2 IMPLICATIONS FOR STOCK ASSESSMENT OF TUNAS AND ASSOCIATED SPECIES

2.1 Description of possible fishing management regimes and the implications for stock assessment research

The goal of any fishing management regime is to control fishing mortality rates in order to achieve management objectives and to maintain the fishery within management constraints. The particular objectives and constraints chosen by managers imply that certain research and monitoring activities be addressed so that scientists may evaluate the effectiveness of the management and predict the likelihood of alternative future management outcomes. One subset of these objectives and constraints includes those associated with Precautionary Approaches of management. These impose some specific research needs for stock assessments and monitoring.

2.1.1 Control rules

One way to characterize precautionary management regimes is through control rules. A control rule describes a variable over which management has some direct control as a function of some other variable(s) related to the status of the stock. In other words, the control rule represents a pre-agreed plan for adjusting management actions, depending on the condition of the stock. In broad terms, the management actions may be designed as strategies to achieve a variety of socio-economic and conservation objectives through imposing (1) a particular exploitation rate (to harvest a specified fraction of the stock each year), (2) a particular escapement (e.g., to maintain a specified spawning stock size), or (3) a particular catch limit. A control rule does not have to adhere strictly to any of these three strategies: managers may choose to have a mixture of these strategies, or they may prefer control rules that achieve different results, depending on the condition of the stock.

In the case of Precautionary Approaches to management, however, there are some specific requirements that are typically imposed. There is the need for criteria for determining the overfished status of the stock (i.e. a stock level under which the stock would be classified as overfished) and criteria for specifying overfishing (i.e. a rate of fishing that would eventually reduce the stock to an overfished state). Once these limits have been established by management, a control rule might, for example, be a feedback rule in which fishing mortality rate is a declining function of stock biomass as that biomass approaches or falls below some overfished level.

Control rules can be defined in terms of limits (conditions to be avoided) and in terms of targets (objectives to be achieved). It is important, however, to acknowledge the role of uncertainty in developing both limit and target control rules. Controls (or, indeed, any management action) cannot be imposed perfectly. There are estimation errors in the status of the stock and in the overfishing or overfished criteria themselves. There are also errors in how broad scientific measures, such as fishing mortality rate, are translated into actual fishing effort or into allowable catches. There are also natural variations in the population dynamics of the stock. In addition, there are implementation errors when regulations are not fully imposed or fully adhered to. Hence, there are practical difficulties in defining targets and limits: one would expect that due to these sources of variation measured status would be above the target half the time and below it half the time. Thus, if the limit is defined as being too close to the target, then there may be an unacceptably large probability of exceeding the limit, just due to the variation associated with trying to achieve the target. In any case, though, the responsibility of the research and assessment process is to evaluate these uncertainties and to make the most appropriate determinations of the probability that the targets are being achieved and that the limits are being avoided. Precautionary actions may be taken by management, based upon its understanding of the risks and implications of these probabilities.

Precautionary limits and targets are a component of the overall management objectives and constraints. Indeed, there are multiple objectives for management, such as maximization of catches, stability of catches over time, equitable distribution among users, and maintenance of market supplies. There will always be tradeoffs between objectives and constraints, including those associated with precautionary criteria, which should be evaluated. Also, Precautionary Approaches per se do not eliminate the possibility of developing new fisheries or new entrants into existing fisheries. If the research and assessment is expected to respond quickly and appropriately to these events, however, then there is an obligation to collect data to do that. In particular, there is a need for timely information in a developing fishery, so that management can respond as status measurements evolve.

While ad hoc approaches to management may be effective when applied by judicious managers, they limit the usefulness of scientific evaluation: one cannot predict in advance how effective management will be relative to the limits and targets, and effects of the decisions may be cumulative. In its guidelines, the UN Fish Stocks Agreement states "reference points shall be used to trigger pre-agreed conservation and management action." This does not need to be interpreted strictly as management requiring a control rule approach, as implied by FAO´s 1996 technical guidelines on the Precautionary Approach to Capture Fisheries and Species Introductions: These guidelines state "scientific evaluation of management options requires specification of operational targets, constraints and decision rules. If these are not adequately specified by managers, then... analysis requires that assumptions be made about these specifications, and that the additional uncertainty resulting from these assumptions be calculated. Managers should be advised that additional specification of targets, constraints and decision rules are needed to reduce this uncertainty." The scientific role in developing rules is one of continual monitoring, re-evaluation, and testing. The scientific knowledge evolves, management objectives change in accordance with the political context, and socio-economic factors are dynamic.

Testing should include performance testing through simulations using an appropriate range of alternative hypotheses. Typically this is done by creating an operating model, i.e. a simulation of the underlying population and fishery dynamics and the fisheries data that are derived from it. Then this simulated fishery system is assessed with the stock assessment model that is being used. Predictions of performance measures are made, assuming the management control rules are implemented. Many iterations are carried out, using alternative inputs and hypotheses for the operating model. For example, a test of an alternative hypothesis might be: how well does the rule perform when the underlying natural mortality rate is different than that assumed? It is important that the operating model be created in sufficient detail and that relevant alternative hypotheses be examined. Otherwise, the test results may be misleading.

The implicit scientific obligations to precautionary management approaches are to determine status relative to limits and targets, to predict outcomes of management alternatives for reaching the targets and avoiding the limits, and to characterize the uncertainty in both of these.

2.1.2 Recovery plans

Recovery plans are required when limits have been exceeded and the stock is in an overfished state. Recovery plans are essentially schedules of actions to improve the stock so that an overfished limit is no longer being exceeded and that subsequently the target is being achieved. Implicit in a recovery plan, in addition to the targets and limits, is specification of the recovery period and the recovery trajectory.

Despite having a well-planned schedule and trajectory, recovery is not guaranteed. Ecological shifts in communities or oceanographic variability may limit both the productivity and the potential of a stock to achieve the recovery goal. Under present knowledge, these sorts of dynamics are not predictable. Thus, there is inherent uncertainty in the trajectories.

The duration of recovery is the time until the status measure, for example the spawning biomass, increases above the limit and on toward the target. Specification of the desired duration of a recovery period is a management prerogative. However, there are some scientific constraints. The recovery period should be long enough to allow an acceptable probability that the status measure(s) exceed the rebuilding limit and target, given the productivity of the stock. If the period is too short, recovery may not be feasible even with no fishing. Basic life history information for, for example, temperate versus tropical tunas, coupled with stock projections with zero fishing, should provide guidance on this. Long-lived species have less annual growth potential, and this should be noted in specifying recovery periods. Biological information on stock productivity should define whether a recovery period is infeasible.

It is more difficult to supply scientific criteria when a recovery period is too long. Suffice it to say that if the period is very long, then the risk of a stock being in an overfished condition becomes increasingly less certain due to uncertainties about future recruitment, and science cannot offer much guidance.

Recovery trajectories are essentially the specification of the interim management objectives during recovery. For example, do managers want to have large immediate reductions in catch to increase the probability of more rapid stock increases, or do they want to phase in reductions with concomitant slower short-term recovery? There are an infinite number of combinations on recovery themes that managers may supply.

In addition, it is important that the timing of these interim milestones is concordant with the basic life history and population dynamics of the fish and with the scientists' ability to measure changes. For example, if the interim milestone were scheduled every three years for a temperate tuna, then that may imply that there should be a research investment in the monitoring of trends in abundance of juveniles in order to detect the response to management.

An essential responsibility of stock assessment research with respect to recovery plans is the determination of feasible trajectories, the evaluation of the uncertainty under various management trajectory scenarios, and the re-evaluation of performance as the plan progresses.

2.1.3 Communication

Communication of the results of the stock assessment analyses to managers, fishermen, and others is an important step in the Precautionary Approach.

When actually preparing communications, one will inevitably have to make tradeoffs on these three principles, according to the audience receiving the results, particularly in regard to the details of transparency. Obviously, not all members of the audience will want to know all the intricacies of an analysis. However, they should be confident that such detail is available and that the detailed reports can be reviewed. Additionally, reports should convey the sources of uncertainty that are included within the analysis and those that are not. This suggests that communication will be in multiple forms, including reports and briefings, to name just two. These principles should provide a basis for communication.

2.1.4 Research links

The implementation of precautionary management approaches suggests important research links with stock assessment processes. It is important that the fisheries be monitored in terms of evaluation of management performance, incorporating evolving changes in management objectives and evaluating management performance relative to interim milestones in recovery stocks. Underlying all of these is the need for measurable limits and targets, the need to characterize uncertainty and to guide the investment priorities to reduce that uncertainty, and the need for operational testing of both estimation models and management schemes.

The nature of the stock assessment approaches is discussed in the subsequent section.

2.2 Stock assessment models

Assessment models represent the essential framework to integrate the information coming from different sources and to produce estimates of reference points and current statuses of the stocks. Furthermore, they have been used for many years for many tunas and tuna-like fishes and, therefore, many advantages and shortcomings have been examined.

2.2.1 Characteristics of tunas and tuna-like fishes as they relate to stock assessment models

Some of the characteristics of tunas and billfishes pose specific complications to the monitoring and assessment of their stocks. Descriptions of the biological and environmental characteristics of these resources are given in Sections 2.4 and 4.2, and other features are discussed below.

Catch rates: Because the fisheries evolve rapidly, utilize different gears, and take place in different areas, catch rates tend to provide snapshots of the relative abundance of particular size ranges in particular areas at particular times of the year. Factors that affect the vulnerability of the fish to capture, such as schooling behaviour, vertical movement, and variations in oceanic conditions, also complicate interpretation of catch rates. Putting together global indices of abundance, using fishery-independent surveys, would, in most cases, be prohibitively expensive.

Fleet dynamics: The mobility of industrial fleets allows them to change their areas of operation as needed to maximize their profits. They often develop new fishing strategies and tactics, such as deployment of new types of gear, modifications of current types of gear or they way the gear is used, and changes in the patterns of cooperation among fishermen. It is often difficult to account for or assess the effects of these changes.

Life history: In addition, there are differences among tunas that affect the stock assessments. In temperate tunas (see Section 2.4.1) the long life span, combined with a late maturity, often causes a lag in reaction time of a stock to management measures taken.

2.2.2 Model types

In terms of data requirements and outputs, a useful distinction can be made between age-aggregated and age-structured models. These two categories comprise the vast majority of assessment models currently used. Age-aggregated or production models characterize the response of an aggregated property of the population, typically biomass, to exploitation. Age-structured models contain an internal representation of the age-structure of the population and how it progresses over time in response to exploitation. More recent extensions of both types of models have included spatial structure and environmental heterogeneity. These approaches are promising, but require detailed information on the movements of the fish.

There a relationship between the type of models applied to the assessment of a species, the management targets and constraints, the amount and types of data available, and the characteristics of the biology or the fisheries associated with that species. For example, the difficulty in determining age in skipjack tuna and the fact that it has extended spawning period complicate the application of age-structured models for these stocks and, therefore, production models have more often been used.

These models differ in a significant way in their data requirements and the outputs that they provide. Production models require knowledge about the removals from the population and one or more measures of relative abundance. Typically, a time series of several years of catch and effort data is used. The parameters of these models can be easily cast in terms of useful reference points, such as the coefficient of fishing mortality, F_msy, and biomass, B_msy, corresponding to the maximum sustainable yield.

Age-structured models require, in addition to information on the catches of the different fisheries involved, data on the size composition of the catch, some way of converting the size data to age data, and one or more indices of abundance. Typical outputs include a matrix of abundance at age (or size) and one or more vectors of selectivities at age. In this case, the estimation of the MSY-related reference points is not as direct as in the case of some production models, and it requires some additional assumptions the about stock-recruitment relationship. However, as will be discussed later, the simplicity of the production models comes at the price of more restrictive structural assumptions.

An assumption common to both types of models is that some or all of the parameters measuring productivity of the population are stable over time. This stability assumption, which may often be violated, affects the ability of the assessment to predict future outcomes, e.g. in projections from age-structured models.

Production models

Production models have a long history of application in tuna fisheries, primarily because they typically require only catch and effort data and because of the easy translation of its parameters to quantities relevant to management. Some of the first tuna stock assessments were done using these models.

These models, however, do not always approximate well the responses of a population to exploitation. For example, production models do not incorporate the time lag associated with transient states in the age composition of the population when the exploitation rate changes. This problem has a greater effect on the analyses of species with long life spans, and could be alleviated by the application of variations of these models, such as delay-difference or age-structured production models that address this problem.

Perhaps more critical are the stability assumptions implicit in most production models. For example, they usually assume a vector of age-specific selectivities that is invariant in time. It is important to note that, in this context, selectivity should be considered the result of not only gear selectivity, but also of particular fishing strategies (i.e. the choice of fishing areas and/or seasons). This stability assumption can be frequently violated in tuna fisheries in which there have been significant changes in fishing strategies, such as the development of a FAD-associated fishery).

A similar problem occurs when a regime shift in environmental conditions affects the productivity of the population, compounded by the difficulty in detecting such shifts.

In these cases the problems might be reduced by introducing a time series structure in the relevant parameters (e.g. carrying capacity or catchability), allowing them to change slowly with time. Parameter estimation may require the use of auxiliary environmental data.

Age-structured models

This category of models can be further divided into models that contain a statistical error structure (e.g. separable models) and those that do not exhibit this feature (e.g., virtual population analysis (VPA) models). Models with a statistical error structure facilitate the estimation of some of the components of the uncertainty surrounding an assessment and carrying that uncertainty throughout the analysis.

The estimation procedure in models with a statistical error structure is based on the comparison between values predicted by the model and values observed in the fisheries.

A common practice in age-structured models is to aggregate the oldest age categories into a "plus group" because of the difficulties in discriminating age classes solely from the length distributions. The number of age classes thus grouped can be significant when the population has a long life span and most of the individual growth takes place early in the life of the fish. In some cases, the dynamics of the plus group cannot be well represented by the model, as when the age-specific selectivities are changing within the plus group. In these cases direct ageing is a valuable tool in disaggregating the plus group, thus avoiding this problem.

Age-specific natural mortality rates may be explicit input parameters required in age-structured models. In general, there is little information with which to estimate these, and this can be a major source of uncertainty in assessment advice.

Length-frequency data might be more informative with respect to growth at a spatial scale smaller than the entire stock range. In other cases, incorporation of spatial structure makes it possible to investigate assumptions regarding stock heterogeneity, a key issue for highly-migratory species.

2.2.3 Measures of abundance

An important requirement of many of the analytical tools discussed above is the availability of a relative or absolute measure of abundance or fishing mortality. Again a useful distinction can be made between the measures of abundance derived from data available on the operations of the fisheries (fishery-dependent indices) and those obtained from fishery-independent data.

Although there are many reasons to prefer a fishery-independent measure of abundance, in practice, most current stock assessments of tunas depend on indices derived from catch and effort data. This is due, in part, to the prohibitively large cost of obtaining measures of abundance that sample adequately the population ranges of tunas and tuna-like fishes. Therefore, statistical methods for standardization have been developed to address such things as spatial structure to the extent that the data allow.

Fishery-dependent measures

The principal assumption behind the classic use of catch-per-unit-of effort (CPUE) as an index of abundance is that there is a linear relationship between standardized CPUE and the abundance of the stock. The many ways in which this assumption can be violated have been discussed and documented. For example, if the degree of mixing of the population were low compared with the exploitation rate in a given location, the CPUE would reflect a local depletion rather than the global relative abundance of the population. Additionally, dynamics of the fishing fleet can alter the degree of sampling coverage.

An important and frequent problem is that unquantified increases or decreases in catchability can lead to an overly optimistic or pessimistic appraisals of the trends in population size. To some extent, standardization of the effort alleviates some of these problems but, unfortunately, the information required to effectively estimate these changes in fishing power is rarely available. One alternative is to replace the assumption of constant catchability in the assessment models by the assumption that the catchability changes slowly over time and allow those changes to be estimated. The incorporation of additional information, such as tagging data or effective fishing power, can help in these structural time-series approaches. In another approach, alternative scenarios can be considered under the assumption of a non-linear relationship between CPUE and abundance.

For example, the changes in fishing strategies brought about by the FAD-associated purse-seine fisheries will require careful consideration of alternative measures of effort, possibly on the basis of new data, in order to obtain meaningful indices of abundance. In another example, the role of fishing depth in the analysis of longline fisheries will also require more analysis. Furthermore, scientists may be unaware of some technological changes due to the proprietary nature of the activity. A good working relationship with the industry is important in this respect.

Fishery-independent measures and scientific experiments

Perhaps the most important source of a fishery-independent measure of abundance for tunas and tuna-like fishes is tagging data, although, strictly speaking, it also depends on the fisheries for the recovery of the tags. Tagging can produce many kinds of useful information, such as estimates of absolute abundance, natural and fishing mortality, and growth, and data on movements of the fish. It is important to consider carefully the design of an experiment in order to maximize its benefits. For example, there are definite statistical advantages to tagging fish of several cohorts, and tagging fish of each cohort more than once during the period of its existence in the population. It is preferable to carry out the analysis of tagging data within the framework of an integrated assessment model.

The use of tagging data requires knowledge about tag shedding and reporting rates and the rate of mixing of the tagged and untagged components of the population. Tag shedding can be estimated through double-tagging experiments and reporting rates by tag-seeding studies, although the value of the latter may be questionable. The assumption of mixing becomes less restrictive if spatial structure is incorporated into the model, so that mixing is required only within each spatial stratum. Large numbers of fish released at the same location at the same time are sometimes recaptured by one vessel a few days later, before they have had time to mix with the untagged fish in the vicinity. In such cases it may be necessary to disregard data for fish that were at liberty only a short time.

Recently, there have been technological advances in tag design, such as archival tags and pop-up tags. At this stage the information that such tags provide is more useful for generating hypotheses on tuna behaviour, including movement, for the purpose of interpreting catchability or alternative model structures than it is for providing estimates of fishing mortality or measures of abundance.

Fishery-independent surveys, e.g. research-vessel and aerial surveys, do not often provide useful information, mainly because of the characteristics of tunas and tuna-like fishes. The extended range of these populations, even when they aggregate for spawning, means that the surveys can cover only a representative fraction of the total area of distribution. In some cases, they can provide only minimum estimates of abundance. Nevertheless, more recent model developments allow the use of measures of some portions of the stock, such as particular age groups or a specific area. Thus, partial surveys can still be useful.

In any case, it is necessary to ensure that appropriate consideration is given to the design of the survey. It is also necessary to consider whether the reduction in uncertainty to be expected from the survey justifies the costs associated with it.

2.2.4 Stock-recruitment relationships and projections

Assumptions about stock-recruitment relationships are particularly critical to the ability of projections to produce useful predictions. Unfortunately, characterizing the uncertainty in model structure and model parameters in the prediction of future recruitment is, in practice, very difficult. For example, the assumption of stability in the stock-recruitment relationship might be violated if there is a major shift in environmental regimes. The occurrence of such shifts usually cannot be predicted, and they are difficult to detect from data. Similarly, changes in the age-specific selectivities can occur as a result of changes in the pattern of exploitation by the fisheries involved.

In some cases, the short time series of data available for estimating a stock-recruitment relationship affects the perception of what could be a reasonable function involved, as the range of observed spawning stocks might be too restricted.

Projections are a standard tool in extending the results of the models to assess the probabilities of various outcomes within specified periods. They require estimates of the most recent population structure and some assumptions about future recruitment and the evolution of the fishery.

These difficulties result in projections that become increasingly unreliable as the length of the period for which the predictions are made increases. In most cases, therefore, only short-term projections are likely to be reliable, as they will depend less on the assumed stock-recruitment relationship.

Projections under certain types of management scenarios might be more variable than others. For example, constant catch scenarios might result in more variable projections because they do not include consideration of future gains in information in light of management responses.

2.2.5 Empirical methods

It should be noted that the modelling framework described above has been developed for a situation for which adequate data are available to estimate a set of reference points with a specified level of confidence. Stocks could be characterized as data-rich, data-moderate, or data-poor, based on the type, amount, and quality of the data and the degree to which they are informative about stock dynamics.

In data poor-situations, the United Nations Convention on the Law of the Sea Relating to the Conservation and Management of Straddling Fish Stocks and Highly Migratory Fish Stocks guideline indicated that "provisional reference points shall be set," and that "the fishery shall be subject to enhanced monitoring so as to enable revision of provisional reference points as improved information becomes available." In such cases empirical methods, such as indices of trends or stability of the catches, trends in CPUE, average size, or age or length composition may be useful indicators of stock condition. Empirical methods, by their nature, however, rely on strong assumptions about the stocks and the fisheries that are not always explicit. These assumptions should be expressed and evaluated. Simulation-based performance evaluations are recognized as useful tools to examine empirical status indicators, helping to judge their robustness, representativeness, and reliability. They may also help to determine sources of uncertainties and prioritize the research needs. It should be noted, however, that the performance of simulation-based models depends largely on model structures and selection of hypotheses.

2.3 Ecological and environmental issues in stock assessments

The environmental and ecological factors affecting tunas and tuna-like fishes (including by-catch issues) are discussed in Sections 4.3 and 4.4. For purposes of stock assessments, there is a need for operational definitions of objectives for ecosystem and/or by-catch management.

2.3.1 By-catches and stock assessment

Assessments of by-catch species in fisheries directed at tunas and tuna-like fishes are difficult for several reasons. Although single-species methods are still applicable, there are practical limitations because of the lack of data. In order to have the data to perform single-species assessments of by-catch species, the level and kind of sampling and, indeed, the information on the general life history patterns of the organisms being sampled would have to be similar to those being conducted for the tunas and tuna-like fishes. Nevertheless, estimating trends in the CPUEs may be useful for monitoring changes in the abundances of by-catch species.

The precision and accuracy of standard fisheries models is conditional, to some extent, on fishing mortality being a significant proportion of total mortality. If a by-catch species is rare, or if it is abundant, but occurs infrequently in the catch, then the applicability of standard fisheries models may be questionable. Fisheries-independent surveys may be useful for estimating relative abundance, but they would usually be expensive.

2.3.2 Environmental issues in stock assessments

Environmental impacts on the dynamics of tuna stocks are known to occur in all regions. These may be on seasonal, interannual, or decadal time scales, and may affect such biological processes as recruitment, natural mortality, growth, movement, and vulnerability to capture. Depending on the complexity of the stock assessment model being used, such effects might be detected specifically on such processes, or in aggregate form as changes in productivity or carrying capacity.

It is possible that environmental effects on shorter time scales could be incorporated to some extent as process error in stock assessment models. Where such effects are significant but not accounted for, the ensuing model mis-specification may result in more variable and/or biased results.

Long-term environmental variation that results in a significant change in stock productivity may require a direct management response. For example, where there has been a long-term, environment-driven change in recruitment, a biomass-based limit reference point that is expressed as a certain percentage of the average virgin biomass may require re-calibration to the current average recruitment levels. If such re-calibration does not occur, the management targets may be unrealistic under the changed stock conditions.

As research develops on the relevant factors important to ecosystem management (see Section 4.4) and management goals for the ecosystem are specified, modelling research will still fit into the control-rule framework of monitoring, evaluating performance, and revising of the models. Whatever form these goals might take, for the foreseeable future the modelling is likely to generate empirical approaches for evaluating progress.

2.4 Uncertainty analysis

The need to provide analyses of the uncertainty associated with the provision of scientific advice is embedded in the Precautionary Approach. In addition, the guidelines in the UN Fish Stocks Agreement for the application of the Precautionary Approach require assessments of the risk of exceeding reference limit points. As such, it is not sufficient to simply provide single "best "estimates of the status of a stock with respect to reference points and predictions about the future consequences of management actions, as estimates of the associated uncertainty, e.g. variances and potential biases, are needed.

The complexity arises because of the variety of data sources, the numerous inputs required, and the limits of knowledge about the underlying process.

2.4.1 Sources of uncertainty

Uncertainty has two components, accuracy and precision. The following is one way of characterizing the uncertainty that leads to estimation error in stock assessments:

• Observation error

  - Measurement error (e.g. uncertainty associated with individually-measured quantities, such as weight or length, or weight of the catch);

  - Sampling error (e.g. uncertainty associated with having obtained data from a sample of the catch or of the fleet);

• Modelling error

  - Input parameter error (e.g. uncertainty associated with parameters, such as natural mortality, required in the assessment models);

  - Model mis-specification error (e.g. uncertainty associated with the underlying model structure and assumptions, such as the form of the stock-recruitment relationship or assumptions of independence).

There are two issues that should be highlighted with respect to uncertainty created during the step that goes from the specification of a management control rule to its implementation in the fishery.

First, it is possible that some of the conditions assumed for the assessment will change upon implementation of management. For example, the assessment may have assumed a certain aggregate age-specific selectivity in recommending the management, but during the process of allocation of total allowable catches the resulting selectivity is changed. The assessment will then have to "play catch-up," by incorporating the changed selectivity in the future. The bias in the assessments will be related to the magnitude of the change and the frequency with which assessments are made. It may be possible to assess this bias and to optimize the periodicity of assessments with a simulation model that incorporates the totality of the stock, fishery, data collection, assessment, and management system.

Second, it is possible that certain management controls, such as individual transferable quotas, may compromise the accuracy of the data (e.g. unreported dumping of lower-value products), resulting in bias in future assessments and jeopardizing the effectiveness of the current management strategy. Implementation uncertainty should be considered in the development and evaluation of control rules. Enhanced monitoring will be required in these situations in order to maintain the integrity of the assessments (see Section 5). The data requirements for stock assessment may be categorized as those relating to measurement and sampling errors, those related to parameter uncertainties, and those related to model uncertainty.

2.4.2 Data required to quantify and reduce uncertainty

Each of these different sources of uncertainty has implications for research in terms of minimizing the overall uncertainty and for ensuring that the uncertainty associated with stock assessments can be adequately quantified.

Measurement and sampling errors

In situations for which the measurement errors may be significant, estimates of its magnitude and any potential biases should be obtained. Similarly, it is important to be able to model the errors associated with any sampling program.

When some aggregation of the data is unavoidable, it is preferable to maintain as fine a scale of resolution as possible. It is always possible to further aggregate the data, but once aggregated, they cannot be disaggregated.

Parameter uncertainties

The main input parameters in stock assessments are biological ones, such as natural mortality rates, weight-length relationships, and age at maturity.

In cases for which these data are not used directly in the stock assessment model, e.g. the likelihood function can be partitioned, the parameter values, e.g., weight-length relationships, can be estimated independently of the stock assessment. In such cases, not only the expected value, but also the statistical distributional properties, must be available for the assessments. In other cases, it may be important to integrate available data and the estimates of the biological parameters, e.g. growth rates, directly into the assessment models. In such cases the raw data may be required. In either case, the assessments should acknowledge the fact that biological parameters can vary over time and space, and data are needed to assess such changes.

Model uncertainty

Model uncertainty is hardest to account for in most assessments. The problem is that there is generally a paucity of information to formulate and distinguish between competing underlying assumptions about the dynamics. Stock assessment models are simplifications that attempt to represent the dominant factors underlying the dynamics of the stock. The complexity that should be introduced in the model will depend upon the complexity of the system and the scale at which model estimates and predictions for management are needed.

2.4.3 Quantification of uncertainty

The main approaches that have been used to quantify uncertainty in fishery stock assessments are sensitivity analyses, Monte Carlo simulation approaches, and Bayesian analyses. There are advantages and limitations to each of these approaches. Regardless of the approach used, the quantification of the overall uncertainty of an estimate conceptually involves a four-step process:

1. For each potential source of error, a set of hypotheses must be developed;

2. For each hypothesis in Step 1, a relative weight or probability must be determined;

3. For all combination of hypotheses in Step 1, the likelihood (fit) of the resulting estimate must be determined;

4. The results from Steps 2 and 3 must be integrated.

Trade-offs will exist between the completeness of the assessment, the magnitude of the problem, the complexity of the task, the required resources, and the judgements on which uncertainties must be considered. Balancing these different trade-offs will be essential in any particular application in order to be able to provide a meaningful and feasible assessment of the uncertainty. In this regard, some elaboration on each of these four steps is provided as Appendix 2.

The implementation of these four steps requires considerable judgement, so it is recommended that the procedures and the bases upon which the decisions are made be fully documented.

There are a variety of approaches involving various degrees of rigor, formality, and complexity that can be used to implement each of these steps. There is a need for additional research on approaches for incorporating the results from both residual and retrospective analyses into the evaluation of overall uncertainty. In general, quantification of uncertainty will ultimately entail complex models and procedures because of the large model, data, and parameter uncertainties associated with fish stock assessments. It is desirable to limit the complexity to the extent possible and to develop approaches that can screen out redundant or implausible alternatives, with an overall aim to provide a realistic appraisal of the uncertainty.

The judgments of different experts about the appropriate sets of alternative hypotheses and about their relative weights will differ. For this reason, it is important to attempt to harmonize and integrate the different judgements.

A comprehensive approach to the quantification of uncertainty is likely to be extremely computationally intensive. If such a process is to be embraced, it is essential that sufficient resources, scientists, and time be allocated to the task.

This would have substantial implications for how stock assessments are conducted by several RFBs. Scientific assessments are frequently accomplished by bringing together the data and initial assessments at a working scientific meeting. Then, based on the discussions at that meeting, a final set of calculations is generated and reviewed prior to the end of the meeting. With the comprehensive approach to quantify uncertainty, either substantially longer meetings would be required, or the process would have to be broken into a multi-stage process. If the time allotted to complete the process is too long, however, this could introduce the possibility that substantial amounts of new catch and effort data might become available before the process is completed.

2.5 Alternative sources of information

Much of the information required for stock assessment will be generated either directly from the fisheries or by sampling the catches. Fishery-independent surveys and scientific experiments (e.g. tagging), when they are feasible, may also provide information on the dynamics of the stocks, and thus make important contributions to reduction of assessment uncertainty. Other sources of potentially valuable information are discussed below.

2.5.1 Adaptive learning through probing

In the case of data-poor and newly-developing fisheries, an overly cautious management approach may hinder "learning" about the resource. The use of an active adaptive management approach^[2] that incorporates "probing" experiments and/or conducting some of the fishing effort in accordance with a survey design may provide a means of increasing the utility of the data collected. If such an approach is used, the following conditions should be met:

1. An operating model is used to design the experiments or surveys to maximize the "learning" value of the data generated;

2. Adequate monitoring regimes are in place to collect the data in a timely fashion;

3. There is rapid assimilation of new information into the operating model, and adaptive management responses occur as learning proceeds;

4. In the case of "probing" experiments involving the infusion of high effort over a designated period of time, there is a management capability to subsequently reduce or re-distribute fishing effort as indicated by the results of the experiment;

5. There is industry consultation and cooperation in the design and implementation of the experiments or surveys.

2.5.2 Industry participation

The involvement of industry in the assessment process has proven to be a useful means of obtaining additional information in many situations. Industry participation has been useful in proposing alternative hypotheses regarding various biological processes, assisting in the interpretation of fisheries data, and planning data collection activities. An additional benefit of industry involvement has been a greater acceptance of the assessment results and the management actions that flow from them.

Some of these include:

1. The technical aspects of fishing gear operation and the interaction of fishing gear with fish populations, leading to better quantification of fishing effort, catchability, and size selectivity;

2. The evolution of fishing technology over time, leading to a better understanding of changes in fishing power and, hence, effective fishing effort;

3. Fishing strategies of the various fleets and the factors that influence them, leading to better quantitative models of fishing effort dynamics;

4. Innovative fishing technological approaches to by-catch mitigation.

Industry participation and cooperation in studies such as these would be crucial for their success.