Table Of Content

Fisheries Research xxx (xxxx) xxx–xxx ContentslistsavailableatScienceDirect Fisheries Research journal homepage: www.elsevier.com/locate/fishres Unraveling the recruitment problem: A review of environmentally-informed forecasting and management strategy evaluation M.A. Haltucha,⁎, E.N Brooksb, J. Brodziakc, J.A. Devined, K.F. Johnsona,e, N. Klibanskyf, R.D.M. Nashd, M.R. Payneg, K.W. Shertzerf, S. Subbeyd, B.K. Wellsh aNorthwestFisheriesScienceCenter,NationalMarineFisheriesService,NationalOceanicandAtmosphericAdministration,Seattle,WA,UnitedStates bNortheastFisheriesScienceCenter,NationalMarineFisheriesService,NationalOceanicandAtmosphericAdministration,WoodsHole,MA,UnitedStates cPacificIslandsFisheriesScienceCenter,NationalMarineFisheriesService,NationalOceanicandAtmosphericAdministration,Honolulu,HI,UnitedStates dInstituteforMarineResearch,Bergen,Norway eSchoolofAquaticandFisheryScience,UniversityofWashington,Seattle,WA,UnitedStates fSoutheastFisheriesScienceCenter,NationalMarineFisheriesService,NationalOceanicandAtmosphericAdministration,Beaufort,NC,UnitedStates gNationalInstituteofAquaticResources,Kgs.Lyngby,Denmark hSouthwestFisheriesScienceCenter,NationalMarineFisheriesService,NationalOceanicandAtmosphericAdministration,SantaCruz,CA,UnitedStates ARTICLE INFO ABSTRACT HandledbyA.E.Punt Studiesdescribingandhypothesizingtheimpactofclimatechangeandenvironmentalprocessesonvitalratesof Keywords: fishstocksareincreasinginfrequency,andconcomitantwiththatisinterestinincorporatingtheseprocessesin Fisheries fishstockassessmentsandforecastingmodels.Previousresearchsuggeststhatincludingenvironmentaldrivers Recruitment offishrecruitmentinforecastingisoflimitedvalue,concludingthatforecastingimprovementsareminimal Forecasting whilepotentialspuriousrelationshipsweresufficienttoadviseagainstinclusion.Thisreviewevaluatesprogress Environment inimplementingenvironmentalfactorsinstock-recruitmentprojectionsandManagementStrategyEvaluations Managementstrategyevaluation (MSEs),fromtheyear2000through2017,byreviewingstudiesthatincorporateenvironmentalprocessesinto recruitmentforecasting,full-cycleMSEs,orsimulationsinvestigatingharvestcontrolrules.Theonlysuccesses identifiedwereforspecieswithashortpre-recruitsurvivalwindow(e.g.,opportunisticlife-historystrategy), wheretheabbreviatedlife-spanmadeiteasiertoidentifyoneoralimitedsetofkeydriversthatdirectlyimpact dynamics.Autoregressivemethodsappearedtoperformaswell,ifnotbetter,forspecieswithalongerpre-recruit survivalwindow(e.g.,seasonal,inter-annual)duringwhichtheenvironmentcouldpotentiallyexertinfluence. Thisreviewsuggeststhattheinclusionofenvironmentaldriversintoassessmentsandforecastingismostlikely tobesuccessfulforspecieswithshortpre-recruitsurvivalwindows(e.g.,squid,sardine)andforthosethathave bottlenecksintheirlifehistoryduringwhichtheenvironmentcanexertawell-definedpressure(e.g.,anadro- mousfishes,thosereliantonnurseryareas).Theeffectsofenvironmentmaybemorecomplicatedandvariable forspecieswithalongerpre-recruitsurvivalwindow,reducingtheabilitytoquantifyenvironment-recruitment relationships. Species with more complexearly life historiesand longer pre-recruit survival windows would benefit from future research that focuses on relevant species-specific spatio-temporal scales to improve me- chanisticunderstandingofabiotic-bioticinteractions. 1. Introduction recruitmentforecastingmodelsresultsinlittleornoimprovementwith respecttofisherymanagementperformance.Thislackofimprovement Interest in incorporating environmental processes in fish stock as- is largelybasedon researchrelatedtogadoid-like life histories.More sessments and forecasting models is increasing in step with studies specifically,since1999,thirtyeightstudieshavecitedBasson(1999), describing and hypothesizing impacts of climate change and environ- which we consider to be a benchmark analysis on the topic of using mentaldriversonvitalratesoffishes.However,muchoftheresearch environmentalfactorsforshort-termrecruitmentpredictionsforaga- on the importance of environmental factors in the design of manage- doid-likelifehistory.Thisincongruitymakesitdifficulttojustifycalls ment procedures found that including environmental indices in for increased complexity that potentially increase risk without ⁎Correspondingauthor. E-mailaddress:[email protected](M.A.Haltuch). https://doi.org/10.1016/j.fishres.2018.12.016 Received31May2018;Receivedinrevisedform13December2018;Accepted17December2018 0165-7836/ Published by Elsevier B.V. Please cite this article as: Haltuch, M.A., Fisheries Research, https://doi.org/10.1016/j.fishres.2018.12.016 M.A.Haltuchetal. Fisheries Research xxx (xxxx) xxx–xxx increasingbenefits. andtheresultantrecruits,ratherthanfocusingonolder,fullyvulner- InthisstudywehighlightthesimulationstudyofBasson(1999)and able ages where survival is less likely to be influenced by the en- then trace developments in the two decades since its publication. vironmentandmorelikelytovarywithfishingpressure. Basson(1999)foundnobenefitstoeitherstockconservationoraverage Since Basson (1999), the inclusion of environmental variables in yield from including an environmental factor in a stock assessment fisheries assessments remains rare. A recent examination of 1250 as- model. Furthermore, reductions in fishery management uncertainty sessments worldwide identified only 24 that included ecosystem in- took place only given a strong environment-recruitment relationship. formation regarding stock productivity, of which 14 were environ- Basson(1999)recommendedthatfutureworkexplicitlyincorporatethe mentalfactors(Skern-Mauritzenetal.,2016).However,areviewwitha assessment procedure within the simulation framework and that the wider scope, which summarized the use of all ecosystem information stock-recruitrelationshipberefittedduringeveryassessment,bothwith usedinstockassessmentsintheU.S.ratherthanjustrecruitment,found and without environmental factors. A more general recommendation that24%ofallU.S.stockassessmentsincludedecosysteminformation wastoexploreadditionallifehistoriestodeterminewhichmaybenefit (Marshalletal.,2018).SinceBasson(1999),severalstudieshavemade frominclusionofenvironmentalcovariatesinstockassessmentmodels. recruitmentforecasts(e.g.,Wardetal.,2014),someofwhichinclude The implicit nature of the relationship assumed between the en- environmentaldrivers.Mostlikely,thecombinationoflife-historytraits vironmental driver and recruitment, and the linkage with the harvest andmanagementtacticswillaffecttheutilityofenvironmentallybased strategy,mayhavecontributedtothenegativeresultinBasson’sstudy. recruitmentforecasting.Weundertookareviewofliteraturefromthe Forexample,Basson(1999)modeledagadoid-likeperiodiclifehistory, year 2000 through 2017, to evaluate progress in implementing en- basedlargelyonthestrongnegativerelationshipbetweenIrishSeacod, vironmental factors in Management Strategy Evaluations (MSEs) and Gadus morhua (ICES sub-area 7a; ICES, 1999), age-0 recruitment and stock-recruitment forecasts. We identified over sixty peer-reviewed sea-surface temperature (SST; Planque and Fox, 1998). Drinkwater studies that address questions of environmentally-driven recruitment (2005) also predicted a reduction in Irish Sea cod recruitment with forecastingandrelatedMSEs.Justoverone-thirdofthesestudieswere increasedSSTwithclimatechange.PlanqueandFox(1998)hypothe- foundtoberelevantduringinitialscreening,whererelevancewasde- sized several mechanisms to explain how warmer water might lower termined as a study going beyond mere correlation to model an en- recruitment, such as decreased fecundity, negative effects on egg or vironmental driver explicitly or implicitly and the stock assessment larvalphysiology,orfoodavailability.Themostimportanteffectofthe and/ormanagementimplicationswereconsidered.Therelevantstudies gadoid life history was likely on the simulated management strategy, weconsiderbelowarebroadlycategorizedaseitherprimarilyfocused where higher predicted temperatures, implying lower recruitment, onrecruitmentforecastingorelseonsimulationsoffull-cycleMSEsor dictated a lower fishing mortality (F) reference point and subsequent harvestcontrolrules(HCRs;Table1). total allowable catch (TAC) for that year. Recruits (age-0) were not Thegoalsofthisrevieware1)toexaminesuccessesandchallenges availabletothefisheryuntilageone,sosettingalowerTACduringa across studies that include environmental considerations in MSEs or yearofpoorerenvironmentalconditionswouldnothaveanimmediate stockrecruitmentforecastsand2)tohighlightstudycharacteristicsthat effect on fishing mortality on recruits from that year. However, tem- resultindifferentoutcomes.Thisreviewislooselystructuredbytopics perature was similar in consecutive years because of modelled auto- relatedtohowtherelevantstudiesaddressorfulfillissuesraisedin,or correlationinthetimeseries,sotheTACsetduringthenextyearwould recommendedby, Basson(1999).Basson’smainrecommendationsin- also be relatively low, relieving some pressure on the previous year’s cluded: 1) the consideration of a wide range of life-history types, ex- recruits. The overall effectof the management tacticon Basson’sper- ploitation patterns, and stock-recruitment relationships, 2) identifica- formancemeasures, yield andspawningstock biomass(SSB),was de- tion of strong interactions between environmental drivers and layedandindirectbecauseage-1fishmadeupasmallproportionofthe recruitment,3)increasedpredictionskillforenvironmentalfactors,4) catchandwerenotfullymatureuntilage3.Thesemanagementtactics increased realism with respect to management actions and im- might be more effective for stocks that matured earlier, had higher plementation, and 5) the consideration of long-term trends in en- natural mortality, or grew more quickly, with more of the spawning vironmentaltimeseries.Section2reviewsstudiesacrossawiderange biomassatyoungerages.Analternativemanagementapproachwould of different stock-specific life histories orgeneral scenarios. Section3 betodelaytheimplementationofstricterTACstomorecloselymatch reviews the need for simulation and forecasting studies with more thetimingoftheentryintothefisheryforcohortsspawnedduringyears realistic assessment, harvest, and management implementation mod- ofpoorenvironment.Thisapproachmightbemoreeffectiveatlower ules. Section 4 summarizes the use of a) either univariate or multi- stocksizesandatasteeperpartofthestock-recruitrelationshipbecause variate mechanistic drivers,b) historical data, c) spatialscale, andd) reducedmortalityonspawningageclassesshouldhaveastrongerpo- temporalforecastscale.Finally,sections5,6,and7provideasynthesis sitive effect on recruitment. Basson (1999) does not report the para- of successes, challenges, and future research recommendations to im- metervaluesforthestock-recruitcurveused,butwhenanalyzingthe proverelevancetofisheriesmanagement. samedata,PlanqueandFox(1998)foundrecruitmenttobeessentially independentofSSB. 2. Life-historystrategiesandenvironmentally-informed InreviewingtheworkthathasbuiltuponBasson(1999),weusethe recruitmentforecasts term ‘recruitment’ to designate the single age class at the end of the interval where population dynamics are governed by a stock-recruit 2.1. Characterizinglifehistories function. In this case, the stock-recruit function generally models the early-stagemortalitypriortorecruitmentasbeingdensity-dependent, Thevariabilityinlife-historystrategiesacrossfishtaxaandamong whereas the mortality rates of individuals older than the recruitment ecoregionsresultsindifferencesinthedurationandabundanceofin- agearetypicallymodeledasbeingdensity-independent.Wedistinguish dividualsindefinedstages(Houde,2002)(Fig.1).Severalauthorshave thisuseof‘recruitment’fromearlierdefinitionswhereitwasdefinedas triedtoclassifyandquantifydifferencesinlifehistorywithrespectto theagewhenfishbecomefullyvulnerabletothefishery(Ricker,1954, implicationsforfisherymanagement(e.g.,WinemillerandRose,1992; 1975). In contemporary stock assessment models, vulnerability to King and McFarlane, 2003; Winemiller, 2005). However, the classifi- fishing gear at all ages can be directly modeled as part of the assess- cations of periodic, opportunistic, (Winemiller and Rose, 1992; ment,aseitherafunctionofsizeorage,andpotentialtemporalchanges Winemiller, 2005) (Fig. 2) and salmonid (McCann and Shuter, 1997) in vulnerability can be accommodated as well (Quinn and Deriso, lifehistoriesareusedtostructureourreviewofthevariedresponsesto 1999). Thus, the recruitment definition that we adopt places the em- environmentally-drivenrecruitmentprocesses. phasisonprocessesthatexertinfluenceonsurvivalbetweenspawning Periodic strategists, such as many groundfish species, are 2 M.A.Haltuchetal. Fisheries Research xxx (xxxx) xxx–xxx e) g a p xt Capturesenvironmentaluncertainty? No No No No No No No Yes Yes No No No NANo No No No Yes Yes NAYes Yes Yes Yes ntinuedonne o c e ( Includeslargspatialscaledrivers? Yes Yes No No No Yes Yes No No Yes Yes No NAYes Yes No No NA No NAYes NA No No e ur Includestemperatmetric? Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes No NAYes Yes Yes Yes NA Yes NAYes NA Yes Yes 1 ForecastLength Annual Near-term Annual Long-term,2008-2100 Annual Near-term,to5yearsLong-term,2006-2100Long-term,2009-2050Annual Annual Annual NAAnnual Long-term,2001-2050 Annual InSeason Long-term,50yearsLong-term,2006-2050NALong-term,50years Long-term,25and50yearsLong-term elationships. >25YearsData No,2000-2010 Yes Yes,1950-1996 Yes,1900-2007 Yes Yes,1969-2000 Yes,1955-1999 Yes,1974-2005and1960-2009Yes,1979-2008 Yes,1970-2009 Yes,1952-2017 Yes,41(recruit),46(environment)YesNo,1997-2015 Yes Yes,1983-2012 No,1987-2000 Yes,1961-2004 Yes,1962-2005 NAYes,1963-2006(Cod) Yes,25and50years Yes,1976-2006 Yes,1963-2008 r nt e nt-environm Mechanisticrelationship? Yes No Yes Yes Yes Yes Yes Yes Yes Yes No Yes NANo Yes Yes No Theoretical Yes NANo Theoretical Theoretical Yes e m theoreticalrecruit Species Chinooksalmon Pacificsardine Pinksalmon Atlanticcroaker SpringSpawningHerring(NSSH)Cohosalmon Sprat Sprat WalleyePollock CohoSalmon CohoSalmon GulfMenhaden GeneralPinksalmon FlatheadSole,NorthernRocksole,ArrowtoothFlounderChinookSalmon Squid(Loligogahi) WalleyePollock WalleyePollock GeneralHerring,Cod,Plaice GeneralizedFlatfish,Hake,Rockfish Sardine WalleyePollock informedMSEstudieswith LifehistoryType,Genus(Classification) Salmon,Oncorhynchus(Salmonid)CoastalPelagic,Clupea(Opportunistic)Salmon,Oncorhynchus(Salmonid)Groundfish,Micropogonias(Opportunistic)CoastalPelagic,Clupea(Opportunistic)Salmon,Oncorhynchus(salmonid)CoastalPelagic,Sprattus(Opportunistic)CoastalPelagic,Sprattus(Opportunistic)Groundfish,Gadus(Periodic) Salmon,Oncorhynchus(Salmonid)Salmon,Oncorhynchus(Salmonid) CoastalPelagic,Brevoortia(Opportunistic)GeneralSalmon,Oncorhynchus(Salmonid)Flatfish,Hippoglossoides,Lepidopsetta,Atheresthes(Periodic) Salmon,Oncorhynchus(Salmonid)Invertebrate,Teuthida(Opportunistic)Groundfish,Gadus(Periodic) Groundfish,Gadus(Periodic) FlatfishCoastalPelagic,Groundfish,Flatfish,Clupea,Gadus,Pleuronectes(OpportinisticandPeriodic)Groundfish,Eopsetta,Merluccius,Sebastes(Periodic) CoastalPelagic,Sardinops(Opportunistic)Groundfish,Gadus(Periodic) y analysestypicall Spatialscale NortheastPacific CaliforniaCurrent NortheastPacific U.S.mid-AtlanticregionNorwegianandBarentsSeaOregonandWashington,USABalticSea BalticSea EasternBeringSea Oregon,US Oregon,US NorthernU.S.GulfofMexicoGeneralSoutheastAlaska,USEasternBeringSea CentralCalifornia FalklandIslands GulfOfAlaska GulfOfAlaska GeneralNorthSea General CaliforniaCurrent EasternBeringSea a studies.Previousdat Publication Burkeetal.,2013 Deyleetal.,2013 Haesekeretal.,2005 Hareetal.,2010 ICES,2013 Logerwelletal.,2003Mackenzieetal.,2008Mackenzieetal.,2012Mueteretal.,2011 Ruppetal.,2012 PacificFisheryManagementCouncil(PFMC,2018Vaughanetal.,2011 Wardetal.,2014Wertheimeretal.,2017Wilderbueretal.,2013 Winshipetal.,2015 Agnewetal.,2002 A’maretal.,2009a A’maretal.,2009b Brooks,2013Bruneletal.,2010 HaltuchandPunt,2011 Hurtadoetal.,2010 Ianellietal.,2011 d / able1ummaryofreviewe Forecasting MSE/Simulation/ReferencePointsHCRs TS 3 M.A.Haltuchetal. Fisheries Research xxx (xxxx) xxx–xxx characterized as slow growing, long-lived, and highly fecund species thathaveevolvedtopersistinhighlyvariableenvironmentswithper- Capturesenvironmentaluncertainty? Yes Yes NA NAYes isftoaetrgsdatiitscgetsgrr,oeiswcstrusiu,cnihgstum,acseshhnoAtartsstl-uamlcnivcateiencdsys,ac(wnWodiatihsPnteaaimcnlitpfiieleclrelmarsgeaaidlcnmiadsotpenRe,ocfeaisecersue,,n1cad9hri9etay3rc.a)h.cSatOaerlrpamicpztooeendrrtiiudzbneysditsritanaics-- termediatebodysize,ageatmaturity,andeggsize,aswellashaving e Includeslargspatialscaledrivers? No NA NA NANA s2shi0toi0orE3ntae)s.crhcloiflreirfeess-phpaionsntso,drfiynasgstttgroarotdewigstythin,ccaatnnsdtealxoghweisb,fietacnaudnndguiemtnybe(reKarilnloygf,bathnioedlodMguicrcFaaatliroltanrannoef-, eachsuccessivelifestageincreasesasafishdevelops.Theinfluenceof e environmental factors such as temperature, salinity, prey levels, and Includestemperaturmetric? No NA NA YesNA pdlirenevkdeealdotipotmoneaenrntev.ieFrxoiprnsemtc,tetenhdteatoltimpcoliannydgiimtoifopnsopsratoawnrntcirnuogeleso,srdweuigtrhigngpdridffoideffrueicrnteginotnrestsiaspgooenfstseeonsf between spawning strategy. For example, the total annual eggs to be m,50m, m,50 m, spawned are fixedprior tothe spawning event fordeterminate spaw- ForecastLength Long-ter2010-20Long-ter80yearsLong-ter2011-20NA NALong-ter50years nfoerrsi,nwdehteerremasineagtgesscpaanwdneevresl.oTphaetseantywtoimsteradtuerginiegstchoensfpearwdniffinegresnetasaodn- vantagesandvulnerabilitiesdependingonthedistributionoffavorable 5YearsData 1956-2006 1951-2010 ˜1968-1978 1978-2010 eLsinpnikflavewuiwrenoniinsncemeg,ettnbyhtepeanerleth(Bciicosanloadsnpit,gairo1wan9nds7ei5ern)se,,txepoo.egfcc.e,teeasndnuvobigrsdroturanarpmithnueigmcntattcyhlopeniendflsaitpnuiaodewnncscnoeinnwdgreiiltlliasoteneaadffswoeitnclotl. >2 Yes, Yes, Yes, NAYes, broadcastpelagicspawners,andenvironmentaleffectsmaybelessin- fluentialforeggsofspecieswithparentalcareorlivebirth.Variability inenvironmentalconditionscaninfluenceeggandlarvaldevelopment Mechanisticrelationship? Theoretical Theoretical Theoretical TheoreticalYes rLmaeetgengsteat(lte.cagon.n,ddPDietiepoBinnlso,ic1sa,9n1991a9)l,s4o)w,ihnleiflcahudesinntcoetvujarunrvi,eanbthillereomuggroohrwtastlthiatgyreartadetusersaa.ntEidonnvthir(eosrenee-- fore mortality rates (Rice et al., 1993; Houde, 1997). At the juvenile Species PacificOceanPerch Rocklobster MultipleAlaskaCrabStocks GeneralizedGadidSnowcrab sjl22uet.0aa2v1gde.6enPt),io.eliermdsioepfdnroioscrmittasytnraatddtefueapglctiesstntoasdrnsednintifcmltuhodorestaewlaithryeeat(hsBeearcrekthleeimtnaiutle.r,dse2ir0ny0e1gxr;toeBunontg,dswstahsdeicpehatrcaaaltn.e, LifehistoryType,Genus(Classification) Rockfish,Sebastes(Periodic) Invertebrate,Panulirus(Equilibruim)Invertebrate,Paralithodes,Chionoecetes,Lithodes(Equilibruim/Periodic)Groundfish,Gadus(Periodic)Invertebrate,Chionoecetes(Equilibruim/Periodic) stpvusreettiesromvceracraadWtpuklte,e-uitgartharrmielaffeeslcettseeru,yeanucrentthiemitdanfpemoasgocprrohlreerlledoeacocdneantcrhlasgkiustrotstioeutbi(srcm(Tgei2dhherae0eunetbn1lrautaat0lanto(gs–iCrdoa2ariainnea0nnssdn5flchcenh0uedaiaep)erellsacnvclbihrocetaeeogelsotrfeteaopaldhxamcelop.rudo,muvlsn2aaag.ai0lh)nmIS0,ntsp4sotuature)dori.rbnveggMsligmoatvtauvdaoenweeildd,mttriiiennaaaarmdtalnkeenvdeetdaanaalnrgotfliadap.enalle-(tgl02irwP-oit0toanaese1ndurtm1eimeirncr)--l ent Seads Sea on ClimateChange (IPCC) climate model output. Results suggest that Spatialscale CaliforniaCurr Australia EasternBering/AluteanIslan GeneralEasternBering rcmeoecmnrWuitn.ihtgmiledenetcht,aedsrpeeasi,wscnaoemgrobpoiadormemdaeswcsh,iaathnnidsfotcircaetcucahnsdtwseorasutslasdnudmliiknienglgyordfaeencndlvionimreoonrvmeecerrnutthiate-l factors that drive walleye pollock recruitment, MSEs suggested that Publication Punt,2011 Puntetal.,2013 Puntetal.,2014 Skagenetal.,2013SzuwalskiandPunt,2012 mppwcorriaeortnphcdiaoousgruicaoetttimnvi,nieathgnnyitgr(ehsAegte’nrirmamvbtaieeriragossineeh,tmsiafanetilsnnd.,tciamo2nlr0ipsp0molr9ironaaadkt)tc.uianhcgTgetehis,verbietcegyoutiuwmrbtrpeeeeeecnnarstfuhotmsihrfemteasntelheardveegseeuolmsslvtteroeeanrfdtatecligaslnttirecraishlsotkewaginneoydrf-, overfishing during periods of stock decline was lower. In later work, A’mar et al. (2009b) found that management strategies incorporating ed) environment often exceeded historical catch levels because manage- u ntin mentdidnotrespondquicklytodecliningproductivity,thusincreasing co the risk of overfishing. The authors postulated that the management ( 1 strategy might be improved when fluctuations or trends in stock pro- e bl ductivityoccurbyusingafisherycontrolrulewithafixedlowerlimit a T for spawning biomass, at which point fishing would be restricted or 4 M.A.Haltuchetal. Fisheries Research xxx (xxxx) xxx–xxx Fig.1.Intrinsicversusextrinsicfactorsthatdrivedifferencesinthedurationandabundanceofindividualsatdefinedstages. eliminated if the model estimated spawning biomass declines below usedaconstantrecruitmentrelationship.Iffuturerecruitmentremains thislimit. similar to past recruitments, then both status quo management, with Ianellietal.(2011)usedIPCCclimateprojectionsacrossarangeof static reference points and current ecosystem considerations, and the scenariosthatincorporatewarmingSSTstoevaluatealternativeman- climate-based alternatives, perform similarly. There is no gain in agement strategies for the Eastern Bering Sea walleye pollock stock. management performance using climate-basedreference points. How- Outcomesfromclimate-basedstockprojectionsresultedinlowercatch ever, if warming conditions result in reduced recruitment, then the and declining biomass due to reduced recruitment compared to pro- status quo management results in lower average catches and an in- jectionsbasedonthecurrentmanagementscenarioandstrategiesthat creased likelihood of fishery closures. The climate-based reference Fig.2.Fourendpointlifehistorystra- tegies,withafifthintermediatestrategy inthecenter.Adaptedfromaprincipal components plot from King and McFarlane (2003, Fig. 1). The five points corresponding to each strategy wereplacedattheapproximatecenter ofeachclusterofspeciesfromKingand McFarlane’s figure. Note that in King andMcFarlane’s(2003)originalfigure, thelabelsforopportunisticandperiodic strategistswereinadvertentlyswitched, but have been placed correctly in the currentfigure. 5 M.A.Haltuchetal. Fisheries Research xxx (xxxx) xxx–xxx pointsandcontrolrulestrackthedeclineinproductivity,allowingthe sole,Eopsettajordani)andopportunisticNorthPacifichake(Merluccius targetstocksizetoshiftlowerwithdecliningrecruitmentresultingin productus) examples. This work suggests that applying the wrong en- higherfisherycatchesandfewerfisheryclosures. vironmental relationship may actually make recruitment forecasts ResearchonNorthSeacod,(Gadusmorhua)aperiodicgadid,byKell worse. Haltuch and Punt (2011) suggested that criteria for including etal.(2005)evaluatedtheeffectofincludingclimateimpactsonshort- environment-recruitment relationships within a management frame- andlong-termmanagementstrategies.Resultssuggestasmalleffectof work could include 1) the use of an a priori conceptual mechanistic climate on short-term stock recovery, which was more dependent on modelbasedonenvironmentaldataatspatialandtemporalscalesthat fishing effort reductions and that incorporating climate into stock as- makebiologicalsense,2)clearlyexplainingtheextentofdatadredging sessment projections did not lead to an improvement. However, over (Burnham and Anderson, 2002), 3) using data-reduction techniques the long-term, climate had a greater effect,although reducing fishing (e.g., principal component analysis) to produce a smaller number of mortalitystillresultedinhighercatchesandstocksize. potentially more informative indices, and 4) using the most parsimo- Investigations of the inclusion of several environmental variables nious model that explains a high proportion of the variability in the intostock-recruitmodelsforthreeEasternBeringSeaperiodicflatfish data(Guisanetal.,2002). species (northern rock sole, Lepidopsetta polyxystra; flathead sole, Evaluatingharveststrategiesforaslow-growing,low-productivity, Hippoglossoideselassodon;andarrowtoothflounder,Atheresthesstomias) periodicspecies(Pacificoceanperch,Sebastesalutus),Punt(2011)in- found that springtime wind pattern was important for predicting re- corporateddifferentdelaysinthespringtransitionperiodaspredicted cruitment for all three species (Wilderbuer et al., 2013). The authors under global warming scenarios. Stock rebuilding was affected by hypothesizedthatwindsactingtotransportpelagiclarvaetofavorable forecasts of spring transition, but the impact could be reduced if the nursery grounds were responsible for above-average recruitment. target F was allowed to adapt during the rebuilding. However, re- However, the Arctic oscillation was a more important recruitment buildingtimedidnotimprovewhenincludingthecorrectformofthe predictor for arrowtooth flounder, which the authors hypothesized environment-stock-recruit relationship. Instead, the current manage- mightbeduetoeffectsonlarvalsurvivalthroughsettlementpatterns. ment system was responsive to climate-induced changes in biological Recruitment forecasts using environmentally modified stock-recruit parametersdespitenotbeingincludedexplicitlyintheoperatingmodel relationshipscoupledwithIPCCclimatemodelpredictionswereusedto orprojections. produce40-yearforecasts,thoughcomparableforecastsexcludingen- AslightlydifferentapproachwasusedbySkagenetal.(2013)ona vironmentaleffectswerenotcompleted. hypothetical periodic species, where trends in environmental factors Winter mortality of estuarine-dependent juveniles has a strong ef- forcedchangesinallbiologicalparametersthatmightbeaffected(e.g., fect on Atlantic croaker (Micropogonias undulatus) recruitment (Hare growth, condition, weight, maturity, and recruitment). Incentives to etal.,2010).Coldairtemperaturescanresultincoolingestuarinewater increase or decrease the total allowable catch were implemented, in temperatures to critically low levels in nursery areas, leading to high addition to the current management tactics. However, the authors mortality of overwintering age-0 croaker and reduced recruitment. found that using only environmentally-driven processes to inform Hare et al. (2010) found that inclusion of minimum winter air tem- management decisions (indicator based management) was not suffi- perature in the stock-recruitment model resulted in a significantly cient.Stockassessmentestimatesofstockstatuswerealsoneeded,and better fit tothe stock-recruit data. This temperature-dependent stock- without that information the environmental indicator was an in- recruit model produced recruitment predictions that were generally sufficientbasisforsustainablemanagementdecisions.Theauthorsex- consistent with the stock assessment. Long-term recruitment forecasts presscautionatgeneralizingthisresulttoobroadlyhowever,astheir incorporatingIPCCclimatemodeltemperaturepredictionsshowedin- primaryintentwastoillustrateasimulationframework. creases in recruitment concurrent with warmer, milder winters (Hare etal.,2010). 2.3. Opportunisticstrategists Workontwoperiodicstrategists(Atlanticcod,Gadusmorhua,and European plaice, Pleuronectes platessa) and one opportunist (Atlantic Researchontwosmallpelagicopportunisticspecies,sprat(Sprattus herring,Clupeaharengus)byBruneletal.(2010)echoedthefindingsof sprattus)andsardine(Sardinopssagax),foundwatertemperaturetobe Basson(1999);thegainfromincorporatingenvironmentalfactorsinto an important predictor of recruitment. Studies of Baltic Sea sprat the management strategy was dependent on the strength of the en- (MacKenzie et al., 2008; MacKenzie et al., 2012) cite several earlier vironment-recruitment relationship. Marginal improvements in man- papersdemonstratingeffectsofwatertemperatureonsurvivalofeggs agementperformancewerefoundgiventheinclusionofenvironmental andlarvae(seecitationsinMacKenzieetal.,2008).Earlierworkshows factors,withthemostimprovementunderunfavorableenvironmental that recruitment of sprat can be predicted by incorporating SST into conditions.Theauthorssuggestthatmovingbeyonduseofrecenthis- models(MacKenzieandKöster,2004).One-yearrecruitmentforecasts torical data to include forecasted environmental data might improve including low predicted SSTare substantially lowerthan forecasts ig- outcomes, but this is hindered by difficulty of making reliable ocea- noringtheenvironment(MacKenzieetal.,2008). nographic forecasts. Historical observations may not fully capture fu- SST was found to influence Pacific sardine recruitment via spatial ture conditions adequately under climate change and it is unknown shifts in the distribution of age-1 recruits during warmer years that whether current environment-recruitment relationships will remain results in increased availability to fisheries (Jacobson and MacCall, stationaryintothefuture. 1995).Later,SSTwasincorporatedintoatemperature-basedHCRfor Using a simulation model parameterized for three different life- Pacificsardine(Hilletal.,2011).Re-evaluationoftheSSTrelationship historytypes(i.e.,long-livedandunproductive,moderatelylong-lived indicatedalackofimprovedrecruitmentpredictions(McClatchieetal., andproductive,andshort-livedandvariableproduction),Haltuchand 2010),butrecentworksuggestsSSTremainsimportant(Jacobsonand Punt (2011) provide further reason for caution when incorporating McClatchie, 2013) and that SST impacts on recruitment vary, de- apparent environment-recruitment relationships into forecasts. They pendingonpopulationsize(Deyleetal.,2013).Infact,includingSST, oftenfoundfalsedetections(i.e.,Type-Ierror)givenacorrelationbe- or a proxy, improved short-term sardine forecast skill by about 30% tween a periodic environmental index and recruitment, for all three (Deyle et al., 2013). Given the persistent influence of SST for Pacific life-historytypes.Failuretodetecttruecorrelations(i.e.,Type-IIerror) sardinerecruitment,Tommasietal.(2017a)usedaMSEtocomparethe wasalsocommon;however,resultsvarieddependingonthelengthof performance of harvest policies that either included or excluded sea- thesimulatedtimeseriesandtheassessmentmethodapplied.Type-II sonal SST forecasts. The harvest policy that incorporated both stock errors were more problematic for the long-lived periodic rockfish biomass andskillful SSTpredictions led toincreases instockbiomass (canaryrockfish,Sebastespinniger)thanfortheperiodicflatfish(petrale andcatchandreductionsintheprobabilityofcatchandbiomassfalling 6 M.A.Haltuchetal. Fisheries Research xxx (xxxx) xxx–xxx belowsocioeconomicorecologicallyacceptablelevels(Tommasietal., et al.,2014). Ifsalmon prey items arenot available,salmon survival, 2017a). However, they recommended combining the SST informed growthandconditionofsalmonarereducedandsalmonpredatorsmay harvest policy with additional catch restrictions at low stock sizes to switchtopreyingonsmaller,poorconditionsalmonjuveniles(Tucker buffer against SST forecast errors. While work on Pacific sardine has et al., 2016). Such spatially restricted dynamics and life-history bot- focused on temperature-dependent harvest policies, SST-recruitment tlenecks, e.g., ocean entry point, suggest that a well-defined mechan- relationshipshavebeenusedtoproducelong-term(90years)recruit- istic conceptual model could lead to improved forecasts for Pacific ment forecasts for another sardine stock that estimated increased salmonpopulations. spawningbiomassduetoincreasingSST(MacKenzieetal.,2012). A large body of research has been dedicated to improving Pacific Vaughan et al. (2011) found that recruitment forecasts of age-0.5 salmon forecast models in the largely bottom-up driven California Gulfmenhaden(Brevoortiapatronus),a smallpelagicestuarine-depen- Current(CheckleyandBarth,2009).Burkeetal.(2013)examinedthe dent clupeiform, are improved using discharge from the Mississippi value of incorporating environmental variables identified in previous River. Early observations (Govoni, 1997) led to the hypothesis that processstudies(rangingfromlocalconditionsandecosystemprocesses greaterriverdischargehindersthemovementofGulfmenhadenlarvae to large-scale oceanographic and atmospheric variables) into the Co- fromthecontinentalshelfspawningareasintoestuarinenurseryareas. lumbiaRiver Chinooksalmon adultforecast ofreturns,findingthata One-year recruitment forecasts dependent on river discharge always forecast model fit retrospective abundance data well. Using a similar contained the observed recruitment within the 95% confidence inter- rangeoflocalandlarge-scaleenvironmentaldrivers,Ruppetal.(2012) valsineachyearofaneight-yearretrospectiveanalysis(Vaughanetal., andLogerwelletal.(2003)successfullyfitenvironmentaldatatotime- 2011). series of Oregon coho salmon (Oncorhyncus kisutch) recruitment. Pro- Incorporating environment into MSEs for opportunistic species is cess studies for Atlantic salmon (Salmo salar) are also demonstrating likelytobeusefulbecausetherecruitingyearclassisalargeproportion clear connections between environmental conditions and recruitment of the total stock size. Agnew et al. (2002) investigated whether in- (e.g.,Millsetal.,2013).Theseexamplesexplicitlyconsiderthebottom- corporating SST six months prior to Falkland squid (Loligo gahi) re- updeterminantsoftheproductivityoftheshelfecosystemandsuggest cruitment could improve single-cohort fishery management. The me- successfulforecastingofsalmonrecruitmentispossible.However,the chanisticlinkbetweenSSTandFalklandsquidrecruitmentisunknown effectofbottom-updeterminantsonsurvivalisneitherdirectnorlinear butlikelyrelatedtogrowthrates.UsingSSTpredictionstosetfishing (Wells et al., 2017). Rather, when production on the shelf is reduced mortality resulted in increased catch and lower risk of early fishery andtheforagetaxaarelessavailable,predatorscanswitchfrompre- closure, but fishing mortality varied widely unless the amount of ferred prey to salmon near shore, imparting significant variability in changeincatchwaslimited. recruitment(Emmettetal.,2006;Wellsetal.,2017).Suchsignificant Hurtado-Ferroetal.(2010)conductedanMSEforJapanesesardine cascadingeffectsrelatedtobottom-updriverscanreducethevalueof (Sardinops melanosticutus), where reference points changed at an en- usingenvironmentalcovariatesinforecastmodels. vironmental threshold. The authors found marginal improvements in Forexample,althoughPacificsalmonhaveaclearlife-historybot- outcomesforlong-termmanagementstrategies,alongwithnobenefit tleneck,analysesbyWinshipetal.(2015)wereunabletoidentifyre- forshort-termmanagementstrategies.Hurtado-Ferroetal.(2010)ar- liable environmental covariates for improving the prediction of adult guedthatgoodenvironmentalpredictionsarenotneededformanage- abundance from sibling models (i.e., estimating remaining cohort ment of this fishery; simply knowing the environmental threshold at abundanceatseafromtheabundanceofyounger,jack,salmonreturns). whichanewmanagementstrategyshouldbeimplementedwasenough. ForecastsofSacramentoRiver,Californiafall-runChinooksalmonadult abundancehavebeeneffectivelyestimatedfromtheabundanceofre- 2.4. Salmonidstrategists(PacificSalmonids) turningjacks(PacificFisheryManagementCouncil(PFMC,2018),but this relationship has weakened (Winship et al., 2015). Regression ThefirstmonthsatseaforPacificsalmonareconsideredacritical models incorporating different temporal combinations of eight en- period during which recruitment may be determined (Beamish and vironmentalvariables,20intotalconsideringseasonalinvestigations, Mahnken, 2001; Wells et al., 2012; Kilduff et al., 2014a, b; Woodson relatedtomaturationorsurvivalwereevaluatedforpredictionofadult and Litvin, 2015). Following Pearcy (1992), researchers focusing on ocean abundance (Winship et al., 2015). While the models demon- Pacificsalmonhaveattemptedtolinkoceanographicconditionsduring strated improved accuracy over models without environmental vari- or near the time of out-migration (juveniles migrating from fresh to ables, cross-validation methods (mimicking the assessment timing re- saltwater)toearlysurvivalandlaterrecruitment.Forsalmonpopula- quirements)didnotconsistentlyidentifyaone-yearforecastmodelthat tions from Alaska through California, Hare et al. (1999) and Mantua includedenvironmentalvariablesasthebestperformer.Simplerauto- etal.(1997)demonstratedtheimpactbasin-scalefactorscanhaveon regressive models based on jack returns had similar or better perfor- influencing salmon recruitment. Roughly, Northeast Pacific Ocean mance(Winshipetal.,2015).Theseresultshighlighttheimportanceof temperatures areinverselyrelatedtoproductionofsalmonintheCa- modelvalidationindevelopingrecruitment-predictionmodels. lifornia Current but positively associated with production of salmon Siblingmodelsmaynotbeavalidmeanstoestimatetheabundance fromAlaska.Mechanically,theserelationshipsaretheresultofagen- ofsalmonatseaforspecieswithashortermarineresidenceperiod,such erallyupwelling-dominatedsysteminCaliforniaCurrentandadown- aspinksalmon(Oncorhynchus.gorbuscha)andcohosalmon,asthereare wellingsysteminAlaska(Hareetal.,1999;Mantuaetal.,1997).Un- few fish that return early to serve as the basis for estimating the re- fortunately,whileinformative,therelationshipsbetweenproductionof mainingcohortstrength.Spawner-recruitrelationshipshavebeenused salmonandsolelybasin-scaledynamicshaveproventoocoarsetore- toestimateabundanceatseaforanumberofpinksalmonpopulations. liably forecast population-specific salmon recruitment. Wells et al. However, escapement (i.e., spawner abundance) surveys can be im- (2016) provides a mechanistic understanding of the relationships be- precise. Marine environmental indicators have been investigated as a tweenthebasin-,regional-,andmeso-scaleenvironmentandsurvivalof mechanismforimprovingtherecruitmentestimates.However,forecast a single central California Chinook salmon (Oncorhyncus tsawytscha) modelsacross43stocksofpinksalmoninthestatesofWashingtonand population.Namely,thestrengthandlocationoftheNorthPacificHigh AlaskasuggestthatrecruitmentestimatesincludingSSTbenefitedonly pressurecellinwinterrelatestolate-winterupwellingalongtheCali- afewstocks(21%)(Haesekeretal.,2005).Inrecentyears,pinksalmon forniacoastwhichprovidesnutrientstothesystemandsetspotential abundanceinSoutheastAlaskahasbeenestimatedusingoceansurvey productivityinspring assalmonout-migrate(Logerwelletal., 2003). estimates ofjuvenile abundanceandbiophysical indicators(e.g., zoo- As spring approaches, regional upwelling and transport dynamics de- plankton abundance, local physical conditions, and basin-scale in- terminetheproductivityandretentionofsalmonpreyitems(Schroeder dicators) (Wertheimer et al., 2017). While model validation suggests 7 M.A.Haltuchetal. Fisheries Research xxx (xxxx) xxx–xxx thatmodelswithbiophysicalindicatorsareanimprovementoverthose Introducing environment-recruitment uncertainty into the forecast without them, the best performing biophysical model has varied be- whenitwasnotincludedinthestockassessmentisalsosuspect. tweencohorts,potentiallyindicatingvaryingprocessesaffectingearly Severalrecentstudiesfeaturegreaterbiologicalcomplexity,testing marinesurvival.EstimatesofoceanabundanceofOregoncoastnatural a range of potential environmental linkages and evaluating whether cohosalmonattempttoaccountforcohort-specificprocessesaffecting forecasts improve when accounting for future environmental state survivalandrecruitmentbyusinganensemblemodelingapproach.The (Tables2and3).Mostforecastvalidationsfitthemodeltoasubsetof ensemble model includes six models with nine marine environmental available data and predict the remaining years so that forecasts are variables indicating basin conditions, low frequency variability, and comparedwithobservations.Forexample,Winshipetal.(2015)eval- regional environmental characteristics (Pacific Fishery Management uates the predictability of the Sacramento Index (a proxy for adult Council(PFMC,2018).Cross-validationhasdemonstratedthattheen- Chinooksalmonabundance)bytestingvariousfunctionalrelationships sembleapproachissuperiortoanyindividualmodel. with explicit termsfor environmental variables (SST, wind speed and The application of MSE for salmon has been uncommon. Winship stress,andotheroceanographicmeasures)orwithflexiblemodelsthat etal.(2013)usedanMSEapproachconsideringstrategiesforreducing allowedforsmoothchanges,reflectinglow-frequencyenvironmentalor fishing mortality on a protected salmon population from Sacramento ecological variation. A greater emphasis on this type of validation is River, California but did not include environmental covariates. This warrantedinfutureworkandcouldhelpreducethefrequencyoffalse lack of examples is, in part, due to the complex biological and en- positives that lead to acceptance of environmental drivers, which are vironmentalinteractionsaffectingsalmonthroughtheirlifecycle.The thendroppedafteroneormoreyearsofadditionaldataareconsidered. practiceofhatcherysupplementationcanalsobeseenasahindranceto Selectionoffishestoafisheryoftenoccursatayoungage,typically conductingMSEsforsalmonbecausehatcheryfishalterlinksbetween less than age two; therefore, when a study explicitly investigates re- stock size and recruitment. Life-cycle modelling may offer a way to lationships between recruitment and the environment, the time scale realisticallymodelprocessesaffectingsalmonsurvivalduringtheirlife waseithercontemporaneous(nolag)oratmostwithaone-yearlag.For cycleandtheinteractionsandfeedbacksoftheprocesses(seeGosselin example,Mackenzieetal.(2008)evaluatedwhetherincorporatingthe et al., 2018). Currently, along the U.S. west coast, life-cycle models North Atlantic Oscillation index value from the same year as the as- inclusiveofenvironmentalfactorsarewellparameterizedfortheearly sessmentwouldimprovetheassessmentestimateofage-1fishthatyear. lifestagesinfreshwater(e.g.,Hendrixetal.,2014).Thesepartiallife- AnexceptiontothetimescaleoflagsinvestigatedwouldbetheMSEof cycle models have informed temperature-dependent incubation sur- Punt (2011), where an environmental driver was incorporated by si- vival, flow release dynamics, and habitat restoration alternatives mulating recruitment data that were linked to the timing of spring (Martinetal.,2017;Dudley,2018).Uponcompletionandintegrationof transitionthreeyearsprior,althoughtheassessmentcomponentofthe marinecomponents(e.g.,Fiechteretal.,2015;Hendersonetal.,2018), MSEdidnotattempt toaccountfortheenvironment. Noneofthere- thesemodelswillbeimportanttoolsforconductingMSEsacrossthefull viewedstudiesinvestigatedmultiplelagsthatcouldoccurwithskipped life cycle including the evaluation of potential mitigation efforts for spawningorrecruitmentatolderages,thoughDeyleetal.(2013)noted climatechangeeffects,improvedhabitats,andmanagedfisheries. that multivariate state-space models have the potential to include multiplelagsifsupportedbythedata. 3. Theneedformorerealisticassessment,harvest,and The complexity of interactions that could occur in the early life- implementationmodules history stages prior to recruitment make it difficult to derive explicit mechanistic models of how the environment directly affects survival. During the past two decades, many studies have attempted to ad- Furthermore,longtimeseriesofdataforhypothesizeddriversthatspan dressquestionsregardingimprovedyieldorstockconservationviain- sufficientcontrastinspawnerabundancearerare.Perhapsthisiswhya corporating environmental factors into forecasts. Several studies es- majorityofstudiesincorporatetheenvironmentimplicitlybyallowing sentially repeat the approach of Basson (1999), simulating the true reference points in HCRs to reflect recent years of recruitment and population dynamics data with observation error before estimating biologicalparameters(e.g.,the“dynamicB0”,MacCalletal.,1985)or future catches (e.g., Brunel et al., 2010; Hurtado-Ferro et al., 2010). bybasingforecastsofrecruitmentonarecenttimeperiodratherthan Basson(1999)acknowledgedthatthis“short-cut”simulationneedsto thefulltimeseries(A’maretal.,2009a,Ianellietal.,2011;Puntetal., be more realistic; ideally, one would use MSE to simulate the entire 2013; Skagen et al., 2013). Using recent recruitment and biological procedure. parametersmakestheimplicitassumptionthatthenear-termenviron- In a full MSE, data are simulated, an assessment is performed to mentalstate,anditseffectonstockdynamics,willbesimilartorecent determine stock status, a HCR is applied to forecast catch, and the observations,withouttheburdenofpredictingtheenvironmentalstate procedure is repeated through time to provide a feedback loop (e.g., andestimatingthefunctionalrelationshipwithrecruitment. Aömaretal.,2009a,b;Hurtado-Ferroetal.,2010;Ianellietal.,2011; Another challenge to developing direct or lagged empirical en- Punt,2011;Puntetal.,2013;SzuwalskiandPunt,2012).Ofthestudies vironment-recruitment relationships is the lack of recruitment ob- reviewed,fullMSEswerelesscommonthanHCRevaluationswithout servations.Surveysofage-0andage-1fishareuncommon(forexcep- the full feedback loop, largely due to computational demands (Punt tions see Vaughan et al., 2011; ICES, 2013), and thus, most studies etal.,2016).Whilesimulationstudiesreflectacontinuumofsimpleto investigaterelationshipsusingstockassessmentoutputthatshouldbe complex, studies fell into two main categories: 1) those that evaluate interpreted with extreme caution (Brooks and Deroba, 2015). When existing and alternative HCRs for a particular or hypothetical stock explicit versus implicit incorporation of the environment has been (e.g.,Agnewetal.,2002;Ianellietal.,2011;Punt,2011;Skagenetal., compared, implicitly including the environment through autocorrela- 2013) and 2) those making forecasts with and without hypothetical tioninrecruitmentormodellingsmoothtrendsovertimeappearsless climatescenariostodemonstratedifferentstocktrajectories(e.g.,Hare prone to prediction error (Punt, 2011; Winship et al., 2015; Johnson et al., 2010; MacKenzie et al., 2012; Mueter et al., 2011; Wilderbuer et al., 2016a). At least one study found an improvement in forecast etal.,2013).Noneofthereviewedstudiesincorporatestheuncertainty performance by explicitly including the environment (Agnew et al., ofmanagementimplementation,i.e.,theinabilitytoachievemanage- 2002), but this was forasimplistic simulation of anannuallife cycle ment targets precisely even though implementation uncertainty is (squid).Futureworkshouldbevettedusingsimulationpriortobeing prevalent in many fisheries. Also, many of the forecast-only studies implementedwithstockassessmentoutput(e.g.,Deyleetal.,2013)or simplyusedassessmentoutputwithoutpropagatingtheuncertaintyand ideally should be included in the stock assessment itself (e.g., A’mar correlation of those model estimates into their forecasts, thereby un- etal.,2009b). derestimating overall uncertainty and potentially introducing bias. No single study performed a full factorial design using a range of 8 M.A.Haltuchetal. Fisheries Research xxx (xxxx) xxx–xxx e) g es pa NForecast PrincipleComponents 1 1 1 NA 4 1 1atatime 1 8covariatesacross6modelswith3variableach.Thisstudyusesmodelaveragingtoaddressuncertainty (continuedonnext UsedtoForecast Aprinciplecomponentsfromallvariables SIOSSTandPDO,oneatatime SST MinimumwinterairtemperatureNA SSTduringwinterpriortoemigration,Springtransitiondateduringemigration,Sealevelduringemigration,SSTduringwinterfollowingemigrationJan-FebNAO Foursetsofoceanmodeltemperatureprojections SST SPR,logspawners,FallMEI,SummerUWI,SpringSST,SpringSSH,FallUWI,JanuarySST 8 NSelected 23 2 1 1 NA 4 3,butoneatatime 4,butoneatatime 2 12variablesacross1selectedmodels Selected All SIOSST,PDO SummerSST MinimumwinterairtemperatureNA SSTduringwinterpriortoemigration,Springtransitiondateduringemigration,Sealevelduringemigration,SSTduringwinterfollowingemigrationEachfitinturnasanexampleAll PC1(averagetemperature),andPC3(predationpressure) May-JulyPDO(y=4yearaverage),SummerUWI,FallNPGO,SPR,LogSpawners,FallMEI,WinterNPI,Spring Es. S M or ed ns, der mulatio NConsi 23 9 1 1 6 10 3 4 10 >100 si esconsideredandusedforforecasting, CovariatesConsidered PacificDecadalOscillation(PDO),OceanicNinoIndex,SST,Springtransition,Upwellingstrength,Upwellinglength,Deeptemperature,Deepsalinity,Freshwaterflow,Freshwatertemperature,Copepodrichnessandnorthernanomaly,Copepodcommunity,Biologicalspringtransition,Ichthyoplanktonbiomass,Ichthyoplanktoncommunityindexscore,Chinookdiet(May,June),Chinookcondition(May),ChinookGrowth(June),Copepodcommunityindexscorefromverticalnets,Junebiomassofbongosamples,Densityofage1anchovy,Densityofadulthake,PrevalenceofRenibacteriumsalmoninarum,JuneChinooksurveyCPUE,JuneandJulySalmonCPUEinVancouverScriptsInstituteofOceanography(SIO)Temperature(surfaceandbottom),NorthPacificIndex(NPI),NewportPierSST,PDO,NorthPacificGyreOscillation(NPGO),SouthernOscillationIndex(SOI),MultivariateElNinoIndex(MEI),andSouthernCaliforniaBight(SCB)SSTSummerSSTNortheastPacificregional Minimumwinterairtemperature 2models,1.predatorabundance(cod),preyabundance(Caplin);2.hatchdate,SST,proportionofrepeatspawners,andpredatorabundance(youngsaithe)NPI,Upwellingintensity,SST,Springtransitiondate,sealevel(alltreatedseasonallyduringemigrationandsomeduringthefirstyearatsea). Maywatertemperature,NAO(Jan-Feb),andicecover(Balticsea)ModeledAugusttemperaturefromthreeBalticSeaocean-biogeochemicalhindcastmodels,andobservedMaytemperatureIceretreat,Springtransitiondate,SummerSST,Summerwinds,Stratification,Predation,andPrinciplecomponents(PCs)1through4fromtheabovecovariatesSeasonalrunningaveragesforMEI,NPGO,NPI,OceanicNinoIndex(ONI),PDO,SpringTransitionDate(SPR),SeaSurfaceHeight,SSTandUpwellingWindIndex(UWI).Calculated3month(2monthfor at onmentalcovari Publication Burkeetal.,2013 Deyleetal.,2013 Haesekeretal.,2005Hareetal.,2010 ICES,2013 Logerwelletal.,2003 Mackenzieetal.,2008Mackenzieetal.,2012 Mueteretal.,2011 Ruppetal.,2012 vir n e of ng ble2mmary orecasti au F TS 9 M.A.Haltuchetal. Fisheries Research xxx (xxxx) xxx–xxx e) NForecast Ensembleof6modelsinclusiveofthreevariableseach 1 NAVariesinter-annually 1 6total:4modelswithJacks,and2modelswithJacksand2environmentalcovariateseach1 1atatime 12eachfrom6downscaledIPCCmodels(72total) NA4,eachinturn (continuedonnextpag UsedtoForecast MeanMay-JulPDOaveragedover4yearspriorreturnyear,MEIOct-Dectheyearpriorreturn,SpringTransitiontheyearpriorreturn,SeaSurfaceHeightApr-Juntheyearpriorreturn,UpwellingJul-SepandSep-Novtheyearpriorreturn,SSTMay-JultheyearpriorreturnandJantheyearofreturn,logSpawners3yearspriorreturnMississippiRiverdischarge NAJuvenileCPUEorEcosystemRanksIndex.Additionalbiophysicalcovariatesarechosenthroughmodelselection. Springwinddirection 22 SST TwotheoreticalregimechangesbasedonobservedfrequencySeasonalIPCCmodeoutputsfor:precipitation,WME,andadvectionofoceanwater NAScenariowheretherangeofenvironmentalvariationissimilartohistoricalobservations.4Scenarios:constanttrendwithno e h t y n nnuall dingo NSelected 9 1 NAVariesinter-a 1to3,depenstock 22 1 1atatime 12 NA1,3,1 Selected SST,winterONI,springSSH,FallUWI,JanuarySST Ensemblemeanpredictor MississippiRiverdischarge NAJuvenileCPUEorEcosystemRanksIndex.Additionalbiophysicalcovariatesarechosenthroughmodelselection. Springwinddirectionandstocksize(northernrocksole);AO,Springwinddirection,andStocksize(arrowtoothflounder);Stocksize(flatheadsole)NA,Manymodelsevaluated. SST TwotheoreticalregimechangesbasedonobservedfrequencySeasonalIPCCmodeloutputsfor:precipitation,WME,andadvectionofoceanwater NAJan-Junetemperature(Cod);NOAlagged2years,feedinggroundtemperaturelagged1year,spawninggroundminimum 5al dot NConsidered 9 1 NA22 6maineffectsaninteractions,11t 22 1 1atatime 10 2,oneatatimeNotClear,<10 CovariatesConsidered MEI)runningmeanstoget12timeseriesforeachvariable.Additionally,Dec-JanmeansforallbutMEI,andJanuarymeans.LogSpawners.MeanMay-JulPDOaveragedover4yearspriorreturnyear,MEIOct-Dectheyearpriorreturn,SpringTransitiontheyearpriorreturn,SeaSurfaceHeightApr-Juntheyearpriorreturn,UpwellingJul-SepandSep-Novtheyearpriorreturn,SSTMay-JultheyearpriorreturnandJantheyearofreturn,LogSpawners3yearspriorreturn MississippiRiverdischarge NASalmonharvestandbiophysicalvariables;JuvenilesalmonsurveyCPUE;Juvenilecondition;Predatorsabundance;Zooplankton;Local-scalephysics(watertemperature,mixedlayerdepth);Basin-scalephysics(PDO,NPI,MEI,NPGO);EcosystemRanksIndexEndingdriftdepth,Endingdriftdistancefromland,AverageAnnualArcticOscillation(AO),MaySST,Springwinddirection,andSpawningstocksize 3seasonalvaluesfor:SST,Windspeed,Northerlywindstress,Easterlywindstress,Windcurl,UpwellingIndex,SealevelandSpringtransitionindex.Also,JackspawningescapementSST Twotheoreticalregimechangesbasedonobservedfrequency WinterPrecipitation,WinterWindMixingEnergy(WME),Springeddyformation,SpringWME,SpringUpwelling,SpringTemperature,SpringPrecipitation,SummerWME,SummerSST,FallSSTTheoreticalgoodorbadenvironmentSimulationassumesrelationshipknowwithouterror.ModelswerefitwithhistoricaldataandselectedusingAIC.Covariates:NOA,temperature,salinity(variousmeasures) d) Publication PacificFisheryManagementCouncil(PFMC,2018 Vaughanetal.,2011Wardetal.,2014Wertheimeretal.,2017 Wilderbueretal.,2013 Winshipetal.,2015 Agnewetal.,2002A’maretal.,2009a A’maretal.,2009b Brooks,2013Bruneletal.,2010 e u e2(contin E/Simulatio-n/ReferencePoints/HCRs bl MS a T 10