A species-specific catchability model to partition hydroacoustic total salmon counts

ABSTRACT

Objective

Fisheries sonar systems can yield accurate and precise total fish counts but cannot unambiguously differentiate fish species with similar or overlapping size and target-strength distributions. To obtain species-specific abundances for the management of fisheries of individual species, the total fish count produced by an acoustic system must be partitioned into abundances of individual species by test-fishing-based species composition methods. However, species composition estimates based on traditional models for catch-per-unit-effort (CPUE) data can be biased due to differences in swimming speed, body size, and spatial distributions between fish. The objective of this study is to establish a generalized CPUE model linking the catch data with the underlying compositions of fish species that display unequal catchabilities and saturation levels to fishing nets.

Methods

We propose a relationship between CPUE and fish abundance through a multispecies disc function model that is linear at low abundances but predicts gear saturation, with initial slope (catchability) that varies with fish species. Fitting this model to historical CPUE data yields maximum-likelihood estimates of species-specific catchabilities and saturation factors for the model. The CPUE-based species estimates (likely biased) are treated as an input to the derived model to obtained corrected species estimates.

Results

The generalized CPUE model was applied to acoustic counts of Pacific salmon returning to the Fraser River in British Columbia from 2008–2018 seasons acquired at two acoustic fish-counting sites on the lower river. The model resulted in substantial corrections for abundance estimates for Sockeye Salmon Oncorhynchus nerka, Pink Salmon O. gorbuscha, and Chinook Salmon O. tshawytscha. Chinook Salmon showed severe gear saturation that led to downward bias in abundance estimates even after correcting for saturation effects.

Conclusions

Our study showed that for a gill-net-based test fishery operation, Pink Salmon had a lower catchability estimate than Sockeye Salmon with a relative catchability of 40−50%, whereas Chinook Salmon’s relative catchability was in the range of 300−400%. These unequal catchabilities must be taken into account when partitioning total acoustic salmon count using CPUE data to avoid the deflation of abundances of Pink Salmon and the inflation of Chinook Salmon.

INTRODUCTION

With rapid advancements in fisheries acoustics since the 1990s (Belcher et al., 2002; Maxwell & Gove, 2007; Moursund et al., 2003; Mulligan, 2000), hydroacoustic technologies have been widely adopted for enumerations of migrating salmon abundances in rivers and spawning streams (Enzenhofer et al., 2010; Holmes et al., 2006; Xie et al., 2024; Xie & Martens, 2014) to replace the traditional CPUE-based abundance-­indexing method. Although hydroacoustics can yield ­accurate and precise counting of total salmon passage, it cannot definitively identify fish species based on highly varied acoustic target-strength vs. fish-length regression relationships (Burwen et al., 2007; Love, 1977) or from directly measured fish length data acquired from imaging sonar systems if species are highly overlapped in their length distributions (Xie et al., 2013). Combinations of underwater video cameras and imaging sonar systems have yielded successful applications for counting fish and estimating species in small river tributaries or through constricted counting fences (Enzenhofer et al., 1998; Helminen & Linnansaari, 2023; Mueller et al., 2006). However, such combined technologies face challenges when applied in large and turbid rivers due to severely curtailed detection ranges. Because salmon fisheries in major production rivers are managed by species and stocks to meet multiple conservation and risk-averse fisheries objectives (Woodey, 1987), in-season management requires accurate information on not only total salmon abundance but also abundances of individual species and stocks. It is a common practice that a test-fishery operation be conducted in the vicinity of a hydroacoustic fish-counting site to obtain catch-based species composition information to partition the total acoustic fish count (Enzenhofer et al., 2010; Nelitz et al., 2018). The catch-based composition estimates have historically been derived from a linear CPUE model of the form

where C_j is catch-per-unit-effort for species j, N_j is its abundance, and q is the catchability coefficient of the fishing gear employed in the test-fishery operation (Cave & Gazey, 1994; Link & Peterman, 1998). This linear model assumes that all salmon species have the same catchability. For species with lower vulnerabilities to fishing nets, composition estimates derived from Equation 1 can negatively bias their abundance estimates while inflating estimates of species with higher catchabilities (Walters, 2015). In this article, we provide a species composition model by generalizing the CPUE model 1 to partition total acoustic fish counts and account for gear saturation when salmon densities and encounter rates with test gill nets are high.

METHODS

Study sites

The hydroacoustic fish-counting sites that were examined in this study are located near the city of Mission and at Qualark Creek of the Fraser River in British Columbia, which are 75 and 170 km upstream, respectively, from the river’s mouth at the Strait of Georgia (Figure 1). The Fraser River is one of the world’s largest production rivers of Sockeye Salmon Oncorhynchus nerka (Northcote & Larkin, 1989), and the two fish-counting sites are below major Sockeye Salmon stock spawning areas.

The river at the Qualark site is 150 m wide, with discharge ranging from 10,000 m³/s (cms) during spring freshet to 500 m³/s during the low water period in winter. Water velocities range from 1.0 m/s near shore to 3–4 m/s in the middle of the river. The high flow velocities in the middle of the river combined with the energy-conserving migration behavior of salmon result in most fish migrating through the site within 15 m of the shore (Enzenhofer et al., 2010). In contrast, the Fraser River reaches at Whonnock and Mission are much wider, with a maximum width up to 400 m and a slower midchannel flow speed of <2 m/s (Xie et al., 2005). Fish can migrate through the entire cross-section of the river.

Test fishery program at Qualark

The daily test fishery consisted of a drift sequence in the morning and another sequence in the evening. A drift sequence, composed of three consecutive approximately 5-min drifts of gill nets of different mesh sizes, began 150 m upstream of the right-bank acoustic system and moved approximately 700 m downstream. The dimensions and mesh sizes for the six daily drifts are summarized in Table 1.

Hydroacoustic fish-counting program at Qualark

Strong flows at the Qualark site forced the majority of the fish to migrate in nearshore water, which allowed for detection and counting of total salmon passage by two (one on each bank) side-viewing dual-frequency identification sonar (DIDSON; http://www.soundmetrics.com/) systems with maximum sounding ranges of 30 m. Detailed descriptions of the sonar deployments, sampling schemes, and data processing procedures are provided in Enzenhofer et al. (2010). A total of 901 d of daily salmon passages at the Qualark fish-counting site were acquired by the DIDSON system from the 2008 to 2018 management seasons.

Test fishery program at Whonnock

Catch data from the Pacific Salmon Commission’s Whonnock test-fishing program have been used for decades to partition the total salmon passage produced at the Mission hydroacoustic fish-counting site into abundances of individual salmon species. The test fishery consisted of two drifts of a seven-panel variable-mesh drift net daily during high slack tide (Table 2), with a total daily fishing time of approximately 60 min. Daily CPUE was calculated from combined catches and efforts from the two drifts to partition the acoustic salmon count for individual species. Details of this test-fishing program and its roles in fisheries management are provided in Michielsens and Martens (2022).

Hydroacoustic fish-counting program at Mission

Weak flows at the Mission site allowed fish to migrate through the entire cross-section of the river. Multiple sonar units, including two shore-based systems deployed on each bank and a vessel-based transect sounding system, were used to fully sample the entire river width. The left-bank unit was an HTI Model 243 split-beam system (http://www.htisonar.com) with a sounding range up to 60 m. The right-bank unit was an ARIS Explorer 1800 with a sounding range of 30 m (http://www.soundmetrics.com/). A Biosonics DTX ­system (https://www.biosonicsinc.com) was employed for the transect survey. Detailed descriptions of the hydroacoustic facility at the Mission site are provided in Xie et al. (2005, 2024). In total, 839 d of daily salmon passages were acquired by the hydroacoustic facility from the 2008 to 2018 management seasons for the analysis.

A generalized CPUE model

Fishing gear, such as a gill net, has a catch limit that invalidates the linear relationship of model 1 when the catch approaches the limit. Link and Peterman (1998) proposed a revised form of model 1 to account for the saturation effect. For a gillnet-based test fishery targeting a single species j, they assume that CPUE is related to the abundance through a nonlinear relationship of the form

where h is the saturation factor; its inverse is the maximum CPUE of the net. In test-fishing operations targeting different salmon species, gill nets of different mesh sizes are employed to catch Sockeye Salmon, Pink Salmon Oncorhynchus gorbuscha, and Chinook Salmon Oncorhynchus tshawytscha (Nelitz et al., 2018). Two main factors should be considered for such operations.

1. Species are not uniformly distributed in the fishing area, resulting in species-specific catchabilities (q). This results in some species being more vulnerable (with a higher catchability) to fishing nets than others. For instance, relative to Sockeye Salmon, Pink Salmon have a lower catchability due to their extremely close to shore and river-bottom migration behavior than Sockeye Salmon, but Chinook Salmon have a higher catchability (Walters, 2015; Xie et al., 1997).

2. For the same net dimension of length and width, a net targeting Chinook Salmon with 8-inch (20.3-cm) meshes has a lower catch capacity than a net targeting Sockeye Salmon with 5.25-inch (13.3-cm) meshes. As a result, saturation factor h is also species-specific.

To account for these two variabilities, we propose a multispecies disc Equation as a generalized gill-net CPUE model, of the form

where k is total number of species a test-fishery is targeting on day t. For Fraser River salmon species, k = 5. These are $j=s,p,c,co,andcm$⁠, denoting, respectively, Sockeye Salmon, Pink Salmon, Chinook Salmon, Coho Salmon Oncorhynchus kisutch, and Chum Salmon Oncorhynchus  keta.

A species composition system based on the generalized CPUE model

Taking species j as an example, its observed proportion P_o,j by the test-fishery CPUE data on day t is

whereas its true proportion on day t is, by definition, that

where $Nt,j$ and $Nt,total$ are, respectively, abundance of species $j$ and abundance of all species on day $t$ and $Nt,total$ is estimated from expanded acoustic counts. Based on the generalized model 3, abundances of the five species are related to CPUE, catchability and net saturation factor as

Substituting Equation 6 for abundances into Equation 5 and noticing that $[1+∑i=1khiqiNt,i]$ is a common factor in the denominator and nominator, we obtain that

Dividing both the nominator and denominator by the total CPUE of $Ct,total$⁠, the true proportion of species $j$ can be expressed as a function of the CPUE-based observed proportion:

Choosing Sockeye Salmon catchability $qs$ as a reference, we can measure catchabilities of all species for their relative variabilities by a metric $Rj$ defined as

The true proportion of individual species $j$ as given by Equation 8 can be expressed for all the species in a compact form of

Equation 10 is a corrected species proportion system for the uncorrected system of Equation 4. Abundance of species$j$ is estimated by multiplying $Nt,total$ to the corrected species proportion $Pj(t)$⁠. It is evident that if catchability coefficients are uniform among all species, then all $Rjs$ become unity. Under this special case, Equation 10 reduces to $Pj(t)=Po,j(t)$⁠. Therefore, Equation 10 is a generalized species proportion system which encompasses the special case of uniform catchabilities among species.

Estimation of catchability and saturation coefficients

System 10 is a function of catchability coefficients (⁠$q$⁠) and saturation coefficients (⁠$h$⁠) of the five species. These coefficients or subsets of them can be estimated by fitting the historical CPUE data to the generalized CPUE model such that the modeled likelihood for the CPUE data reaches its maximum. In this study, we propose a Poisson probability function to characterize the catch data. The Poisson distribution (Haight, 1967) quantifies the probability of the number of fish being captured over a given time interval (fishing time) by a fishing net. For species $j$⁠, we assume that the observed CPUE data follows a Poisson probability mass function of

where $xt,j$ is the observed catch in a fishing time interval on day $t$⁠, and $Ct,j(t)$ is the expected catch as define by Equation 3. For $M$ days of fishing, the probability of catches of species $j$ being $[x1,j,x2,j,…xM,j]T$ (assuming catches are uncorrelated between days) is

Therefore, for the same $M$ days, the probability of catches of all 5 species, denoted by an $M$ × 5 matrix of $X$ of

is

where the column index $j=1,2,3,4,5$ represents Sockeye, Pink, Chinook, Coho, and Chum salmon, respectively. For a given set of historic CPUE data and corresponding acoustic total salmon counts, Equation 14 is a function of species-specific catchability coefficients and the net’s saturation factors. Our objective is to search for a set of $qs$ and $hs$ that maximize the probability function Equation 14. Taking a natural log-transform of Equation 14, we obtain a log form of likelihood function $L$ of $pmf(X)$ as

Noticing that in Equation 15, the second term, being a constant for a given set of historical catch data, can be dropped out without affecting the maximum location of the likelihood function. This leads to a truncated form of $L$ as

where predicted catch $Ct,j$ is given by the disc Equation 3 with qs and hs being its core variables. The maximum of the likelihood function Equation 16 is located at the maximum-­likelihood (ML) estimates of qs and hs.

Best estimates of qs and hs

The acoustically estimated total salmon abundance and the test-fishing acquired CPUE data were used to search for ML estimates of the parameters of Equation 16. Unfortunately, not all of the 10 parameters in Equation 16 are estimable, due to low test catches and inadequate contrast in test catches for Coho and Chum salmon in particular and partial confounding of the $q$ and $h$ effects. To deal with these problems, we first omitted the Coho and Chum salmon parameters by assuming their $qs$ to equal the Sockeye Salmon q, and their hs to be 0. This approximation does not affect $L$ substantially because their catches were almost all very low. Then, we fit a sequence of increasing complex models and compared their performance using ΔAIC criteria (Burnham & Anderson, 2002). These models ranged from the one-parameter “null” model (single $q$⁠, no $h$ effects, i.e., Equation 1) to a six-parameter model with $qs$  $(Rs)$ and $hs$ estimated for the three main species of Sockeye, Pink, and Chinook salmon. Then after comparing these models with independent estimates of abundances from spawning-ground surveys, we developed a final model for which saturation $hs$ parameters were estimated from maximum catches in the historical data. Only visual comparisons with DFO spawners estimates were conducted because it is uncertain how much loss actually occurred each year from the counting sites to the spawning grounds and because spawning estimates are quite noisy (combine various visual survey and mark–recapture procedures, summed over multiple local spawning stocks). Figure 2 is a flowchart showing the procedures leading to the final selections of qs and hs for the best model: model 4.

The initial statistical comparison of simple versus complex estimation models (Table 3) showed that the test fishery catch data were definitely better predicted by models with species-specific catchabilities than with the no-correction base model (Equation 1). Also, there was substantial improvement based on ΔAIC values for including effects of gear saturation (nonzero $h$⁠) although improvement due to using species-specific hs (the six-parameter model 3) was ambiguous, with similar ΔAIC scores for the alternative $h$ model (model 2). However, assuming equal $h$ values over species led to much lower $Rc$ for Chinook Salmon than the other models, and that in turn led to very high Chinook Salmon abundance estimates relative to independent estimates from spawning survey data (acoustic reconstructions compared with DFO spawners abundance estimates are shown in Figure 2C). Our model 4 with the $h$ values estimated from maximum test catch rates came closest to predicting DFO estimates for Chinook Salmon escapement while giving similar performance to the model 3 with fitted values of the $hs$ for Sockeye and Pink salmon.

RESULTS

Catch at Qualark

A total of 4,752 drifts were conducted over the 2008–2018 management seasons using six primary nets with mesh sizes listed in Table 4.

The catch data reveals the following information about the Qualark test-fishery program:

1. Majorities of the Sockeye Salmon were caught by 4- to 5.75-inch (10.2- to 14.6-cm) mesh nets.

2. Majorities of the Pink Salmon were caught by 4.75- to 5.75-inch mesh nets.

3. Adult Chinook Salmon appeared to have similar vulnerabilities to all mesh sizes from 4 to 8 inches.

4. Jack Chinook Salmon appeared to be highly selective to mesh sizes with majorities of the catches by the 4-inch mesh net.

5. Catches of Coho and Chum salmon were negligible in comparison to catches of the three primary species of Sockeye, Pink, and Chinook salmon.

Maximum-likelihood estimates for qs and hs at Qualark

The daily Qualark salmon catches and acoustic abundance observations were fit to the multispecies disc Equation 3. Except as noted the estimates presented below are for model 4. The catch data, most of which were acquired within a daily total drift time of 30 min, were normalized to catches per 30 min. Numerical searches for the estimates were performed with Newton nonlinear algorithms (nlminb) implemented with statistical software package R (version 4.4.1; https://www.r-project.org/). The ML estimates of catchabilities (Table 5) indicate that

1. Pink Salmon showed a significantly lower catchability than Sockeye Salmon, with a relative q of 0.41. Therefore, the uncorrected model 4 would deflate Pink Salmon abundance.

2. Conversely, adult Chinook Salmon showed a significantly higher catchability with a relative q of 4.0. As a result, the uncorrected model would inflate adult Chinook Salmon abundance.

3. The test gill nets appeared to have very different catch capacities for the three species with the highest capacity for Pink Salmon and the lowest capacity for adult Chinook Salmon.

Predicted versus observed CPUE

With the ML estimates in Table 5, daily time series can be compared between the Poisson predicted CPUE $Ct,j$ by Equation 3 and the test-fishing observed CPUE $xt,j$⁠, as shown in Figure 3 for the three species.

The differences (residuals) between predicted and observed CPUEs are randomly distributed (Figure 4). Although the model underestimates exceptionally large-valued CPUEs (left-panel plots in Figure 4), the residuals appear to be normally distributed (right-panel plots).

Corrected versus uncorrected abundances at Qualark

With the Qualark ML catchability estimates, we can partition the daily acoustic total salmon through the correction model of Equation 10. Comparisons between corrected and uncorrected abundance estimates are listed in Table 6. Detailed correction effects on individual species are shown throughout the 2011 management season by the time series plots of corrected versus uncorrected daily abundances (Figure 5).

Corrections vary among the three species in response to the extent of overlapping of their comigrations determined by their differential peak-arrival timings at the site (Figure 5). Although Sockeye and Chinook salmon showed protractive migration durations, the bulk abundance of Pink Salmon did not show up until September. For Sockeye Salmon, this means that prior to the month of September, the uncorrected Sockeye Salmon abundance was only moderately increased by the correction model, which suppressed the uncorrected Chinook Salmon abundance. However, uncorrected Sockeye Salmon abundance in September was significantly downgraded by the correction model, which elevated the uncorrected Pink Salmon abundance. Because corrections are in opposite directions for Pink and Chinook salmon, their uncorrected abundances show remarkable gain and loss by the correction model (Table 7).

The repartitioning of total salmon abundance by the correction model leads to a net decrease of 513,000 fish from the uncorrected Sockeye Salmon, a net decrease of 517,000 from the uncorrected Chinook Salmon, and a net increase of 1.1 million for the uncorrected Pink Salmon. Essentially, the total increase of the 1.1 million Pink Salmon resulted mainly from the comparable decreases of Sockeye and Chinook salmon. Therefore, for the migration scenarios similar to that of September 2011, the downgrade correction of Sockeye Salmon abundance is a result of repartitioning of Pink and Chinook salmon, not just by the correction of Pink Salmon. These corrections are significant for escapement estimates of late-run Sockeye and Chinook Salmon stocks.

Catch at Whonnock

Between the 2008 and 2018 seasons, test-fishing operations recorded total catches of 37,723 Sockeye, 15,350 Pink, 7,219 adult Chinook, 458 jack Chinook, 1,128 Coho, and 3,230 Chum salmon.

Maximum likelihood estimates for qs and hs at Whonnock

Based on the same analysis approach as for the Qualark data, daily Whonnock salmon catches were fit to the multispecies disc Equation 3 using the ML parameter values listed in Table 8. The fitting results are shown in Figure 6.

In comparison to the ML parameters for Qualark (Table 5), the parameters for Whonnock (Table 8) indicate that

1. Sockeye Salmon showed a lower catchability at Whonnock than at Qualark: 0.00105 versus 0.0018;

2. Chinook Salmon at Whonnock had a lower relative catchability: 3.1 versus 4.0;

3. Pink Salmon catchability appears to be higher at Whonnock: 0.48 versus 0.41.

Items (2) and (3) imply that Qualark may require larger catchability corrections for Chinook and Pink salmon. Catchability is treated by model 4 as a site effect, without variation among years—that is, there is only one q per site and per species. The higher catchability for Sockeye Salmon at Qualark is likely due to strong currents at Qualark site forcing Sockeye Salmon to migrate within a more constricted area (nearshore) than Sockeye Salmon at Whonnock where they can distribute across the entire channel. The multispecies CPUE model (Equation 3) appears to fit the data from both sites well (cf. Figures 3 and 6) even though the flow conditions at the two sites differed significantly. Residuals between the predicted and observed CPUEs for the Whonnock data show patterns similar to those for the Qualark data (Figure 4).

Corrected versus uncorrected abundances at Mission

The corrections (Table 9) resulted in significant adjustments for Pink and Chinook Salmon estimates at Mission.

Comparison with historical escapement estimates

The only independent information data that are comparable to the acoustic estimates are historical spawning escapement estimates for Sockeye and Chinook salmon; escapement data for Pink Salmon have been unavailable since 2001 and our estimates for Coho and Chum salmon miss large, later parts of the runs for these species. The Sockeye Salmon escapement estimates are thought to be complete (all stocks monitored in some way, most often with mark–recapture and visual surveys), but the Chinook Salmon data may be missing substantial numbers of fish, particularly to tributaries upstream of Mission but below the Qualark site.

If all years of data are pooled together for between-model comparisons, corrections to the acoustic Sockeye Salmon estimates are negligible (<2%) and the patterns of the estimates are in close agreement with the patterns of DFO Sockeye Salmon escapement estimates for both the whole river system and the stocks upstream of Qualark, as shown by the plots in Figure 7A and C. However, correction effects are evident for Sockeye Salmon during Pink Salmon cycle years when the uncorrected CPUE method likely misallocates Pink Salmon to Sockeye Salmon (Figure 7B and D). The acoustic estimates for Sockeye Salmon in the lower river are generally higher than the escapement estimates near or at spawning grounds, as expected given harvest and other mortality losses between the monitoring and upstream spawning sites due to adverse migration conditions in the river (Rand et al., 2006). The reversal trend observed in 2010 at Mission (Figure 7C) was likely due to saturation of the sonar systems by high-density fish passages, resulting in underestimation of the large Sockeye Salmon run in this dominant cycle year.

Correction effects are most pronounced for Chinook Salmon passing Qualark and Mission (Figure 8). In comparisons with spawning estimates, model 4 resulted in closer matches than other models tested by the ΔAIC analysis (Figure 2). The accuracy improvement by the correction model was measured by improved agreement with the DFO spawners estimates for Sockeye and Chinook salmon.

DISCUSSION

Using the multispecies disc-Equation CPUE model of Equation 3 leads the correction model of Equation 10 for the species proportion estimates that were derived from catch data. This correction model provides more accurate species-specific abundance estimates than are derived from the raw catch data assuming equal catchabilities for salmon species comigrating through the fish-counting sites where total salmon passages is measured acoustically, provided that gear saturation effects are not too severe. Improved abundance estimates by the correction model for individual species should improve the management of salmon fisheries for the Fraser River salmon stocks, especially when comigrating stocks and species of concern limit fisheries. Salmon management in the Fraser River balances the two goals of meeting biologically derived escapement and providing harvest opportunities for abundant co-migrating stocks or species (DFO, 2023). In-season abundance estimates allow for in-season adjustment of fishing plans (i.e., enabling more opportunities when abundance is high or triggering fisheries closures when stocks return at low abundance). The current Fraser River Sockeye Salmon management system, as a result of consistently observed differences between in-river and spawning ground assessments, applies a management adjustment (MA) factor to account for estimation errors (for a comprehensive overview of Fraser Sockeye management, see Cohen, 2012). Improvements in the accuracy of the acoustic estimates will reduce reliance on the MA correction factor and can provide more harvest opportunities when abundance permits. Our findings show that the uncorrected species proportion model (Equation 4) can significantly inflate the late-run Sockeye Salmon escapement estimate especially for the month of September in Pink Salmon return years, thus confounding the observations of high loss en route. As a result of this bias and concerns about stocks of late-run Sockeye Salmon that have been designated as species/stocks of concern by Committee on the Status of Endangered Wildlife in Canada (2017), high MA values have been factored into the management plans for Fraser River Sockeye and Pink Salmon fisheries by the Fraser River Panel under the Pacific Salmon Treaty (https://www.psc.org/publications/pacific-salmon-treaty/). In the 2011 season, the uncorrected model inflated the September Sockeye Salmon escapement at Qualark by 55% or 0.5 million fish. For the five Pink Salmon cycle years between 2008 and 2018, Sockeye Salmon were inflated by 17% or 1.6 million fish.

Fraser River Chinook Salmon are also a mix of stocks of concern and abundant stocks. The current management regime is based on preseason forecasts of abundance and time-area measures to protect at risk stocks (DFO, 2023). In the 11 years of escapement data acquired at the Qualark site, we found that the uncorrected model inflated the Chinook Salmon escapement abundance by 280% or 2.9 million fish. There is a desire to move to a more adaptive approach for in-river Chinook Salmon fisheries management based on in-season abundance estimation. However, the Chinook Salmon estimates with the current methods could mislead management, resulting in openings that may cause adverse impacts on their conservation needs or foregone opportunities due to skepticism about in-season abundance estimates.

The q and h estimates for Chinook Salmon are interesting. Given the larger body sizes and assumed swimming speeds, the relative catchability $Rc$ should be greater than 1.0 due to increased encounter rate with drifting gill nets. A possible explanation for apparently high estimates of $hc$ for the two fishing sites is that only a small proportion of each net has large enough meshes to retain Chinook Salmon at high rates, which would explain the high values of $hc$ due to saturation of the net panel(s) that retain them. However, the panel-specific catch rate data for Chinook Salmon at Qualark indicates that they are caught at similar rates in all panels (Table 4), apparently contradicting this simple small-proportion hypothesis. Another possibility is that Chinook Salmon are more aggressive than the other species and exhibit aggressive behaviors toward fish that are already in the net and are exhibiting various strong body movements, in the same way that “flashers” attract salmon in sport and commercial troll fishing. Another is that our estimates of $h$ are biased upward due to positive covariance in the predicted effects of the two parameters.

Support for our contention of a strong gear saturation effect for Chinook Salmon can be seen in the fall 2023 date when DFO escapement observations demonstrated an unprecedented large (600,000+, at least three times as great as any estimate since 1979) escapement of summer-type Chinook Salmon to several spawning locations in the South Thompson tributary to the Fraser. There was no clear warning of this huge run in any of the gill-net test fisheries (TF) conducted in the lower river (DFO Albion TF site at Haney British Columbia, PSC Whonnock TF site or DFO Qualark TF site). If detected, that run could have provided substantial harvest opportunities for First Nations fishers all along the run’s migration route. When we applied the correction method to preliminary 2023 Qualark data, the method also failed to detect the large run, even when we fit the model to additional years (2019–2022) of data. In fact, the inclusion of more recent years’ data (after 2018) produced acoustic Chinook Salmon estimates deviating farther from DFO’s spawning-ground Chinook Salmon estimates. So, the current monitoring system can certainly detect and warn about very low Chinook Salmon runs (like 2012), in line with DFO’s goal to have highly precautionary harvesting under Canada’s Wild Salmon Policy (https://www.pac.dfo-mpo.gc.ca/fm-gp/salmon-saumon/wsp-pss/ip-pmo/index-eng.html). Unfortunately, the system cannot detect the opposite extreme, unusually good opportunities. Development of system-wide escapement estimates at the different test-fishing sites on the Fraser River based on genetic catch expansion similar to those developed for the Skeena River Chinook Salmon and most recently described in Winther et al. (2020), combined with the use of run reconstruction, could provide a more accurate estimate of total abundance to better describe differential catchability and gear saturation effects. Still, some other method will have to be used to detect large abundances. An alternative approach is to use imaging-sonar-acquired fish length to estimate Chinook Salmon abundances by fitting the acoustic length data to a mixture model or a Bayesian classification model (Xie et al., 2013). There are ambiguities and uncertainties with this approach to separate species that are highly overlapping in their length distributions (such as between Sockeye and Pink salmon). However, the adult Chinook Salmon length distribution is well separated from other Pacific salmon species and can be extracted with high certainty from a mixture of length data of multiple species based on preliminary analyses with newer data from an ARIS system at the Mission site.

Although the correction model rectifies the error in species proportion estimates due to unequal catchabilities among species, it cannot resolve issues caused by the effect of severe net saturation observed for Chinook Salmon in the Fraser River. Gill-net saturation effects like we estimated for Sockeye and Chinook salmon in the Fraser River have also been estimated for the Tyee gill-net test fishery in the Skeena River (Cox-Rogers & Jantz, 1993), British Columbia’s second-largest salmon producing river. We examined Chinook Salmon data for that system and also found much stronger saturation (lower maximum catch rates, steeper relationship) than for Sockeye Salmon similar to what we estimated from the TF data of Fraser stocks, as shown by the comparison plots of yearly catchability versus abundance in Figure 9. The fitted curves shown in Figure 9 are derived from a variant form of Equation 2 for single species:

Data from all the TF sites exhibit a trend of $1/Nj$ as abundances increase due to net-saturation effects (large $h$⁠). Had there been no saturation effects (⁠$h→0$⁠), all the data in Figure 9 would have followed horizontal lines corresponding to catchabilities of individual species targeted by the TF operations at respective sites. It is interesting to note that strong apparent saturation occurs at abundances exceeding 10 million for Sockeye Salmon, whereas nets targeting Chinook Salmon appear saturated at abundance levels as low as a few hundred thousand at Qualark. Among the three TF operations, Whonnock’s catch of Chinook Salmon is subject to the least saturation effect.

The daily probability model of Equation 12 can be improved by taking considerations of autocorrelations of daily catch data series over a correlation time scale of significance. In cases where it is adequate to manage fisheries in time scales greater than daily, daily catch data can be amalgamated into data series of weekly, biweekly or even monthly scales for the correction model. A probability function in the same form as Equation 12 for a longer time series of catch data will likely produce more precise CPUE predictions than the daily probability model.

Finally, the fundamental concept underlying this correction model is that due to differential behavior, spatial distributions, size distributions, or other biological features among species, a sampling gear, being it a fishing net, a sonar system, an underwater camera, or other remote sensing apparatus, is likely to have differential detection probabilities (DPs) among fish species. In the application of the correction model to the gillnet CPUE data, DPs are characterized by differential catchabilities among species. Observed composition data by other means can benefit from the same correction methodology by normalizing detection probabilities to a bench-mark species. For instance, one can use the DP of Sockeye Salmon as a measure for other species DPs. So, a correction model can be established similar to Equation 10 where $Rj$s are weights of DPs of individual species for the employed sampling gear.

DATA AVAILABILITY

The data associated with this study are available from the corresponding author upon reasonable request.

ETHICS STATEMENT

The handling of all fish that were captured for this study was in accordance with the regulations set out by Fisheries and Oceans Canada for scientific test-fishery operations.

FUNDING

This work was partially funded by the Pacific Salmon Commission’s Southern Boundary and Restoration Fund (Project Number: SF-2024-FRP-2).

ACKNOWLEDGMENTS

We thank Jim Krivanek and Charles K. Parken of Fisheries and Oceans Canada for providing the historic acoustic fish-count and test-fishing catch data at the Qualark sonar site, and we thank Catherine G.J. Michielsens and Fiona J. Martens of Pacific Salmon Commission for providing the acoustic fish-count and test-fishing catch data acquired at Whonnock and Mission sites. The Fraser Strategic Review Committee, appointed by the Fraser River Panel, presided over the review of the Fraser River hydroacoustic programs from 2008 to 2015 and provided the initial support for this study.

REFERENCES

Author notes