3.2 The Evolution of Management in Existing Fisheries                        [return to table of contents]

Large-scale exploitation of many fish stocks in the Convention Area began before the establishment of CCAMLR, and many stocks were already overexploited in 1982 when CCAMLR came into force. Most circumpolar abundance estimates of krill stocks were in the range of tens to hundreds of millions of tonnes, and thus were at least two orders of magnitude greater than the annual catches. CCAMLR’s first priority was to conserve fish stocks, not manage the krill fishery, but krill became an important issue in the late 1980s, when krill fishing began to be concentrated in the foraging ranges of krill-dependent predators such as penguins and seals.

(i) The Early Years – Conventional Approaches in the 1980s

The methods used by WG-FSA to assess exploited fish stocks have evolved from more-or-less standard methods used in fisheries assessment worldwide since the 1970s and early 1980s. One of the first methods, which has been used with mixed success, is known as virtual population analysis (VPA). A conventional VPA reconstructs the abundance of a stock over time by adding up the catches of each year class in the stock and accounting for natural mortality. This also gives estimates of recruitment to the stock back to the early years of the fishery. A typical assessment would first use the VPA to estimate stock size and recruitment, and then estimate the future stock size under different proposed management regimes in order to advise on the consequences of these regimes. Unfortunately, accurate estimation of stock trajectories and recruitment depends not only on the reliability of catch statistics, but also on the accuracy of estimates of current stock size, to which the catches and natural deaths are added backwards in time. The VPA can be good at indicating the initial size of the stock, particularly if it has been heavily fished. However, in the absence of other data, it provides no more information on current stock size than other methods.

The basic VPA can be modified to improve estimates of current stock size by ‘tuning’ it to ancillary data on relative or absolute abundance. This method provides the estimate of current stock abundance that gives the best statistical fit to the relative or absolute abundance data.

Although these methods use data collected from the fisheries, such as catch-at-age and effort data, such data alone do not always lead to reliable assessments. In the Convention Area, assessments have been substantially improved by Members conducting scientific surveys in areas of key interest. The use of survey data in conjunction with fisheries-derived data has become CCAMLR’s preferred approach, as it has in many other fisheries conventions. In cases where stock assessments were out of date, or where there has been substantial uncertainty, the Commission has made the conduct of a fishery-independent scientific survey (see section 3.1(iii)) a condition for re-opening a fishery.

When estimates of current abundance are available, it is common practice to calculate a target fishing mortality (instantaneous rate of fishing) for the stock. This calculation is based on estimates of growth rates and natural mortality of the species in question, and is traditionally carried out as a yield-per-recruit analysis. The abundance of, and catch of, a cohort (year class) of fish are calculated throughout the cohort’s life at various levels of fishing mortality. The accumulated catch from the cohort divided by the original size of the cohort at recruitment gives the yield-per-recruit figure.

For some species, the relationship between yield per recruit and fishing mortality (the ‘yield-per-recruit curve’) exhibits a maximum (termed F_max) which has been used as a target fishing mortality for those species. However, for many other species the yield-per-recruit curve does not have a maximum, and so it has been a long-standing practice to set the target fishing mortality at the value at which the tangent to the curve has a value of 10% of the tangent at zero fishing mortality. This value is known as F_0.1. CCAMLR has used F_0.1 as one of the first elements in its management policy for finfish fisheries.

The sustainability of harvesting is largely determined by two factors: the relationship between the size of the spawning stock and the subsequent survival of the offspring on entering the fishery (recruits). The objective of fisheries management should be to maximise yield while keeping the risk of overfishing the stock to an acceptably low level. Fishing at F_max or F_0.1 does not necessarily maximise yield and can deplete the spawning stock biomass to a level where stock recruitment is at risk (referred to in Article II as ‘unstable recruitment’). To overcome this problem, CCAMLR now uses the escapement of the spawning stock as the criterion to determine the allowable level of fishing mortality.

However, the calculations for a yield-per-recruit analysis do not take into account either uncertainty in the biological parameters or random fluctuations in recruitment. For these reasons, CCAMLR has increasingly emphasised stochastic projection methods which use computer simulations to take both these forms of uncertainty into account. Their development is described in the next section.

(ii) Current Approaches – Modelling Studies               [return to table of contents]

(a) Krill            [return to table of contents]

The Krill Yield Model (KYM), developed after the second meeting of the CCAMLR Working Group on Krill (WG-Krill) in 1990, raised concerns about the level of krill exploitation in Subarea 48.3. At that time, the estimates of the krill biomass in part of that subarea averaged only some 0.6 million tonnes, which was barely three times the annual commercial catch of krill in that subarea.

Attempts were made at that meeting to apply a simple approach developed for fish stocks by John Beddington and Justin Cooke in 1983. Their analyses provide a numerical factor (termed g) that can be used to multiply a single estimate of biomass obtained from a survey before harvesting begins to give an estimate of the potential annual sustainable yield. The value of the numerical factor depends on the biological parameters of the stock under consideration. Difficulties immediately became apparent when attempts were made to determine values of some of these parameters for krill, with the result that estimates of the potential annual yield for Subarea 48.3 ranged widely, from 0.2 to 13 million tonnes.

Efforts to improve both the model and the estimates of the parameters were accelerated by the Commission’s request at its 1990 meeting for the provision of best estimates of precautionary catch limits for krill in the various statistical areas.

The essential features of Beddington and Cooke’s approach are:

• a single estimate of only the resource biomass is available;

• annual recruitment is assumed not to fall as the spawning stock size drops, although it does fluctuate around its average level as a result of variability in the environment (these fluctuations mean that the overall biomass will also vary even without harvesting, so the approach takes account of the possibility that the survey may have been made in a year of above (or below) average abundance);

• potential yield is evaluated on the basis of satisfying a risk criterion: that even under harvesting, the probability that the spawning biomass will fall below a level at which recruitment ‘on average’ might be impaired is to be kept small.

The KYM has been modified to allow for:

• strong seasonal effects – unlike the customary situation for fish, where individual animals keep growing over time, seasonal effects in the Antarctic are such that nearly all somatic growth of krill takes place in the three months from November to January;

• the possibility that the fishing season may not run throughout the full year;

• the pre-fishing survey’s estimate of biomass being imprecise, rather than exact;

• uncertainties in the estimates of many of the biological parameters.

Although the results from the KYM also depend on such parameters as the age at sexual maturity and age at recruitment to the krill fishery, early calculations showed that the two key parameters (to which the model was particularly sensitive) were the natural mortality rate of krill and the annual fluctuations of krill recruitment. Initially, the values of these two parameters were little more than guesses. More recently, however, analyses of krill length-distribution data from research surveys have provided better estimates of both these parameters and also better precision of those estimates. The degree of precision is one component of the overall uncertainty, which should be taken into account in the analyses by integrating it over the range of possible values for both (as well as other) parameters. This integration gives greater weight to sets of values that are most consistent with length-distribution information from research surveys.

During the development of the KYM, scientists debated whether the effects of immigration and emigration of krill from a subarea during the course of a year should be taken into account, and if so, how. Clearly, not much (if any) of the krill detected in a survey in Subarea 48.3 would remain resident there throughout the year, because of the general northeastward movement of the water masses in that subarea. Thus, on the one hand it was argued that yield estimates should be based on the total amount of krill passing through a subarea over the year, rather than only that present over the short duration of a survey. On the other hand, such an approach would not take account of the effects of fishing on krill in the other subareas through which they are transported by the currents. Given the difficulties of adjusting properly for both these effects, the present approach is (in principle) to use the yield model to provide precautionary limits for all subareas, based on survey abundance estimates for each. Thus the ‘extra’ catch that arguably could be taken in a subarea because the survey estimate could be adjusted upward to allow for immigration, can instead be taken from adjacent subareas.

The two key outputs of the KYM that are used to decide an appropriate value for the factor g are shown in Figures 11 and 12. The krill survey abundance estimates (termed B₀) are multiplied by g to provide precautionary limits for annual catches. The model assumes that krill fishing takes place throughout the year, as it does in the present fishery. Both plots pertain to the results of a 20-year period of fishing under a fixed TAC. The first shows the probability that the krill spawning biomass will drop below 20% of its median value in the absence of any fishery. As the intensity of krill harvesting grows (i.e. as increases), so does this probability, raising the risk that the spawning biomass is depleted to a level at which recruitment success may be impaired – a situation commonly referred to as ‘recruitment overfishing’. As in Beddington and Cooke’s original work, a value of 10% is used as a standard for this probability. Thus, as is evident from Figure 11, this criterion requires that g be set no higher than 0.149 in fixing precautionary limits.

The arguments of the previous paragraph consider the krill fishery only in a ‘single-species’ context. However, the wording of CCAMLR’s Article II requires that the needs of krill predators are also given consideration in setting precautionary limits for the fishery. At present, detailed modelling of the impact the fishery might have on such predators has yet to provide reliable quantitative results, so an ad hoc approach is being followed for the moment. This is based on the output from the KYM shown in Figure 12, which plots (against g) the median krill spawning biomass after 20 years as a fraction of the corresponding value in the absence of a krill fishery (g = 0). If only krill were to be taken into account, an appropriate target level for this ratio in terms of conventional fisheries management might be 50%. On the other hand, the best situation for the predators would be no fishing at all, i.e. a ratio of 100%. The preliminary target adopted is halfway between these two ‘extremes’, i.e. 75%. Reference to Figure 12 shows that this corresponds to a value of 0.116 for g. The final stage in the application of the results from the yield model to provide a precautionary catch limit is to select the lower of the two values of g (0.149 and 0.116) corresponding to these two criteria (see section 3.2(iii)).

The KYM will be continually refined as more data become available to reduce the uncertainty in estimates of some of the input parameters and as more is learnt about the relationships between these inputs. These factors could affect estimates of the value of g that are appropriate to avoid recruitment overfishing. Possibly more important, however, will be the refinement of the krill and krill–predator models (see section 3.2 (ii)c) to provide a sounder basis for the selection of a target krill escapement value; this would address the concerns of Article II on a more scientifically defensible basis than the ad hoc approach underlying the present selection of 75%.

(b) Finfish                                    [return to table of contents]

A very similar approach to the KYM, termed the ‘Generalised Yield Model’ (GYM), has been applied to some fisheries for finfish. For certain species of finfish, such as the Patagonian toothfish, the predator criterion is not applicable because they are not important prey species. In such cases, the criterion that has been applied is to maintain populations at the level likely to give the ‘greatest net annual increment’ (GNAI), which is conventionally assumed to be around 50% of the unexploited level. The GYM is used to make the same kinds of calculations as the KYM (in fact the KYM can be set up as a special case in the GYM). Precautionary catch limits for the lanternfish Electrona carlsbergi, an important prey species for fur seals, king penguins and squid, have also been calculated from a GYM in a similar way to that used for krill.

The GYM is very flexible, allowing the use of estimates of current or pre-fishing biomass, along with estimates of their uncertainty, in projections of stock biomass. Recruitment fluctuations and uncertainty in demographic parameters are taken into account, as well as the effects of previous catches on the stock. Recruitment can be expressed either as absolute levels (i.e. as numbers of fish), or relative to the pre-fishing spawning stock biomass. For Patagonian toothfish assessments, the GYM is used with recruits drawn from a random distribution, which provides absolute numbers of recruits compatible with estimates of recruitment obtained from trawl surveys. This enables CCAMLR to use the stochastic projections of the GYM to evaluate the effects of different levels of catch even without direct estimates of absolute abundance for the whole stock.

(c) Functional Relationships between Krill and Krill Predators    [return to table of contents]

Krill, the species that shaped so much of the thinking behind the conservation principles of the CCAMLR Convention (Article II, paragraph 3(b)), was initially considered from the single-species perspective, although the inclusion of krill escapement into the decision rules makes some allowance for krill predators over a larger spatial scale, such as a statistical division or subarea.

When CCAMLR first looked at this problem, the KYM had not been developed and less was known about the dynamics of the krill stocks and their interaction with predators. Nevertheless, it was known that the areas of highest krill fishing activity were often close to the land-based breeding colonies of krill-eating birds and seals. These predators depend on krill being within reach of their colonies in order to feed and rear their offspring during the Antarctic summer. The recognition that the most sensitive interactions are probably occurring at much smaller scales than division or subarea means that information from fishing grounds near predator colonies will need to be incorporated into any conservation plan.

To quantify any overlap between the areas in which krill was being fished and the areas in which predators foraged for krill, the concept of the Critical Period Distance (CPD) was developed. This concept is based on the fishery’s catch of krill within 100 km of predator’s land-based breeding colonies between December and March when krill availability to such predators is critical. When expressed as a percentage of the total catch in a subarea, the CPD provides information on the distribution of krill catches in relation to predator colonies.

The CPD is simply a description of the potential spatial and temporal overlap between the fishery and predators foraging for krill. Whether this overlap would have any effect on predators depends on the relationship between the krill fishery, local and regional krill abundance, and the availability of krill to predators. Initial attempts to model these relationships are currently being revised to incorporate temporal aspects of penguin foraging into the model, and new standardised indices based on the niche-overlap theory are being developed to better reflect the foraging–fishery overlap.

Investigations of the interactions of a fishery, the species it harvests and other species dependent on the harvested species are part of what is termed the ‘multi-species approach’. Despite various attempts worldwide to base fisheries management on multi-species approaches, most of the world’s fisheries are still managed on a single-species basis – i.e. the impact of harvesting on only the harvested species is assessed. CCAMLR’s use of a multi-species approach to this problem is innovative; there is little experience elsewhere in this type of assessment.

The main difficulty with a multi-species approach is that a relatively large number of parameter values must be estimated. Furthermore, each parameter estimate has an associated level of uncertainty; the more parameters used in the prediction, the greater the uncertainty that the prediction will prove correct. A multi-species approach is also likely to take longer to develop because of its great complexity. Given these difficulties, the first step in developing a sustainable harvesting strategy for krill was the single-species model of potential krill yield described above.

The next step was to propose a way in which the needs of krill-dependent species could be taken into account. In 1992, WG-Krill suggested, as a first approach, a ‘one-way’ model in which fluctuations in the krill resource have an impact on a predator population but not vice versa (Figure 13). The krill population is represented by a simplified form of the yield model. A simple population dynamics model is used to represent the predator population. The link between these two models (‘consumption by predator’ in Figure 13) is given by a functional relationship between krill abundance (expressed as a proportion of its level in the absence of fishing) and the survival rate of the predator (Figure 14).

The next step in developing the approach outlined in Figure 13 was to choose parameter estimates for the model. The parameter values used in the KYM are retained (including krill recruitment variability), but parameters that have a range of possible values are fixed at the midpoint of their presumed range. The biological parameters required by the predator model had already been monitored by CEMP. This left the functional relationship to be defined. Ideally, this could be determined by using time series of krill biomass data and predator survival rates measured simultaneously in the same areas. Measured krill biomass could then be plotted against measured predator survival and a curve could be fitted to these data, although it would then be necessary to link local krill availability (which is a function of krill abundance and distribution with time) to krill abundance as calculated by the yield model for a particular spatial area.

Unfortunately no such integrated datasets are available. Annual estimates of survival rates of certain predators at various CEMP sites are, however, available. In the absence of estimates of local krill abundance, annual krill abundance values can be calculated from the yield model. These are converted to krill availability values by adding some level of random error. To link these two datasets together, it must be assumed that any changes in the measured predator survival rates are primarily due to fluctuations in krill availability.

An assumption must also be made about the shape of the functional relationship (the example in Figure 14 uses a sigmoidal shape) between predator survival rates and krill availability. Typically, the form that is chosen is defined by the latter two parameters. Given these and the level of variability that relates krill abundance to krill availability, the set of krill abundance values from the yield model can be converted into a set of predator survival rates, which can then be compared to the observed predator survival rates. The parameter values of the functional form are varied until the simulated set of survival rates most closely matches the measured set. This is achieved by comparing their moments – mean, variance and skewness. This ‘one-way’ approach (Figure 13) was applied to data on Antarctic fur seal and black-browed albatross populations from South Georgia.

The nature of the relationship between the krill and predator models is of fundamental importance to the procedure as a whole. It is extremely important, in the future development of this work, that this relationship be carefully investigated. For example,

• Can most variation in predator survival rates be directly attributed to changes in krill abundance?

• How much (including local distribution, size, sex, maturity) of the krill stock is actually available for consumption by seabird and seal predators, and do fluctuations in the availability of krill mask fluctuations in absolute krill abundance?

• What is the impact of environmental conditions such as sea-ice extent and the nature and timing of local weather conditions?

• What is the true shape of the functional relationship and how robust is the model to errors in the assumption of this shape?

• In the krill fishery, factors other than fishing intensity (e.g. timing, extent and location of the fishery in relation to predators’ breeding colonies) might influence krill availability to predators. Does the influence of these factors change from year to year, independently of fishing intensity?

As part of its ecosystem approach, CCAMLR is concerned about ‘indirect effects’ of fishing, i.e. that the removal of prey (krill) at one trophic level can indirectly affect other trophic levels, such as seabirds or marine mammals. Consequently, a second model has been developed.

This model investigates the influence of krill fishing on an Ad?ie penguin population by linking predator survival and reproductive success to local krill availability. This model of the indirect effects of a fishery on krill predators has four main components:

• the spatial and temporal patterns of krill availability;

• the mode of operation of the fishery and its effects on krill;

• the foraging performance (determined by empirical methods or ‘rules of thumb’) and survival of a predator through each of the five stages of its breeding season, incorporating a detailed empirical energy budget for chick rearing; and

• the effect of the removal of krill by the fishery on the reproductive success and adult survival of the penguins.

As with all models, there is a compromise between the level of tractability and the level of biological detail. The model focuses on parental foraging to meet requirements for the growth of a single chick. Parents and offspring are characterised by the difference between the amount of krill they need for maintenance (parents) and development (chick) and the amount of krill that they have actually eaten. Thus, variations in krill availability, due to either natural causes or fishing, will affect the breeding success of penguins.

The model uses ‘offspring survival to fledging’ as a measure of parental reproductive success. Offspring survival and parental survival depend on foraging behaviour, timing of breeding, and the availability of krill. The main aim of the model is to determine possible answers to the following question: ‘If a certain fraction of the available krill is removed by the fishery, what is the reduction in parental reproductive success and survival?’

A typical result is shown in Figure 15(a) for chick survival and in Figure 15(b) for parental survival. The x-axis is the fraction of available krill removed by the fishery. The y-axis is the ratio of survival in the presence of the fishery to survival in the absence of the fishery; hence, it is a relative measure of survival.

Both offspring and adult survival are approximately linear functions of the fraction of krill removed. However, the slope of the relative survival of the chick is about 1.5; thus, for example, removal of 1% of the available krill leads to a reduction of 1.5% in offspring survival and parental reproductive success. On the other hand, the slope of the relationship between adult survival and the fraction of krill removed is less than one (about 0.65 for breeders and 0.5 for non-breeders). Work is continuing on incorporating more detailed spatial structure into the distribution of krill and incorporating krill abundance into post-fledging survival of the offspring.

(d) Other Predator–Prey Relationships               [return to table of contents]

In the early years of CCAMLR, krill was viewed as the central component of the food web and was therefore the focus of CEMP. It is now clear that similar approaches to those developed for krill need to be developed for other important species of the food web. Exploitation of lanternfish in the second half of the 1980s and the recent interest in harvesting squid have highlighted the need to look at some other food chains. Lanternfish are the staple food of king penguins, and also of fur seals in the Indian Ocean sector. Squid feed on zooplankton, including krill, and lanternfish. They are preyed on by toothfish, albatrosses, larger penguins, seals and toothed whales. The life histories of commercial squid species are quite unlike those of finfish and krill. Consequently, although the general principles of ecosystem assessment currently applied to krill might also be used for lanternfish and squid, specific procedures will need to be developed for such assessments.

(iii) The Concept of Decision Rules                       [return to table of contents]

The operational objectives described in section 2 go some way towards interpreting the principles set out in Article II of the Convention. However, they are still not sufficiently specific for objective scientific analysis of different management options. ‘Decision rules’ have therefore been developed. A decision rule specifies the set of decisions that are made in setting, removing or varying management measures, using assessments of the status of a harvested resource.

The determination of the potential yield in the krill fisheries discussed in section 3.2(ii)a is an example of a decision rule that has three parts:

1. choose g₁ so that the probability of the spawning biomass dropping below 20% of its pre-exploitation median level over a 20-year harvesting period is 10% (this is illustrated in Figure 16);

2. choose g₂ so that the median escapement in the krill spawning biomass over a 20-year period is 75% of the pre-exploitation median level (this is illustrated in Figure 17); and

3. select the lower of g₁ and g₂ as the level of g for calculation of krill yield.

As the values of g₁ and g₂ will be different, the third part of the decision rule results in the lower of the two values being applied. A similar decision rule is applied to the fisheries for Patagonian toothfish.

Additional decision rules will be needed as new fisheries or new methods of assessment are developed. For example, they will be needed to enable assessments based on CEMP data to be taken into account when adjusting catch limits or other management measures. Decision rules link the general principles set out in the Convention with the scientific assessments of specific fisheries. Thus they form a fundamental component of a scientific approach to fisheries management.

(iv) Strategic Modelling as a Scientific Basis for Developing Management Strategies  
[return to table of contents]

Applying the ecosystem approach to management presents new scientific challenges. As outlined earlier, this task has to be undertaken in a system with a high level of complexity, even when it is limited to a few key prey and predator species and their interactions with the environment. To make matters more difficult, the scientific research required has to be carried out in a harsh and remote environment, with limited scientific and logistical support. As one way of overcoming these difficulties, CCAMLR has started to develop strategic modelling, using computer simulation, as a tool for setting scientific priorities and developing and evaluating management options.

Strategic modelling relies on the integration of existing computer models used in CCAMLR with new models, which can then be linked together to form ecosystem models. Figure 18 shows an example of how models are linked to form a strategic model for the krill fishery. These integrated models are designed to incorporate the features of an ecosystem that may affect, and may be affected by, conservation and fisheries management. The aim is not to attempt to develop a comprehensive ecosystem model of Antarctica, but rather to develop models that can cast light on particular scientific and management questions. For example, such models can be used to help decide which factors are critical for determining the likely success of a management system for a given fishery, and give guidance on what information is needed to ensure that success. Thus, strategic models can help us to set scientific priorities in terms of critical uncertainties and the scientific resources required to resolve them. As an example, strategic models can be used to answer such questions as how many species and what geographical spread should be monitored to be reasonably certain of detecting adverse effects of krill fishing on dependent species before they exceed those permitted under Article II.

No model of an ecosystem can ever be complete, nor does any one model necessarily include all the important features of an ecosystem. For these reasons a range of models needs to be developed so that the validity of any conclusions drawn from them can be determined.