ebook img

Establishing Functional Relationships between Abiotic Environment, Macrophyte Coverage PDF

19 Pages·2015·2.1 MB·English
by  
Save to my drive
Quick download
Download
Most books are stored in the elastic cloud where traffic is expensive. For this reason, we have a limit on daily download.

Preview Establishing Functional Relationships between Abiotic Environment, Macrophyte Coverage

RESEARCHARTICLE Establishing Functional Relationships between Abiotic Environment, Macrophyte Coverage, Resource Gradients and the Mytilus trossulus Distribution of in a Brackish Non-Tidal Environment JonneKotta1*,KatarinaOganjan1,VeldaLauringson1,MerliPärnoja1,AntsKaasik2, LiisaRohtla1,3,IlmarKotta1,HelenOrav-Kotta1 1 UniversityofTartu,EstonianMarineInstitute,DepartmentofMarineBiology,Mäealuse14,12618Tallinn, Estonia,2 UniversityofTartu,InstituteofEcologyandEarthSciences,ChairofZoology,Vanemuise46, 51014,Tartu,Estonia,3 NovaSoutheasternUniversity,HalmosCollegeofNaturalSciencesand Oceanography,8000NorthOceanDrive,DaniaBeach,Florida,UnitedStatesofAmerica * [email protected] OPENACCESS Citation:KottaJ,OganjanK,LauringsonV,Pärnoja Abstract M,KaasikA,RohtlaL,etal.(2015)Establishing FunctionalRelationshipsbetweenAbiotic Environment,MacrophyteCoverage,Resource Benthicsuspensionfeedingmusselsareanimportantfunctionalguildincoastalandestua- GradientsandtheDistributionofMytilustrossulusin rineecosystems.Todatewelackinformationonhowvariousenvironmentalgradientsand aBrackishNon-TidalEnvironment.PLoSONE10(8): bioticinteractionsseparatelyandinteractivelyshapethedistributionpatternsofmussels e0136949.doi:10.1371/journal.pone.0136949 innon-tidalenvironments.Opposingtotidalenvironments,musselsinhabitsolelysubtidal Editor:FrankMelzner,GEOMARHelmholtzCentre zoneinnon-tidalwaterbodiesand,thereby,drivingfactorsformusselpopulationsare forOceanResearchKiel,GERMANY expectedtodifferfromthetidalareas.Inthepresentstudy,weusedtheboostedregression Received:March26,2015 treemodelling(BRT),anensemblemethodforstatisticaltechniquesandmachinelearning, Accepted:August10,2015 inordertoexplainthedistributionandbiomassofthesuspensionfeedingmusselMytilus Published:August28,2015 trossulusinthenon-tidalBalticSea.BRTmodelssuggestedthat(1)distributionpatternsof Copyright:©2015Kottaetal.Thisisanopen M.trossulusarelargelydrivenbyseparateeffectsofdirectenvironmentalgradientsand accessarticledistributedunderthetermsofthe partlybyinteractiveeffectsofresourcegradientswithdirectenvironmentalgradients.(2) CreativeCommonsAttributionLicense,whichpermits Withinitssuitablehabitatrange,however,resourcegradientshadanimportantroleinshap- unrestricteduse,distribution,andreproductioninany ingthebiomassdistributionofM.trossulus.(3)Contrarytotidalareas,musselswerenot medium,providedtheoriginalauthorandsourceare credited. competitivelysuperiorovermacrophyteswithpatternsindicatingeitherfacilitativeinterac- tionsbetweenmusselsandmacrophytesorco-varianceduetocommonstressor.Tocon- DataAvailabilityStatement:Allrelevantdataare withinthepaperanddownloadableathttp://loch.ness. clude,directenvironmentalgradientsseemtodefinethedistributionpatternofM.trossulus, sea.ee/gisservices2/LiikideInfoportaal/index.html. andwithinthefavourabledistributionrange,resourcegradientsininteractionwithdirect Funding:Fundingforthisresearchwasprovidedby environmentalgradientsareexpectedtosetthebiomasslevelofmussels. InstitutionalresearchfundingIUT02-20ofthe EstonianResearchCouncil.Thestudyhasbeenalso supportedbytheEstonianScienceFoundationgrant 8807,theproject"Thestatusofmarinebiodiversity anditspotentialfuturesintheEstoniancoastalsea" No3.2.0801.11-0029ofEnvironmentalprotectionand technologyprogramoftheEuropeanRegionalFund. PLOSONE|DOI:10.1371/journal.pone.0136949 August28,2015 1/19 FactorsContributingtotheSpatialPatternsofMytilustrossulus TheprojecthasreceivedfundingfromtheBONUS Introduction projectBIO-C3,fundedjointlyfromtheEuropean Union’sSeventhProgrammeforresearch, Akeymissioninecologyistounderstandbioticpatternsandtheirchangesinnature.Inorder technologicaldevelopmentanddemonstrationand toachievesuchanunderstandinginthemarinerealm,ecologistshaveinitiatedamultitudeof fromtheEstonianResearchCouncil.Thefunders projectsaimingtomapmarinebiotaorperformedexperimentstodemonstrateinteractions hadnoroleinstudydesign,datacollectionand betweenphysicalenvironmentandorganisms.However,asdirectmappingofbiotais analysis,decisiontopublish,orpreparationofthe extremelycostlyinmarineenvironment,modellinghasbecomeanunavoidabletool,andsev- manuscript. eralrefinedstatisticalapproacheshavebeenalreadyappliedinthefield[1] CompetingInterests:Theauthorshavedeclared Distributionpatternsofspeciesdependontheirecologicalniche,whichconsistsofamulti- thatnocompetinginterestsexist. dimensionalenvironmentalspace.Ingeneral,non-independenteffectsarecommoninnature [2,3]and,therefore,neitherthespeciesnichenortheresultingdistributionrangecanbepre- dictedfromseparateeffectsofanindividualenvironmentalvariable.Asuitablehabitatisoften definedbycomplexinterrelationshipsamongamultitudeofenvironmentalvariablesthatcan belargelydividedintothreebroadcategories[4].Theseincludeindirectenvironmentalgradi- ents,resourcegradientsanddirectenvironmentalgradients.Indirectenvironmentalgradients canoftenbeeasilymeasured,butrepresentonlyproxiesforasetofunderlyinggradients, whichaffectorganismsdirectlywhileitmaybedifficulttomeasureordisentangletheeffectsof theseunderlyinggradients[5,6].Waterdepthcanbeviewedasatypicalindirectenvironmental gradientinthemarinerealm.Resourcegradientsaresubstancesbeingconsumedanddirect environmentalgradientsrepresentfeaturesthathavedirectphysiologicalimpactongrowth butarenotconsumed.Thepicturegetsmorecomplicatedasthesamefactormayhavean impactsimultaneouslyviadifferentpathways.Forexample,watermovementcanindirectly affectthehabitatofsuspensionfeedingbivalvesbymodifyingsedimentationratesoraffectses- silemusselsdirectlybyphysicallydisturbingordetachinganimals[1,7].Furthermore,theben- thicsuspensionfeedingmodeandsedentarylifestyleofmusselsprescribeanintrinsicneedfor avectoroffooddelivery.Thereby,watermovementcanimpactbenthicsuspensionfeedersalso throughathirdpathway,bymodifyingtheresourcesupplywhilelimitingtheamountoffood reachedbymussels[8,9]. Althoughthenicheconceptintroducedresourceaxesandtheimportanceofcompetition [10],mostoftheecologicalnichemodelinghasbeendealingwithabioticfactorsonly,without consideringinterspecificinteractionsandresources[11–13].Therefore,itislargelyunknown howbioticenvironmentinteractswithHutchinsonianfundamentalnichespaceinstructuring realcommunities[14].Therealizednicheofaspecies,however,dependslargelyonbioticinter- actionswithotherspecies[15–17].Thereby,herewedistinguishbesidesdirect/indirectenvi- ronmentalandresourcegradientsalsothefourthtypeofgradients,namelytheabundancesof ambientspeciesororganismgroupsotherthandirectresources,buteithercompetitors,preda- torsorfacilitators.Werefertothisgradienttypeasbioticinteractiongradients. Benthicsuspensionfeedingmusselsareanimportantfunctionalguildincoastalandestua- rineecosystems.Thisfunctionalguildfeedsonsuspendedfood,usuallymicroalgae,frombot- tom-reachingwatermasses[18].Despitealargebodyoffieldandexperimentalworks[19,20] westilllackknowledgeonhowvariousenvironmentalgradientsseparatelyandinteractively shapethedistributionpatternsofsuspensionfeedingmusselsindifferentecosystems.Therea- sonforthisis,firstly,becausethedistributionofsuspensionfeedersiscontrolledbyalarge numberofprocessesinvolvingbothbenthicandpelagicenvironments(e.g.substratecoloniza- tion,watermovement,phytoplanktonproduction,physicaldisturbances)aswellasmany interactionsbetweentheseprocesses[1,21].Secondly,duetothiscomplexityindrivingforces, thedirectionandmagnitudeofenvironmentalimpactonmusselsisexpectedtovaryhighly amongdifferentecosystems[22,23].Todate,thedistributionpatternsofmusselshavebeen extensivelystudiedintidalhabitats[19,24–26]whereasstudiesonnontidalareasarestill PLOSONE|DOI:10.1371/journal.pone.0136949 August28,2015 2/19 FactorsContributingtotheSpatialPatternsofMytilustrossulus scarce.Contrastingtotidalareas,musselsinhabitsolelysubtidalzoneinnontidalwaterbodies [27]anddrivingfactorsformusselpopulationsareexpectedtodifferfromthetidalareas[23]. Inthetidalzonespeciesareconstantlychallengedbyfluctuatingenvironmentalconditionsand thebioticpatternsoftenreflectthestresstoleranceofspecies[28–30].Ontheotherhand,subti- dalareasofferspeciessomestability;thus,thedistributionpatternsofmusselsareexpectedto beshapedprimarilybyhabitatandfoodavailabilityaswellaspredationpressure[31]. Therisinginterestinmarinehabitatmappinghasresultedinnumerousmodellingstudies focussedonthedistributionofspecies[32–35].However,traditionalstatisticalmodellingmay notbethemostrewardingwaytounderstandenvironmental-speciesrelationships,asitstarts byassuminganappropriatedatamodel,andmodelparametersarethenestimatedfromthe data[36].Duetothelackofasolidknowledgeonhowtheexternalenvironmentimpactsthe speciesthatwearetryingtomodel,thepredictiveperformanceofthesemodelsisexpectedto bemoderate.Ontheotherhand,duetotimeconstraintsandlimitedmanpower,experimental studiescannotresolvecausalconnectionsbeyondoneortwoenvironmentalvariables.More- over,experimentsareonlyseldomreplicatedinspaceandtime.Asaconsequence,theexperi- mentalapproachcanprovideusaverylocalizedsnapshot,butnotagenericunderstandingon environment-biotarelationships. Machinelearningprovidesatheoreticalframeworkthatmovesbeyondtraditionalparadigm boundaries.Considering„complexrealism”andourweaktheoreticalfoundations,modellingis seenhereasasophisticatedtooltoimproveourunderstandingontherelationshipbetween environmentandbiota.Bycontrasttotraditionalmethods,machinelearningavoidsstarting withadatamodelandratherusesanalgorithmtolearntherelationshipbetweentheresponse anditspredictors[37].Butevenheresomeecologicalunderstandingisaprerequisitewhenit comestoselectingenvironmentalvariablesforthemodel.Specifically,inordertosucceedin identifyingandquantifyingrelationshipsbetweentheenvironmentandbiota,themodel shouldincorporateatleastthemostimportantdirectandresourcegradientsaswellasrecap- turemultitudeofinteractionsbetweentheseenvironmentalgradientsandbiota.Thenovelpre- dictivemodellingtechniquecalledBoostedRegressionTrees(BRT)combinesthestrengthsof machinelearningandstatisticalmodelling.BRThasnoneedforpriordatatransformationor eliminationofoutliersandcanfitcomplexnonlinearrelationships.TheBRTalsoavoidsover- fittingthedata,therebyprovidingrobustestimates.Whatisthemostimportantintheecologi- calperspective:itautomaticallydetectsandmodelsinteractiveeffectsbetweenpredictors.The methodcopeswithdifferentnon-linearrelationshipsincludingthresholdsandunimodal responseswhicharecommoninecologicaldatabutdifficulttoanalyseusingmoretraditional methods.Duetoitsstrongpredictiveperformance,BRTisincreasinglyusedinecology [38,39].Although,weadmitthattheresultsofdistributionmodellingarepurelycorrelative andcausalinterpretationsneedtobevalidatedbyfutureexperimentalmanipulations,machine learningalgorithmsenableapowerfulinitialinsighttothekeydriversaswellastotheinterac- tionsbetweentheenvironmentandthebiota. Bluemusselsconsistofagroupofthreecloselyrelatedtaxa,knownastheMytilusedulis complex.ThecommonmusselMytilusedulisinsensustrictoisnativetotheNorthAtlantic, theMediterraneanmusselMytilusgalloprovincialisisnativeintheMediterranean,theBlack SeaandWesternEuropeandthebaymusselMytilustrossulusisnativetoNorthPacific,north- ernpartsoftheNorthAtlanticandBalticSea.Thetaxacanhybridisewitheachother,ifpresent atthesamelocality.M.trossulusinhabitsbothsubtidalaswellasintertidalareas,toleratesa widerangeofenvironmentalconditionsandthereforegainshighbiomassesatdifferenthabitat types[40].Thismakesthespeciesagoodmodelorganismtoimproveourunderstandingon therolesofmultipleenvironmentalgradientsonthedistributionofbenthicsuspensionfeeders. InthebrackishnontidalBalticSea,M.trossulusisanimportantorganisminvarioushardand PLOSONE|DOI:10.1371/journal.pone.0136949 August28,2015 3/19 FactorsContributingtotheSpatialPatternsofMytilustrossulus mixedbottomsubtidalhabitats.HereM.trossuluscoexistswithM.edulis,butasakeyecologi- caldifferentiationM.trossulustolerateslowersalinitycomparedtoM.edulisandtherebydis- tributesalmostthewholerangeoftheBalticSea[41,42].However,therearenopureM. trossulusintheBalticSeawithallmytilidsbeinghybrids,withvaryingfractionsofM.edulis allelesintheirgenomes[43]. Inthepresentstudy,weaimedtodescribetherealizednicheofthemusselM.trossulusin thenortheasternBalticSea,bothintermsofdistributionandthesizeofpopulations.Weused theBRTmodelling(1)toquantifytherelativecontributionofresource,abioticenvironmental andbioticinteractiongradientsonthedistributionofM.trossulusintheBalticSea(2)Wealso soughthowtheavailabilityofresourcesaffectsthestandingstockofspeciesand(3)howbiotic interactionsanddifferentdirectenvironmentalgradientsincludingkeydisturbanceseither separatelyorinteractivelymodulatetheresource-biomassrelationship. Weexpectedthatattheregionalscale,salinityisconsideredasthemainfactordrivingthe distributionofMytilustrossulus[42].Locally,however,alargearrayofenvironmentalvariables suchassubstratetype,watertemperature,flowvelocity,winter-timeicescour,areexpectedto eitherseparatelyorinteractivelyshapethedistributionpatternofmussels[7,44–47].Wealso expectedthatwithinafavourablehabitat,theavailabilityoffoodresourcesdefinesthebiomass ofspecies[47].Nevertheless,resourcegradientsinthisspacemayinteractwithdirectenviron- mentalgradients,whichactasvalvesregulatingtheavailabilityofresources.Asbenthicsuspen- sionfeederslinktwospatiallydistinctsystems,specificabioticenvironmentalconditionsmay beofutmostimportanceforthemindeterminingtheamountofresourcetobereceived[47– 49].Inaddition,disturbancemayreduceorultimatelyevendisruptthelinkbetweenresource parametersandthedistributionofspecies[50].Thismayexplainwhysomehighlytrophic areaswithe.g.sufficientamountofhardbottomandsuitablesalinitylackdensemusselpopula- tions[51].Finally,weexpectthattheinterspecificcompetitionbetweenmusselsandother biotaismoderate,rarelyoutperformingtheeffectsofabioticenvironmentaldisturbances[44]. Itisexpectedthatmacroalgaecompetewithmusselsforsubstrate,although,thishasnotbeen experimentallydemonstratedintheBalticSea.Instead,musselsareknowntofacilitatethe growthofmacroalgae[52]and,thus,mutualisticinteractionsbetweenmusselsandmacroalgae (e.g.dampeningdifferenttypesofdisturbances,intensifyingturbulentflowsatthebottom- waterinterface)mayactuallyoutweighapotentialreductioninadvectionbycanopymacroal- gae[53].Ascomparedtotheoceanicwaters,theBalticSealacksthemajorepibenthicpredators andthereforethepredationpressureonmusselsisalsolow[27,54].Predationbyvertebratesin thestudyareaisrare,decliningandhardlydetectable,therefore,wedecidednottoincludepre- dationtothedistributionmodel[55]. Methods 1.Studyarea ThestudyarealiesinthenortheasternBalticSea,intheEstoniancoastalsea(Fig1).Itischar- acterizedbyfullysubmergedhabitatduetotheabsenceoftides,althoughveryshallowwaters maybeirregularlyexposedbytheactionofwind.Salinityisconstantlylowandclosetothe physiologicaltoleranceofmussels.Opposingtomoresalinerangeofthespecies,invertebrate predationisabsentinthestudyarea[27,54].Thestudyareaencompassesmajorgeomorpho- logicalstructuresincludingdifferenttypesofsoft,limestoneandgranitebedrock,allowingthus togeneralizetheobtainedresultsoverlargepartsoftheBalticSea[56,57].Largepartsofthe studyareaarerelativelyflatandshallow,lackingsteepslopes.Shallowareasmayalsobesub- jectedtointensewinter-timeicescour.Waveenergyislowerthanonthecoastsoflargeoceans, butmaystillberemarkableforthebottomfaunaatshallowexposedsites,especiallyduring PLOSONE|DOI:10.1371/journal.pone.0136949 August28,2015 4/19 FactorsContributingtotheSpatialPatternsofMytilustrossulus Fig1.Mapofthesamplingstationsinthestudyarea.FilledcirclesindicatethelocationsofM.trossulus. doi:10.1371/journal.pone.0136949.g001 autumnandwinterstorms.Someareasaresubjectedtolocalupwellingeventsinducedbywind conditions.Often,angiospermormacroalgalcommunitiesinhabitthesebottomsatdepths downto20m.ThemusselM.trossulusexhibitsgenerallylowbiomassandsparsedistribution. Onlyatveryexposedopen-seaareas,thebiomassmayexceed1kgdwm-2[58]. 2.Biologicaldata Altogether3585stationsweresampledwithintheEstonianterritorialwatersduringtheice- freeseasonsbetween2005and2009.Themajorityofstationsweresampledonlyonce.Within eachwaterbodyapproximately15stationsweresampledannually.Inordertoestablishthe samplingstations,agridofrectangularcellswasgeneratedwithcellsizesof300musingthe SpatialAnalysttoolofArcInfo10[59].Thenwecalculatedthevaluesofwaveexposureand inclinationofcoastalslopesforeachgridcell(seebelow).Theexposureandslopeclasseswere combinedtotheavailableinformationondepthandbottomsediments(dividedintoclay,silt, sand,gravel,boulderandrockbottoms)availableinthedatabasesoftheEstonianMarineInsti- tute.Samplingsiteswerelocatedrandomlyinawaythateachcombinationofexposure,slope, depthandsedimentclasshadacomparablenumberofsamplingsites(Table1). PLOSONE|DOI:10.1371/journal.pone.0136949 August28,2015 5/19 FactorsContributingtotheSpatialPatternsofMytilustrossulus Table1. MeasuredenvironmentalvariablesintheoverallsamplingareaandintheareawhereM.trossuluswasfound. Variable Unit Samplingarea Distributionarea Mean Min Max Mean Min Max Depth m 11.77 0.10 75 8.87 0.2 47 Exposure m2s-1 229020 5672 968957 277950 5672 968957 Slope ° 0.66 0 13.47 0.79 0 10.56 Icethickness m 0.28 0 0.50 0.26 0 0.48 Temperature °C 12.95 0.03 22.23 12.88 0.03 22.23 Salinity psu 6.26 3.70 8.05 6.66 4.42 7.93 Oxygen mmolm-3 319 0 376 325 0 375 Velocity cms-1 3.75 0 15.26 3.58 0 13.34 Siltclaycover % 13.34 0 100 6.22 0 100 Sandcover % 38.12 0 100 21.96 0 100 Bouldercover % 37.87 0 100 58.15 0 100 Chlorophylla mgm-3 19.54 0.66 45 19.00 0.66 45 Plantcover % 31.65 0 100 44.43 0 100 doi:10.1371/journal.pone.0136949.t001 Ateachsamplingsitethecoverageofdifferentsedimenttypes(rock,boulders,pebbles, gravel,sand,silt)andmacrophytes(bothmacroalgaeandhigherordervegetation)wasesti- matedeitherdirectlybydiverorremoteunderwatervideodevice.Theunderwatercamerawas setatanangleof35°belowhorizontomaximisethefieldofviewandtherangeoftheforward viewwasabout2minclearwaters. InadditionateachsamplingsitequantitativesamplesofM.trossuluswerecollectedinthree replicateseitherbyadiverusingastandardbottomframe(0.04m2)onhardbottoms,orbya quantitativeEkman-Lenzgrabsampler(0.02m2)onsoftbottoms.Althoughthesampleswere collectedusingtwodifferentmethodswithdifferentaccuracy,thesetwomethodsarecompara- bleinthecaseofM.trossuluswithlimitedescapeabilitiesandrelativelyhomogenousseafloor area.Samplesweresievedatthefieldon0.25mmmeshscreens.Theresidueswerestoredat −20°Candsubsequentsortingandcountingofspecieswasperformedinthelaboratoryusinga stereomicroscope.Thedryweightofmusselswasobtainedafterdryingtheindividualswith shellsat60°Cfor2weeks. BiomasssamplingandanalysisfollowedtheguidelinesdevelopedfortheHELCOMCOM- BINEprogramme[60].AccordingtotheProtectionRulesoftheEstoniancoastalwaters,bio- logicalsamplingdoesnotrequirespecificpermitsorapprovals.Thestudyareaisnotprivately- ownedandthestudydidnotinvolveendangeredorprotectedspecies. 3.Environmentaldata Asetofenvironmentalvariableswerechosenfortheanalysesbasedonthetheoreticalassump- tionsoftheroleofenvironmentonthemusseldistribution(Table1).Thevaluesofwatertem- perature,salinityandwatervelocitywereobtainedfromtheresultsofhydrodynamicalmodel calculationsfrom2005−2009.Asannualaverages,minimaandmaximaofthestudiedhydro- logicalvariableswerehighlyintercorrelated,weusedannualaveragesinthefinalmodels.The calculationswerebasedontheCOHERENSmodelwhichisaprimitiveequationoceancircula- 0 0 tionmodel.Itwasformulatedwithsphericalcoordinatesona1 ×1 minutehorizontalgrid and30verticalsigmalayers.Themodelwasforcedwithhourlymeteorologicalfieldsof2mair temperature,windspeed,windstressvector,cloudcoverandrelativehumidity.Themeteoro- logicalfieldswereobtainedfromanoperationalatmosphericmodel.Themodelwasvalidated PLOSONE|DOI:10.1371/journal.pone.0136949 August28,2015 6/19 FactorsContributingtotheSpatialPatternsofMytilustrossulus againstwaterlevel,temperature,salinityandwatervelocitymeasurementsfromthestudyarea [61]. Winter-timeicedisturbanceisthekeydisturbanceformacrophyteandbenthicinvertebrate communitiesintheBalticSearange[62,63].FinnishMeteorologicalInstituteprovidedice coveroverthestudyareafortheinvestigatedperiod.Icecoverandthicknesswereproducedon dailybasisatanominalresolutionof500mandwerebasedonthemostrecentavailableice chartandsyntheticapertureradar(SAR)image.Theiceregionsintheicechartswereupdated accordingtoaSARsegmentationandnewiceparametervalueswereassignedtoeachSARseg- mentbasedontheSARbackscatteringandtheicethicknessrangeatthatlocation. Theamountofavailablefoodresourcesaffectsthedensitiesofspecies[64].Forsuspension feeders,thiscouldbetranslatedfromtheamountoforganicsestoninthewater.Waterchloro- phyllaisagoodproxyforthefoodsupplyofmussels[8].Inthisstudyweusedthesatellitesen- sorMODISAquaderivedwaterchlorophyllavalues.Thismeasureislimitedtosurfacewaters only;however,duetointensivemixinginourshallowwaterecosystem,thesatellitederivedval- uesrepresentwellnear-bottomconditions.Satelliteobservationswererecordedonweekly basisoverthewholeice-freeperiod.Cloud,landandotherprocessingflagswereidentifiedand maskedbyNASALevel2OceanColorProcessing.Thespatialresolutionofsatellitedatawas1 km.Erroneouszerochlorophyllavaluesmayoccurduetodifferentproblemsinimagepro- cessingchain.Theerroneousvalueshavetoberemovedpriortostatisticalanalysis.Asinall yearroundchlorophyllavaluesonlyveryrarelydropbelow0.2inthestudyarea,weuseda thresholdof0.1tofilteroutallthesefalsezeroconcentrations. Anothervariableaffectingmusselsalongdifferentpathwaysisexposuretowaves[1,7]. Exposuredefinesthewaterexchangebothbetweencoastalandopenseaaswellasbetween watersurfaceandbottomlayers[65].Thus,theinteractionbetweenchlorophyllaandexposure isexpectedtoindicatethefluxoffoodintothesite[50].Besidesbeingimportantforresource allocation,waveexposureisalsoadirectvariabletransportinglarvaeandaffectingadults directly[45].TheSimplifiedWaveModelmethodwasusedtocalculatethewaveexposurefor meanwindconditionsrepresentedbythetenyearperiodbetween1January1997and31 December2006[66].Anested-gridtechniquewasusedtotakeintoaccountlongdistance effectsonthelocalwaveexposureregime.Theresultinggridshadaresolutionof25m.Inthe modellingtheshorelinewasdividedintosuitablecalculationareas,fetchandwaveexposure gridswerecalculatedandsubsequentlytheseparategridswereintegratedintoaseamless descriptionofwaveexposurealongthestudyarea.Thismethodresultsinapatternwherethe fetchvaluesaresmoothedouttothesides,andaroundislandandskerriesinasimilarwaythat refractionanddiffractionmakewavesdeflectaroundislands. Althoughdepthistraditionallyregardedamongstthemostimportantparametersdescrib- ingspatialpatternofmussels[1,48],initiallywedidnotincludedepthinourmodel.Thisis becausedepthisasurrogateofseveraldirectvariablessuchaslightavailability,temperature, salinity,pressure,waveaction,icescouringortheircombinations[13].Thus,spatialmodels thatincorporatedepthasaindependentvariablearedifficulttointerpretduetoamultitudeof thecause-effectrelationshipsinvolved.Moreover,itislikelythatthedepth-biotarelationship changeswhenmovingfromonegeographicregiontoanother,orwhenextendingthestudy areatoincludealargerregion[1].However,forenvironmentalmanagementitmightbestill appealingtofindagoodapproximationofspatialdistributionasafunctionofasingleandeasy tomeasureparameteraswaterdepth.Therefore,werunadditionalmodelwhereonlydepth wasusedasapredictorofthespatialpatternofmussels. InordertomatchtemporalpatternsrelevanttothelifespanofM.trossulusandtogetridof potentialnoiseduetotheshort-termvariabilityofenvironmentalvariables,annualaveragesof PLOSONE|DOI:10.1371/journal.pone.0136949 August28,2015 7/19 FactorsContributingtotheSpatialPatternsofMytilustrossulus hydrophysicalvariables,waveexposureandwaterchlorophyllaandawintertimeaverageof icedisturbancewereusedwhenmodellingthepatternsofmussels. 4.BoostedRegressionTrees(BRT)modelling PriortomodellingthePearsoncorrelationanalysisbetweenallenvironmentalvariableswas runinordertoavoidsituationsofincludinghighlycorrelatedvariablesintothemodelling.The correlationanalysisshowedthatmostofvariableswereonlyweaklyintercorrelatedatr<0.1. However,moreexposedareaswerealsocharacterizedbyhighersalinity(r=0.60,p<0.001), lowerchlorophylla(r=-0.59,p<0.001)andlowericecover(r=-0.44,p<0.001).Inaddi- tion,thecoverageofstoneswasinverselyrelatedtosandcover(r=-0.59,p<0.001).Neverthe- less,thesevaluesarefarbelowacriticalthresholdwhencollinearitybeginstoseverelydistort modelestimationandsubsequentprediction[67]. ThecontributionofdifferentenvironmentalvariablesonthedistributionofM.trossulus wasexploredusingtheBoostedRegressionTreetechnique(BRT).BRTmodelsarecapableof handlingdifferenttypesofpredictorvariablesandtheirpredictiveperformanceissuperiorto mosttraditionalmodellingmethods(seee.g.comparisonswithGLM,GAMandmultivariate adaptiveregressionsplines,[68,69]).Whileoverfittingisoftenseenasaprobleminstatistical modelling,thisproblemcanbeovercomebyusingindependentdatasets.TheBRTmodelling iterativelydevelopsalargeensembleofsmallregressiontreesconstructedfromrandomsubsets ofthedata.Eachsuccessivetreepredictstheresidualsfromtheprevioustreetograduallyboost thepredictiveperformanceoftheoverallmodel[38]. TheBRTmodellingconsistedofatwo-stageprocess.InthefirstBRTmodelallstudiedenvi- ronmentalvariables(coverageofdifferentsedimentfractions,icethickness,oxygen,salinity, slope,watertemperature,waveexposure,velocity,chlorophylla,coverageofmacroalgae)were regressedtopredictthepresenceofM.trossulus.InthesecondBRTmodelonlythesamples containingM.trossuluswereusedtopredictthebiomassofM.trossulus.Inaddition,thepres- enceandbiomassofM.trossuluswereregressedusingonlydepthasasingleindependent predictor. InfittingaBRTthelearningrateandthetreecomplexitymustbespecified.Thelearning ratedeterminesthecontributionofeachsuccessivetreetothefinalmodel,asitproceeds throughtheiterations.Thetreecomplexityfixeswhetheronlymaineffects(treecomplexity=1) orinteractionsarealsoincluded(treecomplexity>1).Ultimately,thelearningrateandtree complexitycombineddeterminethetotalnumberoftreesinthefinalmodel.Followingthe suggestionsbyElithetal.[38]themodellearningratewaskeptat0.1andtreecomplexityat5 forbothmodels.Itwasalsocheckedthatthefinalmodelshadmorethan1000trees.Neverthe- less,aselectionofmodelparametershadonlymarginalimpactonmodelperformancewith optimalmodelsimprovingpredictionslessthan1%.Inordertoavoidpotentialproblemsof overfitting,unimportantvariablesweredroppedusingasimplifytool.Thistoolisacross-vali- dationbasedprogramdescribedbyElithandcolleagues[[38],detailsinAppendixS2].In ordertoeliminatenon-informativevariables,thetoolprogressivelysimplifiesmodel,thenre- fitsthemodelandsequentiallyrepeatstheprocessuntilsomestoppingcriterionisreached. Suchsimplificationismostusefulforsmalldatasetswhereredundantpredictorsmaydegrade performancebyincreasingvariance.Asaconsequence,ourfinalmodelsdidnotincludeany autocorrelatingvariables.Modelperformancewasevaluatedusingthecrossvalidationstatistics calculatedduringmodelfitting[37].Thus,whenrunningmodelsarandomselectionof80%of thedatawasusedfortrainingthemodelandtherestofthedatai.e.20%wasassignedfortest- ingmodelaccuracy.TheBRTmodellingwasdoneinthestatisticalsoftwareRusingthegbm package[70]. PLOSONE|DOI:10.1371/journal.pone.0136949 August28,2015 8/19 FactorsContributingtotheSpatialPatternsofMytilustrossulus Results 1.Presenceofmussels M.trossuluswasfoundat1635stationsoutof3585.TheBRTmodellingwiththesimplifytool optionondescribed85%ofvariabilityinthepresenceofM.trossulus.Altogethertenindepen- dentvariableswereretainedinthemodel.Over75%ofmodelvariabilitywasduetodirectabi- oticenvironmentalgradientswhereasresourcegradients(exposureandchlorophylla) contributedlessthan25%tothemodel.Ingeneral,directenvironmentalgradientshadstrong separateeffectswhileresourcegradientsimpactedthedistributionpatternofM.trossulus eitherseparatelyorinteractivelywithdirectenvironmentalgradients.FunctionsfittedbyBRT modelswerehighlyvariableinshape,andweremostlynon-linear(Fig2). Thecoverageofboulders,exposuretowaves,watersalinityexplainedover50%ofthe modelvariability.OthervariablescontributedmuchlesstothepresenceofM.trossulus.The increasingcoverofboulders,elevatedexposure,salinityaswellasmoderateicedisturbance separatelyincreasedtheprobabilityofoccurrenceofM.trossulusinthestudyarea.Theproba- bilitytofindM.trossulusincreasedwithalgalcoveruptoathresholdof10%.Abovethislevel furtherincreaseinalgalcoverhadnoeffectonmussels.TheprobabilitytofindM.trossulus increasedbothatlowandhighendsofchlorophyllagradient(Fig2). Exposureandsurfacewaterchlorophyllainteractivelycontributedtothepresenceofmus- selswithchlorophyllabeingimportantatlowexposurevaluesbutnotathighexposurevalues. Interestingly,atlowexposurechlorophyllavaluewasinverselyrelatedtotheprobabilityof occurrenceofM.trossulus.Inaddition,exposurestronglyinteractedwithiceandsiltcover.At lowicethickness,theeffectofexposureonM.trossuluswasonlymarginalwhereasathighice thicknesselevatedexposureexponentiallyincreasedtheprobabilityofoccurrenceofM.trossu- lus.Similarly,atlowexposuretheeffectofsiltonM.trossuluswasmoderatewhereasathigh exposure,elevatedsiltcoverlinearlydecreasedtheprobabilityofoccurrenceofM.trossulus (Fig3). Fig2.Standardizedfunctional-formrelationshipsshowingtheeffectofenvironmentalvariablesonthepresenceofM.trossulusinthestudyarea, whilstallothervariablesareheldattheirmeans.ThevariablesareorderedbytheirrelativecontributionintheBRTmodel(showninbrackets).Upward tickmarksonx-axisshowthefrequencyofdistributionofdataalongthisaxis.Seethesectionofmethodsforfurtherinformationonenvironmentalvariables. doi:10.1371/journal.pone.0136949.g002 PLOSONE|DOI:10.1371/journal.pone.0136949 August28,2015 9/19 FactorsContributingtotheSpatialPatternsofMytilustrossulus Fig3.Three-dimensionalpartialdependenceplotsintheBRTmodelforthepresenceofM.trossulusinthestudyarea. doi:10.1371/journal.pone.0136949.g003 TheBRTmodelincludingonlydepthasasingleindependentpredictorexplainedonly37% ofvariabilityinthepresenceofM.trossulusinthestudyarea. 2.Biomassofmussels Inareaswheremusselswerepresent,thebiomassofM.trossuluswasafunctionofonly3pre- dictors:exposure,coverofmacroalgaeandsalinity.Nevertheless,themodeldescribedonly 65%ofvariabilityinthebiomassofmussels.Atlowexposurevalues,thebiomassofmussel increasedslightlywithincreasingexposure.Abovecertainthreshold,smallincreaseinexposure resultedinadramaticincreaseinthebiomassofmussels.Increaseinbothplantcoverand salinityonlymoderatelyincreasedthebiomassofmussels.Similartothepresencemodel,func- tionsfittedbytheBRTmodelswerehighlyvariableinshape,andnon-linear(Fig4). Importantly,exposureandsurfacewaterchlorophyllainteractivelycontributedtothebio- massofM.trossulusdemonstratingasignificantroleofresourcegradientinthemodelofmus- sels’biomass.Highbiomasseswerefoundeitherunderconditionsoflowchlorophyllaand highexposureorhighchlorophyllaandmoderateexposure.Inadditiontherewerealsostrong interactionsbetweenexposureandthecoverofmacroalgaeandsalinityandexposure.Atlow exposure,relationshipbetweentheplantcoverandM.trossuluswasweak.Athighexposure, Fig4.Standardizedfunctional-formrelationshipsshowingtheeffectofenvironmentalvariableson thebiomassofM.trossuluswithinthedistributionrangeofmussels,whilstallothervariablesare heldattheirmeans.ThevariablesareorderedbytheirrelativecontributionintheBRTmodel(shownin brackets).Upwardtickmarksonx-axisshowthefrequencyofdistributionofdataalongthisaxis.Seethe sectionofmethodsforfurtherinformationonenvironmentalvariables. doi:10.1371/journal.pone.0136949.g004 PLOSONE|DOI:10.1371/journal.pone.0136949 August28,2015 10/19

Description:
Abstract. Benthic suspension feeding mussels are an important functional guild in coastal and estua- biotic interactions separately and interactively shape the distribution patterns of mussels in non-tidal RDC Team (2013) R: A language and environment for statistical computing. R Found Stat
See more

The list of books you might like

Most books are stored in the elastic cloud where traffic is expensive. For this reason, we have a limit on daily download.