THEME 2.
Population biology and resource assessment

The challenges of, and future prospects for, assessing deepwater marine resources: experiences from Australia, New Zealand, Southern Africa and the United States

A.E. Punt
School of Aquatic and Fishery Sciences
University of Washington, Box 355020, Seattle, WA 98195-5020, USA
<[email protected]>

1. INTRODUCTION

The traditional role of fisheries stock assessment has been to provide the "best estimates" of the biomass of a stock and a measure of its potential productivity. Surplus production models (e.g. Punt and Hilborn 1996) and Virtual Population Analysis (e.g. Pope and Shepherd 1985) were techniques that could be used in data-poor (only catch data and an index of abundance) and data-rich (catch data, an index of abundance and catch-at-age data for all years and fishing fleets) situations respectively.

The results from these techniques can be used to determine whether a stock is above or below a target level of biomass (e.g. the biomass at which Maximum Sustainable Yield [MSY] is achieved BMSY) and whether catches are greater than the estimated MSY. In addition, they can be used to determine catch limits using catch control rules such as ^{[39] [40]}

During the 1970s it was realized that the estimates from these methods could be both imprecise and biased. Imprecision arises because of the use of noisy data when fitting population dynamics models and bias arises because some of the assumptions of the population dynamics model (e.g. that CPUE is linearly related to abundance or that catchability has remained constant over time) are violated. Basing management decisions on the "best estimates" for biomass and MSY implicitly ignores the implications of both bias and imprecision. Therefore, if the catch limit is to be 30 percent of the estimate of current biomass, it is 300 irrespective of whether the biomass is estimated to be 1 000±10 or 1 000±1 000. An even worse outcome is the use of evidence for imprecision (when it is assessed) to avoid reducing harvest levels when this is deemed necessary.

Recent evaluations of the ability of traditional methods of stock assessment to estimate quantities of management interest (e.g. Walters and Pearce 1996) suggest, for example, that estimating the biomass available to the fishery within ±40 percent should be considered to be a relatively successful outcome. The ability to estimate the ratio of current biomass to a reference biomass level is better (Punt, Smith and Cui 2002) and control rules exist that do not require any estimates of absolute abundance but instead rely on trends in biomass (e.g. Magnusson and Stefansson 1989).

The realization that assessment results may be both biased and imprecise led, in part, to the development of the Precautionary Approach to Fisheries Management (Anon. 1995). Although the Precautionary Approach deals with more than simply uncertainty related to assessments, it has caused fisheries assessment scientists to focus less on obtaining "best estimates" (although these are still central to the decision rules used to set harvest guidelines for U.S. fisheries) and more on quantifying the risk associated with alternative catch limits that do not satisfy some pre-agreed management objectives or to determine a set of models on which to evaluate the advantages and disadvantages of alternative decision rules.

The uncertainty associated with assessing a stock is directly related to the amount of data and understanding available for assessment purposes. The development of fisheries on the continental slope and in deeper waters has tended to parallel the change from providing "best estimates" to providing information that it is more consistent with the Precautionary Approach to Fisheries Management. However, assessing fisheries in deepwater poses some unique challenges. For example, many deepwater fisheries only started relatively recently with the consequence that the time-series of abundance indices are usually short and traditional methods for obtaining data on stock status, such as tagging, and ageing a large fraction of the catch, have not been applied. Deepwater fisheries are also among the most technologically challenging (and sophisticated); one consequence of this is that fishing efficiency can change rapidly over time, thereby potentially making commercial catch rates inherently unreliable as indicators of stock abundance.

The response of the fisheries assessment community to these challenges has been to develop methods of stock assessment that attempt to synthesize all of the available information into a single analytical framework and to express the results of assessments in probabilistic rather than best estimate terms. The results from studies of similar species in other parts of the world are being used to place prior distributions on the likely values for model parameters. These trends are being paralleled by the wider participation of a broad range of stakeholders in identifying and developing assessment scenarios. Each of these trends is now examined in some detail.

2. SYNTHETIC MODELS AND DEEPENING THE POOL OF MODELS

Deepwater fisheries are frequently data-poor in that data are often sparse (e.g. some length-frequency and catch-at-age information, and occasional survey estimates of relative abundance). In contrast, unlike fisheries that have been operating for many years, the information on historical removals tends to be more complete for deepwater species because, in general, fishing on the slope started relatively recently.

Of the two classes of model on which assessments have traditionally been based, surplus production models are better suited to situations in which a catch time-series and information on relative (or absolute) abundance is available because conventional Virtual Population Analysis methods usually require information on catch-at-age for all years.

Surplus production models can be criticized however for a lack of realism (Hilborn and Walters 1992) and because they cannot use information other than that contained in the relative abundance series. The latter is a major problem for data-poor situations because it could lead to a substantial fraction of the available information being ignored. Delay-difference models (e.g. Deriso 1980, Schnute 1985) overcome some of the problems associated with surplus production models because (subject to certain assumptions) they capture the impact of age-structure dynamics and can hence include information related to growth and natural mortality in assessments. Delay-difference models can be structured to make use of a wider range of data types than surplus production models. For example, they have been extended to be length-structured so that they can be fitted to the moments of catch length distributions (Fournier and Doonan 1987). However, the assumptions that typically underpin delay-difference models (such as knife-edged recruitment and that the age-at-maturity equals that at-recruitment) may be violated to a substantial extent in actual situations.

Hilborn (1990) introduced the idea of fitting relative abundance information to age-structured models that start the population projections prior to the onset of fishing (so that the assumption that the stock was at its pre-exploitation size at this time is not likely to be violated substantially). This approach has been referred to as an age-structured production model (ASPM) by Punt (1994) who showed that the impact of adding age-structure to an assessment model was not necessarily particularly substantial (given sufficiently informative data on relative abundance). Nevertheless, age-structured production models do overcome several criticisms of lack of realism leveled at surplus production models because they include age-structure and represent biological processes such as growth, mortality, recruitment, etc. more explicitly. ASPM models form the conceptual basis for the bulk of the assessments of newly developed deepwater species because most more complicated stock assessment models such as those that underlie the "Integrated Analysis" approach to stock assessment include the ASPM approach as a special case.

The statistical catch-at-age analysis (or Integrated Analysis) method of stock assessment developed by Fournier and Archibald (1982) and Deriso, Quinn and Neil (1985) differs from traditional Virtual Population Analysis methods because it separates the development of the model for the population dynamics from that for the observed data. Integrated analysis forms the basis for the Stock Synthesis program (Methot 1990, 2000). Unlike Virtual Population Analysis there is no need for catch-at-age data for all years so this method of stock assessment can be applied when catch-at-age data are available for only a subset of the years of the assessment period. In fact, this method can be applied with only length-frequency data or no catch-at-age or length-frequency data at all The ability to deal with situations in which there are missing catch-at-age data for some years is one of the main reasons that Integrated Analysis is currently the most commonly applied stock assessment technique for deepwater species off New Zealand, Australia, South Africa, and the west coast of the US. For example, all of the quantitative assessments of stocks in Australia’s South East Fishery are based on some variant of the Integrated Analysis method (Punt, Smith and Cui 2001).

A particularly important feature of the Integrated Analysis approach is that the analyst is not constrained by the structure of the data available for assessment purposes when developing the model of the population dynamics (Smith, Smith and Punt 2001). This allows the analyst to develop a range of alternative models regarding the population dynamics and the relationship between the data collected and the model predictions. McAllister and Kirchner (2001) conducted assessments of orange roughy (Hoplostethus atlanticus) off Namibia based on four alternative models for the cause of the change over time in the index of abundance while the assessment model developed for hoki (Macruronus novaezelandie) off New Zealand includes two stocks, and fisheries and surveys in six areas (Francis, Cordue and Haist 2002). These two assessments also provide an illustration of the types of data used when applying Integrated Analysis (commercial catch rates, acoustic estimates of biomass, and swept-area estimates of biomass for orange roughy; and these three data types as well as survey and fishery age-composition data for hoki). Table 1 overviews the assessments of species caught in substantial quantities on the continental slope off the US west coast in terms of the data types included in the assessments. Length- and age-composition data are both included in some assessments when length but not age information is available for some years. These assessments therefore attempt to make use of as much data as possible.

Table 1
Summary of assessments of species caught to a substantial extent in fisheries on the continental slope of the US west coast

The assessment of hoki off New Zealand referred to above is one of the first that explicitly allows for spatial structure. The trend towards including spatial structure in assessments is likely to continue as assessment scientists examine, for example, length- and age-structure and relative abundance data for spatial patterns. Such patterns are likely to be present when there is spatial heterogeneity in fishing pressure for species (such as rockfish Sebastes spp.) that do not exhibit long-distance movement.

Attempting to use all of the available data in a single analysis should lead to more accurate and precise estimates of management-related quantities. In general, the objective function minimized to find the values for the model parameters is a weighted function of the contributions from individual data sources. Unfortunately, it is not uncommon for the results of an assessment (both quantitative and qualitative) to depend on the weight placed on the different data sources. This is illustrated in Table 2 by the sensitivity of the results of the assessment of hoki off New Zealand to changing the weight assigned to the trawl and acoustic estimates of biomass. The reason for contradictory data is that the model of the population dynamics or of the relationship between one (or all) of the data sources and the model predictions is wrong. The sensitivity of the results to the weights placed on the different data sources can be disconcerting because it indicates that something is wrong but having multiple data sources to test for such sensitivity is better than the comfort that arises when only one data source is included in an assessment and no sensitivity analysis is possible.

Assessments based on Integrated Analysis frequently have many parameters, e.g. the assessment of Pacific Ocean perch (Sebastes alutus) off the US west coast involves the estimation of some 253 "free" parameters - Hamel, Stewart and Punt (2003). The ability to conduct assessments based on the Integrated Analysis paradigm has increased substantially given that computers are becoming increasing powerful and particularly given the availability of the AD Model Builder package^[41]. The AD Model Builder package includes algorithms based on automatic differentiation techniques to compute the gradient of the function to be minimized. The use of automatic differentiation can dramatically reduce the time required to fit stock assessment models with many parameters (Schnute, Richards and Olsen 1998).

Table 2
Sensitivity of the posterior median estimates of biomass in 2002-2003

B2002-03, in absolute terms and expressed relative to the average pre-exploitation biomass B0, for two stocks of hoki off New Zealand, to the weights assigned to the trawl and acoustic data when conducting the assessment (Source: Annala et al. 2003)

Although the trend towards more complicated multi-parameter models is almost certain to continue because it allows analysts the opportunity to include more hypotheses (e.g. that the selectivity of the fishery changes over time) as well as more sources of data in assessments, it is not necessarily true that the ability to make reliable predictions is improved by more complicated models. Unfortunately, research into optimal model complexity and how to select among alternative model formulations has lagged substantially behind the ease with which it has become possible to conceive new models and to fit them to data.

Prior to the availability of software to fit complicated models rapidly, analysts tended to conduct some analyses separately from the main stock assessment to estimate, for example, the growth curve. The results of these "supporting analyses" were then assumed to be known exactly when conducting the actual assessment. However, not including all of the data in the assessment has the potential to miss any trends in, for example biomass and recruitment, which these data suggest (Maunder 2001). Therefore, increasingly, assessment scientists are "integrating" all of the available data into the assessment and dispensing with the idea of supporting analyses. For example, the assessment of orange roughy in New Zealand’s QMA 3B conducted by Smith et al. (2002) estimated the growth curve within the assessment rather than estimating it externally as had been the case in previous assessments (e.g. Francis 2001).

3. MOVING FORMALLY TO A PROBABILISTIC FRAMEWORK

The requirement for scientists to provide information to managers on uncertainty about stock assessments and forecasts arises in part from Article 7.5 of the FAO Code of Conduct for Responsible Fisheries (Anon. 1995) which includes the recommendation: ‘States should apply the precautionary approach widely to conservation, management and exploitation of living resources... In implementing the precautionary approach, States should take into account, inter alia, uncertainties relating to the size and productivity of the stocks, reference points, stock condition in relation to such reference points, levels and distribution of fishing mortality and the impact of fishing mortalities...’. A similar phrasing can be found in Article 5© of the UN agreement on the Conservation and Management of Straddling Stocks (Sainsbury, Punt and Smith 2000).

There is a variety of sources of uncertainty when conducting assessments and examining the consequences of management actions. Francis and Shotton (1997) identify five sources of uncertainty.

i. Process uncertainty ("process error") arises from natural variability. The most common example of process uncertainty is variation in recruitment for environmental reasons.

ii. Observation uncertainty arises through measurement and sampling error although deliberate mis-reporting (of catches for example) also constitutes a form of observation error.

iii. Model uncertainty arises through a lack of understanding of the underlying dynamics of the system being considered.

iv. Error structure uncertainty arises from the inability to correctly identify the sources of error when fitting models to data.

v. Implementation uncertainty reflects the implications of the inability to fully implement management actions.

The need to compute measures of uncertainty is well recognized and considerable attention has been placed on developing methods for this, as reflected by the number of conferences and symposium sessions dedicated to the topic (Francis and Shotton 1997). However, several of the most commonly applied methods of stock assessment based on Virtual Population Analysis (such as Extended Survivors Analysis - Shepherd 1999) are not easily modified so that each of the above sources of uncertainty can be quantified.

Bootstrapping and Monte Carlo approaches (Butterworth, Hughes and Strumpfer 1990, Restrepo et al. 1992) can be used to quantify the uncertainty of model outputs and to form the basis for examining the consequences of alternative management actions for assessments based on Virtual Population Analysis. In contrast, methods of assessment based on Integrated Analysis usually represent the relationship between the data collected and the model predictions through a formal likelihood function. This means that it is possible to apply profile likelihood methods for determining confidence intervals and, in particular, Bayesian methods. The latter are increasingly becoming the method of choice for quantifying uncertainty when applying Integrated Analysis (particularly in Australia and New Zealand) and the number of Bayesian stock assessments of deepwater fish species has increased rapidly over the last ten years (e.g. Francis 1992; Smith and Punt 1998; Ianelli, Wilkins and Harley 2000; Punt et al. 2001; Francis et al. 2002; McAllister and Kirchner 2003; Hamel, Stewart and Punt 2003).

The use of Bayesian techniques when conducting fisheries stock assessments is desirable because inter alia, Bayesian methods provide a single framework within which various sources of uncertainty can be represented (in particular, both parameter and model-structure uncertainty), and because the results from a Bayesian analysis (the probabilities associated with alternative hypotheses) are exactly the information needed when providing scientific management advice to decision makers (Punt and Hilborn 1997, McAllister and Kirkwood 1998). However, the primary reason that most stock assessment scientists choose Bayesian over classical approaches is probably because it becomes possible to formally include knowledge from previous assessments (of species/stocks other than that of current interest) in a new assessment. Hilborn and Liermann (1998) argue that using data for well-studied species to inform data-poor species can be considered to be "standing on the shoulders of giants".

In their assessment of Pacific Ocean perch off the United States west coast, Ianelli et al. (2000) developed informative prior distributions for several of the key parameters of their model. In particular, the prior for the steepness of the stock-recruitment relationship was taken from that developed by Dorn (2002) for rockfishes off the west coast of north America while that for survey catchability was also determined from a meta-analysis (Ianelli et al. 2000).

Liermann and Hilborn (1997) introduced hierarchical meta-analysis to fisheries assessment by conducting a meta-analysis of the impact of depensation at low stock size. In common with tabling estimates of a quantity for which a prior distribution is needed, hierarchical modeling is a technique that can be used to combine data from several independent sources (species/stocks) and represent the outcome in the form of a probability distribution for the quantity of interest. The basic idea is that each species/stock for which data are available has a different value for the quantity of interest but that species/stocks are interchangeable in the sense that the value of the quantity of interest for any given stock can be considered to be a random selection from an underlying distribution (which is the same for all species/stocks after account is taken of explanatory covariates). Hierarchical meta-analysis has now been used to examine the steepness of the stock-recruitment relationship (Myers et al. 2002, Dorn 2002), the maximum rate of increase at low population size (Myers, Bowen and Barrowman 2002), the relationship between catch and abundance (Harley, Myers and Dunn 2001), survey catchability and selectivity (Harley and Myers 2001, Millar and Methot 2002) and carrying capacity (Myers et al. 2001).

The use of priors based on data for other species/stocks is not, however, without controversy. Reasons for this include the representativeness of the species for which data are available. For example, the bulk of the information related to the steepness of the stock-recruitment relationship is for three species Clupea harengus, Gadus morhua, and Melanogrammus aeglefinus. The posterior distributions for steepness for rockfish species derived by Dorn (2002) are centred at lower values than those determined from the stock and recruitment data for all species. Had assessments for rockfish species been based on priors derived from stock and recruitment data for all species rather than just those for rockfish species, the estimates of productivity would have been biased upward. The issue of representativeness is perhaps more problematic for deepwater species as these species tend to be data-poor (so the results of Bayesian assessments are affected more by the choice of the prior distributions) and because it appears that deepwater species are less productive than species found on the shelf and in inshore waters. One consequence of this is that assessment scientists conducting analyses for species off the US west coast have tended to avoid the use of priors based on meta-analyses and have instead made use of priors that are uninformative.

4. BROADING ASSESSMENT GROUP COMPOSITION

The move towards attempting to incorporate as many data types as are available into assessments and to identify a range of alternative models rather than only one requires that assessments be conducted by groups of individuals rather than by a single assessment scientist. Each of the four regions considered in this review has assessment processes that are open to stakeholder groups in various ways. Of the four regions, stakeholder groups (which include managers, industry and conservation groups) probably have the greatest input into assessments conducted for the Australian Fisheries Management Authority (AFMA - the agency that manages fisheries resources on behalf of the Commonwealth Government of Australia) (Smith, Smith and Punt 2001).

In common with assessment groups in the US, New Zealand and South Africa, the primary role of the assessment groups established by the AFMA is to assess the stock or species in relation to its particular management objectives, describe the management implications, and identify the research and monitoring necessary to improve the assessment. These assessment groups do not make management recommendations nor do they attempt to reduce the number of alternative models to only one.

The AFMA assessment groups generally include a broad range of stakeholder groups. The interaction among the participants (although hard to quantify) is one of the major advantages of the process (Smith, Sainsbury and Stevens 1999). It assists with communication and usually leads to trust and mutual respect when what are often conflicting parties are dealing with difficult issues (Smith, Smith and Punt 2001). By participating in the assessment process, industry and managers gain a conceptual understanding of stock assessment and modeling. However, the assessment itself can also be improved through insights by industry on, for example, trends in fishery-dependent data. This is particularly important because stock assessment scientists spend a decreasing amount of time in the field. For some assessments in Australia, several of the hypotheses considered are based on suggestions by industry (Smith, Smith and Punt 2001).

The need for assessment groups to include broad participation of stakeholders is perhaps greatest for situations in which considerable reliance is placed on fishery-dependent data. For example, the assumptions that selectivity and catchability are constant over time is a standard assumption when conducting assessments. Violation of these assumptions can lead to markedly biased results. However, these assumptions are only likely to be examined in detail when there is evidence in the form of auxiliary information that they are false, even though the results of meta-analyses suggest, for instance, that CPUE is unlikely to be related linearly to abundance (Harley et al. 2001). The experience of the author is that the presence of industry participants when conducting assessments is more likely to highlight reasons for changes over time in selectivity and catchability.

The need for a broad range of participants on assessment groups is even greater if the objective of the assessment is to provide the basis for an evaluation of harvest strategies. This is because harvest strategies have to be tested using models that represent ‘the full range of uncertainties’ pertinent to the fishery in question (Butterworth and Punt 1999) and having a large assessment group with broad participation and a diverse background is likely to promote discussion of a wide range of possible factors that might influence the dynamics of the fish population, its supporting ecosystem and the fisheries for it.

5. CONCLUDING REMARKS

This paper has focused on the assessments for deepwater species. However, many of the problems encountered when assessing deepwater species are also common to assessments of pelagic and inshore species, namely lack of appropriate data and understanding. As a result, many of the solutions outlined above are used when assessing data-poor pelagic and inshore species.

There are, however, some monitoring and assessment issues that appear to be specific to deepwater species. For example, the cost of conducting research and monitoring activities in deepwater habitats is substantially greater than in inshore habitats. Some of methods that can be applied relatively straightforwardly to shallow-water species (such as tagging studies and submersible-based indices of abundance) are either more difficult (and costly) or currently infeasible in deepwater habitats.

Assessments are currently conducted for those species that are the most important commercially. The costs associated with data collection in deepwater habitats means that any research targeted towards commercially less important species will yield less precise information and hence to imprecise assessments. There is, therefore, a need to look towards monitoring tools that are able to provide assessment-related information for a wide suite of species rather than just the major target species. This points, for example, to the need to collect length, age and abundance data for all species during surveys rather than for just a few target species. Given the likely lack of substantial amounts of monitoring data, the careful use of the results of meta-analyses is likely to be necessary for many species. One example of this would be to apply the posterior distribution estimated for the catchability coefficient for trawl surveys of rockfish off the US west coast to estimate the biomass of species for which data are not yet sufficient to base an assessment on. Further, the results from the meta-analysis of Liermann and Hilborn (1997), that depensation is more common in fish stocks than previously believed, means that assessments should regularly incorporate this as a possibility rather than ignoring it, which implicitly assumes that there is no probability of depensation at low stock size.

Many deepwater species are long-lived, slow-growing and relatively unproductive. Estimates of the maximum fraction of the exploitable component of an orange roughy population that can be harvested sustainably has been estimated to be less than 3 percent of that in a virgin state (Francis 2001). This suggests that, a priori, assessments of deepwater species should assume that productivity is likely to be low. This can be incorporated into assessments through the selection of prior distributions that give greater a priori weight to low productivity than high productivity scenarios. If the data suggest that productivity is higher than implied by the assumed priors, this will be reflected in the posteriors as data accumulate slowly. However, in the absence of data suggesting higher productivity, the assessment results will be based primarily on scenarios which imply lower productivity.

This paper points to the following as being features of assessments of deepwater species in the future.

• Several alternative models will be included in assessments to attempt to capture alternative hypotheses for the population dynamics and the relationships between the data and the underlying dynamics; hypotheses regarding space and spatial structure will be emphasized to a much greater extent than at present.

• Only a very few of the model parameters will be pre-specified based on analyses of auxiliary information - rather most of the model parameters will be estimated within the assessment.

• Uncertainty will be quantified by means of Bayesian posterior distributions rather than by point estimates and the sensitivity of the results to changes to data weightings and the choice of priors will be employed commonly.

• Prior distributions will be developed for most of the parameters of typical population dynamics models based on meta-analyses. Assessments which are data-poor will therefore be "shrunk" towards a priori expectations rather than assessed using overly simple (and often optimistic) models/assumptions.

6. ACKNOWLEDGEMENTS

Tony Smith (CSIRO Marine Research) is thanked for comments on an earlier version of this paper.

7. LITERATURE CITED

Annala, J.H., K.J. Sullivan, C.J. O’Brien, N.W. Smith & S.M. Grayling (Comps) 2003. Report from the Fishery Assessment Plenary, May 2003: stock assessments and yield estimates. (Unpublished report held in NIWA library, Wellington.)

Anon. 1995. Guidelines on the precautionary approach to capture fisheries and species introduction. FAO Fisheries Technical Paper No. 350/1.

Butterworth, D.S. & A.E. Punt 1999. Experiences in the evaluation and implementation of management procedures. ICES Journal of Marine Science 56: 985-998.

Butterworth, D.S., G.S. Hughes & F. Strumpfer 1990. VPA with ad hoc tuning: implementation for disaggregated fleet data, variance estimation, and application to the Namibian stock of Cape horse mackerel Trachurus trachurus capensis. South African Journal of Marine Science 9: 327-357..

Deriso, R.B. 1980. Harvesting strategies and parameter estimation for an age-structured model. Canadian Journal of Fisheries and Aquatic Sciences 37: 268-282.

Deriso, R.B., T.J. II, Quinn & P.R. Neil 1985. Catch-age analysis with auxiliary information. Canadian Journal of Fisheries and Aquatic Sciences 42: 815-824.

Dorn, M.W. 2002. Advice on West Coast rockfish harvest rates from Bayesian meta-analysis of stock-recruit relationships. North American Journal of Fisheries Management 22: 280-300.

Francis, R.I.C.C. 1992. Use of risk analysis to assess fishery management strategies: A case study using orange roughy (Hoplostethus atlanticus) on the Chatham Rise, New Zealand. Canadian Journal of Fisheries and Aquatic Sciences 49: 922-930.

Francis, R.I.C.C. 2001. Stock assessment for 2001 of orange roughy on the northeast Chatham Rise. New Zealand Fisheries Assessment Report 2002/41.

Francis, R.I.C.C. &R. Shotton 1997. "Risk" in fisheries management: A review. Canadian Journal of Fisheries and Aquatic Sciences 54: 1699-1715.

Francis, R.I.C.C., P.L. Cordue & V. Haist 2002. Review of the 2001 hoki stock assessment. New Zealand Fisheries Assessment Report 2002/42.

Fournier, D. & C.P. Archibald 1982. A general theory for analyzing catch at age data. Canadian Journal of Fisheries and Aquatic Sciences 39: 1195-1207.

Fournier, D.A. & I.J. Doonan 1987. A length-based stock assessment method utilizing a generalized delay-difference model. Canadian Journal of Fisheries and Aquatic Sciences 44: 422-437.

Harley, S.H. &R.A. Myers 2001. Hierarchical Bayesian models of length-specific catchability of research trawl surveys. Canadian Journal of Fisheries and Aquatic Sciences 58: 1569-1584.

Harley, S.H., R.A. Myers &A. Dunn 2001. Is catch-per-unit-effort proportional to abundance? Canadian Journal of Fisheries and Aquatic Sciences 58: 1760-1772.

Hamel, O.S., I.J. Stewart & A.E. Punt 2003. Status and future prospects for the Pacific Ocean Perch resource in waters off Washington and Oregon as assessed in 2003. Pacific Fishery Management Council, 7700 Ambassador Place NE, Suite 200, Portland, OR 97220.

Hilborn, R. 1990. Estimating the parameters of full age-structured models from catch and abundance data. Bulletin of the International North Pacific Fisheries Commission. 50: 207-213.

Hilborn, R. & M. Liermann 1998. Standing on the shoulders of giants: learning from experience in fisheries. Reviews in Fish Biology and Fisheries 8: 273-283.

Hilborn, R. & C.J. Walters 1992. Quantitative Fisheries Stock Assessment: Choice, Dynamics and Uncertainty. Chapman and Hall, New York

Ianelli, J.N., M. Wilkins & S. Harley 2000. Status and future prospects for the Pacific ocean perch resource in waters off Washington and Oregon as assessed in 2000. In: Status of the Pacific coast groundfish fishery through 2000 and recommended acceptable biological catches for 2001. Pacific Fishery Management Council, Portland, OR.

Liermann, M. & R. Hilborn 1997. Depensation in fish stocks: a hierarchic Bayesian meta-analysis. Canadian Journal of Fisheries and Aquatic Sciences 54: 1976-1984.

Magnusson, K.G. and G. Stefansson 1989. A feedback strategy to regulate catches from a whale stock. Reports of the International Whaling Commission (Special Issue 11): 171- 189.

Maunder, M.N. 2001. Integrated Tagging and Catch-at-Age Analysis (ITCAAN): Model development and simulation testing. In G.H. Kruse, N. Bez, A. Booth, M.W. Dorn, S. Hills, R.N. Lipicus, D. Pelletier, C. Roy, S.J. Smith & D. Witherell (Eds.) Spatial Processes and Management of Marine Populations Alaska Sea Grant Program. 123-146.

McAllister, M.K. & G.P. Kirkwood 1998. Bayesian stock assessment and policy evaluation: A review and example using the logistic model. ICES Journal of Marine Science 55: 1031-1060.

McAllister, M.K. and C.H. Kirchner 2001. Development of Bayesian stock assessment methods for Namibian orange roughy Hoplostethus atlanticus. South African Journal of Marine Science 23: 241-264.

Methot, R.D. 1990. Synthesis model: an adaptable framework for analysis of diverse stock assessment data. Bulletin of the International North Pacific Fisheries Commission. 50: 259-277.

Methot, R.D. 2000. Technical Description of the Stock Synthesis Assessment Program. U.S. Dep. Commer, NOAA Tech. Memo, NMFS-NWFSC-43.

Millar, C. & R.D. Methot 2002. Age-structured meta-analysis of U.S. West Coast rockfish (Scorpaenidae) populations and hierarchical modeling of trawl survey catchabilities. Canadian Journal of Fisheries and Aquatic Sciences 59: 383-392.

Myers, R.A., K.G. Bowen & N.J. Barrowman 1999. The maximum reproductive rate of fish at low population sizes. Canadian Journal of Fisheries and Aquatic Sciences 56: 2404-2419.

Myers, R.A., N.J. Barrowman R. Hilborn, & D.G. Kehler 2002. Inferring Bayesian priors with limited direct data: Applications to risk analysis. North American Journal of Fisheries Management 22: 351-364.

Myers, R.A., B.R. MacKenzie, K.G. Bowen & N.J. Barrowman 2001. What is the carrying capacity for fish in the ocean? A meta-analysis of population dynamics of North Atlantic cod. Canadian Journal of Fisheries and Aquatic Sciences 58: 1464-1476.

Pope, J.G. & J.G. Shepherd 1985 A comparison of the performance of various methods for tuning VPAs using effort data. Journal du Conseil International pour l’ Exploration de la Mer. 42: 129-151.

Punt, A.E. 1994. Assessments of the stocks of Cape hake Merluccius spp. off South Africa. South African Journal of Marine Science 14: 159-186.

Punt, A.E. & R. Hilborn 1996. Biomass dynamic models. User’s manual. FAO Computerized Information Series (Fisheries). No. 10. Rome, FAO.

Punt, A.E. & R. Hilborn 1997. Fisheries stock assessment and decision analysis: A review of the Bayesian approach. Reviews in Fish Biology and Fisheries 7: 35-63.

Punt, A.E., A.D.M. Smith & G. Cui 2001. Review of progress in the introduction of management strategy evaluation (MSE) approaches in Australia’s South East Fishery. Marine and Freshwater Research 52: 719-726.

Punt, A.E., D.C. Smith, R.B. Thomson, M. Haddon, X. He & J.M. Lyle 2001. Stock assessment of the blue grenadier Macruronus novaezelandiae resource off south-eastern Australia. Marine and Freshwater Research 52: 701-717.

Punt, A.E., A.D.M. Smith & G. Cui 2002. Evaluation of management tools for Australia’s South East Fishery. 2. How well do commonly-used stock assessment methods perform? Marine and Freshwater Research 53: 631-644.

Restrepo, V.R., J.M. Hoenig, J.E. Powers, J.W. Baird & S.C. Turner 1992. A simple simulation approach to risk and cost analysis, with applications to swordfish and cod fisheries. Fishery Bulletin, US. 90: 736-748.

Sainsbury, K.J., A.E. Punt & A.D.M. Smith 2000. Design of operational management strategies for achieving fishery ecosystem objectives. ICES Journal of Marine Science 57: 731-741

Schnute, J.T. 1985. A general theory for analysis of catch and effort data. Canadian Journal of Fisheries and Aquatic Sciences 42: 414-429.

Schnute, J.T., L.J. Richards & N. Olsen. 1998 Statistics, software, and fish stock assessment. In F. Funk, T.J. Quinn II, J. Heifetz, J.N. Ianelli, J.E. Powers, J.F. Schweigert, P.J. Sullivan & C.-I. Zhang (Eds.) Fishery Stock Assessment Models Alaska Sea Grant Program. 171-184.

Shepherd, J.G. 1999. Extended survivors analysis: An improved method for the analysis of catch-at-age data and abundance indices. ICES Journal of Marine Science 56: 584-591.

Smith, A.D.M. & A.E. Punt 1998. Stock assessment of gemfish (Rexea solandri) in eastern Australia using maximum likelihood and Bayesian methods. In F. Funk, T.J. Quinn II, J. Heifetz, J.N. Ianelli, J.E. Powers, J.F. Schweigert, P.J. Sullivan & C.-I. Zhang (Eds.) Fishery Stock Assessment Models Alaska Sea Grant Program. 245-286.

Smith, A.D.M., A.E. Punt, S.E. Wayte, P.J. Starr, R.I.C.C. Francis, T.K. Stokes, R. Hilborn & A. Langley 2002. Stock Assessment of the Northeast Chatham Rise Orange Roughy for 2001. New Zealand Fisheries Assessment Report 2002/25.

Smith, A.D.M., Sainsbury, K.J. & R.A. Stevens 1999. Implementing effective fisheries management systems - management strategy evaluation and the Australian partnership approach. ICES Journal of Marine Science 56: 967-979.

Smith, D.C., A.D.M. Smith & A.E. Punt 2001. Assessment and process for stock assessment in the South East Fishery, Australia: a perspective. Marine and Freshwater Research 52: 671-681.

Walters, C.J. & P.H. Pearse 1996. Stock information requirements for quota management systems in commercial fisheries. Reviews in Fish Biology and Fisheries 6: 21-42.