Possible ecosystem impacts of applying MSY policies from single-species assessment

Abstract

Ecosim models have been fitted to time-series data for a wide variety of ecosystems for which there are long-term data that confirm the models' ability to reproduce past responses of many species to harvesting. We subject these model ecosystems to a variety of harvest policies, including options based on harvesting each species at its maximum sustainable yield (MSY) fishing rate. We show that widespread application of single-species MSY policies would in general cause severe deterioration in ecosystem structure, in particular the loss of top predator species. This supports the long-established practice in fisheries management of protecting at least some smaller “forage” species specifically for their value in supporting larger piscivores.

Introduction

Single-species assessments and management controls may produce misleading predictions and pathological changes in ecosystems ( Hollowed et al ., 2000 ). Obvious examples include the appropriation of production by fisheries that would otherwise support valued predators such as marine mammals, and the appropriation of production needed to support valued piscivore species by reduction fisheries that target small planktivores. Application at ecosystem scale of harvest rates calculated from single-species assessments (e.g. F _MSY ) might result in widespread degradation in ecosystem function, and in considerably smaller overall yield and value than would be predicted from the sum of corresponding single-species yields. However, there is a counter-argument to this concern: in complex foodwebs, fishing a variety of species at their species-specific F _MSY might reduce both competition among and within these species, as well as predation mortality rates (owing to harvesting predators, so reducing the carrying capacity for unharvested predators).

The ecological basis of sustainable net production and harvest is compensatory response, especially of juvenile survival rates, to reductions in stock size by fishing. In single-species assessment, these responses are seen and measured mainly through stock-recruitment relationships, where “flat” curves (no effect of reduced spawner abundance on recruitment) are caused by compensatory changes in juvenile survival rate. Compensatory responses may be direct and immediate (for example, behavioural changes in foraging time and predation risk), or indirect and delayed owing to ecosystem-scale changes in predation risk and food production. Simple food-chain models predict stronger compensatory responses for any species when interactions with the rest of the system are accounted for, owing to the negative impact of fishing on its predators, and the positive impact on its food organisms. Figure 1 shows Ecosim estimates of equilibrium surplus production plotted against stock size for two fish species and two cases: (i) community structure and abundance (except the target species) held fixed; and (ii) community structure allowed to reach equilibrium in response to changes caused by fishing on the target species. In virtually all such cases that have been examined with Ecosim, regardless of what trophic level the target species occupies, the same pattern is predicted: indirect food-chain effects increase, rather than decrease, the productivity of the harvested species.

Figure 1

Typical Ecosim predictions of the equilibrium relationship between fishing mortality rate and yield for two species, a top predator (cod; left panel) and a mid-foodweb planktivore (herring; right panel) when holding all other biomass pools at Ecopath base biomasses (stationary assessment), and when allowing for ecosystem interactions (full compensation assessment).

If chain-like interactions were important, the main task in ecosystem-scale harvest policy design would be to seek policies that avoid uneconomical or socially unacceptable impacts of fisheries at different trophic levels on one another (and on other valued species), caused by appropriation of potential production and hence weakening of natural compensatory responses. Unfortunately, the web-like structure of interactions, along with the prevalence of trophic ontogeny (most species change trophic position as they grow), can cause much more complex and perverse trophic changes. It is critical to have at least some basic recognition of potentially perverse changes when we engage in developing ecosystem models. There are at least three ways that more complex interactions can cause apparently depensatory rather than compensatory responses to fishing, all involving ways that decreases in the abundance of one species may lead to increases in other species that affect it negatively:

• Cultivation-depensation effects (Walters and Kitchell, 2001). Adults of a dominant species may be abundant in the first place because they exert control on competitors and predators of their own juveniles (trophic triangles). Reducing adult abundance of the dominant species by fishing could result in increased abundance of the competitors and predators, and in turn to reduced juvenile survival, and consequently adult abundance, of the harvested species.

• Competition-predation trade-offs among species at the same trophic level . If the suite of species feeding at approximately the same trophic level varies in competitive ability vs. their ability to avoid predation, the dominant species in the suite is likely to be the one that is less vulnerable to predation. Targeted fishing on that dominant species may enhance its competitors (and hence weaken its compensatory responses). Further, harvesting its predator populations may lead to even stronger negative effects on the target species ( Yodzis, 2001 ).

• Predators feeding on multiple trophic levels . If some piscivore species of harvest interest is fed upon by a predator that is also happy to feed on the smaller fish species preferred by the piscivore (for example, seals and sea lions often feed on herring as well as on piscivores like cod and pollock, that also feed on herring), harvesting the piscivore should enhance (or at least stabilize) the food supply for the predator, which may in turn exert depensatory effects on the piscivore, should its abundance be reduced by fishing (higher predation mortality rate caused by a type II feeding response of the predator).

Fortunately, there is not much field evidence of severe disruptions of compensatory responses (as evidenced by non-stationarity or apparent depensatory patterns in single-species stock-recruitment relationships), except in cases where foodweb structure has been severely disrupted by overfishing (e.g. Newfoundland cod). Still, depensatory effects represent risks that need to be included in prudent designs for ecosystem harvest policy.

Therefore, it is not at all obvious whether we should expect higher or lower yields from an entire ecosystem than would be predicted from application of single-species harvest control policies. Pitcher and Cochrane (2002) , Christensen and Walters (2004) , and Okey and Wright (2004) used Ecosim ( Walters et al., 1997 ) to explore policy options for maximizing total ecosystem yield, net economic value, and other ecosystem-scale measures of total value from fishing. These optimization exercises revealed complex trade-offs between fisheries and management objectives, and one simple answer to such complexity is to argue that a “pretty good” ecosystem value might be obtained just by managing each species to its potential maximum sustainable yield (MSY). We evaluate this policy option using simulation rather than optimization procedures, through Ecopath/Ecosim biomass dynamics and size-structured foodweb models for a variety of marine ecosystems. The models have been used to assemble and summarize a wide variety of recent information on abundances and foodweb interaction patterns from stock assessments and diet studies, mostly for ecosystems that historically have been highly disturbed by fishing.

Methods

Ecosim cases

We selected 11 case studies ( Table 1 ) for which Ecosim has been shown to reproduce reasonably well the past patterns of change in relative abundance of major species (and sometimes catches and total mortality rate), given historical disturbance patterns (fishing mortality rates and/or effort over time, and in some cases changes in oceanographic or nutrient-loading indices of relative primary productivity). In the absence of such response data, ecosystem models might be constructed that either over- or underestimated the importance of various trophic interactions, and the need to account for these in harvest policy design.

Table 1

Biomass pools represented in Ecopath/Ecosim case examples used in MSY comparisons (pools in italics were included in MSY analyses; species code numbers in later figures correspond to row numbers; N/A, no reference available; YOY, young of year).

Ecosystem model, area, reference and pool						Ecosystem model, area, reference and pool
	Central North Pacific	Benguela	Baltic	Chesapeake Bay	Eastern Bering Sea	Eastern tropical Pacific	Georgia Strait	Gulf of Thailand	North Sea	Tampa Bay	West Coast of Vancouver Island
Source:	Cox et al .(2002a , b)	Shannon et al . (2004)	Harvey et al . (2003)	N/A	NRC (2003)	Olson and Watters (2003)	Martell et al . (2002)	FAO/FISHCODE (2001)	Christensen et al . (2002)	N/A	Martell (2002)
Group #
1	Large bigeye	Phytoplankton	Spring phytoplankton	Piscivorous birds	Baleen whales	Pursuit birds	Transient orcas	Rastrelliger spp.	Cod	0–3 Snook	Lingcod
2	Small bigeye	Benthic producers	Other phytoplankton	Non-piscivorous seabirds	Toothed whales	Grazing birds	Dolphins (resident orcas)	Scomberomorus	Haddock	3–12 Snook	Juvenile lingcod
3	Large yellowfin	Microzooplankton	Bacteria	Spot	Sperm whales	Baleen whales	Seals and sea lions	Carangidae	Herring	12–48 Snook	Dogfish
4	Small yellowfin	Mesozooplankton	Microzooplankton	Striped bass YOY	Beaked whales	Toothed whales	Halibut	Pomfret	Mackerel	48–90 Snook	Adult Pacific cod
5	Large albacore	Macrozooplankton	Mesozooplankton	Striped bass resident	Walrus and bearded seals	Spotted dolphin	Lingcod	Small pelagics	Norway pout	90+ Snook	Juvenile Pacific cod
6	Small albacore	Gelatinous zooplankton	Mysids/Invertebrates	Striped bass migratory	Seals	Meso dolphin	Dogfish shark	False trevally	Plaice	0–3 Red drum	Demersal fish
7	Large blue sharks	Anchovy	Meiofauna	Reef fish	Steller sea lions	Sea turtles	Adult hake	Large piscivores	Saithe	3–8 Red drum	Adult herring
8	Small blue Sharks	Sardine	Macrofauna	Bluefish YOY	Piscivorous birds	Large yellowfin	Juvenile hake	Sciaenidae	Sandeel	8–18 Red drum	Juvenile herring
9	Blue marlin	Round herring	Juvenile sprat	Bluefish adult	Adult pollock 2+	Large bigeye	Adult coho	Saurida spp.	Sole	18–36 Red drum	Eulachon
10	Large sharks	Other small pelagics	Juvenile herring	Weakfish YOY	Juvenile pollock 0–1	Large marlin	Juvenile coho	Lutianidae	Whiting	36+ Red drum	Euphausiids
11	Brown sharks	Chub mackerel	Juvenile cod	Weakfish adults	Other demersal fish	Large sailfish	Adult chinook	Plectorhynchidae	Birds	0–3 Sea trout	Pink shrimp
12	Swordfish	Juvenile horse mackerel	Adult sprat	Summer flounder	Large flatfish	Large swordfish	Juvenile chinook	Priacanthus spp.	Gurnards	3–18 Sea trout	Copepods
13	Other billfish	Adult horse mackerel	Adult herring	Atlantic croaker	Small flatfish	Large dorado	Demersal fish	Sillago	Horse mackerel	18+ Sea trout	Herbivorous Zooplankton
14	Mahi mahi	Mesopelagic fish	Adult cod	Menhaden YOY	Shallow pelagics	Large wahoo	Seabirds	Nemipterus spp.	Other predators	0–3 Sand trout	Phytoplankton
15	Small scombrids	Snoek	Salmon	Menhaden adults	Deep pelagics	Large sharks	Small pelagics	Ariidae	Raja	3–12 Sand trout	Detritus
16	Flying squid	Other large pelagics	Seals	Black drum	Deepwater fish	Rays	Eulachon	Rays	Seals	12+ Sand trout
17	Skipjack	Cephalopods	Detritus	Littoral forage fish	Jellyfish	Skipjack	Adult herring	Sharks	West mackerel	0–6 Mullet
18	Large leatherback	Small shallow-water hake		Alewife and herring	Cephalopods	Albacore	Juvenile herring	Cephalopods	Other invertebrates	6–18 Mullet
19	Small leatherback	Large shallow-water hake		American eel	Crustaceans	Auxis	Jellyfish	Shrimps	Juvenile cod	18+ Mullet
20	Large loggerhead	Small deepwater hake		American shad	Infauna	Bluefin	Predatory invertebrates	Crabs and lobsters	Juvenile haddock	0–3 Mackerel
21	Small loggerhead	Large deepwater hake		Bay anchovy	Epifauna	Small yellowfin	Shellfish	Trashfish	Juvenile saithe	3+ Mackerel
22	Jellyfish	Pelagic-feeding demersals		Channel and other catfish	Large Zooplankton	Small bigeye	Grazing invertebrates	Small demersals	Juvenile whiting	0–10 Ladyfish
23	Lance	Benthic-feeding demersals		Other flatfish	Herbivorous zooplankton	Small marlin	Carnivorous zooplankton	Demersal piscivores	Sprat	10+ Ladyfish
24	Squid	Pelagic sharks		White perch YOY	Phytoplankton	Small sailfish	Herbivorous zooplankton	Demersal benthivores	Dab	Jacks
25	Flying fish	Benthic sharks		White perch adults	Discards	Small swordfish	Kelp/seagrass	Shellfish	Copepods	Bay anchovy
26	Mesopelagic fish nekton	Apex sharks		Other elasmobranchs	Detritus	Small dorado	Phytoplankton	Jellyfish	Euphausiids	Pin fish
27	Epipelagic fish nekton	Seals		Gizzard shad		Small wahoo	Detritus	Sea cucumber	Other crustaceans	Spot
28	Epipelagic micronekton	Cetaceans		Blue catfish		Small sharks		Seaweeds	Echinoderms	Silver perch
29	Mesopelagic micronekton	Seabirds		Non-reef-associated fish		Miscellaneous piscivores		Coastal tuna	Polychaetes	Scaled sardine
30	Phytoplankton	Meiobenthos		Sandbar shark		Flying fish		Sergestid shrimps	Other macrobenthos	Mojarra
31	Detritus	Macrobenthos		Hard clam		Miscellaneous epipelagic fish		Mammals	Phytoplankton	Threadfin herring
32		Detritus		Other suspension feeders		Miscellaneous mesopelagic fish		Pony fish	Detritus	Menhaden
33				Other in/epifauna		Cephalopods		Benthos		Menidia (silverside)
34				Mesozooplankton		Crabs		Zooplankton		Catfish
35				Microzooplankton		Mesozooplankton		Juvenile pelagics		Bumper
36				Ctenophores		Microzooplankton		Juvenile Caranx		Caridean shrimp
37				Sea nettles		Large phytoplankton		Juvenile Saurida		Shrimp
38				Blue crab YOY		Small phytoplankton		Juvenile Nemipterus		Stone crab
39				Blue crab adults		Detritus		Phytoplankton		Blue crab
40				Oysters YOY				Detritus		Cyprinodontids
41				Oysters 1+						Poecilids
42				Soft clam						Pigfish
43				Phytoplankton						Gobies
44				Subaquatic vegetation						Rays
45				Benthic algae						Benthic invertebrates
46				Detritus						Macrozooplankton
47										Microzoolplankton
48										Infauna
49										Attached microalgae
50
51										Phytoplankton
52										Detritus

As indicated by the number of biomass pools included ( Table 1 ), the models vary in complexity. All include one or more species for which the basic Ecosim differential equations for biomass dynamics are replaced by age-structured population accounting, using either delay-difference models (“split pools”, Walters et al ., 2000 ) or full size–age “multistanza” accounting, with multiple life history stanzas representing major ontogenetic changes in mortality, feeding, and vulnerability to fishing ( Walters and Martell, 2004 ). Typically, Ecosim model behaviour is dominated not by size–age accounting details, but rather by the “foraging arena” ( Walters and Martell, 2004 ) limitations assumed for trophic interaction rates (food supplies, predation rates), owing to behavioural processes that organize fine-scale overlap patterns between predators and their prey.

For each case model, an informal procedure was used to fit the model to the abundance (and in a few cases, catch and total mortality rate) time-series. Possible causes of model-data discrepancies were first identified by examining modelled time-series patterns in components of change (food consumption, growth, mortality rate), then Ecopath base parameters (biomass, mortality rate, diet composition) and Ecosim dynamic response parameters (prey vulnerability exchange, foraging time adjustment) were varied to improve fit. Promising estimates were then examined more carefully using time-series fitting (maximum likelihood assuming lognormal errors in relative abundance measurement) methods similar to those used for single-species assessments. In no case has this stepwise process produced a “fully validated” model. Rather, the fitting procedure is a developing process, subject to changes over time as new information, and ideas about important interactions that may have been missed, become available.

Mainly, two types of patterns have been encountered during the fitting process: (i) “one-way-trip” ( Hilborn and Walters, 1992 ) declines in abundance under fishery development; and (ii) more complex cyclic or dome-shaped patterns indicative of both more complex fishing regimes, and, in some cases, apparently large changes in species productivity. Fitting one-way-trip data has been largely a matter of adjusting the Ecosim vulnerability exchange parameters that play the same role as compensatory response parameters (particularly recruitment curve slopes) in single-species models. For such species, Ecosim typically predicts similar dynamic response patterns to future fishing policies to those from single-species assessments. Fitting the more complex patterns has typically required examination of multiple alternative hypotheses that might equally well explain the data (e.g. environmental forcing vs. fishing vs. trophic interaction effects).

Key results from fitting process

The following subsections review a few key results from the fitting exercises, particularly in relation to the role of trophic interactions in shaping historical response patterns.

Top-down vs. bottom-up effects

There has been much debate in the field of ecology about the relative importance of top-down (predation) vs. bottom-up (food supply) effects of trophic interactions ( Power, 1992 ; Brett and Goldman, 1996 ). We commonly found strong evidence for top-down effects of harvesting (or protecting) predators on the productivity of their prey, but we did not see correspondingly strong evidence for bottom-up effects of harvesting prey on the productivity of their predators. In other words, a given trophic interaction can have strongly asymmetric effects, with typically stronger effects on the prey than on the predator.

Examples of strong top-down effects involved both marine mammals and piscivores. In the Georgia Strait case, increases in the numbers of seals and sea lions following marine mammal protection, since the mid-1970s, appear to be partly responsible (in conjunction with changes in primary productivity) for declines in survival rate and recruitment of a variety of fish species, ranging from Pacific salmon to demersal cod and flatfish. In the central North Pacific and eastern tropical Pacific, the exploitation of large piscivores (tuna, marlin) has apparently resulted in increased productivity of smaller piscivores (skipjack tuna, mahi mahi). In the Baltic Sea, Bering Sea, and North Sea cases, gadoids are predicted to have had strong predation impacts on small planktivores. Few would doubt the positive impact that cod declines around the Atlantic have had on shrimp productivity ( Worm and Myers, 2003 ), and this interaction is important in the North American case of the West Coast of Vancouver Island ( Martell, 2002 ).

Interestingly, only the relatively simple Baltic Sea model shows strong bottom-up impact of fishing on lower trophic levels. In the Georgia Strait and West Coast of Vancouver Island cases, herring were nearly eliminated by reduction fisheries during the 1960s, but there is no signal of this in the abundance time-series for piscivores. In the Bering Sea case, we have been unable to explain declines in Steller sea lions as having been caused directly by the groundfish fisheries ( NRC, 2003 ). In the case of the West Coast of Vancouver Island, development (and early overfishing) of the shrimp fishery had no apparent impact on cod and other benthic predators. Ecosim predicts that, in each of these cases, predators were able to adjust diet compositions so as to feel little impact of reductions in dominant prey species caused by fishing, and/or that productivity of alternate prey increased owing to reduced competition/predation with/by the harvested prey.

In contrast to the apparent lack of response to depletion of prey, there is good evidence for ecosystem-scale bottom-up effects apparently as a consequence of temporal changes in primary productivity. For several cases (Georgia Strait, West Coast of Vancouver Island, Bering Sea), roughly half the total variance in relative abundance can be explained by fitting a single temporal pattern of relative primary production anomalies. Considering the number of abundance time-series included, the variance reduction is far larger than would be expected by chance alone if each series were subject to an independent anomaly pattern, owing to species-scale environmental influence.

Cryptic predation effects

Ecopath base diet compositions have been estimated mainly from stomach studies aimed at identifying the main prey items that support predators. Unfortunately, an abundant predator may have large predation impacts on a low-biomass prey, without that prey making up a recognizable proportion of its diet. Such cryptic predation impacts may be the most common reason for model “failure” to represent important species interactions.

In the Georgia Strait case, visual observations of marine mammals feeding on juvenile salmon shortly after ocean entry support the assumption that mammal predation has an impact on juvenile survival, but diet studies have failed to sample at the space/time scales that could confirm this assumption. In several cases, calculated juvenile fish biomasses are low compared with adult biomasses, and both cannibalism by the adults and incidental predation by other (temporal) piscivores could have large impacts not evidenced in adult diets. In cases involving gadoids, lack of response to changes in prey availability may well represent recruitment control by cannibalism, rather than food resource availability.

Competition effects

The case models display two types of competition effect, neither particularly common nor dominating overall system behaviour: (i) release (increase) of relatively rare species when dominant species are reduced by fishing or predation; (ii) cultivation–depensation patterns (discussed above).

Examples of simple competitive release effects include predicted (and observed) increases in jellyfish in the Bering Sea following declines in small pelagic fish biomass. Both are planktivores. The observed increases in smaller pelagic predators following reduction in large tuna and billfish appear to reflect a mixed predation–competition effect. In a few cases, competitive effects were stronger than appeared reasonable on the basis of time-series patterns, and these model “errors” were solved by increasing the detail representing prey populations along with diet composition changes to reflect less severe diet overlaps than had been assumed initially.

Cultivation–depensation effects have been noticed for only a few dominant piscivore species, and then only under more extreme fishing scenarios than observed historically. Both pelagic system models show some reduction in juvenile survival rate of yellowfin tuna when this species is severely reduced, owing to increased abundance and predation by skipjack tuna and mahi mahi. However, fisheries for these smaller species may have prevented the effects from being as large as predicted in extreme model scenarios, and we may also have overestimated predation impacts by the smaller species, and diet overlaps.

Bycatch effects

The case studies include several instances where non-target species have suffered relatively severe fishing mortality. The most prominent examples involve billfish and shark populations in the pelagic ecosystem models, which have been reduced by longline fishing directed at tuna. In the Gulf of Thailand, where a wide variety of species are landed, it is difficult to classify species as bycatch or target. Responses in bycatch species have involved mainly simple one-way-trip depletions, with evidence of the usual compensatory response. In the West Coast of Vancouver Island case, a complex interaction exists between fisheries, where the shrimp fishery takes juvenile cod as bycatch while the finfish trawl fishery takes mainly larger cod; Ecosim explains persistence of the shrimp fishery partly by the negative impact on cod by the shrimp fishery itself.

Trophic mediation effects

There has been concern in Ecosim model development about indirect ( Dill et al ., 2003 ; Werner and Peacor, 2003 ) or trophic mediation effects, where interaction rates (or spatial overlap patterns) between predators and prey are influenced by abundances of some third, “mediating” type of organism (e.g. cover provided by seagrass and macroalgae in the Tampa Bay case). None of the cases show enough change in the mediating species to provide a clear test of the importance of such effects, particularly because the main effects might have taken place before observations started.

Estimating F _MSY and equilibrium system responses

We used a simple two-step procedure to evaluate the ecosystem-scale impact of widespread application of single-species MSY policies for each case. First, for each harvested species (each group with non-zero catch and/or bycatch), we ran a long simulation (1000+ years), where fishing mortality rate (F) of that species was incremented or decremented slowly, while holding all other F constant at Ecopath base values. F _MSY for the species was taken to be the F that resulted in maximum average catch. The simulation for each species was repeated for two alternative assumptions about ecosystem change in response to changes in fishing on the one species: (i) no response (food and predator populations held constant, so that the species is buffered from predator/prey effects); (ii) with full ecosystem-scale dynamic response effects in all species (but only to changes in F for the species being tested). As noted in the Introduction, these estimates of F _MSY and MSY are generally higher than those predicted by holding all other biomasses constant, because it includes indirect compensatory responses throughout the ecosystem to changes in abundance of the target species.

Second, for each of the two methods of estimating single-species F _MSY , we ran one additional long-term simulation, with the F for each harvested species set to its F _MSY estimated from the long simulation for that species. Biomasses and catches at the end of this simulation represent Ecosim predicted equilibrium values under the all-species F _MSY policy (not to be confused with a policy aimed at maximizing some multispecies MSY). This simulation amounts to pretending that long-term experience from managing each species under a variable harvesting regime could be applied successfully to the ecosystem as a whole, without (i) uncertainty arising from incomplete response information for each species, (ii) implementation errors associated with uncertain relationships between target Fs and policy instruments used to try and achieve the targets, and (iii) any adaptive learning after implementation about changes in productivity caused by applying F _MSY to other species. Assumption (iii) represents a conservative assessment of our ability to detect changes in productivity in single-species assessment situations.

Results

Estimated F _MSY values vary widely over species compared with the Ecopath base F estimates for all case systems ( Figure 2 ). Every system appears to have at least a few lightly fished species (F ≪ F _MSY ), and most have at least one species that has been overfished historically (F > F _MSY ). In most cases, the simulation to test widespread application of F _MSY over species involves quite a large change in F, compared with Ecopath base values, and such changes might be expected to cause large changes in ecosystem structure whether or not they result in sustainable harvests for all species.

Figure 2

Ratios of Ecopath base fishing rate F to estimated F _MSY (full compensation assessment) for all fished species in the case ecosystems (species codes are the rows in Table 1 ).

Figure 3 summarizes Ecosim predictions of the impact of widespread application of single-species MSY policies, using ratios of MSYs predicted when all species are harvested at their F _MSY rates, to the MSYs predicted for each species when all other species are fished at Ecopath base rates as indicators of relative sustainable (equilibrium) performance, for the two options (interactions held constant and full ecosystem response), respectively. Species are ordered by Ecopath estimates of mean trophic level (calculated from Ecopath base diet compositions). Only in the Georgia Strait case are the predicted performance ratios almost the same for all harvested species, whether they are examined individually or in combination. For all other cases, trophic interactions in combination with widespread F _MSY fishing rates are predicted to cause considerable change in community structure and to MSY performance very different for at least some species than would be predicted from single-species assessment.

Figure 3

Predicted ratios of equilibrium yield when all species are fished simultaneously at k × F _MSY , to equilibrium yield predicted for each species when only that species is fished at F _MSY and all others are fished at Ecopath base F (species code numbers are the rows in Table 1 ): (a) k = 1, applying the F _MSY estimated for each species while holding all other species constant; (b) k = 1, applying the F _MSY estimated for each species while allowing abundances (but not Fs) of other species to vary; (c) k = 0.7, as case (b) except applying 0.7F _MSY for all species, to reduce impact.

We expected to see strong negative relationships, with yields increasing at lower trophic levels at the expense of yields at higher trophic levels, as a consequence of appropriation by fishing of production at the lower levels. This effect is evident in most cases, but is also quite variable. We also expected the pattern to be stronger for the simpler models that do not include many species with complex trophic ontogeny, but surprisingly the pattern is most clear in the Tampa Bay case, one of the most complex models. For most models, the main conclusions from the equilibrium yield comparisons are that (i) yields under a many-species F _MSY policy can diverge grossly from single-species predictions, and (ii) the direction of divergence is not consistently related to trophic level.

Comparison of the yield ratios in Figure 3 reveals that including the full range of ecosystem effects in the estimates of F _MSY can cause quite large differences in predicted patterns of variation among species in the predicted relationship between trophic level and yield ratio. For several models, ratios are more often >1.0, and there are also some drastic shifts in ratios for intermediate trophic level species. Overall, however, the predicted patterns are similar for the two methods of MSY estimation, and for some cases are virtually identical.

Despite predicting large changes in the fish community under widespread F _MSY policy application, for most cases Ecosim predicts equilibrium total ecosystem yield from such policies to be quite similar to the sum of the individual MSY estimates ( Figure 4a ). This surprising result is not just due to the total system yields being dominated by one or a few lower trophic level species, although such species generally do make up the bulk of total yield. Changes in the predicted total landed value of catch for the cases where good data on prices are available ( Figure 4b ) do not show the expected decrease in landed value if the all-species F _MSY policy only caused a shift towards catching lower priced, small fish.

Figure 4

Ratios of predicted total ecosystem (a) yield (biomass), and (b) landed value (based on price multipliers), estimated as the sum of single-species MSY assessments, to equilibrium yield/landed value predicted by Ecosim when all species are fished at F _MSY (landed values only for a subset where reasonable price information was available for all species).

It is possible that the large performance changes shown in Figure 3a and b could be mitigated considerably by use of precautionary F policies, rather than F _MSY . To test this possibility, we did additional long-term simulations with the F for every species set to k × F _MSY , where k < 1 represents a single ecosystem-scale precautionary factor in setting target F. Generally, k < 1 would be expected to produce equilibrium yields of around k(2 − k) times MSY (assuming equilibrium biomass decreases linearly with increasing F), e.g. k = 0.7 implies a sustainable yield of around 0.9 MSY. Applying a precautionary k = 0.7 value in setting all target Fs is indeed predicted to have little impact on total yield, and to have the additional benefit of reducing the departures of catch from those predicted from single-species analysis ( Figure 3c ).

Discussion

The results in Figure 3 support previous analyses that warn of the need in ecosystem management to provide explicit protection for species whose value derives in part from support of other species (“forage fish”; Baxter, 1997 ) as well as from harvesting. It is not true that successful regulation of all harvesting to F _MSY would “solve” the ecosystem harvest management problem without creating severe conflicts among fisheries. Further, in cases such as the Gulf of Thailand, where most of the catch is taken by non-selective gear (trawling), implementation of F _MSY policies that differ widely across species would require major changes in fishing methods, towards more selective practices.

Predicted severity of impacts and trade-offs among different fisheries and values would not have been apparent from simple analysis of pairwise predator–prey relationships. It is fishing whole ecosystems with F _MSY policies for every species that leads to Ecosim predictions of severe erosion, not applying MSY policies to just a few selected species, while managing others more conservatively. When trade-offs between fisheries/species are examined pairwise, Ecosim typically predicts a much less severe reduction (or improvement) in MSY for each species, caused by “buffering” effects, such as availability of alternative prey for piscivores when some subset of the prey field is harvested.

The difference between whole-system vs. pairwise responses very likely explains the apparent contradiction between the finding from Ecosim data fitting about asymmetric effects (bottom-up effects not visible), and the predictions that most often warn of decreased yields from top trophic levels ( Figure 3 ), should there be more intense fishing on lower trophic levels. In most cases, whole-system application of F _MSY would involve fishing a variety of lower trophic level species much harder than in the past.

With growing pressure for future fishery development, and with most opportunities involving lower trophic level species ( Pauly et al ., 1998 ), there will be increasing demand for more precise calculations of just how far towards F _MSY policies it would be possible to move without impacting other fishery and existence values. The results presented will be questioned, perhaps with justification, as overestimating the effects of fisheries on one another, and we cannot claim that results are robust to uncertainties in inputs such as diet composition, assumed trophic ontogeny pattern, and productivity change driven by environmental factors. In some cases, the ratio results can be changed drastically just by changing a few diets or by making adjustments to avoid harvesting younger fish in age-structured populations. In cases involving large changes in predicted trophic level, there is no way to be sure that species currently too rare to be included in the models would not increase to take over the trophic roles of others impacted by fishing. Such possibilities are ecosystem-scale analogues of the “black hole” arguments that fishers often raise in response to assessments that indicate a need for reduced fishing (“there are still plenty of fish, just over the horizon, not included in the assessment calculations”), and they are particularly credible in cases where fisheries practices are selective enough not to depress community abundance in general, through simple swept-area impacts.

In addition to reasonable arguments about why Ecosim may overestimate the effects of trophic interaction, there are also arguments for stronger interaction effects than are currently represented in the structure. In particular, switching behaviour by large predators may cause unexpected, sequential depletion in prey type following depletion of any one type by fishing; for example, Springer et al . (2003) argue that killer whale predation may have caused declines in several marine mammal species in the Bering Sea, following the depletion of other whales.

Despite the large number of case studies and relatively long-term monitoring data now available, there simply is not yet enough empirical experience under widely contrasting fishing patterns to allow precise assessment of whether Ecosim (and other ecosystem and multispecies models) can successfully predict the impacts of complex ecosystem harvest management policies. However, we can conclude that there are significant dangers of adopting single-species management approaches that may be myopic, and that these are large enough to justify continuation and improvement in monitoring programmes aimed at detecting unexpected consequences, should they begin. There is also clear justification when uncertainty is so large for treating all fisheries development policies as adaptive management experiments, with a requirement to monitor trophic interaction and fish community effects along with the usual fishery performance indicators.

We specially thank Niels Daan, Poul Degnbol, and Lynne Shannon for their valued comment on the draft manuscript. Financial support was provided by a NSF grant to JFK for Apex Predators in Pelagic Ecosystems. VC acknowledges support from the few Charitable Trusts.