FISH and , 2003, 4,1^24

Hundred-year decline of North Atlantic predatory ®shes

Villy Christensen, Sylvie Gue¨ nette, Johanna J Heymans, Carl J Walters, Reginald Watson, Dirk Zeller &

Fisheries Centre, University of British Columbia,2204 Main Mall,Vancouver, BC, CanadaV6T1Z4

Abstract Correspondence: We estimate the of high-trophic level ¢shes in the North Atlantic at a spatial , scale of 0.58 latitude by 0.58 longitude based on 23 spatialized ecosystem models, each Fisheries Centre, University of British constructed to represent a given year or short period from1880 to1998.We extract over Columbia, 2204 7800 data points that describe the abundance of high-trophic level ¢shes as a function Main Mall, of year, primary production, depth, temperature, latitude, ice cover and catch composi- Vancouver, BC, tion.We then use a multiple linear regression to predict the spatial abundance for all CanadaV6T1Z4 North Atlantic spatial cells for 1900 and for each year from 1950 to 1999. The results Tel.: ‡1 604 822 5751 Fax: ‡1 604 822 8934 indicate that the biomass of high-trophic level ¢shes has declined by two-thirds during E-mail: the last 50-year period, and with a factor of nine over the century. Catches of high- v.christensen@ trophic level ¢shes increased from 2.4 to 4.7 million tonnes annually in the late1960s, ¢sheries.ubc.ca and subsequently declined to below 2 million tonnes annually in the late 1990s. The ¢shing intensity for high-trophic level ¢shes tripled during the ¢rst half of the time per- Received17Apr 2002 iod and remained high during the last half of the time period. Comparing the ¢shing Accepted30Aug2002 intensity to similar measures from 35 assessments of high-trophic level ¢sh populations from the North Atlantic, we conclude that the trends in the two data series are similar. Our results raise serious concern for the future of the North Atlantic as a diverse, healthy ecosystem; we may soon be left with only low-trophic level species in the sea.

Keywords biomass decline, ecosystem modelling, North Atlantic, predatory ¢sh

Introduction 2

Methodology 3 Ecosystem models from the North Atlantic 3 Assigning models to strata 7 Fisheries catches 8

Regression analysis 9 E¡ect of individual models on the regression analyses 11 E¡ect of catch composition on the regression analyses 13

Predicting biomass of predatory ®shes 13 Catches 14 mortalities 16

Discussion 17

Acknowledgements 21

References 21

# 2003 Blackwell Publishing Ltd 1 Biomass of North Atlantic predatory ¢shes VChristensenet al.

`Yousee something and then you try everything make comparisons to stock portfolio theory: a safe you can think of to make it go away; you turn it portfolio is diversi¢ed, hedging a bet on many di¡er- upside down and inside out, and push on it from ent sectors. Our living marine resources should be every possible angle. If it's still there, maybe managed in a similar way if we are to see but short- you've got something' (Cole1998, p.96) term gain and long-term loss; mining is not a viable option for managing living resources. How much ¢sh is there then in the sea? This is a Introduction crucial question for management of individual stocks How is the world doing today? We often tend to stick in individual areas, and, in that context, a question to Terra ¢rma when re£ecting on this question, but for which we have, at hand, a suite of approaches for the oceans have a role to play as well.We know that addressing it. Our interest in the present study is, global climate is closely linked to the oceans' circula- however, wider: we are asking the question with tion patterns, and that the oceans serve as a major regard to all species in a large area: how much ¢sh is food source, two roles too important to jeopardize. In there in the North Atlantic? that connection, it has been comforting to hear, as Even before embarking on an attempt to quantify we have for decades, that the food supply from the the total ¢sh biomass, we know that any estimate will oceans keeps increasing, but that comfort is begin- be very uncertain. However, just as is the case for ning to erode with reports that the global catches stock assessments, the biomass of ¢sh in itself is not have been decreasing for the last decade (Watson of real importance; what is relevant is how the bio- and Pauly 2001). We hear of a ¢sheries crisis in the mass of ¢sh has changed over time. Recognizing this North Sea, in North-eastern Canada; actually we apriori, we re¢ne the question: how has the biomass have heard of ¢sheries crises about everywhere regu- of ¢sh in the North Atlantic changed over the last larly for the last couple of decades.What is happening 100 years? to the ¢sh in the ocean? We focus our study on the last half of the 20th cen- We have to be concerned for several reasons, with tury, partly because we cannot expect to see any food supply being a major factor. But, our concern clear trends if the time period is too brief, and partly goes beyond this; we have seen drastic changes in because the 50-year period will cover the period fol- ecosystem structure in a number of marine systems, lowing the relative peace (for the ¢sh) of the Second a notable example being the Black Sea (Daskalov WorldWar up through a period with strong industria- 2002), and there is fear that ecosystems may change lizationand expansionof the North Atlantic ¢sheries, toalternate stable states if severelydisturbed.Wehave and onwards to the years of ¢sheries collapses that also seen repeatedly that once ¢sh populations col- have characterized the end of the 20th century lapse, it may take decades for them to rebuild, per- across the North Atlantic. haps because depensatory e¡ects may lead to such Estimating basin-level abundance of ¢sh is a novel changes in ecosystem states (Walters and Kitchell idea, as ¢sheries science has so far always worked on 2001). smaller scales (Pauly and Pitcher 2000) and we are To minimize the risk of adversely impacting the not familiar with any previous attempts we could oceans, we should seek to maintain healthy ecosys- use for guidance. does not have tems. Legislation to ensure this is by now incorpo- much tradition of addressing questions at such level, rated in laws and policy directives of many countries at least not questions that go beyond the amount of (e.g. Canada's Ocean Act, USA's Magnuson-Stevens catch that may be extracted from the oceans (Pauly Act, and the EU Common Policy), as well as 1996). In recent years, however, we have seen more in the UN Convention on the Law of the Sea where interest in reconstructing prior states of ecosystems nations have accepted a mutual obligation to take all (an early example of this is given in Christensen and appropriate actions to preserve the marine environ- Pauly1998), and ¢nd it important to look beyond our ment. An important part of this is to maintain su¤- own time horizon when evaluating the state of the cient stock sizes at all trophic levels as a safety oceans (Pauly1995). margin, avoiding the process of ¢shing downthe food In seeking to estimate the total ¢sh abundance, we web, where predatory species are gradually elimi- may take two di¡erent routes. One is a bottom-up nated (Pauly et al. 1998), since the hope that we may approach where we would attempt to estimate the be able to replace the predators in the sea is abundance of the individual species and sum these unfounded (Christensen 1996). Perhaps, we should abundances up to the North Atlantic level. Such an

2 # 2003 Blackwell Publishing Ltd, F I S H and F I S H E R I E S, 4,1^24 Biomass of North Atlantic predatory ¢shes VChristensenet al.

approach is, however, not likely to succeed; for one, intensive ^ even if pelagics are more known for exhi- we only have abundance estimates of a few popula- biting dramatic collapses (Pauly et al.1998). tions of ¢sh (Caddy et al. 1998), and the chance of As tools of analysis forassessing the biomass of ¢sh actually going out and measuring how much ¢sh in the North Atlantic, we have constructed a series there is in the sea is a formidable task beyond the of ecosystem models of North Atlantic ecosystems as capacity of any research group. Instead, we adopt a part of the `Sea Around Us' project (SAUP), and use modelling approach where we use a number of spa- these together with published models from various tial ecosystem models to quantify how much life areas in the North Atlantic to obtain a wide spatial there is in the area and at the point in time character- and temporal coverage. The models have varying ized by each model. We then use the physical and levels of spatial coverage and details. This paper pro- biological properties of the 0.58 latitude by 0.58 longi- vides an outline for how such a strategy has been tude grid cells in the area covered by the individual implemented to address basin-level questions, and models in a multiple linear regression to search for presents results from the data extraction that has patterns that may predict howabundance is distribu- been conducted based on the models. ted over space and time. The objective to estimate the abundance of ¢shes Methodology in the North Atlantic calls for a level of aggregation, the species level being too detailed. One option is to The methodology we have used to predict the bio- summarize the abundance of ¢shes by trophic level. mass of ¢sh inthe North Atlantic relies ona combina- We know the average trophic level for each group tion of ecosystem modelling, information from from either diet composition studies (e.g. through hydrographic databases, statistical analysis and GIS FishBase) or ecosystem models (e.g. ), and modelling. A £owchart for this approach is presented the models tell us how individual groups are distribu- in Fig. 1to guide further reading. ted between trophic levels. Hence, it becomes feasible to estimate the abundance of ¢sh at, e.g. trophic level Ecosystem models from the North Atlantic 4. However, apart from herrings, we do not have much knowledge about the ¢sh abundances at the The available information about biomasses at the lower trophic levels, e.g. for the smaller forage ¢shes. ecosystem level is very incomplete, making it neces- This re£ects the fact that forage ¢shes have beenof lit- sary to relyon modelling to obtain a coherent picture tle interest historically, and that the sampling meth- of the distribution and abundances of ¢sh in the ods in general use are unable to sample small ¢shes North Atlantic. We can base the modelling on the reliably. array of information that is available at the popula- Indications about historic abundances of, e.g. tion level, mainly due to stock assessments made as menhaden inChesapeakeBay,pointstotheseabeing part of the regulatory process. In addition, we have full of forage ¢sh, while some studies indicate that information from research surveys (which serve as a the abundance of forage ¢shes may have increased major information provider for the assessments), as in recent time due to cascading e¡ects caused by well as from biological oceanographic studies. A decreasing predator abundance as a result of hu- major part of the biological and ecological informa- man exploitation, e.g. for capelin in the Newfound- tion required for construction of the ecosystem mod- land area (Carscadden et al. 2001), and for small els is available from the FishBase database, available pelagics in the Black Sea (Daskalov 2002). However, online at http://www.¢shbase.org. The aim of the the evidence for cascading in marine ecosystems modelling e¡orts is to combine such information to is inconclusive (Pace et al. 1999; Pinnegar et al. derive a realistic picture of biomasses and their inter- 2000), and while the jury is out, we avoid the contro- action ina series of ecosystems throughout theNorth versy here by not dealing with the lower trophic Atlantic. levels. In the present study, we rely on ecosystem models Thus, in this study, we focus on high-trophic level constructed using the widely distributed Ecopath ¢shes, re£ecting that these organisms serve as indi- with Ecosim (EwE) approach and software for which cator species for the health status of marine ecosys- Christensen and Walters (2000) and Pauly et al. tems. The pattern emerging from studying human (2000) gave overviews of its capacityand limitations. impact on a variety of systems shows repeatedly that Ecopath models are intended to summarize the abun- thetoppredatorsarethe¢rsttogowhen¢shingturns dances and interactions of all major functional

# 2003 Blackwell Publishing Ltd, F I S H and F I S H E R I E S, 4,1^24 3 Biomass of North Atlantic predatory ¢shes VChristensenet al.

Figure 1 Schematic overviewof the methodology used for predicting the biomass of high-trophic level ¢shintheNorthAtlantic.

groups in an ecosystem, along with detailed descrip- incorporate time-series information in addition to tions of how we exploit such ecosystems through the year-speci¢c information on which the model ¢shing activities. A typical Ecopath model (such as description is based (see references in Table 2 for the bulk of those on which this study is based) may further details). The time-series information is used include 25^40 functional groups ranging from pri- to assess how well the model can replicate trends over mary producers to marine mammals, and incorpor- time in the ecosystem as part of what may be consid- ating a number of ¢shing £eets for which catches, ered a validation procedure. This, however, has lim- discards and bio-economical details are provided. ited implications for the present study, which does An overview of ¢sh species mentioned in this paper, not incorporate the time-dynamic aspects usually along with elements of their scienti¢c classi¢cation, considered when using the Ecosim routine of EwE is given inTable 1. (seeWalters et al. 1997, 2000). The present study is based on a total of 23 ecosys- For nearly all models, the time periods have been tem models, all of which are available from the ¢rst chosen to take advantage of available data sources. author (see also http://www.ecopath.org). The mod- Notably, the start of biomass data from stock assess- els describe15geographicalareas, and are each made ment has often dictated the period to be used for the to represent agivenyearor short time periodbetween models. The only models that break with this trend 1880 and 1998 (see Table 2). Many of the models are the two historic models for the North Sea

4 # 2003 Blackwell Publishing Ltd, F I S H and F I S H E R I E S, 4,1^24 Biomass of North Atlantic predatory ¢shes VChristensenet al.

Table 1 List of common names of ¢sh species mentioned in addition, a large part of the area is ice covered part of this paper along with corresponding scienti¢c and family the year. names. Our initial selection of available models included two that we later chose to exclude from the analysis. Common name Scientific name Family One of these described the Icelandic waters in 1950, but did not include any biomasses that were based Angler®sh Lophius piscatorius Lophiidae on empirical data. The other, from the Cantabrian Black seabass Centropristis striata Serranidae Sea, covered the narrow shelf area only,and our 0.58 Blackspot Bagellus bogaraveo Sparidae by 0.58 spatial cells did not represent this area in a seabream realistic fashion; hence, we would attribute the bio- Blue®n tuna Thunnus thynnus Scombridae Blue®sh Pomatomus saltatrix Pomatomidae masses to unrepresentative depths. Brill Scophthalmus rhombus Scophthalmidae Because of the uncertainty about abundance of Capelin Mallotus villosus Osmeridae small ¢sh in the North Atlantic in general, we focus Cod Gadus morhua Gadidae on the larger, predatory ¢shes for which much more Dog®sh Squalus acanthias Squalidae information is available, notably through stock Forkbeard Phycis phycis Phycidae Goose®sh Lophius americanus Lophiidae assessment and research surveys.We de¢ne the pre- Greenland halibut Reinhardtius Pleuronectidae datory ¢shes as those ¢sh groups for which the hippoglossoides trophic level is estimated to be 3.75 or more. This Haddock Melanogrammus Gadidae e¡ectively means that we include all ¢sh groups that aegle®nus predominantly eat prey species that feed on ¢sh, zoo- Hake Merluccius merluccius Merluccidae Halibut Hippoglossus Pleuronectidae plankton and/or small benthic organisms (i.e. we hippoglossus excluded all primarily planktivorous, herbivorous Herring Clupea harengus Clupeidae and detritivorous ¢shes). Horse mackerel Trachurus trachurus Carangidae We also exclude marine mammals and birds as Mackerel Scomber scombrus Scombridae well as high-trophic level invertebrates from our ana- Menhaden Brevoortia patronus Clupeidae Pollock Pollachius virens Gadidae lysis. Marine mammals are better dealt with in a Saithe Pollachius virens Gadidae separate study using a di¡erent methodology (see Salmon Salmo salar Salmonidae Kaschner et al. 2001) while for marine birds and Sea trout Salmo trutta trutta Salmonidae invertebrates, it is a consequence of their representa- Silver hake Merluccius bilinearis Merlucciidae tion being fairly super¢cial in the ecosystem models Spiny dog®sh Squalus acanthias Squalidae Striped bass Morone saxatilis Moronidae we have at hand. We also note that the biomasses Sturgeon Acipenser sturio Acipenseridae involved for these groups are negligible in anycase. Summer ¯ounder Paralichthys dentatus Paralichthyidae Generalizing slightly, the high-trophic level ¢shes Turbot Scophthalmus maximus Scophthalmidae may be considered to constitute what is commonly Weak®sh Cynoscion regalis Scianidae called`table ¢sh'.To illustrate this, a list of ¢sh groups Whiting Merlangius merlangus Gadidae included in the high-trophic level ¢sh category is pre- sented in Table 2. It re£ects that the species included (1880s), and for the Newfoundland area (1900). We are those of main interest for human consumption. have included these models to provide extremes on The de¢nition of the trophic level cut-o¡ point cho- the temporal scale, and fully realize that the biomass sen here is somewhat arbitrary, and indeed a few estimates used in these models are more uncertain groups are included, which we would not normally than those in the more current models. Therefore, consider predatory while in a few other cases, some we also investigate the impact that these and other groups one would expect to see included have been models have on the overall results, as is described in excluded. The reason for this may well be that the more detail below. trophic-level estimation depends on how well the We have also sought to include models that are diets (from which the trophic levels were estimated) extreme with regard to other characteristics; a nota- have been de¢ned, something we have not been able ble example is the Lancaster Sound model from to standardize completely between models. However, North-eastern Canada. Re£ecting the typical charac- the general patterns emerging from the selections teristics of such an arctic system, the model includes are very much in accordance with expectations, e.g. a variety of marine mammal groups, but only very few species (but fairly high biomasses on continental limited amounts of high-trophic level ¢shes; in shelves) in the colder, northern areas as compared to

# 2003 Blackwell Publishing Ltd, F I S H and F I S H E R I E S, 4,1^24 5 ims fNrhAlni rdtr ¢shes predatory Atlantic North of Biomass 6 Table 2 Overview of ecosystem models used for estimating abundance patterns of predatory ¢sh in the North Atlantic.

Area Year Cells Fish groups with trophic level > 3.75 Reference

North Sea 1880 369 Blue®n tuna, halibut and turbot, saithe, cod, Mackinson (2001) whiting, sharks, other predatory ®sh, rays and skates, sturgeon, haddock, horse mackerel, salmon and sea trout, gurnards, mackerel, west mackerel, brill, other prey ®sh Newfoundland (2J3KLNO) 1900 563 Greenland halibut, cod, large pelagic Heymans and Pitcher (2002b) feeders, skates, piscivorous small pelagic feeders Faroe Islands 1961, 132 Greenland halibut, cod, saithe, Zeller and Freire (2001), Zeller and Reinert (2001) 1997 other deep water, other demersal ®sh

North Sea 1963, 369 Saithe, cod, whiting, west mackerel, haddock, Christensen et al . (2002) VChristensen 1974 other predators, rays, mackerel, gurnards, horse mackerel, herring Gulf of Biscay 1970, 51 Extra-large pelagics, large sharks, tuna-like ®shes, Ainsworth et al . (2001) 1998 large deepwater ®shes, small sharks

Lancaster Sound 1980 169 Greenland halibut Mohammed (2001) al. et North Sea 1981 369 Saithe, other predatory ®shes, whiting, cod Christensen (1995) #

03BakelPbihn t,FISHadFISHERIES, E I R E H S I F and H S I F Ltd, Publishing Blackwell 2003 Scotian shelf 1982 160 Demersal piscivores, transient pelagics, halibut, dog®sh, cod, Bundy (2002) silver hake, pollock Gulf of Maine 1982 77 Summer ¯ounder, large pelagic feeders, blue®n Heymans (2001) & Georges Bank tuna, blue®sh, cod, large demersal feeders, pollock Morocco 1984 99 Large pelagics, very large demersals, large Stanford et al . (2001) deepwater ®shes, large and medium bathypelagics, small demersals, large demersal. sharks/rays, other sharks/rays Chesapeake Bay 1985 4 Blue®sh, summer ¯ounder, weak®sh, striped bass Baird and Ulanowicz (1989) Gulf of St. Lawrence (4RS) 1986 58 Greenland halibut, large cod, skates, large pelagics Morissette 2001) Newfoundland (2J3KL) 1986 563 Greenland halibut, cod, large pelagic feeders, Bundy et al . (2000) skates, piscivorous small pelagics US South Atlantic States 1996 81 Bill®shes, sharks, tuna, mackerel, snappers, Okey and Pugliese (2001) groupers, jacks, pelagic piscivores, demersal piscivores, benthic piscivores Norwegian-Barents Sea 1997 2307 Cod Dommasnes et al . (2001) Newfoundland (2J3KLNO) 1997 563 Greenland halibut, dog®sh, pollock, transient pelagics, Heymans and Pitcher (2002a) cod, transient mackerel, large demersal piscivores, skates

4 Greenland, west coast 1997 218 Greenland halibut, cod Pedersen and Zeller (2001) ,1^24 Iceland 1997 288 Greenland halibut Mendy and Buchary (2001) Biomass of North Atlantic predatory ¢shes VChristensenet al.

the more species-rich warmer, southern areas. We believe the shear mass of information will outweigh the few cases where the trophic-level estimates were problematic.

Assigning models to strata The coverage of the North Atlantic is incomplete, precluding simple scaling of £ows and rates from the individual ecosystem to the basin level, and calling for a strati¢cation scheme. The scheme wehave chosen builds onthe structure that is applied nette and Morato (2001)

 for catches and other data in the SAUP databases: Okey (2001) Gue 0.58 by 0.58 spatial cells (Watson et al.2001). Each of the ecosystem models covers a distinct geo- graphical area consisting of a variable number of the 0.58 spatial units (see Fig.2). As part of the present study, we have constructed a spatial model for each ecosystem using the Ecospace model incorporated in the EwE Software (Walters et al. 1999). Ecospace spatial cells covered by each model. The lists of ®sh groups indicate the selection

8 incorporates an Ecosim model in each spatial, non- land cell. In total, the models covered 24% of the area , coastal medium predators,

3.75 Reference of the North Atlantic, with the coverage reaching

> 40% in the depth strata where most concentrations of high trophic levels occur (Table 3). Exchange between spatial cells is modelled for each time step (typically monthly) while accounting

Pagellus bogaraveo

, for food availability, predation and ¢shing patterns. coastal sharks, spiny dogfish, jacks,snapper/grouper, black benthic seabass, piscivores, demersal piscivores,and cods hakes, redfish sharks coastal large predators, demersal large predators,

Phycis phycis medium sharks, medium demersal invertebratefeeders,medium rays, demersal predators, medium predators

Figure 2 Map of the15 geographical areas in the North Atlantic for which a total of 23 ecosystem models (shaded polygons, hatched background) were used to obtain estimates for a total of approximately18 000 0.58 by 0.58 spatial cells (shaded grey background).The total water area continued included in the analysis is 28 million km2. All the models for the Newfoundland/Grand Banks area o¡ Canada do not Statistical area codes, where appropriate, are givenused in brackets for for estimating clari®cation. The abundance. third See column indicates the the individual number models of 0.5 for further information about the groups. US Mid Atlantic Bight 1998 48 Billfishes, tunas, bluefish, goosefish, striped bass, weakfish, Azores 1997 240 Pelagic large predators, large deepwater ®shes, large Table 2 Area Year Cellscover the Fish groups with trophicsame level area.

# 2003 Blackwell Publishing Ltd, F I S H and F I S H E R I E S, 4,1^24 7 Biomass of North Atlantic predatory ¢shes VChristensenet al.

Table 3 Area covered by Ecopath models from the North www.me.sai.jrc.it) but for the present study, a 1-year Atlantic, total area and proportion of total area covered by production average representing 1999 was used, as Ecopath models. this was the only yearlyaverage available. The primary productivity maps have a spatial reso- Depth Area covered Total area lution of approximately 0.168, while the database stratum (m) (1000 km2) (1000 km2) Proportion used for the present study operates with 0.58 latitude by 0.58 longitude cells, i.e. with a resolution of one- 0±10 73 200 0.37 ninth of the SeaWiFS resolution. Therefore, a facility 11±50 472 1150 0.41 was included in Ecospace that aggregates the ¢ner 51±100 576 1408 0.41 resolution maps, averaging while maintaining the 101±200 754 2177 0.35 201±1000 1413 3507 0.40 overall mean, and prepares the basemap for the Eco- >1000 3567 19683 0.18 space modelling (for details, see the EwE User'sGuide, available at http://www.ecopath.org). Total 6855 28124 0.24 Temperatures at 10 m depth were obtained from a climatology based on the NOAA World Ocean Atlas 1998 (http://www.nodc.noaa.gov/OC5/wod98v2. The Ecospace models were constructed based on html) as implemented in the `Sea Around Us' Project general information (consulting FishBase) about database. Ice cover information was obtained from habitat and depth preferences for the functional the US National Snow and Ice Data Centre, Boulder, groups of the ecosystems. Primary production was Colorado (http://www.nsidc.org/index.html), in distributed spatially based on SeaWiFS data as form of monthly limits of sea-ice coverage. described below, while ¢shing e¡ort was distributed The environmental parameters are treated as sta- spatially based on distance to coast, depth zone pre- tic in this analysis, even though they,in reality,show ferences of £eets and ¢sh abundance. Fleet de¢ni- considerable interannual variability.We do not, how- tions varied between models, and the characteristics ever, have access to time series of environmental data were based on general knowledge of the ¢sheries of at the North Atlantic level and covering the time per- the North Atlantic. iod of interest (as such data do not exist). For each of the spatial model, the cells were distrib- uted between habitats based on depth only. The fol- Fisheries catches lowing depth strata were used for all models: (i) <10 m, (ii) 11^50 m, (iii) 51^100 m, (iv) 101^200 m, There is a relationship, but not a simple one, between (v) 201^1000 m and (vi) >1000 m (Table 3). Depth the ¢sh biomasses at any given time and how much information at the 0.58 by 0.58 scale was obtained ¢sh mayhave beencaught. If catches werehigh, there from the ETOPO5 data set available on the US must have been some high biomasses present to sup- National Geophysical Data Center'sGlobal Relief Data port these catches. However, high biomasses may CD (http://www.ngdc.noaa.gov/products/ngdc_pro- also be associated with low catches, if the reason is ducts.html) as implemented in the `Sea Around Us' low ¢shing e¡ort. However, we do not have reliable project database (http://saup.¢sheries.ubc.ca). data on development of ¢shing e¡ort over time either The predicted distributions in Ecospace models for the North Atlantic as a whole or for any major show marked sensitivity to primary productivity pat- parts of the basin; hence, it is not straightforward to terns (Martell et al. 2002). We therefore cooperated deduct overall biomass level from total catches. We with the Institute for Environment and Sustainabil- expect, however, that the catch composition will ityof the European Commission'sJoint Research Cen- change as a function of the biomass level of the pre- tre in Ispra, Italy to obtain global primary ferred ¢shing target, i.e. of the high-trophic level spe- productivity maps based on SeaWiFS data. The pri- cies. It is by now well established that ¢sheries mary productivity maps are based on a model that expansions go hand in hand with the process of `¢sh- incorporates the SeaWiFS estimated chlorophyll, ing down the food web' (Pauly et al. 1998), and we photosynthetically active radiation and sea surface can therefore use the catch composition by spatial temperature patterns (Hoep¡ner et al., unpublished unit to draw inferences about the overall biomass of data) based onthe model of Behrenfeld andFalkowski high-trophic level ¢sh species (see below). (1997).The maps areavailable ona monthlyand quar- The catches entering the regression analyses terly basis from October 1997 onwards (http:// come from the ecosystem models, which in turn have

8 # 2003 Blackwell Publishing Ltd, F I S H and F I S H E R I E S, 4,1^24 Biomass of North Atlantic predatory ¢shes VChristensenet al.

utilized numerous sources (see model references in high-trophic level catches (an e¡ect of `¢shing down Table 2). For this purpose, a routine has been added the food web'), and noting that it did not have any to the Ecopath software that allows allocation of observable e¡ect on the regressions; hence, one vari- catches of ecosystem groupings to the catch cate- able less is to be preferred. gories used in the SAUP database, as described Finally, when examining the regression, it was further below. In order to carry out this allocation, clear that the overall catches of tuna and bill¢shes we extracted catch distributions by the International show very little trend over the 50-year period under Standard Statistical Classi¢cation for Aquatic Ani- study (linear slope 0.1% of intercept, r2 ˆ 0.01, SD ˆ mals and Plants (ISSCAAP) categories (see http:// 12% of mean). This is in accordance with expecta- www.fao.org for details of this classi¢cation) for the tions as the catch composition of tuna has changed years and areas covered by the individual models, over the 50-year period; indeed, we now have evi- and used this to guide the distribution for the groups dence for declining mean trophic levels of catches where the allocation was not obvious. within the tunas (Pauly and Palomares 2001). Illus- The catches inthe SAUPdatabase are used forpre- trative of this is that blue¢n tuna catches were esti- dictive purposes based on the biomass regression. mated to be 38 000 tonnes in 1960 and 100 tonnes Catch data for 1900 were obtained from a variety of in 1999, while the decrease was compensated for by published sources and archives (Evermann 1904; increased catches of smaller,lower trophic level tunas Alexander 1905a,b; ICES 1906; Anonymous 1919, so as to maintain (within 1%) the total tuna catch. 1949, 1978; Cushing 1987; Sahrhage and Lundbeck Thus, the tuna and bill¢sh category turned out not to 1992; L o¨ pez Losa 2000). The main source for the beasigni¢cantpredictorofthebiomassofhigh-trophic catches from1950 onwards is the FAO catch database level ¢shes, and the category was omitted as a predic- (http://www.fao.org), with information added from tivevariable fromthe regressionanalysis. the Statlant database maintained by the Interna- tional Council for the Exploration of the Sea (ICES: Regression analysis http://www.ices.dk), as well as from ICES assessment working group reports. Spatial distribution of the We used multiple linear regression techniques as catches was undertaken using an elaborate, rule- implemented in S-Plus 6 software for all regression based procedure implemented and described by analysis (Anonymous 2001b). Prior to performing Wat s on et al. (2001). For this, the statistics were pro- the regression analyses, we used an additive and var- gressively disaggregated based on known distribu- iance stabilizing transformation (AVAS), of S-Plus to tions for the taxa, hydrographic conditions, and on study how individual variables are best transformed where reporting countries were permitted access to obtain linearity (Fig. 3). AVAS seeks for transfor- through ¢sheries agreements in the individual years. mations, Â(y) ˆ 1(x1) ‡ 2(x2) ‡ ÁÁÁ ‡ p(xp) ‡ e, The catches are distributed in 12 categories: (i) which provides a good additive model approximation anchovies, (ii) herrings, (iii) perches, (iv) tunas and for the data, yi, xi1, ..., xip,fori ˆ1, 2, ..., n, while seek- bill¢shes, (v) cods, (vi) salmoniformes, including ing to achieve variance stabilization. Based on the smelts and capelin, (vii) £at¢shes, (viii) scorpion- AVAS analyses, we concluded that logarithmic trans- ¢shes, including red¢sh, (ix) sharks and rays, (x) crus- formations were suitable for primary production taceans, (xi) molluscs and (xii) `other'groups. and biomass, while no transformations were For the regression analysis in the present study, required for year and latitude. For depth, indications we merged herrings and the salmoniformes (the lat- pointed to the use of a quadratic transformation ter being totally dominated by capelin). There are (truncated at 5000 m to avoid extrapolation). Ice indications, both from the catches and ecological cover was treated as a categorical variable (no ice studies, that capelin replaced herring during the cover, ice cover part of the year, and ice cover year 1970 s À1980s when herring abundance in the north- round) and hence required no transformation. The ern Atlantic was low (GjÖsaeter and Bogstad 1998). various catchcategories, as de¢ned above, weretrans- Also, the two species serve as important forage spe- formed using logarithmic transformations (catch in cies for the high-trophic level species considered kg kmÀ2 yearÀ1, with 1 kg kmÀ2 yearÀ1 added to in this study.We chose to combine the two inverte- enable log-transformation of catches of zero). brate groups (x) and (xi) in the regression analysis As data material for the regression analysis, we based on the expectation that high invertebrate extracted 7811records based on the 0.58 by 0.58 spa- catches are associated with low biomass levels of tial cells of the 23 ecosystem models. Each of the

# 2003 Blackwell Publishing Ltd, F I S H and F I S H E R I E S, 4,1^24 9 Biomass of North Atlantic predatory ¢shes VChristensenet al.

Figure 3 Additive and variance stabilizing (AVAS) transformations indicating how parameters (x-axis) may be transformed (y-axis indicate biomass, linear scale) to linearize the individual parameters while considering their joint e¡ects. Results indicate that no transformations are required for year and latitude, while a quadratic transformation is acceptable for depth, and log-transformations for primary production and biomass. Ice cover is treated as a categorical variable. records included estimates of biomass and catch of regression analysis using the inverse of the square high (3.75)-tropic level ¢shes, depth, distance from root of the number of non-land cells as weighting coast, water temperature at 10 m depth, ice cover factor in the models to which the given records category, number of seamounts and reefs, primary belong. production, an upwelling index based on latitude The multiple linear regression takes the following and basin-speci¢c temperature anomalies, catch by form, each of the catch categories de¢ned above, latitude and year of the model. log biomass†ˆa ‡ b1 Á year ‡ b2 Á log PP† 2 We were not able to use the following as predictive ‡ b3 Á depth ‡ b4 Á depth variables: distance from coast (it appears that the ‡ b5 Á latitude ‡ b6 Á ICEPart of year North Atlantic is soaccessible that any ¢shing ground ‡ b7 Á ICEYear round will be exploited; ¢shing was indeed the reason Eur- ‡ b8 Á log catch anchovies† opeans started crossing the Atlantic regularly); num- ‡ b Álog catch herring and smelts† ber of seamounts and reefs (both are negligible) and 9 ‡ b Á log catch perciformes† the upwelling index (there are so few upwelling areas 10 in the study area that no e¡ect can be expected in ‡ b11 Á log catch cods† the regressions). Further, we could not demonstrate ‡ b12 Á log catch flatfishes† any e¡ect of temperature, probably because of the ‡ b13 Á log catch scorpionfishes† inclusion of the latitude and ice cover terms. ‡ b14 Á log catch invertebrates† To prevent the records extracted from models cov- ering large areas from swamping those from other where a is the regression intercept, and b1 to b14 the models, each of the records were weighted in the slopes to be estimated by the regression;

10 # 2003 Blackwell Publishing Ltd, F I S H and F I S H E R I E S, 4,1^24 Biomass of North Atlantic predatory ¢shes VChristensenet al.

biomass is the predicted biomass of predatory ¢shes t-values give indications for the internal `ranking'of (g mÀ2); the parameters, i.e. which ones matter most (or PP is the average primary production (gC mÀ2 where the probability of exceeding the t-value by yearÀ1); chance is smallest). Due to covariation between vari- depth is the average depth (m); ables, we acknowledge that any interpretation of the latitude is the latitude of the observation; `rankings' should be treated with extreme caution.

ICEPart of year and ICEYear round are categorical para- The highest t-value is associated with the year para- meters that take the value 1 if the cell is ice cover- meter, followed by the intercept, latitude and depth. ed part of the year or year round, respectively, Primary production has a surprisingly low t-value, and the value 0 if not and catch variables are in partly re£ecting that depth and primary production kg kmÀ2 yearÀ1, with 1 kg kmÀ2 yearÀ1 being added show covariance, and partly that we do not have to accommodate log-transformations for zero models covering the Gulf Stream region across the catches. North Atlantic where primary production and depth Based on the 7811 records described above, the both are fairly high. regression coe¤cients and test statistics in Table 4 As with any other multiple regression, the results were obtained (see also Fig. 4). The multiple R2 of the are depending on the data material, and we need to regression is 0.859 with 7796 degrees of freedom. consider what we included in the analysis, both with The F-statistic is 3389 on14 and 7796 degrees of free- regard to outliers and predictive variables. To study dom, with a P-value of 0. (Given spatial autocorrela- this further, we have conducted a series of analyses, tion, we do not believe our cells to provide true subsampling from the original data sets. This is degrees of freedom, yet the results indicate that the described in more detail in the following sections. regression is fairly robust.) The residual standard error is 0.1280 on 7796 degrees of freedom. All para- E¡ect of individual models on the regression meters are highly signi¢cant (P < 0.001). analyses Summing up the regression results, we conclude that the predictive variables are able to explain the The regressions we obtainwill depend onwhat obser- major part of the variance in the data set (R2 ˆ 0.86), vations (here, ecosystem models) we include.Tostudy and the slopes have the right sign for the variables the robustness of the regressions, we have analysed where we had expectations about their impact. The the data using a jackknife approach (Sokal and Rohlf

Table 4 Parameterestimates and associated test statistics for multiple linear regressionto predict the biomass (log, g mÀ2)for predatory ¢shes (TL > 3.75) inthe North Atlantic during the second half of the 20thcentury.

Variable Value Standard error t-value Pr(>|t|) Transformation

Year À0.017415 0.000255 À68.3 0.0000000 None (Intercept) 35.873360 0.541 66.3 0.0000000 None Latitude À0.0458485 0.000858 À53.4 0.0000000 None Depth À0.0009162 0.0000194 À47.2 0.0000000 Quadratic Catch, anchovies À0.2390645 0.00731 À32.7 0.0000000 Logarithmic Catch, herring and capelin 0.1216986 0.00387 31.5 0.0000000 Logarithmic Catch, scorpion®shes 0.116684 0.00382 30.5 0.0000000 Logarithmic Catch, perches À0.1420623 0.00472 À30.1 0.0000000 Logarithmic Catch, cods 0.1119097 0.00495 22.6 0.0000000 Logarithmic Depth2 0.000000089 0.000000005 19.5 0.0000000 Quadratic Catch, ¯at®sh 0.0520826 0.00350 14.9 0.0000000 Logarithmic Ice cover, year-round À0.2849061 0.0224 À12.7 0.0000000 Factor Catch, invertebrates À0.0269938 0.00290 À9.3 0.0000000 Logarithmic Primary production 0.1646445 0.0195 8.4 0.0000000 Logarithmic Ice cover, part of year 0.0381208 0.0115 3.3 0.0008919 Factor

The primary production (PP) is in log, gC mÀ2 yearÀ1, while catches are in log, kg kmÀ2 yearÀ1. Depth is included with a linear and a quadratic term. The variables are arranged by t-value (value relative to standard error, given).

# 2003 Blackwell Publishing Ltd, F I S H and F I S H E R I E S, 4,1^24 11 Biomass of North Atlantic predatory ¢shes VChristensenet al.

Figure 4 (a) Observed versus predicted biomass (log-scales, tonnes kmÀ2) for predatory ¢sh in the North Atlantic during the 20th century.Solid line indicates average trend. (b) Plot of residuals (predicted^observed biomass, log-scale) versus observed biomass (log-scale, tonnes kmÀ2)for predatory ¢sh in the North Atlantic during the 20th century.Solid line indicates average trend.

1995), omitting one model at a time from the regres- as high biomass estimates for the North Atlantic sion. The details of jackknife analysis are presented basin, and illustrates the importance of including by Christensen et al. (2001). The jackknife approach extremes (here, a temperature extreme with low ¢sh can be used in a formal context for estimating con¢- biomasses) in the multiple linear regressions. The dence intervals of biomasses but because of the small model, which if omitted would result in the second number of observation groupings (models) and the highest biomasses, is that for the Norwegian Sea and use of a logarithmic scale, the con¢dence intervals the Barents Sea, re£ecting the same point. that could be derived here are too wide to be mean- The most noteworthy ¢nding from the jackknife ingful. We do not ¢nd that the standard method for analysis is that while the absolute estimates of estimating con¢dence intervals based on jackkni¢ng abundance are sensitive to inclusion or exclusion of is applicable to the analyses in the present study,and individual models, the overall trends over time hence we are for the time being not able to associate show remarkably little sensitivity to model deletion. con¢dence intervals with the results. Hence, the overall conclusions from the present The biomass trends resulting from the jackkni¢ng study are not very sensitive to the model selection. are presented in Fig. 5; it is clear that omitting the Rather, they are emergent properties based on many Lancaster Sound model would lead to nearly twice models.

12 # 2003 Blackwell Publishing Ltd, F I S H and F I S H E R I E S, 4,1^24 Biomass of North Atlantic predatory ¢shes VChristensenet al.

Figure 5 Illustrates the e¡ect of excluding individual models from the regression analysis in a jackknife fashion (excludingone model at atime and repeating regression analysis and predictions over time).The thick line with diamond markers indicates the regression with all models included. Jackknifed models are indicated only for the few cases generating stronger deviations from the mean trend.The minus sign indicates removal of that model from the analysis.

regressions omitting individual catch categories are E¡ect of catch composition on the regression fairly stable across the analyses. analyses The overall conclusion from the two series of In an exercise analogous to the jackkni¢ng for quan- regression analyses that omitted parts of the data is tifying the e¡ect of excluding individual models from that the results are robust with regards to the slope the regression analyses, we have investigated the of the resulting biomass trends, whereas the absolute e¡ect of excluding each of the nine individual catch values of the predicted biomasses are more uncer- categories from the regressions. Omitting individual tain. This is in line with the general expectation for catch categories was found to have negligible impact this form of multiple regression, i.e. we expect to be on the estimated biomasses of high-trophic level able to distinguish change better than we can predict ¢shes in the North Atlantic, and as can be seen from absolute values. Fig. 6, nearly all the predicted biomasses fall close to the original regression. Predicting biomass of predatory ®shes The e¡ects omitting catch categories has on the intercepts and slopes of the biomass regressions are We have derived a linear regression to predict the presented in detail by Christensen et al.(2001).Itis abundance of high-trophic level ¢shes in the North concluded that the intercepts and slopes of the Atlantic based on information from a number of

Figure 6 E¡ect on the estimated biomass of high-trophic level ¢sh in the North Atlantic of omitting individual catch groupings from the regression analysis.The thicker line with diamond markers is based on the original regression including all catch categories. Groups that when omitted have a noticeable impact on the results are indicated.

# 2003 Blackwell Publishing Ltd, F I S H and F I S H E R I E S, 4,1^24 13 Biomass of North Atlantic predatory ¢shes VChristensenet al.

Figure 7 Biomass distributions for high-trophic level ¢shes in the North Atlantic in (a) 1900 (b) 1950 (c) 1975 and (d) 1999. The distributions are predicted from linear regressions based on primary production, depth, temperature, year, ice cover, latitude and catch composition. Units for the legend are tonnes kmÀ2. ecosystem models dispersed over the region and in intended to describe general patterns only as they time from the late19th century through to the end of will obviously miss out on speci¢c events, such as the 20th century. The regression is based on a total the emergence of a big year-class of a major popula- of 18 024 spatial units of 0.58 by 0.58, and uses year, tion for reasons we cannot predict.The maps indicate depth, primary production, temperature, ice cover, a strong decline in biomass over the century studied; and catch quantity and composition to predict the we will return to this theme below. biomass. For predictive purposes, we then established a spa- Catches tialized database including the same information for all spatial units globally for 1900, as well as for all The catches of high-trophic level species, i.e. of the years from1950 through1999. For the present analy- main species of interest for human consumption, sis, however, we use the database only to predict bio- increased steadily through to the end of the 1960s, masses in the North Atlantic region to avoid and have declined as steadily since then (Fig. 8). The extrapolation beyond the area covered by the ecosys- catch level in the late1990s was thus lower than that tem models inTable 2. in1950 in spite of major developments in catch capa- Based on the biomass regression analysis applied city and technological progress, along with geogra- to the North Atlantic in 1900, 1950, 1975 and 1999, phical expansion across the North Atlantic region. the maps in Fig. 7 were derived. They indicate how The estimated spatial distributions of the high- biomasses were predicted to be distributed, and are trophic level catches are mapped in Fig. 9. They are

14 # 2003 Blackwell Publishing Ltd, F I S H and F I S H E R I E S, 4,1^24 Biomass of North Atlantic predatory ¢shes VChristensenet al.

Figure 8 Annual catches of high-trophic level ¢sh in the North Atlantic during1950^99. Primarily based on catch data from FAO (see Wat s on et al.2001for details).The catches include only ¢sh species with a trophic level of 3.75 or more.The trophic levels are mainly based on diet compositions and are extracted from FishBase.

Figure 9 Predicted catch distributions of high-trophic level ¢shes in the North Atlantic in (a) 1900 (b) 1950 (c) 1975 and (d) 1999.The catches are based on FAO catch data information supplemented with other sources using a rule-based system for spatial allocation (Watson et al.2001), and are here extracted for ¢sh with trophic level 3.75 (based on trophic levels in FishBase). Units for the legend are tonnes kmÀ2 yearÀ1.

# 2003 Blackwell Publishing Ltd, F I S H and F I S H E R I E S, 4,1^24 15 Biomass of North Atlantic predatory ¢shes VChristensenet al.

based on the rule-based method for distribution of In lieu of a direct measure, we will revert to a classic catches described byWatson et al. (2001), but applied estimation. Beverton and Holt (1957) describe the only to ¢sh species with a trophic level of 3.75 or ratio of catch to biomass for a population as a direct more. measure of ¢shing intensity as a surrogate for what is commonly described as `¢shing mortality', the method of choice in ¢sheries assessment for regulat- Fishing mortalities ing ¢shing e¡ort.We emphasize that the measure of The catch ¢gure and catch maps (Figs 8 and 9) by ¢shing mortality we have derived here is not directly themselves paint a dire picture of what has happened comparable to the mortality rates commonly in the North Atlantic area over the last 50^100 years, reported as the absolute level of the biomasses esti- but they do not directly address a major question:`Do mated here is associated with considerable uncer- we catch less because there are less ¢sh, or is it due tainty.Therefore, we prefer to interpret the measure to catch restrictions imposed to limit catches?' In as a relative indexof `¢shing intensity'only,especially order to address this question, we need to derive mea- since Beverton and Holt proposed this term for use sures of how ¢shing e¡ort has developed over time. in spatial applications. Ideally,we would have a direct measure of the ¢shing Combining information about catch and biomass e¡ort, but such information is pathologically poor levels over time, we obtain the results shown in the even inthis well-studied and highly regulated region. maps in Fig. 10 and in the plot in Fig. 11. The ¢gure

Figure 10 Estimated ¢shing intensity forhigh-trophic level ¢shes (TL  3.75) inthe North Atlantic region in (a) 1900 (b) 1950 (c) 1975 and (d) 1999.The ¢shing e¡ort is derived from spatial estimates of biomasses (Fig. 7) and catches (Fig. 9). Units for the legend are yearÀ1.

16 # 2003 Blackwell Publishing Ltd, F I S H and F I S H E R I E S, 4,1^24 Biomass of North Atlantic predatory ¢shes VChristensenet al.

Figure 11 Estimated catch (dark broken line,106 tonnes yearÀ1)and biomass (dark solid line,107 tonnes) of high-trophic level ¢shes in the North Atlantic during1950^99.The ratio between catch and biomass is an expression of annual ¢shing intensity (light solid line).

summarizes trends over the last 50 years for high- the form of bottom-up approaches summing up the trophic level ¢shes in the North Atlantic. Biomasses biomasses of all major ¢sh populations in the North are found to have been declining steadily over the Atlantic. This is re£ective of the varying time periods period at a rate that was slightly lower in the ¢rst for which assessments have been made for the many 20 years than in the last 30 years. The catches populations in the area. Thus, some form for model- peaked in the late 1960s, and have declined steadily ling is needed to ¢ll in the blanks, i.e. to provide esti- since to the extent that the level in 1999 was lower mates for the years where none have been made. than that in 1950. The resulting measure of ¢shing Also, far from all stocks are being assessed, making a intensity, estimated as the ratio between catch and bottom-up estimate likely to be an underestimate. biomass, provides part of the explanation. Fishing While waiting for a bottom-up approach, we can intensity increased with catches and has remained examine some trends fromvarious stock assessments nearly constant since the late 1960s, while both in the North Atlantic (Fig. 12). Assembling the plots catches and biomasses declined steadily. in the ¢gure was done by reviewing the majority of How long can this continue? There are no indica- the recent stock assessments made for the North tions in the results of a slowing down in the trend of Atlantic, and extracting biomass time series for declining biomass. The results thus predict that high-trophic level ¢shes. The most di¤cult task in high-trophic level ¢shes will be all but gone from the doing this was to decide which populations to include North Atlantic region within a few decades if the cur- here ^ there were so many that virtually all showed rent trend continues. the same patterns, be it target or nontarget species: massive decline during the period for which assess- ments were made, and a presently critical state of Discussion the stocks (see Table 5 for an overview of the state of Overall, we estimate that the biomass of high-trophic a¡airs for the majority of the high-trophic level spe- level ¢sh species in the North Atlantic declined by cies under ICES auspices). In contrast, there were two-thirds during the second half of the 20th cen- very few populations that did not show a clear tury, and by a factor of around nine for the century decline over time (such as cod at the Faroe Islands, as a whole. We should ask then, how reliable is this see Fig. 12). estimate? We note that the ¢nding seems to be fairly Thepatternthatseemstoemergewhenexamining robust to the extent that it did not matter much if biomass trends for a variety of North Atlantic ¢sh we omitted part of the data material on which the populations is one of massive decline, indicating that estimate is built, but despite the jackkni¢ng that the decline over time we are estimating in this study led to Fig. 5, we are at present unable to assign a is at least a feasible scenario. This is also the conclu- formal con¢dence interval to the estimate. It is also sion reached when examining the trends for the a fairly di¤cult task to ¢nd supportive evidence in high-trophic level species included in the stock

# 2003 Blackwell Publishing Ltd, F I S H and F I S H E R I E S, 4,1^24 17 Biomass of North Atlantic predatory ¢shes VChristensenet al.

recruitment database assembled by R. Myers (avail- 1960s, and has remained at what appears to be an able at http://¢sh.dal.ca/myers/welcome.html); see unsustainably high level ever since. For comparison, Christensen et al. (2001)for further details. the trend for ¢shing mortality in 35 populations in Our study indicates that ¢shing intensity in the the North Atlantic based on stock assessments is North Atlantic increased through the 1950s and compared to the ¢shing intensity from our study

Figure 12 Trend over time (1950^2001)inbiomass (thousand tonnes) of avarietyof high-trophic level ¢sh stocks inthe North Atlantic.The ¢gures are arranged byareawith statistical area codes used where appropriate (based on Lilly et al.1998; Brattey et al.2000; NAFO 2000; ACFM 2001; Anonymous 2001a; ICCAT 2001; Lilly et al.2001; O'Brien and Munroe 2001).

18 # 2003 Blackwell Publishing Ltd, F I S H and F I S H E R I E S, 4,1^24 Biomass of North Atlantic predatory ¢shes VChristensenet al.

Figure 12 continued

(Fig. 11) in Fig. 13. We conclude from the graph We are aware that the mean ¢shing intensity of that the two sets of ¢shing intensity bear much simi- assessed stocks presented in Fig. 13 should not be larity. interpreted as the mean ¢shing intensity for high- Several observations require mentioning when trophic level ¢shes in the North Atlantic. For this, examining Fig. 13; one is the di¡erent scaling of the the ¢shing intensities should have been weighted two y-axes. Fishing intensity is calculated as the according to population sizes. However, our inten- annual catch over the biomass and while our study tion is rather to discover something about the aver- indicates a ratio approaching 0.20 yearsÀ1, the indi- age population ^ since the measure of ¢shing cations of ¢shing mortalities from the assessments intensity is calculated as catch over biomass, it is a are three times higher. This indicates that the bio- measure of exploitation rate, a probability of being masses we estimate are considerably higher than caught and as such an ecologically more representa- those originating from averaging over the assessed tive measure. stocks. This apparent di¡erence may have several The maps and ¢gures presented here indicate that causes of which two need to be mentioned. First, only ¢shing intensity and catch levels have been higher some populations are subjected to stock assessment, in the North-eastern Atlantic than in the North-wes- and these tend to be the ones with highest exploita- tern Atlantic. Yet, the decline in biomass of high- tion rates. Second, biomass estimates based on trophic level ¢shes has been most severe in the regressions with log-transformations are quite north-western part of the basin. This may seem uncertain, and indeed we trust the trend in biomass inconsistent, but may well result because of the more than the face value of the estimates. At present, waters of the north-west being colder, deeper and which of the two explanations provide most toward less productive than in the north-east, i.e. the New an explanation is open, but we do expect both factors World waters are less resilient to ¢shing pressure to be contributing. than those in the Old World. Maps of hydrographic

# 2003 Blackwell Publishing Ltd, F I S H and F I S H E R I E S, 4,1^24 19 Biomass of North Atlantic predatory ¢shes VChristensenet al.

Table 5 Status of high-trophic level ¢sh stocks in the North-eastern Atlantic according to the ICESAdvisory Committee for (ACFM 2001).

Species Area State of stock/exploitation

Angler®sh North Sea (IV, VI) Stock is harvest outside of SBL Cod NE Arctic (I, II) Stock is outside of SBL Cod Norwegian coastal Spawning stock is at a historical low Cod Greenland (XIV, NAFO 1) Stock is outside SBL Cod Faroe Plateau (Vb1) Stock harvested outside SBL Cod West of Scotland (VIa) Stock remains outside SBL Cod North Sea (IV, VIId, IIIa) Stock outside SBL Cod Kattegat (IIIa) Stock considered outside SBL Cod Kattegat (IIIa) Stock considered outside SBL Cod Irish Sea (VIIa) Stock remains outside of SBL Cod VIIe-k Stock outside of SBL Cod Icelandic waters (Va) Stock near historic low Greenland halibut NE Arctic (I, II) Stock considered outside SBL Greenland halibut Greenland (V, XIV) Stock harvested outside SBL Haddock Faroe (Vb) Stock outside SBL Haddock West of Scotland (VIa) Stock harvested outside SBL Haddock Rockall (VIa) Stock remains outside SBL Haddock North Sea (IV, IIIa) Stock being harvested outside SBL Haddock Irish Sea (VIIa) Stock harvested outside of SBL Hake Southern (VIIx, IXa) Stock outside SBL Hake Northern (IIIa, IV, VI, VIII, VIIIa,b,d) Stock is outside SBL Red®sh NE Arctic (I, II) Stock considered outside SBL Saithe NE Arctic (I, II) Stock within SBL following good year classes Saithe Icelandic waters (Va) Stock considered outside SBL Saithe Faroe (Vb) Stock harvested outside SBL Saithe North Sea (IV, IIIa, VI) Stock is within SBL Whiting Irish Sea (VIIa) Stock remains outside of SBL

ICES statistical area codes are given in brackets. Only two smaller stocks (of saithe) are considered within safe biological limits (SBL).

and productivity patterns lend some credibility to know if this demand was met by a higher biomass of such a hypothesis. If this observation has any merit, the forage species and/or by higher mortality rates it means that care should be exercised when transfer- for the groups. On the other hand, we can be certain ring experience on managing North-easternAtlantic that the product of these two, i.e. the production of stocks to the North-western Atlantic. prey species must have been higher.We note in pas- In the present study, we were not able to reliably sing that there are ways of obtaining supporting evi- estimate the abundance of forage ¢shes, and chose dence ^ egg and larval surveys have been conducted to omit these from the results. This is re£ective of our for a century, and even if they were rather sporadic limited knowledge of these groups and is indicative in the early part of the 20th century, there is a wide- of ¢sheries science focusing on the exploited target spread coverage of standardized egg surveys from species, and largely ignoring the of the sys- the 1960s through to the 1980s or beyond. Unfortu- tems on which the ¢sheries rely. nately, the surveys have typically focused on target Ecosystem models may indeed help one to draw species only, and the eggs or larvae of the species of inferences about prey abundance from predator lower trophic level may not have beenanalysed. Since demand.Wecan conclude that if the biomasses of pre- the samples are stored in many laboratories, it is, at datory ¢shes were indeed much higher in past eco- least in principle, still possible to obtain such infor- systems (as all evidence points to), they must have mation given su¤cient interest and resources. been consuming more than todays' impoverished Another source of evidence may come from the size fauna would lead one to think. However, we do not compositions of forage species from`old'diet composi-

20 # 2003 Blackwell Publishing Ltd, F I S H and F I S H E R I E S, 4,1^24 Biomass of North Atlantic predatory ¢shes VChristensenet al.

Figure 13 Trend in mean annual ¢shing mortality for assessments of 35 populations of high-trophic level ¢sh species from the North Atlantic (broken line, primary y-axis). Sources are the same as for the biomasses in Fig. 12.The solid line (secondary y- axis) indicates the annual ¢shing intensity from the present study (Fig. 11).The insert shows the two series plotted versus each other, with the values from this study on the x-axis.

tion studies of predatory ¢shes. Based on the size funded by the Pew Charitable Trusts (http:// distributions, mortality rates can be estimated given www.pewtrusts.com). DrJoshua S. Reichert,Director readily available growth parameters. of the Environment Programme of the Trusts, posed It should be noted that the overall ¢ndings of this several challenges to us in the conceptualization study are not caused by catch trends over time or for phase of the project, including the question:`What is that matter bya systematic error in catch trends over the health status of the North Atlantic?' It is the quest time. Even if the catches are totally omitted from the to answer his question that lead to this publication. regression analysis, we obtain a highly signi¢cant A brief version of this contribution was presented at regression in which the year-term explains most of the AAAS Annual Meeting in Boston, February the change in biomass. Hence, the regression we pre- 2002, and we thank SeaWebb/Compass, notably sent in this study does not serve to explain what is Nancy Baron, for excellent cooperation. causing the changes in biomass, be it environment, and Daniel Pauly acknowledge support from Cana- ¢shing pressure or a combination. We do ¢nd, how- da's National Scienti¢c and Engineering Research ever, that the decrease in biomass is associated with Council.We thank Alida Bundy, Paul Fanning, Nico- a marked increase in ¢shing pressure.While we fully las Hoep¡ner, George Lilly, Ernesto Lo¨ pez Losa, Ðse recognize that environmental changes may impact MobrÔten, Ransom Myers, David Preikshot, Don ocean productivity patterns (see Beaugrand et al. Stansbury, Kim Stobberup, Rashid Sumaila, Haral- 2002), we have no reason to believe that the environ- dur Porbjo« rnsson, Hreikar Valty¨ sson and other col- mental trend over time would lead to a consistent leagues for making data available, and helping us decrease in the biomass of high-trophic level ¢shes. access and/or interpret it.We also thank Tony Smith We have developed and applied a methodology to and two reviewers for providing useful comments assess the state of the high-trophic level ¢sh popula- on the manuscript. tions of the North Atlantic, and have concluded that the biomass of these commercially and ecologically References important species is dwindling rapidly.We stress that what happens to the high-trophic level species serves ACFM (2001) Report of the ICES Advisory Committee on Fisheries Management. ICES CM 2001/ACFM. as an indicator for what we do to the ocean, and Ainsworth, C., Ferris, B., Leblond, E. and Gue¨ nette, S. (2001) hence we conclude that all is not well with the ocean. The Bayof Biscay,France,1998 and1970 models. In: Fish- eries Impacts on North Atlantic Ecosystems: Models and Acknowledgements Analyses (eds S. Gue¨ nette, V. Christensen and D. Pauly). Fisheries Centre Research Reports 9 (4), 271^313. This contribution is a result of the `Sea Around Us' Alexander, A.B. (1905a) Statistics of the Fisheries of the New project (http://saup.¢sheries.ubc.ca), initiated and England States, 1902. Report of the Bureau of Fisheries

# 2003 Blackwell Publishing Ltd, F I S H and F I S H E R I E S, 4,1^24 21 Biomass of North Atlantic predatory ¢shes VChristensenet al.

1904, US Commission of Fish and Fisheries. Government Christensen,V.(1996) Managing ¢sheries involving predator Printing O¤ce,Washington, pp.245^325. and prey species. Reviews in Fish Biology and Fisheries 6, Alexander,A.B. (1905b) Statistics of the Fisheries of the South 417^442. AtlanticStates, 1902. Report of the Commissioner, Part Christensen,V., Gue¨ nette, S., Heymans, J.J.,Walters, C.J.,Wat- XXIX, USCommission of Fish and Fisheries. Government son, R., Zeller, D. and Pauly, D. (2001) Estimating ¢sh Printing O¤ce,Washington, pp.343^410. abundance of the North Atlantic,1950^1999.In:Fisheries A nony mous (1919) Census of Industry, part III: Fisheries Sta- Impacts on North Atlantic Ecosystems: Models and Ana- tistics1917. Dominion Bureau of Statistics, J. de Labroqu- lyses (eds S. Gue¨ nette,V.Christensen and D. Pauly). Fish- erieTache¨ ,Ottawa,p.104. eries Centre Research Reports 9 (4),1^25. Anonymous (1949) Fisheries Statistics for Canada 1947. Christensen,V., Beyer, J.E., Gislason, H. andVinter, M. (2002) Dominion Bureau of Statistics, Department of Trade and A comparative analysis of the North Sea based on Eco- Commerce, Ottawa, p.104. path with Ecosim and Multispecies Virtual Population Anonymous (1978) Historisk Statistikk 1978. Norges O¤- Analysis. In: Proceedings of the INCO-DC Conference sielle Statistikk XII 291. Published by Statistisk Sentral- Placing Fisheries in Their Ecosystem Context Gala¨ pagos byrÔ, Oslo, ISBN 82-537-0758-4. Islands, Ecuador,4^8 December 2000 (edsV.Christensen, Anonymous (2001a) Report of the North-WesternWorking G. Reck and J.L. Maclean). ACP-EU Fisheries Research Group. ICES CM 2001/ACFM:20. Reports, Brussels12,39. Anonymous (2001b) S-Plus 6 forWindows Guide to Statistics, Christensen, V. and Pauly, D. (1998) Changes in models of Vol.1. Insightful Corporation, Seattle, WA. aquatic ecosystems approaching carrying capacity. Eco- Baird, D. and Ulanowicz, R.E. (1989) The seasonal dynamics logical Applications 8,S104^S109. of the Chesapeake Bay ecosystem. EcologicalMonographs Christensen,V.andWalters, C.J.(2000) Ecopathwith Ecosim: 59,329^364. methods, capabilities and limitations. In: Methods for Beaugrand,G.,Reid,P.C.,Ibanez,F.,Lindley,J.A.and Assessing the Impact of Fisheries on Marine Ecosystems of Edwards, M. (2002) Reorganization of North Atlantic the North Atlantic (eds D. Pauly and T.J. Pitcher). Fisheries marine copepod biodiversity and climate. Science 296, Centre Research Reports 8 (2),79^105. 1692^1694. Cole, K.C. (1998) The Universe and theTeacup:The Mathematics Behrenfeld, M.J. and Falkowski, P.G. (1997) Photosynthetic ofTruth and Beauty.HarcourtBraceandCo.,NewYork. rates derived from satellite-based chlorophyll concentra- Cushing, D.H. (1987) The Provident Sea. Cambridge Univer- tion. Limnology and Oceanography 42,1^20. sity Press, Cambridge. Beverton, R.J.H. and Holt, S.J. (1957) On the Dynamics Daskalov, G.M. (2002) Over¢shing drives a of Exploited Fish Populations, Facsimile Reprint 1993. in the Black Sea. Marine Ecology Progress Series 225, Chapman & Hall, London. 53^63. Brattey, J., Cadigan, N.G., Lilly, G.R., Murphy, E.F., Shelton, Dommasnes, A., Christensen,V., Ellertsen, B., Kvamme, C., P.A. and Stansbury, D.E. (2000) An assessment of the Melle,W., Nottestad, L., Pedersen,T. et al. (2001) Ecosys- cod stock in NAFO Subdiv 3Ps in October, 2000. DFO tem model of the Norwegian Sea and Barents Sea. Canadian Stock Assessment Secretariat Research Docu- In: Fisheries Impacts on North Atlantic Ecosystems: ment134. Models and Analyses (eds S. Gue¨ nette, V. Christensen Bundy, A. (2002) Exploring multispecies harvesting strate- and D. Pauly). Fisheries Centre Research Reports 9 (4), gies on the eastern Scotian Shelf with Ecosim. FAO/UBC 313^240. Fisheries Centre Research Reports10(2),112^117. Evermann,B.W.(1904)Statistics of the Fisheries of the Middle Bundy, A., Lilly, G. and Shelton, P.A. (2000) A mass balance AtlanticStates . Report of the Commissioner, Part XXVIII, model of the Newfoundland^Labrador shelf. Canadian US Commission of Fish and Fisheries. Government Technical Report of Fisheries and Aquatic Sciences Printing O¤ce,Washington, pp.433^540. 2310. GjÖsaeter, H. and Bogstad, B. (1998) E¡ects of the presence Caddy,J.F., Carocci, F.and Coppola, S. (1998) Have peak ¢sh- of herring (Clupea harengus) on the stock^recruitment ery production levels been passed in continental shelf relationship of Barents Sea capelin (Mallotus villosus). areas? Some perspectives arising from historical trends Fisheries Research 38,57^71. in production per shelf area. Journal of Northwest Atlantic Gue¨ nette, S. and Morato, T. (2001) The Azores Archipelago Fishery Science 23, 191^219. i n 1997. I n: Fisheries Impacts on North Atlantic Ecosys- Carscadden, J.E., Frank, K.T.and Leggett,W.C. (2001)Ecosys- tems: Models and Analyses (eds S. Gue¨ nette,V.Christensen tem changes and the e¡ects on capelin (Mallotusvillosus), and /surname> D. Pauly). Fisheries Centre Research a major forage species. Canadian Journal of Fisheries and Reports 9 (4), 241^270. Aquatic Sciences 58, 73^85. Heymans, J.J. (2001)The Gulf of Maine,1977^86. In: Fisheries Christensen, V. (1995) A model of trophic interactions in Impacts on North Atlantic Ecosystems: Models and the North Sea in 1981, theYear of the Stomach. Dana11, Analyses (eds S. Gue¨ nette, V. Christensen and D. Pauly). 1^28. Fisheries Centre Research Reports 9 (4),128^150.

22 # 2003 Blackwell Publishing Ltd, F I S H and F I S H E R I E S, 4,1^24 Biomass of North Atlantic predatory ¢shes VChristensenet al.

Heymans, J.J. and Pitcher,T.J. (2002a) A model of the marine Mohammed, E. (2001) A preliminary model for the Lancas- ecosystem of Newfoundland and Southern Labrador ter Sound region in the 1980s. In: Fisheries Impacts on (2J3KLNO) in the time periods1985^87 and1995^97. In: North AtlanticEcosystems: Models and Analyses (eds S. Ecosystem Models of Newfoundland for the Time Periods Gue¨ nette,V. Christensen and D. Pauly). Fisheries Centre 1995, 1985, 1900 and 1500 (eds T.J. Pitcher and J.J. Hey- Research Reports 9 (4),99^110. mans). Fisheries Centre Research Reports10(5),5^43. Morissette, L. (2001) Mode¨ lisation E¨ cosyste¨ mique Du Nord Du Heymans, J.J. and Pitcher,T.J. (2002b) A Picasso-esque view Golfe Du Saint-Laurent. The© se de maitrise. Universite du of the marine ecosystem of Newfoundland and southern Quebec a Rimouski, Canada,170 pp. Labrador: models for the time periods1500 and1900. In: NAFO (2000) Report on Standing Committee on Fisheries Ecosystem Models of Newfoundland for the Time Periods (STACFIS), June 2000. 1995, 1985, 1900 and 1500 (eds T.J. Pitcher and J.J. O'Brien, L. and Munroe, N.J. (2001) A Report of the 4th Heymans). Fisheries Centre Reports10(5),44^70. Transboundary Resources Assessment CommitteeMeet- ICCAT (2001) Report of the ICCAT SCRS.West Atlantic Blue- ing Assessment of the Georges Bank Atlantic Cod Stock ¢n tuna stock assessment session, Madrid, Spain,18^22 for 2001. Northeast Fisheries Science Center Reference September 2000. Detailed report. Document1^10. ICES (1906) Bulletin Statistique Des Peª ches Maritimes Des Okey,T.A. (2001) A`straw-man' Ecopath model of the Middle Pays Du Nord de l'Europe,Vol.1. HÖst and Sons, Copenha- Atlantic Bight continental shelf, United States. In: Fish- gen. eries Impacts on North Atlantic Ecosystems: Models and Kaschner, K., Watson, R., Christensen, V., Trites, A.W. and Analyses (eds S. Gue¨ nette, V. Christensen and D. Pauly). Pauly, D. (2001) Modeling and mapping trophic overlap Fisheries Centre Research Reports 9 (4),151^166. between marine mammals and commercial ¢sheries in Okey,T.A. and Pugliese, R. (2001) A preliminary Ecopath the North Atlantic. In: Fisheries Impact on North Atlantic model of the Atlantic continental shelf adjacent to the Ecosystems: Catch, E¡ort and National/Regional Ecosys- southeasternUnited States. In: Fisheries Impacts onNorth tems (eds D. Zeller, R. Watson, T. Pitcher and D. Pauly). AtlanticEcosystems: Models and Analyses (eds S. Gue¨ n- Fisheries Centre Research Reports 9 (3),35^45. ette, V. Christensen and D. Pauly). Fisheries Centre Lilly, G.R., Shelton, P.A., Brattey, J., Cadigan, N.G., Healey, Research Reports 9 (4),167^181. B.P., Murphy, E.F. and Stansbury, D.E. (2001) An assess- Pace, M.L., Cole, J.J., Carpenter, S.R. and Kitchell, J.F. (1999) ment of the cod stock in NAFO Divisions 2J‡3KL. DFO Trophic cascades revealed in diverse ecosystems. Trends Canadian Stock Assessment Research Document 44. in Ecology and Evolution14, 483^488. Lilly,G.R., Shelton, P.A., Brattey,J., Cadigan, N., Murphy,E.F., Pauly, D. (1995) Anecdotes and the shifting baseline syn- Stansbury, D.E., Davis, M.B. et al. (1998) An assessment drome of ¢sheries.Trends in Ecology and Evolution10,430. of the cod stock in NAFO Divisions 2J‡3KL. DFO Cana- Pauly,D. (1996) One hundred milliontonnes of ¢sh, and ¢sh- dian Stock Assessment Research Document15,102. eries research. Fisheries Research 25,25^38. Lo¨ pez Losa, E. (2000) Fishing and ¢sh processing in the Bas- Pauly,D., Christensen,V., Dalsgaard, J., Froese, R. andTorres, que country,1800^1936. An economic history.PhD The- F., Jr (1998) Fishing down marine food webs. Science sis. University of the Basque Country, Spain, 596 pp. (in 279, 860^863. Spanish). Pauly, D., Christensen, V. and Walters, C. (2000) Ecopath, Mackinson, S. (2001) Representing trophic interactions in Ecosim, and Ecospace as tools for evaluating ecosystem the North Sea in the1880s, using the Ecopath mass-bal- impact of ¢sheries. ICES Journal of Marine Science 57, ance approach. In: Fisheries Impacts on North Atlantic 697^706. Ecosystems: Models and Analyses (eds S. Gue¨ nette, V. Pauly,D. and Palomares, M.L.D. (2001)Fishing down marine Christensen and D. Pauly). Fisheries Centre Research food web: it is far more pervasive than we thought. In: Reports 9 (4),35^98. Proceedings of the Conference on Sustainability of Fisheries, Martell, S.J.D., Beattie, A.I.,Walters, C.J., Nayar,T. and Briese, Rosenstiel School of Marine Sciences, University of Miami, R. (2002) Simulating ¢sheries management strategies in November,26^28, 2001. the Strait of Georgia ecosystem using Ecopath and Eco- Pauly, D. and Pitcher,T.J. (2000) Assessment and mitigation sim. In:The Use of Ecosystem Models to Investigate Multi- of ¢sheries impacts on marine ecosystems: a multidisci- species Management Strategies for Capture Fisheries (eds T. plinary approach for basin-scale inferences, applied to Pitcher and K. Cochrane). Fisheries Centre Research the North Atlantic. In: Methods for Assessing the Impact Reports 10 (2), 16^23. of Fisheries on Marine Ecosystems of the North Atlantic Mendy, A.N. and Buchary, E.A. (2001) Constructing an Ice- (eds D. Pauly and T.J. Pitcher). Fisheries Centre Research landic marine ecosystem model for 1997 using a mass Reports 8 (2),1^12. balance modeling approach. In: Fisheries Impacts on Pedersen, S.A. and Zeller, D. (2001) Multispecies interac- North AtlanticEcosystems: Models and Analyses (eds S. tions in the West Greenland marine ecosystem: impor- Gue¨ nette,V. Christensen and D. Pauly). Fisheries Centre tance of the shrimp ¢sheries. In: Fisheries Impacts on Research Reports 9 (4),182^197. North AtlanticEcosystems: Models and Analyses (eds

# 2003 Blackwell Publishing Ltd, F I S H and F I S H E R I E S, 4,1^24 23 Biomass of North Atlantic predatory ¢shes VChristensenet al.

S. Gue¨ nette,V.Christensenand D. Pauly). Fisheries Centre Walters, C., Pauly, D. and Christensen, V. (1999) Ecospace: Research Reports 9 (4),111^127. prediction of mesoscale spatial patterns in trophic rela- Pinnegar, J.K., Polunin, N.V.C., Francour, P., Badalamenti, F., tionships of exploited ecosystems, with emphasis on Chemello, R., Harmelin-Vivien, M.L., Hereu, B. et al. the impacts of marine protected areas. Ecosystems 2, (2000) Trophic cascades in benthic marine ecosystems: 539^554. lessons for ¢sheries and protected-area management. Walters, C., Pauly,D., Christensen,V.and Kitchell, J.F.(2000) Environmental Conservation 27,179^200. Representing density dependent consequences of life Sahrhage, D. and Lundbeck, J.A. (1992) A . history strategies in aquatic ecosystems: Ecosim II. SpringerVerlag, Berlin. Ecosystems 3,70^83. Sokal,R.R.andRohlf,F.J.(1995)Biometry: the Principles and Watson, R., Gelchu, A. and Pauly, D. (2001) Mapping ¢sh- Practice of Statistics in Biological Research,3rdedn.W.H. eries landings with emphasis on the North Atlantic. In: Freeman Company,NewYork. Fisheries Impact on North Atlantic Marine Ecosystems: Stanford, R., Lunn, K. and Gue¨ nette, S. (2001)A preliminary Catch, E¡ort and National and Regional Data Sets (eds D. ecosystem model for the Atlantic coast of Morocco in the Zeller, R.Watson and D. Pauly). Fisheries Centre Research mid-1980s. In: Fisheries Impacts on North AtlanticEcosys- Reports 9 (3),1^11. tems:ModelsandAnalyses (eds S. Gue¨ nette,V.Christensen Watson, R. and Pauly, D. (2001) Systematic distortions in and D. Pauly). Fisheries Centre Research Reports 9 (4), world ¢sheries catch trends. Nature 414, 534^536. 314^344. Zeller, D. and Freire, K. (2001) A preliminary North-east Walters, C., Christensen,V.and Pauly, D. (1997) Structuring Atlantic marine ecosystem model: Faroe Islands and dynamic models of exploited ecosystems from trophic ICES area Vb. In: Fisheries Impacts on North Atlantic mass-balance assessments. Reviews in Fish Biology and Ecosystems: Models and Analyses (eds S. Gue¨ nette, Fisheries 7, 139^172. V. Christensen and D. Pauly). Fisheries Centre Research Walters, C. and Kitchell, J.F. (2001) Cultivation/depensation Reports 9 (4), 207^212. e¡ects on juvenile survival and recruitment: implica- Zeller, D. and Reinert, J. (2001) Evaluating spatial closures tions for the theory of ¢shing. Canadian Journal of and e¡ort restrictions inthe Faroe Islands marine ecosys- Fisheries and AquaticSciences 58,39^50. tem. Ecological Modelling, in press.

24 # 2003 Blackwell Publishing Ltd, F I S H and F I S H E R I E S, 4,1^24