Back-Cas(ngofAdultStripedBassOtolithsUsingaRandomForestsModeltoDetermineCri(cal

CoastalNurseryHabitatsHughes,C.S.1andR.A.Rulifson1,2

1EastCarolinaUniversity,DepartmentofCoastalResourcesManagementandIns(tuteforCoastalScienceandPolicy,Greenville,NC27858

2EastCarolinaUniversity,DepartmentofBiology,Greenville,NC27858

ABSTRACT FishspeciesresidingintheAlbemarleEstuarineSystem(AES),a shallow wind-driven lagoonal system in North Carolina,USA, have mul(-element signatures in otoliths that reflectthe water chemistry of tributary rivers used as nurseryhabitat. We hypothesized that we could iden(fy whichwatershednurseryhabitatsweretheprincipalcontributorstothesurvivalandreturnofadultStripedBassMoronesaxa*lisin the spawningpopula(on.RandomForests (JMPPro11.2)analysiswas used to assign adults to their nursery habitats.ThemodelwastrainedusingAge0fish(the2008yearclass)thathadknownriverofsummernurseryhabitatandknownotolithelementalsignatures.Wethenusedthetrainedmodeltoanalyzethe60-to120-daypost-hatchregionofAge2fish(from the2008year class collected in2010)otoliths for theelementalsignaturesandgot100%agreement. Fromthere,weexpandedthedatatoincludemul(pleyearclasses(1996to2006)foundinthe2010spawningrun. Results indicatedthat thismethod can be used to back-cast adult fish to therespec(venurseryhabitatatAge0,andthereforedeterminethe rela(ve contribu(on of each habitat to the adultspawningpopula(on. Results canassistwith selec(ng sitesfor designa(on as “Strategic Habitat Areas” by the state ofNorthCarolina.

Example:SurvivedTitanicPassengers

Random Forests ModellingRandom Forests, or random decision forests, is a machinelearning algorithmmethod for classifica(on, regression andother tasks, that operates by construc(ng a mul(tude ofdecisiontreesattraining(meandoutpuangtheclassthatisthe mode of the classes (classifica(on) or mean predic(on(regression)oftheindividualtrees.Arandomdecisionforestcorrects for the tendency of decision trees to overfit thetraining set. An example is shown in Fig. 1.We canmodelwhoshouldbeexpectedtosurvivetheS.S.Titanicdisasterbyusingtheship’smanifest,thecostofthefares,thesex,age,andwhethertherewerespousesand/orsiblingsofsurvivorsor those lost at sea. Blue indicates survivors, and redindicatesthosethatdied.

Fig.1.Predic(ngthesurvivorsoftheS.S.Titanicsinkingbasedontheship’smanifest,sex,costofthe(cket,age,andwhethertherewerechildren,siblings,orspousesonboard.Bestsurvivorswerethosefemalesthatpaidmorethan$49.50infareandwereaparentwith1childornochildren.

METHODOLOGY Water Chemistry Determine spa(al and temporal stability of habitatwatershedsintheAlbemarleEstuarineSystem(AES).•  Eightriversystems:NorthRiver,PasquotankRiver,Lihle

River, Perquimans River, Chowan River, Roanoke River,ScuppernongRiver,AlligatorRiver(Fig.2).

•  Water samples takenwith peristal(c pump at the sub-surface (0.5 m), deep (0.5 m from bohom), upstreamanddownstream(n=398samplesover17months).

•  Induc(vely Coupled Plasma Op(cal EmissionSpectrometry(ICP-OES)usedtomeasurestron(um(Sr),barium(Ba),manganese(Mn),andmagnesium(Mg).

•  Spa=al and temporal variability – ANOVA and LinearDiscriminantFunc(onAnalysis(LDFA).

STUDY QUESTIONS •  Arewater chemistries of themul1ple Albemarle

Sound watersheds stable enough to be used as“watershedfingerprints”?

•  If so, can otolith chemistry of adult Striped BassMorone saxa1lis be used to determine whichnursery habitats contribute the most adultsreturningtospawn?

Fish Collections •  Adult Striped Bass collected March through May 2009

and2010 duringthepre-spawning,spawning,andpost-spawnperiod.

•  Adults sampled from spawning grounds of theRoanokeRiver near Weldon, North Carolina by State agencyelectrofishingsurvey.

Fig.2.MapofAlbemarleSound,NorthCarolina,studyareawherewhitecirclesindicatewatersamplecollec(onsites.

Fig3.(Lem)LasertransectsfromtheotolithcoretoouteredgeofAge0fish.NurseryHabitatsdefinedaselementconcentra(onsinthesec(onof60to120dayspost-hatch.

Fig.4.LDFAofwaterchemistrysignaturesinriversusingSr:Ca,Ba:Ca,Mn:Ca,Mg:Cara(osinwatersamples(n=398)collectedfromOctober2010toJuly2012.Groupcircles=95%confidencelimit.

Otolith Chemistry •  Sagihal otoliths used for ageing and micro-chemical

analysis.•  LA-ICP-MStransectswereusedtomeasureSr,Ba,Mn,Mg

inAge0andadultotoliths(Fig.3).•  Otolith chemistry from Age 0 fish held in cages within

watersheds(Mohanetal.2012)usedtotraintheRandomForests model. Twenty-eight (28) predictors used. Fishclassified with < 35% predictability were “unknown” tothemodel(e.g.,fishfromotherriversystems).

•  Back-cas=ng Adult Nursery habitats based on 60-120daysofotolithchemistry–RANDOMFORESTS

RESULTS Water Chemistry Watershedhabitatsshowedspa1aldifferences(ANOVA&LDFA)(Fig.4).WatershedhabitatsshowliOletemporalvariability.•  Sr, Ba, Mn = No significant varia(on over seasons or

years.•  Mg=Significantshort-termfluctua(on.

Dataset Modeled Total N ALLI N ALLI % CHOW N CHOW % PASQ N PASQ % PERQ N PERQ % UNKN N UNKN %

Mohan (2008 YC) 14 1 7.14% 3 21.43% 3 21.43% 7 50.00% 0 0.00%

Hughes (2008 YC) 17 0 0.00% 3 17.65% 1 5.88% 13 76.47% 0 0.00%

Hughes (2007 YC) 23 0 0.00% 7 30.43% 0 0.00% 14 60.87% 2 8.70%

Hughes (2006 to

1994 Year Classes) 206 0 0.00% 25 12.14% 6 2.91% 142 68.93% 33 16.02%

Table1. DatasetsmodeledfromMohan’s2008yearclasscomparedto60to120daysofHughes’2008yearclass,Hughes’2007yearclass,andHughes’2006 to1994yearclass.Tableshowstotalnumberoffish,numberoffishpredictedtoeachriver,andpercentageof fish predicted to each river. (ALLI=Alligator River, CHOW=Chowan River,PASQ=PasquotankRiver,PERQ=PerquimansRiver,UNKN=UnknownRiver).

OtolithChemistry•  TheChowan(CHOW),Perquimans(PERQ),&

Pasquotank(PASQ)riverswerewheremostspawningadultshadspenttheirfirstsummersoflife.

•  Habitatsignatureswere:•  PerquimansRiver–highMn(Fig.7)&most

returningfishasadults(Table1).•  ChowanRiver–lowSr&highBa;secondmost

returningadultfish.•  PasquotankRiver–highSr,lowBa,&highMg

(Fig.7).

60days 120days

Fig.5.Differencesofthefourtraceelementsinotolithsfromadultsusingnurseryhabitatsinthethreeimportantwatersheds:Chowan,Perquimans,andPasquotankriversbetween60and120days(Hughes2008yearclassdata).

CONCLUSIONS •  TheRandomForestsmodelworksextremelywelltodetect

pahernsinotolithchemicalsignaturesandmatchesthemtoknownhabitats.

•  Researchersmustfirstdeterminestabilityofhabitatwaterchemistrytoensurethatyoungfishinhabi(ngthathabitatwillacquireasignaturecharacteris(cofthathabitat.

•  Anthropogenicinfluences(farming,urban)canbeusedaspredictorsinthemodel.

•  ThePerquimansRiveristhemostimportantnurseryhabitatforStripedBasstosurvivetoadulthood.

OtolithChemistry•  TheChowan(CHOW),Perquimans(PERQ),&Pasquotank

(PASQ) rivers were where most spawning adults hadspenttheirfirstsummersoflife.

•  Habitatsignatureswere:•  PerquimansRiver–highMn(Fig.5)&mostreturning

fishasadults(Table1).•  ChowanRiver–lowSr&highBa;secondmostreturning

adultfish.•  PasquotankRiver–highSr,lowBa,&highMg(Fig.5).

Mohan, J.E.,R.A.Rulifson,D.R.Corbeh,andN.M.Halden.2012.Valida(onofoligohaline elemental otolith signatures of striped bass using in situ cagingexperiments andwater chemistry.Marine and Coastal Fisheries: Dynamics,Management,andEcosystemScience.4(1):57-70.

FundingbytheNCDepartmentofEnvironmentalQuality,CoastalRecrea(onalFishingLicenseProgram