Estimating the ecological impacts of fisheries: What data are needed to estimate bycatches?

Martín A. Hall

Head, Tuna-Dolphin Programme, Inter-American Tropical Tuna Commission, 8604 La Jolla Shores Dr., La Jolla, CA 92037, USA.
Email: [email protected]

1 INTRODUCTION

In many fisheries, the focus of attention has shifted from the traditional single-species management concerns to a new awareness of effects that may cause far reaching ecological impacts on target and non-target species. The magnitude of these impacts should be estimated and their causes understood in order to achieve the elusive goal of ecosystem management (Alverson et al. 1994, Dayton et al. 1995, Hall 1996). When a fishery places one or more species in danger, or when it has adverse effects on a whole ecological community, management plans to eliminate or mitigate its effects are necessary. These plans require a good knowledge of several elements of the population dynamics, such as recruitment, abundance and mortality of the species involved. Based on these statistics, and guided by the precautionary approach (FAO Code of Conduct for Responsible Fisheries, Article 7.5), targets and objectives can be set to control the problem. Trends in relative or absolute abundance estimates can be monitored to assess the changes and make the policy decisions. Because it is also one of the key elements in determining the effects of fishing on ecosystems, the emphasis of this paper is on the estimation of incidental mortality caused by fishing practices. Although this is largely a statistical problem, the approach discussed here should be useful to the many non-statisticians who are dealing with these issues.

2 ESTIMATING THE INCIDENTAL MORTALITIES OF THE MAIN BYCATCH SPECIES

Very frequently the drive to assess the ecological impacts of a fishery will come from concerns over the fate of one or a few species which are suspected of being taken at unsustainable levels or which are charismatic enough to generate special attention from the public or from managers. When that is the case, the first priority is to get the best possible estimate of that impact on the "main bycatch species." In exceptional cases (e.g. Hall 1998), it is possible to have full coverage of the fleet causing the incidental mortality. More often, however, economic or logistic constraints allow only sampling of the activity of the fleet in question. The discussion of sampling design which follows addresses the following issues: biases, level of precision, representativeness of samples, observer effects, and other uncertainties (cryptic mortality).

The estimate in question will, in some cases, be a composite estimate, with its components being:

Mortality = (number of individuals captured) x (proportion of those captured that die)

In some cases it may be known or assumed that all individuals captured are killed, the proportion becomes one and the estimate is limited to the first term.

2.1 Estimating the numbers of individuals captured

Direct and indirect methods have been used to estimate the number of individuals captured. Direct estimates are those in which the simple average caught or killed per unit of fishing effort is used. An example of this is the number of individuals caught per crab pot or per lobster trap and in this case the unit of effort is reasonably homogeneous. When the bycatch is correlated with some other variable, such as the total catch of a set of pots or nets, ratio or regression estimates can be used (Cochran, 1977). In this case records are kept, not only of the mortality, but also of the catch per unit of effort. For instance, if the bycatch species is a predator of the target species in a fishery, it may be that for ecological reasons there is some relationship between the size of a school of the target species and the number of predators that follow it, feed from it, etc. The knowledge of this relationship may provide a way to improve the estimation of the bycatch, if good estimates for the overall distribution of catches per set, or at least, a cumulative total, can be obtained. In the case of the tuna purse-seine fishery of the eastern Pacific, the average mortality of dolphins per ton of tuna captured was used (Hall and Boyer 1986).

In other fisheries, there is heterogeneity in the use of the gear that affects bycatches (i.e. gillnets of different lengths or trawls hauled for different periods). If the sampling unit is a net set, or a haul, there is no a priori selection of the length or duration, and these are treated as random variables. The probability of an individual being taken is proportional to the area of the gillnet (or to its length if depth is constant), or to the volume filtered by the trawls, which is proportional to the duration of the tow for trawls with similar dimensions, towed at similar speeds. Thus information on the characteristics of the fishing operation increases our ability to estimate the bycatch. If there is information available to extrapolate to total gillnet length, or to total hours trawled, then the sampled ratio of mortality to gillnet length or hours trawled can be used effectively. As the length of the net sampled is a random variable, the sampling process produces both a random mortality and a random length that have to be combined in a ratio.

These methods require an estimate of the average impact of each unit of effort, which can be called the "bycatch per unit of effort" (BPUE) which parallels the concept of catch per unit effort (CPUE). BPUE and CPUE are not equivalent, however, and the effort measure used for both of them in the same fishery may be different (Hall 1996). In the purse-seine fisheries, for example the CPUE can be expressed in units of catch per hour searched, but the BPUE must be expressed in units of mortality per set (or per ton of tuna caught), because a set is the unit of fishing activity that can result in bycatch. Searching for fish doesn't cause mortality.

The estimates of BPUE have to be extrapolated to the total level of "effort" to complete the estimation process. The fisheries literature provides many examples of this type of extrapolation. Again, there are two options, one is used when effort is known, or assumed to be known, without error, and the other is used when there is also error in this estimate.

The new formula can be written as:

Mortality = BPUE x (total level of effort) x (proportion of individuals captured that die).

2.2 Estimating BPUE

The conditions, considerations and caveats relating to the estimation of CPUE and detailed in section 2.1 also apply to the estimation of BPUE.

2.3 Estimating the level of effort

The effort level that has to be estimated or measured is the one used in the extrapolation of the BPUE ratio. It can be obtained from observer data, fisher's logbooks, port records, or landings data. If it is estimated with error, this source of variance must be included in the overall estimate.

2.4 Estimating the proportion of individuals captured that die

When some of the individuals captured survive after having been released, it is necessary to include an estimate of the probability of death for those captured. This value could come from a best guess, or from an experiment performed for that purpose. The variance coming from this term is easy to obtain, but its validity when applied to fishery data needs to be explored.

2.5 Biases

Ratio estimates of the kind described above are biased when sample sizes are small (Cochran 1977), and frequently data obtained from new observer programmes falls into this category. There are several ways, other than the simple one of increasing sample size, to address this issue. These include (a) the use of one of the several formulas with bias-adjusting terms that have been proposed for this purpose (Rao 1969) and (b) the use of a bootstrap correction (Efron 1982, Efron and Tibshirani 1993).

The sample coverages during the early years of the tuna-dolphin programme were very low and there were questions concerning how much coverage was needed to produce reliable estimates, and what happened to the estimates when coverage was very low.

To answer these questions, a simulation approach was used to study the magnitude of the bias, and our ability to adjust for it by using the various formulae available for this purpose. Data were available for a series of years with observer coverage of around 25% - 30%. As there were no clear trends within this period, samples from three consecutive years were combined into one, with roughly the same number of trips as the total for a single year. This became a "simulated universe," whose properties were perfectly known. From this universe, samples of different coverage levels were extracted, and several hundred replicates were made at each level.

After testing several formulae, the three with the lowest biases (the Hartley-Ross and Pascual formulae (Rao 1969)), and one with bootstrap bias correction (Efron and Titbshirani 1993) were selected for a more detailed comparison of performance. Figure 1 shows the results for these three cases. The biases were almost always positive, tending to overestimate the mortality per ton ratio; as sample coverage increases, the bias approaches zero. The bootstrap adjustments for bias, applied to the classical or to the bias-adjusted formulae, are effective in all cases, improving the accuracy of the estimates.

Figure 1. Relative bias (%) in MPT ratio in the estimates of mortality of the eastern spinner dolphin. H-R = Hartley-Ross formula, Clas. = Classical formula, P. = Pascual formula.

2.6 Precision

The precision of the estimates will depend on the three sources of variance previously identified: one coming from the estimate of BPUE, one from the estimate of the level of effort, and one from the estimate of the probability of death. Of these, the level of effort can be known with enough accuracy that the error may be considered negligible. The error of the probability of death is likely to be set in an experiment, so it may become a fixed component, unchanging over time, unless the experiment is repeated or expanded. The sampling design, therefore, will be dictated mostly by the error in the BPUE estimate. The statistical distribution of the bycatches is a complicated issue because of the variety of cases that are routinely observed. This variety ranges from rare species that encounter the gear at random, which probably follow a Poisson distribution, to schooling species that are either absent or present in large numbers. To handle all the variety, it is advisable to use bootstrap methods to estimate the variance of the BPUE (Efron and Tibshirani 1993); again, simulation methods can provide a good approach to produce realistic estimates which are independent of assumptions about the underlying statistical distributions. Variance estimates for the BPUE and the probability of death can be combined to obtain an estimate of variance for mortality from formulae for the variance of a product of random variables (Goodman 1960) or the delta method (Rice 1988), if all three components have associated error.

In most cases the "main bycatch species" will be the reference point for the sampling decisions, including those concerning the level of precision. Figure 2 shows the changes in variance with increasing sample size, obtained from the simple simulation described above. This type of curve makes the trade-offs between precision and cost visible to those involved in decision-making on the subject. Other than at low sampling rates (<10%) it appears that all three formulae give essentially the same answer.

Figure 2. Changes in variance with sampling coverage using mortality of eastern spinner dolphin per ton for a combined data set. Average of 1,000 simulations. Bootstrap estimates M=100. True ratio = 0.026. H-R = Hartley-Ross formula, Clas. = Classical formula, P. = Pascual formula.

2.7 Representativeness

Many factors can affect the representativeness of the data, and may seriously bias bycatch estimates.

2.7.1 Selection of units of effort to be sampled

It is expected that the sampling design follows a pattern established by the researcher (random, systematic, stratified, etc.), but deviations are common. One example of this are the programmes where vessels take observers on a voluntary basis. It is impossible to know if the tendency to volunteer is correlated with some performance level (e.g. only vessels without problems agree to take observers). In international programmes, different national fleets may have different levels of experience, different regulations, etc., and the extrapolation of data from one to the other may not be valid. Logistics may prevent the adequate sampling of some sectors of the fleet (e.g. small vessels that don't have the space to take observers, or vessels operating from ports with difficult or costly access may get a smaller share than they should). Again, for logistical reasons, random or systematic designs may have to be modified to accommodate constraints in the number of observers, their location, etc. Complex stratified sampling designs may be developed which take into account the logistical constraints imposed on the programme by the fishery. Whatever the case, sampling coverage must include the main areas of the fishery and the entire fishing season, to provide the necessary information on possible spatial and temporal sources of heterogeneity. Simulations with real data coming from pilot samples may again help sort these issues.

2.7.2 Observer effect on crew performance

It is possible that the presence of an observer may affect in some way the behaviour of the captain or the performance of the crew. A captain may chose to fish in areas with lower mortality rates when an observer is on board, or the crew may make extra efforts to release alive the individuals captured. Compliance with regulations, and use of safety equipment will most likely be different when they are being reported.

Can these effects be detected? By comparing the fishing areas, species and size composition of the catch, catch rates, trip duration, etc., of the observed sector of the fleet with the rest, it may be possible to test whether the selection of fishing grounds, and other operational decisions depart from the unobserved fleet pattern (assuming that a combination of data, and samples at landing ports could provide these data). But without information on unobserved vessels, comparisons of crew performance in release procedures, compliance with requirements and regulations, etc. are not possible.

2.7.3 Unobserved mortality

The presence of an observer on a vessel is no guarantee that all the mortality will be observed. The reasons for this include; negligence (e.g., not being present during the fishing operation), temporary disability (e.g., the observer may be sick or injured), poor visibility (e.g., sets in darkness or in rough seas), disappearance of the dead individual prior to observation (e.g., dropping from a hook or falling out of a net, taken by a predator from a net or hook, etc.). A special category of unobserved mortality is that of the individuals that die as a result of the fishing operations, as a result of delayed effects (e.g., animals released with internal injuries, or external injuries that were erroneously believed not to lead to mortality, facilitated predation as a result of alterations in schooling or other behaviour as a result of the fishing operation; stress-or fatigue-induced mortality, etc.).

2.7.4 Underreporting of mortality

Even though it is possible to conceive of some circumstances under which an observer may intentionally overestimate mortality (antagonism with crew, overzealous protection of the species involved, etc.), it is much more likely that the mortality will be under-reported. There are three basic motivations that could lead to this: (i) observers spending prolonged trips at sea may develop bonds of friendship with captains and crews, and that may affect their reports; (ii) captains and crews may intimidate the observer, and the observer may underreport out of fear; and (iii) the observers may be bribed to falsify their reports. Of these three, intimidation should disappear when the vessel arrives in port, and the observer has an opportunity to correct the data and report the incident for the corresponding sanction. The other two are more difficult to identify and correct. The records of the observers could be followed over long periods of time, and those consistently reporting below average values, compared to other observers, could be monitored very closely or eliminated from the programme. In such cases their data would be eliminated from some or all the analyses. Besides these statistical checks, there are very few alternatives to detect these biases. Lie detector tests are not fully accurate, and their legal value is not clearly established. Sting operations or observations on spending patterns by some individuals may identify a few guilty parties, but they do not make it possible to quantify the bias. Placing a sub-sample of "trusted" observers or volunteers mixed with the regular pool could help quantify this difference, by comparing the figures from this and the other group.

2.8 Factors affecting BPUE

All the considerations of the previous sections are focused on the objective of obtaining a good estimate of the incidental mortality of the main bycatch species but none of them contributes to understanding its causes. In addition to producing an estimate of a problem's magnitude, an observer programme should serve as a tool to solve the cause of incidental mortality. To achieve this goal it is necessary to identify possible factors that cause the incidental mortality or increase the average BPUE. These factors can be of several types, depending on the fishery. Some can be environmental factors (visibility, sea state, presence of currents, etc.) and others are related to the gear and its deployment (is the right gear used? Is it deployed in the right way?, etc.). Unfortunately, at the beginning of the studies on estimation and mitigation of bycatches it is not known which factors affect BPUE, so it is recommended that a broad approach be taken, trying to include as many factors as possible. The list of factors is potentially very long, but those affecting the ability of the animals to detect the gear, their behaviour, and the behaviour of the gear should be considered first. Once an adequate database becomes available, statistical techniques, such as generalised linear models (McCullagh and Nelder 1989, Stefánsson 1996) can be used to determine which factors are significant.

Acquiring knowledge about the factors affecting BPUE allows one to: (a) improve the estimation of the bycatch levels; and (b) develop mitigation procedures (regulations, technology, education, etc.) to address them. The latter constitutes the basis for most bycatch reduction programmes. It is recommended that input from fishers be given a high priority when establishing the list of factors to consider.

3 ESTIMATING THE MORTALITY OF OTHER, MOSTLY NON-TARGET, SPECIES

Most of the problems, and solutions, mentioned for the target species apply for the other components of the bycatch. If the sampling design is based on the main bycatch species, however, it is quite possible that the estimates for other species have broader confidence intervals, and may be biased. This could be a serious problem if the bycatch of the "main" species has a uniform or random distribution, and some of the secondary species are very patchy in their distribution.

4 SPATIAL AND TEMPORAL DISTRIBUTION OF THE BYCATCHES

Knowledge of the spatial and temporal distribution of the bycatches is crucial to the quality of the estimates, and to the mitigation programmes. From the estimation point of view, the stratification of the data into spatial and temporal units that reflect real heterogeneities, will be an effective way to improve the estimates and reduce the costs of the sampling programmes. From the point of view of the mitigation programmes, it provides a quick assessment of the feasibility of spatial and temporal closures as mitigation measures.

5 RELATIVE OR ABSOLUTE ABUNDANCE OF THE SPECIES TAKEN

To put in perspective the impact of a fishery, it is very important to compare the level of mortality with the population size, and its net recruitment. The "relative mortality" is the ratio of mortality to population size. These data may come from the fishery, but most commonly will require special surveys, tagging experiments, or other procedures. Without them, the mortality data have only limited value because the assessment of their significance is left to the "gut feelings" of those interested. Time series of BPUE data could be used to monitor trends in the populations taken, but only after the same careful procedures that should be used in the interpretation of CPUE data. An additional problem in the use of BPUE data is that many possible actions taken to reduce bycatches would result in lowering the BPUE without reflecting any population changes. A system where the index of abundance is also a performance measure is not likely to be very informative over time because the trends in the population will become confounded with the performance changes that may be the objective of management.

6 BYCATCH TO CATCH RATIOS

Bycatches should be expressed as a function of the catches in the same fishing operations to facilitate the comparison among areas, gears, etc. Some ratio estimates may require the catches In other cases the catches will put the bycatches into perspective by showing the ecological costs of different operations under a comparable standard (Hall 1996.)

7 CONCLUSIONS

This paper provides a brief description of the data requirements to implement an effective bycatch mitigation programme. The value of simulations performed on real data from pilot samples is emphasised as a tool to provide statistical insights into the problem without the need for complex theoretical analyses. The use of resampling techniques to deal with bias and precision problems is also proposed as a major component of the estimation process. Finally, in order to contribute to the solution of bycatch problems, the exploration of the causes of the bycatch must be an integral part of the sampling scheme.

REFERENCES

ALVERSON, D.L., FREEBERG, M.H., MURAWSKI, S.A. & POPE, J.G. 1994. A global assessment of fisheries bycatch and discards. F.A.O. Fish. Tech. Paper No. 339, Rome, 233 pp.

COCHRAN, W.G. 1977. Sampling Techniques. 3^rd ed. J. Wiley & Sons, London, 428 pp.

DAYTON, P.K., THRUSH, S.F., AGARDY, M.T. & HOFMAN, R.J. 1995. Environmental effects of marine fishing. Aquatic conservation. Mar. Freshw. Ecosystems, 5: 205-232.

EFRON, B. 1982. The jack-knife, the bootstrap and other resampling plans. CBMS Regional Conference Series in Applied Mathematics 38, S.I.A.M., Philadelphia, 92 pp.

EFRON, B., & TIBSHIRANI, R.J. 1993. An introduction to the bootstrap. Monographs on Statistics and Applied Probability 57, Chapman and Hall, New York, 436 pp.

HALL, M.A. 1996. On bycatches. Rev. Fish Biol. Fisheries, 6: 319-352.

HALL, M.A. 1998. An ecological view of the tuna-dolphin problem: impacts and trade-offs. Rev. Fish Biol. Fisheries, 8: 1-34.

HALL, M.A. & BOYER, S.D. 1986. Incidental mortality of dolphins in the eastern tropical Pacific tuna fishery: description of a new method and estimation of 1984 mortality. Rep. Int. Whal. Commn., 36: 375-381.

GOODMAN, L.A. 1960. On the exact variance of products. American Statistical Association Journal, 55: 708-713.

MCCULLAGH, P. & NELDER, J.A. FRS. 1989. Generalized Linear Models, Second edition. Chapman & Hall, London, 511 pp.

RAO, J.N.K. 1969. Ratio and regression estimators. In N.L. JOHNSON & H. SMITH Jr., eds., New developments in survey sampling. Wiley Interscience, London. 732 pp.

RICE, J.A. 1988. Mathematical Statistics and Data Analysis. Wadsworth & Brooks/Cole, Pacific Grove, California, 595 pp.

STEFÁNSSON, G. 1996. Analysis of groundfish survey abundance data: combining the GLM and delta approaches. ICES J. Mar. Sci., 53: 577-588.

Monitoring tuna fisheries in the western Pacific

Antony D. Lewis

Oceanic Fisheries Coordinator, Secretariat of the Pacific Community, Ocean Fisheries Programme, P.O.Box D5, Noumea Cedex 98848, New Caledonia.
Email: [email protected]

Abstract: Existing arrangements for monitoring the large western pacific tuna fishery, primarily for scientific purposes, are described. The annual catch of 1.2-1.4 million tonnes of the four main tuna species is taken by around 7000 vessels. Primary coverage of the catch and effort by domestic and distant water fleets is achieved through logsheet coverage at national level, and is currently 80% of the catch in the Secretariat of the Pacific Community area. Port sampling and observer programmes have been established relatively recently, and provide limited coverage. The various data are compiled by the SPC/Oceanic Fisheries Programme (OFP) and maintained in an accessible Regional Tuna Fishery Database. Current levels of monitoring are not adequate for some purposes, with other types of data increasingly being required from the fishery and its operational environment. It is hoped that an arrangement being developed for the conservation and management of regional highly migratory fish stocks will provide a framework for increased monitoring efforts.

1 THE WESTERN PACIFIC TUNA FISHERY

Fisheries that capture tuna in the western Pacific Ocean range over a vast area, eastwards from the coasts of Asia out into the Central Pacific, and from northern Japan, at nearly 50° N, to similar latitudes in the Southern Hemisphere. The major volume of the catch is however, taken in tropical waters, and most detailed statistics until recently, have been compiled for the so-called South Pacific Commission (now Secretariat of the Pacific Community (SPC)) statistical area (Fig. 1), which covers 30 million km² of ocean including the EEZs of 24 coastal states and large areas of high seas. Recognising the need to collect information throughout the range of tuna stocks exploited by the fishery, there has been an increasing tendency to collect statistics over a wider area, from 50° N to 50° S and 120° E to 150° W (Fig. 1). The area, conveniently called the Western and Central Pacific Ocean (WCPO), was recently adopted for the collection of statistical data by the 11th Standing Committee on Tuna and Billfish (Anon. 1998). The breadth of this area at the Equator represents one quarter of the earth's circumference.

The WCPO tuna fishery is now the world's largest, taking between 1.2 and 1.4 million tonnes of the four main market species of tuna (skipjack tuna (Katsuwonus pelamis), yellowfin (Thunnus albacores), bigeye tuna (Thunnus obesus) and albacore tuna (Thunnus alalunga)), as well as some hundreds of thousands of tonnes of tuna-like species, such as frigate and bullet tunas (Auxis spp.), mackerel tuna (Euthynnus affinis) and a range of non-target, associated and dependent species (NADS). Figure 2 shows the evolution of the western Pacific tuna catch, with a rapid rise in the total catch during the 1980s, due to the development of the international purse seine fleet, and relative stability since 1990.

Figure 1. The Western and Central Pacific Ocean, showing the WCPO boundaries (heavy and dotted line) and the SPC statistical area (thin line). EEZ areas are clear and high seas shaded.

In recent years, the SPC statistical area has contributed 65% of the WCPO tuna catch, the domestic fisheries of eastern Indonesia and the Philippines, 25% of the catch, and the balance (10%) mostly in the temperate domestic fisheries of Japan and Taiwan. In the SPC statistical area, most of the catch is taken by purse seine gear (80%), followed by longline gear (12%), and lesser amounts by pole-and-line which was once the dominant fishery, troll gear and a variety of artisanal fisheries. The Indonesian/Philippine fisheries involve a range of artisanal gear types, often unclassified. Statistics from these fisheries are seriously incomplete.

The tuna catch is still mostly taken by the mobile international fleets of the distant water fishing nations (DWFNs); Japan, Republic of Korea, Taiwan and the United States of America. Table 1 lists the number of vessels by gear type active in the fishery in the SPC area. This is close to 1,500 for the SPC area, comprising around 180 large purse seine vessels, 1,200 longliners of various sizes and operational modes, 160 pole-and-line vessels and a small number of troll vessels. Details of vessels active in the SPC area are generally available on the Regional Register of Fishing Vessels maintained by the Forum Fisheries Agency (FFA). The number of vessels active in the WCPO probably approaches 7000, the large number of additional vessels consisting mostly of Japanese, Indonesian and Philippine coastal vessels (OFP 1998b).

Figure 2. Annual tuna catch, 1970 - 1997, by gear type, in the Western and Central Pacific Ocean. Catches of skipjack, yellowfin, bigeye and albacore tuna, the four main target species, are included.

Skipjack is the dominant species in the total catch (>60%), followed by yellowfin (20-30%), with smaller amounts of bigeye and albacore (Table 2). Skipjack are caught almost entirely in the surface fishery (purse seine and pole-and-line), whereas the other species are taken in both surface and longline fisheries (Table 3).

Unlike most other ocean areas where important tuna fisheries exist, the majority of the WCPO tuna catch is taken within the Economic Exclusion Zones (EEZ) of coastal states. This proportion (within-EEZ Vs high seas) varies by gear type, with the equatorial purse seine fishery taking a relatively high proportion of its catch in the EEZs of coastal states, and the widely distributed longline fishery taking the majority of its catch on the high seas. The WCPO tuna catch, although widely distributed, is concentrated in equatorial areas, 10° N to 10° S, as this is the main operational area of the purse seine fleet, which dominates catches by weight. The surface fisheries also show meridional shifts in concert with large-scale oceanographic events, notably the El Niño/Southern Oscillation (ENSO) phenomenon (Lehodey et al. 1997). The spatio-temporal distribution of fishing effort also has seasonal components and is affected by a range of other factors such as the existence of access agreements, market forces etc.. The fishery as a result is very dynamic.

Table 1. Number of active vessels in the Secretariat of the Pacific Community (SPC) statistical area since 1990 (1997 figures may be incomplete).

Table 2. Tuna catch by species in the Western and Central Pacific Ocean (WCPO), since 1990.

Table 3. Catch of tuna species by gear in the Western and Central Pacific Ocean (WCPO), in 1997. Catches in tonnes; albacore catches are for the Pacific south of the Equator.

The WCPO tuna catch supplies numerous markets and is unloaded or transhipped at many locations. The great majority of the purse seine catch is canned after unloading, either directly to canneries in Pago Pago (American Samoa), Philippines, Thailand and Japan, or transhipped from regional ports to those and other canneries further abroad e.g. Puerto Rico, Europe. The longline catch, mostly destined for the higher priced sashimi markets of Japan, is either airfreighted fresh from a variety of locations throughout the region, or in the case of frozen catch from larger, long-trip, longline vessels, unloaded direct or transhipped frozen, mostly to Japan. A relatively small proportion of the catch enters local markets, although this is increasing in the case of by-catch and lower grade tunas.

2 MONITORING THE FISHERY

Monitoring activity is focused in three main areas;

1. Monitoring catches; to address a variety of industrial and scientific needs;
2. Monitoring vessel activity; usually to meet compliance requirements, but also as an integral part of monitoring catches and interpreting catch rates, and
3. Electronic monitoring; a relatively recent development for both compliance and enforcement needs.

2.1 Why monitor?

The Agreement for the Implementation of the Provisions of the United Nations Convention on the Law of the Sea of 10 December 1982, relating to the Conservation and Management of Straddling Fish Stocks and Highly Migratory Fish Stocks, otherwise known as the UN Implementing Agreement (UNIA), reaffirms generally accepted requirements to monitor catches and vessel activity, in support of conservation and management objectives. These include requirements to, inter alia,

1. "Ensure" that measures (to ensure the long term sustainability of straddling fish stocks (SFS) and highly migratory fish stocks (HMFS) are based on the best scientific advice available;
2. Assess the impacts of fishing, other human activities and environmental factors on target stocks and species belonging to the same ecosystem or associated with, or dependent on, the target stocks;
3. Apply the precautionary approach in accordance with Article 6 of UNIA;
4. Take measures to prevent or eliminate overfishing and excess fishing capacity and to ensure that levels of fishing effort do not exceed those commensurate with the sustainable use of fishery resources;
5. Take into account the interests of artisanal and subsistence fishers;
6. Collect and share, in a timely manner, complete and accurate data concerning fishing activities on, inter alia, vessel position, catch of target and non-target species and fishing effort, as set out in Annex 1 of UNIA as well as information from national and international research programmes, and
7. Implement and enforce conservation and management measures through effective "monitoring, control and surveillance".

2.2 Who monitors?

Under flag state obligations, States are required to collect and verify various fishery data from vessels flying their flag. Data are also collected from foreign vessels fishing in EEZs under access agreements by the coastal state providing such access. Such data can be generally made available to appropriate regional fishery organisations in agreed formats, given the highly mobile nature of both fleets and fish stocks, and the need for central co-ordination of such information. In the western Pacific, the 22 Pacific Island SPC member states, as well as so-called metropolitan countries (Australia, France, New Zealand, USA, and the United Kingdom) have long delegated the co-ordinating role for oceanic fishery statistics in the SPC area, to the SPC. These statistics are published as quarterly Bulletins and as a Tuna Fishery Yearbook, and are also made available in other ways (Oceanic Fisheries Programme 1997, 1998a, 1998b).

Under the aegis of the Standing Committee on Tuna and Billfish, statistics have more recently been collated from a wider area more closely approximating the known range of the stocks of at least several of the main tuna species i.e. 120° E to 150° W, and 50° N to 50° S, as noted. This area is contiguous with those of other regional organisations monitoring, collecting and compiling similar statistics; the Indian Ocean Tuna Commission (IOTC) in the west and south-west, and the Inter-American Tropical Tuna Commission (IATTC), whose area of interest extends west to 150° W, in the east. In areas mostly to the south, the Commission for the Conservation of Southern Bluefin Tuna (CCSBT) manages the temperate water southern bluefin tuna throughout the range of the stock, and even further south, starting at 60° S, the Commission for the Conservation of Antarctic Marine Living Resources (CCAMLR) monitors those resources, which however do not include tuna.

2.3 What is monitored?

The basic fishery data collected as part of monitoring activity are listed in the UNIA Annex 1 and include the following:

1. Time series of catch and effort data, by fishery (gear) and fleet;
2. Total catch in number, weight or both, by species (target and non-target) and fishery;
3. Discard statistics, including estimates where necessary;
4. Effort statistics and fishing location;
5. Catch composition by length, weight and sex;
6. Other biological data, such as information on age, growth, recruitment, and stock distribution, and
7. Other relevant research data (including studies on environmental, oceanographic and ecological factors affecting stock abundance).

Data verification mechanisms, including vessel monitoring systems (VMS), observer programmes, vessel reports and port sampling are also specified. The situation in the WCPO is described below.

2.3.1 Catch effort data

Since much of the catch in the SPC area is taken by DWFNs licensed to fish in the EEZs of coastal states, catch and effort data are supplied under bilateral access agreements to the country concerned, then sent to SPC for entry and verification. Catch and effort for the entire treaty area (EEZs and high seas) are provided under the United States Multilateral Treaty on Fisheries via the Forum Fisheries Agency as treaty administrator. Catch and effort data are also provided directly from domestic flag vessels. Data on high seas catches in the SPC area are in some cases provided with bilateral access data to adjacent coastal states, and in the case of Japan, as aggregated five or one degree square data. In most cases, the catch effort data are provided on standardised regional data forms (logsheets).

In recent years, the coverage of tuna catches in the SPC area by catch and effort logsheet data has been 80%. A significant improvement in this coverage came about with the adoption of a set of Minimum Terms and Conditions of Access by the Parties to the Nauru Arrangement, usually known as the PNA group, in 1993. These prohibited high seas transhipments (by purse seine vessels in the first instance), requiring such activity to take place in designated ports in the region. Coverage of purse seine catches improved markedly for several fleets thereafter. The main gaps in the logsheet coverage now concern high seas catches by the fleets, mostly longline, of some countries, and the domestic fleets (various gears) of some coastal states.

Data on effort statistics appropriate to each gear are collected on standardised logsheets for the SPC statistical area, as are data on fishing location, date, time fished and other operational data. Additional information on vessel and fishing gear specifications are available either on the logsheets or from the Regional Register of Fishing Vessels maintained by FFA. These data may be used in attempts to standardise fishing effort as required.

The logsheet catch and effort data are compiled and maintained in the Regional Tuna Fishery Database (RTFD), which now includes over 7 million records of fishing operations. Access to these data is covered by a confidentiality agreement which allows access for research purposes to data aggregated at an agreed level, usually 5 degree squares by month for longline data, and one degree squares by month for purse seine and pole-and-line data. The SPC/OFP, with the approval of countries supplying such data, has discretionary powers in the release of such data. Unaggregated logsheet data, because of their commercial sensitivity, are made available from the RTFD only under strict conditions and with authorisation from the original sources of the data.

In the wider WCPO area, logsheet coverage of the domestic fleets of eastern Indonesia and the Philippines, largely comprising of artisanal-scale vessels, is poor or even non-existent, even though these fleets contribute 25% of the total WCPO catch, based on estimated landings.

2.3.2 Total catch

Logsheet coverage of each fishery or gear is typically incomplete to some degree, and estimates of total catch are usually derived from landings data and industry sources. The SPC Tuna Fishery Yearbook, published since 1993 and now also available on the SPC website, lists annual total catch, by flag, gear and species, for the four main commercial species of tuna taken in the SPC statistical area. The best available information for tuna catches in eastern Indonesia and the Philippines is also included. The 1997 Yearbook was completed in October 1998 (Lawson 1998). Total catch estimates for the main tuna species in the western Pacific tuna fishery since 1990 are presented in Table 2.

Total catch estimates should include both target and non-target species, as well as discards i.e. total removals from the fishery, but as by-catch is rarely recorded on logbooks, and discards, including both by-catch species and small or damaged individuals of target tuna species, are also rarely recorded, such estimates are generally not available for the western Pacific tuna fishery.

Bailey et al. (1996) attempted to document by-catch and discard practices in the WCPO tuna fishery from available logsheet data, observer information and published reports. They estimated that purse seine by-catch may be in the range 0.35% to 7.3% of the total catch by weight, depending on school association. That by-catch in longline fisheries was typically much higher but could not be reliably estimated from available data, and that by-catch and discards in pole-and-line, troll and handline fisheries are relatively minor. Lawson (1998), drawing on a larger observer database (1992-97), estimated that discards and by-catch from the four main purse seine fleets (1994-96) comprised 3.46% and 0.71 % respectively of the total weight, and for selected longline fleets (1992-97), discards and by-catch comprised 3.8% and 42.5%, respectively, of the total catch by weight. A comparison with the Eastern Pacific Ocean (Hall and Williams in press), where observer coverage is much higher, suggest that levels of by-catch and discards are both much higher in this fishery.

2.3.3 Catch composition data

The size and species composition of the catch, other than indicative weight data on some logsheets for the four main tuna species, is mainly obtained from port sampling of the catch. This also affords the opportunity to corroborate declared catches on logsheets in some cases.

Although port sampling is now carried out at over 20 unloading or transhipment points in the region with the co-operation of national fisheries departments, coverage (in this sense, coverage is taken to mean the proportion of unloading vessels sampled) of the total catch remains low and patchy. Under the provisions of the United States Multilateral Treaty on Fisheries USMLT, the US purse seine catch is sampled by National Marine Fisheries Service (NMFS) staff in Pago Pago. Other purse seine fleets are sampled intermittently in the region, with the exception of the Japanese fleet which largely unloads in Japanese ports and is subject to sampling by the National Research Institute of Far Seas Fisheries (NRIFSF) programmes. Domestic longliners enjoy extensive coverage by port sampling programmes, but large freezer longliners, which provide the bulk of the longline catch and mostly unload in Asian ports, receive little coverage, with the exception of some albacore longliners unloading in Fiji and Pago Pago. Pole-and-line and troll landings receive limited coverage.

One problem affecting the representative nature of purse seine sampling is the practice of high grading i.e. sorting and storing the catch according to size or species, to meet differing market requirements. This is particularly the case with larger yellowfin tuna, where wells sampled in port during transhipment or unloading may not be representative of the total or stratified catch. Without access to detailed well loading data, assuming such data are maintained, this problem will be difficult to overcome without extensive and costly observer coverage.

One benefit of port sampling purse seine catches is that it enables an important species identification problem to be addressed, that being the separation of juvenile yellowfin and bigeye. These are not separated in catch reporting, and are not always readily distinguishable to the untrained eye. Port sampling remains the primary source of information on the occurrence of juvenile bigeye in the catch, an issue of increasing importance as concerns surrounding levels of bigeye exploitation increase.

Port sampling provides limited information on retained by-catch species, especially those from the longline catch (e.g. billfish, mahi mahi, wahoo, opah, escolar etc.), but no information on discards, by definition. Observers remain the primary source of information on total catch composition.

2.3.4 Observers

The value of observer activity, scientific and/or compliance, in collecting a range of information not otherwise available, but increasingly required, on total catch and size composition by species, discards, technology change etc. is well recognised. Observer programmes to meet national requirements have long been established in several countries in the region e.g. Australia, New Zealand and since 1981, the Federated States of Micronesia, but regional observer programmes are a relatively recent development. Under the provisions of the United States Multilateral Treaty on Fisheries, an observer programme was established by FFA, as treaty administrator, in 1988 with the goal of achieving 20% trip coverage of the US purse seine fleet. Ten years of data are now available from this successful programme. The SPC/OFP, with European Union (EU) funding support has also operated a scientific observer programme, based on four full-time and other part-time observers, since 1994. Although some coverage of all fleets and gear operating in the SPC area has been achieved, overall coverage of trips by the large number of vessels (circa 1500) remains low (<2%). National observer programmes have, or are being, established in several countries, supported in some cases by financial provisions built into access agreements, and in terms of training and technical support, by FFA and SPC. Coverage is likely to remain low in the short to medium term, given the availability of resources. It is estimated that achieving reasonable coverage of distant water fleets (assuming 20% coverage for purse seine and 10% for longline and pole-and-line), covering around 800 vessels, and utilising Pacific Island staff (requiring an estimated pool of 200 observers) might cost as much as US$ 2.8 million.

The existing programmes, although limited in scope, have already produced a wealth of information and have allowed levels of by-catch in the western Pacific tuna fishery, by species, to be realistically estimated for the first time.

2.3.5 Other data

A range of other biological and research data, not generally regarded as monitoring, may nonetheless be useful in support of stock assessment. These could include information on distribution and stock structure, age and growth, data from tagging (mark-recapture) experiments, oceanographic and environmental variables known to influence distribution, abundance and catchability, and surveys of stock abundance (acoustic, aerial etc.).

2.3.6 VMS/surveillance

Electronic vessel monitoring systems (VMS) have been in place for some time at the national level in the region (e.g. Australia, New Zealand) and attempts are ongoing by the FFA to introduce, with the support of member countries and DWFNs, a harmonised VMS system for vessels operating in the region. Apart from the rather complex legal and political aspects of this initiative, such a system when implemented will have considerable implications for vessel monitoring, both in temporal and spatial dimensions, and will supplement existing logsheet-based data collection systems. It is doubtful however if VMS, despite its near-real time advantages, will completely replace logsheet-based data collection in the short term - catch data supplied in both cases are unverified but logsheets provide hard-copy back-up, and VMS coverage is never likely to be complete e.g. inclusive of domestic and artisanal vessels. Similarly, observer programmes provide data verification, and additional information on operations, by-catch, discards etc. which are generally not feasible for fishermen to collect and supply.

2.4 Is the current level of monitoring of the western Pacific tuna fishery sufficient?

Current levels of monitoring for primarily scientific purposes are probably adequate for detailing general trends in target tuna catches by gear, fleet and area. They also appear to be adequate for monitoring significant changes in catch rates, and for gross changes in the status of tuna stocks on a regional basis. Regional stock assessments are currently available for three of the four main tuna species exploited in the fishery, and in all cases suggest that current levels of exploitation are low to moderate.

Trends in catch rates of target tuna species are monitored for selected gear/fleet combinations (and some efforts have been made to standardise fishing effort (Hampton et al. 1999). With the adoption of the precautionary approach and reference point-based management, data requirements can be expected to increase, with the need to incorporate estimates of risk and uncertainty in assessments, and with the possible application of complex data-intensive operational models. Key areas where data collection/monitoring of western Pacific tuna fisheries could be improved include:

1. Improved statistical coverage of the fisheries of eastern Indonesia and the Philippines;
2. Consolidated size composition data for most species throughout the range of their geographical distribution, and
3. Improved understanding of biological processes affecting abundance, catchability etc.

The situation with species other than the target tuna species i.e. non-target, associated and dependent species, is very much data-poor, and a considerable increase in present levels of monitoring of these components of the total catch would be necessary to even begin to consider impacts of fishing on their stocks, let alone comprehensive stock assessments of this range of species. This would ideally require a significant increase in observer-based activity, but the observer programme is currently very much under-resourced, and unlikely to improve significantly. If this is to be regarded as a priority activity, other options, such as the retention and/or reporting of all catch and subsequent sampling of the catch, along with dedicated at-sea research programmes would need to be considered.

Currently, considerably less than 2% of the landed value of the tuna resource in the SPC area is expended on monitoring, research and management-related activity, a figure which is low by most standards. Some intensively managed fisheries report research and monitoring expenditure of between 5% and 8% of the landed catch value. In addition, there is presently little or no cost recovery in the western Pacific of research/monitoring costs, these generally being met from development assistance funds. Not surprisingly, it is only a small proportion of the access fees/resource rental accruing to coastal states (5% of landed value on average) which is applied to such activity.

Coastal states and fishing nations in the WCPO have embarked, since December 1994, on a process to develop a regional mechanism for the conservation and management of highly migratory fish stocks in the western and central Pacific. Three high-level consultations involving all parties with an interest in these resources, coastal states and fishing nations alike, have been convened, with the third considering draft articles for a Convention for the conservation and management of these resources. The goal is to establish, possibly by June 2000, a Commission to support the operation of such a Convention. This would provide a framework for future co-ordinated monitoring of the WCPO tuna fishery, and represents possibly the most realistic opportunity to increase current monitoring efforts to the required levels.

REFERENCES

ANON. 1998. Report of the Eleventh Meeting of the Standing Committee on Tuna and Billfish, 28 May - 6 June 1998, Honolulu, Hawaii. August 1998, 108 p.

BAILEY, K., WILLIAMS, P.G. & ITANO, D.G. 1996. By-catch and discards in western Pacific tuna fisheries: a review of SPC data holdings and literature. SPC Oceanic Fisheries Programme Technical Report No. 34. Secretariat of the Pacific Community, Ocean Fisheries Programme, P.O.Box D5, Noumea Cedex 98848, New Caledonia.

HALL, M.A. & WILLIAMS, P.G. (in press). Bycatches in Pacific tuna fisheries. National Coalition for Marine Conservation: Symposium on Managing Highly Migratory Fish of the Pacific Ocean, November 4-6, 1996, Monterey, California.

HAMPTON, J., BIGELOW, K. & LABELLE, M. 1999. A summary of current information on the biology, fisheries and stock assessment on bigeye tuna (Thunnus obesus) in the Pacific Ocean, with recommendations for data requirements and future research. SPC Technical Report No.36. Secretariat of the Pacific Community, Ocean Fisheries Programme, P.O.Box D5, Noumea Cedex 98848, New Caledonia.

LAWSON, T.A. 1998. Tuna Fishery Yearbook 1997. Secretariat of the Pacific Community, Noumea, New Caledonia. 133p.

LEHODEY, P., BERTIGNAC, M, HAMPTON, J., LEWIS, A. & PICAUT, J. 1997. El Niño Southern Oscillation and tuna in the western Pacific. Nature 389: 715-718.

OCEANIC FISHERIES PROGRAMME 1997. Estimates of bycatch and discards in central and western Pacific tuna fisheries; preliminary results. SPC/OFP Internal Report No. 33, 30p.

OCEANIC FISHERIES PROGRAMME 1998a. Coverage of western and central Pacific tuna fisheries by data held by the SPC Oceanic Fisheries Programme. 11th Meeting of the Standing Committee on Tuna and Billfish, 30 May - 6 June, Honolulu, Hawaii. Working Paper 4, 21p.

OCEANIC FISHERIES PROGRAMME 1998b. Estimates of annual catches of target species in tuna fisheries of the western and central Pacific Ocean. Ibid, Working Paper 5, 67p.

Sampling and estimation of discards in multi-species fisheries

Tatsuro Matsuoka

Kagoshima University, Faculty of Fisheries, Shimoarata 4-50-20, Kagoshima 890-0056, Japan.
Email: [email protected]

1 INTRODUCTION

The assessment of bycatch discards is one of the most important tasks in monitoring capture fisheries. Discards in each sector must be evaluated accurately in order that problems unique to each sector are identified and solved. As it is nearly impossible to observe all discarding practices across the diverse range of fishing activities, the discarded amount by fishery is usually statistically estimated on the basis of partial coverage of onboard sampling. The methodology for discard estimation, however, has been confused due to misunderstanding the characteristics of discard practices in multi-species fisheries. A large portion of world capture fisheries is characterised by small-scale, multi-species and multi-gear fisheries particularly in tropical and subtropical coastal waters. These fisheries are seldomly monitored by onboard observers at high coverage ratios. Given the large, relative size of this fishery sector, it is therefore, of immediate concern to establish a methodology for the estimation of discards, in order to assess their impacts on the stocks and the environment of these fisheries. This paper reviews methods for discard estimation, analyses the nature of discards in multi-species fisheries and introduces a new approach for discard estimation and assessment.

2 METHODOLOGICAL REVIEW OF DISCARD ESTIMATION

The majority of discard analyses have assumed a positive correlation between discard and landing amounts (FAO 1996). This is reflected in the methodology that adopts the use of discard ratios, which is generalised as the mean value of the ratio of discarded catch to retained catch, estimating total discard, r, as;

r = E (d / c) (1)

where d represents discarded and c represents retained catch in a fishing sector, determined through sampling. A discard amount, D in a sector is described as;

D = r · C (2)

where C is the total catch landed in the sector.

Using these principles, discard ratios have been applied widely to estimate total discard in a number of different ways. The variables d, c, D and C may either be the sum of all species, target species or each species which can make such calculations confusing. In this paper, the term `sum' refers to amalgamation of species and `total' refers to amalgamation over a sector. A `sector' is defined as a unit of a capture fishery which is uniquely identified by a fishing gear and method and a typical catch composition of harvested species.

2.1 Estimation of discards from summed landings

In an earlier paper (Matsuoka 1996), a realistic approach to the definition of the discard ratio was recommended to avoid various limiting factors in discard estimation. The proposed discard ratio was based on the summed discard, d_i, and summed retention, c_i, in a fishing sector-i with the summed discard amount, D_i, in a sector estimated as a function of the discard ratio, r_i and the summed landing, C_i;

r_i= E (d_i / c_i) (3)

D_i = r_i · C_i (4)

The estimation procedure recognises that discard ratios must be based on the factor which has the best correlation with the discarded amount in a sector. The ratio is also easy to apply and, therefore, a useful tool due to the fact that summed landings in each sector are generally available for most fisheries.

The procedure also allows an estimation of individual species discard to be made once reliable data, based on summed landings, are available. When a specific discard ratio, r_ij is available, a specific discard amount, D_ij of the j-th species in the i-th sector may be estimated by;

r_ij = E (d_ij / c_i) (5)

D_ij = r_ij · C_i (6)

where d_ij is the observed discard of species-j within a range of sector-i.

The latter part of this estimation procedure is re-evaluated in part in this paper.

2.2 Estimation of discards from the landing of target species

The method applied in Alverson et al. (1994) to provide a global estimate of discards was:

r_(t) = E (d_(t) / c_t) (7)

D_(t) = r_{(t) ·} C_t (8)

where c_t is the retained catch of target species-t during sampling, d_(t) is the summed discards during the sampling period when species-t is targeted, r _(t) is the discard ratio, C_t is the total landing of species-t over all sectors and D_(t) is the estimated, summed discards when species-t is targeted. This relationship holds when one sector is analysed with landing statistics which are independent and not affected by other sectors. When applied, however, to many fisheries and multi-species fisheries in particular, this method was found to over-estimate the total discards (Matsuoka 1996). It is therefore necessary to review the global discard estimate of 27 million tonnes produced by this method given the inaccuracies of this model.

2.3 Estimation of discards by fishing gear type

Alverson and Hughes (1995) revised their method to estimate specific discards, i.e. D_ij of each species-j in each sector-i. Using this method, discard ratios, r_ij are defined for species-j which are retained, c_ij, and discarded, d_ij, by fishing gear-i;

r_ij = E (d_ij / c_ij) (9)

D_ij = r_ij · C_ij (10)

Although this model seems an ideal approach to the evaluation of discards in detail, it does not apply particularly to multi-species fisheries both theoretically and in practice.

2.4 Discard per unit effort (DPUE)

A method of discard estimation dependent on fishing effort presents a solution to the problems experienced with estimates derived from other sources (Matsuoka 1998). The discard per unit effort (DPUE) is simply defined as the mean value of the discarded amount per unit fishing effort, e.g. per vessel per day. Discards, D_i, in sector-i are estimated on the basis of DPUE, q_i and the total fishing effort, E_i as;

D_i = E_i · q_i (11)

This method may also be used to estimate discard by species-j and sector-i if appropriate data for q_ij are available;

D_ij = E_i · q_ij (12)

3 CATCH AND DISCARDS IN MULTI-SPECIES FISHERIES

A small-scale, coastal Danish seine fishery in southern Japan was surveyed to examine the applicability of differently defined discard ratios and to consider appropriate methodologies for the estimation of discards. Occurrences of species retained and discarded were analysed to identify the characters of discard practices in multi-species and multi-gear fisheries. This fishery is composed of four sub-sectors which target red seabream, cuttlefish, conger eel and assorted finfish using different gear types and mesh sizes, respectively. On the basis of these characters this fishery is assumed to be a good model of a multi-species and multi-gear fishery.

3.1 Species composition of the Danish seine fishery in southern Japan

A total of 145 finfish and crustacean species were recorded from 37 sampling occasions over a period of three years. A total of 97 species were recorded as being regularly or occasionally landed and 48 were recorded as regular discards. The top ten species in landings accounted for 75% by weight of the total catch retained. In each sub-sector, 24 to 31 species were landed. In these landings the target species constituted between 28% and 48% of the retained catch. These data indicate the fact that the landings of target species or any other species of commercial importance are not representative of summed landings. The target species concept is inapplicable and, therefore, ineffective. Between 58 and 79 species were discarded in each sub-sector. Discard of each species did not exceed 10% of the summed discards in each sub-sector and there was no major discard species.

3.2 Operational practices for uncommon species

During 14 periods of observation in the cuttlefish sub-sector, 29 species were retained and 79 species were discarded (Table 1). A total of 17 species were common to both retained and discarded catches and 72 species occurred either in retained or discarded catch. The 17 species occurring in both retained and discarded catches comprised 79% of the retained catch and 24% of the discarded catch by weight. The remaining 62 species comprised 76% of the discarded catch by weight and did not occur in the retained catch. The correspondence between the species composition of retained and discarded catch was, therefore, very poor.

In order that an estimate of discard can be based on individual species discard ratios, r_ij, discard species have to occur in both the retained and discarded catches. In this instance the discard ratio, r_ij, could not be evaluated for the 62 species which constituted three-quarters of the summed discards. As multi-species fisheries are characterised by the fact that the majority of discard species are never landed, discard estimates based on individual species discard ratios, r_ij, are impossible to conduct in many cases.

Table 1. Common species occurring in discards and landings by number and weight, recorded on 14 sampling occasions over 3 years, from the cuttlefish sub-sector of a Danish seine fishery in southern Japan.

3.3 Correlation of discards and landings

The correlation between discarded and retained catches was examined for data from the cuttlefish sub-sector. The correlation coefficient of summed discards to summed landings was 0.661 and that of the summed discards to the retained target-species catch (cuttlefish), was 0.617. This was lower than the correlation of the summed discards with the summed landings.

The individual species correlations of the 17 species common to both the retained and discarded catch ranged from -0.252 to 0.649 with an average of 0.111 (Fig. 1). The correlation of individual species discarded to the summed landings ranged from -0.479 to 0.593 with an average of 0.106. It is noted that the correlation coefficients of discards to landings were negative in many cases. This implies that there is no scientific basis to assume a proportional relationship between discards and landings. The correlation of individual species discards to the target catch retained was also investigated. The correlation coefficients ranged from -0.474 to 0.658 with an average of 0.143 recorded. This assessment is similar to that recorded in the cases above.

These data describe a multi-species fishery where fishermen expect a mixed catch of profitable species which are retained and a related amount of bycatch which is discarded. The amount discarded, therefore, depends on the fishing success for target species. Negative correlation of the discarded catch, by species, to the retained catch is typical of the discard practices conducted in multi-species fisheries.

3.4 Distribution of discard data

The complete range of discard species recorded in sub-sector catches is rarely present in individual discard records. It is also commonly the case that some discard species are seldom encountered during `normal' fishing operations. Although summed landing and summed discard data are normally distributed, the distributions of discard data by category and species are more widely distributed and more skewed. The pattern of the distribution and the related skewness are exaggerated by the number of zero-catch records which increase if discards are classed in smaller categories or by individual species. Such a distribution is characterised by large standard deviations and poor resolution of discard weight data making specific discard estimation difficult.

3.5 DPUE and the index of variation

Table 2 compares indices of variation (standard deviation/mean value) of discard ratios to summed landings and target-species landings and discard per unit effort (DPUE) derived from sampling in the four sub-sectors of the fishery. The statistical variation of the discard ratio to the summed landings is usually smaller than that of the discard ratio to the target landings. The index for the DPUE tends to be constantly small. This implies that the DPUE is the most reliable parameter to represent discard practices. The larger values of the indices most likely reflect variations in both landings and discards during fishing operations.

Figure 1. The relationship of the correlation coefficients derived from the discard and landing of individual species and those derived from the correlation of the discard of individual species and the summed landing of all species.

Table 2. Indices of variation of the discard ratio and the discard per unit effort (DPUE) derived from data recorded on 37 sampling occasions over 3 years from a Danish seine fishery in southern Japan.

		Discard ratio		DPUE
Sub-sector	No. of trials	Sum./Sum.	Sum./Tgt.	Sum.
Red seabream	9	0.814	1.039	0.549
Conger eel	8	0.535	0.671	0.454
Cuttlefish	14	0.303	0.503	0.502
Assorted fish	6	0.523	0.770	0.577

Sum./Sum. = The discard ratio of summed discards to summed landings.
Sum./Tgt. = The discard ratio of summed discards to target landings.

3.6 Estimation of discards

Summed discards in the cuttlefish sub-sector, which were estimated on the basis of DPUE and discard ratios are compared in Table 3. Summed discards estimated on DPUE according to Equation (11) were 32.1 t. The estimation of the discard ratio derived from the summed landing, r_i, of 1.59, was 41.4 t (equation (4)), which was more or less similar to that estimated by DPUE. Since cuttlefish comprises only 37.4% of the retained catch in this sub-sector, the discard to target-species retained ratio, r_(t) of 4.86, was much greater than the discard to summed landings ratio. On the other hand, cuttlefish landings by this sub-sector were only about half the total landings of cuttlefish in the whole fishery. The estimation of the discard, r _(j), based on the catch of the target species (equation (8)) was 125.9 t and much greater than that estimated by the other two methods due to the varied nature of fishing practices and landings of cuttlefish as non-target species in other sub-sectors.

This analysis indicates that the ratio of discard to target species in such multi-species and multi-gear fisheries grossly over-estimates discards relative to the estimates generated using DPUE and summed landings and verified by at sea observation.

Table 3. Estimates of the annual discard in the cuttlefish sub-sector of the Danish seine fishery in southern Japan, derived from data recorded on 37 sampling occasions over 3 years.

Sum./Sum. = The discard ratio of summed discards to summed retained catch.
Sum./Tgt. = The discard ratio of summed discards to target species retained.

4 DISCUSSION

4.1 The nature of discard practices in multi-species fisheries

Two important factors are demonstrated by the present case study. These are firstly, that many species do not commonly occur in both discards and landings and secondly, that there is a wide range and negative correlation between discards and landing.

It is a typical characteristic of multi-species fishing that a large number of species are routinely discarded and never landed. This reflects the fact that the catch is sorted onboard mainly by species. This characteristic is of particular importance when considering the most appropriate choice of discard ratio and estimation procedure to be applied to landing data, e.g. it is impossible to obtain discard ratios, r_ij of individual species discards to landings. The wide range of both positive and negative correlation coefficients observed between individual species discards and retained catch is of particular significance in multi-species fisheries. Positive correlation suggests that species are sorted mainly by size and negative correlation, with landings, suggests that the catch of discard species increases if fishing for preferable species is unsuccessful.

4.2 DPUE and the summed discard ratio

Although an apparently similar term to DPUE, based on bycatch, i.e. bycatch per unit effort (BPUE), was proposed by Hall (1996), this was, in effect, a bycatch (discard) ratio. The concept of DPUE is based on totally different criteria.

Although this analysis recommends the estimation of discards using discard per unit effort (DPUE), key fishing effort statistics are not always available to fulfil this recommendation. Fishing effort devoted in a sector-i can, however, be estimated from total landings and catch per unit effort (CPUE) data, E (c_i) as;

E_i = C_i / E (c_i) (12)

although the estimation of discards will be less reliable than those made with actual effort data.

It is important to understand that a discard ratio is not simply a ratio between discards and retained catch. The retained catch is used as a parameter to estimate fishing effort to induce discards (Matsuoka 1998). This is endorsed by the relationship;

D_i = q_i · E_i

= E (d_i) · C_i / E (c_i)

_ E (d_i___c_i) · C_i _

______= r_i_· C_i_ (13)

This relationship is not always true particularly when the variation of c_i is large and the distribution is skewed. This also indicates that the estimation of discard using a discard ratio based on the summed retained-catch is the best alternative to DPUE for the estimation of discards in the absence of effort data.

There is also a variety of other discard ratios based on a range of different factors, e.g. landings by numbers of individuals. The choice of the most appropriate ratio depends on which factor is most representative of the fishing effort, shows the highest level of correlation with observed discards and is available in landing statistics.

4.3 Application of the discard ratio based on the target catch

The over-estimation of discards calculated on the basis of target-species landings, presented by Alverson et al. (1994), can be explained by the way in which the method of estimation is applied. The over estimation was, in fact, due to the multiple counting of fishing effort (Equation (8)), in the production of discard estimates for the different sectors. The over-estimation was mainly due to the fact that no consideration had been given to the by-product landings of a species in the sectors which do not target that species (Matsuoka 1996). The definition of the discard ratios on the basis of the retained target species catch (Equation (7)) is possibly an alternative approach in this instance, although the reliability of this method is lower than that based on the sum of retained catch. The method can be utilised, however, once it is limited to a single and independent sector. The target-species concept may be applicable in mono-species fisheries or where a target species constitutes the majority of retained catches. This concept and subsequent estimation are not, however, applicable to multi-species fishery sectors.

Application of this concept is only appropriate where the bycatch of targeted species in other sectors are peripheral and landing statistics of species are available from all sectors. Discard estimation is theoretically impossible if species statistics alone are available where species are harvested with many types of fishing gears.

These caveats regarding the application of estimate methodologies further strengthen the view that discard estimation at a small scale, e.g. assessing the impact of discards on individual resources within a single sector and at a large scale, e.g. assessing the wastage by national, regional or global fishing industries over multiple sectors have different preconditions demanding different approaches.

4.4 The estimation of individual species discards

The estimation of individual species discards is not rejected, in theory, despite the results of the present analyses. The partial estimate of individual species discards in a sector or across sectors may be possible under such circumstances where the statistical distribution of the individual species discards is well known. The estimation of the total discarded in a sector using the sum of individual species discard estimates is, however, beyond reality in multi-species fisheries due to the complexity of the discard species assemblage.

The only realistic method to estimate individual species discard is that based on the DPUE. Even in this case, sample distribution of individual species discards must be carefully examined to determine whether the method is applicable as the properties of the distribution of discards is likely to differ from species to species.

The weakness of individual species discard ratios, r_ij, is primarily due to the uncommon occurrence of species between landings and discards and occasionally a negative correlation of discards to retained catches. It is also impossible to analyse the correlation between discards and retained catch in the case of protected or rare species, although this is one of the most important tasks required of discard estimation in many fisheries.

4.5 Recommendations for discard estimation

The estimation of discards must be based on the principle that parameters with small indices of variation produce highly accurate estimates. Discard estimates based on DPUE are, therefore, the most statistically reliable and accurate estimates available. In addition the onboard sampling to obtain DPUE, q_i or q_ij, required by this method, is much easier than the sampling required in the application and verification of discard ratios. The DPUE approach is also universal because it does not require detailed knowledge of the catch and discard composition as preconditions to its application. The method also has the distinct advantage that it is not effected by changes in resources. Long-term assessments of discards are possible, therefore, without being effected by changes in landings. If DPUE is to be applied successfully it must also reflect the biological characteristics of discard species where the fishing efficiency of gear is standardised. The advantages attached to this method are true for the majority of multi-species and multi-gear fisheries and can also be expanded to provide estimates of the discards in most mono-species and mono-gear fisheries.

A potential problem of the DPUE method is that the magnitude of the total discard in a fishery is difficult to visualise, as it is the sum of a number of DPUE estimates by sector, gear type and vessel sizes. This suggests that the illustration of discard practices and the estimation of discards are not the same tasks. Discard ratios may be a more easily understood means to describe discard practices and DPUE, a more reliable and accurate method for discard estimation when DPUE and fishing effort statistics are used in combination.

4.6 Future requirements of discard estimation

The accuracy of discard estimates is governed by the application of the most appropriate methodology and a sufficient volume of reliable data. Application of the most appropriate method infers the use of the best parameter for the purpose of estimation. This issue has not been considered in detail, however, because most discard estimates have assumed discard ratios as a precondition. Although methodology is currently being discussed, mainly from the coverage viewpoint, this is not a preferred approach. It is strongly recommended that discard practices must be further analysed and methodologies further developed for the assessment of discards.

It is also important to stress that careful consideration is required when dealing with a large and extensive variety of species in both retained and discarded catches in multi-species fisheries. This has special consequences in tropical and subtropical fisheries which are conducted with a variety of fishing gears for a variety of species. Fishermen and fisheries scientists in such regions currently face complex and difficult problems in their attempts to overcome the discard problems in these fisheries. It is ironic, however, that key data required for discard estimation are lacking in most tropical countries due to insufficient information collection systems in many coastal fisheries. Discard practices, as a result, have received little attention and few statistics are available for landings by sector or fishing effort. It is strongly recommended that future strategic and methodological development of discard monitoring give due consideration to multi-species fisheries which prevail in a large portion of world fisheries and also address similar issues in commercially important mono-species fisheries.

REFERENCES

ALVERSON, D.L., FREEBERG, M.H, MURAWSKI, S.A. & POPE, J.G. 1994. A global assessment of fisheries bycatch and discards. FAO Fisheries Technical Paper 339, FAO, Rome, 233pp.

ALVERSON, D.L. & HUGHES S. E. 1995. Bycatch: from emotion to effective natural resource management. In: Solving Bycatch: Consideration for Today and Tomorrow, p3-28.

FAO 1996. Technical consultation on reduction of wastage in fisheries. FAO Fisheries Report, 547, FAO, Rome, 27pp.

HALL, A.M. 1996. On bycatches, Rev. Fish Biol. Fish., 6: 319-352.

MATSUOKA, T. 1996. Discards in Japanese marine capture Fisheries and their estimation, in Technical Consultation on Reduction of Wastage in Fisheries. FAO Fisheries Technical Paper 547 suppl., FAO, Rome, 309-329.

MATSUOKA, T. 1998. Methodology to estimate bycatch and discards, In: Regional Workshop on Responsible Fishing. Southeast Asian Fisheries Development Center, Bangkok, p207-217.

Evaluating the costs and benefits of alternative monitoring programmes for fisheries management

André E. Punt

CSIRO Marine Research, GPO Box 1538, Hobart, TAS 7001, Australia.

Email: [email protected]

Abstract: An approach to the quantitative evaluation of the costs and benefits of different levels of observer coverage is outlined. This approach attempts to value observer programmes in terms of their ability to better satisfy management objectives rather the precision with which quantities that may be of interest to management (e.g. discard rates) are estimated. The approach is illustrated using the observer programme which estimates the discards of blue grenadier by the trawl fishery off eastern and southern Australia and the observer programme that which estimates by-catch of Hooker's sea lions in the fishery for arrow squid off southern New Zealand. The benefits of increased precision for the latter are shown to be small while the impact of different levels of observer coverage on management advice for blue grenadier is found to be substantial for some types of management approaches.

1 INTRODUCTION

In Australia, observers on fishing vessels are responsible inter alia for validating catches, monitoring catch quotas, advising fishing masters of the correct radio reporting procedures, and observing whether operations and activities comply with Australian government procedures (Nicol 1992). The first of these tasks includes species identification of the catch, measuring a random subsample of the catch, and collection of biological data for ageing, feeding, reproduction, growth, stock-structure and migration studies. Furthermore, it is possible to record recaptures of tagged southern bluefin tuna from vessels carrying observers separately from those captured by vessels not carrying observers, and consequently estimate the probability of tag-reporting for vessels without observers (T. Polacheck, CSIRO Marine Research, pers. commn). Data collected by observers can also form the basis for addressing by-catch reduction issues (e.g. Kennelly 1995). The data collected in Australia are typical of those collected in other observer programs (e.g. Fitzgerald et al. 1993, Lee et al. 1996).

A question that arises frequently when observer programmes are designed, is how to select an appropriate level of observer coverage. From the viewpoint of the scientists conducting assessments, the appropriate level is often stated as 100% while from that of the fishing industry (who increasingly have to bear the full costs associated with observer programmes), this level should be as low as possible. In general, the level of observer coverage is selected taking account of the trade-off between cost and the precision with which quantities of interest can be estimated (e.g. Smith et al. 1997). This approach can be criticised; however, because it does not really examine the benefits of collecting observer data, which should relate to an increased ability to satisfy agreed management objectives.

Many of the Fishery Assessment Groups (FAGs) established by the Australian Fisheries Management Authority (AFMA) have as their objectives to "conduct assessments" and "provide advice on the benefits of alternative harvest strategies". As will be shown later, one by-product of the latter is the ability to "value" additional research, including observer programmes.

This paper proceeds by first outlining an example of how uncertainties can be dealt with by applying a precautionary approach when making management decisions. It then overviews the process of evaluating alternative "management procedures"¹ and, in particular, how the results of such an exercise can be used as the basis for a quantitative evaluation of the costs and benefits associated with different levels of observer coverage. Finally, the results for two case studies are presented. These results should be considered to be illustrative only and do not constitute a formal evaluation of the observer programmes concerned.

1.1 Eastern stock of gemfish

Landed catches by the trawl fishery off eastern Australia were not validated until the introduction of the quota management system in 1986. Nevertheless, a programme was established in 1975 to estimate the landings by the demersal trawl fleet off eastern Australia by monitoring the catches landed through co-operatives (Rowling 1990). As part of this programme, catches of gemfish (Rexea solandri) which was one of the major components of this trawl fishery during the 1970s and 1980s, were monitored. However, it is now recognised that this monitoring programme under-estimated landings because an unknown fraction of the catch was not landed at the co-operatives, and because at-sea discarding occurred.

One implication of this is that assessments of this stock are hampered by uncertainty regarding the historical catches. In order to overcome this problem, assessments have been based on two alternative series of historical catches. One of these reflects recorded catches only, while the other makes some allowance for under-reporting of catches and discarding (Punt et al. 1997a). There is no agreement in the assessment group regarding which of these two series of catches is closer to reality, although it is agreed that they span the plausible range adequately (Smith and Punt (in press)).

The results from stock assessments based on the two series of catches differ quite markedly. For example, depending on which catch series is assumed when conducting the assessment, there is either a 90% probability that the depletion of the resource in 1996 lies in the interval [0.19, 0.36] or in the interval [0.24, 0.48]. More importantly from the management point of view, the probability of the resource recovering to the biomass at which Maximum Sustainable Yield, MSY, is achieved under a total allowable catch, TAC, of 1000t differ markedly (0.15 and 0.40 respectively). TAC decisions for this resource have been based on the more pessimistic of the two assessments.

1.2 Quantitative basis for evaluating the benefits of an observer programme

Conceptually, the "value" of an observer programme can be assessed in terms of how the data it provides allow a management procedure to improve the ability to satisfy the management objectives (McDonald and Smith 1997, McDonald et al. 1997). Evaluation of the performance of candidate management procedures involves the following steps (Punt 1992).

1. The development of a model (the "operating model") that reflects inter alia the underlying biological system being managed, how data are collected through logbooks, and survey and observer programmes, and how management decisions impact the resource;
2. The identification of a set of performance measures to quantify the management objectives (probability of dropping below some threshold biomass, average annual yield, etc.);
3. The selection of a set of alternative candidate management procedures, and
4. The use of a simulation approach to evaluate how well different combinations of management procedure, including its associated data collection scheme, perform in terms of the management objectives.

The evaluation of different levels of observer coverage can be examined by assessing the extent to which the performance measures change when the quality of the data obtained from the observer programme is changed. The results from studies into the relationship between the precision with which data can be collected and the extent of observer coverage (e.g. Smith et al. 1997) are used to define the scenarios regarding observer coverage to investigate. This approach is in contrast to the "traditional" approach of plotting measures of the change in the coefficient of variation of the estimate of some quantity determined from the data collected from the observer programme (e.g. discard rates) against the extent of observer coverage.

Two important features of using simulation to evaluate the performance of an observer programme are: (a) the evaluation is linked closely to how the data will be used for management purposes, and (b) the effect of other uncertainties will be to impact the extent to which increased observer coverage will improve management. It would be expected, for example, that the ability to satisfy the management objectives is related to the accuracy and precision of the least reliable source of information. The utility of increasing observer coverage is related to the extent to which it is to provide data on the source of uncertainty that is limiting performance.

2 CASE STUDY 1: THE BLUE GRENADIER FISHERY OFF SOUTHERN AUSTRALIA

An Integrated Scientific Monitoring Programme (ISMP) was introduced recently into Australia's south-east fishery. This programme aims, thorough at-sea and port sampling, to provide `statistically robust' estimates of total catch (retained and discarded) of quota and non-quota species, and of the size/age composition of the total catch of selected species (Smith et al. 1997). The ISMP was designed to allocate sampling effort between 14 target strata (based on target species, region of fishing and home-port) to achieve desired levels of precision for these quantities (measured by coefficients of variation). These levels of precision were selected `based on predetermined stock assessment priorities and the budgetary implications of the different levels' (D. Alden, Australian Fisheries Management Authority, pers. commn). However, the selected levels of precision were not linked directly to achievement of the management objectives.

It is known that inappropriate allowance for the impact of discards can bias the conclusions drawn from stock assessment models (e.g. Pikitch 1991, Alverson et al. 1994). The impact of the ISMP on stock assessment in the south-east fishery has to date been most marked for the blue grenadier resource. Blue grenadier (Macruronus novaezelandiae) is currently the second most valuable species in the south-east fishery (Tilzey 1998) and has been subject to considerable dumping of small fish in recent years (Smith 1998). The assessment for this species is based on a Bayesian variant of the `stock synthesis' (Methot 1989, 1990) approach (Punt et al. in prep). The assessment uses data on absolute abundance from the egg production method, catches (both landed and discarded), catch-at-age (both landed and discarded), and standardized catch-rate indices (Smith 1998). The catch and catch-rate data are disaggregated by fleet. Figure 1 shows posterior distributions for the time-trajectory of recruitment (1-year-class strength) for variants of the assessment in which the data from the ISMP on discards are included in the assessment and in which these data are ignored (and discarding is consequently assumed to be inconsequential).

Figure 1 shows clearly that the estimates of recruitment for recent years depend strongly on the inclusion or otherwise of the data on discards from the ISMP. In particular, an assessment that excludes this information fails to identify that the 1995 year-class is as strong as estimated by an assessment that includes the discard data. The commercial catch-at-age data provide an indication that the 1996 year-class is strong. However, including the discard data in the assessment leads, as expected, to more precise estimates of the strengths of recent year-classes, as reflected by the widths for the 90% credibility intervals (Figure 1). The consequences of different levels of future catch are also sensitive to whether or not the discard data are included in the assessment. For example, Figure 2 shows time-trajectories for the probability that the spawner biomass remains above 40% of the virgin level, B₀, and the expected spawner biomass if the landed catch is 10,000 t from 1998 onwards. The analysis that incorporates historical discarding also allows for discarding in the future. The trajectories of expected spawner biomass are the same until 1999 when the 1995 cohort enters the spawner biomass. Somewhat surprisingly, given that it corresponds to weaker recent recruitments, the assessment based on excluding the ISMP data is more optimistic (higher probabilities of remaining above 40% B₀ and high expected spawner biomasses). This is because future discarding impacts the 1995 and 1996 year-classes when discarding is included in the assessment.

The question of the appropriate level of precision for the estimates of discard catch (and its age-structure) for this species can be addressed in a variety of ways. For the purposes of this study, the value of different levels of observer coverage can be associated with the accuracy and precision with which the estimates of recruitment are determined². This, in turn, can be assessed using a Monte Carlo simulation approach. This approach involves specifying a "true" time-trajectory of recruitment (obtained by applying the stock assessment approach to the existing data for blue grenadier - Figure 1 upper panel), generating 500 pseudo data sets (which include pseudo discard data) from this trajectory, applying the stock assessment approach to each data set to obtain 500 estimates of the recruitment time-trajectory, and finally comparing the "true" with the estimated series. The artificial data sets for absolute abundance, landed catch and catch-at-age, and catch-rate are generated assuming the levels of precision determined by fitting the assessment model to the actual data, while a range of levels of precision are considered for the discarded catch and catch-at-age data. The latter is a proxy for different levels of observer coverage.

Figure 3 shows the distribution of the relative error (estimated value less the true value) against the change to the variance of the information on discards. Results are shown for the 1994, 1995 and 1996 year-classes as well as for the 1997 spawner biomass. Following Punt et al. (in prep), for the base-case analysis the size of discarded catch each year is assumed to be log-normally distributed with a coefficient of variation, CV, of 0.3 while the proportion of the discarded catch that is of age a during year y, , is assumed to be log normally distributed with a CV of . The means for these log-normal distributions are taken to be the true values. The rationale for the choice of a log-normal distribution for the age-composition data is given by Punt (1997) while the specific values 0.3 and 0.47 are based on fits to the actual data for blue grenadier.

The accuracy and precision of the estimates of the 1994 year-class and of the 1997 spawner biomass (Figures 3a and 3d) are not impacted particularly noticeably by changes to the extent of observer coverage (as reflected by the CVs for the estimates of the discards and their age-structure). This is perhaps not surprising because the 1994 year-class has already entered the commercial fishery and is consequently determined primarily by the data from the landed catch (and its age-structure). Similarly, the spawner biomass consists of animals aged 5 and older so its estimation is also determined primarily by the data from the landed catch.

The medians of the distributions of relative error for the 1995 and 1996 year-classes decline and the widths of the 90% intervals increase as the precision with which the discard data is determined is reduced. The bias in the estimates of year-class strength is a function of the precision of the discard data because the estimation procedure underlying the assessment "shrinks" recruitment towards that expected from a deterministic (Beverton-Holt) stock-recruitment relationship (Punt, et al. in prep). Including "shrinkage" is common when applying methods of stock assessment that use catch (or discard) age-composition data (e.g. Butterworth et al. 1990)³. The extent of shrinkage depends on the amount of information about a year-class. If little is known, the amount of shrinkage can be substantial. The bias becomes more negative as the precision of the discard data is reduced because the 1995 and 1996 year-classes are estimated to be substantially stronger than expected (Fig. 1).

Figure 1. Posterior distributions for the time-trajectories of recruitment, expressed as a fraction of that expected from the deterministic stock-recruitment relationship (1980-97) for blue grenadier off eastern Australia. The solid lines are distribution medians and the dotted lines posterior 90% credibility intervals, while the dashed lines correspond to expected recruitment. Results are shown for assessments that include and exclude the data from the ISMP.

Figure 2. Time-trajectories (1995-2017) of the probability that the spawner biomass of blue grenadier remains above 40% of the virgin level, and the expected spawner biomass, for an annual landed catch of 10,000t. Results are shown for assessments that include and exclude the data from the ISMP.

Figure 3. Medians and 90% intervals of relative error of four management-related quantities plotted against different levels of variance of the discard information. The four quantities are: (a) the 1994 year-class strength, (b) the 1995 year-class strength, (c) the 1996 year-class strength, and (d) the 1997 spawner biomass.

Figure 4 explores whether, by improving the precision of either the size of discarded catch or its age-structure, major improvements to the estimates of the 1995 year-class strength can be achieved. This allows the question whether it is preferable to transfer effort from estimating discarded catch to estimating the age-structure of the discarded catch or vice versa to be addressed. The results in Figure 4 suggest that the relative errors are slightly more sensitive to the precision of the estimates of the discarded catch than to its age-structure.

Figure 4. Medians and 90% intervals for the relative error for the estimates of the 1995 year-class strength against different levels of variance for the discard information. Results are shown for analyses that consider different levels of precision of: (a) both the size of discarded catch and its age-structure, (b) only the size of the discard catch, and (c) only the age-structure of the discarded catch.

It is noteworthy that there is little improvement in accuracy and only a small improvement in precision as the variance of the discard data is reduced from the current level⁴ (Figs. 3 and 4). This occurs because the accuracy and precision of the estimates of year-class strength also depend on the precision with which the landed catch, its age-structure and the relative abundance index are determined. The precision of these latter data is such that improving the precision of the information on discards does not improve overall performance markedly, implying that improved performance requires increasing the precision of some of the other data used in the assessment.

3 CASE STUDY 2: THE HOOKER'S SEA LION-ARROW SQUID FISHERY INTERACTION

Hooker's sea lion, Phocarctos hookeri, has been listed as vulnerable by the IUCN (IUCN, 1996), and as threatened by the New Zealand government. The government of New Zealand, through its Department of Conservation, has established a programme to enable this species to recover under the terms of the 1978 Marine Mammals Protection Act and the 1996 Fisheries Act. The main cause of human-induced mortality of Hooker's sea lions is the fishery for arrow squid, Notodarus Sloanii, in New Zealand subantarctic waters near the Auckland Islands. The estimated average annual catch of sea lions by this fishery over the period 1988-96 was 72.5 while its 1996 abundance is estimated to be 13,558 (SE 905) (Gales 1995). The sea lion population was estimated to lie between 14 and 76% of its pre-exploitation level at the start of 1997 with 90% probability (Maunder et al. in press). A management decision rule has been established to manage the interaction between the squid fishery and the sea lion population. This rule involves closing the squid fishery if the estimated by-catch of Hooker's sea lions exceeds the Maximum Allowable Fishing-Related Mortality (MALFRM). The MALFRM is determined annually using a variant of the Potential Biological Removals approach developed by Wade (1998). Maunder et al. (in press) examine the trade-off between the recovery of the sea lion population and the frequency with which the squid fishery has to be closed, associated with different formulae for determining the MALFRM.

The estimate of the sea-lion by-catch used to decide whether or not to close the squid fishery is based on in-season monitoring on a weekly basis (P.J. Starr, New Zealand Seafood Industry Council, Wellington, New Zealand, pers. commn). A target level of 20% observer coverage is currently used to monitor this fishery. The protocol developed by Maunder et al. (in press) can be used to evaluate the implications of different levels of observer coverage.

The approach used by Maunder et al. (in press) will not be discussed in detail here. Briefly, it involves assessing the current status (1996) of the sea lion population assuming that its dynamics follow a Pella-Tomlinson (Pella and Tomlinson 1969) model and using a Bayesian estimation framework. The protocol then involves projecting the population forward from 1997 to 2016. For the projection period, the fishing mortality rate in the absence of a MALFRM (i.e. if the fishery for squid were to be restricted through its TAC only) is assumed to be distributed about the historical (1988-96) rate, and survey estimates of the abundance of sea lions are assumed to be obtained every fourth year starting in 1996.

The protocol developed by Maunder et al. (in press) is modified in two ways for the calculations of this paper. First, it is assumed that the CVs supplied with the estimates of abundance used to calculate the MALFRM are negatively biased. This assumption is made because these CVs are based solely on sampling variation. Recent analyses (e.g. for gray whales (Wade 1996) and minke whales (Punt et al. 1997b)) suggest that sampling variances for abundance surveys for marine mammals can be much smaller than the true extent of variance. For the purposes of this study therefore, the overall CV for the abundance estimates is taken to be 0.2 although that used when calculating the MALFRM is only 0.07, thus reflecting the assumption that most of the observation-error variance is undetected. The choices of 0.2 and 0.07 are based on the CV assumed for abundance estimates by Maunder et al. (in press), and the observed CVs for the 1994 and 1995 surveys by Gales (1995).

The second modification is to make allowance for errors when estimating the by-catch of sea lions using the observer programme. The by-catch during year y in the absence of a fishery closure, , is given by the product of an annual exploitation rate, , and the number of sea lions, . An estimate of this by-catch, , is then generated using the equation where . The parameter z reflects the "cost of uncertainty" (see below). If is less than the MALFRM for year y, , the actual removals from the sea lion population, , is set equal to (i.e. the fishery is not closed). In contrast, if then is set equal to which reflects the impact of closing the fishery.

Following Maunder et al. (in press), the management objective for the sea-lion population is assumed to be related to the probability of recovering to above 90% of the pre-exploitation level by 2016. Different levels of observer coverage can be evaluated in terms of the change in the proportional loss of squid catch (defined as ; i.e. the loss in squid catch is assumed to be related linearly to the reduction in potential sea lion mortality), given a fixed probability of recovering to above 90% of the pre-exploitation level by 2016. The fixed probability, 0.73, was determined by assuming perfect information about the by-catch, i.e. , and applying the MALFRM rule simulated by Maunder et al. (in press). When the CV of the by-catch, , is increased from zero, the value for the parameter z is selected to achieve the fixed probability of recovery.

Figure 5 shows the median and 90% intervals for the loss in squid catch (averaged over the 20-year projection period) as a function of . There is relatively little change in the loss in squid catch with . A similar result is obtained even if the productivity of the resource is reduced and the estimates of abundance used when calculating the MALFRM are assumed to be negatively biased. The reason for this is that the estimates of by-catch are assumed to be unbiased (apart from the factor, z, which allows for uncertainty) and the productivity of the population is relatively low. In such a situation, it is the total by-catch over a time period, not the distribution of that by-catch over time, that is important. The assumption of unbiasedness implies, in this case, that "what you lose on the swings, you gain on the roundabouts". Whether the effect of different levels of observer coverage is as small as estimated here depends on whether the loss in squid catch is in fact a linear function of when the fishery is closed, as is assumed here and by Maunder et al. (in press). For example, the current formalism implies that closing the squid fishery at the start of one year followed by not closing the fishery at all the next year is equivalent in terms of its effect on the profitability of the squid fishery to closing it in both years half way through the year.

Figure 5. Medians and 90% intervals for the proportional loss in squid catch as a function of the CV of the by-catch of sea lions in the arrow squid fishery.

4 DISCUSSION

Embedding the evaluation of observer programmes within an evaluation of alternative candidate management procedures implies that the impact of changes to the observer programme on the ability to satisfy the management objectives can be assessed. For example, although Figures 3 and 4 indicate that the estimates of recent recruitment for blue grenadier are sensitive to the precision with which discard data can be obtained, if management is based on the estimates of spawner biomass, the ability to satisfy the management objectives will not be particularly sensitive to the level of observer coverage. In contrast, if management is based on tracking recent recruitment (which will be the case if plots like Figure 2 are used to specify TACs), the ability to satisfy the management objectives will be highly dependent on the level of observer coverage.

The improvement in performance associated with increased observer coverage is not "linear" (Figures 3 -5). This is either because another uncertainty (not controlled by the observer programme) is more important or because the structure of the management procedure itself is the primary determinant of performance. The sea lion case study is an example of the latter. The ability to achieve a 0.73 probability of recovering to 90% of the unexploited level by 2016 depends on constraining the by-catch to some average level, the exact time-sequence of by-catch is of much less importance.

Although the evaluation approach provides a means for linking the extent of observer coverage (and hence the cost of the observer programme) to the management objectives, the results are only meaningful if the management objectives have been quantified. Unfortunately, this is seldom the case in practice (Pearse and Walters 1992, Francis and Shotton 1997). For example, the lack of clearly defined management objectives for blue grenadier off southern Australia precluded a full evaluation of the ISMP for this species. Similarly, the benefits determined using the evaluation approach will only be realised if the management procedure considered is actually implemented. In fact, it would be hoped that adoption of this approach for valuing observer programmes (and other forms of monitoring) would encourage efforts to quantify the (usually vague) management objectives and agree management procedures. Finally, the approach requires that cognisance be taken of the `precautionary approach' to fishery management, so that yields are lower in the absence of information. This implicitly provides an incentive for the collection of data.

The approach can, of course, only evaluate quantifiable benefits; the value of the unquantifiable benefits of an observer programme such as improved liaison between fishers, scientists and managers cannot be determined by the approach. However, it is not clear that any approach exists for evaluating such benefits. Similarly, observer programmes (such as the ISMP) may relate to several species, and the multi-species nature of the fishery and the accompanying management objectives have to be taken into account in the evaluation of the benefits of such observer programmes. A particular problem in this regard for some programmes (such as the ISMP) is that one of their goals could be to document information (in this case on bycatch) for as yet unspecified species. In such a situation, there are no management objectives to satisfy, and the approach described above could not be applied.

The analyses of this paper are based on the assumption that the behaviour of a fisher is independent of whether or not the vessel carries an observer, and this is often not the case (Cramer et al. 1994, Liggins et al. 1997). In principle, it is possible to include this "observer effect" when evaluating an observer programme by incorporating it in the operating model. One advantage of doing so is that it then becomes possible to quantify the impact of having observers for enforcement rather than monitoring purposes. However, a problem that arises when including an "observer effect" is that there is rarely a clear basis for specifying its magnitude.

Although the approach of using the process of comparing alternative candidate management procedures to evaluate the costs and benefits of observer programmes is not ideal because it cannot cover all considerations, it nevertheless makes explicit the link between the collection of data and the ability to satisfy the management objectives. This basic approach for cost-benefit evaluation has been adopted by AFMA. A formal objective for Federally managed fisheries in Australia is to "implement efficient and cost effective management" (Commonwealth of Australia, 1998), and the AFMA Research and Environment Committee recently agreed (for fisheries for which this was feasible) to base research prioritisation on the outputs from evaluation of management procedures.

ACKNOWLEDGEMENTS

The Blue Grenadier Assessment Group (BGAG) and the Central Ageing Facility (CAF) are thanked for supplying the data used for the blue grenadier case study. Dave Alden, Doug Butterworth, Jean Chesson, Dave Smith, Tony Smith and Paul Starr are all thanked for their comments on an earlier version of the manuscript.

REFERENCES

ALVERSON, D.L., FREEBERG, M.H., MURAWSKI, S.A. & POPE, J.G. 1994. A global assessment of fisheries by catch and discards. FAO Fishery Technical Paper No 339, Rome.

BUTTERWORTH, D.S., HUGHES, G.S. & STRUMPFER, F. 1990. VPA with ad hoc tuning: implementation for disaggregated fleet data, variance estimation, and application to the horse mackerel stock in ICSEAF Divisions 1.3+1.4+1.5. S. Afr. J. mar. Sci., 9, 327-57.

BUTTERWORTH, D.S., COCHRANE, K.L. & DE OLIVERIA, J.A.A. 1997. Management procedures: A better way to manage fisheries? The South African experience. In PIKITCH, E.K., HUPPERT, D.D. & SISSENWINE, M.P. eds. Global Trends: Fisheries Management. p. 83-90. American Fisheries Society Symposium 20, Bethesda, Maryland, USA.

COCHRANE, K.L., BUTTERWORTH, D.S., DE OLIVERIA, J.A.A. & ROEL, B.A. 1998. Management procedures in a fishery based on highly variable stocks and with conflicting objectives: experiences in the South African pelagic fishery. Rev. Fish Biol. Fish., 8, 177-214.

COMMONWEALTH OF AUSTRALIA. 1998. Fishery Management Act 1991. Reprinted as in force on 31 January 1998 (including amendments up to and including Act No. 120 of 1997). Reprint No 2. Australian Government Publishing Service, Canberra, ACT, Australia.

CRAMER, J., BERTOLINO, A.R. & SCOTT, G.P. 1994. Estimates of the catch of undersized swordfish by the U.S. large pelagic fleet based on logbook reports and scientific observations. Collective Volume of Scientific Papers ICCAT, 42: 294-302.

FITZGERALD, S.M., MCELDERRY, H., HATANAKA, H., WATANABE, Y., PARK J. S., GONG, Y. & YEH S-Y. 1993. 1990-1991 North Pacific high seas driftnet scientific observer programs. Bull. N. Pacific Fish. Comm., 53: 77-90.

FRANCIS, R.I.C. and SHOTTON, R. 1997. "Risk" in fisheries management: a review. Can. J. Fish. Aquatic Sci., 54: 1699-715.

GALES, N.J. 1995. Hooker's sea lion recovery plan. Threatened Species Recovery Plan Series No 17. Department of Conservation, Wellington, New Zealand.

IUCN 1996. 1996 IUCN Red List of Threatened Species. IUCN, Gland, Switzerland.

KENNELLY, S.J. 1995. The issue of by-catch in Australia's demersal trawl fisheries. Rev. Fish Biol. Fish., 5: 213-34.

LEE, D.W., BROWN, C.J. & JORDAN, T.L. 1996. Overview of the SEFSC pelagic observer program in the northwest Atlantic from 1992-1994. Collective Volume of Scientific Papers ICCAT, 45: 355-65.

LIGGINS, G.W., BRADLEY, M.J. & KENNELLY, S.J. 1997. Detection of bias in observer-based estimates of retained and discarded catches from a multi species trawl fishery. Fish. Res., 32: 133-47.

MAUNDER, M.N., STARR, P.J. & HILBORN, R. (In press) A Bayesian analysis to estimate loss in squid catch due to the implementation of a sea lion population management plan. Mar. Mamm. Sci. 00, 00-00.

MCDONALD, A.D. & SMITH, A.D.M. 1997. A tutorial on evaluating expected returns from research for fishery management. Natural Resource Modelling, 10: 185-216.

MCDONALD, A.D., SMITH, A.D.M., PUNT, A.E., TUCK, G.N. & DAVIDSON, A.J. 1997. Empirical evaluation of expected returns from research on stock structure for determination of total allowable catch. Natural Resource Modelling, 10: 3-29.

METHOT, R. 1989. Synthetic estimates of historical abundance and mortality for northern anchovy. Am. Fish. Soc. Symp., 6: 66-82.

METHOT, R. 1990. Synthesis model: an adaptable framework for analysis of diverse stock assessment data. Bull. N. Pacific Fish. Comm., 50: 259-77.

NICOL, R. 1992. Observing the observers. Australian Fisheries, 51: 20-2.

PEARSE, P.H. & WALTERS, C.J. 1992. Harvesting regulations under quota management systems for ocean fisheries. Marine Policy, 16: 167-82.

PELLA, J.J. & TOMLINSON, P.K. 1969. A generalized stock production model. Bull. I-ATTC, 13: 421-96.

PIKITCH, E.K. 1991. Use of a mixed species yield-per-recruit model to explore the consequences of various management policies for the Oregon flatfish fishery. Can. J. Fish. Aquatic Sci., 44: 349-59.

PUNT, A.E. 1992. Selecting management methodologies for marine resources, with an illustration for southern African hake. S. Afr. J. mar. Sci., 12: 943-58.

PUNT, A.E. 1997. The performance of VPA-based management. Fish. Res., 29: 217-43.

PUNT, A.E., ROWLING, K. & PRINCE, J. 1997a. Summary of the data for use in the assessments of the eastern stock of gemfish. Australian Fisheries Management Authority, Canberra, ACT, Australia.

PUNT, A.E., COOKE, J.G., BORCHERS, D.L. & STRINDBERG, S. 1997b. Estimating the extent of additional variance for Southern Hemisphere minke whales from the results of the IWC/IDCR cruises. Reports of the International Whaling Commission, 47: 431-34.

PUNT, A.E., SMITH, D.C. & HADDON, M. (In. prep) Stock assessment and risk analysis for blue grenadier, Macruronus novaezelandiae, off southern Australia.

ROWLING, K.R. 1990. Changes in the stock composition and abundance of spawning gemfish Rexea solandri (Cuvier), Gempylidae, in south-eastern Australian waters. Aust.J.Marine freshwater Res., 41: 145-63.

SMITH, A.D.M. & PUNT, A.E. 1998. Stock assessment of gemfish (Rexea solandri) in eastern Australia using maximum likelihood and Bayesian methods. In QUINN II, T.J., FUNK, F., HEIFETZ, J., IANELLI, J.N., POWERS, J.E., SCHWEIGERT, J.F., SULLIVAN, P.J. & ZHANG, C-I, eds. Fisheries Stock Assessment Models. Alaska Sea Grant College Program, AK-SG-98-01.

SMITH, D.C. (Ed.) 1998. Blue grenadier 1998, Report prepared by the Blue Grenadier Assessment Group for the South East Fishery Assessment Group. Australian Fisheries Management Authority, Canberra, ACT, Australia.

SMITH, D.C., GILBERT, D.J., GASON, A. & KNUCKEY, I.A. 1997. Design of an integrated scientific monitoring program for the South East Fishery. Australian Fisheries Management Authority, Canberra, ACT, Australia.

TILZEY, R.D.J. (Ed.) 1998. The South East Fishery 1997, Fishery Assessment Report compiled by the South East Fishery Assessment Group. Australian Fisheries Management Authority, Canberra, ACT, Australia.

WADE, P.R. 1996. Gray whale stock assessment. International Whaling Commission, Cambridge, UK.

WADE, P.R. 1998. Calculating limits to the allowable human-caused mortality of cetaceans and pinnipeds. Mar. Mam.. Sci., 14: 1-37.

1 A management procedure can be defined as a set of rules which utilise pre-specified data to provide recommendations for management actions (Butterworth et al. 1997, Cochrane et al. 1998).

2 As will be discussed later, accuracy and precision of the estimates of recruitment has been taken as the "management objective" in the absence of agreed and clearly defined management objectives for blue grenadier.

3 `Shrinkage' in the context of blue grenadier involves including in the assessment a prior for year-class strength based on an (estimated) deterministic stock-recruitment relationship. If the assessment data provide little information on a particular year-class, it is "shrunk" to the value predicted from the stock-recruitment relationship.

4 The current level of sampling intensity for blue grenadier was selected before the recent strong year-classes entered the fishery and large-scale dumping started to occur.