The fishing and natural mortality of large, piscivorous Bull Trout and Rainbow Trout in Kootenay Lake, British Columbia (2008–2013)

Raw data

The acoustic receiver VRL download files were processed using Vemco^® VUE software to calculate the number of hourly detections of each transmitter by each acoustic receiver. The reported recaptures were recorded by MFLNRO staff and the rewards administered by the Freshwater Fish Society of British Columbia (FFSBC). Based on lake geomorphology and the locations of the receivers (Fig. 1) the study area was divided into 30 sections (Fig.  2).

Hourly receiver data

The hourly detection, receiver deployment, fish capture and recapture data sets and sectional shapefiles were manipulated using R version 3.3.2 (R Core Team, 2015). During data manipulation, two or less detections of a fish at a receiver in an hour or any detections after the expected tag life were discarded. The resultant clean and tidy (Wickham, 2014) data sets were bundled together in an R data package called klexdatr (Article S1).

Daily sectional data

The hourly receiver detection data were then aggregated into daily sectional detection data using the lexr R package (Article S1). During the aggregation process, acoustically tagged individuals that 30 days after release were no longer detected or were only detected at a single section were classified as post-release mortalities and were excluded (Hightower, Jackson & Pollock, 2001). Detections of the same fish in more than one section in the same day (7% of the total) were resolved in favour of the section with the most detections. Ties were broken by choosing the section with the least receivers and if still tied the smaller section (sectional areas were unique). The total percent receiver coverage of each section in each interval was calculated (Fig. 3) assuming a detectional radius of 500 m (Gutowsky et al., 2013; Gutowsky et al., 2016; Huveneers et al., 2015).

Seasonal sectional data

The lexr package was then used to group the daily data into seasonal periods for analysis. Following Gutowsky et al. (2013), the seasons were winter (January–March), spring (April–June), summer (July–September) and autumn (October–December). The final data included logical matrices indicating for each fish in each seasonal period whether it was monitored (active acoustic transmitter) and/or reported (recaptured) and whether it moved (detected in two or more sections) and/or spawned.

Spawning in Bull Trout was identified by a hiatus in detections for at least four weeks during August and September when the fish was deemed to have moved out of the main lake (O’Brien, 2001; Andrusak & Andrusak, 2014). For Rainbow Trout spawning was identified by either a detection in April or May at the outflow of Trout Lake (section S02) or a hiatus in detections for three or more weeks during April and May (Irvine, 1978; Irvine, Baxter & Thorley, 2013) with the bracketing detections occuring in the top 9% of Kootenay Lake (sections S07–S09).

CJS model

The probabilities of being recaught (and reported) by an angler versus dying of other causes were estimated from the seasonal data using an individual state-space (Royle, 2008; Kéry & Schaub, 2011) Cormack–Jolly–Seber (CJS) (Cormack, 1964; Jolly, 1965; Seber, 1965) survival model. In the individual state-space formulation of the CJS model, the ith individual is alive when it is initially tagged at time period f_i, e.g., (1)${\displaystyle z_{i,f_i}=1.}$ Its latent state (alive versus dead) at subsequent periods is modelled as the outcome of a series of Bernoulli trials (2)${\displaystyle z_{i,t+1}\sim Bernoulli\left(z_{i,t}\cdot {\varphi }_{i,t}\right)}$ where ϕ_i,t is the predicted survival for the ith fish in the tth time period. An individual that is alive at period t also has a probability ρ_i,t of being recaptured, i.e., (3)${\displaystyle y_{i,t}\sim Bernoulli\left(z_{i,t}\cdot {\rho }_{i,t}\right).}$

Base model

In the base model of the current study, ρ_i,t is the probability of being recaptured (and reported) by an angler. To reduce the number of necessary assumptions, reported recaptures are excluded from the analysis for all subsequent time periods, i.e., subsequent detections or recaptures of any released individuals were ignored. In addition, a living individual with an active transmitter (T_i,t = 1) also has a probability (δ_i,t) of being detected moving (m_i,t) between sections by the receiver array (4)${\displaystyle m_{i,t}\sim Bernoulli\left(z_{i,t}\cdot T_{i,t}\cdot {\delta }_{i,t}\right).}$ Finally, in the spawning season (S_i,t = 1) a fish has a probability κ_i,t of entering the state of spawning (x_i,t = 1) for the period (5)${\displaystyle x_{i,t}\sim Bernoulli\left(z_{i,t}\cdot S_{i,t}\cdot {\kappa }_{i,t}\right).}$ In the base model, the terms ϕ_i,t, ρ_i,t, δ_i,t and κ_i,t, which represent probabilities, are specified by four parameters, i.e., (6)${\displaystyle logit\left({\varphi }_{i,t}\right)={\beta }_{\varphi 0},}$ (7)${\displaystyle logit\left({\rho }_{i,t}\right)={\beta }_{\rho 0},}$ (8)${\displaystyle logit\left({\delta }_{i,t}\right)={\beta }_{\delta 0}}$ and (9)${\displaystyle logit\left({\kappa }_{i,t}\right)={\beta }_{\kappa 0}}$ where (10)${\displaystyle logit\left(p\right)=log\left(\frac{p}{1-p}\right).}$

Full model

The full model extends the base model through the inclusion of seven additional parameters (β_κL, β_ϕx, β_δS, β_κY, β_ρY, β_ϕY and β_δY). The first parameter (β_κL) allows the log odds probability of spawning to vary with the calculated fork length (L_i,t) (11)${\displaystyle logit\left({\kappa }_{i,t}\right)={\beta }_{\kappa 0}+{\beta }_{\kappa L}\cdot L_{i,t}}$ while the second (β_ϕx) allows the log odds survival to vary with spawning (12)${\displaystyle logit\left({\varphi }_{i,t}\right)={\beta }_{\varphi 0}+{\beta }_{\varphi x}\cdot x_{i,t}.}$ The fork lengths were calculated based on the measured length at capture L_i,fi plus the length increment expected under a Von Bertalanffy Growth Curve (Walters & Martell, 2004) (13)${\displaystyle L_{i,t}=L_{i,f_i}+\left(L_{\infty }-L_{i,f_i}\right)\left(1-e^{-0.25\cdot k\left(t-f_i\right)}\right)}$ where the 0.25 in the exponent adjusts for the fact that there are four time periods (seasons) per year. The calculation assumes a L_∞ of 1,000 mm with a k of 0.13 for Bull Trout and a k of 0.19 (Andrusak & Thorley, 2014) for Rainbow Trout.

The third additional parameter (β_δS) allows the log odds probability of being detected moving to vary with the spawning season (14)${\displaystyle logit\left({\delta }_{i,t}\right)={\beta }_{\delta 0}+{\beta }_{\delta S}\cdot S_{i,t}}$ while the last four parameters (β_κY, β_ρY, β_ϕY and β_δY) allow the probability of spawning, recapture, survival, and detection moving between sections to vary with the standardised year (Y_i,t), i.e, (15)${\displaystyle logit\left({\kappa }_{i,t}\right)={\beta }_{\kappa 0}+{\beta }_{\kappa Y}\cdot Y_{i,t}.}$ The parameters in the full model are defined in Table 1.

Parameter estimation

The parameters were estimated using Bayesian methods (Ntzoufras, 2009; Kéry, 2010; Kéry & Schaub, 2011). The prior distribution for each parameter was a normal distribution with a mean of 0 and a standard deviation of 3, i.e., (16)${\displaystyle {\beta }_{\kappa 0}\sim Normal\left(0,3\right).}$ The posterior distributions of the parameters were estimated using a Monte Carlo Markov Chain (MCMC) algorithm. To avoid non-convergence of the MCMC process, five chains were run, starting at randomly selected initial values. Each chain was run for at least 10⁵ iterations with the first half of the chain discarded for burn-in followed by further thinning to leave a grand total of 10,000 samples. Convergence was confirmed by ensuring that $\hat{R}\le 1.05$ for each of the parameters in the model (Brooks & Gelman, 1998; Kéry & Schaub, 2011).

The vagueness of the priors was assessed by a sensitivity analysis (Kéry & Schaub, 2011). More specifically, the model was refitted with the prior distribution for each parameter a normal distribution with a mean of 0 and a standard deviation of 6. The sensitivity of the posteriors to the change in priors was assessed using $\hat{R}$. Although developed to quantify the convergence between chains, $\hat{R}$ can also be used as a metric of the convergence between models by combining all the samples for each model. The vagueness of the priors was confirmed by ensuring that the between model $\hat{R}$ values were all ≤1.2. A higher minimum $\hat{R}$ value was accepted for the between model (than the within model convergence) because the purpose of the test was to confirm that the posteriors were not unduly altered by the change in the priors, i.e., except in the case of simple models with lots of data it is not realistic to expect that a substantial change in the priors will have no effect at all on any of the posteriors. The chosen $\hat{R}$ value of 1.2 is within the accepted range of values used to confirm the convergence of chains within a single model (Kéry & Schaub, 2011).

The reported point estimates are the mean and the 95% credible intervals (CRIs) are the 2.5 and 97.5% quantiles (Gelman, 2014). Unless stated otherwise all presented and plotted estimates are for a representative individual which in the current study was judged to be a 650 mm non-spawner in the autumn of 2011.

Model selection

Model selection was achieved by using indicator variable selection (Hooten & Hobbs, 2015) to estimate probabilities for the inclusion of the seven additional parameters from the full model in a final model. More specifically, “stochastic search variable selection” (George & McCulloch, 1993; Hooten & Hobbs, 2015) was used to assign an indicator variable to each parameter such that (17)${\displaystyle {\gamma }_{\kappa L}\sim Bernoulli\left(1∕2\right)}$ (18)${\displaystyle {\beta }_{\kappa L}\sim {\gamma }_{\kappa L}\cdot Normal\left(0,3\right)+\left(1-{\gamma }_{\kappa L}\right)Normal\left(0,0.03\right).}$ In other words, if the indicator is 1 then the parameter is drawn from the standard vague prior, otherwise the parameter is drawn from a prior that is so constrained that its value is effectively zero. The values of γ_ρY and γ_ϕY indicate the support for a change in F and M, respectively, over the course of the study.

Software

The analyses were performed using R version 3.3.2 (R Core Team, 2015), JAGS 4.2.0 (Plummer, 2003) and the klexr R package (Article S1), which was developed specifically for this paper.

Model assumptions

The validity of a statistical model’s estimates depends on the extent to which its assumptions are violated (Kéry & Schaub, 2011). Like the standard CJS model, the modified form used in the current analysis assumes that

• 1.

  The tagged individuals are a random sample from the population.

• 2.

  There are no effects of tagging.

• 3.

  There is no loss of anchor or acoustic tags.

• 4.

  All individuals are correctly identified.

• 5.

  There is no emigration out of the main lake.

• 6.

  There are no unmodelled individual differences in the probability of recapture in each time interval.

• 7.

  There are no unmodelled individual differences in the probability of survival in each time interval.

• 8.

  The fate of each individual is independent of the fate of any other individual.

• 9.

  The sampling periods are instantaneous and all individuals are immediately released.

In addition, the modified form used in the current study also assumes that

• There are no unmodelled individual differences in the probability of a living individual with an active acoustic transmitter being detected moving among sections in each time period.

• Anglers report all anchor tags.

• Spawning events are correctly identified.

• Growth increments are correctly calculated.

The extent to which each assumption may have been violated was assessed based on the literature and, when available, auxilary data. In particular, the random nature of the population sample was assessed by plotting the spatial distribution of captures, daily detections and recaptures. The extent of any outmigration was evaluated from the proportion of acoustically tagged individuals last detected at the top of the North Arm (S07), bottom of the South Arm (S32) and at the upper end of the West Arm (S20). Only individuals that were last detected at least 120 days before the end of their transmitter life and were not subsequently recaptured were considered to be possible outmigrants. Finally, anchor tag loss was estimated from the proportion of double-tagged individuals that had lost an anchor tag at recapture using a simple Bayesian zero-truncated binomial tag-loss model (Fabrizio et al., 1999).

Captures

Between 2008 and 2013, 88 large (≥500 mm) Bull Trout and 149 large (≥500 mm) Rainbow Trout were marked with at least one $100 reward anchor tag (Tables 2 and 3). A total of 69 of the Bull Trout and 114 of the Rainbow Trout were also acoustically tagged. Most Bull Trout were caught in the central portion of the main lake or the top of the North Arm (Fig. S1) while most Rainbow Trout were caught in the central portion of the main lake (Fig.  S2).

Post-release mortalities

Of the acoustically tagged individuals five (7%) of the Bull Trout and 28 (25%) of the Rainbow Trout were no longer detected moving between sections 30 days after release. They were deemed to be post-release mortalities and were excluded from the data. The daily detections are plotted by section and date in Fig. 4 for the remaining 83 Bull Trout and 121 Rainbow Trout.

Detections

The spatial distribution of the daily detections indicates that fish use of the main lake was relatively evenly distributed for both Bull Trout (Fig. S3) and Rainbow Trout (Fig. S4). A grand total of 27 acoustically tagged Bull Trout, which were not subsequently recaught, were last detected moving in the main lake 120 days before the end of their transmitter life (Table 4). Of the 27 Bull Trout, seven (26%) were last detected moving at section S07, one (4%) at S20 and zero (0%) at S32. For Rainbow Trout, the grand total was 48 individuals (Table 5), of which seven (15%) were last detected at section S07, four (8%) at S20 and three (6%) at S32.

Recaptures

A total of 26 of the 83 Bull Trout and 24 of the 121 Rainbow Trout were reported to have been recaught by an angler (Fig. 4). In the main lake, most of the Bull Trout were recaught in the central portion or the top of the North Arm (Fig. S5). In contrast the Rainbow Trout recaptures were relatively evenly distributed throughout the main lake with the exception of the most northerly sections of the North Arm and the most southerly sections of the South Arm (Fig. S6).

Of the 25 Bull Trout recaptures that had been initially tagged with a $100 and a $10 tag, 18 had both anchor tags, three had just the $100 tag and four had just the $10 tag. For the 22 equivalent Rainbow Trout recaptures, 20 had both anchor tags, zero had just the $100 tag and two had just the $10 tag. Analysis of the number of individuals with one versus two anchor tags using the simple Bayesian tag-loss model estimates the probability of a single-tagged individual losing its anchor tag between release and recapture to be 0.18 (95% CRI [0.08–0.32]) for Bull Trout and 0.07 (95% CRI [0.01–0.17]) for Rainbow Trout. The corresponding probability of a double-tagged individual losing both its anchor tags (assuming the fates of both tags are independent) is 0.04 (95% CRI [0.01–0.10]) and 0.006 (95% CRI [0.000–0.027]), for Bull Trout and Rainbow Trout, respectively.

Parameter estimation

Model convergence was confirmed by an $\hat{R}\le 1.02$ for Bull Trout and ≤ 1.02 for Rainbow Trout. The vagueness of the priors was confirmed by a between model $\hat{R}\le 1.05$ for Bull Trout and ≤ 1.12 for Rainbow Trout.

Recapture

The final CJS survival model estimated that the annual probability of being recaptured (and reported) by an angler (ρ) was 0.17 (95% CRI [0.11–0.23]) for Bull Trout (Table 6) and 0.14 (95% CRI [0.09–0.19]) for Rainbow Trout (Table 7). The indicator variable probabilities for γ_ρY were 0.09 and 0.09 for each species, respectively, revealing little to no support for a change in the probability of recapture with year.

Survival

The survival probability for Bull Trout was largely unaffected by spawning (γ_ϕκ = 0.29) but changed over the course of the study (γ_ϕY = 0.95). More specifically, the annual Bull Trout survival probability was estimated to have declined from 0.91 (95% CRI [0.76–0.97]) in 2009 to just 0.46 (95% CRI [0.24–0.76]) in 2013 (Fig. 5).

In contrast, Rainbow trout survival was strongly affected by spawning (γ_ϕκ = 1.00) and may have changed over the course of the study (γ_ϕY = 0.76). If present, the change in survival over time was a decline (Fig. 5). The CJS model estimated that spawning reduced the annual Rainbow Trout survival probability from 0.77 (95% CRI [0.68–0.85]) to 0.41 (95% CRI [0.30–0.53]) (Fig. 6).

Spawning

For both species, the probability of spawning was clearly dependent on the fork length as γ_κL = 0.97 for Bull Trout and 1.00 for Rainbow Trout. Figure 7 indicates that the probability of spawning increases from 0.40 (95% CRI [0.16– 0.75]) for a 500 mm Bull Trout to 0.94 (95% CRI [0.78–0.99]) for a 800 mm Bull Trout. The equivalent values for Rainbow Trout are 0.00 (95% CRI [0.00–0.00]) and 0.90 (95% CRI [0.76–0.98]), respectively.

There was little support (γ_κY = 0.23) for a change in the probability of spawning for Bull Trout from 2009 to 2013. However, in contrast, there was strong support (γ_κY = 1.00) for a decline in the probability of spawning for Rainbow Trout over the course of the study  (Fig. 8).

Movement

The last key parameter in the CJS model—the probability of being detected moving among sections while alive with an active transmitter—varied by spawning season for Bull Trout (γ_δS = 1.00) but not Rainbow Trout (γ_δS = 0.20). In the case of Bull Trout, the probability was 0.49 (95% CRI [0.40–0.58]) for being detected moving during the summer spawning season compared to 0.91 (95% CRI [0.88–0.94]) during the other seasons. The probability of being detected moving was 0.95 (95% CRI [0.93–0.97]) for Rainbow Trout. Over the course of the study, the probability of being detected moving among sections may have (γ_δY = 0.64) declined (β_δY =  − 0.37) for Bull Trout but not Rainbow Trout (γ_δY = 0.10).

Sample

The model assumes that the tagged individuals are a random sample from the population. The distribution of captures shows that fish were not sampled from all areas equally (Figs. S1–S2). However, the detections of acoustically tagged fish (Figs. S3–S4) indicated a high degree of mixing between areas. The possibility that fish may have differed in their susceptibility to (re)capture by angling is discussed below.

Tagging

The second assumption is that there are no effects of tagging. To reduce the chances of post-release mortality inflating M, fish that were no longer detected moving between sections 30 days after release were excluded from the analysis (Hightower, Jackson & Pollock, 2001). The fact that 7% of the Bull Trout versus 25% of Rainbow were judged to be post-release mortalities suggests that in the short-term at least Bull Trout are less affected by capture, handling and surgery. To ensure any longer-term effects of tagging were minimal, the transmitter to body weight ratio was ≤0.9% (Brown et al., 1999). Nonetheless, it is possible that tagging could have increased natural or fishing mortality or reduced the probability of spawning due to infection or increased energetic expenditures.

Anchor tag and transmitter loss

The CJS model assumed that there was no anchor tag or transmitter loss (Brenden, Jones & Ebener, 2010). To minimize anchor tag loss, all fish were double-tagged. Analysis of double versus single-tagged recaptures indicates that the probability of losing both anchor tags was 4% for Bull Trout and 0.6% for Rainbow Trout. The higher loss rate for Bull Trout may be due to the spawning migration they undertake up steep turbulent tributaries (Andrusak & Andrusak, 2014). The exclusion of fish that had ceased moving within 30 days of initial capture reduced the chances that transmitter loss would affect the mortality estimates (Hightower, Jackson & Pollock, 2001).

Identification

Anglers were required to return all tags. Consequently it is unlikely that fish were misidentified.

Emigration

The model assumed that there was no movement outside of the main lake other than temporary seasonal spawning migrations by Bull Trout in the summer and Rainbow Trout in the spring. Bull Trout spawn in tributaries to Kootenay Lake (Andrusak & Andrusak, 2014) as well as tributaries to Trout Lake and Duncan Reservoir (O’Brien, 2001). The seven Bull Trout that were last detected at the top of the North Arm (sections S07 and S09) in the spring (Table 4) may have been early-leaving spawners headed to Trout Lake or Duncan Reservoir (O’Brien, 2001) that failed to return. The implications of the misidentification of spawning events is discussed below. The 18 Rainbow Trout that were last detected in the spring at the top of the North Arm or at section S02 at the outflow of Trout Lake (Table 5) are accounted for as spawning mortalities by the model. With the exception of the seven spring Bull Trout migrants at the top of the North Arm there is no indication of any significant movement outside the main lake other than temporary seasonal spawning migrations that are accounted for by the model.

Recapture

A key assumption of all mark-recapture studies is that there are no unmodelled individual differences in the probability of (re)capture in each time interval (Biro, 2013). As is often the case, the reliance on a single capture method means it is not possible to test this assumption (Biro, 2013). Depending on whether any individual differences were fixed or learned the estimated potential fishing mortality would be an over or under-estimate, respectively (Askey et al., 2006; Biro, 2013).

Survival

The full model allowed survival to vary with spawning status and year. Simulations by Matechou et al. (2013) indicate that the survival estimates are likely to be robust to any remaining time-dependent variation.

Independence

Individuals exhibited no obvious shoaling or coordination of movements other than spawning migrations. Consequently the fate of each individual was likely independent of the fate of any other individual.

Instantaneous

While individuals were almost immediately released, the sampling periods for captures, recaptures and detections were effectively continuous. However, because the seasonal survival was high (≥82% for Bull Trout and ≥80% for Rainbow Trout) and because the purpose was to estimate overall annual survival across four seasonal time periods, substantial biases are not expected (Barbour, Ponciano & Lorenzen, 2013).

Movement

The CJS model assumes that mortalities exhibit no movement. Consequently if the model was to overestimate the probability of a living individual with an active transmitter being detected moving between sections then M would also be overestimated. However, this is unlikely for three reasons. Firstly, the model allowed the probability of movement to vary with spawning season and year. Secondly, the seasonal probability of detecting an individual moving between sections was relatively high (91% for Bull Trout and 95% for Rainbow Trout). Finally, the state-space component of the model means that a temporarily inactive individual or one that is subsequently recaptured will not be misidentified as a natural mortality. Consequently any bias in M is expected to be small.

Reporting

To maximize the tag reporting rate, which was assumed to be 100%, at least one of the tags was worth $100 tag (Bacheler et al., 2009). Previous studies on Common Snook (Centropomus undecimalis) in Florida in the late 1990s (Taylor et al., 2006) and multiple freshwater fish species in Idaho from 2006 to 2009 (Meyer et al., 2012) suggest that $100 is sufficient to achieve reporting rates of approximately 99%.

Spawning

The model assumed that spawning events are correctly identified. Large Rainbow Trout from Kootenay Lake almost exclusively spawn at the head of the Lardeau River (Irvine, 1978; Irvine, Baxter & Thorley, 2013). As this spawning area was acoustically monitored duing the spring it is likely that most of the Rainbow Trout spawning events were correctly identified.

For Bull Trout the situation is more complicated. Large Bull Trout from Kootenay Lake spawn in multiple tributaries of Kootenay Lake, Trout Lake and Duncan Reservoir (O’Brien, 2001; Andrusak & Andrusak, 2014). Consequently it was not possible to monitor spawning locations and spawning was inferred by a hiatus in detections for four or more weeks during August and September. As a result only fish which successfully returned to Kootenay Lake in the fall post-spawning were identified as spawners. Bull Trout that succumbed to spawning-related injuries, stress or predation prior to returning to Kootenay Lake were identified as inlake natural mortalities As a result M is likely biased high for non-spawning Bull Trout and biased low for spawners. This bias may explain why an effect of spawning on mortality was not detected for Bull Trout.

Growth

The model assumed that growth was described by a seasonal Von Bertalanffy Growth Curve with a L_∞ of 1,000 mm and a k of 0.13 for Bull Trout and a k of 0.19 for Rainbow Trout. The curve for Bull Trout was partly based on large Bull Trout in Adams Lake, British Columbia (Bison, O’Brien & Martell, 2003). The growth curve for Rainbow Trout was based on lake specific data (Andrusak & Thorley, 2014) and is likely a good approximation.

Mortality estimates

The implications of any violations of the assumptions are negligible for the mortality estimates with two exceptions. The reliance on a single capture method means that F could be an over or under-estimate for either or both species. Similarly, the difficulty identifying Bull Trout spawners means that spawning Bull Trout may experience a higher M than non-spawners.

Post-release mortality

The current study indicates that the large Bull Trout in Kootenay Lake are relatively robust to angling and surgical implantation of an acoustic transmitter with a 30 day post-release mortality of 7%. For comparison, Post et al. (2003) in their age-structured model of an adfluvial bull trout population presumed post-release mortality was 10% but considered values of 2% and 25%. The current study estimated that large Rainbow Trout experience a post-release mortality of 25%. For comparison, meta-analyses indicate that the mean post-release mortality in salmonid recreational fisheries is 16–18% although the distribution is positively skewed with most estimates ≤ 11% (Bartholomew & Bohnsack, 2005; Huhn & Arlinghaus, 2011). It is worth noting that the current estimates include surgical implantation of a acoustic transmitter and may therefore be inflated relative to the post-release mortality of fish caught-and-released by recreational anglers. It is however also possible that the care taken playing, landing and recovering the fish resulted in a comparable or even lower post-release mortality rate than in the recreational fishery.

Catchability

Remarkably few estimates of catchability have been published for large piscivorous trout. A notable exception is Post et al. (2003) who based on creel data from Lower Kananaskis Lake and experimental fishery data from Marie Lake and Quirk Creek considered a q of 0.07 fish caught.vulnerable fish⁻¹ angler-hr⁻¹ ha⁻¹ yr⁻¹ to be representative for Bull Trout. They assumed fish were invulnerable to capture at ≤200 mm and completely vulnerable by 400 mm. Their nominal value of 0.07 is more than double the current Bull Trout estimate of 0.034 angler-hr⁻¹ ha⁻¹ yr⁻¹. However, Post et al. (2003) did consider the possibility that q ranged between 0.035 and 0.14. The reasons for the relatively low catchability of Bull Trout in Kootenay Lake are unknown but could include less efficient angling methods (anglers primarily target Rainbow Trout) or more uniform fish distributions.

Catchability is typically strongly negatively related to the population density—a phenomenon known as hyperstability (Ward et al., 2013). Based on a meta-analysis of Lake Trout (Salvelinus namaycush) in 12 lakes in Ontario, Canada, Shuter et al. (1998) modelled the relationship between catchability and density (D) using a modified form of the equation (20)${\displaystyle q=exp\left(-\frac{0.14\cdot E}{1+0.35\cdot D}\right)\cdot E^{-1}.}$ In Kootenay Lake in 2011, Bull Trout occured at a density of 0.75 fish.ha⁻¹ and experienced an effort of 4.88 angler-hr ha⁻¹ yr⁻¹. If large Bull Trout have the same catchability as Lake Trout, Eq. (20) predicts a catchability of 0.12 angler-hr⁻¹ ha⁻¹ yr⁻¹. Although less then the upper range of 0.14 considered by Post et al. (2003), 0.12 is still much higher than the estimated value of 0.034 angler-hr⁻¹ ha⁻¹ yr⁻¹.

Ward et al. (2013) concluded that hyperstability for stocked Rainbow Trout in 18 British Columbian lake was caused by differences in angler skill with less experienced anglers tending to target lakes with higher fish densities. The lakes were all small waterbodies (<45 ha) with high fish densities (>15 fish ha⁻¹) and no natural recruitment. As the predictions were for small insectivorous trout at high densities they could not be reliably extrapolated to large piscivorous Rainbow Trout. Nonetheless, Ward et al.’s (2013) findings suggest that Catch-Per-Unit-Effort may need to be scaled by angler effort when attempting to model Bull Trout and Rainbow Trout abundance in Kootenay Lake.

Inter-annual variability

There was little to no support for a change in the fishing mortality for either Bull Trout or Rainbow Trout over the course of the study. This finding is important because, as discussed below, it suggests that angler effort is not the primary driver of any short-term fluctuations.

Inter-annual variability

The CJS model strongly supported an increase in the natural mortality of Bull Trout over the course of the study. Further support for an increase in Bull Trout natural mortality is also provided by a preliminary redd count analysis for Kootenay Lake (Andrusak & Andrusak, 2014) that suggests a lake-wide decrease in the number of spawners from a peak in 2009 to a low in 2013 (the last year of the redd count study). As the CJS model provided little support for a change in the probability of Bull Trout spawning the decline in the number of Bull Trout spawners likely reflects a decline in Bull Trout abundance. The finding that M has changed over the course of the study is important because it is consistent with Kokanee abundance or another biological factor as a driver of short-term Bull Trout population dynamics.

The CJS model only provided moderate support for an increase in the natural mortality of Rainbow Trout. More specifically the CJS model estimated that the annual instantaneous natural mortality of a non-spawning Rainbow Trout increased from 0.13 in 2008 to 0.47 in 2013. In contrast, Rainbow Trout escapement, as estimated by the peak count with a multiplier of 3.08 (Hagen et al., 2010), increased from 1,583 in 2008 to a high of 3,289 in 2012 before falling slightly in the spring of the final year of the current study (Fig. 10). While the escapement was increasing to record levels the probability of spawning was declining dramatically. This finding is important because it suggests that the actual increase in the number of large Rainbow Trout was even more dramatic than indicated by the escapement.

Spawning

For Bull Trout the CJS model did not identify an increase in the mortality rate associated with spawning. However, as discussed above spawners which died prior to returning to Kootenay Lake would have been misidentified as inlake mortalities. It is therefore likely that the natural mortality of spawners and non-spawners are under and over-estimated, respectively.

For Rainbow Trout, spawning increased the annual interval natural mortality rate from 23% to 59%. The high spawning mortality is consistent with the limited spawning area and high levels of antagonistic interactions at the outflow of Trout Lake (Hartman, 1969; Hartman & Galbraith, 1970). The high spawning mortality may also explain why the probability of spawning is low for Kootenay Lake Rainbow Trout smaller than approximately 650 mm in length, i.e., the cost of spawning selects for delayed maturation.