APPENDIXES

APPENDIX I - Fishery checklist

The Marine Stewardship Council (www.MSC.org) has developed a series of principles and criteria which allow the fishery management to be measured against the FAO Code of Conduct for Responsible Fisheries.

Many of the criteria can be directly related to CITES requirements. However, the MSC and FAO Code of Conduct go well beyond CITES minimum requirements in their definition of good management.

By applying good management defined by the Code of Conduct, not only will CITES requirements be met, but various pressures leading to overfishing will be removed.

The principles, criteria and guidelines are provided herewith a commentary for conch and can be used to assess a fishery to see how well management is being applied. All the issues raised should be considered by the fishery managers in developing a sound management system. The priority of requirements (first and second) is also listed. (The guidelines have been adapted from Moody Marine Ltd. Scoring guidelines are available, with other documentation, from the Marine Stewardship Council website www.MSC.org) Those components considered to be most relevant to the CITES regulations and requirements are Principle 1 and all its sub-components, 2C, 3A, 3B1 and 3G.

APPENDIX II - Interpreting mean meat weight

1. Factors controlling meat weight

Mean weight of conch meat pieces may change in a fishery due to three causes:

• Increasing fishing mortality will tend to decrease the average weight as there will be an increased chance that the conch will be caught when they are younger.

• Selectivity will not only affect the population in the same way as fishing mortality, but if selectivity changes the size composition and average weight will change.

• As recruitment fluctuates, so the relative abundance of young, small animals will change in the catches.

Controlling mean size by itself will be inadequate in protecting the stock. In observing any particular catch composition, it may be difficult to interpret the meaning with respect to the underlying population from which the catch is drawn. For example, a high proportion of mature conch in the catches may indicate that the stock is healthy with a large spawning stock. It may also indicate that recruitment has failed and fishers are switching to depleting the spawning stock in deeper waters.

2. Reference points

Mean size could be used for two purposes. If mean size can be linked to the proportion of the catch which is mature, some minimum size can be set to ensure that this proportion remains high. Conceptually, this will help to ensure that animals are only exploited once they have had a chance to spawn and that the spawning stock will be maintained.

Interpreting such a policy in terms of fishing mortality will depend on selectivity, but in general the implication would be that the fishery is directed at the mature stock and that spawning would be less likely to be affected. Even a high fishing mortality should allow a significant proportion to spawn before being caught if the maximum size is set high enough. Ensuring that some minimum proportion of animals reach maturity is usually managed by spawners-per-recruit analyses.

Given the variability in size at age, size at maturity and so on, it is not possible to specify a meat weight where all conches will be mature. A reasonable limit reference point would be the mean weight where 50 percent (or some other percentage) are mature. Maturity is relatively easy to measure from the shell (flared lip) as well as the body prior to processing. Size composition data would have to be collected on individuals before and immediately after processing to enable appropriate estimates to be made. Such conversion models do not have to rely on random samples. Specifically larger and smaller individuals can be selected to get a good range. The analysis is straightforward using generalized linear models.

A more specific proposal can be made in the case of avoiding growth overfishing. For any animal which is caught, a choice could be made between leaving it in the sea, allowing it to grow at the risk of dying from natural causes or catching it now. Balancing the losses from mortality against gains from growth is usually managed by yield-per-recruit analyses (See Section 11.3).

Per-recruit analyses can be carried out without data as long as a reasonable growth and mortality model is available (Figure 4). The following yield-per-recruit analysis uses the Gompertz growth and maturity models, and parameters used in the Belize conch stock assessment workshop 1999 (CRFM, 1999). The model and parameters are presented as an example and would need to be reviewed and, ideally, estimated for each fishery.

Although meat weight is shown to be approximately linearly related to fishing mortality, the change is relatively small (Figure 4). It is possible that measurement errors will give a wide range of possible fishing mortality estimates for any given meat weights. Nevertheless the mean weight can be interpreted as long as the fundamental assumptions are not violated.

Table 7: Parameters used in the yield-per-recruit model. Parameters and models are taken from CRFM (1999), but are not the only parameters and models which could be proposed. Alternatives should be considered before applying this approach.

A mean weight based on yield-per-recruit assumes a particular selectivity and fixed recruitment. If these assumptions are incorrect, the SPR and YPR based reference points may be poor estimates. Size frequency data in many cases will show such assumptions are false and hence interpretation of mean meat weight invalid. An alternative more robust approach is desirable.

Figure 4: Spawning biomass per-recruit and yield per-recruit indicating the SPR_40%, F_0.1 and F_MSY for conch using the growth and mortality parameters indicated. SPR_40% is the fishing mortality where the spawning biomass has been reduced to 40 percent of its unexploited state. F_0.1 and F_MSY are the fishing mortality where the YPR slope is 10 percent of the initial slope and the YPR maximum respectively. Each fishing mortality reference maps directly to an equivalent mean weight. Hence, given a per-recruit analysis, mean weight might be used as a proxy for fishing mortality. However, the method assumes constant recruitment and knife edge selectivity.

3. Example decision rule

The following example illustrates an approach where data can be immediately used to make a decision. The method focuses on growth, but a similar method should be developed for maturity, looking to maintain a desirable ratio of mature to immature conch in the catches. Such an approach will require the expected ratio in a lightly exploited fishery, which was not available for this analysis.

An alternative but closely related approach to yield-per-recruit would look at actual size compositions in the catches and decide whether overall it would have been better to leave the animals in the sea longer to grow rather than catch them now. Such a system would be independent of selectivity and recruitment, but would reflect current catch composition and would change in relation to these effects. Hence, a strong recruitment might well lead to advice to reduce fishing mortality. The approaches are best encapsulated as decision rules, whereby actions taken on an indicator’s values in relation to reference points.

For any particular size composition, the net gain in leaving the animals in the sea can be derived from the growth model:

where C is the catch weight, Y₁ is the yield of the i^th conch in the catch and W_it is the weight of the i^th conch in the catch (sample) given that its age is t and natural mortality is M. Where discounting is used, M becomes natural mortality plus the discount rate. The calculation for the marginal yield (dY_i/dt) can be summed over the sample. A positive number will indicate that overall greater gains would have been made by leaving these conches in the sea and catching them later.

For the Gompertz growth model, the marginal yield equation for the catch composition becomes:

The critical point will be where the marginal yield is zero, where it switches from being positive to negative. This represents the optimal point for the yield. Through simplification and substitution this point can be defined as:

For a particular sample, where the score S is negative, the implication is yield would be improved by increasing the fishing mortality. Conversely, when S is positive, yield would be increased by reducing fishing mortality. The score itself might therefore be used as an indicator determining the performance of catches in relation to growth. The absolute sum of the score measuring the difference of each size from the optimal size can also be used to indicate the efficiency of the current selectivity. These indicators might be used to measure and quantify various management controls which could improve selectivity as well as advice on fishing mortality levels.

This analysis was conducted on Belize data (from CRFM, 1999) for illustration purposes using the previous growth and mortality parameters (Table 7). The data show a large change in catch composition due to different sampling activities (Figure 5). This change is clearly picked up both by the score and the mean weight indicator, both implying rightly that the 1998 size composition would represent growth overfishing.

Where only a mean meat weight is available, the score can be applied to the single value. This point would be the size where, for example, an aquaculturist or farmer would harvest the resource. For the Belize parameters, the optimum would be 190g mean meat weight. This ignores the variation in size in the catches, but nevertheless gives some indication of whether meat weights are too small.

Figure 5: Catch size composition reported for Belize 1996-1998 (CRFM, 1999). The compositions show a significant change from 1996/97 to 1998 due to a change in sampling practice and does not represent catches. The data are used for illustration purposes only.

Table 8: Mean weight and scores for Belize weight frequency sample data. The mean weight fell in 1998. The score indicates that if 1998 composition represented the catch, growth overfishing would occur. Conversely 1996-97 suggests fishing mortality could increase. The absolute score measures the inefficiency of the gear selectivity. The lower this score, the closer the sizes will be to the optimum size. Controls which reduce this score could increase yield without increasing fishing mortality.

4. Growth model

The decision rule is sensitive to the choice of growth model and parameters. It is important that a growth model is used that is as reliable as possible.

The size frequency data from Belize suggests the asymptotic size of 240 g is probably too small. Carrying out the procedure for a more realistic 300 g asymptotic size suggests an optimal mean meat weight of 236 g, and that 1996/97 is very close to this optimum (i.e. fully exploited).

The growth model is deterministic in this analysis. A slightly more sophisticated approach would allow the model to describe the mean growth and allow for growth variability. Further work would allow uncertainty in parameter values to be represented. Decision rules are particularly good at coping with uncertainty as they allow the costs of making the wrong decision to be explicitly taken into account.

APPENDIX III - Participatory fisheries stock assessment (ParFish)

ParFish implements a multi-criterion decision-making methodology to provide management advice for data-poor, artisanal fisheries in developing countries. The method applies Bayesian decision analysis by identifying the "Bayes action" (see Lindgren, 1976 p.374, for example). All analyses have explicit risks with the optimum action maximizing the expected utility. The method focuses on rapid inexpensive assessment methods to initiate adaptive management.

The standard stock assessment models still form the basis of the method. The software allows the method to be applied without knowledge of the numerical routines. Robust statistical techniques have been chosen where ever possible to improve results.

The method works by specifying a simulation model representing the behaviour of the fishery. Simulation model parameter sets are drawn at random from a probability distribution constructed from the available information to run simulations. These are used to simulate the possible behaviour of the fishery in response to a fishery control. The fishery controls supported are effort, catch and closed area controls. During the simulation, the state of the stock and the preference score as indicated by the fishers is recorded. These are used to calculate the state of the stock and the fishery in relation to the relevant reference points.

The probability modelling is based on parameter frequencies. This allows complex models to be broken down into simpler components. Modelling frequencies, as opposed to modelling data directly, allows a wide range of information sources to be used. Probabilities are modelled using multi-dimensional non-parametric kernel smoothers. Smoothing is carried over all dimensions and random draws (selections) from the posterior are very fast.

Kernel smoothing functions are a non-parametric way to estimate probability densities from frequency data. They spread individual points in smooth functions around the point. They work on the same principle as histograms, but have better statistical properties. The smoothing, and hence weighting amongst the different data sources, is estimated directly from the data themselves using a least squares cross-validation technique (see Silverman, 1986).

Fisher interviews form an important component of the method. Interviews are used to:

• calculate a preference score as a proxy for utility.

• estimate a prior probability for the logistic stock assessment model.

The method supports fishing experiments as well as standard catch-effort data models. Fishing experiments allow catchability (fishing mortality) to be estimated rapidly.

Field testing in the Turks and Caicos Islands was used to test whether interviews can be relied upon to provide sensible management advice. It was shown in the Turks and Caicos case that interviews provided advice which, if applied in 1974, would be expected to have obtained much greater benefits from the fishery than that which was obtained under no catch control.

1. Turks and Caicos Islands Retrospective Analysis

The example illustrates an important tenet of the precautionary approach, that a lack of information should not be used as a reason to delay action. Methods exist, such as ParFish, for which data can be both rapidly collected and inexpensive. These can lead to making decisions in the short term which will be a very great improvement over taking no action.

Using interview data in the assessment, it was tested against standard catch and effort stock assessment using decision analysis (see PFSA, 2003). Interviews were conducted in July 2003 to obtain stock assessment information from fishers. The Turks and Caicos Islands has a 30-year catch and effort time series which can generate not only an assessment for comparison with the interview decision analysis but also can indicate what would have happened had a quota based on the interviews been applied. It was of interest to see how well management would do if actions were based only on the interview data.

The fishery is managed through a quota, so this is the appropriate control. A standard stock assessment using the logistic model fitted to the catch-effort time series indicated the current quota of 1.675 million pounds as too high; and recommended lowering it to 1.6 million pounds landed weight. Using the preference information, the stock assessment based upon both the interview and catch-effort model combined and the catch-effort model alone suggest a lower quota around 1.53 and 1.38 million pounds respectively. Interviews by themselves are much less accurate (as indicated by the much lower limit control), but nevertheless recommends a target of 1.68 million pounds, reasonably close but above the other targets.

Table 9: Target and limit controls for the Turks and Caicos Islands conch fishery based on catch-effort and interview data.

If it is assumed that fishers knew as much in 1974 as they do now, we can use the interview data as representative of what would have been obtained had the interviews been conducted at the beginning of the time series. Hence, the interview-only target quota can be applied at that point to see what might have happened to the fishery had this stock assessment method been applied, assuming that the logistic and maximum likelihood parameter estimates are correct.

The actual total catch over the period 1975-2002 was 45.47 million pounds. Had the 1.68 million pound quota been applied, the results suggest a total catch of 47.00 million pounds. This quota would realize higher catches in the longer term by foregoing higher catches in the late 1970s. A discount rate of around 5 percent yields approximately the same net present value between the two options.

The real gain, however, would have been the rise in catch rate (Figure 6). The catch-effort model suggests the stock was in an overfished state in 1974 and an enforced quota would have led to stock recovery. In other words, the catch would be met with much less work and costs than is now applied (from 3 300 boat days down to 2 500 boat days to realize the same catch). It indicates considerable benefits to using just interviews in this case, but would need more testing to make the case as a general statement. In particular, in cases where it turns out the logistic is not the best model, it needs to be shown that interviews may still have value in setting initial targets.

Figure 6: Expected catch per boat day (CPUE) from the fitted logistic model and the projected CPUE with 1.68 million pound quota.

The cost of applying the quota is that, without the depletion in the mid 1980s, less information would now be available on the behaviour of the stock, so that the current stock assessment would be less reliable. This problem would have to have been addressed through increased research activities.