Table Of ContentRESEARCHARTICLE
Modeled Population Connectivity across the
Hawaiian Archipelago
JohannaL.K.Wren1,2*,DonaldR.Kobayashi3,YanliJia4,RobertJ.Toonen2
1 JointInstituteforMarineandAtmosphericResearch,UniversityofHawai‘iatMānoa,Honolulu,Hawai‘i,
UnitedStatesofAmerica,2 Hawai‘iInstituteofMarineBiology,SchoolofOceanandEarthScienceand
Technology,UniversityofHawai‘iatMānoa,Kāne‘ohe,Hawai‘i,UnitedStatesofAmerica,3 Ecosystemsand
OceanographyProgram,PacificIslandsFisheriesScienceCenter,NationalOceanographicandAtmospheric
Administration,Honolulu,Hawai‘i,UnitedStatesofAmerica,4 InternationalPacificResearchCenter,
UniversityofHawai‘iatMānoa,Honolulu,Hawai‘i,UnitedStatesofAmerica
a11111
*[email protected]
Abstract
Wepresentthefirstcomprehensiveestimateofconnectivityofpassivepelagicparticles
releasedfromcoralreefhabitatthroughouttheHawaiianArchipelago.Potentialconnectivity
OPENACCESS
iscalculatedusingaLagrangianparticletransportmodelcoupledofflinewithcurrentsgener-
Citation:WrenJLK,KobayashiDR,JiaY,Toonen
atedbyanoceanographiccirculationmodel,MITgcm.Theconnectivitymatricesshowasur-
RJ(2016)ModeledPopulationConnectivityacross
prisingdegreeofself-recruitmentanddirectionaldispersaltowardsthenorthwest,fromthe
theHawaiianArchipelago.PLoSONE11(12):
e0167626.doi:10.1371/journal.pone.0167626 MainHawaiianIslands(MHI)tothenorthwesternHawaiianIslands(NWHI).Weidentify
threepredictedconnectivitybreaksinthearchipelago,thatis,areasinthemidandnorthern
Editor:GiacomoBernardi,UniversityofCalifornia
SantaCruz,UNITEDSTATES partofthearchipelagothathavelimitedconnectionswithsurroundingislandsandreefs.
Predictedregionsoflimitedconnectivitygenerallymatchobservedpatternsofgeneticstruc-
Received:April22,2016
turereportedforcoralreefspeciesintheuninhabitedNWHI,butmultiplegeneticbreaks
Accepted:November17,2016
observedintheinhabitedMHIarenotexplainedbypassivedispersal.Thebettercongru-
Published:December8,2016
enceinourmodelingresultsbasedonphysicaltransportofpassiveparticlesinthelow-lying
Copyright:Thisisanopenaccessarticle,freeofall atollsoftheuninhabitedNWHI,butnotintheanthropogenicallyimpactedhighislandsofthe
copyright,andmaybefreelyreproduced,
MHIbegsthequestion:whatultimatelycontrolsconnectivityinthissystem?
distributed,transmitted,modified,builtupon,or
otherwiseusedbyanyoneforanylawfulpurpose.
TheworkismadeavailableundertheCreative
CommonsCC0publicdomaindedication.
DataAvailabilityStatement:Dataisavailablehere:
https://www.bco-dmo.org/dataset/665467. Introduction
Funding:ThispaperwasfundedbyNSF(OCE12- Determininglevelsandpatternsofconnectivityisvitalforunderstandingmetapopulation
60169),andinpartbyagrant/cooperative
dynamicsandpersistence,andisessentialforeffectiveresourcemanagementsee[1–5].Over
agreementfromtheNationalOceanicand
ecologicaltimescales,populationpersistencedependsoneithertheabilitytoretainlocally
AtmosphericAdministration,ProjectR/SS-13,
whichissponsoredbytheUniversityofHawaiiSea producedlarvae,i.e.self-recruitment,ortheabilitytoimportlarvaefromnearbyareas,i.e.
GrantCollegeProgram,SOEST,underInstitutional connectivity[6–8].Self-recruitmentisametricdescribinghowopenorclosedapopulationis,
GrantNo.NA14OAR4170071fromNOAAOfficeof whichinturndescribesitsresilience[7,9].Openpopulationsreceiveaninfluxoflarvaefrom
SeaGrant,DepartmentofCommerce(forRJTand
outsidesources,makingthemmoreresilienttolocaldisturbancesbutlimitedinpotentialfor
JLKW).Theviewsexpressedhereinarethoseof
localadaptation[10,11].Closedpopulationsaremoresensitivetolocaldisturbancesandpos-
theauthorsanddonotnecessarilyreflecttheviews
sessagreaterpotentialforlocaladaptationsincetheyaredependentonlocallyproducedoff-
ofNOAAoranyofitssubagenciesThisisSea
GrantcontributionUNIHI-SEAGRANT-JC-13-16, springandhaveamoredirectlinkbetweenlocalproductionandrecruitment.Marine
PLOSONE|DOI:10.1371/journal.pone.0167626 December8,2016 1/25
PopulationConnectivityinHawaii
SOESTcontributionnumber9638andcontribution populationstudieshavehistoricallyworkedundertheassumptionthatmarinefishpopulations
number1661fromtheHawai‘iInstituteofMarine areopen—thatis,theyreceivelarvaefromotherpopulationssomedistanceaway[12]dueto
Biology.
thedispersalabilityandrelativelylonglarvaldurationofmarinefishlarvae.However,studies
CompetingInterests:Theauthorshavedeclared inrecentyearshavechallengedthisnotion,showingthatdespiteastronglarvaldispersalabil-
thatnocompetinginterestsexist. itymanymarinereefpopulationsappearclosed,withlarvaestaying“closetohome”[6,13–16].
Wenolongerassumeallmarinepopulationstobeopen,andthefocusisnowondetermining
theextenttowhichmarinepopulationsexchangelarvae(see[1,2]).Knowingtheconnected-
nessofapopulationisvitalineffectivelymanagingthepopulationanddesigningfunctioning
marinereserves.
Mostcoastalmarinespecieshaveabiphasiclifecycle,inwhichdispersaltakesplacepre-
dominantlyduringthepelagiclarvalstageofthelifecycle[17].Somespecieslaybenthiceggs
thatdevelopintopelagiclarvae,whereasothersspawngametesdirectlyintothewatercolumn,
wheretheydriftaspassiveparticlesuntiltheydevelopswimmingabilitiessimilartobenthic
hatchlings.Larvaecanbefeedingornon-feedinginthewatercolumn,andthepelagiclarval
phasemaylastforminutestomonthsinthepelagosbeforetheyreturntothebenthostosettle.
Eachoftheselife-historydifferenceshavepredictableimpactsonobservedpopulationgenetic
structure[18,19],butthebiologicalandphysicalfactorsdrivingdispersalintheseaarenot
wellunderstoodanddifficulttogeneralize.Factorscontrollingsuccessfuldispersalcanbespe-
ciesspecific[15,20–22],dependontimingofspawningevents[23,24],andvaryamongloca-
tions[25–29].
TheHawaiianArchipelago,locatedinthesubtropicalNorthPacificOcean,isa2,500km
longchainofvolcanicislandsandatolls,stretchingfrom19˚NintheMHIto30˚Ninthe
NWHI.TheHawaiianArchipelagoisoneofthemostisolatedontheplanet,andhometoone
ofthelargestmarinereservesintheworld,PapahānaumokuākeaMarineNationalMonument
(PMNM).ThereisahighlevelofendemismintheHawaiianArchipelago[30,31],anddueto
itsremotelocation,hasuniquemanagementneeds[32,33].WhiletheMHIarepopulatedwith
activefisheriesandheavyanthropogenicloading,theNWHIareuninhabitedandfullypro-
tectedwithlittleanthropogenicinfluence[34].OneofthehopesforestablishingPMNM,
whichwasthelargestMPAontheplanetatthattime,wasaspillovereffectwheretheprotected
fishpopulationsinPMNMwouldreplenishfishstockintheMHI.Unfortunately,thishope
hasbeenlittlesupportedamongstudiestodateofbothinvertebratesandfishes[35–38].The
lackofspilloverfromPMNMtotheMHIhasbeenattributedtotheprevailingsurfacecurrents
movinglarvaeuptheislandchainfromtheMHItowardstheNWHI[35,39].
BecausemanagementneedsvarygreatlybetweentheheavilypopulatedMHIandtheunin-
habitedPMNM,itisvitalthatweunderstandthepopulationdynamicsbetweentheseareasas
wellaswithinthem.Well-connectedpopulationswithnumerousdispersalpathwaysamong
sitesaremoreresilient,thatis,morelikelytorecoverfromdisturbance.Conversely,isolated
populationsthatarehighlydependentonself-recruitmentforpopulationmaintenanceareless
likelytorecoverafteradisturbanceandfaceagreaterriskofextinction[3–5,7].
Extensivepopulationgeneticworkhasbeendonetocharacterizepopulationstructurefor
fishandinvertebratestoinferexchangeamongsitesthroughouttheHawaiianArchipelago
(reviewedby[20,22]),butonlyahandfulofstudieshavefocusedonestimatingdispersaldur-
ingthelarvalstage[35,36,39–45].Todate,allsuchstudiesfocusoneitherasinglespeciesof
interest,asmallregionofthearchipelagooraverylimitedtimeperiod.Here,wepresentthe
firstcomprehensivedatasetdescribingmodeledpotentialconnectivityamongsitesthroughout
theentireHawaiianArchipelagousingabiophysicalmodelcoupledwitheddyresolvingocean
currents.Weuseapurelyphysicalmodelwithpassiveparticlestodeterminelikelypatternsof
potentialconnectionswithinthearchipelagoandJohnstonAtollbecausedetailedinformation
onlarvalbehavior,mortalityratesandpopulationsizesarenotcurrentlyavailableforthevast
PLOSONE|DOI:10.1371/journal.pone.0167626 December8,2016 2/25
PopulationConnectivityinHawaii
majorityofspeciesinHawai‘i.Theresultsfromthisstudywillsetthegroundworkforfuture
studiestousemorerealisticbiophysicalmodelsthatincorporatesuchfactorsaslarvalbehavior
astheybecomeavailable.
Methods
Dispersalmodel
MITgcm. TheMassachusettsInstituteofTechnologygeneralcirculationmodel
(MITgcm)solvestheincompressibleNavier-Stokesequationsonasphereindiscretizedforms
employingafinite-volumetechnique[46].TheregionalMITgcmimplementationforthe
HawaiianArchipelagoextendsfrom175˚Eto150˚Wandfrom15˚Nto35˚Nata0.04˚(~4km
intheregion)resolution.Intheverticaldirection,thewaterdepthisdividedinto50layers
withathicknessrangingfrom5mnearthesurfaceto510mnearthebottom.Itisforcedatthe
surfacebywindsderivedfromtheAdvancedScatterometer(ASCAT)observationswitha
0.25˚resolution,andbyheatandfreshwaterfluxesobtainedfromtheEuropeanCenterfor
Medium-RangeWeatherForecast(ECMWF)InterimReanalysisata1.5˚resolution.The
oceanstate,asestimatedbytheglobalHYCOMpredictionsystemata0.08˚resolution[47],is
usedtodefinetheinitialandopenboundaryconditions.Thesimulationperiodrunsfrom
May2009toMay2014.Weusetheflowfieldsinthe100mmodellayertodisperseparticlesin
ourLagrangiantrackingexperiments(seebelow),asthislayerhasshowntobethebestpredic-
torofsettlementintheregion[42,43].
Habitat. Forthisstudy,weincludedallavailablecoralreefhabitatintheHawaiianarchi-
pelagoandJohnstonAtoll.JohnstonAtollisthenearestreeftotheHawaiianArchipelago,
located1390kmsouthwestoftheBigIslandofHawai‘i.WechosetoincludeJohnstonAtollin
thehabitatdefinitionbecausethereareshownbiogeographictiesbetweenJohnstonAtolland
theHawaiianArchipelago[41,48–51].Togenerateourhabitatmap,weusedhabitatdefinedas
‘coralreef’inIKONOS-deriveddatafortheNorthwestHawaiianIslands[52,53]andthedata
setpresentedin[54]fortheMHI,andcreateda4-km2gridofthathabitat,totaling687habitat
pixels.Thehabitatpixelswereadditionallygroupedinto31differentislands/banks/atolls
(hereafterreferredtoasislands)toallowforislandscalecomparisons(Fig1).
Modelinitialization. Toinvestigatetheexchangeofparticlesamonglocationsinthe
HawaiianArchipelago,weusedaLagrangianbio-physicalparticletrackingmodel[40,42]cou-
pledwiththeflowfieldsfromtheMITgcmsimulationdescribedabove.Eddydiffusivitywas
setto250m2/sec,consistentwithdriftersinHawaiianwaters(following[35]).Wereleased50
particles(virtuallarvae)dailyfromMay2,2009,untilApril10,2014,from687coralreefhabi-
tatpixelstotalingjustover62millionreleasedparticlesforeachmodelrun.Weusedapelagic
larvalduration(PLD)of45days,representativeformostreeffish[55].Previousstudiesshow
thatPLD’slongerthan45daysdonotsignificantlyaltersettlementprobabilitiesintheMHI
[42].Foraparticletobeconsidered“settled”,ithadtobewithina5kmradiusofthecenterof
ahabitatpixelonthelastdayofitsPLD(onday45)(Table1).Preliminarystudiesshowno
changeinconnectivitypatternswhenparticlesareallowedtosettleacrossarangeofPLD,so
wechoseastaticPLDasopposedtorangeofPLD’sorasettlementwindowbecausewewere
interestedinthephysicaldriversofdispersal.Thedispersalmodelwasrunthreetimesandthe
outputaveragedforconsistency.
Totesttherobustnessofthemodelwithrespecttooceancirculationmodelresolution[56]
weranidenticalbiophysicalmodelruns,forcingthemodelwithcurrentvelocitiesfromthe
globalHYCOMata0.08˚resolutionandaregionalimplementationofHYCOMat0.04˚
(availablefortheMHIonly)resolutioninadditiontotheMITgcm(S1File).
PLOSONE|DOI:10.1371/journal.pone.0167626 December8,2016 3/25
PopulationConnectivityinHawaii
Fig1.MapoftheHawaiianArchipelago.ToppanelshowingtheHawaiianArchipelagodomainofthedispersalmodelwiththemajorsurfacecurrents(in
green)identified(afterLumpkin1998).Bathymetrylinesdenote1000and50misobaths.BottompanelsshowcoralreefhabitatpixelsfortheNorthwestern
HawaiianIslandsandMainHawaiianIslandsrespectively,witheachisland’shabitatpixelsshownasaseparatecolor.BathymetrylinesinNWHIdenote
50mdepthisobath.
doi:10.1371/journal.pone.0167626.g001
PLOSONE|DOI:10.1371/journal.pone.0167626 December8,2016 4/25
PopulationConnectivityinHawaii
Table1. Glossaryofoftenusedterms. TermsanddefinitionsusedthroughouttheHawaiianArchipelago
connectivitystudy.
Potential Themodeledestimateofconnectivityofasiteusingphysicaloceanographicattributes
connectivity andlimitedbiologicalfactorsinfluencingdispersalabilityoftheparticles.
Settlement Inthisstudywedefinesettlementasanyparticlewithin5kmofthecenterofahabitat
pixelonday45afterrelease
Self-recruitment Aparticlethatsettlesbackontothesameislandfromwhichitwasreleased
Connectivitybreak Anareaoverwhichfew,ifany,particlesareexchanged
Source-sinkindex Theratioofexportofparticlesandimportofparticlestoanarea,dividedbythetotal
numberofsuccessfullytransportedparticles.
doi:10.1371/journal.pone.0167626.t001
Statisticalanalysis
Wearefocusingonpotentialconnectivityinthisstudy,whichestimatestheconnectivityofa
siteusingphysicaloceanographicattributesandlimitedbiologicalfactorsinfluencingdispersal
ability(inourcasePLDandhabitat)[57,58].Toevaluatepatternsofpotentialconnectivityin
theHawaiianArchipelago,wecreatedaconnectivitymatrixthatmeasuresthelikelihoodof
particleexchangebycurrentsamongsites.Themodelgeneratesa687x687settlementmatrix
(Sij)containingthenumberofparticlesreleasedfromhabitati(sourcesite)thatsuccessfully
reachedhabitatj(receivingsite)forthefullrunofthemodel(fiveyears).Tocreatetherear-
wardprobabilitymatrix,wescaledSijtoislandspecifictotalreleasedparticles.Rearwardprob-
abilitymatricesreportoriginsitesofparticlesarrivingatthereceivingsiteandcanbewritten:
P =S /∑S.
ij(rearward) ij j
Wethenbinnedthe687habitatsitesusedinthedispersalmodelbyisland,resultingina31
x31islandmatrix,toallowforasimplercomparisonofpotentialconnectivity.Theresulting
probabilitymatrix(P )showstheoriginislandofsuccessfullytransportedparticlesateach
ij
island.ThenumberineachcelloftheP matrixistheprobabilityofaparticletransportedto
ij
islandjhavingoriginatedfromislandiforthefiveyearsthemodelwasrun,andeachrowin
thematrixsumsto1.Thediagonaloftheprobabilitymatrixshowstheself-recruitmentfor
eachisland.ForwardprobabilitymatriceswerealsogeneratedandaredescribedinS1File.
BecausethemajorityofcoralreeffishspawnduringMay-June[59,60],wecalculatedallmet-
ricsonbothyear-roundreleasesandreleasesrestrictedtoMay-Juneofeachyear.Allmatrices
wereplottedusingthesoftwareprogramGenericMappingTools(GMT)4.5.11[61].
Subtractionmatricesweregeneratedbysubtractingtheprobabilitymatrixforyearround
releasesfromthematrixforMay-JunereleasesusingthesubroutinegrdmathinGMT4.5.11.
Theresultingsubtractionmatrixshowswherethetwoconnectivitymatricesdiffer.Onlyfor-
wardmatriceswerecomparedwitheachother,andrearwardmatriceswitheachother.We
usedmanteltestsforeachpairofconnectivitymatricesusingfunctionmantelintheVegan
packageversion2.2–1inthestatisticalsoftwareR[62]tocalculatethecorrelationbetweenthe
probabilitymatrices.
Successfultransport,definedasanyparticlewithin5kmofthecenterofahabitatpixelon
day45afterrelease,wascalculatedbytallyingthedailynumberofsuccessfullytransportedpar-
ticlesforallislandsanddividingitbythetotalnumberofdailyparticlesreleasedforthefive-
yearmodelrun,allowingustodetermineannualandseasonalvariability.Additionally,wecal-
culatedislandspecific“settlement”successoverfiveyears.
Dispersaldistance,thegeographicdistancebetweenthereleasesiteandreceivingsitefora
successfullytransportedparticleatthereceivingsite,wasdeterminedbyfirstcalculatingdis-
tancesbetweenall687habitatpixelsusingthedistancematrixfunctiondistmwiththeHaver-
sineformulaintheR-packagegeosphere[63].Thedistmfunctioncalculatesthegreatcircle
PLOSONE|DOI:10.1371/journal.pone.0167626 December8,2016 5/25
PopulationConnectivityinHawaii
distance(Haversineformula)betweentwopointsusingtheirlatitudesandlongitudesin
degreesandcreatesa687x687distancematrix(Dij)withthereleasesites(i)onthex-axis
(rows)andreceivingsites(j)onthey-axis(columns).Wemultipliedthesettlementmatrix
(Sij)generatedbythebiophysicalmodel(seeabove)withthedistancematrix(Dij)togenerate
aproductmatrix(Pij).Becausethereismorethanonespawningandsettlementsite(hence-
forthhabitatsite)perisland(forexample,BigIslandhas129habitatsites,O‘ahuhas62and
KureAtollhas13),weaddedallthedistancesforallthehabitatsitesintheproductmatrix
belongingtoeachisland,generatinga31x31matrixcontainingthesumofallthedistancesof
alltheparticlesforeachislandcalledtheislandproductmatrix(PIij).Thesameprocedurewas
followedtogenerateanislandsettlementmatrix(SIij);asquare31x31matrixcontainingthe
totalnumberofsuccessfulsettlersforeachisland.Wethendividedthecolumnsumsfromthe
islandproductmatrixwiththecolumnsumsoftheislandsettlementmatrixtoobtainthe
meandispersaldistanceforsuccessfullysettledparticlesateachisland.Thesecalculationswere
performedforyear-roundreleases,aswellasforparticlereleasesconfinedtoMayandJuneof
eachyeartoallowustoexploreseasonalpatterns.
Self-recruitment,definedastheproportionofsuccessfullytransportedparticlesateach
islandthatoriginatedfromthatsameisland,isanimportantmetricwhenevaluatingtheper-
sistenceofapopulation[8,64].Wecalculatedself-recruitmentforthedurationofthemodel
runforeachislandbydividingthenumberofreleasedparticlesfromanislandthatweretrans-
portedbacktothatislandbythetotalnumberof“settlers”there.Thisallowsustodetermine
howdependentanislandisonrecruitmentfromoutsidelocationstomaintainthepopulation.
Source-sinkdynamicswereassessedbycalculatingasource-sinkindexfollowingHolstein
etal.[21].Wedefineasourceasanislandthatexports(outgoing)moreparticlesthanit
imports(incoming),andasinkislandimportsmoreparticlesthanitexports[21,64].The
source-sinkindexisaratioofthedifferencebetweensuccessfultransportoutoftheisland
(export)andsuccessfultransportintotheisland(import),dividedbythetotalofallsuccessfully
transportedparticlesinandoutoftheisland[21,64].Becausetheindexlooksatthedifference
inthetotalfluxofparticlesintoandoutofeachisland,itallowsustocompareislandswith
varyingamountofhabitatandislandsthathavetotalnumbersoftransportedparticlesthatdif-
fersbyordersofmagnitude.Theindexspansfrom-1to1,andapositiveindeximpliesasource
siteandanegativeindeximplyasinksite.Thestrongertheindexthemorelikelythesiteisto
beapersistentsourceorsinksite.Azeroindexindicatesthatthefluxofparticlesthataresuc-
cessfullytransportedontotheislandandoutoftheislandsarethesame.Thisindexallowsus
tocompareislandsinthearchipelagoandevaluatessource-sinkdynamicsonaregionalscale,
whereasself-recruitmentallowsustocharacterizeislandsassourcesorsinksonalocalscale.
Results
Potentialconnectivity
Theprobabilitymatrixshowsanisolation-by-distancepatternwithsitesfarawayfromeach
otherhavinglittleornopotentialconnectivityandconsiderableself-recruitmentformost
islandsacrosstheArchipelago(Fig2).RestrictingparticlereleasetothetypicalMay-June
spawningseasonminimallyalterstheoverallpotentialconnectivitypatterns(r=0.932)(Fig2b,
S1Fig).DuringspawningseasonO‘ahuandMaroReefshowstrongerconnectionswithneigh-
boringislandswhileNi‘ihauandKaua‘ibecomelessconnected.Self-recruitmentismore
importantforKureandMidwayAtolls(Fig2)duringspawningseason,whereasRaitaismore
dependentonself-recruitmentyearround(Fig2).
ThereislimitedpotentialconnectivitybetweentheNWHIandtheMHI,andthedirection
ofdispersalispredominantlyfromtheMHItotheNWHI.ParticlesoriginatingintheMHI
PLOSONE|DOI:10.1371/journal.pone.0167626 December8,2016 6/25
PopulationConnectivityinHawaii
Fig2.PotentialConnectivitymatrixfortheHawaiianArchipelago.(A)Thevaluesineachcellare
“settlement”probabilitiesscaledtothereceivingsiteforyear-roundparticlerelease.Arrowsindicatedthe
breaksmentionedinthetext.Eachrowinthematrixaddsupto1.Highvalues(red)indicatehighconnectivity
andlownumbers(blue)indicatelowconnectivity,andwhitecellsdenotenoconnectivity.(B)Differencematrix
showingthedifferenceinconnectivitybetweenyear-roundandMay-Juneparticlerelease.TheMay-June
PLOSONE|DOI:10.1371/journal.pone.0167626 December8,2016 7/25
PopulationConnectivityinHawaii
releasematrixissubtractedfromtheyear-roundreleasematrix(inAabove).Positivevalues(red)denotea
higherconnectivityvalueforyear-roundreleasesandanegativenumber(blue)denoteshigherconnectivity
forMay-Junereleasedparticles.
doi:10.1371/journal.pone.0167626.g002
form37uniqueconnectionswithsitesintheNWHI(markedareaintheupperrightcornerof
Fig2a),whileparticlesoriginatingfromtheNWHIonlyform24uniqueconnectionswith
sitesintheMHI(markedareasinthebottomleftinFig2a).Fourtimesasmanyparticlesare
successfullytransportedfromtheMHItotheNWHIthanviceversa(3.1%fromMHIto
NWHIand0.77%fromNWHItotheMHIofthetotalsuccessfullysettledlarvae).TheMHI
donotexportanyparticlesnorthwestofMokumanamana,andislandslocatedbetweenKaula
andNihoainthecenterofthearchipelagoaretheonlyislandsintheNWHItocontributepar-
ticlestotheMHI.MostparticlesreleasedfromNihoaarelosttothesystem,indicatedbythe
lowself-recruitment(<1%)andlowcontribution(6.067E-4%–0.72%)tothe“settlement”at
nearbyislands(Fig2).
WhiletransportbetweentheMHIandtheNWHIispredominantlynorthwestward,total
transportbetweenallislandsinthearchipelagoisreversed,with37%ofthesuccessfullytrans-
portedparticlesarriveatislandstothesouth(islandsbelowthediagonalinFig2a),while32%
ofparticlesaretransportedtoislandstothenorth(islandsabovethediagonalinFig2a).How-
ever,96.3%ofthesuccessfullytransportedparticlesoriginatingintheMHIendupsettling
withintheMHIand3.73%aresuccessfullytransportedtotheNWHI,while95.1%ofsuccess-
fullytransportedparticlesoriginatingintheNWHIaresuccessfullytransportedtositeswithin
theNWHIand4.87%ofparticlesaresuccessfullytransportedtotheMHI.
Therearethreebreaksintheconnectivitymatrixpresentforbothyear-roundandseasonal
particlerelease.Veryfewparticlessuccessfullycrossthesebreaks.Thesebreaksaremorepro-
nouncedduringspawningseasonreleases(Fig2b),andaremoredistinctintheforwardmatri-
ces(S2andS3Figs).ThesouthernmostbreaklocatedbetweenNihoaandMokumanamanais
themostpronounced.NoparticlescrossthisbreakintooroutoftheMHI,effectivelycutting
theMHIofffromtheNWHI.ThecentralbreakatGardnerPinnaclesandMaroReefistra-
versedonlybyparticlestoandfromRaitaBank.ThenorthernbreakbetweenLisianskiand
PearlandHermesAtolleffectivelyisolatesKureAtollandMidwayIslands,resultinginhigh
self-recruitmentforthenorthernmostislandsinthearchipelago.
Usingflowfieldsfromdifferentoceanographiccirculationmodelsatdifferentspatialreso-
lutionsallowsustotestwhetherthepotentialconnectivitypatternsarerobusttomodelresolu-
tion.Thereisastrongcorrelationbetweenthepotentialconnectivitydescribedaboveandthe
connectivitymatrixgeneratedfromadispersalmodelrunthatusedcurrentvelocitiesfromthe
coarserglobalHYCOM(r=0.9291)(S4Fig).FortheMHI,wecomparedconnectivitymatrices
generatedfromthreedispersalmodelrunsthatusedcurrentvelocitiesfromthe0.04˚MITgcm
(S5aFig),0.04˚regionalHYCOM(S5bFig),and0.08˚globalHYCOM(S5cFig).Potential
connectivityfortheMHIgeneratedfromthemodelrunusingdifferentresolutionsofthe
HYCOMcurrentsshowedthestrongestcorrelation(r=0.974)(S6aFig),followedbyconnec-
tivitymatricesgeneratedfromthemodelrunswiththesamespatialresolutionoftheflow
field,MITgcmand0.04regionalHYCOM(r=0.9533)(S6bFig).Weobservedthelargestdif-
ferencebetweenpotentialconnectivitygeneratedfrommodelrunsusingMITgcmandthe
0.08kmHYCOMflowfields(r=0.9305)(S6cFig).
Total“settlement”
Successfultransportacrossallislandsishighlyvariablewithameanof1.416%(SE7.708e-5)of
allreleasedparticlessuccessfullyarrivingatareceivingsiteoverthefive-yearmodelrun.The
PLOSONE|DOI:10.1371/journal.pone.0167626 December8,2016 8/25
PopulationConnectivityinHawaii
Fig3.TotalpercentsuccessfulsettlementforallsitesintheHawaiianArchipelagoforthefive-yearmodelrun.Thegreenbarsrepresentparticles
spawnedduringpeakspawningseasonMay-Juneeachyear.
doi:10.1371/journal.pone.0167626.g003
lowesttotalsuccessfultransportwasseenonJuly6,2011(0.682%),andthehighesttotalsuc-
cessfultransportonNovember2,2012(2.405%).Thereisnodiscernibleseasonalpatternin
totalarrivalsobservedforthearchipelagoasawhole(Fig3).Thehighestratesofsuccessful
transportin2009(2.27%)and2010(2.22%)coincidedwithparticlesreleasedduringpeak
spawningseason(markedbygreenbarsinFig3);however,thefollowingthreeyearshadsome
ofthelowestratesofsuccessfultransportforparticlesreleasedinMay-June(0.68%,0.95%,and
0.96%).
TheMHIhaveoveralllargerrelativeprobabilityofsuccessfularrivalswhileislandsnearthe
observedbreaksintheconnectivitymatrixhavethelowestrelativeprobabilityofsuccessful
arrivals.JohnstonAtollhasthelowestrelativesuccessfularrivalvalueofallat0.0637%for
year-roundreleaseand0.0337%forpeakspawningseasonreleases(Fig4).AtKaula,therela-
tivearrivalsuccessisalmosttwoordersofmagnitudelargercomparedwithJohnstonAtoll,
with3.574%foryear-roundspawning.Weseethelargestrelativearrivalsforseasonalrelease
atLāna‘iwith2.713%.Hawai‘iIslandistheonlyMHItoshowhigherarrivalsuccessforparti-
clesreleasedduringspawningseason(2.523%)comparedtoyear-roundreleases(2.298%).
Distancetraveled
Thespatiallyaveragedmeandistancetraveledis112.32km(SE=1.705)foryear-roundparti-
clerelease.Particlesreleasedduringpeakspawningseasontravelfurther,withmeandistance
of124.37km(SE=2.372).Mediandistancesareshorter,101.39kmand110.80kmforyear-
roundandMay-Junereleaserespectively,indicatingthatafewparticlesdisperse,traveling
PLOSONE|DOI:10.1371/journal.pone.0167626 December8,2016 9/25
PopulationConnectivityinHawaii
Fig4.Totalpercentsuccessfulsettlementateachislandforthefive-yearmodelrun.GreenbarsshowsettlementforparticlesspawnedduringMay-
June,graybarsshowsettlementforyear-roundspawning.
doi:10.1371/journal.pone.0167626.g004
significantlylongerdistancesanddrivingupthemean.Thisisalsoevidentfromthelongright
tailonthedensitykernel(Fig5).
Particlesarrivingatislandsinthecenterofthearchipelagotravelthelongestmeandis-
tances,whileparticlesatJohnstonAtolltraveltheshortest(100%self-recruitment)(Fig6).Par-
ticlessuccessfullytransportedtothebankjustsouthofNihoadispersedmorethandoublethe
meandistanceforothersitesacrossthearchipelago:onaverage341.4kmduringMay-June
release,and277.2kmduringyear-roundrelease.Consistentwithtotaldispersaldistancesfor
allislands,island-specificdispersaldistancesaregreaterforparticlesreleasedduringspawning
season,for23outof31islands(Fig6).IntheMHI,dispersaldistancesareconsistentthrough-
outtheyearexceptforparticlesreleasedfromKaua‘iwhichhasamuchlongerdispersaldis-
tanceduringMay-Junerelease.Kaua‘idispersaldistancesaremoresimilartoislandslocated
inthecenterofthearchipelago,likelyduetothepredominantlynorthwestdirectionofdis-
persal(Fig2)andthelongerdistancesbetweenhabitatsintheNorthwesternHawaiianIslands.
Particlesreleasedfromislandlocatedtothenorthwestofeachconnectivitybreak(Pearland
HermesAtoll,MaroReefandMokumanamanaIsland)haveshorterdispersaldistancescom-
paredtotheislandjustsoutheastofthebreak(LisianskiIsland,GardnerPinnacles,andNihoa
Island)by45.6%,63.4%,and73.9%respectively.
Self-recruitment
Themeanself-recruitmentforthearchipelagois25.2%(SE=0.0414)butvariesgreatlyfrom
islandtoisland.JohnstonAtollreliessolelyonself-recruitment(100%)forpopulationpersis-
tence,whileatNihoaIslandself-recruitmentaccountsforlessthan1%oftotalsettlement(Fig
7).DuringpeakspawningseasonNihoa,alongwithGardnerPinnacles,importalltheirparti-
cles.Islandspecificself-recruitment(Fig7,anddiagonalintheconnectivitymatrixinFig2)is
strongestatKure(year-round50.5%),PearlandHermesAtoll(year-round80.67%,May-June
PLOSONE|DOI:10.1371/journal.pone.0167626 December8,2016 10/25
Description:released from coral reef habitat throughout the Hawaiian Archipelago. tests for each pair of connectivity matrices using function mantel in the Vegan .. the interaction of particles with the physical environment [66–69] Current (HLC) and the North Hawaiian Ridge Current (NHRC), flow along the