© 2004 by ICES/CIEM International Council for the Exploration of the Sea/Conseil International pour l'Exploration de la Mer
Possible solutions to some challenges facing fisheries scientists and managers
School of Resource and Environmental Management, Simon Fraser University Burnaby, British Columbia, Canada V5A 1S6
*Correspondence to R. M. Peterman: tel: +1 604 2914683; fax: +1 604 2914968. e-mail: peterman{at}sfu.ca.
The purpose of this paper is to review recent work on four key challenges in fisheries science and management: (1) dealing with pervasive uncertainties and risks; (2) estimating probabilities for uncertain quantities; (3) evaluating performance of proposed management actions; and (4) communicating technical issues. These challenges are exacerbated in fisheries that harvest multiple stocks, and various methods provide partial solutions to them: (i) risk assessments and decision analyses take uncertainties into account by permitting several alternative hypotheses to be considered at once. (ii) Hierarchical models applied to multi-stock data sets can improve estimates of probability distributions for model parameters compared with those derived through single-stock analyses. (iii) Operating models of complete fishery systems provide comprehensive platforms for testing management procedures. (iv) Finally, results from research in such other disciplines as cognitive psychology can facilitate better communication about uncertainties and risks among scientists, managers, and stakeholders.
Keywords: Bayesian, decision analysis, hierarchical model, Kalman filter, operating model, risk assessment, risk communication, uncertainty
Received 5 January 2004; accepted 5 June 2004.
 |
Introduction |
|---|
 Top Introduction Challenges and some possible...ConclusionReferences |
|---|
Fisheries scientists and managers face many significant challenges; here are four.
- Uncertainties and the risks they create are pervasive owing to natural variability in components of aquatic ecosystems, imperfect information about those components, and lack of perfect control over fisheries.
- It is difficult to estimate probabilities for the uncertain factors in stock assessments.
- Fisheries scientists who provide advice to managers must comprehensively take into account uncertainties and risks in their analyses of management options.
- Scientists must communicate complex and technical results effectively to decision makers and the public.
This paper has two purposes. First, it elaborates on these challenges facing fisheries scientists and managers. Second, it describes some potential solutions to each challenge by reviewing recent research. Although most examples here are from Pacific salmon (Oncorhynchus spp.) fisheries, the lessons learned are applicable to other species.
|
Challenges and some possible solutions |
|---|
|
TopIntroductionChallenges and some possible...ConclusionReferences |
|---|
Challenge 1: uncertainties and risks are pervasive
To put the challenges facing fisheries scientists and managers into context, consider a typical fishery system (Figure 1). The natural aquatic system is sampled by scientists and harvesters, and the resulting data are used by stock assessment scientists to estimate abundance, productivity, recruitment, and other attributes of a stock. The resulting model is used to estimate how several potential management actions, such as various harvest rates or enhancement activities, might affect outcomes. Scientists then provide stock assessment advice to fisheries managers and interested parties (stakeholders), ideally with some iterative feedback. Managers consider their management objectives along with input from stock assessment scientists and stakeholders before recommending a particular action, such as a harvest rate, which then affects the natural system.
|
Such fishery systems contain numerous sources of uncertainty (ellipses in Figure 1). Five are:
- the natural variability across space and time in distribution, abundance, and productivity of fish populations;
- observation error (i.e. imperfect information), which arises from measurement error as well as sampling error (Mace and Sissenwine, 2002);
- the difficulties associated with communication among scientists, managers, and stakeholders about technical scientific information and its associated uncertainties;
- unclear management objectives;
- implementation error, which is the difference between a management goal and the actual realized spawning-stock biomass or fishing mortality rate, for example.
These uncertainties can be large and can affect interpretation of data, the results of analyses, the rank orders of management options, and the effectiveness of those options. Uncertainties are therefore important because they create risks: biological risks for fish populations, economic losses for those in the fishing industry, and social disruptions for people in fishing-dependent communities. Uncertainties are pervasive and occur in all fishery systems to varying degrees. Consequently, most decisions in fisheries management should take uncertainties into account. This applies to decisions not only concerning harvest regulations, but also the design of monitoring schemes and activities such as ocean ranching or other attempts to increase abundance of fish stocks.
Potential solutions to Challenge 1
That stock assessments can account for uncertainties is well known; the challenge of pervasive uncertainties has been met by increasingly sophisticated technical tools. For instance, it is no longer widely acceptable to provide scientific advice to managers on possible consequences of management actions based only on best-fit, or point, estimates of current stock biomass and productivity parameters of stocks. Stock assessments in many regions now routinely take several sources of uncertainties into account quantitatively [National Research Council (NRC), 1998; Quinn and Deriso, 1999]. This includes assessments made for the stocks investigated through the International Council for the Exploration of the Sea (ICES), an organization in which growing emphasis on conservation concerns and application of a precautionary approach (FAO, 1995) has led many analysts to estimate probabilities that stock indicators will cross reference points (Lassen and Sparholt, 2000). Further, the European Commission is actively encouraging policy-orientated research that takes uncertainties into account and includes risk assessments. Risk assessments and decision analysis have also been particularly useful for evaluating a broad range of management options in the context of uncertainties (Francis and Shotton, 1997; McAllister and Kirkwood, 1999).
Risk assessment
To avoid misunderstandings, fisheries scientists, managers, and stakeholders should always clearly state what they mean by the term "risk". Technically, it has two components, the magnitudes of adverse consequences that will arise from events that are uncertain, and the chances (i.e. probabilities) of those events and their consequences occurring. "Risk assessment" (i.e. risk analysis) refers to the general process of estimating both components of risk, not just one or the other.
As yet, there is no standardized risk-assessment procedure for fisheries situations, let alone ecological systems in general, although various broad frameworks for the latter have been developed (Power and McCarty, 2002). This lack of standardized methodology is partly because those that have been developed were mainly derived in the early 1990s from specialized procedures for estimating human health risk from toxic chemicals [National Research Council (NRC), 1993]. Therefore, ecological risk assessment is a relatively new field and methods are continually evolving (and can be found within the pages of such journals as Risk Analysis and Human and Ecological Risk Assessment).
Generally, a typical risk assessment in fisheries management includes five components. (i) First is a management objective, which often includes factors such as expected catch, variation in catch over time, and probability of the spawning biomass or other variable crossing a limit reference point (a condition to be avoided). Indicators are identified to measure how well a management objective is expected to be met. (ii) Several management options are considered for achieving the objective. (iii) A stochastic risk-assessment model of system processes includes (iv) a wide range of quantified hypotheses about those processes, i.e. different parameter values or structural forms of relationships among variables. (v) Uncertainties are taken into account by weighting these alternative hypotheses and their consequences by the degree of belief in them or their probability of occurrence. The model then estimates the probability distribution of outcomes or other indicators for each proposed management action. Risk assessments are usually linked to further analyses, such as the procedure for decision analysis described next.
Decision analysis
It is not sufficient for decision makers merely to see a description and quantification of uncertainties and risks. They want to know how uncertainties and risks affect the ability of each potential management option to meet particular management objectives. To provide this information, scientists often conduct formal quantitative decision analyses (Walters, 1986; Clemen, 1996; Peterman and Anderson, 1999), which add three new components to the five already mentioned for risk assessment: (vi) a decision tree or decision table to help structure the analysis and communicate its content, (vii) a ranking of management options that results from conducting the decision analysis, and (viii) extensive sensitivity analyses to show decision makers the effects of changing various assumptions on that rank order of management options. In this context, procedures for risk assessment can be thought of as a subset of decision analysis (Figure 2).
|
Decision analysis has several advantages over standard approaches to decision making. First, by taking uncertainties into account explicitly, decision analysis often indicates that the best management option for meeting an objective will be different from that recommended by a simpler analysis based only on point estimates of parameters and state variables, i.e. an option that ignores uncertainties (Reckhow, 1994; Frederick and Peterman, 1995). For example, Robb and Peterman (1998) found that the adult abundance estimate for returning Nass River (British Columbia, Canada) sockeye salmon (O. nerka) that was optimal for opening an upstream First Nations fishery was 40 000 fish when only point estimates of model components were used. In contrast, when a decision analysis was conducted that took into account uncertainty in both the structural form of the stock-recruitment relation as well as its parameters, that optimal abundance tripled to 120 000 fish. The main reason for a decision-analysis result being different from the deterministic analysis is that, in fisheries systems, losses associated with deviating from an optimal state are usually asymmetric (e.g. where loss in long-term value of catch is higher for a spawning biomass that is 50% below some desired level than if it is 50% above). Similarly, probability distributions for uncertain quantities are often asymmetric. Given either of these conditions, it usually becomes optimal to choose an action that "hedges" away from the higher potential losses (Reckhow, 1994). When decision makers consider political, economic, and social pressures, the final recommended action may or may not still hedge in this direction.
A second benefit of decision analysis is that it can include various structural forms of models as alternative hypotheses in a single analysis. This is important, because mis-specification of a model's components (compared with the real-world situation) may produce inaccurate estimates of outcomes, and yet we usually do not know the model's correct specification. A significant point is that decision analysis does not require scientists, stakeholders, or others to agree on which single model should be used in analyses of management options. Instead, several alternative models can be included.
Such alternatives create two important limitations of decision analysis; analysts must choose not only which alternative models are legitimate and necessary for inclusion, but also they must assign a probability to each model. Such choices may influence the rank order of actions, particularly in cases where data are relatively uninformative about the alternative models (i.e. unable to distinguish among them) and where different models lead to considerably different predictions. As yet, there is no definitive answer to this problem. This is a complex topic beyond the scope of this paper, but Punt and Hilborn (1997) and McAllister and Kirchner (2002) provide useful advice.
Perhaps the most extensive example of a decision analysis in fisheries management is a recent evaluation of recovery plans for seven depleted populations of spring and summer chinook salmon (O. tshawytshca) from the Snake River sub-basin of the Columbia River system in the northwestern United States (Peters and Marmorek, 2001). Those stocks were listed under the US Endangered Species Act. Adult and juvenile fish migrate through several reservoirs and dam systems, and also face problems from nearby agricultural lands, harvesting, hatcheries, predation, and changing ocean conditions. Large uncertainties about the various factors that contributed to reduced stock abundance over several decades led to contentious debates about interpretations of data and which salmon population model was most appropriate for evaluating management options designed to achieve their recovery (Marmorek and Peters, 2001)1. A decision-analysis framework defused some of the debate by including multiple hypotheses and models in one analysis along with uncertainties in them (Peters and Marmorek, 2001).
The decision analysis was aimed at identifying acceptable actions to be implemented by the US National Marine Fisheries Service. One example of a quantitative management objective (the "recovery" objective) was to find a management action with at least a 50% chance of having six of the seven Snake River stocks exceed their respective desired target spawner abundances during the last 8 of the next 48 years (Peters and Marmorek, 2001). Other management objectives considered had a similar format. This approach (using the top six of seven stocks) assigned priority to the best-off stocks, while recognizing that there was some non-zero probability that the recommended action would not be successful for all stocks. This approach to structuring a multi-stock management objective may be useful in other fisheries.
A decision tree reflects some key elements of the Snake River chinook salmon problem (Figure 3). These elements dealt largely with uncertainties in data and hypotheses about mechanisms operating during downstream freshwater migration by juveniles, as well as delayed mortality effects in the ocean. The uncertainties, hypotheses, and outcomes were incorporated into a stochastic simulation model. A wide range of weightings for alternative hypotheses were evaluated and one of the management options (A3), removing the four lower Snake River dams, was the highest-ranked and most robust option after extensive sensitivity analyses were conducted across a wide range of assumptions (Peters and Marmorek, 2001).
|
This example for the endangered Snake River chinook salmon stock also illustrates that decision analysis is a useful framework for focusing efforts of members of a diverse multi-stakeholder team, and taking into account their sometimes strongly differing views about hypotheses and uncertainties (Marmorek and Peters, 2001).
Three recent examples also illustrate the benefits of decision analysis; the first two apply to problems in the ICES region. Kuikka et al. (1999) explored how environmental uncertainties affect recruitment and growth of Baltic cod (Gadus morhua), which in turn affect the optimal mesh size for managing that fishery. That decision analysis demonstrated that increased mesh size would reduce the probability of stock collapse and also meet other management objectives. That method of analysis has been accepted formally within the ICES region as a basis for scientific advice to managers for Baltic cod. Another decision analysis on Baltic cod also showed that reduced fishing mortality was necessary to lower the probability of stock collapse substantially; this result was robust to various assumptions about the structure of the model (Jonzén et al., 2002). Finally, Punt et al. (2002) used decision analysis extensively to evaluate harvesting options for Australia's multi-stock, multispecies Southeast fishery.
Clearly, risk assessment and decision analysis are useful methods for dealing systematically with some of the uncertainties and risks facing fisheries scientists and managers. Alternative hypotheses can be incorporated into a single analysis, numerous uncertainties can be taken into account explicitly, and the rank order of management options can be identified under a variety of assumptions through sensitivity analyses. Actions may emerge that are robust to such a range of assumptions.
Despite these benefits, risk assessment and decision analysis cannot resolve all issues related to uncertainties and risks. The methods have serious limitations in addition to the two already mentioned (choosing which alternative models/hypotheses to include, and how to weight them). First, where a management objective includes conflicting components, such as maximizing catch but minimizing the chance of stock collapse, decision analysis cannot indicate the action that makes the best trade-off between those components; managers must make that trade-off. The only exceptions are if those components are expressed in common units such as utility, or if one or more components imposes an absolute constraint. A second limitation arises from deciding on the variables and processes to include in an analysis. Because the omitted processes likely also have associated uncertainties, any risk assessment or decision analysis can lead to overconfidence in the results. This is also true of any other quantitative approach. Third, risk assessments and decision analyses are difficult to describe, especially to non-technical people who use the results. Finally, not only is expertise in these methods limited, so is time. There may not be enough expertise or time to apply advanced methods as part of annual stock assessments, which already consume considerable time and effort. To help deal with time constraints, comprehensive risk assessments and decision analyses could be most useful for developing pre-agreed state-dependent and time-independent management procedures (control rules, management strategies) that are intended to be in place for a considerable period before being re-evaluated (McAllister and Kirkwood, 1999). This topic is expanded upon later under Challenge 3.
Risk management
Risk management is the process in which decision makers "manage the risks" by selecting a particular action, or set of actions. They do so after taking into account scientific advice from a risk assessment, decision analysis, or stock assessment, as well as other factors not considered explicitly in those analyses (Figure 2). Because of these "other factors", risk management is not a purely scientific process; it involves subjective judgements about compromises or trade-offs. There is no scientifically "correct" weighting for catch, year-on-year variation in catch, or probability that a fish stock will fall below a biomass limit reference point. Nevertheless, results from risk assessments can indicate how much of one of these indicators will be lost for a given gain in another, under each management option.
Clear communication is critical to risk management. To improve the efficiency and effectiveness of decision making, and to ensure that all scientific information is understood, there needs to be an iterative, two-way flow of information among people responsible for risk analysis, decision analysis, and risk management, as well as between these people and stakeholders.
Challenge 2: estimating probabilities for uncertain quantities
Another challenge for fisheries scientists is to estimate probabilities, or degrees of belief, for parameters that are considered uncertain. We can
- directly calculate probabilities from a lengthy data set, such as annual water levels in a river,
- use expert judgement, or
- use the data available along with Bayesian methods to produce a posterior probability distribution (Ellison, 1996; Punt and Hilborn, 1997).
There are difficulties with all three approaches. The first is not commonly used because lengthy data series for uncertain components are rare in fisheries. Even where such series exist, questions arise about relevance of very old data because underlying processes can change. The second approach, seeking expert opinion, is used widely. However, such elicitations of expert judgements are well known by cognitive psychologists to be subject to bias, and incorrect estimates of precision are attributable to many factors (Morgan and Henrion, 1990, p. 102). For instance, if a question is ambiguous concerning the exact quantity about which an opinion is being sought, each expert in a group might think about a different location, season, life stage, etc., when giving their opinion. This would make the distribution wider than it should be and might also bias it. An unambiguous question will ensure that experts' responses reflect only uncertainty about the parameter's value, rather than uncertainty about which entity the parameter represents (the "clarity test" of Morgan and Henrion, 1990, p. 50).
The third approach for describing uncertainties in stock assessments, using available data in conjunction with Bayesian statistical methods, is increasingly used, but it is far from widespread [National Research Council (NRC), 1998]. A prior probability distribution can be combined with a likelihood distribution derived from data, and the resulting posterior probability distribution can quantify the degree of belief in different values of some parameter. Such posterior probabilities can then be used in a risk analysis and decision analysis to weight various hypotheses about the value of the parameter. One major challenge about this third, or Bayesian, approach to describing uncertainties is to find means of deriving defensible prior probability distributions. When data are not too informative about an uncertain parameter owing to data paucity, low contrast, or large natural variability and observation error, for example, the shape of the posterior probability distribution is greatly affected by the choice of the prior probability distribution (Ellison, 1996). This can have important management implications. For example, if the resulting posterior probability distribution is too narrow, it may underestimate the probability of extreme cases that lead to deleterious conservation outcomes. This general problem is acute for relatively unproductive stocks that are a conservation concern (Rivot et al., 2001); such stocks typically have relatively few data and there is a potentially high cost of incorrectly estimating the probability of decline or recovery of a stock. For this reason, some researchers argue that, given relatively non-informative data, it is most appropriate to use a completely non-informative prior probability distribution for a parameter to avoid biasing the posterior (Walters and Ludwig, 1994; Punt and Hilborn, 1997). Others argue for using independent biological information where it is available to create an informative prior (e.g. a narrow normal distribution; McAllister et al., 1994, 2001). In situations with data that are very informative about the uncertain quantity, the nature of the prior is relatively unimportant, because the narrow likelihood will dominate the calculation of the posterior.
Potential solutions to Challenge 2
A hierarchical model is a quantitative tool that can help produce defensible informative priors through use of large sets of data on multiple populations. Rather than assuming that each population's parameter values are statistically independent from those of other populations, hierarchical models allow for some underlying structure or pattern in the parameters. For instance, all stocks of a given species might be assumed to have a maximum reproductive rate that is drawn from a single normal probability distribution that has a mean and variance (e.g. Myers et al., 1999). Such models are hierarchical in the sense that each population's value of some uncertain parameter, such as parameter ai of the Ricker (1975) stock-recruitment model [Equation (1) below], represents a sample from a distribution that is described by unknown parameters, which must also be estimated. Hierarchical models include mixed-effects models (fixed and random effects) estimated by classical or Bayesian methods. Such models have proven very useful for combining information across multiple populations of the same species, as well as across species, or across years (Liermann and Hilborn, 1997; Myers et al., 1997, 1999, 2001; Adkison and Su, 2001; Myers, 2001; Su et al., 2001; Chen and Holtby, 2002; Mueter et al., 2002). Those analyses identified generally consistent and limited ranges of values for particular parameters, such as maximum annual reproductive rates across fish species worldwide (Myers et al., 1999), and a narrow range of coefficients reflecting the effects of summer sea surface temperature on survival rates of Pacific salmon populations (Mueter et al., 2002; Su et al., 2004). In the absence of other information, such results are useful either for establishing prior probability distributions for parameters to be used in Bayesian updating, or for directly specifying posterior probability distributions or weightings to be used in decision analyses.
To illustrate the usefulness of hierarchical models, consider the common problem of estimating parameters of fisheries models in the presence of large natural environmental variation, which tends to mask parameter values. To the extent that multiple stocks share common environments, they should show similar responses to environmental variation. Indeed, numerous studies have documented positive covariation among stocks in particular variables across several hundreds of kilometres, and as stocks become increasingly separated, the correlation approaches zero. Such studies cover a variety of species and locations, including Pacific herring (Clupea pallasaii; Ware and McFarlane, 1989); North Sea and North Atlantic fish (Shepherd et al., 1984), Baltic salmon (Salmo salar; McKinnell and Karlström, 1999), Northeast Pacific sockeye salmon, pink salmon (O. gorbuscha), and chum salmon (O. keta; Peterman et al., 1998; Pyper et al., 2001, 2002), and numerous other marine and freshwater species (Myers et al., 1997). Hierarchical models take advantage of such situations by attributing some of the observed variation to responses that are shared among stocks (i.e. are common to them), thus permitting more precise estimates of model parameters. Such models therefore "borrow strength or information" from stocks with similar parameter values.
An example for 43 pink salmon stocks in the Northeastern Pacific Ocean demonstrates the benefit of applying a hierarchical Bayesian model (HBM). The pink salmon stocks with ocean entry points for seaward-migrating juveniles that are less than 
500 km apart show positive covariation in residuals from their stock-specific, best-fit Ricker stock-recruitment models (Pyper et al., 2001). In our hierarchical Bayesian analysis of pink salmon, this positive covariation among stocks permitted stocks to be treated as "statistical replicates" when a model was fitted; this tended to average out observation errors across stocks (Su et al., 2004). We used a generalized Ricker model:
|
| (1) |
i the coefficient reflecting the effect of summer sea surface temperature (SSTit) in the region where each stock's juveniles spend their first four months in the ocean, and 
it is the residual variation. We used spatially correlated prior distributions to reflect possible regional similarity of the stock-specific ai and i parameters (Su et al., 2004).
This multi-stock hierarchical Bayesian model gave more precise estimates of the ai and i parameters than separate analyses of each stock (Figure 4; Su et al., 2004). These narrower posterior probability distributions permit improved estimates of biological reference points that are affected by these parameters, because some of the environmentally induced variation in productivity has been better accounted for than in single-stock analyses. The probability distributions can also be used to weight different combinations of parameter values in decision analyses.
|
Therefore, although multi-stock situations normally create problems for scientists and managers (perhaps caused by simultaneous harvesting of several stocks with different productivities) in situations where several stocks respond similarly to some variable, hierarchical models can improve stock assessment information. Such models provide a consistent method for estimating informative prior probability distributions that are more precise for particular parameters than if populations were analysed separately.
Nevertheless, hierarchical models and Bayesian approaches are not panaceas; they have limitations too. A hierarchical model will not be beneficial for estimating parameters unless they are at least somewhat similar across data sets. For instance, quantities such as a salmon stock's unfished equilibrium abundance may differ considerably between even nearby stocks as a consequence of human-induced or natural differences in freshwater habitat. In such a case, the uncertainty about any single stock's unfished equilibrium abundance will be reduced little, if any, by applying a hierarchical model that uses data from other nearby stocks. In addition, the hierarchical modelling approach will not be particularly advantageous when there are only a few data sets on different populations. Also, although estimates of parameters from hierarchical Bayesian models usually have lower mean-squared errors (are statistically "more efficient") than estimates from separate Bayesian analyses of single data sets (Gelman et al., 1995), the performance of such hierarchical estimators needs to be evaluated with "operating models" (see Challenge 3 below) to generate numerous realizations of multiple data sets in order to determine the performance of the Bayesian estimates.
Bayesian methods for estimating probabilities for uncertain quantities also have limitations. Complex fish stock assessment models that contain numerous uncertain parameters require sophisticated and computationally intensive techniques. In such situations, joint and marginal posterior probability distributions can be estimated with sampling-importance-resampling (SIR) algorithms (Rubin, 1988) or Markov Chain Monte Carlo (MCMC) methods (Gelman et al., 1995). The latter can be used by employing relatively new software (e.g. WinBUGS; Spiegelhalter et al., 1999). However, problems can result from the nonlinear nature of most fisheries models, in which estimates of two different parameters may be correlated, for instance. This can cause convergence problems for MCMC methods. In such cases, careful re-parameterization of the model is necessary (e.g. Meyer and Millar, 1999). Care should be taken to choose the appropriate scale for non-informative priors in fisheries models. For example, parameters such as 
in lognormal models should have a flat prior on the log() scale (Gelman et al., 1995; Millar, 2002). It is often difficult to find non-informative priors, especially for variances of random effects in hierarchical models. Therefore, sensitivity analyses must also be conducted with different priors to determine the extent to which they affect the posterior distribution.
Finally, all fisheries models, including hierarchical Bayesian ones, will result in incorrect estimates of parameters and inappropriate management advice if the models assume (as most do) that parameters are constant over time when they in fact vary temporally. For instance, body size and stock productivity can change with climate, so models should reflect these possibilities. State-space models (Chatfield, 1989) can deal with time-varying parameters, and many fisheries scientists have used this approach, most commonly by applying a Kalman filter (Collie and Sissenwine, 1983; Walters, 1986; Mendelsohnn, 1988; Pella, 1993; Schnute, 1994; Millar and Meyer, 2000; Peterman et al., 2000, 2003). However, state-space models have been limited to relatively simple fisheries situations.
Challenge 3: evaluating the performance of management options
Even if the challenges described above are met, scientists and managers will still be uncertain about the management option that will best meet a given management objective or set of objectives, owing to complex interactions and feedbacks among components of fishery systems (Figure 1).
Potential solutions to Challenge 3
Given a clearly stated objective, simulations can be done to evaluate relative performances of proposed management options. Although fisheries scientists routinely conduct stochastic simulations, the most comprehensive method to evaluate options is to simulate the entire feedback system shown in Figure 1 (not just part of it) using an "operating model" (Linhart and Zucchini, 1986). Such models are analogous to flight simulators; the latter include detailed dynamic feedback processes to help pilots determine which decision-making protocols are best in the presence of a wide range of possible, but uncertain, simulated contingencies. Similarly, operating models of fisheries typically aim to identify robust decision-making rules by simulating six components:
- the stochastic dynamics of a "true" (i.e. assumed) natural population;
- samples of indices of abundance or other data from that "true" population, including observation error;
- where needed, a stock assessment step that uses those sampled data to update annual estimates of state variables and parameters;
- state-dependent harvest control rules that specify the effect of either sampled indices or stock assessment estimates on the choice of management actions;
- implementation of those actions, including error; and
- the effect of those realized actions on the "true" population (Punt, 1992; de la Mare, 1996, 1998; Sainsbury, 1998).
Normally, operating models are used to explore numerous situations and structures for the first four components above. For example, a wide range of alternative hypotheses about scenarios for the "true" population are considered routinely in successive simulations. The specific nature and form of components (ii), (iii), and (iv) (collectively often called a management procedure or strategy; de la Mare, 1996) can be varied across runs of the operating model to determine the best combination of sampling procedure (e.g. sampling methods and times/places to sample), types of models and parameter-estimation methods (e.g. constant or time-varying parameters, maximum likelihood or Bayesian updating), and harvest control rules (functional forms and parameter values). The final result is usually a relative ranking of management procedures based on those that are most robust to a wide range of conditions. Just as with decision analysis, sensitivity analyses should also be conducted to indicate how that ranking changes with different management objectives.
The fifth component of an operating model, implementation error, is extremely important. It is the deviation between a desired state and the actual realized outcome (Rosenberg and Brault, 1993; Rice and Richards, 1996). This error typically arises from a combination of non-compliance with regulations by harvesters, changing catchability, and other dynamic processes in the fishing fleet. Implementation error can be a large source of variation, yet it is rarely included in the exploration of management options in stock assessments. Such omission leads to overconfidence in the effectiveness of proposed management actions. To rectify this situation, analysts can include implementation error by employing a stochastic harvesting process in an operating model based on historical data or hypotheses.
Such comprehensive operating models of entire fishery systems provide a strong test of robustness of management options (Cooke, 1999). An excellent and early example of operating models to evaluate management procedures was the International Whaling Commission's (IWCs) development of the Revised Management Procedure (RMP; IWC, 1994; de la Mare, 1996; Kirkwood, 1997). The IWCs analyses explored many shapes of functions for the harvest control rule while taking into account uncertainties in estimates of whale abundance. Those analyses also examined model performance under numerous combinations of uncertainty in stock identity in multi-stock fisheries and temporal trends in abundance of whales resulting from environmental changes or interactions with other species. The harvest control rule that performed best for whales was robust to these sources of uncertainty (de la Mare, 1996). Operating models have also been applied in many other situations [see Smith, 1993; Butterworth and Punt, 1999, and several papers published in Volume 56(6) of the ICES Journal of Marine Science; Peterman et al., 2000]. European Commission and ICES scientists are currently actively developing operating models for a wide variety of fisheries in the ICES region to derive robust management procedures (including harvest control rules; Kell et al., 1999). The general conclusion from past work on this topic is that such comprehensive simulations of sources of uncertainties provide different recommendations to decision makers than if only a subset of those uncertainties were analysed.
Nevertheless, operating models can be formidable to implement. Not only do they require advanced expertise, but more processes must be included than in most standard stock assessment models. These processes include a description of potential sampling errors, which are usually unknown. Intangible or other factors that are not simulated may strongly influence the performance of some control rule when it is applied "in the field", leading to overconfidence in the results of the operating model. Further, time constraints often preclude analyses with operating models.
Challenge 4: communication
Communication among scientists, managers, and stakeholders is another source of uncertainty or error that influences fishery systems (Figure 1). Stock assessment, risk analysis, and decision analysis are highly technical endeavours. It is difficult to convey assumptions, results, and implications effectively to people who are not actively involved in the analyses. Many sources of communication problems are obvious, but some are more subtle. As an example of the latter, cognitive psychologists who conduct research on how people reason about uncertainties and risks have found widely different intuitive interpretations of such seemingly straightforward terms as "probability". Teigen (1994) found that people interpret "probability" in six different ways. It can reflect:
- the chance of seeing a given outcome for a stochastic process (which most fisheries scientists would intend in a stock assessment context);
- the tendency or ease with which some event is perceived to occur (if a stock size has been low recently, people may perceive a higher tendency or probability of its dropping dangerously low, even if a calculated chance probability indicates otherwise);
- the knowledge or awareness of the range of possible outcomes (if you are aware of only one possible outcome, you will assign it a high knowledge probability);
- the confidence or degree of subjective belief in some outcome based on one's experience;
- control, with more management influence over the outcome generally leading to a higher perceived probability of an outcome occurring;
- the plausibility of the scenario (how convincingly the case is presented).
Potential solutions to Challenge 4
There is no simple answer to the problem of communicating technical information. It takes concerted effort by managers, scientists, and stakeholders through ongoing involvement and interaction in analyses to improve mutual understanding (Smith et al., 1999). Of course, scientists could be trained better in how to communicate technical concepts more effectively to non-technical audiences. For example, after extensive experiments, Gigerenzer and Hoffrage (1995) found that people were more likely to interpret the chance probability noted above correctly when results for a wide variety of problems were stated in frequency format rather than as decimal probabilities. Fisheries scientists should exploit this fact when presenting the probability of some outcome occurring (i.e. chance as described by Teigen, 1994). For example, compare the following statements about the effect of a proposed fishing mortality rate:
- "There is a probability of 0.2 (or a 20% chance) that the stock biomass will drop below its limit reference point within 5 years";
- "Two out of every 10 situations like this will lead to the stock biomass dropping below its limit reference point within 5 years".
To extend this idea of frequency format to uncertain events in a multi-stock fishery, consider a case in which five fish stocks are simultaneously vulnerable to harvest, but they differ in their limit reference points and current stock biomasses relative to those reference points. Say that stock assessment scientists evaluated a particular proposed management regulation via Monte Carlo simulation. If they used the recommended frequency format, they would, for example, report that "In three out of ten situations like this, any two of the five fish stock biomasses would drop below their limit reference points during the period considered." According to the studies of Gigerenzer and Hoffrage (1995), fisheries managers and stakeholders would find this statement more understandable (and would act more logically and consistently on the information) than a statement using the more typical probability format, such as "There is a probability of 0.3 that 40% of the stocks would drop below their limit reference points during the period considered."
This simple idea of using frequency format has another benefit; it may help reduce the confusion over the term "risk" discussed earlier. Mislabelling a probability of an undesirable outcome as the only measure of "risk" reflects a failure to understand the dimensions (units) of risk and its components. In the concretely pictured sets produced in peoples' minds by presenting information about uncertainties in a frequency format, the tangled concept of risk and its attendant arithmetic and dimensional errors would scarcely arise. Thinking in frequencies automatically and intuitively separates the two components of risk described earlier into two activities everyone does easily from an early age: they visualize each of the possible outcomes/costs as separate cases, and they count the cases.
A special problem of communication is that perceptions of risk are often quite different from experts' estimates of risk (Slovic, 1987). Perceptions of risks by stakeholders tend to be higher than estimated risks when:
- they have less control over uncertain events;
- they are not actively involved in the decision-making process;
- the sources of risks are completely new;
- the risks are not being shared equally among stakeholder groups (Slovic, 1987).
Of course, using frequency format and being aware of risk perception issues can only help with a small portion of the challenges related to communicating uncertainties and risks. Good documentation, early and frequent interactions among interested groups, and "gaming" workshops using models can also only go so far in bridging the gap between technical specialists and others. We need many more innovative approaches to facilitating two-way communication.
|
Conclusion |
|---|
|
TopIntroductionChallenges and some possible...ConclusionReferences |
|---|
There is one last important point to make about uncertainties and risk management. Sometimes, decision makers become unjustifiably worried about the reliability of biological information provided by fish stock assessment scientists because of the numerous uncertain components that are included in analyses, such as alternative structural forms of models and probability distributions of parameter values. However, managers and stakeholders should keep these detailed descriptions of uncertainties in perspective. They should not put low weight on biological information simply because fisheries scientists have been so explicit about describing major sources of uncertainties. Such uncertainties also exist for economic and social factors; they are just not usually described as well as uncertainties associated with physical and biological factors. Fisheries managers and stakeholders should therefore set the same standards for accepting information as evidence for economic and social indicators as they do for physical and biological indicators. Of course, the response will be, "we don't have the same rigorous data on economic and social indicators." This may be true, but economic factors such as discard rate and price per tonne of fish show considerable variation. Clearly, therefore, there should be a united call for more research on economic and social processes, such as movement of vessels and discarding behaviour of vessel crews (Hilborn, 1985; Dorn, 2001; Ulrich et al., 2002). Results of such research should aim to incorporate harvesters into models as dynamic, not static, components, and to reflect uncertainties in processes and parameter values.
This review has hopefully highlighted some of the major challenges in fisheries science and management. Potential solutions to these challenges are partly provided by advanced quantitative methods such as decision analysis, hierarchical models, and operating models. Methods and lessons learned from other disciplines such as cognitive psychology can also help improve communication among scientists, managers, and stakeholders. Considerable research is already being conducted on many of these topics by scientists in the ICES family, and elsewhere. Nevertheless, despite these advances, there is still substantial work to be done to reduce the problems and limitations described above for each potential solution to the challenges presented here.
|
Acknowledgements |
|---|
For many past discussions on the topics in this paper, I thank Zhenming Su, Brian Pyper, Franz Mueter, Bill de la Mare, Judith Anderson, Steve Haeseker, and students in my graduate course, "Risk Assessment and Decision Analysis in Management of Natural Resources", who have provided excellent feedback and forced a clarification of my own thinking on the topics covered here. I also greatly appreciate constructive feedback on a draft manuscript provided by Bill Silvert, Tony Smith, Murdoch McAllister, and Judith Anderson. Funding was provided by the Canada Research Chairs Program, Ottawa, Canada, and the Natural Sciences and Engineering Research Council of Canada.
|
Footnotes |
|---|
1 Available on line at http://www.consecol.org/vol5/iss2/art8
2 Available only on line via the Internet at http://www.consecol.org/vol2/iss1/art2
|
References |
|---|
|
TopIntroductionChallenges and some possible...ConclusionReferences |
|---|
-
Adkison M.D. and Su Z. (2001) A comparison of salmon escapement estimates using a hierarchical Bayesian approach versus separate maximum likelihood estimation of each year's return. Alaska Fisheries Research Bulletin 58:1663–1671.
Anderson J.L. (1998) Embracing uncertainty: the interface of Bayesian statistics and cognitive psychology. Conservation Ecology 2:2–30.
Butterworth D.S. and Punt A.E. (1999) Experiences in the evaluation and implementation of management procedures. ICES Journal of Marine of Science 56:985–998.[CrossRef]
Chatfield C. (1989) The Analysis of Time Series: an Introduction 4th edn (Chapman & Hall, London) 241 pp.
Chen D.G. and Holtby L.B. (2002) A regional meta-model for stock-recruitment analysis using an empirical Bayesian approach. Canadian Journal of Fisheries and Aquatic Sciences 59:1503–1514.
Clemen R.T. (1996) Making Hard Decisions: an Introduction to Decision Analysis 2nd edn (Duxbury Press, Belmont, California) 664 pp.
Collie J.S. and Sissenwine M.P. (1983) Estimating population size from relative abundance measured with error. Canadian Journal of Fisheries and Aquatic Sciences 40:1871–1879.
Cooke J.G. (1999) Improvement of fishery-management advice through simulation testing of harvest algorithms. ICES Journal of Marine Science 56:797–810.
de la Mare W.K. (1996) Some recent developments in the management of marine living resources. In Floyd R.B., Sheppard A.W., De Barro P.J. (Eds.). Frontiers in Population Ecology(CSIRO Publishing, Melbourne) pp. 599–616.
de la Mare W.K. (1998) Tidier fisheries management requires a new MOP (management-oriented paradigm). Reviews in Fish Biology and Fisheries 8:349–356.[CrossRef][Web of Science]
de Young B., Peterman R.M., Dobell A.R., Pinkerton E., Breton Y., Charles A.T., Fogarty M.J., Munro G.R., Taggart C. (1999) Canadian marine fisheries in a changing and uncertain world. Canadian Special Publications in Fisheries and Aquatic Sciences 129:1–199.
Dorn M.W. (2001) Fishing behaviour of factory trawlers: a hierarchical model of information processing and decision-making. ICES Journal of Marine Science 58:238–252.
Ellison A.M. (1996) An introduction to Bayesian inference for ecological research and environmental decision-making. Ecological Applications 6:1036–1046.[CrossRef][Web of Science]
FAO. (1995) Precautionary approach to fisheries. Part 1. Guidelines on the precautionary approach to capture fisheries and species introductions. FAO Fisheries Technical Paper, 350/1. 52 pp. [Also reprinted in 1996 with the same title, but in the series FAO Technical Guidelines for Responsible Fisheries, 2. 60 pp.].
Francis R.I.C.C. and Shotton R. (1997) "Risk" in fisheries management: a review. Canadian Journal of Fisheries and Aquatic Sciences 54:1699–1715.
Frederick S.W. and Peterman R.M. (1995) Choosing fisheries harvest policies: when does uncertainty matter? Canadian Journal of Fisheries and Aquatic Sciences 52:291–306.
Gelman A., Carlin J.B., Stern H.S., Rubin D.B. (1995) Bayesian Data Analysis(Chapman & Hall, London) 526 pp.
Gigerenzer G. and Hoffrage U. (1995) How to improve Bayesian reasoning without instruction: frequency formats. Psychological Review 102:684–704.[CrossRef][Web of Science]
Hilborn R. (1985) Fleet dynamics and individual variation: why some people catch more fish than others. Canadian Journal of Fisheries and Aquatic Sciences 42:2–13.
Hilborn R. and Peterman R.M. (1977) Changing management objectives. In Ellis D.V. (Ed.). Pacific Salmon: Management for People(University of Victoria Press, Victoria, B.C.) pp. 68–98.
IWC D.V. (1994) Annex H. Revised management procedure (RMP) for baleen whales. Report of the International Whaling Commission 44:142–152.
Jonzén N., Cardinale M., Gårdmark A., Arrhenius F., Lundberg P. (2002) Risk of collapse in the eastern Baltic cod fishery. Marine Ecology Progress Series 240:225–233.[Web of Science]
Kell L.T., O'Brien C.M., Smith M.T., Stokes T.K., Rackham B.D. (1999) An evaluation of management procedures for implementing a precautionary approach in the ICES context for North Sea plaice (Pleuronectes platessa L.). ICES Journal of Marine Science 56:834–845.
Kirkwood G.P. (1997) The revised management procedure of the International Whaling Commission. In Pikitch E.K., Huppert D.D., Sissenwine M.P. (Eds.). Global Trends: Fisheries Management. American Fisheries Society Symposium, 20(American Fisheries Society, Bethesda, Maryland) pp. 91–99.
Kuikka S., Hildén M., Gislason H., Hansson S., Sparholt H., Varis O. (1999) Modeling environmentally driven uncertainties in Baltic cod (Gadus morhua) management by Bayesian influence diagrams. Canadian Journal of Fisheries and Aquatic Sciences 56:629–641.
Lassen H. and Sparholt H. (2000) ICES framework for the implementation of the precautionary approach in fisheries management advice. Precautionary Approach Reference Points. Estimation Procedures. ACFM Working Paper, May 2000.
Liermann M. and Hilborn R. (1997) Depensation in fish stocks: a hierarchic Bayesian meta-analysis. Canadian Journal of Fisheries and Aquatic Sciences 54:1976–1984.
Linhart H. and Zucchini W. (1986) Model Selection(John Wiley, New York) 301 pp.
Mace P.M. and Sissenwine M.P. (2002) Coping with uncertainty: evolution of the relationship between science and management. In Berkson J.M., Kline L.L., Orth D.J. (Eds.). Incorporating Uncertainty into Fisheries Models. American Fisheries Society Symposium, 27(American Fisheries Society, Bethesda, Maryland) pp. 9–28.
Marmorek D.R. and Peters C.N. (2001) Finding a path towards scientific collaboration: insights from the Columbia River Basin. Conservation Ecology 5:28.
McAllister M. and Kirchner C. (2002) Accounting for structural uncertainty to facilitate precautionary fishery management: illustration with Namibian orange roughy. Bulletin of Marine Science 70:499–540.[Web of Science]
McAllister M.K. and Kirkwood G.P. (1999) Applying multivariate conjugate priors in fishery-management system evaluation: how much quicker is it and does it bias the ranking of management options? ICES Journal of Marine Science 56:884–899.
McAllister M.K., Pikitch E.K., Babcock E.A. (2001) Using demographic methods to construct Bayesian priors for the intrinsic rate of increase in the Schaefer model and implications for stock rebuilding. Canadian Journal of Fisheries and Aquatic Sciences 58:1871–1890.
McAllister M.K., Pikitch E.K., Punt A.E., Hilborn R. (1994) A Bayesian approach to stock assessment and harvest decisions using the sampling importance resampling algorithm. Canadian Journal of Fisheries and Aquatic Sciences 51:2673–2687.
McKinnell S.M. and Karlström Ö. (1999) Spatial and temporal covariation in the recruitment and abundance of Atlantic salmon populations in the Baltic Sea. ICES Journal of Marine Science 56:433–443.
Mendelsohnn R. (1988) Some problems in estimating population sizes from catch-at-age data. Fishery Bulletin, US 86:617–630.
Meyer R. and Millar R.B. (1999) Bayesian stock assessment using a state-space implementation of the delay difference model. Canadian Journal of Fisheries and Aquatic Sciences 56:37–52.
Millar R.B. (2002) Reference priors for Bayesian fisheries models. Canadian Journal of Fisheries and Aquatic Sciences 59:1492–1502.
Millar R.B. and Meyer R. (2000) Non-linear state space modelling of fisheries biomass dynamics by using Metropolis–Hastings within-Gibbs sampling. Applied Statistics 49:327–342.
Morgan M.G. and Henrion M. (1990) Uncertainty: a Guide to Dealing with Uncertainty in Quantitative Risk and Policy Analysis(Cambridge University Press, Cambridge) 332 pp.
Mueter F.J., Peterman R.M., Pyper B.J. (2002) Opposite effects of ocean temperature on survival rates of 120 stocks of Pacific salmon (Oncorhynchus spp.) in northern and southern areas. Canadian Journal of Fisheries and Aquatic Sciences 59:456–463.
Myers R.A. (2001) Stock and recruitment: generalizations about maximum reproductive rate, density dependence, and variability using meta-analytic approaches. ICES Journal of Marine Science 58:937–951.
Myers R.A., Bowen K.G., Barrowman N.J. (1999) The maximum reproductive rate of fish at low population sizes. Canadian Journal of Fisheries and Aquatic Sciences 56:2409–2419.
Myers R.A., MacKenzie B.R., Bowen K.G., Barrowman N.J. (2001) What is the carrying capacity of fish in the ocean? A meta-analysis of population dynamics of North Atlantic cod. Canadian Journal of Fisheries and Aquatic Sciences 58:1464–1476.
Myers R.A., Mertz G., Bridson J. (1997) Spatial scales of interannual recruitment variations of marine, anadromous, and freshwater fish. Canadian Journal of Fisheries and Aquatic Sciences 54:1400–1407.
National Research Council (NRC) J. (1993) Issues in Risk Assessment(National Academy Press, Washington, D.C.) 356 pp.
National Research Council (NRC) J. (1998) Improving Fish Stock Assessments(National Academy Press, Washington, D.C.) 177 pp.
Pella J.J. (1993) Utility of structural time series models and the Kalman filter for predicting consequences of fishery actions. In Kruse G., Eggers D.M., Marasco R.J., Pautzke C., Quinn T.J. (Eds.). Proceedings of the International Symposium on Management Strategies for Exploited Fish Populations(Alaska Sea Grant College Program, AK-SG-93-02, University of Alaska, Fairbanks) pp. 571–593.
Peterman R.M. and Anderson J.L. (1999) Decision analysis: a method for taking uncertainties into account in risk-based decision making. Human and Ecological Risk Assessment 5:231–244.[CrossRef][Web of Science]
Peterman R.M., Pyper B.J., Grout J.A. (2000) Comparison of parameter estimation methods for detecting climate-induced changes in productivity of Pacific salmon (Oncorhynchus spp.). Canadian Journal of Fisheries and Aquatic Sciences 57:181–191.
Peterman R.M., Pyper B.J., Lapointe M.F., Adkison M.D., Walters C.J. (1998) Patterns of covariation in survival rates of British Columbian and Alaskan sockeye salmon (Oncorhynchus nerka) stocks. Canadian Journal of Fisheries and Aquatic Sciences 55:2503–2517.
Peterman R.M., Pyper B.J., MacGregor B. (2003) Use of the Kalman filter to reconstruct historical trends in productivity of Bristol Bay sockeye salmon (Oncorhynchus nerka). Canadian Journal of Fisheries and Aquatic Sciences 60:809–824.
Peters C.N. and Marmorek D.R. (2001) Application of decision analysis to evaluate recovery actions for threatened Snake River spring and summer chinook salmon. Canadian Journal of Fisheries and Aquatic Sciences 58:2431–2446.
Power M. and McCarty L.S. (2002) Trends in the development of ecological risk assessment and management frameworks. Human and Ecological Risk Assessment 8:7–18.[Medline]
Punt A.E. (1992) In Payne A.I.L., Brink K.H., Mann K.H., Hilborn R. (Eds.). Selecting management methodologies for marine resources, with an illustration for southern African hake. South African Journal of Marine Science 12:943–958 Benguela Trophic Functioning.
Punt A.E. and Hilborn R. (1997) Fisheries stock assessment and decision analysis: the Bayesian approach. Reviews in Fish Biology and Fisheries 7:35–63.[CrossRef][Web of Science]
Punt A.E., Smith A.D.M., Cui G. (2002) Evaluation of management tools for Australia's south east fishery. 3. Towards selecting appropriate harvest strategies. Marine and Freshwater Research 53:645–660.[CrossRef][Web of Science]
Pyper B.J., Mueter F.J., Peterman R.M., Blackbourn D.J., Wood C.C. (2001) Spatial covariation in survival rates of Northeast Pacific pink salmon (Oncorhynchus gorbuscha). Canadian Journal of Fisheries and Aquatic Sciences 58:1501–1515.
Pyper B.J., Mueter F.J., Peterman R.M., Blackbourn D.J., Wood C.C. (2002) Spatial covariation in survival rates of Northeast Pacific chum salmon. Transactions of the American Fisheries Society 131:343–363.[CrossRef]
Quinn T.J. and Deriso R.B. (1999) Quantitative Fish Dynamics(Oxford University Press, New York) 542 pp.
Reckhow K.H. (1994) Water quality simulation modeling and uncertainty analysis for risk assessment and decision making. Ecological Modelling 72:1–20.[CrossRef][Web of Science]
Rice J.C. and Richards L.J. (1996) A framework for reducing implementation uncertainty in fisheries management. North American Journal of Fisheries Management 16:488–494.[CrossRef]
Ricker W.E. (1975) Computation and interpretation of biological statistics of fish populations. Bulletin of the Fisheries Research Board of Canada, 191:.
Rivot E., Prévost E., Parent E. (2001) How robust are Bayesian posterior inferences based on a Ricker model with regards to measurement errors and prior assumptions about parameters? Canadian Journal of Fisheries and Aquatic Sciences 58:2284–2297.
Robb C.A. and Peterman R.M. (1998) Application of Bayesian decision analysis to the management of a sockeye salmon fishery. Canadian Journal of Fisheries and Aquatic Sciences 55:86–98.
Rosenberg A.A. and Brault S. (1993) In Smith S.J., Hunt J.J., Rivard D. (Eds.). Choosing a management strategy for stock rebuilding when control is uncertain. Canadian Special Publication of Fisheries and Aquatic Sciences 120:243–249 Risk Evaluation and Biological Reference Points for Fisheries Management.
Rubin D.B. (1988) Using the SIR algorithm to simulate posterior distributions. Bayesian Statistics 3:395–402.
Sainsbury K. (1998) Living marine resources assessment for the 21st century: what will be needed and how will it be provided? In Funk F., Quinn T.J., Heifetz J.N., Ianelli J.E., Powers J.F., Schweigert J.F., Sullivan P.J., Zhang C.-I. (Eds.). Fishery Stock Assessment Models(Alaska Sea Grant College Program, AK-SG-98-01, University of Alaska, Fairbanks) pp. 1–40.
Schnute J.T. (1994) A general framework for developing sequential fisheries models. Canadian Journal of Fisheries and Aquatic Sciences 51:1676–1688.
Shepherd J.G., Pope J.G., Cousens R.D. (1984) Variations in fish stocks and hypotheses concerning their links with climate. Rapports et Procès-verbaux des Réunions Conseil International pour l'Exploration de la Mer 164:108–112.
Slovic P. (1987) Perception of risk. Science 236:280–285.
Smith A.D.M. (1993) In Smith S.J., Hunt J.J., Rivard D. (Eds.). Risks of over- and under-fishing new resources. Canadian Special Publication of Fisheries and Aquatic Sciences 120:261–267 Risk Evaluation and Biological Reference Points for Fisheries Management.
Smith A.D.M., Sainsbury K.J., Stevens R.A. (1999) Implementing effective fisheries management systems – management strategy evaluation and the Australian partnership approach. ICES Journal of Marine Science 56:967–979.
Spiegelhalter D.J., Thomas A., Best N.G. (1999) WinBUGS Version 1.2 User Manual, MRC, Biostatistics Unit(Cambridge University, Cambridge, U.K.).
Su Z., Adkison M.D., Van Alen B.W. (2001) A hierarchical Bayesian model for estimating historical salmon escapement and escapement timing. Canadian Journal of Fisheries and Aquatic Sciences 58:1648–1662.
Su Z., Peterman R.M., Haeseker S.L. (2004) Spatial hierarchical Bayesian models for stock-recruitment analysis of pink salmon (Oncorhynchus gorbuscha). Canadian Journal of Fisheries and Aquatic Sciences 61:.
Teigen K.H. (1994) Variants of subjective probabilities: concepts, norms, and biases. In Wright G. and Ayton P. (Eds.). Subjective Probability(John Wiley, New York) pp. 211–238.
Ulrich C., Pascoe S., Sparre P.J., De Wilde J.-W., Marchal P. (2002) Influence of trends in fishing power on bioeconomics in the North Sea flatfish fishery regulated by catches or by effort quotas. Canadian Journal of Fisheries and Aquatic Sciences 59:829–843.
Walters C.J. (1986) Adaptive Management of Renewable Resources(MacMillan, New York) 374 pp.
Walters C.J. and Ludwig D. (1994) Calculation of Bayes posterior probability distributions for key population parameters. Canadian Journal of Fisheries and Aquatic Sciences 51:713–722.
Ware D.M. and McFarlane G.A. (1989) In Beamish R.J. and McFarlane G.A. (Eds.). Fisheries production domains in the northeast Pacific Ocean. Canadian Special Publication of Fisheries and Aquatic Sciences 108:359–379 Effects of Ocean Variability on Recruitment and an Evaluation of Parameters Used in Stock Assessment Models.
CiteULike 
Connotea 
Del.icio.us What's this?
This article has been cited by other articles:
|
|
 |
|
  M.-J. Rochet and J. C. Rice Simulation-based management strategy evaluation: ignorance disguised as mathematics? ICES J. Mar. Sci., May 1, 2009; 66(4): 754 - 762. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
M. Dickey-Collas, M. A. Pastoors, and O. A. van Keeken Precisely wrong or vaguely right: simulations of noisy discard data and trends in fishing effort being included in the stock assessment of North Sea plaice ICES J. Mar. Sci., December 1, 2007; 64(9): 1641 - 1649. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
L. T. Kell, I. Mosqueira, P. Grosjean, J-M. Fromentin, D. Garcia, R. Hillary, E. Jardim, S. Mardle, M. A. Pastoors, J. J. Poos, et al. FLR: an open-source framework for the evaluation and development of management strategies ICES J. Mar. Sci., May 1, 2007; 64(4): 640 - 646. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||