Pearsonsss258.com

MARINEECOLOGYPROGRESSSERIESMarEcolProgSerVol.
486:257–276,2013doi:10.
3354/meps10397PublishedJuly12INTRODUCTIONSpawningstrategiesofmarinefishesfrequentlyappeartotargetoceanographicconditionsthatpro-moteenhancedretention,survivalandeventualrecruitmentoflarvae.
Thismaybethroughfavorablefeedingconditions,lowdensitiesofplanktonicpre-dators,localretentionordeliverytowardsnurserygrounds,orbyprovidingsuitabletemperaturesforhatching,growthandlarvaldevelopment.
Suchstrategiesdiffermarkedlydependingontheecologyandlifehistoryofthespeciesandthecharacteristicsofthenurseryhabitattowhichpost-larvaemustrecruit.
Forexample,severalcold-waterfishesintheInter-Research2013·www.
int-res.
com*Email:barbara.
muhling@noaa.
govComparisonbetweenenvironmentalcharacteristicsoflarvalbluefintunaThunnusthynnushabitatintheGulfofMexicoandwesternMediterraneanSeaBarbaraA.
Muhling1,*,PatriciaReglero2,LorenzoCiannelli3,DiegoAlvarez-Berastegui4,FranciscoAlemany2,JohnT.
Lamkin5,MitchellA.
Roffer61CooperativeInstituteforMarineandAtmosphericStudies,UniversityofMiami,Miami,Florida33149,USA2InstitutoEspaoldeOceanografía,CentreOceanogràficdelesBalears,MolldePonents/n,07015PalmadeMallorca,Spain3CollegeofEarth,OceanandAtmosphericSciences,104CEOASAdministrationBuilding,OregonStateUniversity,Corvallis,Oregon97331-5503,USA4BalearicIslandsCoastalObservingandForecastingSystem,ParcBit,Naorte,BlocA2-3,07121PalmadeMallorca,Spain5NationalOceanicandAtmosphericAdministration,NationalMarineFisheriesService,SoutheastFisheriesScienceCenter,Miami,Florida33149,USA6Roffer'sOceanFishingForecastingService,Inc.
,60WestoverDrive,WestMelbourne,Florida32904,USAABSTRACT:DespitebeingwelladaptedforfeedingincoldwaterontheirNorthAtlanticfeedinggrounds,Atlanticbluefintunaundertakelongmigrationstoreachwarm,lowproductivityspawn-inggroundsintheGulfofMexicoandMediterraneanSea.
Environmentalconditionswithinspawningareashavebeenpresumedtobenefitlarvalsurvival,throughappropriatefeedingconditions,andenhancedlarvalretentionandgrowthrates.
However,fieldcollectionsandstudiestoexplorethepotentialmechanismsarerare.
Inthisstudy,acomparisonoftheenvironmentalcharacteristicsofbothspawningsiteswascompletedusingstandardizedenvironmentaldataandmodelingmethods.
Predictivemodelsoflarvaloccurrencewereconstructedusinghistoricallarvalcollections,andenvironmentalvariablesfrombothinsituandremotelysensedsources.
Resultsshowedthatlarvaeonbothspawninggroundsweremostlikelytobefoundinwarm(23to28°C),lowchlorophyllareaswithmoderatecurrentvelocitiesandfavorableregionalretentioncondi-tions.
IntheGulfofMexico,larvaewerelocatedinoffshorewatersoutsideoftheLoopCurrentandwarmeddies,whileinthewesternMediterranean,larvaloccurrenceswereassociatedwiththeconfluenceofinflowingAtlanticwatersandsaltierresidentsurfacewaters.
AlthoughourresultssuggestedcommonthemeswithinpreferredspawninggroundsonbothsidesoftheAtlanticOcean,theecologicalprocessesgoverninglarvalsurvivalandeventualrecruitmentareyettobefullyunderstood.
KEYWORDS:Bluefintuna·Spawning·MediterraneanSea·GulfofMexicoResaleorrepublicationnotpermittedwithoutwrittenconsentofthepublisherMarEcolProgSer486:257–276,2013NorthAtlanticOceanspawnwithintheGeorgesBankGyreduringthespringplanktonbloomtoalignlarvaewithfavorablefeedingconditionsandpro-motelarvalretention(Shermanetal.
1984).
Atlanticherringalsospawninconcertwithoceanographicretentionfeatures,whichmaymaximizereturnoflarvaetonurseryhabitats(Sinclair&Iles1985).
Alongupwellingcoasts,speciesmayspawnup-streamoffavorablelarvalfeedingareas(Roy1998,Landaeta&Castro2002),orduringtimesoflowupwellingactivitytomaximizecoastalretention(Castroetal.
2000,Shanks&Eckert2005).
Tempera-ture-sensitivespeciesmayspawninareasoflocallyenhancedproductiononcesuitablewatertempera-turesarereached(Motosetal.
1996),whileothersmigratetospawninggroundsbasedontemperaturecues(Simsetal.
2004)andothercues(Cury1994,Corten2001).
AtlanticbluefintunaThunnusthynnus(BFT)isahighlymigratoryspeciesthattravelslongdistancesbetweencold-waterforagingareasandwarm,oligo-trophicspawningareas(Blocketal.
2005,Rookeretal.
2007,Galuardietal.
2010).
AdultfishconsumecephalopodsandsmallbenthicandpelagicfishesaspreywhileonfeedinggroundsintheNorthAtlanticOcean(Chase2002,Estradaetal.
2005).
Theyarecapableofdivingtodepthsofupto1000minwatersascoldas3°Ctoexploitthisprey(Blocketal.
2001,Wallietal.
2009).
TheuniquecirculatorysystemofadultBFTallowsthemtokeeptheirmuscles,eyes,brainandviscerawarmerthanambientwatertem-peratures,whichhasfacilitatedtheirexpansionintocoldnorthernAtlanticwaters(Graham&Dickson2004,Fromentin&Powers2005).
Atpresent,al-thoughtheyarecapableoftransatlanticmigrations,BFTintheNorthAtlanticaremanagedas2separatestocks(Blocketal.
2005,InternationalCommissionfortheConservationofAtlanticTunas2011).
ThewesternstockgenerallyfeedsintheNorthwestAtlantic,andspawnsintheGulfofMexico(GOM)duringthespring(Stokesburyetal.
2004).
Theeast-ernstockhasfeedinggroundsoffPortugal,MoroccoandtheBayofBiscay,althoughsomefishalsoappeartosharenorthwesternAtlanticfeedinggroundswithfishfromthewesternstock(Blocketal.
2005,Fromentin&Powers2005).
AlladultfishfromtheeasternstockarepresumedtospawnintheMedi-terraneanSeaduringthesummer(Oray&Karakulak2005,Rookeretal.
2008,Alemanyetal.
2010).
Despitebeingwelldesignedforlifeincoldwaters,adultBFTmigratelongdistancestoreachwarm,oligotrophicseasinwhichtospawn.
Arrivalandres-idencyonspawninggroundsvariesgreatly:notallfishappeartomigrateeachyear(Wilsonetal.
2005),andthosethatdocanspendaslittleas2wkinspawningareasorasmuchasseveralmonths(Blocketal.
2001,Galuardietal.
2010).
Asthespawningseasonprogresses,watertemperaturesonspawninggroundscanwarmto>28°C.
Althoughwelladaptedtocoldwaters,BFTarenotwellsuitedforverywarmwaters.
Astemperaturesapproach30°C,theircar-diacfunctionmaybeinhibitedandtheyrequireverylargeamountsofoxygentosurvive(Blanketal.
2004).
Inaddition,feedingconditionsforadultfishmaybepoorcomparedwithNorthAtlanticfeedinggrounds,asboththeGOMandMediterraneanSeaareoligotrophicinoffshorewaters,andlowinprima-ryandsecondaryproductivity(Azov1991,Estrada1996,Gilbesetal.
1996).
Despitetheselimitations,adultBFTmigratelongdistancestoreachthesespawninggrounds.
Thisimpliesthatenvironmentalconditionsonspawninggroundsarebeneficialforlarvalsurvival.
SomepreviousstudiesexaminingtheassociationsbetweenlarvalBFTdistributionsandenvironmentalconditionshavebeencompletedusingdifferentmethodologiesanddifferentenvironmentalpredic-tors.
TheseanalysesfoundthatBFTlarvaetendedtobelocatedwithinparticularoceanographiczones.
InthewesternMediterraneanSea,larvaeweremostabundantnearthesalinityfrontwhereinflowingAtlanticwatersconvergewithhighersalinityresi-dentwatersaroundtheBalearicIslands(Garciaetal.
2005,Alemanyetal.
2010,Regleroetal.
2012).
Lar-vaehavealsobeencollectedinotherregionswithintheMediterraneanSea,mainlyinthecentralMedi-terraneanaroundSicilyandothersmallerislands,inareasalsocharacterizedbytheconfluenceofinflow-ingsurfaceAtlanticwatersandsaltierresidentwaters(Tsujietal.
1997,Kochedetal.
2012).
Morerecently,BFTlarvaehavebeenfoundnearCyprus(Oray&Karakulak2005).
However,thesewereassumedtocomefromaresidentsubpopulationthatdoesnotmigrateoutoftheMediterraneanSea.
IntheGOM,BFTlarvaewerefoundmostfrequentlyin'commonwaters'outsideofthewarmLoopCurrentandassociatedwarm-corerings,andawayfromhigherchlorophyllcontinentalshelfwaters(Muhlingetal.
2010,Lindo-Atichatietal.
2012).
Spawningsea-sonsinbothregionsappearedlikelytoextendforaround6to8wk,andwerecloselytiedwithseatem-peratures(Muhlingetal.
2010).
Theseresultscon-trastwiththemoreprotractedspawningoftropicaltunas,whichcanlastformorethan6mooftheyear(Schaefer2001).
However,adirectcomparisonofconditionsonthe2spawninggrounds,usingthe258Muhlingetal.
:Larvalbluefintunahabitatcomparisonsameenvironmentalvariablesandthesamemeth-ods,islacking.
Despitetheirconsiderablecommercialvalue,ourunderstandingoftheecologicaldriversofmigrationandspawningbehaviorinadultBFTislimited.
ThereasonswhyadultBFTexpendlargeenergyreservestomigratetospawninglocationsthattheyfindphysiologicallystressfulremainunclear.
Bycompar-ingandcontrastingtheenvironmentalconditionsatspawninggroundsonbothsidesoftheAtlanticOcean,thegeneralconditionsadultfishappeartotargetforspawningcanbedefined.
Inaddition,wecanbegintodefinerelationshipsbetweenlarvaloccurrencesandthephysicalenvironment,anddis-tinguishbetweenthosethatareprobablyduetophysiologicalconstraintsandthosethatareproxiesforotherfactors,suchaswatermassorfeedingecolo-gy.
Improvingourunderstandingintheseareaswillalsoallowexistingindicesofabundanceusedinstockassessmentstoberefined(Ingrametal.
2010).
Inthefuture,inferencescanbedrawnaboutfactorsinfluencingrecruitment,potentialresponsestocli-matechangeandtheimplicationsofmanagementactions(Garciaetal.
2013).
Inthisstudy,weusedthesameenvironmentalvariablesandmodelingmethodstodefinetheenvi-ronmentalcharacteristicsoflarvalBFTcollectionlocationsinthenorthernGOMandthewesternMediterraneanSea.
LarvaloccurrencemodelswereconstructedusinghistoricallarvalsurveydataandenvironmentalvariablesfrombothinsituCTDcastsandremotelysensedsatellitedata.
Weaimedtocom-pareandcontrastconditionsateachsamplingsite,andtousethoseresultstoconsiderthepotentialeco-logicalrequirementsthatmaybenefitlarvalsurvival.
MATERIALSANDMETHODSLarvalsurveysLarvalBFTwerecollectedinthenorthernGOMthroughtheNationalMarineFisheriesServiceSoutheastAreaMonitoringandAssessment(SEA-MAP)Program(Fig.
1,Table1).
Datawereavailable259YearGOMGOMNo.
ofGOMGOMmeanBIBINo.
ofBIBImeanstartendstationsper10m2(SE)startendstationsper10m2(SE)199826April30May471.
05(0.
66)199924April30May611.
18(0.
48)200020April25May623.
16(1.
58)200118April27May615.
99(2.
34)16June7July1143.
48(1.
07)200219April28May503.
00(1.
51)7June26June554.
59(2.
01)200313May30May423.
97(1.
50)4July29July1175.
44(2.
51)20046May30May306.
14(3.
50)18June10July1547.
94(5.
68)200528June23July1592.
79(0.
75)200624April28May661.
51(0.
54)200718April23May353.
62(1.
82)200821April30May561.
57(0.
58)200914May31May2012.
99(4.
80)Table1.
No.
ofstationswithallenvironmentalvariablesavailable,andthedatestheyweresampledintheGulfofMexico(GOM)andBalearicIslands(BI),1998to2009.
Meanno.
ofAtlanticbluefintunaThunnusthynnus(BFT)larvaeper10m2isalsoshownforeachsampledyear(bongonetsonly),withstandarderrorsinparenthesesFig.
1.
SamplingareasintheGulfofMexico,1998to2009andBalearicIslands,2001to2005.
Samplesizesofstationsareshownonaonedegreesquaredgrid,andsummedtoshowsamplingeffortMarEcolProgSer486:257–276,2013for1982to2009,althoughonlydatafrom1998to2009wereusedforhabitatmodeling,toalignwiththeremotelysensedchlorophyllrecord(seebelow).
Obliquebongoandsurfaceneustontowswerecon-ductedasdescribedinScottetal.
(1993)andMuh-lingetal.
(2010)acrossagridofstationscompletedbetweenmidAprilandtheendofMayeachyear.
Bongonetswerefittedwitha333mmeshontwo61cmdiameterroundframesandtowedobliquelyto200mdepth,ortowithin10mofthebottominshal-lowerwater.
Neustonnetswerefittedwith0.
95mmmeshona1*2mframe,andweretowedatthesur-face.
Samplesfromtherightbongonetandtheneustonnetweresorted,andlarvaeidentifiedtothelowestpossibletaxaatthePolishPlanktonSortingandIdentificationCenterinSzczecin,Poland.
TheidentificationsofallscombridlarvaewereconfirmedattheSoutheastFisheriesScienceCenterinMiami,Florida.
InthewesternMediterraneanSea,larvalBFTwerecollectedduring5annualsurveyscompletedfrom2001to2005duringJuneandJulyaroundtheBale-aricIslands(BI)(Fig.
1,Table1).
TheareacoveredbythesesurveyswaslessthaninthenorthernGOM(~180*220nmilevs.
~700*330nmile).
However,stationspatialdensitywasgreater,withbetween185and221stationsperannualsurveyvs.
between31and81stationsintheGOM(Alemanyetal.
2010,Regleroetal.
2012).
SimilartotheGOMsurveys,lar-vaewerecollectedwithbongonetsof60cmdiame-ter,howevernetsweretowedobliquelytoonly70mdeptharoundtheBI(vs.
200mintheGOM).
Duringthefirst2yrofsurveysaroundtheBI,bothnetswerefittedwitha250mmesh,andfrom2003onwards,onenetwas200m(notsortedforlarvae)andonewas333m(usedforlarvalcollections).
Larvaewereremovedfromplanktonsamplesandidentifiedtospecieslevel.
IntheGOM,between20and66sta-tionsperyearwereavailablewithallenvironmentalvariablespresent,withbetween55and159peryearintheBI(Table1).
EnvironmentalvariablesAsuiteofenvironmentalvariableswereselectedforinclusioninpredictivehabitatmodels.
Tempera-tureisanimportantphysiologicalcontrolofmanybiologicalprocesses,includingspawning(e.
g.
Masu-maetal.
2006),andsotemperaturesat3discretedepths(0,100and200m)wereincluded.
Inadditiontotemperature,salinityisaneffectivetracerofwatermass,andsothisvariablewasalsoincludedatthesame3depths.
Chlorophyllisaproxyforprimaryproductivity,sosurfacechlorophyllvaluesavailablefromremotelysenseddatawereincorporated.
Geo-strophicvelocitiescanhighlightbroadfrontalzonesandregionsofdifferentretentionconditions.
Finally,larvalbehaviorsandcatchabilitymayfollowadielcycleinsomespecies,andsotimeofdaywasalsoincludedinhabitatmodels.
InboththenorthernGOMandBIsurveys,insituhydrographicdatawerecollectedusingaSeabirdSBE9/11CTDwithtemperature,pressure,conduc-tivityanddissolvedoxygensensors.
Temperatureandsalinityatthesurface(meanofupper5m),100and200mdepthwereavailableforthemajorityofsampledstations,andwerechosenasrepresentativedepthsforanalysis.
Temperatureatthesurfaceisinfluencedbyseasonalwarming,whilesalinityatthesurfaceisprimarilyaffectedbyriverineoutflowintheGOM,andtheinflowofAtlanticOceanwatersintheBI.
Physicalmeasurementsat100and200mdepthwerechosentoindicatewatermassstructureandsubsurfacefeatures.
WhilewatermassesintheBIregioncanfrequentlybedifferentiatedbysurfacevariables,particularlysalinity,theGOMbecomeslargelyisothermalbylatespring,andwatermassesaremuchmoreeasilydefinedatdepth.
RemotelysensedvariablesReliableinsitumeasurementsofchlorophyllawerenotavailableforbothsurveys,soremotelysenseddatawereusedinstead.
Surfacechlorophyllvaluesinmgm3wereextractedforeachstationdateandlocationfrom8dcompositesofdatafrom2sensors:Sea-viewingWideField-of-viewSensor(SeaWiFS)andModerateResolutionImagingSpectro-radiometer(MODIS)-Aqua.
Level3StandardMap-pedImageproductspublishedbyNASA'sGoddardSpaceFlightCenterOceanColorGroupwereselectedforuse.
SeaWiFSdatawereavailablefrom1998to2010,withsomeperiodicdatagapsfromsatelliteout-agesinmorerecentyears.
MODIS-Aquadatawereavailablefrom2003onwards.
Compositesof8dwerechosenasasuitablecompromisebetweenspatialandtemporalaccuracy,andminimizingcloudim-pacts.
DatavalueswereextractedusingtheMarineGeospatialEcologyToolbox(MGET,Robertsetal.
2010)inArcGIS9.
3(Esri).
Wheredatafromonlyonesensorwereavailableforaparticularstation,thatvaluewasused,andwheredatafrombothsensorswereavailable,ameanwastaken.
Correlationsbe-tweenvaluesfromthe2sensorswerestronginboth260Muhlingetal.
:LarvalbluefintunahabitatcomparisontheGOM(Pearson'sr=0.
8),andtheBI(Pearson'sr=0.
86).
GeostrophicvelocityintheGOMwasobtainedfromdailyAvisoglobalaltimetrydataataspatialresolutionofapproximately0.
333°,andextractedinArcGISusingtheMGETtoolbox.
Owingtothepres-enceofislandsandcomplexmesoscaleflows,geo-strophicvelocitymeasurementsaroundtheBIwerebestobtainedfrominsitumeasurements(Regleroetal.
2012).
Inthisregion,geostrophicvelocitywasesti-matedbydifferentiatingtheinterpolateddynamicheightfield,whichwascalculatedfromCTDdata(Torresetal.
2011).
Thesedatahadavariablespatialresolution,dependingonthesamplinggrid,butaver-agedapproximately0.
167°.
ModeledclimatologiesClimatologiesoftemperatureandsalinityatthesurfaceand100mdepth,surfacechlorophyllandsurfacecurrentvelocitieswereconstructedtobettervisualizetypicaloceanographicconditionsineachregionatthetimeofpeakspawning.
Surfacechloro-phyllclimatologieswereconstructedusingMODIS-Aquadata,whiletemperatureandsalinityclimatolo-gieswereconstructedusingtheHYbridCoordinateOceanModel(HYCOM)+NavyCoupledOceanDataAssimilation(NCODA)Global1/12DegreeAnalysis(GLBa0.
08).
AlthoughHYCOMperformedquitewellinreplicatingobservedtemperatureandsalinityfieldsinbothregions,andsurfacecurrentfieldsintheGOM,itwaslesssuccessfulforestimat-ingsurfacecurrentsintheBIregion.
Forthisreason,meangeostrophiccurrentsfromAvisomeanabsolutedynamictopographieswereusedforclimatologiesofsurfacecurrentsinbothregions,despitelowerspa-tialresolution.
SpawningpeaksaroundthesecondhalfofMayintheGOM(Muhlingetal.
2010)andinearlyJulyintheBIregion,sothesemonthswerecho-senforclimatologicalanalyses.
Datafrom2003to2012wereusedforeachclimatology,allofwhichwerecompiledusingMGETinArcGIS.
HabitatmodelsHabitatmodelswereconstructedbycombiningall8environmentalvariables(surfacetemperature,sali-nity,chlorophyll,temperatureandsalinityat100and200mdepth,geostrophiccurrents)withdayoftheyearandhourofsampling,intomultivariatepre-dictivehabitatmodels.
Aninitialproblemwiththeinputvariableswasthehighdegreeofcollinearity(Table2).
Thiscanresultinmodelnon-convergence,inconsistentresultsanddegradedmodelperfor-mance(Walczak&Cerpa1999,Neumann2002).
Pat-ternsofcollinearitydifferedbetweenthe2regions.
261SSTSSST100mS100mT200mS200mYeardayHourChlGOMSSS0.
04T100m0.
510.
10S100m0.
000.
300.
08T200m0.
410.
090.
900.
13S200m0.
250.
220.
670.
260.
72Yearday0.
490.
050.
020.
060.
020.
01Hour0.
070.
010.
020.
010.
010.
030.
01Chl0.
280.
460.
310.
050.
280.
260.
010.
03GeoVel0.
400.
120.
600.
090.
470.
480.
140.
080.
25BISSS0.
12T100m0.
020.
64S100m0.
040.
650.
69T200m0.
030.
510.
560.
40S200m0.
020.
330.
360.
670.
20Yearday0.
880.
260.
050.
120.
050.
16Hour0.
020.
100.
030.
040.
030.
040.
06Chl0.
190.
120.
200.
110.
220.
080.
270.
03GeoVel0.
010.
220.
280.
190.
060.
170.
070.
080.
04Table2.
PearsoncorrelationsamongenvironmentalvariableswithintheGulfofMexico(GOM)andBalearicIslands(BI).
SSdenotesseasurface,or0mdepth,Tistemperature,Sissalinity,ChlischlorophyllaandGeoVelisgeostrophicvelocity.
Correlations>0.
5areshowninboldMarEcolProgSer486:257–276,2013IntheBI,surfacesalinitywascorrelatedwithtemper-atureandsalinityatdepth,andsurfacetemperatureswerealsostronglycorrelatedwithdayoftheyear(Table2).
However,intheGOM,temperatureandsalinityatdepthwerestronglycorrelatedwitheachother.
Toaddressthisproblem,wefollowedtheap-proachofMalmgren&Winter(1999)andSousaetal.
(2007),andusedprincipalcomponentanalysis(PCA)toreducethedimensionalityoftheenvironmentalvariablematrixandprovideanewsetofuncorre-latedaxes.
PCAwasappliedtoaEuclideandistancematrixofnormalizedenvironmentaldatafromeachregionseparatelyinPrimer-6software(Clarke&Gorley2006),withallenvironmentalvariablesin-cluded.
Resultingprincipalcomponentaxesweretheninputaspredictorvariablestoamultilayerper-ceptronartificialneuralnetworkmodeltoprovideestimatesoflarvalBFTdistributionsanddefineenvi-ronmentalconditionsinbothregionswherelarvaehavebeenmostlikelytobecollectedhistorically.
Artificialneuralnetworksareflexibleandaccuratepredictivemodelsthatdonotrequirenormallydis-tributeddataandcanmodelhighlynonlinearfunc-tions.
Theygenerallyperformwellforaddressingquestionsrelatedtospeciesdistributionmodeling(Segurado&Araujo2004).
Modelslearnbyself-adjustingasetofparameters,andthenusinganalgorithmtominimizetheerrorbetweenthedesiredoutput(observed)andthenetworkoutput(pre-dicted)(Malmgren&Winter1999).
Multilayerper-ceptronneuralnetworksconsistofsystemsofinter-connectednodeswithaninputlayer,ahiddenlayerandanoutputlayer,witheachlayercontainingoneormoreneurons.
Eachlayerisconnectedbynon-lineartransferfunctions(Gardner&Dorling1998).
Theinputlayercontainsoneneuronforeachpredic-torvariable.
Thehiddenlayercontainsavariablenumberofneurons:moreneuronstendtoimprovemodelfit,howevertoomanycanleadtooverfitting,thusinternalmodelvalidationisrequired(Oldenetal.
2008).
Inthisstudy,overfittingwaspreventedbyholdingback20%oftrainingdatarowsasthemodelfitwasbeingperformedandcontinuallyevaluatingmodelfitagainstthehold-outdataset(Sherrod2003).
Theneuralnetworksusedweresupervised,feed-for-wardmodelswithonehiddenlayer,trainedbyacon-jugategradientback-propagationalgorithm.
Thesetypesofneuralnetworksarepopularinecologybe-causetheycanapproximateanycontinuousfunction(Oldenetal.
2008).
Logisticfunctionswereusedforhidden-andoutput-layeractivation(Sherrod2003),andv-foldcrossvalidationwasusedformodeltest-ingandvalidation.
Therelativeimportanceofeachpredictorvariablewascalculatedusingsensitivityanalysis,wherethevaluesofeachvariableareran-domizedandtheeffectonthequalityofthemodelismeasured(Sherrod2003).
Variablesnotcontributingtoafinalmodelareexcluded.
Thisprocessresultsinascoreoutof100foreachpredictorvariable,with100alwaysdenotingthemostimportantpredictor.
Aseparatemodelwasconstructedforeachsampledregion(GOMandBI),usingDTREGsoftware(Sher-rod2003).
AsBFTlarvalabundanceswereoftenlowandvari-able,andthesamplinggearsbetweenregionswerenotdirectlycomparable,modelswereconstructedtopredictthepresenceorabsenceofBFTlarvae,ratherthanabundance.
NettowsintheGOMweremuchdeeperthanintheBI(200vs.
70mdepth).
LarvalBFTareusuallymostabundantintheupper10to20mofthewatercolumn(Muhlingetal.
2012),sotheGOMsurveysprobablyunderestimatedthepresenceofBFTlarvaesubstantially.
Topartiallyaccountforthis,ifeitherorbothgears(bongoorneuston)ateachstationintheGOMcaughtBFTlarvae,thestationwasconsideredtobea'positive'station(seeMuhlingetal.
2010).
RecentexperimentsusingagearmorecomparabletothetechniqueusedintheBIshowthevalidityofthismethodology(Muhlingetal.
2012).
Tocompensateforthispotentialdifferenceincatcha-bilitybetweenregions,andgeneralexpectedineffi-cienciesinsamplingofrarelarvaeduetogearavoid-ance,misclassificationcostswereassignedtoeachmodelfromeachregion(Sherrod2003).
Thispara-meterwasadjustedinaniterativefashion,andsetatthelowestpossiblevalueatwhichmodelsensitivityremainedabove80%(Muhlingetal.
2010).
Toinvestigatetherelationshipsbetweenmodelpredictionsandtherawenvironmentalvariables,scatterplotsbetweeneachvariableandthepredictedprobabilityoflarvaloccurrencefromeachneuralnetworkmodelwereconstructed.
Althoughtheseplotsdidnottakeintoaccountinteractionsamongvariables,theyprovidedanindicationofwhichvari-ablesweremostimportantinpredictingsuitablehabitat.
Inaddition,theyshowedthenatureoftherelationshipbetweenlarvaloccurrencesandeachvariable(i.
e.
positive,negative,parabolic),usingpolynomiallinesofbestfit.
Observedproportionsofpositivestationsforeachvariable,dividedintobins,wereoverlaidonthescatterplots.
Tobrieflyinvestigatepotentialinteractionsamongvariables,wireframecontourplotsofmodeledprob-abilitiesofoccurrencewereconstructedfor2influen-tialenvironmentalvariables,foreachstudyregion.
Aseachneuralnetworkmodelrankedtheinputvari-262Muhlingetal.
:Larvalbluefintunahabitatcomparisonables(i.
e.
principalcomponentaxes)intermsofcon-tributiontothemodel,weselectedanenvironmentalvariablestronglycorrelatedwiththemostimportant,andsecondmostimportant,inputprincipalcompo-nentaxisforeachmodelforinclusioninthewire-frameplots.
Generatedprobabilitiesofoccurrencefromneuralnetworkmodelsfrom2selectedfromeachregionwerecontouredacrosseachsamplingarea,usingkriginginSurfer9(GoldenSoftware),andBFTlar-vaecatchlocationsforeachoftheseyearswereoverlaid.
Weselected2yearsfromeachregionwithdifferingoceanographicconditionstoinvestigateifandhowlarvaloccurrencesmovedwithoceano-graphicfeatures.
RESULTSRegionaloceanographicconditionsTheGOMduringMaywastypicallycoolerinsur-facewatersonthecontinentalshelfthanoffshore,andwarmerwithinthemainbodyoftheLoopCur-rent(Fig.
2).
SurfacesalinitiesontheshelfwerelowerthanintheopenGOM,inassociationwithriverineoutflow.
Atdepth,bothtemperatureandsalinitywerehigherwithintheLoopCurrentandinthegeneralpositionoflargeanticyclonicrings,whichareperiodicallyshedfromthecurrent.
Surfacechlorophyllwashighestonthecontinentalshelf,andlowestwithintheLoopCurrentregion.
Surfacecur-rentvelocitiesweremarkedlyhigherintheareaofinfluenceoftheLoopCurrent,fromthewesternCaribbeanSeathroughtotheeasternGOMandFloridaStraits(Fig.
2).
SurfacetemperaturesduringJulyinthewesternMediterraneanSeawerewarmestoffthecoastsofSpainandAlgeria,andaroundtheBI(Fig.
3).
CoolerwaterswereassociatedwiththesouthcoastofFrance.
Salinitiesatthesurfaceandatdepthwerelowestinthewesternportionofthestudyarea,inassociationwithinflowofAtlanticwater.
Tempera-tureatdepthwasspatiallyvariable,withmaximumvaluesoffthenorthernAfricancoast.
SimilartotheGOM,surfacechlorophyllwasgenerallylow,withhighervaluesnearthesouthcoastofFrance.
SurfacecurrentvelocitieswerehighestintheAlboranSea,263Fig.
2.
Monthlyclimatologiesfor6environmentalvariablesintheGulfofMexicoduringMay.
Temperatureandsalinityatthesurfaceandat100mdepthwerederivedfromtheHYbridCoordinateOceanModel(HY-COM)+NavyCoupledOceanDataAssimilation(NCODA)Global1/12DegreeAnalysis(GLBa0.
08),2003to2012.
Geostrophiccurrentveloci-tieswereobtainedfromAvisoaltimetrymeanabsolutedynamictopogra-phyvalues.
SurfacechlorophyllwasderivedfromModerateResolutionImagingSpectroradiometer(MODIS)-Aqua8dcomposites,2003to2012.
AschematicoftheLoopCurrent(solidblackarrow)andananticyclonicring(dashedblackarrow)arealsoshown(Lindo-Atichatietal.
2012)MarEcolProgSer486:257–276,2013andalongthenortherncoastofAlgeria(Fig.
3).
ModeratemeancurrentswerealsoobservednorthoftheBI,however,ingeneral,velocitiesintheBIregionweremuchlowerthanintheAlboranSea,andalsolowerthanintheLoopCurrentareaofinfluenceintheGOM.
LarvalhabitatmodelsOnaweeklybasis,theproportionofsampledsta-tionswith>0BFTlarvaeincreasedwithincreasingsurfacetemperatureswithinbothstudyregions(Fig.
4).
OccurrencesoflarvaeintheGOMincreasedfromaround5%inmidAprilto35%towardstheendofMay.
Samplesfrom1998to2009wereavailableonlyuntiltheendofMay,howevertheproportionofpositivestationsfromhistoricaldata(1982to2009)includedsomeJunestations,whicharealsoshowninFig.
4.
Duringthesampledtimeperiod,surfacetem-peraturesfromCTDcastsatsampledstationsin-creasedfromameanof~24.
5°Cto27°C,althoughvariabilityaroundthesemeanswashigh.
IntheBIregion,theproportionofpositivestationsincreasedfrom~5%inearly-midJunetoapeakof20to25%bythefirstweekofJuly.
Larvaloccurrencesthendecreasedbacktoaround5%bytheendofJuly.
Duringthis8wkperiod,meansurfacetemperaturesatsampledstationsincreasedfromlessthan21°Ctogreaterthan27°C.
SpawningintheBIregionthustendedtobeginatcoolertemperaturesthanintheGOM,buttherateoftemperatureincreasethroughthesamplingperiodwashigher(Fig.
4).
PCAanalysisofenvironmentalconditionsateachsampledstationineachregionshowedageneralseparationofwatermassalongthefirstprincipalcomponent,PC1(Fig.
5).
IntheGOM,stationswithnegativevaluesalongPC1hadhighertemperaturesat100and200mdepth,highersalinityat200m,highergeostrophicvelocitiesandlowsurfacechloro-phyll.
ThesestationswereprobablyassociatedwiththeLoopCurrentorwarmringsofLoopCurrentori-gin.
Alongthesecondprincipalcomponent,PC2,sta-tionswithpositivevaluesweresampledlaterinthe264Fig.
3.
Monthlyclimatologiesfor6environmentalvariablesinthewesternMediterraneanSeaduringJuly.
Temperatureandsalinityatthesurfaceandat100mdepthwerederivedfromtheHYbridCoordinateOceanModel(HYCOM)+NavyCoupledOceanDataAssimilation(NCODA)Global1/12DegreeAnalysis(GLBa0.
08),2003-2012.
GeostrophiccurrentvelocitieswereobtainedfromAvisoaltimetrymeanabsolutedynamictopographyvalues.
SurfacechlorophyllwasderivedfromModerateReso-lutionImagingSpectroradiometer(MODIS)-Aqua8dcomposites,2003to2012.
AschematicofinflowingAtlanticOceanwater,theNorthernCurrentalongtheFrenchcoast(blackarrows)andregionsoftypicalmesoscaleeddyactivity(dashedblackarrows)arealsoshown(Regleroetal.
2012)Muhlingetal.
:Larvalbluefintunahabitatcomparisonyear,andshowedhighersurfacetempera-tures.
IntheBI,stationswithpositiveva-luesalongPC1hadhighertemperaturesatdepth,andlowersalinitiesthroughoutthewatercolumn.
Thesestationswereprob-ablyassociatedwithinflowingAtlanticwater.
AlongPC2,stationswithpositivevaluesweresampledlaterintheyear,andshowedhighersurfacetemperaturesalongwithlowersurfacechlorophyll(Fig.
5).
Inbothregions,6principalcomponentaxesencompassedmorethan90%ofthesamplevariability,andsothefirst6axesfromeachPCAanalysiswereinputtoneuralnetworkmodels(Table3).
IntheGOM,PC2wasthemostimportantvari-able,suggestingastronginfluenceofdateandsurfacetemperature.
IntheBI,PC1wasthemostinfluentialpredictor,im-plyingthatwatermasswasthemostimpor-tantconsiderationwhenpredictinglarvaloccurrences.
IntheGOM,thestronginflu-enceofdayoftheyearwasapparent,withgenerallyincreasingprobabilitiesofoccur-rencethroughtime.
Thistrendwasalsoevidentfromahistogramofpositivesta-tionsthroughtimefromtherawdata(Fig.
6).
Predictedandactualprobabilitiesoflarvaloccurrencedecreasedwithin-creasingsalinityandtemperatureatdepth,suggestingloweroccurrenceswithinLoopCurrentwaters.
Therelationshipbetweenlarvaloccurrencesandsurfacetempera-tureswasparabolic,withmaximumprob-abilitiesofoccurrencebetween25and27°C.
IntheBI,dayoftheyearwasalso2651920212223242526272829051015202530Apr19Apr26May3May10May17May24May31June7June14June21June28July5July12July19July26August2Surfacetemperature(°C)Proportionpositivestations(%)CurrentGOMHistoricalGOMCurrentBISurfacetemperatureFig.
4.
Proportionofstationswith>0bluefintunalarvaebyweekfortheGulfofMexico(GOM,1998to2009)andBalearicIslands(BI,2001to2005).
Proportionswerecalculatedforallstationssampledduringeachweek,forallsampledyearstogether.
Graydashedseriesshowsthepropor-tionofpositivestationsbyweekfortheGOMfortheyears1982to2009,toshowresultsfromsomeJunesam-plescollectedinthemid1990s.
Mean(±SD)surfacetemperaturesfromCTDcaststakenatsampledstationsarealsoshown.
r:GOM;e:BIGOM–6–4–20246PC1(36.
6%oftotalvariation)-4–2024SSTSSST100mS100mT200mS200mDateTimeChlGeoVelBI–50PC2(15.
2%oftotalvariation)510–50PC1(31.
1%oftotalvariation)5DateTimeSSTSSST100S100T200S200ChlGeoVelPC2(19.
7%oftotalvariation)Fig.
5.
Principalcomponent(PC)analysison10environmentalvariablesfromtheGulfofMexico(GOM)andBalearicIslands(BI).
Vectorsofeachvariableareshown,whereTistemperature,Sissalinity,ChlischlorophyllaandGeoVelisgeostrophicvelocity.
Thefirst2principalcomponentaxesareshown,alongwiththepercentoftotalvariationexplainedbyeachaxisMarEcolProgSer486:257–276,2013influential,withmodeledandactualprobabilitiesofoccurrenceincreasingthroughtoearlyJuly,beforedecreasingagain.
Larvaewerealsomostlikelytobecollectedatstationswithlowsurfacechlorophyll(36.
4vs.
3437).
Therangeoftemperaturesat200mdepthvariedtoamuchgreaterdegreeintheGOM(~12°C)thanintheBI(~1°C).
Rangesofsurfacechlorophyllweresimilar.
However,althoughprobabilitiesofoccurrenceintheBIwereproportion-allygreateratlowchlorophyllvalues(0bluefintunalarvaearealsoshowninsolidblackdots,forbinnedenvironmentalvariables.
NotenonlinearscalesforsurfacechlorophyllandgeostrophicvelocitiesMarEcolProgSer486:257–276,20132680–1010203040506018202224262830010203040506036.
537.
53738.
538010203040506012.
51313.
514.
51415010203040506037.
437.
637.
83838.
238.
638.
4010203040506012.
513.
51314010203040506037.
93838.
138.
238.
338.
438.
538.
6010203040506026-May10-Jun25-Jun10-Jul25-Jul9-Aug0102030405060010203040506001020304050600.
090.
170.
310.
0406:0012:0018:000:0024:000.
030.
10.
301.
0R2=0.
17R2=0.
16R2=0.
09R2=0.
08R2=0.
16R2=0.
02R2=0.
05R2=0.
24R2=0.
00R2=0.
20Seasurfacetemperature(°C)SeasurfacesalinityTemperature100m(°C)Salinity100mProbabilityoflarvaloccurrence(%)Probabilityoflarvaloccurrence(%)Temperature200m(°C)Salinity200mDateTimeSurfacechlorophyll(mgm–3)GeostrophicVelocity(ms–1)Fig.
7.
ScatterplotsofenvironmentalvariablesandpredictedprobabilitiesofoccurrencefromtheneuralnetworkmodelfortheBalearicIslands.
Thegeneralnatureoftherelationshipbetweenlarvaloccurrencesandeachvariableareshownusingpoly-nomiallinesofbestfit(blackline).
Observedproportionsofstationswith>0bluefintunalarvaearealsoshowninsolidblackdots,forbinnedenvironmentalvariables.
NotenonlinearscalesforsurfacechlorophyllandgeostrophicvelocitiesMuhlingetal.
:LarvalbluefintunahabitatcomparisonWhenresultsofhabitatmodelswereappliedto2exampleyearsintheGOM(1999and2006),mostpositivestationswereplacedintofavorablehabitat(Fig.
9).
Twopositivestationsfrom2006wereinlessfavorablehabitatneartheedgesoftheLoopCurrentandawarmeddy.
Theeffectofmesoscalefeaturesonpredictedprobabilitieswasevidentfromplotsoftemperatureat100mdepthandgeostrophicvelocity.
BFTlarvaeweregenerallycollectedoutsideoftheLoopCurrent(indicatedbytemperaturesatdepth0.
15mgm3)inboth2004and2005.
ContoursofsurfacesalinityshowedthedelineationbetweenlowersalinityAtlanticwaterinthesouthofthestudyareaandhighersalinityresidentwaterinthenorth.
Inbothyears,larvaeweremostabundantinmoder-atesalinities(37.
2to37.
8),presumablyclosetowatermassboundaries.
Interestingly,surfacechlorophyllvalueswereataminimuminthevicinityofthefrontalzone.
DISCUSSIONResultsfromthisstudygenerallyagreedwellwithpreviouspublishedresearch,suggestingthatBFTspawninwarm,oligotrophic,offshoreareaswithintheGOMandwesternMediterraneanSea(Blocketal.
2005,Teoetal.
2007,Alemanyetal.
2010,Muh-lingetal.
2010,2012,Regleroetal.
2012).
Spawninginbothareasappearedtocommencewhensurfacetemperatureswereinthelow20s(°C),peakaroundthemid-20s,andtailoffatwarmertemperatures,althoughsamplingrestrictionsintheGOMmadedeterminationoftheendofthespawningseason269Prob.
ofoccurrenceSt.
DeviationSt.
DeviationProbabilityofoccurrence(/1)Probabilityofoccurrence(/1)0.
40.
30.
20.
122Surfacetemperature(°C)Surfacetemperature(°C)Surfacetemperature(°C)Temperatureat100m(°C)2324252627282927262524232221201918170.
40.
30.
20.
10.
30.
20.
10.
30.
20.
1221920212223242526272819202122232425262728Surfacetemperature(°C)Temperatureat100m(°C)SurfacesalinitySurfacesalinity23242526272829272625242322212019183837.
837.
637.
437.
23736.
836.
63837.
837.
637.
437.
23736.
836.
6StandardDeviationStandardDeviationGOMBI0.
40.
30.
20.
10.
0750.
050.
02500.
10.
0750.
050.
02500.
30.
250.
20.
150.
1Fig.
8.
Wireframeplotsofmeanmodeledprobabilitiesoflarvalbluefintunaoccurrencewithsurfacetemperatureandtemper-atureat100mintheGulfofMexico(left),andsurfacetemperatureandsalinityintheBalearicIslands(right).
Gridpointswithlessthan4observationshavebeenblankedout.
Standarddeviationspergridpointarealsoshown(bottom)MarEcolProgSer486:257–276,2013therelesscertain.
Bytheendofthespawningseason,afterapproximately6to8wk,surfacetemperaturesapproachphysiologicallystressfulvaluesforadultBFT.
Atthistime,adultspresumablyleavetheirspawningareasandmigratebacktocoolerfeedinggroundsintheNorthAtlantic(Blocketal.
2005).
TemperaturestressmaybemoreofafactorintheGOM,wheresurfacetemperaturesduringsummermonthsreach29to31°C,withnocoolerwatersnear-by(Muller-Kargeretal.
1991).
AsadultBFTshowrelativelyshallowdivingbehaviorintheGOM(Teoetal.
2007)andrarelycrossthethermocline(Wallietal.
2009),theirabilitytoavoidverywarmwatersinlatespringandsummerislimited.
Incontrast,theMediterraneanSeastayscoolerduringthesummer,particularlyintheBIregion.
Althoughourstudydidnotexamineresidencetimeonspawninggrounds,thisislikelytobevariable,withadultfisharrivingandleavingthroughoutthespawningseason(Farley&Davis1998,Blocketal.
2001,Galuardietal.
2010).
AdirectcomparisonofenvironmentalconditionsonbothAtlanticspawninggroundsrevealedseveralsimilaritiesbetweentheBIandGOMregions.
Inbothareas,larvaewerefoundinoligotrophicwaters,evenwherehigherproductivitywatermasseswereob-servedcloseby.
ThisrelationshipwasstrongerintheBIthanintheGOM,wherelowestregionalchloro-phyllisusuallywithintheLoopCurrent,whichadultBFTavoid(Teoetal.
2007).
AlthoughtheMedi-terraneanwasmuchmoresalinethantheGOM,lar-valdistributionsonbothspawninggroundsshowedrelationshipswithsalinity.
Thesewereshowntobeproxiesforwatermass:intheGOM,larvalBFTwererarelycollectedwithintheLoopCurrentorwarm2702006:Habitat(%)2006:Temperature100m(°C)1999:Habitat(%)1999:Temperature100m(°C)PositiveStation+NegativeStation1999:GeostrophicVelocity(ms–1)0.
010.
040.
100.
25200.
63304030°N28°26°24°98°96°94°92°90°88°86°84°82°W98°96°94°92°90°88°86°84°82°W98°96°94°92°90°88°86°84°82°W98°96°94°92°90°88°86°84°82°W98°96°94°92°90°88°86°84°82°W98°96°94°92°90°88°86°84°82°W30°N28°26°24°30°N28°26°24°30°N28°26°24°30°N28°26°24°30°N28°26°24°2725232119172006:GeostrophicVelocity(ms–1)Fig.
9.
Krigedpredictedprobabilitiesfromneuralnetworkmodelsoflarvalbluefintunaoccurrenceforleg2ofthe1999and2006cruisesintheGulfofMexico(top).
Allsampledstations,andlarvalbluefintunacatchlocationsareindicated.
Twoenvi-ronmentalvariableshighlightedasinfluentialinFig.
6arealsoshown:temperaturesat100mdepthin°C(middle),andgeo-strophicvelocitiesinms1(bottom),overlaidwithlarvalcatchlocations.
Notenonlinear(log10)scaleforgeostrophicvelocitiesMuhlingetal.
:LarvalbluefintunahabitatcomparisonLoopCurrenteddies,whichwerebestidentifiedbyelevatedsalinityandtemperatureatdepth.
IntheBI,larvaewereassociatedwithmoderatesurfacesalini-tiesaroundthefrontalzonewherelesssalineincom-ingAtlanticwatersmeetmoresalineresidentwaters.
InboththeGOMandBI,migratingadultBFTap-pearedtomovepastareasofveryhighcurrentveloc-ity,intheLoopCurrentandAlboranSearespec-tively,andconcentratespawninginregionsofmoderatecurrentflow(<0.
3ms1).
IntheGOM,thiscorrespondedtotheoffshorewatersoutsideofhighchlorophyllcontinentalshelfzonesandlargeener-geticeddyfeatures.
IntheBI,larvaewereconcen-tratednearafrontalzonewithlowsurfacechloro-phyllandnostrongdirectionalcurrentflow.
Althoughlarvaewerefoundin1to2°CcoolerwatersaroundtheBIthanintheGOM,itappearslikelythattemperatureisanimportantphysiologicalcontrolintheinitiationofspawning.
IntheAtlantic,PacificandIndianoceans,temperaturehasbeenfoundtobeadrivingforcebehindspawningactivity(Yukinawa1987,Miyashitaetal.
2000,Medinaetal.
2712005:Habitat(%)2004:Habitat(%)2004:Surfacechlorophyll(mgm–3)2004:Surfacesalinity2005:Surfacechlorophyll(mgm–3)2005:Surfacesalinity100.
220.
1950.
170.
1450.
120.
0950.
0738.
237.
937.
637.
33736.
710304040°N39°38°0.
5°1.
5°4.
5°E2.
5°3.
5°0.
5°1.
5°4.
5°E2.
5°3.
5°0.
5°1.
5°4.
5°E2.
5°3.
5°0.
5°1.
5°4.
5°E2.
5°3.
5°0.
5°1.
5°4.
5°E2.
5°3.
5°0.
5°1.
5°4.
5°E2.
5°3.
5°40°N39°38°40°N39°38°40°N39°38°40°N39°38°40°N39°38°PositiveStation+NegativeStationFig.
10.
Krigedpredictedprobabilitiesfromneuralnetworkmodelsoflarvalbluefintunaoccurrenceforthe2004and2005cruisesintheBalearicIslands(top).
Allsampledstations,andlarvalbluefintunacatchlocationsareindicated.
Twoenviron-mentalvariableshighlightedasinfluentialinFig.
6arealsoshown:surfacechlorophyllinmgm3(middle),andsurfacesalinity(bottom),overlaidwithlarvalcatchlocationsMarEcolProgSer486:257–276,20132002,Blocketal.
2005,Tanakaetal.
2006,Alemanyetal.
2010)andadultphysiology(Carey&Teal1966,Blanketal.
2004).
Medinaetal.
(2002)suggestedthatwatertemperaturetriggersBFTspawningintherelativelyshortdistancebetweentheStraitsofGibraltarandtheBI,althoughitremainsunclearwhethertheabsolutetemperatureortherateoftem-peratureincreaseismoreimportant.
AdultBFTincaptivityhavealsobeenobservedtospawnoncewatertemperaturesreach~22°C(Sawadaetal.
2005).
LarvalcollectionsinboththeAtlanticandPacificoceanshavemostlybeenrestrictedtowatersbetween~22and28°C(Masumaetal.
2006,Tanakaetal.
2006,Alemanyetal.
2010,Muhlingetal.
2010).
WithintheMediterraneanSea,spawningtakesplaceearlierintheeasternbasin,wherewaterswarmtosuitabletemperaturesearlierintheyear(Oray&Karakulak2005).
Egghatchingandlarvaldevelop-mentarealsoinfluencedbytemperature,witharound25°Csuggestedtobeoptimal(Miyashitaetal.
2000).
Fewstudiesexistontheeffectsoftempera-tureonlarvaltunagrowth,withsomestudiesfindingnoassociation(Jenkinsetal.
1991,Tanakaetal.
2006)andothersapositiveassociation(Regleroetal.
2011,Garciaetal.
2013).
Itmaybethataboveapar-ticulartemperaturethreshold,larvaegrowequallywell,orthatstudiescomparingtemperaturesacrossabroadenoughrangearelacking.
Conversely,itispossiblethatconcentrationsofpreyitemstypicalofwarm,oligotrophicwatersaremoreinfluentialindetermininggrowthandsurvivalthantemperatureitself.
Whilemoderatelywarmwatertemperaturesappeartoberequiredforspawning,thespatiallyandtemporallyrestrictednatureofspawninggroundsandlarvaloccurrencessuggestthatotherbiophysicalfactorsarealsoimportant.
StudiesconcerningthedietsoflarvalBFTarerare,buttheyappearlikelytofeedoncopepods,copepodnauplii,cladoceransandappendiculariansatpre-flexionstages,beforebecomingmorepiscivorouspostflexion(Uotanietal.
1990,Llopizetal.
2010,Catalánetal.
2011,Regleroetal.
2011).
Theadvan-tages(ifany)forlarvaebeingspawnedintooligo-trophicregionsremainunclear,althoughsomehypo-theseshavebeenproposed.
Bakun(2012)suggestedthatpredationratesonlarvalBFTmaybelowerinlessproductiveareas,andthattheveryhighfecun-dityofadultBFT(Medinaetal.
2002)mayenabletheoverwhelmingandsatiationofresidentplanktonicpredatorsbysheernumbersofspawnedlarvae.
Oceanographicfeaturessuchaseddieshavebeenproposedasoptimalspawningareas,providinglocallyenhancedfeedingopportunitiesandsufficientconcentrationsoftunalarvaetoenablesuccessfulpiscivory(Bakun2012).
Inaddition,spawninggroundsmaybecharacterizedbyoptimaltransportconditionsforspawnedlarvaetonurseryareas(Kita-gawaetal.
2010).
ResultsfromthisstudyfoundthatlarvaldensitiesintheGOMwerelowerinanti-cycloniceddies,andinboththeGOMandBI,larvaewerenotassociatedwithstrong,directionalcurrentflows.
AlthoughoceanographicconditionsonBFTspawninggroundsintheGOM,BIandthePacificOceanshowseveralsimilarities,itwouldseemthatadultBFThavemodifiedtheirspawningstrategiestobestexploitlocalconditions.
Whileentrainmentinstrong,directionalflowsmaybebeneficialforlarvaeifitfacilitatesdeliverytonurseryareas(Kitagawaetal.
2010),entrainmentintheLoopCurrent(forexam-ple)wouldresultintherapidnorthwardstransportoflarvaeintheGulfStream,awayfromassumednurseryareasintheGOM(Brothersetal.
1983).
Similarly,whilesomemesoscaleeddyfeaturespro-videenhancedfeedingandretentionconditionsforpelagiclarvae(Bakun2012),warmanticyclonesmaybemuchlessbeneficialthancoolercyclonicfea-tures.
ThecomplexmesoscaleeddyfieldsinthewesternMediterranean,resultingfrominflowingAtlanticwaters,contrastwiththeperiodicsheddingoflarge,warmanticycloniceddiesintheGOM(Mil-lot,1999,Lindo-Atichatietal.
2012),andbothhavedifferentimplicationsforlarvalfeeding,transportandsurvival.
LarvalBFTonbothsidesoftheAtlanticappeartogrowataround0.
3to0.
4mmperdayduringearlylife(Scottetal.
1993,Garciaetal.
2013),withapossibleaccelerationofgrowthassociatedwiththeonsetofpiscivory(Kajietal.
1996,Tanakaetal.
2012).
Somestudiessuggestthatthefastest-growinglarvaearemostlikelytoultimatelysurvive,andthereforelarvaewhichencounterconditionsmostconducivetorapidgrowthmaybeatanadvantage(Brothersetal.
1983,Tanakaetal.
2006).
However,larvaeinhigh-densitypatchesmaygrowmoreslowlyduetodensity-dependentcompetitionforprey(Jenkinsetal.
1991).
Improvedunderstandingofecologicalprocessesgov-erninglarvalBFTgrowthandsurvivalisclearlyre-quiredbeforegeneralizationscanbeapplied.
BFTlarvaeinboththeGOMandBIweremostlikelytobecollectedinregionsofmoderatemeso-scaleoceanographicactivityandeddygeneration,butnotinareasofstrong,directionalflow.
IntheBI,thesalinityfrontbetweenAtlanticandMediter-raneanwatersappearedmostfavorable.
Thestron-gestfrontsintheGOMareassociatedwiththeLoopCurrent,andithasbeensuggestedpreviouslythat272Muhlingetal.
:Larvalbluefintunahabitatcomparisonthisregionmaybeafocalpointforspawningactivity(Richardsetal.
1989).
However,adoptionofanew,shallowersamplinggearintheGOMin2010allowedimprovedassessmentofthedistributionofverysmallBFTlarvae,andresultsshowedthattheywerefoundacrossthenorthernGOM(Muhlingetal.
2012).
WhileitispossiblethatspawningintheGOMisassociatedwithfiner-scalefronts,eddiesorotheroceanographicfeatures,thecoarsenatureoftheSoutheastAreaMonitoringandAssessmentProgramsamplinggridhasnotallowedthoroughinvestiga-tionofthis.
Thehabitatmodelsdescribedinthisstudyareaffectedbyseverallagsintimeandspace,asaresultoftemporalaveragingandspatialinter-polationofpredictorvariables,larvalagesandlarvalpatchiness.
Larvaecollectedareprobablyseveraldaystomorethanaweekold,implyingconsiderabledriftsincespawning.
Inaddition,thescalesofsam-plingarealsolikelytobemuchcoarserthanscalesoflarvalpatchiness.
Allocationofspawningtofiner-scaleoceanographicfeaturesisthereforenotpossi-bleatthisstage,andwillprobablyrequireamodifiedsamplingdesignforbothlarvaeandenvironmentalvariables.
Althoughthisstudyconcentratedon2well-knownmajorspawninggroundsforBFT,larvaehavebeencollectedinotherlocations.
InthewesternAtlantic,atleastsomespawningactivityappearstotakeplaceinthesouthernGOMandfarwestCaribbeanSea,althoughonlyverysmallnumbersoflarvaehavebeencollectedtodate(Olvera-Limasetal.
1988,Muhlingetal.
2011,NationalOceanicandAtmo-sphericAdministrationNationalMarineFisheriesServiceunpubl.
data).
TheimportanceofthewiderCaribbeanSeaasaspawningarearemainsuncer-tain,althoughtaggedfishhavenotbeenreportedtotravelsouthofCubaontheirwaytotheGOM.
PartsofthewesternCaribbeanarealsowarmtoagreaterdepththanthenorthernGOM(Gordon1967),whichmayinhibittheabilityofadultfishtothermoregulatebydiving.
IntheeasternAtlantic,BFTlarvaehavebeencollectedinthefareasternMediterraneanSeanearCyprusduringJune(Oray&Kurakulak2005)andinthecentralMediterraneanduringJuly(Tsujietal.
1997,Kochedetal.
2012).
TaggedadultBFThavebeenshowntomigratefromtheopenAtlanticintotheMediterraneanasfarasSicily(Blocketal.
2005),butmigrationpatternsoffishspawninginthefareasternMediterraneanarelesswellknown.
TheymaynottravelouttotheAtlanticOceanafterspawn-ing(DeMetrioetal.
2004),butinsteadcomprisearesidentsubpopulation.
Druonetal.
(2011)usedremotelysensedsurfacetemperaturesandchloro-phyllvaluestomappotentialspawningandfeedinghabitatforBFTintheentireMediterraneanSea.
Pre-dictedsuitablespawninglocationsfromthisstudygenerallycoveredagreaterspatialextentthancon-firmedspawningareas.
Thissuggeststhateitheradditionalenvironmentalvariablesareimportantindeterminingactualspawninglocationsorthataddi-tionalunconfirmedspawningareasexist.
Overall,thiscomparativestudyofenvironmentalconditionsonBFTspawninggroundsintheGOMandBIsuggestedseveralstrongsimilarities,includingapreferenceforlowchlorophyll(<0.
2mgm3),warmwaters(24to27°C)withmoderatecurrentflows(<0.
3ms1)andhighoceanographiccomplexity.
De-spiteregionaldifferencesinwaterproperties,observedrelationshipswithsalinityandwithtemperaturesatdepthwereusefulindefiningwatermassestargetedforspawning.
However,toelucidatetheecologicalimplicationsoflarvaespawnedintotheseconditions,furtherresearchofadultreproductivesystems,larvaldiets,growth,behaviorandstructureofplanktonicfoodwebsisclearlyrequired.
Whilemodelingandconceptualstudiesarevaluable,dataonbasiclarvalprocessesarelacking.
Inaddition,identificationofactualspawninglocationsthroughcollectionsofeggs,larvalbacktrackingstudiesorobservationsofspawningadultswouldbeofgreatbenefitindeter-miningthepreciseconditionsrequiredforspawning.
Nevertheless,theworkdescribedherecomprisesanimportantfirststepinthesearchforunderstandingthespawningecologyofBFT.
Acknowledgements.
IntheBalearicIslands,initialworkwasfundedbytheBALEARESproject(CTM2009-07944MAR)andATAME,2011-29525-004-02,withbasedataprovidedbytheTUNIBALproject(REN2003-01176),whichwerecompetitiveprojectsoftheSpanishGovernmentI+D+I(Research+Development+Innovation)NationalPlan.
WealsoexpressourgratitudetothecrewoftheRV'VizcondedeEza'(2001to2002)andRV'CornidedeSaavedra'(2003to2005),andtoallthescientificstaffthatparticipatedinsurveysandcarriedoutlabanalysis,especiallyD.
Cortés,T.
Ramírez,M.
FernándezdePuelles,J.
Jansà,P.
Vélez-Belchi,J.
M.
RodríguezandC.
GonzálezPola.
SpecialthanksarealsoextendedtoA.
GarcíaandJ.
L.
LópezJurado,fortheircrucialroleinthedevelopmentoftheTUNIBALproject.
TheauthorsalsothankstaffattheSeaFisheriesInstitutePlank-tonSortingandIdentificationCenter,GdyniaandSzczecin,Poland,includingM.
KoniecznaandL.
Ejsymont,forpro-cessingofplanktonsamplesfromtheGulfofMexico.
WealsoextendourgratitudetoallthecaptainsandcrewofalltheNOAAshipswhocollecteddataonSEAMAPplanktoncruises.
StaffattheEarlyLifeHistorylaboratoryofNOAAinMiamiarethankedforcollaboration,adviceanddatapro-cessing,inparticularW.
J.
Richards.
J.
RobertsandtheMarineGeospatialEcologyToolboxatDukeUniversitywereimmenselyhelpfulforprocessingremotelysensed273MarEcolProgSer486:257–276,2013environmentaldata.
L.
C.
acknowledgesfurthersupportfromNSF-CMGgrant0934961.
B.
M.
acknowledgestheNASAClimateandBiologicalResponse:ResearchandApplicationsprogram.
D.
A.
wassupportedbytheICTS-SOCIB(focusresearchprogrambluefintuna).
LITERATURECITEDAlemanyF,GarciaA,Gonzalez-PolaC,JansaJ,LopezJuradoJL,RodriguezJM,VelezBelchiP(2010)Atlanticbluefintunaandrelatedspeciesspawninghabitatchar-acterization:influenceofenvironmentalfactorsonlarvalabundanceanddistributionoffBalearicarchipelago(westernMediterranean).
ProgOceanogr86:2138AzovY(1991)EasternMediterranean—amarinedesertMarPollutBull23:225232BakunA(2012)Oceaneddies,predatorpitsandbluefintuna:implicationsofaninferred'lowrisklimitedpayoff'reproductiveschemeofa(former)archetypicaltoppredator.
FishFish,doi:10.
1111/faf.
12002BlankJM,MorrissetteJM,Landeira-FerandezAM,Black-wellSB,WilliamsTD,BlockBA(2004)InsitucardiacperformanceofPacificbluefintunaheartsinresponsetoacutetemperaturechange.
JExpBiol207:881890BlockBA,DewarH,BlackwellSB,WilliamsTDandothers(2001)Migratorymovements,depthpreferencesandthermalbiologyofAtlanticbluefintuna.
Science293:13101314BlockBA,TeoSLH,WalliA,BoustanyAandothers(2005)ElectronictaggingandpopulationstructureofAtlanticbluefintuna.
Nature434:11211127BrothersEB,PrinceED,LeeDW(1983)Ageandgrowthofyoung-of-the-yearbluefintuna,Thunnusthynnus,fromotolithmicrostructure.
In:PrinceED,PulosLM(eds)Pro-ceedingsoftheinternationalworkshoponagedetermi-nationofoceanicpelagicfishes:tunas,billfishesandsharks.
NOAATechRepNMFS8:4959CareyFG,TealJM(1966)Heatconservationintunafishmuscle.
ProcNatlAcadSciUSA56:14641469CastroLR,SalinasGR,HernándezEH(2000)EnvironmentalinfluencesonwinterspawningoftheanchovetaEngraulisringensoffcentralChile.
MarEcolProgSer197:247258CatalánIA,TejedorA,AlemanyF,RegleroP(2011)TrophicecologyofAtlanticbluefintunaThunnusthynnuslarvae.
JFishBiol78:15451560ChaseBC(2002)DifferencesindietofAtlanticbluefintuna(Thunnusthynnus)atfiveseasonalfeedinggroundsontheNewEnglandcontinentalshelf.
FishBull100:168180ClarkeKR,GorleyRN(2006)PRIMERv6:usermanual/tuto-rial.
PRIMER-E,PlymouthCortenA(2001)Theroleof'conservatism'inherringmigra-tions.
RevFishBiolFish11:339361CuryP(1994)Obstinatenature:anecologyofindividuals.
Thoughtsonreproductivebehaviorandbiodiversity.
CanJFishAquatSci51:16641673DeMetrioG,OrayI,ArnoldGP,LutcavageMandothers(2004)JointTurkish-ItalianresearchintheeasternMediterranean:bluefintunataggingwithpop-upsatel-litetags.
CollectVolSciPapICCAT56:11631167DruonJN,FromentinJM,FlorianA,JukkaH(2011)Poten-tialfeedingandspawninghabitatsofAtlanticbluefintunaintheMediterraneanSea.
MarEcolProgSer439:223240EstradaM(1996)PrimaryproductioninthenorthwesternMediterranean.
SciMar60:5564EstradaJA,LutcavageM,ThorroldSR(2005)DietandtrophicpositionofAtlanticbluefintuna(Thunnusthyn-nus)inferredfromstablecarbonandnitrogenisotopeanalysis.
MarBiol147:3745FarleyJH,DavisTLO(1998)Reproductivedynamicsofsouthernbluefintuna,Thunnusmaccoyii.
FishBull96:223236FromentinJM,PowersJE(2005)Atlanticbluefintuna:pop-ulationdynamics,ecology,fisheriesandmanagement.
FishFish6:281306GaluardiB,RoyerF,GoletW,LoganJ,NeilsonJ,LutcavageM(2010)ComplexmigrationroutesofAtlanticbluefintuna(Thunnusthynnus)questioncurrentpopulationstructureparadigm.
CanJFishAquatSci67:966976GarciaA,AlemanyF,Velez-BelchiP,LopezJuradoJLandothers(2005)Characterizationofthebluefintunaspawn-inghabitatofftheBalearicarchipelagoinrelationtokeyhydrographicfeaturesandassociatedenvironmentalconditions.
CollectVolSciPapICCAT58:535549GarciaA,CortesD,QuintanillaJ,RamirezT,QuintanillaL,RodriguezJM,AlemanyF(2013)Climate-induceden-vironmentalconditionsinfluencinginterannualvariabil-ityofMediterraneanbluefin(Thunnusthynnus)larvalgrowth.
FishOceanogr22:273–287GardnerMW,DorlingSR(1998)Artificialneuralnetworks(themultilayerperceptron)—areviewofapplicationsintheatmosphericsciences.
AtmosEnviron32:26272636GilbesF,TomasC,WalshJJ,Muller-KargerFE(1996)AnepisodicchlorophyllplumeontheWestFloridaShelf.
ContShelfRes16:12011207GordonAL(1967)CirculationoftheCaribbeanSea.
JGeo-physRes72:62076223,doi:10.
1029/JZ072i024p06207GrahamJB,DicksonKA(2004)Tunacomparativephysiol-ogy.
JExpBiol207:40154024IngramGWJr,RichardsWJ,LamkinJT,MuhlingBA(2010)AnnualindicesofAtlanticbluefintuna(Thunnusthyn-nus)larvaeintheGulfofMexicodevelopedusingdelta-lognormalandmultivariatemodels.
AquatLivingResour23:3547InternationalCommissionfortheConservationofAtlanticTunas(2011)Reportofthe2010Atlanticbluefintunaassessmentsession.
CollectVolSciPapICCAT66:505714JenkinsGP,YoungJW,DavisTLO(1991)Densitydepen-denceoflarvalgrowthofamarinefish,thesouthernbluefintuna,Thunnusmaccoyii.
CanJFishAquatSci48:13581363KajiT,TanakaM,TakahashiY,OkaM,IshibashiN(1996)PreliminaryobservationsondevelopmentofPacificbluefintunaThunnusthynnus(Scombridae)larvaerearedinthelaboratory,withspecialreferencetothedigestivesystem.
MarFreshwRes47:261269KitagawaT,KatoY,MillerMJ,SasaiY,SasakiH,KimuraS(2010)TherestrictedspawningareaandseasonofPacificbluefintunafacilitateuseofnurseryareas:amod-elingapproachtolarvalandjuveniledispersalprocesses.
JExpMarBiolEcol393:2331KochedW,HattourA,AlemanyF,GarciaA,SaidK(2012)SpatialdistributionoftunalarvaeintheGulfofGabes(easternMediterranean)inrelationwithenvironmentalparameters.
MeditMarSci41:514LandaetaMF,CastroLR(2002)SpringspawningandearlynurseryzoneofthemesopelagicfishMaurolicusparvi-274Muhlingetal.
:LarvalbluefintunahabitatcomparisonpinnisatthecoastalupwellingzoneoffTalcahuano,cen-tralChile.
MarEcolProgSer226:179191Lindo-AtichatiD,BringasF,GoniG,MuhlingB,Muller-KargerFE,HabtesS(2012)VaryingmesoscalestructuresinfluencelarvalfishdistributioninthenorthernGulfofMexico.
MarEcolProgSer463:245257LlopizJK,RichardsonDE,ShirozaA,SmithSL,CowenRK(2010)Distinctionsinthedietsanddistributionsoflarvaltunasandtheimportantroleofappendicularians.
LimnolOceanogr55:983996MalmgrenBA,WinterA(1999)ClimatezonationinPuertoRicobasedonprincipalcomponentsanalysisandanarti-ficialneuralnetwork.
JClim12:977985MasumaS,TezukaN,KoisoM,JinboTandothers(2006)Effectsofwatertemperatureonbluefintunaspawningbiologyincaptivity.
BullFishResAgency4(Suppl):157172MedinaA,AbascalFJ,MeginaC,GarciaA(2002)Stereo-logicalassessmentofthereproductivestatusoffemaleAtlanticnorthernbluefintunaduringmigrationtoMedi-terraneanspawninggroundsthroughtheStraitofGib-raltar.
JFishBiol60:203217MillotC(1999)CirculationinthewesternMediterraneanSea.
JMarSyst20:423442MiyashitaS,YujiT,YoshifumiS,OsamuM,NobuhiroH,KenjiT,ToshioM(2000)Embryonicdevelopmentandeffectsofwatertemperatureonhatchingofthebluefintuna,Thunnusthynnus.
SuisanZoshoku48:199207MotosL,UriateA,ValenciaV(1996)Thespawningenviron-mentoftheBayofBiscayanchovy(Engraulisencrasico-lusL.
).
SciMar60:117140MuhlingBA,LamkinJT,RofferMA(2010)Predictingtheoccurrenceofbluefintuna(Thunnusthynnus)larvaeinthenorthernGulfofMexico:buildingaclassificationmodelfromarchivaldata.
FishOceanogr19:526539MuhlingBA,LamkinJT,QuattroJM,SmithRH,RobertsMA,RofferMA,RamirezK(2011)Collectionoflarvalbluefintuna(Thunnusthynnus)outsidedocumentedwesternAtlanticspawninggrounds.
BullMarSci87:687694MuhlingBA,RofferMA,LamkinJT,IngramGWJrandothers(2012)OverlapbetweenAtlanticbluentunaspawninggroundsandobservedDeepwaterHorizonsurfaceoilinthenorthernGulfofMexico.
MarPollutBull64:679687Muller-KargerFE,WalshJJ,EvansRH,MeyersMB(1991)OntheseasonalphytoplanktonconcentrationandseasurfacetemperaturecyclesoftheGulfofMexicoasde-terminedbysatellites.
JGeophysResOceans96:12645–12665NeumannDE(2002)Anenhancedneuralnetworktech-niqueforsoftwareriskanalysis.
IEEETransSoftwEng28:904912OldenJD,LawlerJJ,PoffNL(2008)Machinelearningmeth-odswithouttears:aprimerforecologists.
QRevBiol83:171193OlveraLimasR,CerecedoJL,CompeanGA(1988)Distribu-ciondelarvasdetunidosenelGolfodeMexicoyMarCaribe:abundanciaybiomasadetresspeciesenlazonaeconomicaexclusive.
CiencPesq6:119140OrayIK,KarakulakFS(2005)Furtherevidenceofspawningofbluefintuna(ThunnusthynnusL.
,1758)andthetunaspecies(AuxisrocheiRis.
,1810,EuthynnusalletteratusRaf.
,1810)intheeasternMediterraneanSea:prelimi-naryresultsofTUNALEVlarvalsurveyin2004.
JApplIchthyology21:236240RegleroP,UrtizbereaA,TorresAP,AlemanyF,FiksenO(2011)Cannibalismamongsizeclassesoflarvaemaybeasubstantialmortalitycomponentintuna.
MarEcolProgSer433:205219RegleroP,CianelliL,Alvarez-BerasteguiD,BalbínR,Ló-pez-JuradoJL,AlemanyF(2012)GeographicallyandenvironmentallydrivenspawningdistributionsoftunaspeciesinthewesternMediterraneanSea.
MarEcolProgSer463:273284RichardsWJ,LemingT,McGowanMF,LamkinJT,Kelley-FragaS(1989)Distributionoffishlarvaeinrelationtohy-drographicfeaturesoftheLoopCurrentboundaryintheGulfofMexico.
RappConsIntExplorMer191:169176RobertsJJ,BestBD,DunnDC,TremlEA,HalpinPN(2010)MarineGeospatialEcologyTools:anintegratedframe-workforecologicalgeoprocessingwithArcGIS,Python,R,MATLAB,andC++.
EnvironModelSoftw25:11971207RookerJR,BremerJRA,BlockBA,DewarHandothers(2007)LifehistoryandstockstructureofAtlanticbluefintuna(Thunnusthynnus).
RevFishSci15:265310RookerJR,SecorDH,DeMetrioG,SchloesserR,BlockBA,NeilsonJD(2008)NatalhomingandconnectivityinAtlanticbluefintunapopulations.
Science322:742744RoyC(1998)Anupwelling-inducedretentionareaoffSene-gal:amechanismtolinkupwellingandretentionpro-cesses.
SAfrJMarSci19:8998SawadaY,OkadaT,MiyashitaS,MurataO,KumaiH(2005)CompletionofthePacificbluefintunaThunnusorientalis(TemmincketSchlegel)lifecycle.
AquacultRes36:413421SchaeferKM(2001)Reproductivebiologyoftunas.
In:BlockBA,StevensED(eds)Tuna:physiology,ecology,andevolution.
AcademicPress,SanDiego,CA,p225270ScottGP,TurnerSC,GrimesCB,RichardsWJ,BrothersEB(1993)Indicesoflarvalbluefintuna,Thunnusthynnus,abundanceintheGulfofMexico:modelingvariabilityingrowth,mortality,andgearselectivity:ichthyoplanktonmethodsforestimatingfishbiomass.
BullMarSci53:912929SeguradoP,AraujoMB(2004)Anevaluationofmethodsformodellingspeciesdistributions.
JBiogeogr31:15551568ShanksAL,EckertGL(2005)PopulationpersistenceofCali-forniacurrentfishesandbenthiccrustaceans:amarinedriftparadox.
EcolMonogr75:505524ShermanK,SmithW,MorseW,BermanM,GreenJ,Ejsy-montL(1984)Spawningstrategiesoffishesinrelationtocirculation,phytoplanktonproduction,andpulsesinzoo-planktonoffthenortheasternUnitedStates.
MarEcolProgSer18:119SherrodPH(2003)DTREG:classificationandregressiontreesfordataminingandmodeling.
Availableatwww.
dtreg.
com/DTREG.
pdf,accessedSeptember2012SimsDW,WearmouthVJ,GennerMJ,SouthwardAJ,HawkinsSJ(2004)Low-temperature-drivenearlyspaw-ningmigrationofatemperatemarinefish.
JAnimEcol73:333341SinclairM,IlesTD(1985)Atlanticherring(Clupeaharen-gus)distributionsintheGulfofMaine—ScotianShelfareainrelationtooceanographicfeatures.
CanJFishAquatSci42:880887SousaSIV,MartinsFG,Alvim-FerrazMCM,PereiraMC(2007)Multiplelinearregressionandartificialneuralnetworksbasedonprincipalcomponentstopredictozoneconcentrations.
EnvironModelSoftw22:97103275MarEcolProgSer486:257–276,2013StokesburyMJW,TeoSLH,SeitzA,O'DorRK,BlockBA(2004)MovementofAtlanticbluefintuna(Thunnusthynnus)asdeterminedbysatellitetaggingexperimentsinitiatedoffNewEngland.
CanJFishAquatSci61:19761987TanakaY,SatohK,IwahashiM,YamadaH(2006)Growth-dependentrecruitmentofPacificbluefintunaThunnusorientalisinthenorthwesternPacificOcean.
MarEcolProgSer319:225235TanakaY,MinamiH,IshihiY,KumonKandothers(2012)Relationshipbetweenpreyutilizationandgrowthvaria-tioninhatchery-rearedPacificbluefintuna,Thunnusori-entalis(TemmincketSchlegel),larvaeestimatedusingnitrogenstableisotopeanalysis.
AquacultRes,doi:10.
1111/j.
1365-2109.
2012.
03258.
xTeoSLH,BoustanyAM,DewarH,StokesburyMJWandothers(2007)Annualmigrations,divingbehavior,andthermalbiologyofAtlanticbluefintuna,Thunnusthyn-nus,ontheirGulfofMexicobreedinggrounds.
MarBiol151:118TorresAP,RegleroP,BalbinR,UrtizbereaA,AlemanyF(2011)CoexistenceoflarvaeoftunaspeciesandotherfishinthesurfacemixedlayerintheNWMediterranean.
JPlanktonRes33:17931812TsujiS,NishikawaY,SegawaK,HiroeY(1997)DistributionandabundanceofThunnuslarvaeandtheirrelationtotheoceanographicconditionintheGulfofMexicoandtheMediterraneanSeaduringMaythroughAugustof1994(draft).
CollectVolSciPapICCAT46:161176UotaniI,SaitoT,HiranumaK,NishikawaY(1990)FeedinghabitofbluefintunaThunnusthynnuslarvaeinthewesternNorthPacificOcean.
NipponSuisanGakk56:713717WalczakS,CerpaN(1999)Heuristicprinciplesforthedesignofartificialneuralnetworks.
InfSoftwTechnol41:107117WalliA,TeoSLH,BoustanyA,FarwellCJandothers(2009)Seasonalmovements,aggregationsanddivingbehaviorofAtlanticbluefintuna(Thunnusthynnus)revealedwitharchivaltags.
PLoSONE4:e6151WilsonSG,LutcavageME,BrillRW,GenoveseMP,CooperAB,EverlyAW(2005)Movementsofbluefintuna(Thun-nusthynnus)inthenorthwesternAtlanticOceanre-cordedbypop-upsatellitearchivaltags.
MarBiol146:409423YukinawaM(1987)Reportonthe1986researchcruiseoftheR/VShoyoMaru.
DistributionoftunaandbillfisheslarvaeandoceanographicobservationintheeasternIndianOceanJanuaryMarch1987.
RepResDivFishAgencyJpn61:1100276Editorialresponsibility:NicholasTolimieri,Seattle,Washington,USASubmitted:February11,2013;Accepted:April30,2013Proofsreceivedfromauthor(s):July5,2013