrise and when daily tidal changes were relatively small. In Pacific salmon fisheries, commercial and recreational anglers often note that catches are influenced by daylight, tides, and lunar cycles (e.g., Island Fisherman Magazine 2023). We note, however, that these effects were relatively modest compared to seasonal and spatial effects on abundance.

Size effects on abundance and distribution

Chinook Salmon size and age covary such that differences among Chinook Salmon size‐classes in abundance and distribution reflect ontogenetic processes. In our study, catch rates of sublegal and small size‐classes increased throughout the summer, likely due to dispersal from early marine rearing areas (Beamish et al. 2011; Tucker et al. 2011) as well as a transition toward piscivory, which may result in changes in habitat use and increased vulnerability to troll gear. Conversely, the abundance of medium and large individuals peaked in July. Although individual stocks have run timings along the southwest coast of Vancouver Island that are earlier or later in the year, midsummer coincides with particularly high stock diversity and total abundance (Freshwater et al. 2021).

The spatial distribution of each size‐class was not stable but instead evolved from spring to early fall, resulting in seasonal hot spots of abundance. Large and medium individuals were initially most abundant offshore, but they became more evenly distributed in July, August, and September, presumably reflecting stock‐specific distributions and migration timing (additional details below). Small and sublegal individuals showed a more complex distribution wherein they became increasingly concentrated outside of Barkley Sound during late summer. Trawl survey data showed a shift from predominantly Columbia River yearling populations during the early summer (i.e., populations that migrate rapidly north and further from shore) to local populations that begin to migrate out of WCVI inlets in late summer and fall as they grow and begin to feed on fish (Tucker et al. 2011; Hertz et al. 2016; Freshwater et al. 2024b). Although we did not explicitly model changes in the abundance and distribution of mature and immature life stages, size‐specific patterns will be highly correlated with maturation stage (e.g., most sublegal and small individuals are immature).

Stock effects on abundance and distribution

Stock‐specific abundance within the study area changed from week to week, consistent with variation in migration timing among Chinook Salmon populations (Healey 1991). Stocks showed relatively distinct but seasonally stable spatial distributions. Lower Columbia River and Puget Sound populations were found throughout the study area; however, higher resolution data are necessary to determine whether individuals commonly move between the two habitats or have relatively constrained distributions after juvenile dispersal. Notably, Fraser River fall populations, which are also largely resident in southern areas (Weitkamp 2010), did not show a similar distribution. The remaining stocks are far north or offshore migrants. The majority of Fraser River summer 4.1, Washington/Oregon coastal, and upriver Columbia River fish were found further from shore, but they were intermittently captured in nearshore habitats as well. We hypothesize that while the main migratory corridor from northern rearing areas for these stocks is near the shelf break, it is not uncommon for individuals to deflect inshore, perhaps responding to prey cues. The opposite pattern appears to hold for WCVI stocks, which predominantly migrate close to shore (DFO 2012). The majority of California Chinook Salmon rear south of Washington (Shelton et al. 2019). Since many of the California individuals that we encountered were immature, we expect that a relatively small proportion of California fall‐run Chinook Salmon arrive in the year prior to spawning.

Stock‐specific behaviors are most clearly resolved by the stock model, but they will also contribute to variation among size‐classes and between rearing histories. This is because the relative contribution of each stock varies (1) among size‐classes due to differences in marine life history and (2) among rearing history types due to differences in hatchery supplementation among stocks.

Rearing history effects on abundance and distribution

Hatchery‐reared fish were approximately twice as abundant as wild individuals. Due to the high marking rates at Washington and Oregon hatcheries (Anderson et al. 2020; Washington Department of Fish and Wildlife 2021; Oregon Department of Fish and Wildlife 2022) and widespread parentage‐based tagging at British Columbia hatcheries, relatively few individuals had an unknown rearing history. Seasonal trends in abundance differed among groups; hatchery fish increased gradually from moderate to high abundance, while wild fish and individuals with an unknown rearing history had more peaked seasonal patterns. Unlike stocks or size‐classes, the three rearing history groups had relatively similar spatial distributions, with hot spots of abundance in both nearshore and offshore areas. Furthermore, this pattern persisted when we examined different rearing histories within a stock. Seasonal patterns in the abundance of rearing history groups are likely driven by stock composition since certain populations and marine life histories have a greater proportion of hatchery production than others (e.g., Puget Sound and lower Columbia River).

Common distributions among hatchery and wild individuals support previous evidence that estimates of distribution and exploitation rates from hatchery indicator stocks are reasonable proxies for closely related wild populations (Weitkamp 2010). Nevertheless, differences among Chinook Salmon stocks in spatiotemporal distribution, body size, and energy content emphasize that hatchery fish from coastal stocks, such as Puget Sound and the lower Fraser River, are not ecologically equivalent to interior yearling populations, which are less heavily enhanced in southern British Columbia (DFO, Salmonid Enhancement Program, unpublished data). Finally, similar distributions between wild and hatchery fish highlight that ecological interactions have the potential to occur across multiple marine life stages—not only immediately after ocean entry or prior to spawning migrations. The ecological characteristics of hatchery fish, as well as their potential interactions with wild populations, should be considered when designing enhancement strategies that are intended to support ecosystem‐based management objectives.

Limitations

Our estimates of abundance may be influenced by several limitations to our analysis. We were unable to account for the full suite of environmental variables that might influence Chinook Salmon habitat use, and our study addressed only a small portion of the species' range. For example, Chinook Salmon distributions (Shelton et al. 2021) and vertical habitat use (Sabal et al. 2023) may be sensitive to changes in temperature, particularly within the southern portion of the species' range. We found weak effects of temperature on behavior in our study area (Freshwater et al. 2024a), perhaps because seasonal and interannual variability was relatively modest (standard deviation = 1–2°C). Future studies that quantify spatial variability in the effect of environmental drivers across diverse oceanographic domains would provide valuable information on how climate change impacts may differ among Chinook Salmon life history strategies.

Similarly, our study focuses on only a portion of the species' range. Although it is plausible that Chinook Salmon are concentrated in high‐relief areas on the shelf regardless of location, size‐ and stock‐specific differences may arise due to variation in the life history strategies that are present. We also note that we did not use stratified random sampling. Although our sensitivity analysis suggests that the present estimates of spatial relationships were robust to sampling design, our predictions in sparsely sampled areas may be less reliable.

Estimates of abundance inferred from catch data may deviate from true abundance due to variation in catchability. In the case of hook‐and‐line gear, which requires the fish to strike a lure, our predictions may not be representative in locations where non‐foraging‐related behaviors (e.g., migration and predator avoidance) occur. Similarly, catch data provide a discrete estimate of abundance but no information on individual movement. Chinook Salmon appear to transition from relatively small individual spatial distributions and slow migration speeds while occupying foraging grounds (<50 km; traveling 1–10 km/day) to rapid movements (up to 50 km/day) when undergoing directed migrations (Courtney et al. 2021). Preliminary results from tagging data indicate that some individuals were consistently detected within the study area over multiple weeks or months, but additional analyses are required to understand seasonal and stock‐specific patterns of habitat use (Freshwater, unpublished data).

Finally, due to logistical constraints we were unable to provide inference on all Chinook Salmon populations migrating through southwestern Vancouver Island. We conducted relatively little inshore sampling because the program was focused on resident killer whale critical habitat extending offshore (DFO 2017). As a result, we rarely encountered WCVI populations—consistent with previous evidence that they are strongly nearshore oriented (Beacham et al. 2008; DFO 2012)—and our model predicted low abundance relative to recent returns (CTC 2022). Similarly, we were unable to develop distribution models for less abundant stocks, such as Fraser River spring‐ and summer‐run yearling populations, due to insufficient sample sizes. Both issues could be addressed through more intensive sampling of specific locations or time periods.

Implications

Knowledge of fine‐scale stock‐specific distributions is particularly relevant for fisheries management decisions that are intended to minimize harvest impacts on less productive stocks. In many cases, management interventions that are meant to protect nontarget populations in Pacific salmon fisheries can vary across short time horizons (days or weeks) and small spatial domains (<10 km²; DFO 2023). Our results are broadly consistent with management interventions that are used to protect WCVI Chinook Salmon by restricting commercial harvest in nearshore areas (Beacham et al. 2008; DFO 2012) as well as time–area closures that are used to minimize impacts on Fraser River Chinook Salmon stocks of concern (Dobson et al. 2020). Spatiotemporal models, such as the one presented here, can serve as additional tools as management agencies refine interventions to minimize impacts on nontarget stocks while preserving sustainable harvest opportunities.

Stock‐specific distributions also play a key role in identifying environmental drivers of productivity and predicting how populations will respond to changing environmental conditions. Although Chinook Salmon populations do exhibit coherent responses to regional environmental drivers (Kilduff et al. 2015), there is growing evidence that population trajectories may covary with life history strategies and marine distributions (Freshwater et al. 2022; Atlas et al. 2023). More closely linking physical processes to growth and survival is necessary to identify the mechanisms driving variability in Pacific salmon population dynamics (Crozier et al. 2019; Wells et al. 2020).

Agencies are increasingly tasked with implementing ecosystem‐based fisheries management (EBFM), which incorporates interactions among species into management frameworks. For example, EBFM may prioritize reducing fisheries mortality to increase the prey biomass available to predators (Cury et al. 2011) or modifying dynamic spatiotemporal closures to minimize incidental harm to nontarget species (Santora et al. 2020). In the case of Chinook Salmon, one dimension of EBFM is ensuring that adequate prey are available to sustain at‐risk populations of resident killer whales, which rely on Chinook Salmon (Ford et al. 1998). Chinook Salmon that are large and belong to populations with higher mean lipid content are presumably the most valuable food resource for resident killer whales (O'Neill et al. 2014). Thus, the quality of the prey field for resident killer whales will vary spatially and temporally with stock‐ and size‐class‐specific patterns of abundance. Lipid‐rich stocks were relatively less abundant than lipid‐poor stocks, which may result in tensions between prey availability and prey quality for Chinook Salmon predators; however, we sampled only a portion of the critical habitat. We also note that the foraging success of resident killer whales is impacted by factors beyond prey quantity or quality, such as bathymetric features and anthropogenic disturbance (Holt et al. 2021; Thornton et al. 2022), which are the focus of ongoing work.

CONCLUSIONS

We used hierarchical models to quantify patterns of variability in Chinook Salmon size, condition, abundance, and distribution in a region with substantial life history diversity. We found evidence of a physiological transition from somatic growth to lipid storage among immature Pacific salmon that were predicted to remain at sea for at least one additional year. We also identified differences among Chinook Salmon stocks in lipid content, even after accounting for body size and sampling date. Many of the most lipid‐rich stocks were relatively less abundant in the study area than lipid‐poor stocks.

By sampling throughout the domain and accounting for variable effort, we were able to corroborate anecdotal evidence that Chinook Salmon abundance is strongly influenced by spatial location. Nearshore reefs and offshore areas of high bathymetric relief likely provide foraging opportunities, and ongoing work is attempting to determine how these features influence interactions among Chinook Salmon, their prey, and their predators. Chinook Salmon distributions varied among size‐classes and stocks but not with rearing history. Size‐specific spatial distributions provided evidence of fine‐scale ontogenetic changes in habitat use as juvenile Chinook Salmon begin to transition away from inshore habitats.

ACKNOWLEDGMENTS

We thank Willem Offerein, Simon Offerein, Erika Nielsen, Katie Innes, Kelsey Flynn, Dylan Glaser, and Colin Bailey for assistance with fieldwork. We also thank David Huff and Joe Smith for providing aging data. We are grateful to Dylan Glaser and Michelle Charbonneau for providing constructive feedback on a previous version of the manuscript. Reviews from Joe Langan, Dick Beamish, and two anonymous reviewers considerably improved the manuscript.

CONFLICT OF INTEREST STATEMENT

The authors declare no conflicts of interest.

DATA AVAILABILITY STATEMENT

The data and code underpinning these analyses are available from https://github.com/Pacific‐salmon‐assess/chin_catch, and an archived version is available at https://doi.org/10.5281/zenodo.13844414.

ETHICS STATEMENT

All scientific sampling was conducted in accordance with an S34 scientific permit issued by the DFO's Regional Director of Science.

REFERENCES