Long-term changes and recurrent patterns in fisheries landings from Large Marine Ecosystems (1950–2004)

(1)

and education use, including for instruction at the authors institution

and sharing with colleagues.

Other uses, including reproduction and distribution, or selling or

licensing copies, or posting to personal, institutional or third party

websites are prohibited.

In most cases authors are permitted to post their version of the

article (e.g. in Word or Tex form) to their personal website or

institutional repository. Authors requiring further information

regarding Elsevier’s archiving and manuscript policies are

encouraged to visit:

(2)

ContentslistsavailableatSciVerseScienceDirect

Fisheries

Research

j o ur na l h o me p a g e :w w w . e l s e v i e r . c o m / l o c a t e / f i s h r e s

Long-term

changes

and

recurrent

patterns

in

ﬁsheries

landings

from

Large

Marine

Ecosystems

(1950–2004)

Lorenza

Conti

a,b,∗

_,

_Gaël

_Grenouillet

b

_,

_Sovan

_Lek

b

_,

_Michele

_Scardi

c

a_Department_of_Ecology_and_Sustainable_Economic_{Developement,}_University_of_Tuscia,_Largo_{dell’Università,}_01100,_Viterbo,_Italy b_Laboratoire_Evolution_&_Diversité_Biologique,_Université_Paul_Sabatier,₁₁₈_route_de_Narbonne,_31062,_Toulouse_cédex_9,_France c_Department_of_Biology,_Tor_Vergata_University_of_Rome,_Via_della_Ricerca_{Scientiﬁca,}_00133,_Roma,_Italy

a

r

t

i

c

l

e

i

n

f

o

Articlehistory: Received3March2011

Receivedinrevisedform1December2011 Accepted2December2011

Keywords:

LargeMarineEcosystems Functionalgroups SelfOrganizingMaps Timeseries

a

b

s

t

r

a

c

t

TheregionaldynamicsofindustrialfisherieswithinLargeMarineEcosystems(LMEs)boundarieswere investigatedbymeansofahistorical-descriptiveapproach.LandingsdatafromtheSeaAroundUsProject databasewereusedtodetecttrendsintotalyieldsandvariationsinlandingscompositionbyfunctional groupsovertime.Thetemporalandspatialscalescoveredbythisstudyallowedgeneralissuestobe addressedsuchasthedetectionofrecurrentpatternsandsynchroniesinfisherieslandings.An unsuper-visedartificialneuralnetwork,SelfOrganizingMap(SOM),isusedasatooltoanalyzefisherieslandings compositionvariationoverfivedecadesin51LMEsallovertheworld.Fromthehistoricalanalysisof “fishingbehaviors”withinLMEsabroaddistinctionbetweentwomaintypesoffisheriesemerged:(1) smallandmediumpelagicsfisheries,withstablecompositionsorcyclicbehaviors,occurredinLMEs whichsharecommonproductivefeatures,despitedifferentgeographicallocationsand(2)demersal fisheries,whicharemoreaffectedbyeconomicdriversandtendtoconcentrateinLMEsintheNorthern Hemisphere.Ouranalysiscanberegardedasafirststeptowardsthechallengingscopeofdescribingthe relativeinfluenceofenvironmentalandeconomicdriversonexploitedecosystems.

1. Introduction

The regional dynamics of industrial ﬁsheries within Large Marine Ecosystems (LMEs) were investigated by means of a historical-descriptiveapproach.Thisapproachisparticularly effec-tivewhenaddressingecologicalissues,inparticularinthedomain ofﬁsheriesoceanography(FrancisandHare,1994),where reduc-tionismandexperimental–predictivemethodscouldbeineffective indealingwithuncertaintyandcomplexinterplays.

While historical time series of industrial ﬁsheries landings are available, few reviews have been published up until now

(Christensen et al., 2009; FAO, 2010; Garibaldi and Limongelli,

2003),andpublishedpapersfocusedmainlyonsinglespeciesand selected LMEs, often from North Atlantic or North Paciﬁc (e.g.

Drinkwater,2009;Rose,2005).In particular,multi-decadal

eco-logical time series have beenlargely used for thedetection of gradualor abruptchanges in ecosystems,suchas regimeshifts

(seeOverlandetal.,2008for“regimeshifts”deﬁnitions),andfor

theanalysisofteleconnections(Stein,1998),i.e.co-variationsand synchroniesofsinglespeciesindifferenthemispheres(e.g.Alheit

etal.,2005;Bakun,1998;Fréonetal.,2003;Lluch-Beldaetal.,1992;

∗ Correspondingauthor.Tel.:+390561556735;fax:+390662275147. E-mailaddress:lorenza.conti@cict.fr(L.Conti).

Schwartzloseetal.,1999).Typically,theseassociationsaredriven

bycyclicorabruptevents,asENSO(i.e.ElNi ñoSouthernOscillation) andregimeshiftsthemselves,whoseeffectsspreadfarbeyondlocal influencethroughpoorlyknownlinks.Shiftsinclimateregimescan rearrangemarinecommunitiesandtropho-dynamicrelationships andinducechangesintheproportionsofdominatingspeciesover decadaltimescales(Alheitetal.,2005).Recentfindingssuggest thatoverexploitation,andnotonlyclimateregimeshifts,can pro-motesuchlong-termchangesinmarineecosystems(Curyetal.,

2008;PaulyandMaclean,2003).Fisheries-inducedregimeshifts

involvenotonlythespecies-level,butalsoentirefunctionalgroups.

Savenkoffetal.(2007)demonstratedthattheGulfofSt.Lawrence

ecosystemshiftedfromamixedpiscivorousgroundﬁshand small-bodiedforagespeciesstructuretoadominanceoflowtrophiclevel pelagicspecies,asaconsequenceofremovalbyﬁshingof large-bodieddemersalpredators.Othershiftsfromdemersal-dominated topelagic-dominatedecosystemshavebeendocumentedinthe AtlanticOceanandtheBalticSea(Bundy,2005;Franketal.,2005;

WormandMyers,2003).

In this study,landings datafromthe SeaAroundUs Project

database(SAUP;availableon-lineatwww.searoundus.org)from

1950to2004wereusedtodetecttrendsintotalyieldsand vari-ations in landingscomposition by functional groups over time. The temporal and spatial scalescovered by this studyallowed generalissuestobeaddressedsuchasthedetectionofrecurrent 0165-7836/$–seefrontmatter © 2011 Elsevier B.V. All rights reserved.

(3)

Fig.1. Mapoftheworld’sLargeMarineEcosystems.ForLMEslegendanddescriptionseeTable1.(Source:LargeMarineEcosystemsoftheWorld,http://www.lme.noaa.gov/). Modiﬁed.

patterns and synchronies in fisheries landings. These kinds of ordered responses represent the result of change in economic conditions,resourceexploitation andfishingpressures interact-ingwithenvironmentaldynamicsandclimatechangeovermore thanfiftyyears.Inthiscontext,emergingpatternscouldrepresent afirststeptowardsabettercomprehensionofcomplexinterplays andsynergiesbetweenecosystemsandmanagement.

Inordertocopewiththecomplexityofecologicaldatasets, pow-erfulandﬂexibletoolssuchasartiﬁcialneuralnetworkscouldplay akeyrole,bothindescriptiveandpredictiveanalysis,thus pro-vidingsyntheticandinformativeinsightsintolargescaledynamics (e.g.Almeida,2002;Laëetal.,1999;LekandBaran,1997;Lekand

Guégan,1999;Leketal.,1996).Inthisstudy,theSelfOrganizing

Map(SOM),anunsupervisedartificialneuralnetwork,isproposed asatooltoanalyzethevariationsinfisherieslandingscomposition overfivedecadesin51LMEsallovertheworld.

2. Materialsandmethods

2.1. Dataset

Fifty-fiveyears(1950–2004)ofreportedfisherieslandingsfrom theworld’sLMEswereextractedfromtheSAUPdatabase.Fifty-one LMEswereselectedfortheanalysis,withtheexceptionofLMEs fromPolarOceans,whichpresentedscarceandlowdifferentiated landings(Fig.1,Table1).LMEsaredefinedashomogeneousregions ofoceanandcoastalspacethatencompassriverbasinsand estuar-iesandextendouttotheseawardboundaryofcontinentalshelves andtheseawardmarginsof coastalcurrentsystems,which are delineatedaccordingtocontinuitiesintheirphysicalandbiological characteristics(ShermanandDuda,1999).

TimeserieswererepresentedbyannualLME-speciﬁclandings compositionofﬁsheriesharvestsbyfunctionalgroups,asreported

bySAUP(Table2).Functionalgroupswerechosenascloser

descrip-torsoffisheriesdynamicsatlargerspatialscaleswithrespecttoa finertaxa resolution(e.g.singletargetspecies)asalready men-tionedinotherstudies(Franketal.,2006;Hughesetal.,2005).In otherwords,thedatasetwascomposedby2805records(i.e.51 LMEstimes55years),eachrepresentingatypicalcatchprofileor “fisherybehavior”inspaceandtime.

2.2. Trendanalysisofannualtotalﬁsherieslandings

Spearman’s rank correlation was used to detect monotonic trends in time series of landings. In order to capture trends

discontinuities,ineachtimeserieswelookedforthemost signif-icantturningpoint,i.e.theoptimalsubdivisionofthetimeseries forwhichtheweightedrankr2

s [rs(w)2 =(n1r21+n2r22)/n]was

maxi-mized.Insomecasesthiscorrespondedtoasinglemonotonicseries (55-yearslong),inotherstwosub-serieswerefound.Wedidnot takeintoaccounttrendsshorterthan6years.

2.3. Landingscompositionanalysis

In ordertoremovetheeffectof thetrend inlandings abun-dances,therelativecontributionofeachfunctionalgrouptototal yieldswasusedasadescriptorofﬁsheriesharvests.Inparticular, theproportionsof13among28functionalgroups(i.e. character-ized by>1% ofaverage contributionacrossLMEs) and a mixed category(i.e.totalcontributionoftheother14functionalgroups reportedintheSAUPdatabase,withanaveragecontribution<1%)

(Table2)wereusedtotrainaSelfOrganizingMap(SOM).

TheSOMsareaclassofneuralnetworksbasedoncompetitive unsupervisedlearning.Their“neurons”areplacedatthenodesofa squareorhexagonallatticethatisusuallytwo-dimensional (lower-orhigher-dimensionalSOMsareseldomused).Astheycompete witheachothertobeactivated,onlyoneofthemcanbethe win-neratanyonetime.Theneuronsbecomeselectivelytunedtoa setofinputpatternsduringthecompetitivelearningandthe loca-tionsofthetunedneuronsbecomeorderedrelativetoeachotherin suchawaythatameaningfulcoordinatesystememergesfromtheir arrangement(Kohonen,1990).ASOMcanthereforeberegardedas amapoftheinputpatternsinwhichthecoordinatesoftheneurons inthelatticearerelatedtotheirfeaturesandsimilarpatternswill bemappedontoneighboringneurons(Fig.2).

Theessentialelementsandparametersofthealgorithmare:(1) acontinuousinputspaceofactivationpatternsthataregenerated inaccordancewithacertainprobabilitydistribution;(2)atopology ofthenetworkintheformofalatticeofneurons,whichdeﬁnesa discreteoutputspace;(3)atime-varyingneighborhoodfunction h(t)thatisdeﬁnedaroundawinningneuronandshrinksduring thelearningphase,and(4)alearningrate(t,r)thatstartsatan initialvalue0andthendecreasesgraduallywithtimet,butnever

goesexactlytozero.

Therearethreebasicstepsinvolvedintheapplicationofthe algorithmafterinitialization:sampling,similaritymatching,and updating. Thesethree stepsarerepeateduntilformation ofthe featuremaphasbeencompleted.Thealgorithmissummarizedas follows:(1)initialization,i.e.chooserandomvaluesfortheinitial weightvectorswi(0).Theonlyrestrictionhereisthatthewi(0)be

(4)

Table1

The51LargeMarineEcosystems(LMEs)analyzedinthisstudy.LMEsidentiﬁcativenumber,acronyms,names,surfaceareas,coordinatesandoceanicbasinarereported.

LMENr. Acronym LMEname Area(km2₎ _Lat_N _Long_E _Ocean

1 EBS EastBeringSea 1,397,933 57.3 −167.5 Paciﬁc

2 GAL GulfofAlaska 1,460,365 54.3 −139.9 Paciﬁc

3 CAL CaliforniaCurrent 2,273,942 34.9 −120.4 Paciﬁc

4 GCA GulfofCalifornia 222,713 33.4 −110.4 Paciﬁc

5 GME GulfofMexico 1,549,163 30.2 −92.9 Atlantic

6 SUS SoutheastUSContinentalShelf 319,775 33 −81.8 Atlantic

7 NUS NortheastUSContinentalShelf 308,656 48.2 −75.8 Atlantic

8 SCS ScotianShelf 300,258 45.6 −62.1 Atlantic

9 NFL Newfoundland-LabradorShelf 898,803 51.5 −60.6 Atlantic

10 IPH InsularPaciﬁc/Hawaiian 982,811 23.3 −166.6 Paciﬁc

11 PCA PaciﬁcCentral-AmericanCoastal 1,990,321 9.1 −90.5 Paciﬁc

12 CAR CaribbeanSea 3,285,047 12.9 −75.2 Atlantic

13 HUM HumboldtCurrent 2,529,645 −29.1 −71 Paciﬁc

14 PAT PatagonianShelf 1,167,969 −37.6 −61.5 Atlantic

15 SBR SouthBrazilShelf 567,996 −22.5 −48.6 Atlantic

16 EBR EastBrazilShelf 1,079,113 −11.3 −45.6 Atlantic

17 NBR NorthBrazilShelf 1,058,516 1.3 −53 Atlantic

22 NOR NorthSea 698,130 54.6 10.7 Atlantic

23 BAL BalticSea 389,252 59.6 21.1 Atlantic

24 CBS Celtic-BiscayShelf 769,121 51.1 −5.1 Atlantic

25 IBE IberianCoastal 303,958 40.4 −6.1 Atlantic

26 MED MediterraneanSea 2,561,659 36.4 17.7 Atlantic

27 CAN CanaryCurrent 1,141,648 23.9 −1.3 Atlantic

28 GUI GuineaCurrent 1,927,373 4.5 3.8 Atlantic

29 BEN BenguelaCurrent 1,436,847 −20.9 17.8 Atlantic

30 AGU AgulhasCurrent 2,632,932 −22.1 34.9 Indian

31 SCC SomaliCoastalCurrent 843,937 0.6 38.7 Indian

32 ARA ArabianSea 3,945,355 28.4 51.7 Indian

33 RED RedSea 462,210 18.5 31.9 Indian

34 BBE BayofBengal 3,679,296 25 90.1 Indian

35 THA GulfofThailand 395,780 8.4 102.2 Paciﬁc

36 SCH SouthChinaSea 3,183,503 17.2 105.5 Paciﬁc

37 SUL Sulu-CelebesSea 1,017,861 7.8 121.4 Paciﬁc

38 IND IndonesianSea 2,275,957 −3.9 119.9 Paciﬁc

39 NAS NorthAustralianShelf 782,956 −17.8 133.8 Paciﬁc

40 NEA NortheastAustralianShelf 1,284,441 −18 149.8 Paciﬁc

41 ECA EastCentralAustralianShelf 654,158 −28.6 149.4 Paciﬁc

42 SEA SoutheastAustralianShelf 1,192,306 −40.5 143.2 Paciﬁc

43 SWA SouthwestAustralianShelf 1,052,046 −31.6 126 Indian

44 WCA WestCentralAustralianShelf 545,539 −26.9 118.6 Indian

45 NWA NorthwestAustralianShelf 915,060 −18 118.9 Indian

46 NZS NewZealandShelf 967,616 −40.7 172.8 Paciﬁc

47 ECH EastChinaSea 780,554 37.4 105.3 Paciﬁc

48 YEL YellowSea 440,387 41.7 110.1 Paciﬁc

49 KUR KuroshioCurrent 1,322,524 32.4 133.5 Paciﬁc

50 SJA SeaofJapan 997,858 43.6 134 Paciﬁc

51 OYA OyashioCurrent 532,831 46 150.4 Paciﬁc

52 OKH OkhotskSea 1,570,523 54.5 146.4 Paciﬁc

53 WBS WestBeringSea 2,170,639 58.2 174.4 Paciﬁc

60 FAR FaroePlateau 150,558 60.4 −11.5 Atlantic

62 BLA BlackSea 461,958 43.8 39.8 Atlantic

differentfori=1,2,...,n,wherenisthenumberofneuronsinthe lattice.Itmaybedesirableinitiallyusesmallmagnitudesforthe weights.Anotherwayofinitializingthealgorithmistorandomly selecttheweightvectorsfromtheavailablesetofinputvectors;(2) sampling,i.e.drawasamplexfromtheinputspacewithacertain probability.Thevectorxrepresentstheactivationpatternthatis appliedtothelattice.Thevectorxdimensionisequaltothe dimen-sionoftheweightvectorswi;(3)similarityMatching,i.e.ﬁndthe

best-matching(winning)neuron(BestMatchingUnit,BMU)attime steptbyusingaminimum-distance(usuallyEuclidean)criterion; (4)updating,i.e.adjustthesynapticweightvectorsofallneuronsby usingtheupdateformulawi(t+1)=wi(t)+(t,h(t))[x(t)−wi(t)],

here(t,h(t))isthelearningrateparameter,andh(t)isthe neigh-borhoodfunctioncenteredaroundthewinningneuron;both(t, h(t))andh(t)arevarieddynamicallyduringthelearningphase;and (5)continuation,i.e.gobacktostep2untilnonoticeablechanges inthefeaturemapareobserved.

Thetraining procedurecan becarried out oncemore, start-ingwithasmallerlearningrateandsmaller(orevennull)radius

ofthetrainingneighborhood.Thissecondtrainingphaseis usu-allyreferredtoastheﬁnetuningphase.MoredetailsaboutSOM trainingcanbeobtainedfromseveraldifferentsources(e.g.

Hecht-Nielsen, 1990; Kohonen, 1982, 1995; Lippmann, 1987; Zurada,

1992).TheSOMPAKpackage,a classicimplementationofSOMs, wasdevelopedbyTeuvoKohonenandhisco-workersandisnow available fromtheweb site

http://www.cis.hut.ﬁ/research/som-research/nnrc-programs.shtml. SOMs have been successfully

appliedtoabroadspectrumofproblemsincludingapplicationsto ecology,becausetheyarenotsensitivetosomeoftheproblemsthat oftenaffectecologicaldata,hinderingmoreconventionalmethods (non-linearrelationshipsbetweenvariables,non-normal distribu-tionsofdata,etc.).Theycanbeviewedasanon-linearextension ofPrincipalComponentAnalysis(Ritteretal.,1992),orasahybrid betweenclusteringandordinationtechniques.

BymeansofaSOM,the2805records(i.e.LME/year)were clas-siﬁedintoanoutputmapof63 units(i.e.7×9hexagons),each representingavirtuallandingscompositionproﬁle,towhicha sub-setofobservationswasassigned.TheSOMdimensionwaschosenin

(5)

Fig.2. StructureofaSOM:theinputlayer(X)isconnectedtothefeaturemap(Y) andeachconnectionisassociatedwithaweight(w).FromFausett(1994).

ordertoavoidemptyunitsandtopreserveclearpatternsinthe dis-tributionsofvariables.SOMunitswererandomlyinitializedbefore theﬁrsttrainingphase. Weightsobtainedfromthisphasewere thenusedtoinitializethesecond,orﬁnetuning,phase.

TheoverallevolutionofeachLMEfisheryoverthetimerange wascapturedbyLME-specifictemporaltracksdrawnonthemap, i.e.brokenlinesconnectingthecellsinwhich1950–2004 obser-vationsforeachLMEfell.Inordertopointoutrecurrentpatterns, ahierarchicalclassification(UPGMAalgorithm)oftheseLME tra-jectorieswasperformed,basedonthehexagondistancebetween therelativepositionsoftheLMEobservationsforthesameyear. The“elbow(orknee)method”wasusedtodeterminetheoptimal partition,whichwascomprisedof5clusters.

2.4. Mantel’stest

Mantel’s tests wereperformed between:(1) distance matri-cesderivedfromlandingsabundances(Bray–Curtisdistance)or landingsproportions(Euclideandistance)andgeographic(and lat-itudinal)distancebetweenLMEs’centroids.TheMantelstatistics timeserieswereanalyzedbylinearcorrelationtodetect signiﬁ-canttrends.GeographicandlatitudedistancesbetweenLMEswere computedbymeansoftherdist.earthfunctioninR.

3. Resultsanddiscussion

3.1. Trendanalysisofannualtotalﬁsherieslandings

Monotonictrendsweredetectedinannualtotalfisheries land-ingstimeseriesbySpearman’srankcorrelation(Fig.3).AllLMEs showasignificantandpositivetrendfortheentireseriesorinthe firstsub-series,withtheonlyexceptionbeingtheIberianCoastal ecosystem(#25)whichshowednosignificanttrendupuntil1994, followedbyasteepdecreaseintotallandings.

Fifteen LMEs werebestﬁtted bya single55-year long posi-tivetrend(n1=55inFig.3),wheretotalabundancesexhibiteda

significantmonotonicincreaseovertime.TheseLMEsaremostly equatorialorsubequatorialecosystems[e.g.PacificCentral Amer-icanCoastal (#11), CaribbeanSea (#12),Guinea Current(#28), SomaliCoastal Current(#31),Arabian Sea(#33), Bayof Bengal (#34),Sulu-CelebesSea(#37),NorthAustralianShelf(#39), North-westAustralianShelf(#45)].TemperateLMEsshowingapositive trendarealllocatedinthesouthern [e.g.Southeastand South-west Australian Shelves (#42 and 43) and New Zealand Shelf (#46)]and boreal Pacific Ocean [e.g.East ChinaSea (#47) and

Table2

The28functionalgroupsusedfortheanalysis,modiﬁedfromSeaAroundUsProject (www.searoundus.org).The13functionalgroups(i.e.characterizedby>1%of aver-agecontributionacrossLargeMarineEcosystems)retainedfortheSelfOrganizing Maptrainingareshowninbolditalic.Theotherfunctionalgroupsarerepresented byamixedcategory(i.e.,“Other”).

Functionalgroup Size(Lmax)

Vertebrates Bathydemersals Large(>90cm) Medium(30–89cm) Small<30cm Bathypelagics Large(>90cm) Medium(3089cm) Small<30cm Benthopelagics Large(>90cm) Medium(30–89cm) Small<30cm Demersals Large(>90cm) Medium(30–89cm) Small<30cm Flatﬁshes Large(>90cm) Smalltomedium(<90cm) Pelagics Large(>90cm) Medium(30–89cm) Small<30cm

ReefAssociatedFishes Large(>90cm)

Medium(30–89cm) Small<30cm Rays Large(>90cm) Smalltomedium(<90cm) Sharks Large(>90cm) Smalltomedium(<90cm) Invertebrates Cephalopods Crustaceans(nectonic) Crustaceans(benthonic)

OtherdemersalInvertebrates

YellowSea(#48)],withtheexceptionoftheFaeroeShelf(#60). Thiswidespreadincreasein totalfisheries yieldsrepresentsthe direct effectof thecorrespondingincrease in fishingeffortand efficiencydrivenbygrowingmarketdemandanddeterminedby theavailabilityofimprovedtechnologiesfornavigationand fish-ing activitiesfrom 1950onwards (Mullon et al., 2005). Fishing vessels becameable toreachmoredistant fishinggroundsand displayedincreasingcapacityforharvestingmarineresources.The reportedlandingsincreasewaslargestinthe1960s,whenthe tra-ditional fishinggroundsoftheNorthAtlanticand NorthPacific became fullyexploitedand newfisheries openedat lower lati-tudesandintheSouthernHemisphere(WatsonandPauly,2001),as demonstratedbythetropicalandsub-tropicalecosystems.Similar conclusions may alsoconcern those LMEs which showed posi-tiveand significanttrendsin landings abundances followed by shortandnon-significantsub-series,encompassingonlythelast 6–7yearsoftheseries[e.g.GulfofAlaska(#2),CanaryCurrent (#27),Northeastand WestCentral AustralianShelves (#40and 44)]andtheIndonesianSea(#38)characterizedbytwopositive series,thefirstencompassingonly7years(1950–1956). Regard-ingtheincreaseobservedfortheEastChinaSeaandYellowSea,it hasbeendemonstratedthatthismayarisefromsystematic distor-tionsinlandingsreportingbytheChinesegovernmentratherthan fromanactualincreaseinharvestedresources(WatsonandPauly,

2001).

Regardingthetwo-phaselandingsseries,adistinctionbetween trendbreaksandtrendinversionswasobserved,thelatterbeing characterizedbyanoppositesignofrankcorrelation(rs)forthe

two sub-series. Trend breaks, i.e. two positive signiﬁcant sub-series,wereobservedinelevenLMEs,althoughinmostcasesthe decrease in landingsabundancesat thebreak point wasfeeble enoughtoallowregroupingtheseLMEswiththosecharacterizedby asinglemonotonouspositivetrend(i.e.,trapezium-shapedsecond

(6)

Fig.3.Trendsinlandingsabundancesinthe51LMEsfrom1950to2004.MonotonictrendsweredetectedbySpearman’srankcorrelation(rs).Fromtheleft:LMEIDnumber,

numberofyearsintheﬁrstsub-series(n1),Spearman’srfortheﬁrstsub-series(r1),representationofthetrend,numberofyearsinthesecondsub-series(n2),Spearman’sr

forthesecondsub-series(r2).Significanttrendsarefilledinshadedgrey.Whiteshapesrepresentnon-significanttrends(forLMElegendanddescriptionsseeTable1).

sub-seriesase.g.LME#3or#16inFig.3).Incontrast,otherLMEs seriesshowedasharperbreakpoint,wheredrasticreductionsin landingsabundancesdraggedtheseriesdowntothelowestvalues comparablewiththosefromthe1950s(i.e.,triangle-shaped sec-ondsub-seriesinFig.3).Thispatternisoftenassociatedwiththose LMEswheresmallpelagicsarethemostabundantcomponentof landings[e.g.HumboldtCurrent(LME #13)and BlackSea(LME #62)].Naturalandoftensharposcillationsofthesetargetspecies controltotalﬁsheriesyields.

Finally,trend inversions werefoundin eighteenLMEseries. Mostofthesepatternsweregeographicallyclusteredinthe North-eastAtlanticOceanandintheNortheastPacificOcean.Alongthe Atlanticcoast,fromtheGulfofMexico(#5)goingnorthuntilthe Scotian Shelf (#9), a significant decrease in totallandings was observedineachLME,withaprogressivelyearlierturningpoint fromsouthtonorth(e.g.,from1981intheGulfofMexicoto1962 intheScotianShelf).Asfor thePacificOcean,allLMEslandings seriesfromtheAsiancoastsdroppedinthe1980s(e.g.,LMEs#49to #53).Thesepatternssupportthewidelyrecognizedgloballandings declinesbegan in thelate 1980s, when theunderlying ecosys-temsinthemajorfishinggroundscollapsedunderunsustainable exploitationpressures(Christensenetal.,2003;Jacksonetal.,2001;

MyersandWorm,2003;Paulyetal.,2005).

3.2. StabilityversusvariabilityinLMElandings

SigniﬁcanttrendsinLMElandingstimeseriesimplystrong auto-correlation.Therefore,focusingondifferencesbetweensubsequent yearsmayrevealinterestingpropertiesofthetimeseries.InFig.4, thedistributionofEuclideandistancesbetweenlandings composi-tionbyfunctionalgroupsinsubsequentyearsisshownbymeans ofboxandwhiskersplots.Distanceswerecomputedonrawdata (leftpanel)andonproportions(rightpanel).

Themainfeaturefromthisanalysiswasthedistinctionbetween “stable”(i.e.,narrowrangeofdistances)and“variable”(i.e.,wider rangeofdistances)ecosystems.Itisimportanttostressthatinthis

particularcomparison,“stability”and“variability”donotreferto thewholetimeseries,buttochangesbetweensubsequentyears. ThedistinctionbetweenstableandvariableLMEswasparticularly evidentwhenrawdatawereused(Fig.4,left),whilethe differ-encesweremuchmoredampenedwhenproportionsoffunctional groupsweretakenintoaccount(Fig.4,right).Thevariability associ-atedwithchangesinharvestabundanceswasspreadalongawide scale,wherefewLMEswerecharacterizedbyhighandunstable ﬁsherieslandings[e.g.,NortheastUSContinentalShelf(#7), Hum-boldtCurrent(#13), NorthSea(#22),EastChinaSea(#47) and YellowSea(#48)],whileanumberofmorestableLMEsshowed modestorlimitedvariabilityofharvests[e.g.,Australianshelves (#39–45)].Whenproportionsbyfunctionalgroupwereused,the rangeofvariationbetweensubsequentyearsnarrowed,and vari-abilitybecamesimilarbetweenLMEs(i.e.,inmostcasesEuclidean distanceswere<0.2).

Comparingthetwoboxandwhiskersplots,mostoftheLMEs withmorestableandlowerlandings(narrowboxesinleftpanel,

Fig.4)seemedtobeassociatedwithrelativelyvariablecomposition byfunctionalgroups(widerboxesinrightpanel,Fig.4)[e.g.,Insular Pacific/Hawaiian(#10)andAustralianshelves(#39–45)].Inother words,lowlandingsseemedtobemorevariableintheir compo-sitioncomparedwithmoreabundantlandings,inwhichfisheries tendedtotargetspecificfunctionalgroups.Anexceptiontothis featurewastheHumboldtCurrent(#13),wherethedramatic fluc-tuationsofthePeruviananchoveta(Engraulisringens),drivenby ENSOoscillations,leadtoamarkedvariabilityinlandings compo-sitionbetweenENSOandnon-ENSOyears(widestboxesinboth panels,Fig.4).

In contrast, the Mediterranean Sea (#26) stands out for its stability,bothwhenrawdataandproportionswereused,a fea-ture that hasbeen described in previousstudies. Althoughthe MediterraneanSeapresentsoneofthemostdiversecompositions inﬁsherieslandings,itisalsoamongthemoststableecosystems intermsofthecontributionsoffunctionalgroupstototallandings

(7)

Fig.4. BoxandwhiskerplotsshowingthedistributionofEuclideandistancesbetweensubsequentyearsoftotallandingsabundancesintonkm−2year−1(left)andoflandings compositionbyfunctionalgroupsastheirpercentagecontributiontototallandings(right)(seeTable1forLMElegend).

3.3. Landingscompositionanalysis 3.3.1. SelfOrganizingMap(SOM)

Aslandingstimeserieswerebiasedbythewidespreadincrease intotalyieldsfrom1950to2004,changeinfisherieslandings com-positionwasbetterdescribedby variationin functionalgroups’ proportions rather than by variation in abundances. The SOM trainedontherelativecontributionof14functionalgroupstototal annuallandingsineachLMEis showninFig.5.The numberof observationsassignedtoeachBMUrangedfrom7(unit3,4,i.e., unitinlinexandcolumny)to108(unit5,2).Theabsenceofempty unitsshowedthateachBMUrepresentsarealharvestcomposition, whichwasthenassociatedwithanumberofrealobservations.No “virtual”profileswherethusrepresentedonthemap.Eachunit intheSOMrepresentsatypicalcombinationofthe14variables (functionalgroupandsizeclass;Table2),whichcanbevisualized asahistogramrepresentingtherelativecontributionofeach func-tionalgrouptototallandings(Fig.6).Theleastrepresentedprofile (seeunit3,4)wasassociatedwiththeHumboldtCurrentlandings from1950to1956.Thehistogramshowedamixedcomposition oflargeandsmallpelagics,wherethemixedcategory(i.e.,“Other”) wasalsowellrepresented.Themostfrequentprofile(seeunit5,2) wasassociatedwiththetimeseriesfromtheGuineaCurrentand Sulu-CelebesSea,togetherwithlandingsfromtheCanaryCurrent (1970–1971),EastChinaSea(1987–1989)andKuroshioCurrent (roughlyfromtheendofthe1970suntil1993).ThisBMU repre-sentedacatchprofiledominatedbymediumpelagiclandings.

ThestructureoftheSOMdependsonthepartitioningofthe con-tinuousvariationoftheinputvariablesintheoriginaldataset;the

resultingdiscretesetofvariationforeachfunctionalgroupisshown inFig.7.Thecomparisonofthesepatternsallowsan understand-ingoftherelativecontributionofeachvariabletotheSOM,and providesastraightforwardwaytoidentifythosevariableswhich arelessdeterminant.Regardingthesemaps,adistinctionbetween regularandirregularpatternsofdistributionofthedifferent func-tionalgroupswasobserved.Functionalgroupsshowedhighestand lowestvaluesoccurringinunitsthatwerelocatedinwell-defined SOMregions(e.g.,largebenthopelagicsintheupperrightcorner andsmallpelagicsintheupperleftcorner),whileothergroups showedamoreirregularpatterninwhichextremevaluesdidnot occurinSOMunitsthatwereclosetoeachother(e.g.,cephalopods, crustaceansandsmallflatfishes).ThecomplexstructureoftheSOM wasinfluencedbyregularandstrongpatterns,asthoserepresented bylargebenthopelagics,mediumpelagicsandsmallpelagics.The modulationoftherelativedominanceofthesefunctionalgroupsin totallandingsinfluencedtheLMEdisplacementsontothemap,i.e. thetemporalevolutionoflandingscompositionsineachLMEfrom 1950to2004,representedasabrokenlineontheSOM.Theother groupsexertedalocal(inspaceortime)influenceastheywere relevantonlyinasmallnumberofobservations.Theseless repre-sentedgroupswereoftenassociatedwithmixedlandginsprofiles, whereas themostfishedfunctional groupstended todominate reportedlandings.

3.3.2. Fisherytracks:temporalvariationofLMElandingsprofiles Theevolutionofthefisheries51regionalLMEscanbedescribed byLME-specifictemporaltracksdrawnontheSOM.Anexample of thesetemporal tracksisgiven in Fig.8for theScotianShelf

(8)

Fig.5. SelfOrganizingMap.Thenumberofobservationsassociatedwitheachunitisshown.

ecosystem,wherethetypicaldemersal-dominatedharvest com-positionshiftedtowardslowertrophiclevels(e.g.mediumpelagics andbenthicinvertebrates)intheearly1990s.Thispatternhasbeen describedfor theecosystemoff NovaScotia, inresponse tothe collapseofthebenthicﬁshcommunity(Franketal.,2005).

Fromahierarchicalclassificationofthese51“fisherytracks”, 5clusterswereidentified(Fig.9).Thefrequencyofoccurrenceof eachclusterhasbeenplottedontheSOMtovisualizetheregion ofthemapwhere theLMEtimeseriesclassifiedineachcluster tendedtoconverge(Fig.10).Thedetailforeachclusterisshown

Fig.6.SelfOrganizingMap.Landingscomposition,describedbytherelativeabundanceofthe14functionalgroups,isshownineachSOMunitasahistogram.Fromtheleft, barsrepresentthepercentagecontributionof:largedemersals,mediumdemersals,smalldemersals,smallﬂatﬁshes,largebenthopelagics,mediumbenthopelagics,large pelagics,mediumpelagics,smallpelagics,cephalopods,crustaceans(nektonic),crustaceans(benthonic),demersalinvertebratesandamixedcategory(i.e.,“Other”).

(9)

Fig.7.SelfOrganizingMap.Theweightofthe14inputvariablesisshowningrayscale(whitelow,darkgreyhigh).

inFig.11.Abettercomprehensionofthemainfeaturesdescribed foreachclustercanbeattainedbycomparingthemapsshownin

Figs.6and7,withthoseinFig.11.Adescriptionofthemainfeatures

foreachclusterisgivenbelow.

3.3.2.1. Cluster1–mixedcatches. Thisclusterwasthelargestone, encompassing18LMEs.Theseregionswereprobablytheless char-acterizedwith respectto LMEs classified in other clusters and forthis reasontheyhavebeengroupedtogether.Nonetheless,a number ofLMEs classifiedin this cluster sharedcommon fish-eriesprofilesinwhichlowtrophiclevelspecies,assuchmedium pelagicsanddemersalinvertebratesfunctionalgroups,werewell

represented(compareFig.11 withFig.7,e.g.medium pelagics, demersalinvertebratesandcrustaceans).

3.3.2.2. Cluster2–demersal,benthopelagicsandinvertebrates. This clusterregrouped12LMEsfromNorthernHemisphere,bothfrom thePaciﬁcandtheAtlanticOcean,andthePatagonianShelf.These LMEs showed homogeneous patterns in landings composition variation overtime, where harvestswerebroadlycharacterized by bottomfunctional groups, suchasbenthopelagicsand dem-ersals (compare Fig. 11 with Fig. 7, e.g. large benthopelagics, largedemersals and medium demersals).In particular,thetwo currents’ systems (i.e., the Oyashio Current and the California

Fig.8.SelfOrganizingMap.ThetemporalevolutionofﬁsherieslandingscompositionintheScotianShelfisshownontheSOM,from1950until2004.Landingscomposition intermsoffunctionalgroupsisshownasahistograminsideeachunit.

(10)

Fig.9. Hierarchicalclassiﬁcation(UPGMA)ofSOMunits.Fiveclusterswereidentiﬁedfollowingthe“elbowmethod”shownbesidethedendrogram.

Current) showed the most variable patterns among the LMEs inthiscluster.TheCaliforniaEasternBoundaryCurrent, charac-terizedbyperiodicstrengtheningofcoastalupwellinglinkedto climaticteleconnections(Bakun,1990),showedcyclicoscillations betweenpelagic[Californiananchovy(Engraulismordax)andSouth American pilchard (Sardinops sagax)] and benthopelagic (North Pacifichake, Merlucciusproductus) catches,while landingsfrom theOyashioCurrent,characterizedbycoldsub-polarnutrientrich waters,weredominatedbySouthAmericanpilchardandPacific saury(Cololabissaira)atthebeginningoftheseries(upuntilearly 1960s)andthenturnedintoamixedpelagicandbenthopelagics (i.e.,Alaska pollack,Theragrachalchogramma)fishery withmore dampedoscillationswithrespecttoits eastern Pacific counter-part.AsfortheotherwesternPacificLMEs(i.e.,OkhotskSea,Sea ofJapan),theyalsoexperienceda commontrendfrommedium pelagic-dominatedlandings tolarge benthopelagicslandings.A similarpatternwasobservedfortheFaeroePlateau,wherecatches boosted upat thebeginning of the1970s, when medium ben-thopelagics(i.e.BluewhitingMicromesistiuspotassou)becamethe dominantfractionoftotalharvests.

Cluster2encompasseda“Canadiansub-cluster”,whichwould standaloneifapartitionof8clustershadbeenchosen,inwhich twoNorthwestAtlanticLMEs(i.e.,Newfoundland-LabradorShelf andScotianShelf)showedageneralrestructuringofﬁshery land-ingsfromdemersalandlargebenthopelagicdominatedharveststo

Fig.10. SelfOrganizingMap.Proportionsofobservationsbelongingtothe5clusters areshowningrayscale.Cluster2(white)issplittedintotwoparts,theunit5,3falling intotheregionoftheSOMbelongingtoCluster1.Thisunitrepresentsthe“Canadian sub-cluster”describedinSection3.3.2.2.

mediumpelagicsanddemersalinvertebratescatches(e.g.Scotian ShelfﬁsheryevolutionisshowninFig.8).

These ecosystems experienced an abrupt (and well docu-mented) change in the 1990s, when mixed demersal landings werelargelyreplacedbymediumpelagicsanddemersal inverte-brates. In particular,while theScotianShelf ecosystemshowed mixed demersal and pelagic landingssince the 1970s, and the shifttopelagicandinvertebrateslandingswasobservedin1994, theNewfoundland-Labrador Shelf wascharacterizedby demer-sal dominated landingsuntil 1992,when demersal ﬁsheswere replacedbyinvertebrates.

Many Atlantic cod (Gadusmorhua) stocksin the northwest-ern Atlantichave experiencedextremepopulationdeclines due tooverexploitationandenvironmentallyinducedchangesin pro-ductivity (Dutilet al., 1999). The collapse of six of seven cod stocksrangingfromsouthernLabradortotheScotianShelfinduced Canada to imposebans oncommercial exploitation for several speciesin1992and1994(Myersetal.,1997).Codhadbeenthe maintargetspeciesoftheNewfoundlandfisherysincethe1500s, butin1992,afteralmostfivecenturiesofexploitation,thestock collapsedleadingtothedisruptionoftheAtlanticCanadianfishing industry.Afterthe“codmoratorium”wasimposedbythe Cana-diangovernmentin1992,fishingvesselsandgearswereconverted towardsnewlowtrophicleveltargetspecies,mainlyrepresented bydemersalinvertebrates[e.g.Northernshrimp(Pandalus bore-alis), Atlantic seascallop (Placopecten magellanicus), Snow crab (Chionoecetesopilio),Oceanquahog(Arcticaislandica)andAtlantic lobster(Homarusamericanus)].Thisnewfisheryregimewas suc-cessfulineconomicterms,asthemonetaryvalueofthecombined shrimpandcrablandingsexceededthatofthecollapsed ground-fishfisheries(Franketal.,2005),butthedisruption(andsuccessive reorganization)ofthetrophicwebwhichfollowedtheremovalof top-predators, raisesimportantissuesaboutecosystemstability andresiliencetowardsfutureperturbations(Franketal.,2006).

Fromatropho-dynamicperspective,thefisheriestracks exhib-itedbytheseLMEscouldrepresentanexampleof“fishingdown marinefoodwebs”(Paulyetal.,1998),i.e.adeclineofthe aver-agetrophicleveloflandingsovertime,eventhoughintheseLMEs suchatrendwasforcedbymanagementmeasuresratherthanby autonomousreorganizationofthefishingfleet.

OnlyoneSouthernHemisphere’s ecosystemwasclassiﬁedin cluster2:thePatagonianShelf.ThissouthAtlanticLMEhasbeen

(11)

Fig.11.SelfOrganizingMap.Proportionsofobservationsbelongingtoeachofthe 5clustersareshowningrayscale.SeealsoFigs.6and7forSelfOrganizingMap interpretation.Moredetailsoneachclusteraregiveninthetext(Section3.3.2).

characterized by large demersal-dominated landings until the beginningofthe1980s,whencephalopodsbecameimportant com-ponentsoftotal.

3.3.2.3. Cluster3–smallandmediumpelagics. Thisclusteris com-posedbyelevenLMEswhichsharethecommonfeatureoflandings representedmainlybysmallandmediumpelagics(compareFig.11

withFig.7,e.g.smallandmediumpelagics).LMEsencompassed into this clustercould besubdivided in three main subgroups: (1) enclosedandsemi-enclosedbasins(i.e.BalticSea, BlackSea andMediterraneanSea),(2)upwellingecosystems[i.e.three East-ern Boundary Currents (HumboldtCurrent, CanaryCurrentand BenguelaCurrent),ArabianSea,SouthBrazilShelf,IberianCoastal andPaciﬁcCentral-AmericanCoastal]and(3)enrichedbasins(i.e. GulfofMexico).Withtheexceptionoftheoligotrophic Mediter-raneanSea, LMEsclassiﬁedin Cluster3 aremoderate tohighly productiveecosystems(NOAA,2002)eventhoughnutrient avail-abilityistriggeredbydifferentprocessesinupwellingregionsand enclosedbasins,wherehumaninducedeutrophication,riverrunoff andlackofrapidexchangewithadjacentoceansplayamajorrole

(Caddy,1993;GaribaldiandLimongelli,2003;Kullenberg,1986;

NOAA,2002).

From an industrial perspective, small and medium pelagic speciesrepresentthemostabundantlandingsinworld’sﬁsheries. Ithasbeensuggestedthattheselowtrophic(andlow economi-callyvaluable)speciesaremoreabundantandheavilyexploitedin thoseecosystemsinwhichprimaryproductivityexperienceslow temporal(seasonalorintra-annual)variability(ContiandScardi,

2010).

3.3.2.4. Cluster4–smalldemersals.ThisclusterencompassedLMEs fromtheIndianOceanand SouthwestPacific,whichsharedthe commonfeatureoflandingstimeseriesdominatedbysmall dem-ersalharvests(compareFig.11withFig.7:smalldemersals).Going deeperintaxonomicresolution,itcanbeobservedthattheseLMEs actuallyshowedalargeamountofreportedlandingsinthe“mixed group”category.Thelow-resolvedlandingscompositionofthese ecosystemsis relatedtothelow taxonomicresolutionof tropi-calandsubtropicalfisheries,wheremixedcategories(e.g.,“Marine fishesnotidentified”)oftenrepresentthebulkofreportedlandings. 3.3.2.5. Cluster5–frommediumpelagicstosmallandmixed demer-sals. Cluster5wastheleastrepresentedclass,whichencompassed onlytwoLMEs:theNorthSeaandtheAgulhasCurrent(compare

Fig.11withFig.7,e.g.mediumpelagics).Althoughthese ecosys-temsshowedself-evidentdifferencesintheirgeographiclocation andmainfeatures(e.g.,hydrology,bathymetry,productivity,etc.), theysharedacommontrendinlandingscomposition,inwhich pelagicsrepresentedthedominantfractionintotallandingsinthe ﬁrsthalfof thetimeseries (i.e.,upuntil themid1970s), while demersalsbecamepredominantinmorerecentyears.

3.4. Mantel’stest

ThetimeseriesofMantel’sstatisticsresultingfromthe com-parison of the geographic distancematrix and annual distance matrices(Bray–Curtisdistance)offisherieslandingsbyfunctional groups (tonkm−2year−1)is shown in Fig. 12. A negative trend wasobservedforthisrelationshipwhentotallandingsabundances weretakenintoaccount(r=_−0.4442,p<0.01).Incontrast,Mantel’s statisticbetweenthegeographicdistancematrixandtheannual distancematrices(Euclideandistance)offisherieslandings com-position(i.e.,percentagecontributionofeachfunctionalgroupto totalyields)showednosignificantrelationship.Whenaddressing landingsdatabymeansoftheBray–Curtisdistance(i.e.takinginto accountlandingsabundances),thenegativetrendobservedinthe

(12)

Fig.12. Mantel’sstatisticstrendbetweenthegeographicdistancematrix(i.e., dis-tancebetweenLMEcentroids)andannualdistancematricesofﬁsheriescatches compositionbyfunctionalgroups.Dataareplottedbylandingsandproportions (e.g.Fig.4).

Mantel’sstatistictimeseriesbetweenlandingsprofilesand geo-graphicdistancesbetweenLMEscouldbeinterpretedasageneral homogenizationoftheseprofilesacrossLMEsovertime.Thistrend couldbeexplainedasaresponsetoincreasingfishingpressures andmarketglobalization.Onthecontrary,whenfunctionalgroup proportionsweretakenintoaccount(i.e.aqualitativeevaluation oflandingscompositionvariation),notrendwasdetected, demon-stratingthatthecompositiongenerallyremainedunchangedover time,showingthusaregionalpatternonly.

Althoughtheseexplanationsmayappearascontradictory,we suggestthatboth areplausibleandactatdifferentlevels:while ona quantitativebasis, increased fishingpressuresand market globalizationmayhaveweakenedtherelationshipbetween land-ingsprofilesandthegeographicallocationfromwhichtheseare extracted,theinfluenceofthesesamepressuresarecomparable amongallLMEs sothatthere isnoglobaltrendwhen landings compositionsareconsideredasproportions,i.e.regionalpatterns continuetodominatetheanalysisonaproportionalbasis.

4. Conclusions

Analyzingthetemporalvariationoffifty-onemultivariatetime serieshasbeenachallengingobjective.Weaimedatdescribing thevariationboth intime and spaceoffisheries landings com-positionintheselectedecosystems,inordertodetectpotential recurrentpatterns.Theresultsobtainedinthisstudyprovide sig-nificantinsightsintofisheriesdynamicsataglobalscaleoverthe lastfiftyyears.

Onaquantitativebasis,timeseriesofLMElandingsabundances werecharacterizedbydistinctpatterns,rangingfromawidespread trendofregularincreasingharveststomorecomplexvariations, wheretimeseriesshowedinterruptionsandturningpoints. More-over,landingscompositionbyfunctionalgroupsineachLMEseems torepresentafairlyconservativeandregionalfeature,despitethe ﬂuctuationsinharvestedabundances.

RegardingLMElandingscompositionbyfunctionalgroups,the rangeofvariationpresentedsomediscontinuities,whichenable ﬁveclustersofLMEstobedistinguished,eachrepresentinga dis-tinct“ﬁshingbehavior”.Inparticular,fromthehistoricalanalysis oftheseharveststemporaltracks,abroaddistinctionbetweentwo

majorapproachestofisheriesemerged.Ononeside,fisherieswhich relyonsmallandmediumpelagicsproductiontendtoexhibit sta-bleorcyclicbehaviorsinlandingscomposition.Thispatternhas beenrelatedtointrinsicfeaturesoftargetedspeciespopulations, whichtypicallyexhibita“wasp-waist”controlresultingincyclic outburstoftheexploitedsmallpelagicstocks(Curyetal.,2000). Moreover,ithasbeenobservedthattheLMEscharacterizedbythis typeoflandingsseemedtosharecommonproductivefeatures(i.e., upwellingregimesorenrichedbasins),whichmaysustainspecies at low trophic levels. It follows that pelagic-dominated land-ingsarelocatedinthoseregionswhichpresentspecificfeatures despitetheirgeographicallocation.Nevertheless,togeneralizethis relationship – between theecosystem productive features and pelagic-dominatedharvests–furtherinvestigationsareneeded.

Ontheotherhand,ademersalfisheriesclusterwasidentified, whereharvestsweredominatedbybottom-associatedresources. The geographical localization of the LMEs encompassedin this second category suggests that this cluster is more affected by economicdrivers(e.g.investmentsinfishinggearsandnew tech-nologies)ratherthanenvironmentalfeatures.Infact,theeleven LMEstendedtoconcentrateintheNorthernHemisphere,where fishingpressureshavebeenhistoricallyhigher(Anticamaraetal., 2011).It couldbefurther suggestedthat NorthernHemisphere LMEsarealsogenerallycharacterizedbywidercontinentalshelves, whichrepresentacriticalfeaturefordemersalexploitation.Allthe harvestsprofilesexceptthelandingsencompassedinthesetwo groupsrepresentedmixedandlow-resolvedharvests,whichare alsocharacterizedbylowertotalyieldswithrespecttodemersal or,especially,pelagicdominatedcatches.

Fromamanagementperspective,thepatterningofLMEfishing historiesproposedinthisstudyactuallyopennewscenariosfor thedecision-makers.Thecontrolstrategiesenactedforpurseseine (pelagic) and demersal(groundfish and invertebrates) fisheries must relyondifferentapproaches,yetwiththesameobjective. Themanagementofpelagicfisheriesusuallydealswithasingle tar-getspecies(orasingletargetfunctionalgroup,e.g.smallpelagics), whosestockdynamicsaredependentonnaturalvariability. There-fore,longtimeseriesofbothlandingsandenvironmental descrip-torsareneededinordertoeffectivelycontrolfishingpressures.

Incontrast,effectivemanagementstrategiesfordemersal fish-eriesneedtotakeintoaccountboththeintrinsicvariabilityofthe underlyingecosystemandtheeconomicpotentialoftheexploiting fisheryindustry(Contietal.,inpreparation).Moreover,thelargest amongthepreviouslymentionedclustersencompassedfishing his-toriesbasedonmixedlandings:thisfindinghighlightstheneedfor betterresolutionoflandingsdatainordertocapturesignificant trendsandtoprovideusefulindicationstopolicymakers.

The analysis of regional fisheries harvests variation over fifty-five years represents a first step towards the challenging goalofdisentanglingtherelativeinfluenceofenvironmentaland economicalforcingonexploitedecosystems.Whilesomeattempts proposednewinsightsintospecificrelationships,e.g.betweentotal yieldsand primaryproductivity(Conti andScardi,2010)orSST

(Biswasetal.,2008)thereisstillalackofcomprehensionofother

dynamics.Inthiscontext,long-termmulti-decadalstudiescapture more spatio-temporal structures and natural phenomena, thus providingthebestwaytodescribeboththeanthropogenicand nat-uralbiologicalsignalsdrivingexploitedecosystems(Edwardsetal.,

2010).

Acknowledgment

WewouldliketothankDr.MontserratDemestre(Instituteof MarineScience,Barcelona)forreviewinganearlierversionofthe manuscriptprovidingusefulsuggestions.

(13)

References

Alheit,J.,Möllmannb,C.,Dutza,J.,Kornilovsc,G.,Loewed,P.,Mohrholza,V., Was-mundaet,N.,2005.SynchronousecologicalregimeshiftsinthecentralBaltic andtheNorthSeainthelate1980.ICESJ.Mar.Sci.62(7),1205–1215. Almeida,J.F.,2002.Predictivenon-linearmodelingofcomplexdatabyartiﬁcial

neuralnetworks.Curr.Opin.Biotech.13(1),72–76.

Anticamara,J.A.,Watson, R.,Gelchu, A., Päuly,D.,2011. Globalﬁshing effort (1950–2010):trends,gaps,andimplications.Fish.Res.107,131–136. Bakun, A., 1990. Global climate change and intensiﬁcation of coastal ocean

upwelling.Science247,198–201.

Bakun,A.,1998.Oceantriadsandradicalinterdecadalstockvariability:baneand boonforﬁsheriesmanagement.In:Pitcher,T.,Hart,P.J.B.,Pauly,D.(Eds.), Rein-ventingFisheriesManagement.ChapmanandHall,London,pp.331–358. Biswas,B.K.,Svirezhev,Y.M.,Bala,B.K.,Wahab,M.A.,2008.Climatechangeimpacts

onﬁshcatchintheworldﬁshinggrounds.ClimaticChange93,117–136. Bundy,A.,2005.StructureandfunctioningoftheeasternScotianShelfecosystem

beforeandafterthecollapseofgroundﬁshstocksintheearly1990.Can.J.Fish. Aquat.Sci.62,1453–1473.

Caddy,J.,1993.Towardacomparativeevaluationofhumanimpactsonﬁshery ecosystemsofenclosedandsemi-enclosedseas.Rev.Fish.Sci.1,57–96. Christensen,V.,Guénette,S.,Heymands,J.J.,Walters,C.J.,Watson,R.,Zeller,D.,Pauly,

D.,2003.HundredyeardeclineofNorthAtlanticpredatoryﬁshes.FishFish.4, 1–124.

Christensen,V.,Walters,C.J.,Ahrens,R.,Alder,J.,Buszowski,J.,Christensen,L.B., Cheung,W.W.L.,Dunne,J.,Froese,R.,Karpouzi,V.,Kaschner,K.,Kearney,K., Lai,S.,Lam,V.,Palomares,M.L.D.,Peters-Mason,A.,Piroddi,C.,Sarmiento,J.L., Steenbeek,J.,Sumaila,R.,Watson,R.,Zeller,D.,Pauly,D.,2009.Database-driven oftheworld’slargemarineecosystems.Ecol.Model.220,1984–1996. Conti,L.,Scardi,M.,2010.Fisheriesyieldandprimaryproductivityinlargemarine

ecosystems.Mar.Ecol.Progr.Ser.410,233–244.

Cury,P.M.,Bakun,A., Crawford,R.J.M.,Jarre, A.,Qui ˜nones,R.A.,Shannon, L.J., Verheye,H.M.,2000.Smallpelagicsinupwellingsystems:patternsof inter-actionandstructuralchangesin“wasp-waist”ecosystems.ICESJ.Mar.Sci.57, 603–618.

Cury,P.M.,Shin,Y.-J.,Planque,B.,Durant,J.M.,Fromentin,J.-M.,Kramer-Schadt,S., Stenseth,N.C.,Travers,M.,Grimm,V.,2008.Ecosystemoceanographyforglobal changeinﬁsheries.TrendsEcol.Evol.23(6),338–346.

Drinkwater,K.,2009.ComparisonoftheresponseofAtlanticcod(Gadusmorhua)in thehigh-latituderegionsoftheNorthAtlanticduringthewarmperiodsofthe 1920–1960sandthe1990–2000s.DeepSeaRes.II56(21–22),2087–2096. Dutil,J.D.,Castonguay,M.,Gilbert,D.,Gascon,D.,1999.Growth,condition,and

envi-ronmentalrelationshipsinAtlanticcod(Gadusmorhua)inthenorthernGulf ofSt.LawrenceandimplicationsformanagementstrategiesintheNorthwest Atlantic.Can.J.Fish.Aquat.Sci.56(10),1818–1831.

Edwards,M.,Beaugrand,G.,Hays,G.C.,Koslow,J.A.,Richardson,J.A.,2010. Multi-decadaloceanicecologicaldatasetsandtheirapplicationinmarinepolicyand management.TrendsEcol.Evol.25(10),602–610.

Fausett,L.V.,1994.FundamentalsofNeuralNetworks:Architectures,Algorithms, andApplications.PrenticeHall,EnglewoodCliffs,NJ,xvi+460pp.

FoodandAgricultureOrganizationoftheUnitedNations(FAO),2010.TheState ofWorldFisheriesandAquaculture(SOFIA).FAOFisheriesandAquaculture Department,Rome.

Francis,R.C.,Hare,R.H.,1994.Decadal-scaleregimeshiftsinthelargemarine ecosys-temsoftheNorth-eastPaciﬁc:acaseforhistoricalscience.Fish.Oceanogr.3(4), 279–291.

Frank,K.T.,Petrie,B.,Choi,J.S.,Leggett,W.C.,2005.Trophiccascadesinaformerly cod-dominatedecosystem.Science308,1621–6123.

Frank,K.T.,Petrie,B.,Shackell,N.L.,Choi,J.S.,2006.Reconcilingdifferencesintrophic controlinmid-latitudemarineecosystems.Ecol.Lett.9,1096–1105. Fréon,P.,Mullon,C.,Voisin,B.,2003.Investigatingremotesynchronouspatternsin

ﬁsheries.Fish.Oceanogr.12(4–5),443–457.

Garibaldi,L.,Limongelli,L.,2003.Trendsinoceaniccapturesandclusteringoflarge marineecosystems:twostudiesbasedontheFAOcapturedatabase.FAOFish. Tech.Pap.435.FoodandAgricultureOrganizationoftheUnitedNations,Rome. Hecht-Nielsen,R.,1990.Neurocomputing.AddisonWesleyPublishingCompany. Hughes,T.P.,Bellwood,D.R.,Folke,C.,Steneck,R.S.,Wilson,D.,2005.Newparadigms

for supportingtheresilience ofmarineecosystems.Trends Ecol.Evol.20, 380–386.

Jackson,J.B.C.,Kirby,M.X.,Berger,W.H.,Bjorndal,K.A.,Botsford,L.W.,Bourque, B.J.,Bradbury,R.H.,Cooke,R.,Erlandson,J.,Estes,J.A.,Hughes,T.P.,Kidwell, S.,Lange,C.B.,Lenihan,H.S.,Pandolﬁ,J.M.,Peterson,C.H.,Steneck,R.S.,Tegner, M.J.,Warner,R.R.,2001.Historicaloverﬁshingandtherecentcollapseofcoastal ecosystems.Science293,629–638.

Kohonen,T.,1982.Self-organizedformationoftopologicallycorrectfeaturemaps. Biol.Cybern.43,59–69.

Kohonen,T.,1990.Theselforganizingmap.Proc.IEE78,1464–1480. Kohonen,T.,1995.Self-OrganizingMaps.Springer,Berlin.

Kullenberg,J.,1986.Long-termchangesintheBalticecosystem.In:Sherman,K., Alexander,L.M.(Eds.),VariabilityandManagementofLargeMarineEcosystems. AAASSelectedSymposium99.WestviewPress,Boulder,CO,USA.

Laë,R.,Lek,S.,Moreau,J.,1999.PredictingﬁshyieldofAfricanlakesusingneural networks.Ecol.Model.120,325–335.

Lek,S.,Baran,P.,1997.Estimationsoftroutdensityandbiomass:aneuralnetworks approach.NonlinearAnal.30(8),4985–4990.

Lek,S.,Delacoste,M.,Baran,P.,Dimopoulos,I.,Lauga,J.,Aulagnier,S.,1996. Appli-cationofneuralnetworkstomodellingnonlinearrelationshipsinecology.Ecol. Model.90(1),39–52.

Lek,S.,Guégan,J.F.,1999.Artiﬁcialneuralnetworksasatoolinecologicalmodelling, anintroduction.Ecol.Model.120,65–73.

Lippmann,R.P.,1987.Anintroductiontocomputingwithneuralnets.IEEEAcoust. SpeechSignalProcess.April,4–22.

Lluch-Belda,D.,Schwartzlose,R.A.,Serra,R.,Parrish,R.,Kawasaki,T.,Hedgecock,D., Crawford,R.J.M.,1992.Sardineandanchovyregimeﬂuctuationsofabundancein fourregionsoftheworldoceans:aworkshopreport.Fish.Oceanogr.1,339–347. Myers,R.A.,Hutchings,J.A.,Barrowman,N.J.,1997.Whydoﬁshstockscollapse?The

exampleofcodinAtlanticCanada.Ecol.Appl.7(1),91–106.

Myers,R.A.,Worm,B.,2003.Rapidworldwidedepletionofpredatoryﬁsh commu-nities.Nature423,280–283.

Mullon,C.,Fréon,P.,Cury,P.,2005.Thedynamicofcollapseinworldﬁsheries.Fish Fish.6,111–120.

NOAA–NationalOceanicandAtmosphericAdministration,2002.Largemarine ecosystemsoftheworld.<http://www.edc.uri.edu/lme/default.htm>. Overland,J.,Rodionov,S.,Minobe,S.,Bond,N.,2008.NorthPaciﬁcregimeshifts:

deﬁnitions,issuesandrecenttransitions.Progr.Oceanogr.77,92–102. Pauly,D.,Christensen,V.,Dalsgaard,J.,Froese,R.,Torres,F.,1998.Fishingdown

marinefoodwebs.Science279,860–863.

Pauly,D.,Maclean,J.,2003.Inaperfectocean:ﬁsheriesandecosystemintheNorth Atlantic.IslandPress,Washington,DC.

Pauly,D.,Watson,R.,Alder,J.,2005.Globaltrendsinworldﬁsheries:impactson marineecosystemsandfoodsecurity.Phil.Trans.R.Soc.B360,5–12. Ritter,H.,Martinetz,T.,Schulten,K.,1992.NeuralComputationandSelf-Organizing

Maps.Addison-WesleyPublishingCompany,Reading,MA.

Rose,G.A.,2005.Capelin(Mallotusvillosus)distributionandclimate:asea“canary” formarineecosystemchange.ICESJ.Mar.Sci.62,1524–1530.

Savenkoff,C.,Castonguay,M.,Chabot,D.,Hammill,M.O.,Bourdages,H.,Morissette, L.,2007.ChangesinthenorthernGulfofSt.Lawrenceecosystemestimatedby inversemodelling:evidenceofaﬁshery-inducedregimeshift?Estuar.Coast. ShelfS73,711–724.

Schwartzlose,R.A.,Alheit,J.,Bakun,A.,Baumgartner,T.R.,Cloete,R.,Crawford,R.J.M., Fletcher,W.J.,Green-Ruiz,Y.,Hagen,E.,Kawasaki,T.,Lluch-Belda,D., Lluch-Cota,S.E.,MacColl,A.D.,Matsuura,Y.,Nevarez-Martinez,M.O.,Parrish,R.H., Roy,C.,Serra,R.,Shust,K.V.,Ward,M.N.,Zuzunaga,J.Z.,1999.Worldwide large-scaleﬂuctuationsofsardineandanchovypopulations.SouthAfr.J.Mar.Sci.1, 289–347.

SeaAroundUsProject.Databaseonlineat<www.seaaroundus.org>.

Sherman,K.,Duda,A.M.,1999.Largemarineecosystems:anemergingparadigmfor ﬁsherysustainability.Fisheries24(12),15–26.

Stein,M., 1998. Integratingﬁsheriesobservationswith environmentaldata – towardsabetterunderstandingoftheconditionsforﬁshinthesea.J.Northw. Atl.Fish.Sci.23,143–156.

Worm, B.,Myers, R., 2003. Meta-analysis of cod–shrimp interactionsreveals top–downcontrolinoceanicfoodwebs.Ecology84,162–173.

Watson,R.,Pauly,D.,2001.Systematicdistortionsinworldﬁsheriescatchtrends. Nature414,534–536.

Zurada,J.M.,1992.IntroductiontoArtiﬁcialNeuralSystems.WestPublishing Com-pany,NewYork.