Fish Eating Birds and Salmonids in Scotland

7. Models, experiments, and the way forward

SUMMARY

The results of the present studies have enabled calculations of the consumption of commercially important fish by birds. However, with improved information on birds, modelling the impact of their predation on fish populations has become more complex.

Most estimates of the impact of bird predation on fish catches have relied upon either experiments or modelling, but the results of the present studies suggest that when used in isolation, neither of these approaches is likely to prove entirely satisfactory. Large scale catchment experiments would be expensive and of limited application. However the alternative, of modelling predation, is theoretical and too dependent upon untested assumptions to provide a reliable scientific tool for fisheries management.

We suggest that the most efficient way forward involves an iterative 'model-field experiment-remodel' approach. Priorities for future research are given, with the view to constructing models that can be tested by experiment and refined to produce useful fisheries management tools.

INTRODUCTION

The information collected in this research programme improves our understanding of the biology of fish-eating birds and certain aspects of the population dynamics of salmonid fishes. We now have a better knowledge of the distributions of sawbill ducks throughout the seasons, an insight into fluctuations in their populations between years, and the degree of seasonal mobility of birds in relation to their potential to recolonise areas in which others have been shot. We also have a much more comprehensive picture of the diet of fish-eating birds in Scotland than was previously available. Under some circumstances goosanders, mergansers and cormorants can consume large numbers of brown trout or Atlantic salmon. The over-riding trend is that salmonid fishes constitute a greater proportion of the diet of birds at more northerly latitudes, probably reflecting the relative abundance of salmonids within communities of fish. Larger species of bird took larger sizes of fish but cormorant, goosander and merganser generally took similar sizes of salmon (predominantly large parr and smaller than average smolts) mainly during the early part of the smolt run. In terms of the numbers of salmon per bird per day, most were taken by red-breasted mergansers and goosanders from the mainstems of northern rivers and relatively few by cormorants. Some of the trout taken by cormorants from rivers and lochs were of a size that could be kept by anglers.

In our calculations of the numbers of salmon consumed by sawbills, the highest figures were for goosander ducklings consuming salmon parr of age one year and older on the River Dee in July and August. Experimental removals of salmon parr in a variety of streams, and the main stem of the River Dee, resulted in recolonisation indices ranging from 7 to 54% (mean 21.1%). This suggests that there may be some scope for local compensation by immigration during the late summer months. A review of recent studies found little evidence to suggest that populations of parr can compensate for losses by increased growth or survival of the remaining fish but the possibility that compensation occurs at some sites cannot be ruled out. The scope for releasing fish from predation by birds is also limited by existing methods of bird control. Even intensive efforts to kill birds over 20 to 170 km of river only reduced bird use by about a half.

IMPACTS OF PREDATION ON FISHERIES; MODELLING OR EXPERIMENT?

There is a need to establish (i) the circumstances where fish-eating birds might have an impact on fish catches, and (ii) that such impact occurs. Most estimates of impact have relied upon either experiments or modelling, but it is clear that when used in isolation, neither of these approaches is entirely satisfactory. The present work was initiated to provide information for a quantitative model of the impact of bird predation on fish catches (e.g. Shearer et al. 1987). We have here calculated the numbers of fish removed by birds, but it still remains difficult to model impact on fish populations or catches.

The main problem is that birds take most fish from the wider, deeper parts of the river. We tried to resolve this by calculating fish consumption with respect to stream width to enable comparisons between fish removed per hectare and fish densities in streams of specific widths. Nevertheless, the problem persists because estimates of fish density on wider streams are mainly available for shallow riffle rather than for pools and deeper channels. Similarly, there is virtually no direct information on the population dynamics of juvenile salmon living on the mainstem and all good long term data come from studies in narrow streams. For modelling purposes we must assume that the same dynamics prevail downstream but because fish population dynamics are complex and involve fish community aspects that remain unstudied, this assumption is untested. Considering the problems associated with modelling, it is tempting to argue that a bird removal experiment would be more informative.

An experiment to demonstrate an effect of birds on fish catches would involve removing birds and then measuring the outcome compared with places where birds were not removed. This would have the advantage over modelling of establishing 'cause and effect' if fish catches were significantly increased at the experimental site. This is because, with all other factors held constant, the result could be directly attributable to bird presence. Such an experiment would be the only direct way to demonstrate the impact of birds but it would also have the disadvantages usually associated with large-scale field manipulations. For instance, it would be difficult to find adequate control sites, i.e. places where "all other factors" can be considered to be held constant, because river systems (and their community ecology) vary so much. The conventional way around such a problem is to replicate the experiment, and to reverse the treatment (bird removal) between experimental and control sites. As each treatment would have to be carried out over many years to accommodate the life cycle of salmon, the experiment would be very expensive. Moreover, long-term variations in natural fish mortality may mask any effect of bird predation. Finally, the result would only apply to the circumstances of the experiment and it would be difficult to extrapolate to other river systems.

Compared with the experimental approach, modelling the impact of birds on fish catches may seem more practical and cheaper. Once devised, the model could be applied to a variety of circumstances by varying the input parameters. Within the context of ecological investigation, such mathematical models can be powerful tools, allowing quantitative predictions based on specific assumptions. However, this process is not an alternative to experimentation because models are theoretical and their validity is best judged by testing their predictions with field observation or (preferably) with field experiments. The result can then be used to refine the original model (to give a closer approximation to 'reality') which can then be tested again.

This reiterative procedure (model-prediction-experiment-result-refined model) leads to increasing levels of understanding, so such models are very useful for research irrespective of their initial accuracy. In contrast a model can only be useful as a fisheries management tool if its predictions are sufficiently accurate; the use of an untested model for management purposes is inherently risky because its implementation could result in wasted effort or an unexpected outcome.

For example, our simple model of the potential impact of predation by goosander ducklings on salmon parr on the River Dee in summer (Chapter 4) is useful. The calculations suggest large numbers of parr are consumed and, with limited scope for compensation in the short term (Chapter 5), their population density may be lowered by predation in areas used by ducklings. The model is crude but it has the advantage of producing quantitative predictions that are testable with field experiment. The potential impact (a reduction in fish density of up to 1.9% per day) is of a scale that would be detected by measuring parr densities after the exclusion of sawbill ducklings from areas of the River Dee mainstem. If the predictions were fulfilled the model could then be useful in management decisions.

In contrast, it is not yet possible to test a model of the impact of goosander predation on large parr on the lower sections of the Dee in winter. This is because it has so far not been possible to measure (i) the populations of these fish in situ, (ii) smolt output in the following spring, or (iii) the numbers of returning adult fish, so a bird removal experiment seems futile. It may be argued that many of the fish consumed might otherwise have returned to support the fishery, but for the present, this assertion is based on a number of untested assumptions and remains speculative.

The most difficult situations for both research and management occur where complex models are necessary (to aid our understanding of bird-fish interactions) but cannot be easily tested. For example, on rivers in southern Scotland, goosanders and cormorants consume fewer salmon per bird than in more northerly rivers; they feed mainly on other fish species. Any comprehensive model of bird impact on salmon catches on these rivers necessarily involves knowledge of the whole fish community, including descriptive data (e.g. fish breeding parameters) and quantitative interactions (e.g. compensatory mechanisms) within the fish community. Unless fish communities in Scottish rivers are "non-interactive" (Oberdorff et al. 1998), such a model would be difficult to construct. Moreover, even if a complex community interaction model was devised and the impact of bird predation on salmon predicted, it would be difficult to test. As juvenile salmon are only a small proportion of bird diet on these rivers and the birds are highly mobile in spring, an experiment could not produce a statistically detectable effect on fish catches without the removal of very large numbers of birds over many years.

LIMITATIONS ON THE QUALITY OF FISH DATA

The quality of model predictions depends on the quality of input information and the validity of the assumptions made. At present good data on the population dynamics of salmonid fisheries are available from only a few studies. The most comprehensive work has been that of Elliott (1994) who has studied a population of brown trout over several decades and modelled the life cycle from numbers of eggs to numbers of spawning adult fish. Other population studies have generally been less extensive and have monitored limited components of the fish life cycle (e.g. Egglishaw & Shackley 1977, Buck & Hay 1984). Because of the paucity of information, it is necessary to generalise from those data that are available without any indication of the errors that might result from this process. The most detailed studies of fish populations have been on small streams (Elliott 1994, Egglishaw & Shackley 1977) whereas most predation by fish-eating birds on salmon parr and trout is in larger mainstem areas of rivers. There have been few studies of mainstem sites in Scotland (Gardiner 1983) and none has been conducted during winter and spring months when large numbers of parr are consumed by birds on some northern rivers.

To develop more robust models of impacts of fish-eating birds, it is necessary to quantify the variation in dynamics of fish populations in relation to habitat characteristics. The fish removal experiments in the present study covered a wide range of stream and river habitats. Because levels of recolonisation by salmon parr varied between sites, it is possible that the scope for compensation varies in relation to habitat. Moreover, fish population dynamics are known to vary between years due to differences in habitat conditions and/or population densities. Studies of both temporal and spatial variations in fish population dynamics by direct field experiment can be time-consuming and expensive. Therefore, current studies aim to develop a mechanistic understanding of the population dynamics of salmon parr that will provide guidance to target large scale, but effective, field experiments.

For example, new methods for automatically tracking movements of parr (Armstrong et al. 1996, 1997) have been used to determine whether de-watering of shallow areas of stream encourages fish to colonise vacant areas (Armstrong et al. 1998). These tracking methods are also being coupled with new models of fish distributions (Ruxton & Armstrong, in prep.) to explore the form of density-dependent growth in relation to population structure (Armstrong, Huntingford & Herbert, in prep.). By coupling model and experiment, it should be possible to explore the effects of scale on population dynamics. This will enable extrapolation, from the detailed observations of natural populations (that are often possible only in small streams) to larger systems.

The interactive use of modelling and experiment is a powerful procedure for developing our understanding of fish population dynamics. It would be ideal to conduct experiments at the whole-catchment level but numbers of smolts and returning adult fish can be measured accurately at only a few sites. The further development of fish counters and smolt and adult traps is needed to provide greater opportunities for large scale experiments. Even when it is impractical to test a full catchment-scale model over the whole fish life-cycle, it may be possible to test components by experiment. For example, we have shown we could construct a testable model to predict the impact of goosander ducklings on densities of salmon parr in the mainstem of the River Dee. With the provision of smolt traps, we could also develop testable models to predict the impact of birds on the smolt output from northern Scottish river catchments.

PREDATOR-PREY RELATIONSHIPS

Although sawbills and cormorants behave like specialist predators in their high mobility and their rapid recruitment to local areas of high fish availability, it is clear from their diet that they are generalists. Thus, in theory they have the potential for major impact on the populations of some of their prey species (Chapter 5, 'numerical and functional responses' et seq.). We therefore need to clarify the nature of their functional responses to the availability of their various prey. We have suggested that the extent to which they consume salmonids is perhaps determined by the availability of more easily caught prey and we now require information on both bird diet and prey availability to predict functional responses. An understanding of functional responses is fundamental to the prediction of the circumstances predisposing numerical responses, which in turn influences the recruitment to specific foraging sites and the turnover of birds there. It is now clear that these parameters need be considered both in devising experiments to control bird predation, and in predicting their outcome.

PRIORITIES FOR FUTURE WORK

With a view to developing testable models over the long-term, we propose the following priorities, for information on:

Fish

salmonid fish abundance, behaviour and year-round population dynamics at a catchment scale; fish community ecology in low, wide sections of both northern and southern river systems; the relative vulnerability of fish to predation on the mainstem compared with tributaries; continued improvements in fish counting methods to provide long-term data series and enable large-scale experiments; more detailed information on the ability of trout to recolonise depleted areas; the response of fish communities to selectively depleting single species and single age classes.

Birds

the long-term variation in sawbill abundance on specific rivers to identify local factors influencing the large variation in the survival of young ducklings and the apparent stability of wintering goosander numbers; cormorant and sawbill densities at stillwaters and rivers where there is also information on diet and fish abundance, to describe predator-prey functional responses, and to estimate the proportions of fish populations consumed by birds; precise estimates of the daily food intake of birds for modelling purposes; the variation in the diet of cormorants over a larger range of rivers and stillwaters; methods for distinguishing parr from smolts in material from bird stomachs; scaring/shooting experiments in conjunction with marked birds to investigate turnover. site selection by foraging birds to elucidate the vulnerability of fish in relation to habitat structure.