CCAMLR

The ASPM model used by Agnew and Kirkwood (2004, WG-FSA-04/82) has been modified through the introduction of a new annual selectivity function, modelled as a unimodal, algebraically decaying function of age, with three easily interpretable parameters.
When only a single selectivity curve was fitted for all years, the change to the new selectivity function improved the fit to length frequency data achieved last year, but the fit was still poor. The fit to the CPUE was also poor.
With selectivity parameters allowed to be estimated on an annual basis the fits are substantially improved. However, by allowing all parameters to be estimated in this way the model is over-parameterised and regularly reaches preset parameter boundaries. With one or two parameters of the selectivity estimated only once the fit to length frequencies is good but infinite populations are estimated. If the CPUE data are given high weight in the fitting, then the model has sufficient flexibility to change the selectivity to fit the CPUE very closely. Convergence problems were noted for the highly parameterised versions of this model.
Revisions to the selectivity model this year have led to an ASPM with superior performance. However, this study demonstrates that the ASPM is still not suitable for use when assessing South Georgia toothfish populations by fitting to the available catch length frequency data and CPUE data.

We investigate the spatial and temporal (i.e. fishing season) variation both systematic and random in haul-by-haul catch and effort data from the longline fishery for Dissostichus eleginoides in Subarea 48.3. We concentrate on presenting the results of preliminary work on the potential effects of changing fishing patterns on the standardisation of the CPUE time series. We consider whether the spatial pattern of fishing might have changed over the years, whether the current measure of fishing effort (number of hooks on a line) has changed in its fishing properties over time, and if soak time is a useful additional measure of effort. We then consider how these results might impact on the assessments, particularly the standardisation of the CPUE series. The results indicate that fishing for Dissostichus eleginoides is not uniformly distributed in its range. In some years, though not all, the fishery becomes concentrated into distinct grounds. In the CPUE standardisation procedure we found that although soak time was a significant predictor of catch the fitted relationship was not consistent with using soak time as a co-measure of effort by expressing CPUE as kg per hook-hour. In contrast the fitted log-linear relationship between catch and number of hooks set was consistent with using this variable directly as the measure of effort. The analysis of the interaction between the fishing season and the fitted coefficient for log(Hooks Set) in the regression model for catch suggested a declining trend in the coefficient over time. This may indicate that catchability has been changing and has perhaps become greater over time as a result of effort concentrating in productive locations. We recommend further consideration of the spatial characteristics of the fishery and its implications for using CPUE as an indicator of abundance.

This paper describes a possible method for implementing the precautionary approach in assessments of Dissostichus stocks when a number of different datasets are being integrated for estimating current biomass and age structure. Consideration is given to its application to the assessment of Dissostichus eleginoides in Division 58.5.2. CASAL and GYM are key elements in the implementation described in this paper but the methodology is laid out in such a way that alternative software solutions could be used in either or both of their places. The framework for this procedure has four main components. The first is the integrated assessment, undertaken external to the GYM by software such as CASAL. The second step is the sampling of the “current state” and associated parameters to provide inputs to the GYM. The third step is to use the GYM to undertake a number of projection trials, determining the median pre-exploitation spawning biomass and future states of the stock under the nominated harvest strategy. The fourth step is to test the performance of the harvest strategy against the decision rules. A controller is proposed to be used to manage the entire process including calling the CASAL or other software, the GYM and for evaluating the harvest strategy. The structure of this approach will help explore the value of integrating the datasets for Dissostichus elegionoides assessments such as are available for Division 58.5.2, which include recruitment surveys, commercial catch data from trawl and longline activities (biomass, length composition and age), and mark-recapture data. This method is an extension of current practice, although the framework is newly described and better coordinates the integration of the different steps in the precautionary approach used by WG-FSA. It is recommended that this approach be implemented for use by WG-FSA at its next meeting.

Three Operating Models (OMs) reflecting an optimistic, an intermediate, and a pessimistic current status for the toothfish resource in the Prince Edward Islands are developed. These are planned for use for initial trials of candidate Management Procedures (MPs) to provide future TAC recommendations for this resource. These will aim to provide an appropriate balance between avoiding risk of severe depletion and utilising potential harvest from the resource, in circumstances where the resource status is very uncertain.

A version of the confidence interval calculating program TrawlCI was recompiled to enable it to be run within the DOS emulator of recent versions of the Microsoft Windows operating system. The performance of the recompiled version was compared with that of the original version.
The recompiled version of TrawlCI produced very similar, though not identical, results to the original version. We attribute this to differences in the minimisation routines of the recompiled version. We conclude that the differences evident from these tests are unlikely to significantly influence the estimated long-term yield of Dissostichus eleginoides or Champsocephalus gunnari

Response-biased sampling in the context of regression modelling occurs when sample units are selected with a probability that is a function of the response. In sampling to obtain fish to both age and measure for length there are two potential response-biased sampling processes when length is the response and age is the predictor variable. These sample processes are (i) the actual fishing process involving a particular gear, and (ii) the method of on-deck sub-sampling fish for ageing from the random length frequency sample. When the selectivity of the gear combined with availability of fish to be caught is length-dependent in (i) and when fixed sample sizes per length bin or class are employed in (ii) then both these sampling processes are response-biased. Response-biased sampling and its effect on parameter estimation has been studied for linear and generalized linear, and linear mixed models but since population-average length given age is assumed to follow the von Bertalanffy growth relationship this work extends previous work to general nonlinear models and combines two response-biased sampling processes. Maximum likelihood, naïve least squares, and inverse probability weighted least squares estimation are used to estimate the von Bertalanffy parameters for simulated and real data on the growth of the Patagonian toothfish (Dissostichus eleginoides) given a known selectivity function.

This report outlines the development of a Bayesian sex and age structured population model for the assessment of Antarctic toothfish (Dissostichus mawsoni) in the Ross Sea (Subareas 88.1 and 88.2), and initial progress towards evaluation of spatially explicit models. Three model scenarios were investigated. The first scenario considered the Ross Sea fishery as a single homogeneous area (single-area). The second and third scenarios (2-area and 3-area respectively) split the Ross Sea into either two or three discrete areas, with migrations of fish between areas. The 2-area model appeared to provide a better representation of the some of the observations, although there was a lack of any observations defining migration between areas. The 3-area model, showed many problems in fitting the data adequately. In both of the spatial models, there was a lack of useful data on area splits, migration rates, or migration patterns.
Simulations experiments suggested that reliable, albeit uncertain, estimates of biomass could be obtained from each model when the operating model was the same as the estimation model. However, the single area estimation model was strongly negatively biased when using tag-recapture data simulated from the 2-area model.
All the multi-area models investigated here could do with considerable improvement. The choice of area (i.e., the boundary between areas), selectivity functions (i.e., either domed or logistic), selectivity types (i.e., either age based or length based), and migration ogives need further investigation. Additional tag-recapture data, specifically data that allows movement rates between areas to be quantified, are required to develop more realistic stock structure hypotheses.

Abundance estimates from tag-release and tag-recapture data require that the number and type of errors in data for analysis are minimised. Accurate recording of the sequence numbers on tags at time of release and recapture form an important part of data accuracy. Check digits, as a part of the tag sequence number, can assist in the identification of errors in recorded data.
A computer program to calculate, check, and validate a check digit scheme based on Verhoeff's Dihedral Group D5 check algorithm (Mohr 2005, Verhoeff 1969) is described.

A descriptive analysis of the toothfish tagging programme carried out in the Ross Sea since 2001 is presented for the first time. Tag-release and tag-recapture data are presented for both toothfish species for Subareas 88.1 and 88.2 for New Zealand vessels only. This is because data from non-New Zealand vessels were unavailable at the time of the analysis. A total of 4903 Antarctic toothfish have been released and 89 recaptured, and 443 Patagonian toothfish released and 9 recaptured. For the last two years, when tagging has been part of the Conservation Measure, New Zealand vessels have tagged between 1.0 and 1.37 toothfish per tonne of catch. Tagging rates by area over the past three years have been in the same proportion as the catch by area. However, recapture rates have tended to be higher in the northern and eastern SSRUs 88.1C and 88.2E.
The maximum movement of Antarctic toothfish from the New Zealand data set has been about 200 km. However, most (80%) Antarctic toothfish have moved less than 50 km. Consequently, nearly all fish have been recaptured from the same SSRU where they were released. The mean size of tagged Antarctic toothfish has increased since 2001, but is still smaller than the mean size of fish taken in the commercial catch. Larger toothfish (>35 kg) are difficult to tag without significant damage to fish, and there appears to be a trade-off between maximising size of released fish and minimising tagging mortality. Growth rates of Antarctic toothfish that have been at liberty for 2–3 years have averaged 5– 7 cm per year, which is consistent with growth rates predicted from the von Bertalanffy growth curve. The preliminary estimate of tag loss from double tagging experiments is 13% per year.

This paper describes an approach, using CASAL, to undertake operating model/estimation model experiments for Dissostichus spp., and methods for calculating CCAMLR yields using CASAL. We present an example model for the fictitious fishery on a fictitious species Dissostichus spurious (Everson, 2004) and investigate model performance from alternative types of observations. In general, most models were fitted with an expected percent root mean squared error (%RMSE) of less than 20%. The inclusion of additional data (i.e., more observation types) assisted in providing better estimates.
Further research on the expected performance of integrated models is required that investigates a range of alternative “true” states with data that includes bias and variance in observations, as well as the robustness of the estimation model to alternative operating model assumptions. In addition, the expected uncertainty that may arise from an MCMC approach has not been considered here.
In general, operating/estimation modelling experiments provide a means of evaluating alternative approaches to the assessment of stocks, however it should be noted that simulation studies often under-estimate the uncertainty that would be found in a real assessment.