Cusk (Brosme brosme) COSEWIC assessment and status report 2012: chapter 10

• Previous page
• Table of contents
• Next page

Population Sizes and Trends

Sampling Effort and Methods

While a number of fishery independent surveys have been conducted in the species range of Cusk, only a relatively small number are suitable for estimation of trends in Cusk abundance and biomass. These include:

• DFO summer bottom trawl survey
• NMFS fall bottom trawl survey
• Industry Atlantic Halibut survey

The NMFS spring bottom trawl survey is also useful but it is a shorter series than the fall survey. As the two series exhibit similar trends, focus in this assessment is placed on the fall survey. In addition to these surveys, commercial catch rates (CPUE) of the longline fleet operating in NAFO Divisions 4X and 5Z have been used as estimates of biomass trends (Harris and Hanke 2010).

The DFO summer bottom trawl survey has been conducted every July – August since 1970. Employing a stratified – random design and standardized sampling protocol (Halliday and Kohler 1971), about 200 bottom trawl sets are annually made on the Scotian Shelf. Over the duration of the time series, the sampling effort has increased. The spatial distribution of the strata was described earlier. Harris and Hanke (2010) consider that this survey does not representatively sample the population because the survey does not sample rocky bottom areas, the preferred habitat of Cusk. It was argued that as Cusk abundance changes, it will either spread out from this habitat (high abundance) or retreat to these areas (low abundance). This can lead to hyperdepletion where the survey index declines faster than the population biomass, implying that survey catchability, q, is a function of Cusk abundance. This hypothesis was tested in population modelling by Davies and Jonsen (2011), who considered it to be supported by the information available. This is discussed further below.

The NMFS fall bottom trawl survey has been conducted every October – December since 1963. Also employing a stratified-random design and standardized sampling protocol (Grosslein 1974), the survey annually undertakes about 350 bottom trawl sets in the Gulf of Maine area. Harris and Hanke (2010) contend that this survey suffers the same catchability issue as the DFO survey.

The industry Atlantic Halibut survey is conducted during late May – late July every year since 1998 using both a commercial and fixed station design. For the purposes of this assessment, only the data from the fixed stations that have been consistently occupied (which reside in 4VWX) have been used. The protocol of the Halibut survey is discussed by Trzcinski et al.. (2011). As the protocol was in development during the first year (1998) of the survey, these data are not used in this assessment. Also, while set catch was standardized to 1000 hooks, no standardization for soak time was undertaken. This is consistent with the treatment of these data by Harris and Hanke (2010). Sampling has averaged about 55 stations per year with little variation.

As in COSEWIC (2003), commercial landings and effort data for longline vessels were used to estimate an annual CPUE (tons per trip) index of mature biomass. Longliners deploy their gear in a variety of habitats including deep water along the edge of the continental shelf and in rocky bottom areas that may not be well sampled by survey trawls.

Developing a catch per unit effort (CPUE) index requires careful data selection to avoid potential biases introduced by fishing. The design of the CPUE index here is based upon that of Harris et al.. (2002) and Harris and Hanke (2010) with some minor modifications. In their analysis, Harris and Hanke (2010) limited the CPUE data to NAFODiv 4X, excluding those in 5Z. The number of trips made in each area has been similar since 1986, except in the mid-1990s when effort in 5Z was considerably lower. Given the distributional range of Cusk, it was decided to use both the 4X and 5Z data in this assessment.

Longline vessels operating in this area target a range of groundfish species including Cod, Haddock, and Pollock. Cusk are generally a bycatch to this fishing. Prior to 1999, a longline fishery for ‘shack’ has been recognized, which was directed for Cusk and White Hake combined. Consistent with Harris and Hanke (2010), all trips that caught any of these five species, regardless of the target species, were used in the CPUE index.

Harris and Hanke (2010) reported that the reliability of data prior to 1986 was questioned anecdotally, presumably by fishermen, when a landing cap of 1000 t for Cusk was implemented. Previously, there was no landing limit for Cusk. It has been suggested that other species, such as Atlantic Cod, were reported as Cusk when target species quotas were exceeded. As well, misreporting and discarding were considered prevalent in the fishery. For this reason, they did not use the CPUE data prior to 1986. In 2003, the landing cap was reduced to 750t where it has remained. Here, the 1986 – 2010 CPUE data are used in the CPUE index.

Harris and Hanke (2010) based their CPUE index on the data from tonnage class two and three vessels. Tonnage class 2 and 3 vessels consistently contributed about 50 – 60% of the longliner fleet’s landings until the mid-1990s after which time, their share declined to the 20% level observed more recently (Figure 18). The contribution of small tonnage class zero and one vessels has, on the other hand, increased. Reliable effort data for these smaller vessels do not exist. The data for the TC 2 and 3 vessels were considered adequate to estimate a CPUE index.

Figure 18. Share of longline fleet landings in all NAFO areas by tonnage class.

Harris and Hanke (2010) produced their CPUE index through averaging of monthly catch rates of tonnage class two and three longliners operating in NAFODiv 4X during July – September. There have been significant changes over the time series in the months during which the fishery occurs, with it becoming focused in July – September since the late 1990s. To ensure consistency of the time series, as with Harris and Hanke (2010), the analysis of the CPUE data was limited to these months.

For the subset of selected data since 1986, there has been a significant decline in the number of trips (Figure 19) with current levels (278) being well below peak values (1500 – 2500) in the early 1990s.

Figure 19. Trend in annual number of trips fished by tonnage class 2 and 3 longlines in NAFO Divs. 4X and 5.

Very little observer information is available for this fleet (Table 6). Since 1985, only 314 of the possible 24,516 trips have been observed. In addition, there were even fewer trips where length measurements of Cusk were taken. Therefore, the DFO Maritimes Science port sampling data were used to characterize longline landings size composition. An overview of port sampling intensity is given in Table 7. Since 1960, 459 samples have been taken with the majority of these since the mid-1990s. Sampling was focused on the longline fleet with few samples on trawl and gillnet landings and no samples of traps and other gears.

Abundance

Table 8 summarizes the abundance indices of the DFO summer bottom trawl survey. These were estimated for the whole survey area (4VWX) rather than just for 4X, as was done by Harris and Hanke (2010). While the two series are very similar, it was considered that the broader area index was more representative of abundance trends for the entire DU. This survey indicates a significant decline in mature (42 cm+) numbers since the start of the time series (Figure 20). The average indices in the survey since the last COSEWIC assessment (2002-2010) are lower than from the 9 years prior to that assessment (1993-2001). The average mature numbers declined from 341,000 to 218,000 while the mature biomass declined from 695 t to 407 t. These are underestimates of the true abundance and biomass because Canadian landings annually averaged 865 t over the 2002-2010 period and between 250-300 t was estimated to have been discarded annually in the Lobster fishery (see section on Threats and Limiting Factors). Thus, current catch is well in excess of the minimum estimates of mature biomass from the DFO summer bottom trawl survey. Therefore, these are considered minimum estimates of abundance and biomass and they are taken as an index of relative abundance.

Figure 20. Trend in abundance of immature (<42 cm) and mature (42 cm+) Cusk in NAFO Divs 4VWX, based upon DFO summer bottom trawl survey.

The length composition of Cusk taken in the trawl survey has contracted over the 40-year time series (Figure 21). In the 1970s and 1980s, the population on the Scotian Shelf was dominated by 55 cm + individuals. Since then, abundance of these length groups has declined considerably in comparison to the smaller Fish.

Figure 21. Decadal change in length frequency of Cusk in NAFODiv 4VWX as observed by DFO summer bottom trawl survey.

Table 9 summarizes the NMFS fall bottom trawl survey indices. The trends and CVs are similar to those in the DFO summer survey (Figure 22). The fall survey provides a minimum estimate of current abundance in the order of 450,000 individuals and a minimum estimate of mature biomass in the order of 580. Note that again these data are considered as relative and not absolute estimates of abundance and biomass.

Figure 22. Trend in abundance of immature and mature (53 cm+) Cusk in NAFO Divs 5Z-6, based upon NMFS fall bottom trawl survey.

Similar decadal changes in length frequencies were observed in the NMFS survey as in the DFO summer survey (Figure 23), although the decline in mature individuals since 1988 is more evident in the NMFS fall survey.

Figure 23. Decadal change in length frequency of Cusk in NAFODiv 5Z-6 as observed by NMFS fall bottom trawl survey.

Harris and Hanke (2010), in their analysis of Cusk catch rates in the Atlantic Halibut survey, averaged the kg/1000 hooks of the 53 core stations to produce annual estimates of a biomass index. This analysis was repeated here. In addition, two analyses using the core station results and GLM models were used. One assumed a lognormal error distribution and the other assumed a negative binomial distribution as was done for Atlantic Halibut (Trzcinski et al. 2011). The R code and diagnostics for both the linear and GLM models are provided in Appendix 2*. The indices of the three models are provided in Table 10 and the trends in Figure 24. The results of the negative binomial model were used in trend analysis.

Figure 24. Trends in Cusk biomass indices (kg / 1000 hooks) based on Industry Atlantic Halibut survey. The legend is as follows: LM is the GLM estimate assuming lognormal error, GLM NB is the GLM estimate assuming negative binomial error (this is the index used in trend analysis), and Average is the simple mean of the fixed station catch rates.

The trends of the three indices are similar except for the 2011 increase in kg/1000 hooks observed in the average index. The fit of the models was poor (e.g. R² of 34% in the linear model) with influential observations evident. All models indicate a small decline at the beginning of the survey time series and a modest increase since 2006. There is some evidence of a recent declining trend. Overall, though, these effects are small with the overall trend being relatively flat.

Cusk lengths have not been consistently sampled on this survey although a reasonable amount of data is available. These data indicate that the Halibut survey has been catching almost exclusively mature (42cm+) Cusk throughout the times series (Figure 25). A comparison of size frequencies of the DFO summer and Halibut industry surveys for 2000 – 2010 indicates that the latter currently catches a significantly higher proportion of Cusk at lengths greater than 60 cm (Figure 26). These are sizes of Cusk seen by the DFO survey during the 1970s and 1980s (Figure 21). Indeed, this survey caught even larger Cusk at that time. Thus, the DFO trawl survey can catch this size of Cusk but does not appear to be doing so during the 1990s and 2000s.

Figure 25. Temporal change in Cusk proportion at length in industry Atlantic Halibut survey.

Figure 26. Comparison of Cusk average proportion at length observed in the Halibut and DFO summer trawl survey during 2000 – 2010.

A number of explorations of the commercial longline fishery’s catch rates were undertaken, using month and unit area as factors in both linear and GLM negative binomial models. As observed by Harris and Hanke (2010), the model explorations did not produce satisfactory statistical fits (available upon request from the status report writer). Therefore, the NAFODiv 4X 5Z average was used as the preferred CPUE index of biomass (Table 11). Figure 27 compares this series to that of Harris and Hanke (2010) which is reported in Davies and Jonsen (2008). The two series indicate the same overall trend.

Figure 27. Trends in CPUE indices for tonnage class 2 and 3 longliners fishing in NAFODiv 4X5 during July – September; note that the Harris & Hanke (2010) index was only for NAFODiv 4X.

The commercial port sampling data indicate that the landings during quarter 2 and 3 reported by tonnage class 2 and 3 longliners consist of almost exclusively mature (>42 cm) individuals. Decadal changes in the size composition of the landings were evident (Figure 28). Modal size has become smaller, paralleling the trend in the DFO and NMFS surveys. There were some samples from 1961 – 62. Surprisingly, these indicated a smaller size range exploited than in the 1970s. This may indicate either inadequate sampling or changes in size composition not associated with population changes (e.g. regulatory changes).

Figure 28. Decadal changes in landings of tonnage class 2 and 3 longliners operating in NAFODiv 4X5 during the 2^nd and 3^rd quarters of the year.

As indicated earlier, the declines estimated based on trawl survey results were considered to have been overstated (Harris and Hanke 2010) due to a possible relationship between trawl survey catchability and Cusk abundance (hyperdepletion): as abundance declines, Cusk may retreat to their preferred rocky habitat that is relatively inaccessible to the survey trawl gear. Davies and Jonsen (2008, 2011) explored the possibility of a change in trawl survey catchability with Cusk biomass using a Bayesian surplus production model (equation 6):

B_t = (B_t-1 + rB_t-1 (1 –B_t-1/ K) – C_t-1) η_t      (6)

B_t-1 and C_t-1 denote biomass and landings in year t – 1 respectively, r is the intrinsic rate of population growth, K is the carrying capacity (population biomass at equilibrium before exploitation and ht is a lognormal random variable with a mean of zero and variance σ² to account for stochasticity in population dynamics. The observation model (equation 7) incorporated a shape parameter to explore hyperdepletion in the survey indices:

I_i,t = q_iB_t^βε_i,t     (7)

I_i,t is survey i in year t, q_i is the catchability of survey i, and β is the shape parameter. If β is less than one, hyperstability is indicated (the index declines at a slower rate than the populaton biomass; common in many fisheries). If it is above one, hyperdepletion is indicated (catch rate falls faster than biomass decline). Initial versions of the model (Davies and Jonsen 2008) explored a number of data inputs and configurations. Ultimately, they settled on using NAFODiv 4X landings, along with the 4X CPUE (average) index and the DFO summer bottom trawl survey to characterize Cusk population dynamics (Davies and Jonsen 2011). The industry Halibut and 4VsW sentinel surveys were not used, as they were uninformative of stock dynamics due to being short time series. While the estimated parameters of these models were highly uncertain, hyperdepletion in the trawl surveys was shown to improve model fit. Cusk biomass was estimated to have declined 59% during 1970 – 2001 and 64% during 1970 – 2007, substantially lower than the rate estimated when assuming survey catchability is independent of Cusk biomass.

This model (Appendix 3* – equivalent to Model 3 of Davies and Jonsen 2011) was updated with some changes. First, the 1970 – 2010 Canadian total catch from all sources, rather than the landings for NAFODiv 4X + 5, were used. This includes the discard data provided in the Threats and Limiting Factors section below. Second, the 1986 – 2010 CPUE index (average) for NAFODiv 4X+5 rather than just 4X was used (Table 11). Finally, the 1970 – 2010 DFO summer bottom trawl survey index of mature biomass for NAFODiv 4VWX rather than the stratified mean kg/tow for NAFODiv 4X was used (Table 8). Similar to Davies and Jonsen (2011), model convergence was tested using two chains (300,000 total iterations with 260,000 burn-in iterations and a thinning rate of 20) resulting in a Gelman – Rubin Diagnostic of Rhat = 1, providing strong evidence of convergence (Ntzoufras 2009). The model fit the two biomass indices without apparent trends in residuals (Figure 29). The posterior distributions of the model parameters are provided in Figure 30 and the trend in the annual proportion biomass of the carrying capacity (K) is given in Figure 31. A summary of the posterior quantiles of key model parameters is provided in Table 12.

Figure 29. Observed (dots) and model predicted (lines) ln(indices) of Cusk biomass; 4X5 longline CPUE (top panel) and DFO summer survey (bottom panel).

Figure 30. Posterior density plots of model parameters; tau.com and tau.rv are the observation error on the CPUE and DFO survey indices, σ is the process error, Commercial and Survey Q are the CPUE and DFO survey catchability; the remaining legends are self-explanatory.

Figure 31. Trend in proportion that annual Cusk biomass is of carrying capacity (K) from state – space model; 25^th, 5^th (median) and 75^th percentiles provided.

These results are very similar to those of Davies and Jonsen (2011) recognizing the uncertainty in the parameters. The intrinsic rate of population growth, r, is 0.09, close to the previous estimate of 0.12. The carrying capacity, K, is slightly higher at almost 69 kt compared to the previous estimate of 52 kt. MSY and BMSY are estimated to be 1.6 ktand 34 ktrespectively, compared to the previous estimates of 1.5 ktand 26 kt. Most significantly, the shape parameter on the DFO summer survey is well above one, with the median estimate being 2.1 (the previous estimate was 2.5). This is strong support for the hyperdepletion hypothesis.

Davies and Jonsen (2011) indicated that there is high uncertainty in the parameter estimates of the model. For instance, biomass is estimated to be close to virgin levels in 1970. However, Cusk have been fished since the early 1900s. Canadian reported landings in the 1960s averaged about 4000t annually. Thus, it is highly unlikely that the stock was at carrying capacity. More credence should be given to the model’s relative rather than absolute trends in biomass. Given the parameter uncertainties, Davies and Jonsen (2011) showed that the model could not be used to reliably predict future states under different catch scenarios. While an annual catch of 750 t (close to recent values) should be sufficient to allow biomass increase, this in fact has not occurred. Davies and Jonsen (2011) consider that this could be due to 1) high bycatch mortality, 2) lack of data on recent recruitment, 3) poor recruitment during 2000 – 2007 or 4) reduced productivity / increased natural mortality. While the model can explain historical trends, it does not appear to be informative in the prediction of future states. Given this uncertainty, it was decided not to undertake analyses of recovery trajectories.

Fluctuations and Trends

The overall trend in most indices of mature abundance and biomass are of long term decline. The times series of Cusk abundance from the DFO summer survey adjusted for hyperdepletion, commercial longline CPUE, and Halibut longline survey, standardized to their 2000 – 2010 means, are provided in Figure 32. The DFO summer survey indicates a continuous decline since the mid-1970s to the present. The commercial CPUE index has declined continuously since it began in 1986 and at a rate comparable to the adjusted DFO trawl survey index. The Halibut survey time series is too short to indicate long-term changes. However, it has been stable since it began in 1999.

Figure 32. Cusk abundance indices from the DFO trawl survey adjusted for hyperdepletion, the commercial longline CPUE, and the Halibut longline survey, standardized to their 2000 – 2010 means.

The annual instantaneous rate of change (α) in the various indices were determined as the slope of the ln(index) versus year (log-linear regression). Data on the number of mature individuals were available for the DFO trawl surveys, and for mature biomass for the commercial CPUE and Halibut survey indices.

 The percent change (%Δ) over a specific time period t was estimated using equation 5. The regressions are plotted in Figure 33.

%Δ = 100 * (exp(α * t) – 1) (5)

Figure 33. Log linear regressions of Cusk abundance from the DFO trawl survey index (1974-2010), the same index adjusted for hyperdepletion, the commercial CPUE index (1986-2010) and the Halibut longline survey (1999-2011). The range of the y-axis values in each panel is the same thus allowing visual comparison of the estimate slopes.

The estimated change in mature population numbers from the DFO summer survey over a 3-generation period (1974-2010) was −98%, based on a statistically significant slope estimate of −0.103 (Table 13). A regression over the last 2-generations (1986-2010) produced a statistically significant slope estimate of −0.117 which gives an estimated change over this time period of −94%. The slope estimate from the latest one-generation time period (1998-2010) was not significant but the estimated change was −70%. For the CPUE time series, the slope estimate for the last 2 generations (1986-2010) was statistically significant and indicated a change of −73%. The estimated change over the last generation was −55%, based on a significant slope estimate of −0.067. The regression of the Halibut survey results produced a positive but non-significant slope estimate of 0.018, and this indicates a change of +24% over the time period. The results from the NMFS fall survey indicate a change in mature numbers of −83% over the latest 3 generations.

The surplus production model results indicate that biomass was close to virgin levels in 1970 and declined to 14% of this by 2010. This is an 85% (equation 5 applied to proportion data of Figure 31) decline in biomass over 3 generations (Figure 34). The estimated shape parameter β was used to adjust the DFO survey results using the equation Î_t ≈ I_t^1/β and a log-linear regression of the adjusted index produced an estimated 3-generation decline of 83%, and a decline of 73% over the past 2 generations, the same time period covered by the CPUE index. In other words, the trends in the adjusted DFO survey and CPUE index are similar.

Figure 34. Trend in Ln Proportion annual biomass of carrying capacity from Bayesian Surplus Production Model.

The overall estimated decline of 85% is more extreme than the 63% estimated by Davies and Jonsen (2011). It should be noted that some of this difference is related to how the percent decline is estimated. As noted above, Davies and Jonsen (2011) estimate a 59% decline in Cusk biomass between 1970 and 2001 and a 64% decline from 1970 to 2007. The latter is based upon the ratio of the 1970 to 2007 proportion biomass (of carrying capacity) or 1 – 0.34/0.91 = 63%. The estimate based upon equation 5 is 75%. To further explore the source of the change in perception between Davies and Jonsen (2011) and this assessment, the original model was run using updates to the 1970 – 2007 catch, CPUE and DFO summer trawl survey data both separately and combined. The updates produce similar percent declines in biomass during 1970 – 2007 to those of Davies and Jonsen (2011), suggesting that it is the addition of the 2008 – 2011 data that is causing the perception of further decline (Table 14).

A trend that the model does not capture is the stability (or slight increase) in mature biomass since 1999 reflected in the Atlantic Halibut survey (Figure 24). The slope of the Halibut survey is not significantly different from zero, whereas those of the other indices (except for the 1998 – 2010 DFO Summer Survey), are significantly different from zero (Table 13). The model indicates a continuing decline during this period. Davies and Jonsen (2011) did not include this index in their model, as was done here, due to the shortness of the time series.

The following is a summary of the main conclusions of the trend analyses. The DFO summer bottom trawl survey indicates a long-term decline in abundance and biomass, even when hyperdepletion is taken into account. Associated with this decline has been a shift to smaller sizes and a reduction in the IAO. The longline CPUE has also declined at a similar rate and there has been a shift in the catch’s size composition to smaller individuals. Of note is that the fishery and survey occur in generally the same area, so that this correspondence in trends indicates that they are measuring the same process. The population model also indicates a continuous decline in Cusk biomass over the past 3 generations. The Halibut survey indicates that large Cusk are still present on the Scotian Shelf and that biomass has been stable since 1999. The Halibut survey samples deeper waters off the Shelf that are outside the coverage of the trawl survey. It is possible that these waters still contain large Cusk, whereas they have been depleted on the Shelf proper.

Rescue Effect

The main source of a rescue effect for Cusk in Canadian waters would be from Cusk in the Gulf of Maine in US waters. As indicated in the population trends above, Cusk in this region have declined to a similar extent as in the Canadian zone. Indeed, NMFS conducted a workshop in November 2009 to describe its status from an endangered perspective and many of the issues noted for Canadian Cusk were also mentioned. Thus, this and the low suspected movement rates of Cusk suggest that the likelihood of a rescue from the US Gulf of Maine is low. There is thought to be little movement of Cusk between West Greenland and Canadian waters.