Progress in Oceanography - cpb-us-e1.wpmucdn.com€¦ · Progress in Oceanography ... Seasonal-to-interannual prediction of North American coastal marine ecosystems: Forecast methods,

Contents lists available at ScienceDirect

Progress in Oceanography

journal homepage wwwelseviercomlocatepocean

Seasonal-to-interannual prediction of North American coastal marineecosystems Forecast methods mechanisms of predictability and prioritydevelopments

Michael G Jacoxab Michael A Alexanderb Samantha Siedleckic Ke Chend Young-Oh KwondStephanie Brodieae Ivonne Ortizfg Desiree Tommasieh Matthew J Widlanskyi Daniel BarriejAntonietta Capotondibk Wei Chengfl Emanuele Di Lorenzom Christopher EdwardsnJerome Fiechtern Paula Fratantonio Elliott L Hazena Albert J Hermannfl Arun KumarpArthur J Millerq Douglas Pirhallar Mercedes Pozo Builae Sulagna Rayc Scott C SheridansAneesh Subramaniant Philip Thompsonu Lesley Thornev Hariharasubramanian AnnamalaiuKerim Ayding Steven J Bograda Roger B Griffisw Kelly Kearneyfg Hyemi KimvAnnarita Mariottij Mark Merrifieldq Ryan Rykaczewskix

a Environmental Research Division NOAA Southwest Fisheries Science Center Monterey CA United Statesb Physical Sciences Division NOAA Earth System Research Laboratory Boulder CO United Statesc Department of Marine Sciences University of Connecticut Groton CT United Statesd Physical Oceanography Department Woods Hole Oceanographic Institution Woods Hole MA United Statese Institute of Marine Sciences University of California Santa Cruz Santa Cruz CA United StatesfUniversity of Washington Joint Institute for the Study of the Atmosphere and Ocean Seattle WA United StatesgNOAA Alaska Fisheries Science Center Seattle WA United StateshNOAA Southwest Fisheries Science Center La Jolla CA United Statesi Joint Institute for Marine and Atmospheric Research School of Ocean and Earth Science and Technology University of Hawairsquoi at Mānoa Honolulu HI United StatesjNOAA Climate Program Office Silver Spring MD United Statesk Cooperative Institute for Research in Environmental Sciences University of Colorado Boulder Boulder CO United StateslOcean Environment Research Division NOAA Pacific Marine Environmental Laboratory Seattle WA United Statesm Program in Ocean Science and Engineering Georgia Institute of Technology Atlanta GA United StatesnOcean Sciences Department University of California Santa Cruz Santa Cruz CA United StatesoNOAA Northeast Fisheries Science Center Woods Hole MA United StatespNOAA Climate Prediction Center Camp Springs MD United Statesq Scripps Institution of Oceanography La Jolla CA United StatesrNOAA National Center for Coastal Ocean Science Silver Spring MD United StatessDepartment of Geography Kent State University Kent OH United StatestDepartment of Atmospheric and Oceanic Sciences University of Colorado Boulder CO United Statesu Department of Oceanography University of Hawairsquoi at Manoa Honolulu HI United Statesv School of Marine and Atmospheric Sciences Stony Brook University Stony Brook NY United StateswOffice of Science and Technology NOAA Fisheries Silver Spring MD United StatesxNOAA Pacific Islands Fisheries Science Center Honolulu HI United States

A R T I C L E I N F O

KeywordsPredictionPredictabilityForecastEcological forecastMechanismSeasonalInterannual

A B S T R A C T

Marine ecosystem forecasting is an area of active research and rapid development Promise has been shown forskillful prediction of physical biogeochemical and ecological variables on a range of timescales suggestingpotential for forecasts to aid in the management of living marine resources and coastal communities Howeverthe mechanisms underlying forecast skill in marine ecosystems are often poorly understood and many forecastsespecially for biological variables rely on empirical statistical relationships developed from historical ob-servations Here we review statistical and dynamical marine ecosystem forecasting methods and highlight ex-amples of their application along US coastlines for seasonal-to-interannual (1ndash24 month) prediction of

httpsdoiorg101016jpocean2020102307Received 25 June 2019 Received in revised form 17 February 2020 Accepted 18 February 2020

Corresponding author at Environmental Research Division NOAA Southwest Fisheries Science Center 99 Pacific St Ste 255A Monterey CA 93940 UnitedStates

E-mail address michaeljacoxnoaagov (MG Jacox)

Progress in Oceanography 183 (2020) 102307

Available online 20 February 20200079-6611 Published by Elsevier Ltd This is an open access article under the CC BY license (httpcreativecommonsorglicensesBY40)

Large marine ecosystem properties ranging from coastal sea level to marine top predator distributions We then describe known me-chanisms governing marine ecosystem predictability and how they have been used in forecasts to date Thesemechanisms include physical atmospheric and oceanic processes biogeochemical and ecological responses tophysical forcing and intrinsic characteristics of species themselves In reviewing the state of the knowledge onforecasting techniques and mechanisms underlying marine ecosystem predictability we aim to facilitate forecastdevelopment and uptake by (i) identifying methods and processes that can be exploited for development ofskillful regional forecasts (ii) informing priorities for forecast development and verification and (iii) improvingunderstanding of conditional forecast skill (ie a priori knowledge of whether a forecast is likely to be skillful)While we focus primarily on coastal marine ecosystems surrounding North America (and the US in particular)we detail forecast methods physical and biological mechanisms and priority developments that are globallyrelevant

1 Introduction

Businesses government agencies and the public rely on weatherforecasts for planning and responding to changing environmental con-ditions While these forecasts are useful to ocean-related sectors thereis also high and increasing demand for ocean forecasts to supportproactive decision making in the marine environment (eg for fish-eries aquaculture shipping energy tourism and public health)Forewarning of ocean conditions can help decision makers reduce ne-gative impacts in unfavorable conditions and maximize opportunities infavorable conditions (Tommasi et al 2017a Hobday et al 20162018) The rise of marine ecological forecasting research in recent yearsbuilds upon a substantial knowledge base and infrastructure in climateforecasting (Doblas-Reyes et al 2013) and has been driven largely byincreasing demand for decision-support tools to help marine stake-holders prepare for and adapt to ecosystem variability and change(Payne et al 2017 Tommasi et al 2017a) including extreme oceanconditions such as marine heatwaves (Holbrook et al 2019) ecosystemhealth stressors (Siedlecki et al 2016) and the lack of sea ice in sub-arcticarctic waters (Stabeno et al 2017)

Marine policy and resource planning decisions are made based onforecasts and outlook information on various timescales In contrast toclimate projections which are used to explore long-term trends basedon various scenarios and mostly inform policy level decisions and long-term planning forecasts for days to years aim to provide prognosticinformation of ocean conditions on the timescales relevant to marineresource managers and other end-users A number of marine ecosystemforecasts have been cited in the literature (eg Payne et al 2017)though many are in fact nowcasts or target short lead times (ie days)For example nowcast tools that predict marine species distributionsbased on observed environmental conditions have been developed tobetter manage fisheries bycatch and ship-strike risks (Howell et al2008 Hazen et al 2017 2018 Welch et al 2019a Thorne et al2019) and three-day probability forecasts for blooms of the harmful

algal bloom (HAB) forming phytoplankton Pseudo-nitzchia alert man-agers to potential shellfish toxicity that could necessitate fishery clo-sures (Anderson et al 2016) Similarly coastal high water levels havehistorically been a more suitable target for short-lead prediction with acrucial focus on inundation impacts on timescales from hours to daysHowever in many cases longer lead times (ie 1ndash24 months) arebetter aligned with management and industry decisions Forecast sys-tems for southern bluefin tuna in Australia with several months leadtime have been useful products for fishers to plan the start of theirfishing season (Eveson et al 2015) and for managers to set seasonalarea closures based on impending ocean conditions (Hobday et al2011) In the eastern Bering Sea preliminary seasonal predictions ofthe ldquocold poolrdquo (ie the area where bottom temperature islt 2 degC) wereincluded in Ecosystem Status Reports from 2015 to 2018 to allowmanagers to consider fishing quotas in light of expected conditions(Siddon and Zador 2018) Indeed marine ecosystems are well suited toseasonal-to-interannual (S2I 1ndash24 month) prediction given the con-vergence of ecosystem variability physical drivers and managementand industry decisions that take place on these timescales A number ofefforts are underway to develop marine ecosystem forecasts for S2Itimescales along the coasts of the United States (US) leveraging im-proved technological capabilities model development and scientificunderstanding (Fig 1 Table 1) The S2I timescale is the focus of thispaper though we briefly discuss shorter and longer timescales as ap-propriate

The forecasts outlined above differ greatly in the models observa-tions and assumptions upon which they rely and the specific end-userneeds they are meant to address Of particular note for this paper theyalso exploit a wide range of physical biogeochemical and ecologicalprocesses that give rise to predictability on different spatial and tem-poral scales However documentation of an apparently predictableresponse in the marine ecosystem often precedes a thorough under-standing of the mechanisms underlying that predictability For ex-ample while recent studies have suggested multiple potential S2I

Fig 1 Locations of active S2I marine ecosystem forecasting efforts along US coastlines Label numbers correspond to entries in Table 1 Ocean bottom depth (m) isindicated in bluewhite shading (For interpretation of the references to colour in this figure legend the reader is referred to the web version of this article)

MG Jacox et al Progress in Oceanography 183 (2020) 102307

Table1

Summaryof

season

al-to-interann

ual(S2I)marineecosystem

forecastsalon

oastlin

esinclud

ymecha

ictability

Correspon

location

sareshow

initalicsun

derldquoForecast

rdquoindicate

rrespo

ndingsubsection

Section3

ldquoMarineEc

osystem

ForecastingMetho

dsrdquo

Forecast

ictability

ration

alstatus

Referen

sternBe

Summer

(July1)

xsea-ice

amically

downscaledph

ysical

forecast

Persistenc

esea-iceproc

throug

hUnimak

talprod

uctfor

Herman

in(201

4)httpsaccessafscno

cificNorthwest

physicsan

emistry

sardine

distribu

amically

downscaledph

ysical

emical

forecaststatisticalspecies

distribu

el(3123

21322)

Persistenc

SOteleco

nnection

sspeciesrespon

environm

ebprod

arSied

leckie

tal(20

16)Kap

oosorg

uctsj-sco

homeph

physics

speciesdistribu

s(swordfi

shleathe

rbackturtles

Californiasea

lionsb

luesharksb

luewha

amically

downscaledph

ysical

forecast

statisticalspeciesdistribu

els(312

Persistenc

SOteleco

nnection

sspeciesrespon

environm

velopm

eziani

etal(20

09)forph

ysical

dAbrah

9)forspeciesdistribu

Seasurfacetempe

ratureseasurfacehe

Statisticalforecast

(linearinve

rsemod

Persistenc

ical-extratrop

teleco

nnection

etal(20

ggiani

Diaset

Pacific

Island

sealeve

lGloba

namical

forecaststatistical

forecasts(3113

Persistenc

eanicRossbywav

ebprod

tedmon

Widlansky

etal(20

httpsuh

slcsoesth

awaiie

dusea-le

forecasts

6Coterminou

coasts

sealeve

lStatisticalforecast

tworkmod

Persistenc

eatmosph

forcing

opticatmosph

patterns)

velopm

eridan

etal(20

7Northeast

Shelftempe

ratures

pecies

distribu

interan

llowtailflou

nderAtlan

dsilver

butterfi

Statisticalforecasts(m

ultipletype

s)(313

Persistenc

eGulfStream

lfinteraction

vection

atmosph

teleco

nnection

velopm

etal(20

15)Dav

8Northeast

Seasurfacetempe

raturespeciesdistribu

ilotwha

leriver

herring)

Statistically

downscaledclim

ateforecast

(delta

d)statistical

speciesdistribu

els(3143

Persistenc

especiesrespon

environm

entalch

velopm

Thorne

etal(20

9Gulfof

Lobsterland

Statisticalforecast

basedon

rature

Speciesrespon

environm

perimen

Millset

wwgmriorgo

research

projectsg

ulf-maine

-lobster-forecasting

10Gulfof

toxicdino

ellate

ndrium

Statisticalforecast

basedon

observed

sembleof

itions

history

speciesrespon

environm

entalch

ualforecastspa

ersonet

httpswwww

ebsiten

ortheast-

forecasting

predictors for temperature and sea level on the Northeast US Shelf itis not yet clear how individual predictors are dynamically related andwhy statistical correlations with basin-scale climate indices hold uponly over portions of the observational record (eg Li et al 2014)Furthermore while physical forecast systems have been developedlargely from first principles many predictions particularly of biologicalquantities rely on empirical statistical relationships whose underlyingmechanisms are not well understood Uncertainty in the drivers ofobserved relationships increases concerns of nonstationarity ie that aseemingly robust relationship will not hold up over time (eg Myers1998) and forecasts based on it will fail as a result To the extent thatstatistical models can be informed by mechanistic understanding ofphysical-biological relationships or other ecological responses (egSection 42) they will likely perform better in a predictive capacity(Cuddington et al 2013)

The primary aims of this paper are to review forecasting approachesbeing developed and implemented for marine ecosystems documentmechanisms underlying predictability in these regions and explorehow these mechanisms can be exploited for marine forecasting effortsWe focus primarily on North American large marine ecosystems butemphasize that many of our findings and recommendations are ap-plicable to coastal regions around the globe We focus on the S2Itimescale shorter and longer timescales are mentioned briefly and areof importance to marine decision making but are outside the scope ofthis review The forecasts and mechanisms addressed in this paperapply to physical biogeochemical (chemistry phytoplankton andzooplankton) and ecological (higher trophic level) properties Weoutline these properties in Section 2 and then detail dynamical andstatistical forecast methods (Section 3) mechanisms of predictability(Section 4) and pressing challenges and priority developments (Section5) Finally concluding remarks including a summary of recommenda-tions for future work are presented in Section 6

2 Scope of marine ecosystem forecasts

Marine ecosystem research encompasses topics from physical at-mosphereocean dynamics to coupled social-ecological systems andmarine ecosystem forecasts can be similarly diverse in their aims Forexample marine ecosystem forecasts are expected to aid in the im-plementation of ecosystem-based management which requires quanti-tative methods criteria and thresholds to assess overall ecosystemstatus evaluate trade-offs among ecosystem services and guide

management actions (Levin et al 2009 Leslie and McLeod 2007) Tothat end forecasts are desired for various biotic and abiotic componentsof marine ecosystems in the water column and on the seafloor as wellas the structure and function of marine ecosystems and ecologicalcommunities (Levin and Schwing 2011)

In this paper we address predictability associated with a range ofphysical biogeochemical and ecological properties for which prog-nostic information is beneficial to end-users and for which the neces-sary observations are available to develop initialize and evaluateforecasts (Capotondi et al 2019a) Atmospheric state variables ofparticular interest include near surface fields such as winds pressuretemperature and precipitation the last of which is also used to estimateriver outflow Commonly targeted physical oceanic variables includesea level derived quantities such as mixed layer depth and statevariables (eg temperature salinity currents) that may be 2D (surfacebottom) or 3D Biogeochemical variables of interest (eg pH aragonitesaturation state oxygen concentration primary and secondary pro-duction) are similarly needed at different depth levels depending on theapplication though depth-integrated values may also be sufficient insome cases Metrics for physical and biogeochemical forecasts will alsovary as different applications may be concerned with the absolute stateof the ecosystem deterministic or probabilistic anomalies or pheno-logical changes Finally characteristics of living marine resources(LMR) are targeted in forecast applications that include improving theefficiency of fishing practices (eg Eveson et al 2015 Kaplan et al2016 Turner et al 2017) mitigating impacts of public health threatssuch as HABs (eg Stumpf et al 2009) and alleviating human threatsto protected species (Thorne et al 2019) LMR occurrence abundanceand distribution are some of the most common quantities forecasted inecological applications (Table 1 Tommasi et al 2017a Payne et al2017) as they often respond to environmental variability in predictableways

3 Marine ecosystem forecasting methods

A number of dynamical and statistical approaches are available forS2I forecasts (Fig 2) each with advantages and disadvantages Dyna-mical prediction systems can resolve non-linearities in natural systemsand use them to improve model skill and can forecast unprecedentedsituations or nonstationarity in the climate system In contrast statis-tical prediction systems generally require long historical records formodel training and development and are not well equipped to

Fig 2 Summary of dynamical and statis-tical marine ecosystem forecasting ap-proaches Numbers accompanying in-dividual arrows indicate correspondingsections in the text Note that the LinearInverse Model is an ldquoempirical dynamicalmodelrdquo as it evolves the system forward intime but is built on statistical relationshipsand for our purposes is included with sta-tistical methodsAdapted from Capotondi et al (2019a)

anticipate unprecedented conditions or handle nonstationarity in phy-sical-biological systems However statistical prediction systems aremuch less resource intensive than their dynamical counterparts are notsubject to biases that arise from model error in dynamical predictionsystems and offer an alternative for systems (especially in ecologicalapplications) in which we lack the understanding needed to constructand parameterize dynamical models Several statistical and dynamicalforecast approaches are described in more detail below given theircomplementary nature both are needed for maximum realization offorecast skill

31 Physical forecasting methods

311 Coupled global climate forecastsS2I forecasts from dynamical coupled prediction systems are now

run routinely at multiple operational centers (Graham et al 2011)Most commonly dynamical prediction systems consist of several com-ponent models each with detailed implementation of part of the earthsystem (ie atmosphere ocean sea ice land) that are coupled tocapture climate feedbacks and controls A number of these predictionsystems additionally include ocean biogeochemical fields such ascarbon nutrients oxygen phytoplankton and zooplankton Severalnational or international programs including the North AmericanMulti-Model Ensemble (NMME Kirtman et al 2014) and the WMOLead Centre for Long-Range Forecast Multi-Model Ensemble (LC-LRFMME Graham et al 2011 wwwwmolcorg) have been developedto consolidate predictions from individual modeling centers and enableinvestigation and use of multi-model ensemble forecasts On timescalesbeyond one year (and up to 10 years) forecasting efforts are beingadvanced primarily through the World Climate Research Program(Kushnir et al 2019) phases 5 and 6 of the Coupled Model Inter-comparison Project and nascent efforts at various operational centers(eg the Community Earth System Model Decadal Prediction LargeEnsemble Yeager et al 2018) While they are not the focus of thispaper subseasonal (2 week ndash 2 month) forecasts are similarly beingactively developed through programs including the National WeatherService Global Ensemble Forecast System (Hamill et al 2013) theSubseasonal Experiment (SubX Kirtman et al 2017 Pegion et al2019) and the Sub-Seasonal to Seasonal Prediction Project (S2S Vitartet al 2017)

Dynamical prediction systems have been extensively studied fortheir abilities to capture relevant phenomena and predictability sourcesfor seasonal (1ndash12 month) timescales (Huang et al 2014) For ex-ample they have been shown to provide skillful predictions for large-scale modes of climate variability including the El Nintildeo-Southern Os-cillation (ENSO httpsiricolumbiaeduour-expertiseclimateforecastsensocurrent) and associated teleconnections (Trenberthet al 1998) for monthly sea surface temperature anomalies (SSTa) incoastal Large Marine Ecosystems around the US and elsewhere (Stocket al 2015 Hervieux et al 2017) for SST and precipitation anomaliesover insular Hawaii and US-affiliated islands (Annamalai et al 2014)and for sea surface height (SSH) anomalies in much of the tropicalPacific Ocean including around US-affiliated islands (Widlansky et al2017) However the relatively coarse resolution of dynamical predic-tion systems is not well suited for capturing many complex fine-scaleprocesses (eg freshwater discharge tidal forcing coastal upwelling)that are of interest for coastal marine ecosystem prediction In thesecases skillful forecasts of large-scale modes of variability and theirinfluence on regional features together with statistical or dynamicaldownscaling techniques discussed in the following sections may pro-vide more useful information

In addition to real-time forecasts many climate prediction centersprovide extensive sets of reforecasts (ie forecasts made for past per-iods using observations available at the time of the forecast) using thesame model configuration as that used for the real-time forecasts In thecontext of S2I predictions reforecasts are crucial to (i) develop

appropriate forecast calibration methods and correct for dynamicalprediction system biases (which can be of the same order of magnitudeas the quantities being predicted) (ii) assess the skill of the predictionsystem for specific applications and (iii) determine and understandsources of predictability

312 Dynamically downscaled regional forecastsDynamical downscaling allows one to leverage atmosphere and

ocean state estimates from relatively coarse resolution oceanatmo-sphereclimate models while better resolving fine-scale dynamics in aspecific area of interest In marine science dynamical downscaling ty-pically involves running a regional ocean model forced at the oceansurface and lateral boundaries by output from a global model (Fig 2)In the case of retrospective analyses boundary conditions are oftenderived from oceanatmosphere reanalyses while output from globalclimate models or earth system models is used to force dynamicallydownscaled climate projections (eg Auad et al 2006 Sun et al2012 van Hooidonk et al 2015 Hermann et al 2016 Xiu et al 2018)and forecasts (eg Siedlecki et al 2016)

In the case of S2I forecasts that are of interest here dynamicaldownscaling has the potential to resolve fine-scale responses to pre-dictable large-scale climate forcing For example global models cangenerally forecast ENSO-related wind stress anomalies in the NortheastPacific with some skill but they are unable to resolve the regionalocean response to those anomalies In particular anomalies of upwel-ling intensity and other key variables (eg temperature pycnoclinedepth nutrient fluxes) are greatest within tens of km of the coast inglobal models these processes are sub-grid scale resulting in signalsthat are too weak and diffuse (Jacox et al 2017) Similarly globalmodels with data assimilation can reasonably simulate open oceanvariability in the Northwest Atlantic (such as Gulf Stream variability)while downscaling using a regional model is needed to accurately si-mulate the responses of Northeast US Shelf circulation to the openocean forcing (Chen and He 2015) However increased resolutionalone does not correct all shortcomings of global forecasts While re-gional models can resolve fine-scale anomalies they rely on globalmodels to impart the forcing that generates those anomalies An im-portant area of investigation is the degree to which signals that arepoorly resolved in global models (eg coastal trapped waves polewardundercurrents buoyancy-driven coastal currents) can be transmittedthrough regional model boundaries into the domain where they arebetter resolved Furthermore regional models inherit not only skill butalso bias from global models In both of these respects downscaledforecasts may benefit from global fields that are statistically correctedto remove biases and recover fine-scale variability at the lateral andsurface boundaries of the regional domain (see Section 314)

There are several active efforts to produce dynamically downscaledseasonal ocean forecasts along US coastlines (Fig 1 Table 1) In theNortheast Pacific these efforts include one for the OregonWashingtonshelf (JISAOrsquos Seasonal Coastal Ocean Prediction of the EcosystemJSCOPE Siedlecki et al 2016) one for the California Current Systemusing the UC Santa Cruz CCS configuration of the Regional OceanModeling System (ROMSe) (Veneziani et al 2009) and one for theeastern Bering Sea using the Bering10K ROMS configuration (Hermannet al 2013 Ortiz et al 2016 Kearney et al 2019) While each of theseefforts shares a common overall approach they differ in their detailsand intended applications (Table 1) Dynamically downscaled S2Iocean forecasts along the east coast are in a relatively nascent stagepartly due to the absence of a dominant interannual climate signal toimpart predictive skill (as ENSO does along the US west coast) Cur-rently operational forecast systems for the east coast eg the MARA-COOS models for the Mid-Atlantic Bight (httpassetsmaracoosorg)the US Northeast Coastal Forecast System (httpfvcomsmastumassdedunecofs) and the coupled Northwest Atlantic PredictionSystem (httpomgsrv1measncsuedu8080CNAPS) target muchshorter lead times up to 72 h

313 Statistical downscaling and forecastingStatistical downscaling offers an approach for predicting marine

ecosystem variability based on the underlying premise that regionalecosystem variability can be explained by large-scale climate dynamicsand their interaction with regional influences including coastal oro-graphy bathymetry shelf-basin exchange and characteristics of theland-sea interface The approach generally involves developing quan-titative relationships between larger-scale climatic variables (pre-dictors) that are typically well predicted by climate and weatherforecasting models and local variables and conditions of interest(predictands) (Fig 2) There is a large variety of statistical forecastingapproaches some of which are univariate namely they relate thepredictand to just one predictor (eg Davis et al 2017) while othersincorporate multiple predictors (multivariate approaches) In manycases the predictor is a single variable representing some pattern oflarge-scale ocean andor atmosphere variability eg the Gulf Streamvariability (Davis et al 2017) the North Atlantic Oscillation (Xu et al2015) or ENSO (Faggiani Dias et al 2018) Widely used multivariateapproaches rely on multiple linear regressions for determining the re-lative influence of different predictors at a given lead time (egSanchez-Franks et al 2016) but more advanced non-linear statisticaltechniques have also been used particularly for predictions of atmo-spheric variables (eg Gaitan et al 2014) Statistical downscaling hasbeen successfully used for S2I forecasting of climate variables to reduceuncertainty associated with application of downscaled climate in-formation across multi-scale models and for the development of futureclimate scenarios (Schoof 2013 Wilby et al 2004) In all cases thepotential of statistical methods for forecasting depends substantially onthe availability of consistent reliable observational data sets (eg insitu or satellite) that adequately capture large-scale atmosphereoceanvariability regional oceanic responses and relationships between the

twoOne type of statistical downscaling used to better understand

complex time-space mechanisms associated with weather variation andoceanic responses is synoptic climatology (Lee and Sheridan 2015)Synoptic climatology represents a holistic approach in first categorizingatmospheric conditions into one of multiple modes of variability andthen assessing the relationship between these atmospheric modes andan environmental outcome (Yarnal 1993) The predictor variables aretypically derived from globalregional reanalysis data sets and can in-clude atmospheric properties near the earthrsquos surface such as sea-levelpressure or winds or higher in the atmosphere such as 700 hPa or850 hPa air temperature or geopotential height The synoptic metho-dology serves as an effective statistical downscaling tool for outputfrom weather forecasting models or global climate models (Schoof2013) and in some cases has proven to be superior to finer-scale model-generated atmospheric data (Wetterhall et al 2009) Synoptic clima-tology is especially useful for forecasting on timescales of a week orlonger where models would be unable to render atmospheric featureswith sufficient precision for dynamic modeling as in the case of coastalecosystems where shelf- and estuarine-scale processes and air-sea in-teractions are primarily controlled by weather events and not clearlylinked to global scale climate forcing (Sheridan et al 2013 Pirhallaet al 2015)

A unique multivariate approach is Linear Inverse Modeling (LIM)where information on the statistics of the system is used to develop amodel of the system with basic features that are independent of theforecast lead time The LIM framework describes the system of interestin terms of an anomaly state vector usually constructed from monthlyanomalies of the key system variables whose evolution is modeled interms of linearly damped and stochastically perturbed dynamics(Penland and Sardeshmukh 1995) The LIM framework has been

Fig 3 July 2006 SST in the California Current System (CCS) from (a) the CanCM4 global forecast system (Merryfield et al 2013) (b) a 10 km regional ocean modelsimulation using the Regional Ocean Modeling System (ROMS) forced by CanCM4 fields that were interpolated directly to the ROMS grid (c) a ROMS simulationforced at the surface and lateral boundaries by CanCM4 fields that were first bias corrected and (e) the observed ocean state as obtained from a ROMS CCS historicalreanalysis (Neveu et al 2016) (d) Monthly climatological SST in a central CCS region (345ndash43degN 0ndash100 km offshore black lines in a-ce) with line colorscorresponding to labels in other panels

extensively applied to the study of ENSO in the tropical Pacific (Penlandand Sardeshmukh 1995 Newman et al 2009 2011 Capotondi andSardeshmukh 2015 2017) and for examining the prediction of NorthPacific SSTa and the Pacific Decadal Oscillation (PDO Alexander et al2008) Recent studies have shown the LIM to have seasonal forecastskill comparable to that of the NMME for conditions in the tropicalPacific (Newman and Sardeshmukh 2017) and North Pacific (FaggianiDias et al 2018)

314 Hybrid statisticaldynamical approachesSeveral forecasting efforts leverage both dynamical and statistical

methods and their respective strengths to improve understanding ofregional scale ocean responses to broad-scale forcing (Fig 2) In oneapproach atmospheric andor oceanic fields from global models arestatistically corrected over the region of interest before using them toforce a regional ocean model Statistical correction of the global fieldsmay be as simple as bias correction or may use more involved methodsincluding canonical correlation analysis (eg Huth 2002 Fowler et al2007) or multiple linear regression (eg Goubanova et al 2011) Atthe ocean surface statistically downscaling the wind forcing has beenshown to improve the representation of ocean dynamics in EasternBoundary Upwelling Systems by more accurately capturing featuresthat drive environmental gradients near shore (Machu et al 2015) Ifglobal fields are directly interpolated onto the regional model grid theincreased regional resolution alone will not necessarily correct biasesinherited from the global model a problem alleviated by statisticalcorrection of the surface forcing (Fig 3) Similarly statistical down-scaling of ocean conditions at the lateral boundaries can reduce thetransfer of biases into the regional domain and may improve thetransfer of features such as coastal trapped waves which have anidentifiable signature in global models even though they are not re-solved

A second approach aims to retain benefits of dynamical downscalingwhile reducing concerns related to computational expense and thelimited availability of forcing fields As in the synoptic forecasting ap-proach described in the previous section output from a small ensembleof dynamical downscaling runs is used to establish multivariate statis-tical relationships between the large-scale multivariate forcing and thesmall-scale multivariate response (Hermann et al 2019) Once suchrelationships are established these relationships may be used to sta-tistically project global forcing from a wider ensemble of models (tocapture more structural uncertainty) and more realizations of eachmodel (to capture more intrinsic uncertainty) onto the dynamicallyderived multivariate modes effectively yielding a downscaled en-semble much larger than would be feasible with dynamical down-scaling alone

32 Biogeochemical and ecological forecasting methods

321 Dynamically downscaled regional forecastsIn a subset of global forecast systems (Section 311) biogeochem-

ical variables are included in addition to physical ones enabling thedynamical downscaling approach (Section 312) to be extended fromphysics to biogeochemistry However downscaling biogeochemicalfields introduces the added challenge that variables available at theglobal scale and those in the regional model may not fully correspondor may be governed by a different set of equations This mismatchcomes primarily from the fact that even though most biogeochemicaland lower trophic level ecosystem models adhere to a similar generalformulation (ie nutrients-phytoplankton-zooplankton-detritus) thelevel of complexity with which each component or trophic level is re-presented typically varies (eg number of limiting nutrients or phy-toplankton functional groups) as do the details of relationships be-tween model components In addition the resolution of the modelsdictates the ability for the model to simulate the processes being re-solved For example coarse resolution models cannot resolve shelf

processes essential for capturing hotspots of respiration of organicmaterial on the shelf which result in corrosive and low oxygen con-ditions in the Northern CCS (Siedlecki et al 2015 2016) On the otherhand global models may carry additional nutrients (eg iron) andresolve additional functional groups (eg diazotrophs) that are notessential for capturing the dominant modes of biogeochemical andecosystem variability in a region of interest

Ideally the global and regional models would use biogeochemicaland ecosystem formulations of identical complexity to guaranteecompatibility (eg Van Oostende et al 2018) but this approach isoften impractical for several reasons (1) the models used in thedownscaled regional domain are generally application-specific andemphasize different biogeochemical processes (eg carbon cycling topredict ocean acidification or phytoplankton-zooplankton assemblagesto predict forage fish species distributions) (2) significant effort andlocal knowledge have already been devoted to the complexity cali-bration and evaluation of models used in the downscaled regional do-main and (3) global earth system models do not use a common for-mulation for biogeochemistry so a regional model could only bematched to one or a subset of global models In regions where physicaland biogeochemical variability is generated predominantly by localforcing the predicted ecosystem response may be less sensitive to dis-crepancies in the downscaled variables at the modelrsquos open boundariesIn contrast for a region strongly influenced by remote forcing a morecareful matching of the biogeochemical and ecosystem variables maybe necessary (eg additional downscaling of phytoplankton and zoo-plankton functional groups common to the global and regional modelsor aggregating total phytoplankton biomass and redistributing it intoregional functional groups based on local ecosystem properties) Un-derstanding the sensitivity of regional ecosystem responses to thesebiogeochemical and ecological formulations is an important area forfuture research

When the coarser model being used to force a regional model doesnot include a biogeochemical or ecosystem component regionaldownscaling is still possible (eg Fennel et al 2006 Davis et al 2014Siedlecki et al 2015 2016) However it is essential to ensure that thelarge-scale biogeochemical variability and the predictability it impartsis appropriately downscaled to the regional simulation Therefore someapproaches such as relying on climatological nutrient conditions at theboundaries are not suitable for applications where nutrient variabilityoutside of the regional domain significantly influences the ecosystemvariability inside the domain One approach to recover large-scalebiogeochemical variability from a physical global model is to derivebiogeochemical fields (eg oxygen nitrate DIC total alkalinity) fromphysical fields (eg temperature salinity) using local empirical re-lationships (eg Davis et al 2014 Siedlecki et al 2015 2016) thathave in some cases already been established from local observations(eg Alin et al 2012 Jacox et al 2018) This approach ensures thatbiogeochemical variability associated with large-scale anomalous phy-sical conditions is translated to the downscaled simulation but it islimited by the stationarity of the empirical relationships being appliedFrom a predictability standpoint it remains to be demonstrated whe-ther errors introduced by statistical downscaling of global fields in-troduces larger uncertainties in the regional solution than those in-herently associated with the global model output especially whenconsidering impacts on broader ecosystem characteristics such as dis-tribution or abundance of higher trophic level species

322 Statistical forecastsMany ecological forecasts follow a lsquobottom-uprsquo approach in which

physical forecasts are linked to ecological components of interest mostoften using empirical statistical models (Fig 2 Table 1) Ecologicalforecasting typically relies on four components (i) skillful forecasts ofphysical variables (ii) predictable responses of an ecological metric ofinterest to physical change (iii) relevance to and engagement withforecast end-users (Payne et al 2017 Dietze et al 2018) and (iv)

timeliness of the forecasts to be included within the end-user decision-making process and timeline (Zador et al 2016) In lieu of process-based or mechanistic models of ecological dynamics for which theo-retical knowledge and detailed data required for parameterization(Buckley and Kingsolver 2012) may be lacking statistical models(sometimes termed correlative models) offer a pragmatic approach toecological forecasting These models can approximate mechanisticforcing through correlations (Dormann et al 2012) but they should bedeveloped with a priori mechanistic understanding (Fourcade et al2018) and must account for both physical forecast error and un-certainty in empirical statistical relationships that can further degradeforecast skill (eg Turner et al 2017 see also Section 422) Alter-natively some ecological forecasts rely on ecological lags rather thanphysical forecasts to generate predictability Such forecasts can bebased on the life history of individual species for example springsummer abundance of the toxic dinoflagellate Alexandrium fundyensein the Gulf of Maine is forecast based on counts of resting cysts (theirdormant stage) the previous fallwinter (Anderson et al 2014) Insome cases observed environmental conditions are statistically corre-lated with some later response in the ecosystem Ocean temperaturemeasurements can be used to predict the beginning of the annual lob-ster migration in Maine with 1ndash3 months lead (Mills et al 2017) andsilver hake distributions on the Northeast US shelf have been found tolag shifts in the Gulf Stream path by roughly 6 months a relationshipmediated by the covariance of Gulf Stream position and bottom tem-perature on the shelf (Nye et al 2011 Davis et al 2017) In othercases ecological forecasts are generated by extrapolating from time-series data without the need for covariates or exogenous variables(Stergiou and Christou 1996 Stergiou et al 1997) This type of ap-proach is most commonly used to forecast annual fisheries landings buthas broader application to forecasting populations in general (Wardet al 2014)

4 Mechanisms of predictability in marine ecosystems

In this section we document mechanisms that impart predictabilityto physical biogeochemical and ecological properties of marine eco-systems These mechanisms include oceanic processes atmospheric andcoupled ocean-atmosphere phenomena sea-ice dynamics biogeo-chemical and ecological responses to environmental forcing and lifehistory properties of marine populations (Table 2) Together thesemechanisms have the potential to be leveraged in marine ecosystemforecasts for wide-ranging applications We focus on predictability as-sociated with these mechanisms on S2I timescales though in somecases they have broader application (eg see Liu and Di Lorenzo (2018)for mechanisms associated with predictability of Pacific decadalvariability) Similarly our discussion uses regionally specific examples(ie highlighting the manifestation of mechanisms along NorthAmerican coasts) but they occur throughout the worldrsquos oceans

41 Mechanisms of physical predictability

411 Oceanic processes4111 Persistence Ocean anomalies once formed by processes asdiverse as surface fluxes and advection can persist for extended periodsin a given location This persistence of anomalies carries inherentpredictability which is much higher in the ocean than in theatmosphere For example given the high specific heat capacity ofwater anomalies in temperature and other ocean variables can remainin place for months enabling skillful forecasts on S2I timescales basedon persistence alone This persistence can be influenced by a wide rangeof dynamical processes and is seasonally dependent at mid-latitudesdue to changes in the mixed layer (ML) depth and consequently itsthermal inertia

Among North American Large Marine Ecosystems (LMEs) those inthe Pacific (Eastern Bering Sea Gulf of Alaska California Current

Insular Pacific Hawaiian) exhibit higher SSTa forecast skill from per-sistence than those in the Atlantic (Southeast US Continental ShelfNortheast Continental US Shelf) (Fig 4 Hervieux et al 2017) ForLMEs in the North Pacific persistence SSTa forecasts provide significantskill for lead times of at least 3 months regardless of when they areinitialized and for up to 12 months (and likely longer) for certain in-itialization months (Fig 4) In the Northeast and Southeast US LMEspersistence forecast skill for SSTa typically extends only 1ndash2 months(Fig 4) though Gulf of Maine SSTa formed in early spring can persistfor ~6ndash8 months much longer than those formed in early summer(Hervieux et al 2017 Chen and Kwon 2018) In the Mid Atlantic Bight(MAB) the relative contributions of the atmosphere-ocean heat flux andocean advection and their seasonality influence decorrelation time-scales of temperature anomalies which can vary by a factor of two dueto strong interannual variability in both atmospheric and oceanic pro-cesses As a result the predictability of spring temperature anomalies inthe MAB is tractable on relatively short (le 2-month) timescales (Chenet al 2016) On both the east and west coasts decorrelation timescalesand associated persistence forecast skill are depth dependent depth-averaged temperature or water column heat content in the Gulf ofMaine has longer decorrelation timescales and greater predictabilitythan shelf temperature in the MAB (Chen et al 2016) Similarly sub-surface anomalies (eg bottom temperature) in the Northern CCS areforecast with greater skill than surface anomalies (Siedlecki et al2016) the reasons for this finding are under investigation but in-creased persistence at depth likely plays a role

4112 Re-emergence Seasonal variations in the depth of the surfaceML can influence the evolution of ocean temperatures Outside thetropics temperature anomalies that form at the surface and spreadthroughout the deep winter ML remain at depth after the ML shoals inspring These anomalies are incorporated into the summer seasonalthermocline between approximately 20 and 100 m depth in the NorthPacific where they are insulated from damping by surface fluxes Whenthe ML deepens again the following fall the deep anomalies are re-entrained into the surface layer and influence SST In addition topredictability of winter SSTa based on observed conditions the prior

Table 2Summary of mechanisms underlying seasonal-to-interannual (S2I) marineecosystem predictability along US coastlines including associated timescales ofpredictability and the sections of the text in which they are discussed Thetimescale refers to marine ecosystem predictability not necessarily the me-chanism itself for example tropical ndash extra-tropical atmospheric teleconnec-tions can be quite fast (ie weeks Alexander et al 2002) but may impartpredictability on much longer timescales in the ocean (eg 1ndash12 months Jacoxet al 2017) We focus on the dominant timescales of predictability in the S2Itimeframe though in some cases these mechanisms are associated with pre-dictability on shorter (eg for persistence) or longer (eg for advection)timescales as well

Mechanism Timescales of predictability Section

Physical predictabilityPersistence 1ndash10 months 4111Re-emergence 6ndash12 months 4112Coastal waves 1ndash3 months 41131Baroclinic Rossby waves 1 month ndash 2 + years 41132Advection 1ndash2 + years 4114Tropical ndash extra-tropical

connections1 monthndash2 years 412

Sea-ice processes 1ndash12 months 413

Biogeochemical and ecologicalpredictability

Biogeochemical response tophysical forcing

Consistent with physicalpredictability

Species response to environmentalchange

Consistent with environmentalpredictability

Species life history 1ndash2 + years 423

winter this process can generate predictability of subsurfacetemperature anomalies (below the ML) in summer First noted in SSTrecords by Namias and Born (1970 1974) and later termed theldquoreemergence mechanismrdquo (Alexander and Deser 1995) this processhas been shown to occur over large portions of extratropical oceansincluding coastal regions (eg Alexander et al 1999 2001 Hanawaand Sugimoto 2004 Byju et al 2018) and a similar process can occurfor salinity anomalies (Alexander et al 2001)

4113 Ocean waves In addition to the familiar wind-driven surfacewaves other types of ocean waves can influence marine ecosystems onmuch longer time and spatial scales Some like Kelvin wavespropagate along boundaries including the equator as well as thecoasts of continents The effects of Kelvin and other ldquocoastallytrapped wavesrdquo are confined to a narrow near-shore region on theorder of tens of kilometers wide Others like Rossby waves (alsoreferred to as planetary waves) can be thousands of kilometers long andmove slowly westward across the open ocean outside of the equatorialband The vertical structure of the ocean with a rapid change in densityin the pycnocline leads to a prominent class of waves with a ldquofirstbaroclinic mode structurerdquo which produce anomalies of opposite signin sea surface height and pycnocline depth While slower than thesurface waves first baroclinic mode Kelvin waves move relatively

rapidly and can propagate along the eastern Pacific margin from theequator to Alaska in ~2 months First baroclinic mode Rossby wavespropagate relatively slowly taking several years to cross the PacificOcean from east to west Predictability associated with these waves is ofinterest as they alter ocean currents and hydrography and can havelonger term effects on ocean physics and biology even after the waveshave moved on

41131 Coastal waves Coastal boundaries support baroclinicKelvin and other coastal trapped waves that propagate at speeds of~200 km dayminus1 (eg Gill 1982) with the boundary on the right (left)in the northern (southern) hemisphere and thus act as a source ofpredictability in the downstream region In the Pacific basin baroclinicmode equatorial Kelvin waves with SSH and thermocline depthanomalies of opposite sign are an integral part of ENSO events Windvariations along the equator generate equatorial trapped Kelvin wavesthat propagate eastward and excite Kelvin and other coastal trappedocean waves upon reaching the continental boundary Along their pathof propagation these coastal waves also are driven by variability inlongshore winds whose influence increases with latitude along theNorth American west coast (eg Enfield and Allen 1980) Coastaltrapped waves have been observed to propagate to high latitudes alongthe west coasts of North and South America generating substantialvariability in SSH coastal currents and thermocline depth (Enfield and

Fig 4 SSTa reforecast skill measured by anomaly correlation coefficient (ACC) as a function of lead time and initialization month for seven Large Marine Ecosystemsalong North American coasts Gray dots indicate ACC significantly above 0 at the 95 confidence level White triangles indicate ACC significantly above persistenceat the 90 level with upward triangles showing ACC gt 05 and downward triangles showing ACC lt 05 Forecast SSTa was evaluated against the NOAA OptimumInterpolation SST version 2 (Reynolds et al 2007 Banzon et al 2016)Adapted from Hervieux et al (2017)

Allen 1980 Chelton and Davis 1982 Clarke and van Gorder 1994Frischknecht et al 2015 Jacox et al 2015a) and potentially affectinglower trophic level production (Clarke and Dottori 2008) While thepropagation of coastal trapped waves is a clear mechanism forpredictability of downstream conditions the trapping scale of thesewaves is typically on the order of tens of kilometers so they are sub-grid scale phenomena in current global climate models High-resolutionregional ocean models have the potential to better harnesspredictability associated with coastal trapped waves (eg Kurapovet al 2017) though proper modeling of coastal trapped wavepropagation must account for frictional decay as well as wavescattering at topographic features

Similarly coastal trapped waves could carry anomalous signalsalong the Northwestern Atlantic coast from the Labrador Sea to theNortheast US shelf This coastal (or topographic) waveguide has oftenbeen suggested as a major equatorward pathway for high-latitude sig-nals associated with the deep limb of the Atlantic meridional over-turning circulation (Kawase 1987 Yang 1999 Johnson and Marshall2004) However the viability of coastal trapped wave propagation as amechanism of predictability along the North American east coast ismuch less clear than it is along the west coast The topographic wa-veguides on the east coast are primarily aligned with the continental

slope well offshore in comparison to the west coast and it is not yetclear how these propagating signals impact the broader shelf environ-ment In addition there are some dynamical barriers to coastal trappedwaves along the east coast including the Northwest Corner near theeastern edge of Grand Banks where the Gulf StreamNorth AtlanticCurrent impinges upon the continental slope (Rossby 1996) and theLaurentian Channel where a significant freshwater plume exits the Gulfof St Laurence (Richaud et al 2016) Previous studies have shown thatequatorward propagation of anomalies from the subpolar gyre alongthe Northwest Atlantic shelf and upper slope is predominantly drivenby advection (Section 4114 Chapman and Beardsley 1989 Rossbyand Benway 2000 Pentildea-Molino and Joyce 2008 Shearman and Lentz2010 Xu et al 2015) which is a much slower process than coastaltrapped wave propagation

41132 Baroclinic Rossby waves In the extratropics firstbaroclinic Rossby waves readily apparent in SSH measurements fromsatellites (Fig 5) propagate westward at speeds of ~2ndash8 cm sminus1

depending on latitude (Chelton and Schlax 1996) Rossby waves canform near the coast from refraction of coastal waves and subsequentlypropagate westward across the North Pacific (Jacobs et al 1994Meyers et al 1996 Clarke and Dottori 2008) However open-oceanwind forcing rather than eastern boundary waves appears to be the

Fig 5 (top) Observed (CMEMS satellite al-timetry) and forecast (CFSv2 at 0 and 6-month leads) monthly SSH anomaliesaveraged over the 18ndash23degN latitudinal bandin the Eastern and Central Pacific as afunction of longitude and time The verticalgreen line indicates the longitude ofHonolulu Hawaii (bottom) RetrospectiveSSH anomaly forecast skill for CFSv2 (blue)and persistence (orange) forecasts in a2deg times 2deg region centered on Honolulu Lineartrends were removed from forecasts as wellas the CMEMS verification dataset prior tocalculating skill (For interpretation of thereferences to colour in this figure legend thereader is referred to the web version of thisarticle)

dominant driver of Rossby wave formation in the North Pacific (egMiller et al 1997 Capotondi and Alexander 2001 Fu and Qiu 2002Andres et al 2011) Once formed the relatively slowly propagatingRossby waves provide a mechanism for SSH predictability within theocean basins and along western boundaries Rossby wave propagationis readily apparent as diagonal lines in Hovmoumlller (time-longitude)plots of SSH (Fig 5) In the central Pacific around Hawaii sea levelvariability is affected by westward-propagating planetary Rossby waves(Firing et al 2004) as well as by regional wind-stress anomalies viadynamical (eg Ekman pumping) and thermodynamical (egevaporative cooling) oceanic processes (Long et al 2020) Whileglobal forecast systems are not able to resolve the latter fine-scaledrivers of coastal sea level variability on seasonal timescales they areable to skillfully predict large-scale SSH anomalies associated withRossby wave propagation in regions including the tropical Pacific(Widlansky et al 2017) and around Hawaii (Fig 5)

The generation and structure of Rossby waves in the North Atlanticappear to be more complex than in the Pacific (Osychny and Cornillon2004) Nevertheless Rossby wave propagation has been shown togenerate predictability for sea-level anomalies in Bermuda (Sturges andHong 1995) and along the US east coast (Hong et al 2000) and to alesser degree SSTa in some portions of the basin (Zhang and Wu2010) On interannual and longer timescales wind-generated Rossbywaves can influence Gulf Stream transport (eg Sturges and Hong2001) and on reaching the western boundary generate fast boundarywaves that modulate the annual cycle of sea level along the east coast(Calafat et al 2018) While Gulf Stream transport is difficult to predicton sub-annual timescales it has important implications for the pre-dictability of sea level in east coast regions prone to nuisance flooding(Sweet et al 2014 Ezer 2016)

4114 Advection Changes in along-shore currents can lead toanomalous coastal conditions and may provide a basis for physicalpredictability For example during 2002 anomalously cold freshconditions along much of the North American west coast

characteristic of subarctic waters were attributed at least in part to astrengthening of the equatorward California Current (Barth 2002Bograd and Lynn 2003 Freeland et al 2003 Murphree et al 2003)When temperature anomalies are balanced by salinity anomalies so thatthere are no density perturbations (ie spiciness anomalies) heatcontent anomalies can be advected over long distances For examplesubsurface ocean temperature and salinity anomalies can propagateslowly eastward along isopycnals advected by the mean North Pacificcurrents and influence ocean conditions along the west coast of NorthAmerica years later (Taguchi and Schneider 2014 Pozo Buil and DiLorenzo 2017 Taguchi et al 2017)

Similarly the poleward flowing California Undercurrent (CUC)transports relatively warm and salty Pacific Equatorial Water as farnorth as the Aleutian Islands affecting the properties of NorthAmerican west coast shelf and slope waters along its path (Reed andHalpern 1976 Hickey 1979 Bograd et al 2008 Thomson andKrassovski 2010 Connolly et al 2014 Nam et al 2015 Turi et al2016 Durski et al 2017) Changes in the depth of the CUC appear to beespecially important more so than changes in the composition of theCUC water for influencing shelf waters through seasonal upwelling(Meinvielle and Johnson 2013) Therefore predicting undercurrentvariability has the potential to contribute to predictability of bothsurface and subsurface conditions in the CCS While the CUC is tooweak and not properly resolved in global climate models (Hickey et al2016) its presence is promising in terms of mechanistic pathways forpredictability of ocean conditions within the CCS For example the CUCin the Climate Forecast System Reanalysis (CFSR) strengthens anddeepens in summer to early fall and periodically surfaces along theentire CCS consistent with observations in the region (Thomson andKrassovski 2010 Pierce et al 2000 Durski et al 2017) While globalforecast systems have yet to be evaluated with respect to the CUCaccurate prediction of variability in its strength and position wouldlikely impart predictive skill for oceanographic conditions along thewest coast and this skill may be enhanced through dynamical down-scaling with regional ocean models that can better resolve the CUC

Fig 6 SSTa forecast skill in the CCS as a functionof initialization month and lead time for (a) per-sistence forecasts (b) the CanCM4 global forecastsystem (c) a forecast that uses CanCM4 for ENSO-neutral years and persistence for years followingmoderate-to-strong ENSO events (1983 19871988 1989 1992 1998 1999 2000 2003 2008)and (d) a forecast that uses CanCM4 for years fol-lowing moderate-to-strong ENSO events and per-sistence for all other years White dots indicatesignificant skill above persistence (95 confidencelevel) Anomaly Correlation Coefficient (ACC) wascalculated for 1982ndash2009 using NOAArsquos 025degOISSTv2 as truth

In the Northwest Atlantic large-scale ocean circulation such as theGulf Stream has been linked to ocean temperature on the Northeast UScontinental shelf offering promise for ecosystem predictability (Nyeet al 2011) The position of the Gulf Stream in turn has been found torespond to NAO extrema with ~1-year lag (Joyce et al 2000Frankignoul et al 2001) Therefore the dynamical and statistical linkbetween the large-scale ocean circulation and the Northeast US shelftemperature provides a robust source of predictability for the coastalenvironment on the Northeast US shelf In addition to the Gulf Streamtransport along-shelf advection from the Labrador Sea has been shownto have major impacts on downstream temperatures (Chapman andBeardsley 1989 Rossby and Benway 2000 Shearman and Lentz 2010Xu et al 2015) Pentildea-Molino and Joyce (2008) showed it takes oneyear for SSTa to propagate equatorward from Nova Scotia to CapeHatteras along the continental slope In the Gulf of Maine subsurfacetemperature anomalies have been shown to lag the NAO by two years(Mountain 2012) while SSTa can be traced upstream to the LabradorShelf four years earlier (Xu et al 2015) Additional evidence of theimpact of alongshore geostrophic advection on shelf temperature waspresented by Forsyth et al (2015) who found a significant correlationbetween coastal sea level and depth-averaged MAB shelf temperaturestwo years later though predictive skill associated with this correlationneeds to be explored further Further influences of alongshore currentsfor example Gulf Stream warm-core rings that can induce significantshelf-slope exchange (eg Joyce et al 1992 Chen et al 2014 Zhangand Gawarkiewicz 2015) influence the hydrography of the shelf altercritical habitats and even introduce species from other regions (egHare and Cowen 1996) are beginning to be explored (eg Monim2017) but are not yet associated with known predictability

412 Tropical-extratopical connectionsWhile the atmosphere itself tends to be less predictable than the

ocean on S2I timescales (eg Davis 1976) atmospheric forcing (egsurface winds and heat fluxes) can give rise to ocean predictabilityoften associated with large-scale modes of climate variability Thedominant interannual climate signal impacting the Northeast Pacificregion is ENSO and one of the ways ENSO impacts the North Americanwest coast is through an atmospheric teleconnection in which tropicalconvection excites atmospheric Rossby waves that modify the strengthand position of the Aleutian Low jet stream and storm track (Trenberthet al 1998) Consequent impacts along the coast during El Nintildeo in-clude poleward (downwelling-favorable) wind stress anomalies warmSSTa and increased precipitation and river runoff along the Californiacoast (eg Alexander et al 2002 Park and Leovy 2004 Jacox et al2015a) ENSO variability has been shown to be the primary source ofseasonal predictability for air temperature and precipitation anomaliesover North America (Quan et al 2006 Peng et al 2012) and for windand precipitation anomalies over Hawaii and US-affiliated islands(Annamalai et al 2014) and the same may be expected for oceanicanomalies that respond to the same teleconnections Indeed along thewest coast SSTa forecast skill above persistence derives primarily frompredictable ENSO-related anomalies (Fig 6) Furthermore a latitudinalgradient in SSTa forecast skill along the west coast with higher skill inthe north indicates that ENSO-related forecast skill in global forecastsystems comes primarily through the atmospheric teleconnection ratherthan the oceanic teleconnection (Section 41131 Jacox et al 2017)The Northern CCS response to ENSO variability is dominated bychanges in the wind while the oceanic teleconnection (ie coastaltrapped wave propagation) dominates the Southern CCS response toENSO (Hermann et al 2009 Frischknecht et al 2015) Since theformer is resolved in relatively coarse resolution global models whilethe latter is not the predictable ENSO-driven variability in the north isbetter captured (Jacox et al 2017) These results from global climatemodels are consistent with those from statistical approaches namelythe LIM (Section 313) which showed a large fraction of North PacificSSTa predictability originating from ENSO likely through the

atmospheric teleconnection given the coarse-grained data used in theLIM (Alexander et al 2008) The degree of influence that any parti-cular ENSO event exerts in the Northeast Pacific is likely to be affectedby the spatial pattern and evolution of the event (Capotondi et al2019b) as well as atmospheric internal variability (Deser et al 2017)

The ENSO teleconnection also influences coastal waters in theNorthwest Atlantic although the impact is not as strong as for the westcoast (Alexander et al 2002 Alexander and Scott 2002 2008) ENSOexhibits a weak negative correlation with the North Atlantic Oscillation(NAO) especially in winter (Li and Lau 2012) which results in a basin-wide tripole SSTa pattern primarily through turbulent surface heat fluxanomalies (Alexander and Scott 2002 Hu et al 2013) SSTa associatedwith ENSO exhibit particularly strong amplitude along the northernflank of the Gulf Stream Northeast US shelf and the Gulf of Mexicocoast (Kwon et al 2010) and thus act as a potential source for pre-dictability for the coastal ocean along the east coast Similarly changesin the North Pacific can potentially be used as a predictor for theNortheast US shelf as the two regions are linked via the Pacific-NorthAmerica atmospheric teleconnection pattern (McKinnon et al 2016Dai et al 2017) In particular the PDO as well as CentralNorth Pacificspring SST precipitation and outgoing longwave radiation anomalieslead Gulf of Maine SSTa by 1ndash3 months (Chen and Kwon 2018) A si-milar lagged response of Long Island Sound air temperature and watertemperature to North Pacific anomalies was found to be consistent withatmospheric Rossby wave trains emanating from the Western Equa-torial Pacific (Schulte et al 2018) It is not yet clear whether the PacificSST anomalies are at least partly driving the atmospheric teleconnec-tions or both the Pacific and Atlantic SST anomalies are responding to acommon teleconnection with a time lag Surface and bottom waterproperties on the Northeast US shelf have also been associated withthe NAO 2ndash4 years earlier (Mountain 2012 Xu et al 2015) thoughthat relationship is mediated by changes in alongshore advection(Section 4114) rather than resulting from direct atmospheric forcing

Though less well studied than the ENSO teleconnections describedabove additional tropicalextra-tropical coupling may extend predict-ability to slightly longer timescales Boreal winter variability in theNorth Pacific Oscillation (NPO) and its oceanic expression the NorthPacific Gyre Oscillation (NPGO) can energize sea level pressure andsurface temperature anomalies the following winter through a tropicalbridge (Di Lorenzo et al 2015 Di Lorenzo and Mantua 2016) Spe-cifically weakening of the off-equatorial winds associated with a po-sitive winter NPO pattern can trigger the growth of meridional modes(Vimont et al 2003 Chiang and Vimont 2004) with SSTa reaching thetropical Pacific in springsummer and activating an ENSO feedbackwith teleconnection back to the extra-tropics the following fallwinterThis connection from extra-tropics to tropics and back again can po-tentially offer predictability on timescales of 1ndash2 years (eg Joh and DiLorenzo 2017 Capotondi et al 2019b) and is predicted to intensifyunder climate change (Liguori and Di Lorenzo 2018)

413 Sea-ice related processes in subarctic regionsThe regional seas of the US subarctic (including the Bering

Chukchi and Beaufort Seas) are seasonally covered by sea ice whoseextent thickness and timing of arrival and retreat vary considerablyfrom year to year In the Eastern Bering Sea sea ice area and thicknessare influenced by advection and local formation and melting which aregoverned in turn by surface forcing and regional oceanography (egCheng et al 2014) and these interactions potentially give rise topredictability of the ocean through several mechanisms For examplethe pan-Arctic ice area has a ldquomelt-to-freezerdquo season memory(Blanchard-Wrigglesworth et al 2011) effectively a reemergencemechanism in which the ice edge in fall returns to where it was inspring (after a retreat in summer) in response to SSTa that have per-sisted in the region of the spring ice melt (Bushuk et al 2014) Arcticice area also has a ldquofreeze-to-meltrdquo season memory which meanssimply that anomalous ice area and thickness in fallwinter persist to

the following spring though dynamical forecast skill often exceedspersistence forecast skill for regional Arctic sea-ice extent (Bushuket al 2017)

In addition to providing predictability for sea ice itself persistenceof sea ice can cascade into predictability of the marine ecosystem as icepresence and subsequent melting impacts ocean properties from springinto summer (Stabeno et al 2012 Brown and Arrigo 2013) In theEastern Bering Sea winter sea ice exerts control over the extent of thebiologically important summer cold pool on interannual timescales(Fig 7) as well as in the multidecadal trend which has shown de-creasedthinning sea ice and reduced cold pool extent since the early1980s (Mueter and Litzow 2008) While the US east coast does notexperience sea ice locally it receives cold and fresh waters of subarcticorigin via advection through an extensive interconnected coastalboundary current system (Section 4114) Sea-ice variability in theLabrador Sea which exhibits considerable forecast skill for lead timesup to at least 11 months (Bushuk et al 2017) could have predictabledownstream impacts on the Northeast US shelf

42 Mechanisms of biogeochemical and ecological predictability

421 Biogeochemical response to physical forcingThe mechanisms of physical predictability described in Section 41

can also lead to predictability of ocean biogeochemical tracers andecosystem variables For example equatorward wind anomalies alongthe North American west coast which respond predictably to ENSOvariability (Section 412) have a twofold impact on the nutrient fluxinto the upper ocean of the CCS region (Jacox et al 2015b) In a ca-nonical El Nintildeo event weaker equatorward winds drive weaker up-welling and also draw waters from shallower ocean depths both ofwhich reduce upward nutrient flux as well as increasing pH along thecoast The opposite is true during La Nintildea when anomalously strongequatorward winds drive strong upwelling increasing the nutrient

supply to the euphotic zone and reducing pH in near-surface coastalwaters (Jacox et al 2015b Turi et al 2018) Durski et al (2017) foundthat in addition to wind-driven upwelling alongshore currents andlocal productivity on the shelf also drive interannual oxygen variabilityin the CCS though the predictability of these latter two drivers is yet tobe quantified Off the OregonWashington coast seasonal forecast skillhas been demonstrated for biogeochemical properties including sub-surface oxygen pH and aragonite saturation state (Siedlecki et al2016) While the mechanisms underlying this skill on interannualtimescales are still being investigated seasonal oxygen declines aredriven jointly by local nutrient trapping and physical transport pro-cesses (Siedlecki et al 2015) including advection in the CaliforniaUndercurrent which has relatively low pH and dissolved oxygen con-tent and relatively high inorganic carbon and nutrient concentrations(Hickey 1979 MacFadyen et al 2008 Pierce et al 2000)

On longer (multiannual) timescales predictability of biogeochem-ical variables in the Northeast Pacific has been attributed to advectionof anomalies in the North Pacific gyre and ocean memory (Chikamotoet al 2015 Sasaki et al 2010 Taguchi and Schneider 2014 Pozo Builand Di Lorenzo 2015 2017) as well as ldquodouble integrationrdquo effects ndashintegration of the atmospheric forcing by ocean physics which are thenintegrated by ocean biogeochemistry (Ito and Deutsch 2010 Kilpatricket al 2011 Di Lorenzo and Ohman 2013)

422 Species responses to environmental changeMarine species behaviors movements growth and mortality are

driven by extrinsic and intrinsic mechanistic forcing and for manyecological metrics applicable to forecasting (Table 1) these mechanisticlinks are approximated by statistical models For example water tem-perature is an important covariate in ecological forecasting and reflectsphysiological functioning and limits (eg thermal tolerance and biolo-gical rates for metabolism growth digestion performance and ac-tivity) In the Eastern Bering Sea the extent and timing of winter sea ice

Fig 7 (a) Eastern Bering Sea sea ice covergt 10 in mid-March (top) and bottom water temperature for the week of July 1 (bottom) during a warm (2004) and acold (2008) year from output of the Bering10K model hindcast Bottom waters with temperaturelt 2degC form the cold pool (b) Ice cover anomaly and bottomtemperature anomaly indices over the Eastern Bering Sea shelf (the negative of bottom temperature anomaly is plotted to enable comparison with ice cover anomly)The ice cover index is the average ice concentration in the region 56-58degN 163-165degW for Jan 1-May 31 Ice concentration data are obtained from the National Snowand Ice Data Center (NSIDC) using the bootstrap algorithm for historical data (through ~2010) and the NASA Team algorithm for more recent data Mean annualbottom temperatures were calculated from NOAANMFS bottom trawl surveys of the Eastern Bering Sea conducted June-August and sampling ~250ndash300 stations ina region spanning approximately 545ndash62degN and 179ndash158degW Both the ice cover and bottom temperature anomalies were standardized by removing their mean anddividing by their standard deviation (c) Cold pool index from observations hindcasts and nine-month lead forecasts from the Bering10K model Predictions of thecold pool are for July 1 (the middle of the summer bottom trawl survey) In 2018 the forecast failed largely due to unprecedented southeasterly winds that preventedsea ice formation Subsequent skill evaluation has suggested skillful prediction is limited to shorter lead times (~3 months) when forecasts are initialized in fall whilelonger lead (~6 month) forecasts have skill when initialized in spring

control springsummer water temperature (Fig 7 Section 413) whichin turn alters the timing of the phytoplankton spring bloom crustaceanzooplankton availability and the condition factor metabolic ratesdemand for prey and survival of young walleye pollock Anomalouslywarm water and associated reduced spring phytoplankton biomass leadto lower abundance of large crustacean zooplankton Juvenile pollockconsumption rates are increased in higher temperatures despite thereduction of available zooplankton prey leading to fish that are rela-tively long but in poor condition (Fig 8 Moss et al 2009 Sigler et al2014 Duffy-Anderson et al 2017) and lowering survival the followingwinter

While the use of a mechanistically sound set of predictor variablesmay reduce the problem of overfitting to the dependence structure ofthe response methods to evaluate transferability will remain essentialfor model selection and skill assessment of empirical models particu-larly when predicting outside the range of observed conditions (Wengerand Olden 2012 Roberts et al 2017 Yates et al 2018) In complexecological systems where biological processes are driven by inter-species relationships and a range of environmental drivers whose re-lative importance shifts over time the most successful empirical eco-logical forecasts may be the ones of biological processes linked to adirect physical driver (eg temperature and growth) rather than onewhose effect is mediated by trophic relationships and other processes(Cuddington et al 2013 Payne et al 2017) Thus a mechanistic linkbetween environmental forcings and biological responses will be key toforecast reliability and first principle approaches may help in this re-spect Ecological models for prediction may also have to be more sim-plistic than those describing historical change as some physical vari-ables included in an explanatory model will not be forecast skillfullyand should be excluded from a predictive model For example a speciesdistribution model for dolphinfish Coryphaena hippurus used fourphysical covariates to describe historical patterns (Brodie et al 2015)but only one physical covariate for a seasonal forecast (Brodie et al2017) Similarly the spatiotemporal scales of environmental variabilityassociated with biological responses must be consistent with the scalesof predictability in the environmental variability itself For exampledecomposing SST variability into a seasonal climatology a low-fre-quency (~interannual) component and a high-frequency component(sub-annual) shows that skill in a swordfish distribution model (Brodieet al 2018) is most strongly associated with the climatological com-ponent followed by the low-frequency and high-frequency components(Fig 9) One would expect some distribution predictability based on theclimatology and skillful forecasts of interannual SST variability (Jacoxet al 2017) but distributional changes associated with high frequency(eg eddy) variability are unlikely to be predictable on S2I timescales

423 Speciesrsquo life historyJust as the forecast horizon for physical ocean variables is longer

than atmospheric ones due to the intrinsically longer timescale of oceanprocesses (Goddard et al 2001) the lead time of ecological forecastscan be enhanced by the relatively slow turnover of animal populationsDeriving predictive skill from the influence of observed populations onfuture year classes requires that the species life history be well studiedwhich is true for many fish species Most of the variability in survival offish populations occurs during early life stages before fish are recruitedto the fishery (Walters and Martell 2004) Therefore outside of thisearly life history knowing the number of fish in one age class can in-form how many fish of the next age class to expect the following yearIndeed stock assessment models use observations of abundancebio-mass to track cohorts over time and make predictions of biomass one tothree years in the future Such predictions are like persistence forecastsas the ldquoobservedrdquo biomass in the current year (actually a model esti-mate obtained from assimilating catch and survey data) is used topredict biomass the next year In such forecasts processes such as re-cruitment growth and mortality that may contribute to gains or lossesin biomass are generally assumed constant at recent levels rather than

mechanistically forecasted based on environmental conditions How-ever multi-species statistical catch-at-age models allow for naturalmortality rates to change annually (Holsman et al 2016) as is done inthe climate-informed multi-species stock assessment for walleye pol-lock Pacific cod and arrowtooth flounder in the Eastern Bering Sea(Holsman et al 2017) To the extent that these models include tem-perature prey quality and other environmental factors the persistenceobserved will be passed on to the population dynamics of the stocksAlternatively a range of plausible productivity (recruitment growthand mortality) values sampled from historical variability can be used toproduce an ensemble of future biomass forecasts The predictabilityhorizon of such fisheries forecasts is dependent on a speciesrsquo life ex-pectancy and population demography with longer lead times for spe-cies that have lower natural and fishing mortality (Fig 10 Brander2003) Starting from an observed abundance natural and fishingmortality rates can be used to estimate the point in time when a po-pulation will be comprised by equal parts of fish that were observed atforecast initialization and subsequent recruits (Brander 2003) Afterthis point skill is less dependent on persistence of observed age classesand more on the validity of model assumptions concerning futureproductivity

For fast-growing short-lived species particularly those in single-cohort fisheries such as squid (Agnew et al 2002) little to no pre-dictive skill can be derived from persistence of the population from yearto year In these cases incorporation of biogeochemical or physicalforecasts is key to enhancing predictability For instance use of a sea-sonal SSTa forecast can enhance the recruitment predictability horizonand lead to more effective catch targets for Pacific sardine (Tommasiet al 2017b) Similarly recruitment forecasts for yellowfin flounderare more accurate when informed by an environmental covariate(Miller et al 2016) For semelparous species like salmon which arerecruited to the fishery as mature spawning adults and die afterspawning recruitment forecasts are essential to assess the strength ofthe incoming adult year-class Ocean conditions impact salmon survivalduring their first few months at sea and that survival is later reflectedin adult populations when they return to spawn Thus observed oceanconditions can be used to forecast recruitment to the fishery based onsalmon life history with lead times up to 2ndash3 years (Peterson et al2014) In many cases environmentally informed salmon abundanceforecasts exhibit increased skill relative to the sibling models commonlyused in management (which rely on estimated freshwater returns topredict ocean abundance or freshwater returns at a later date) (Winshipet al 2015) However the reliability and optimal construction of en-vironmentally informed forecasts vary over time (Winship et al 2015)and their skill is likely a function of the ocean state both bottom-up(food availability) and top down (predation) impacts on salmon sur-vival may be more prominent when ocean conditions are poor (Wells

Fig 8 Dependence of pollock length at age 1 on bottom temperature in theEastern Bering Sea Based on 1975 ndash 2017 samples collected as part of theannual NOAANMFS Eastern Bering Sea summer bottom trawl survey Notethat warmer temperatures are associated with longer fish but those fish tend tobe skinny and in worse condition (Section 422)

et al 2017 Henderson et al 2018)

5 Challenges and priority developments

51 Better resolution of key processes in general circulation models

The last two decades have witnessed remarkable advances in thefidelity of ocean model simulations due to (i) steadily increasing com-putational resources that have enabled enhanced model resolution (ii)development and implementation of sophisticated parameterizations ofunresolved ocean processes (eg vertical mixing) and (iii) algorithmicimprovements to crucial numerical implementations (eg tracer ad-vection schemes) Nonetheless additional improvements in physicalmodels are required to better represent key processes in the future Incoastal waters increases in resolution are particularly important aslength scales of physical variability (eg the first baroclinic Rossbyradius) decrease with ocean depth approaching zero near the coast Forexample small-scale changes in atmospheric forcing (eg due tocoastal orography) and variability in physical boundaries such ascoastlines and bathymetry (eg capes bays canyons and banks)

Fig 9 (a) Daily mean SST in the CCS on Nov 23 2007 (during a La Nintildea event) and Apr 15 2015 (during the 2013ndash2016 Northeast Pacific warm anomaly) takenfrom the UC Santa Cruz historical CCS reanalysis (Neveu et al 2016) Also shown are three temporal components of the SST calculated from 1980 to 2016 daily SSToutput For each 01deg grid cell a monthly climatology is calculated and then removed from the daily SST time series to produce anomalies The anomalies are thensmoothed with a 12-month running mean to produce the low frequency (interannual) component with the remaining signal making up the high frequency (sub-annual) component such that Full = Climatology + Low Frequency + High Frequency (b) Predictive performance of swordfish distribution models built usingBoosted Regression Trees (Brodie et al 2018) with different temporal components of the SST as predictors Model predictive performance is measured using the areaunder the receiver operating characteristic curve (AUC) statistic

Fig 10 Contribution of biological persistence to prediction skill for Pacificsardine Albacore tuna and shortfin mako shark as measured by the percentageof observed abundance remaining in the population in later years All speciesare assumed to experience a fishing mortality rate of 025 yearminus1 Naturalmortality is 06 for sardine (Hill et al 2017) 04 for albacore (ISC 2017) and0128 for mako shark (ISC 2018)

meaningfully influence the coastal ocean circulation Biogeochemicalproperties are likely to exhibit strong gradients on even smaller spatialowing to their fast and sometimes nonlinear responses to the physicsthat drive them (eg Mahadevan and Campbell 1926) In particularupwelling that is a fundamental driver of coastal ecosystems exhibitsconsiderable alongshore heterogeneity in both amplitude and depth ofsource waters and consequently nutrient transport and primary pro-duction Tidal exchanges with estuaries combined with estuarinemixing complicate the physical and biogeochemical signatures of riveroutflows that unavoidably influence local coastal ecosystems Internalwaves generated offshore by the barotropic tide are directed onshore inheterogeneous beams that may reflect at the continental slope and shelf(depending on the local topographic slope and seasonally varyingstratification) or continue onshore ultimately breaking and con-tributing to local vertical mixing and nutrient transport Surface wind-waves influence local circulation and ecosystem elements in at least twoways through alteration of near-surface circulation due to wave-in-duced surface stress and Stokesrsquo Drift of near-surface organisms Finallywe require a much better understanding of the impacts of near-surfacesubmesoscale (1 km or less hours to days) processes including intenselocal vertical velocities local changes in vertical mixing frontogenesisfrontolysis and nonlinear Ekman transport These processes are in-creasingly modeled yet their ecosystem impacts are still unclear andpredictability associated with them remains poorly understood

52 Reforecast skill assessments

Skill assessments in which long-term reforecasts (ie retrospectiveforecasts of past conditions) are tested against atmosphere and oceandatasets or reanalyses (ie data assimilative historical model runs)provide the means to investigate (i) the spatial and temporal predict-ability of particular variables like SST SSH and precipitation (egAnnamalai et al 2014 Stock et al 2015 Hervieux et al 2017Widlansky et al 2017) (ii) the benefits of regional downscaling(Siedlecki et al 2016) and (iii) mechanisms underlying forecast skill(Jacox et al 2017) Metrics used to assess marine forecast skill in theliterature largely mirror those from the meteorological community and

can be deterministic (eg anomaly correlation coefficient root meansquare difference) or probabilistic (eg Brier score) To this pointmarine forecasts have most commonly been assessed for their ability topredict monthly average anomalies of SST and SSH metrics that areinfluenced by large-scale climate variability influence marine speciesor coastal communities and have been adequately measured to enableforecast verification over multiple decades Moving forward more at-tention should be paid to predictability of additional environmentalvariables (Section 2) that influence marine species in different ways aswell as suites of variables that may act as co-stressors for particularorganisms

Verification of rapidly proliferating marine forecasts will requireextension of skill assessments not only to additional physical and bio-geochemical variables but also to integrated quantities deemed im-portant as ecosystem indicators (eg undersaturation days summerhypoxic volume) and to higher trophic level characteristics of interestto end-users Similarly evaluation of ocean hindcasts (historical modelruns forced by observed boundary conditions typically from oceanat-mosphere reanalyses but with no ocean data assimilation) to determinemodelsrsquo ability to simulate processes of interest is a key part of buildingeffective forecast systems (Tommasi et al 2017a Siedlecki et al 2016Hobday et al 2016) While comprehensive skill assessments likelyrequire decades of reforecasts initialized at regular intervals no stan-dard protocol or threshold is currently in place as to the number ofseasons years or cycles of a climate oscillation that must be included inskill assessments in order to provide confidence in the forecasts goingforward Such protocols are likely dependent on details of a givenforecasting application including the specific region and variable(s)but effort should be made to establish protocols for reforecast experi-ments to ensure they are adequate for building confidence in forecasts

53 Forecast uncertainty and ensembles

Sources of forecast uncertainty include initial condition errors aswell as uncertainty due to model errors The distinction between thesetwo sources of forecast uncertainty is useful pedagogically but they arenot independent in a real physical system like the atmosphere or ocean

Fig 11 Dependence of CCS SSTa forecastskill on forecast ensemble size (top) Skill(Anomaly Correlation Coefficient ACC) as afunction of initialization month and leadtime for persistence forecasts individualCanCM4 runs and CanCM4 ensembles of 5and 10 members (bottom) Skill as a func-tion of lead time for persistence the NMMEensemble mean and CanCM4 ensemblesizes of 1ndash10 members All possible combi-nations of ensemble members were testedfor each ensemble size As CanCM4 has 10ensemble members there are 10 possibleone-member ensembles (ie individualruns) 45 possible two-member ensembles120 possible three-members ensembles andso on up to one possible 10-member en-semble For each ensemble size the skill ofeach possible combination is shown as a thinline and the mean skill of all possible com-binations is shown as a thick line In allcases ACC was calculated for 1982ndash2009using NOAArsquos 025deg OISSTv2 as truth

Owing to both model and observational limitations initial conditionsfor any forecast of a chaotic system can only be estimated within acertain accuracy Decades ago Lorenz (1982) quantified forecast errorgrowth as a function of initial condition errors by studying the rate atwhich solutions of a numerical weather prediction model divergedNow there is growing interest in quantifying forecast uncertainty formany areas of the earth system and on multiple timescales Methods foruncertainty quantification need to cope with sensitivities to initialconditions interactions of many spatial and temporal scales in the earthsystem and the fact that the sources of uncertainty are themselvesuncertain

Moving from physical forecasts to biogeochemical and ecologicalforecasts additional uncertainty is introduced An ecosystem model ingeneral will exhibit variability associated with a response to physicalforcing plus intrinsic variability associated with nonlinearities in theecology itself Therefore even if a physical model can skillfully predictthe environmental changes in the system and an ecological model canproperly represent the complexity of the ecosystem response errorsmay propagate nonlinearly through the food web in ways that limit ourability to predict the ecosystem response For example errors in phy-sically driven nutrient fluxes will lead to errors in phytoplanktongrowth which then cascade to errors in zooplankton growth and so onup the food chain to fish and higher-order predators If the error growthis linear predictions of target species may not be adversely affectedBut if the error growth is exponential even small errors in physicalforcing could lead to saturated error statistics in the mid-to-high trophiclevels This type of error propagation through trophic levels has notbeen quantified in typical ecosystem models currently being used andshould be explored and quantified both in its dependence on para-meters and its dependence on the complexity of the model food webdynamics

One common method of quantifying uncertainty and improvingforecast skill in physical systems is to use model ensembles For de-terministic forecasts which offer a single predicted outcome with nostatement of uncertainty (Wilks 2011) the issued forecast may be themean or median of the ensemble In this approach forecast skill for amultimodel ensemble mean tends to be as good as or better than theskill of the best model in the ensemble even when the ensemble in-cludes models with relatively poor skill (Hervieux et al 2017) Simi-larly for a single model the mean forecast computed from multipleensemble members is more skillful than the forecast from any singlemodel realization (Kumar and Hoerling 2000) In the example of SSTaforecasts in the CCS ~3ndash4 ensemble members are needed to reliablybeat a persistence forecast on 2ndash4 month lead times At longer leads(gt 6 months) a single model run can beat the skill of a persistenceforecast but larger ensembles improve skill up to a point with eachadditional ensemble member offering diminishing returns (Fig 11)Ecological model ensembles may consist of models that are funda-mentally different in their construction (ie using very different sta-tistical techniques) rather than of multiple realizations from modelsthat are structurally very similar (as is typical in the physical realm)Nonetheless there are indications that the same general finding (ie anensemble mean forecast outperforms the individual models that con-tribute to it) holds at least for species distribution modeling (Marmionet al 2009) However the use of an ensemble mean as a deterministicforecast omits potentially useful information the ensemble can provideabout forecast uncertainty (Demargne et al 2010) In contrast to de-terministic forecasts probabilistic forecasts issue an expected outcometaken from a set of possible outcomes by providing a summary measureof the ensemble or assigning a probability distribution to the forecastdata (Potts 2011) Probabilistic forecasts provide explicit statements ofuncertainty regarding how well the future state of a system is known(Wilks 2011) Knowing the probabilities of different outcomes is ben-eficial from a risk management and decision-making perspective anduse of probabilistic information is commonplace with respect toweather forecasts (eg deciding whether to carry an umbrella based on

forecast likelihood of rain) Examples of probabilistic forecasts inmarine resource management are less common though probabilisticSST forecasts are used to predict the likelihood of coral bleaching (Liuet al 2018) as well as El Nintildeo events (Kousky and Higgins 2007)which in turn have been used to inform fishery closures to avoid by-catch (Welch et al 2019b) In addition retaining the distribution ofpotential outcomes will be important for forecasting extreme eventswith large forecast ensembles (eg 50ndash100 members) likely necessaryto forecast conditions in the tails of distributions (Doi et al 2019)particularly if the ocean extreme is not driven by a predictable de-terministic forcing mechanism (Jacox et al 2019)

54 Computing resources

S2I prediction requires significant computing infrastructure interms of both processing power and data storage The climate modelsrun at these time scales include more components and complexity thanweather forecast models (Mariotti et al 2018) and large suites of re-forecasts are needed to evaluate model skill and quantify biases Forexample the typical contribution of an individual modeling center toPhase I of the NMME included 10 ensemble members each run for30 years of reforecasts initialized monthly and extending to lead timesof 12 months totaling 3600 years of model simulations Output fromthese runs must then be stored and while 2D physical fields such as SSTand SSH have most commonly been analyzed developing marine eco-system forecasts necessitates saving 3D ocean fields including tem-perature salinity currents and biogeochemistry (when available) attemporal resolutions not less than monthly and ideally much higherWhen high-resolution regional ocean models are needed to properlyrepresent nearshore dynamics critical to marine ecosystem forecasts anadded layer of computational demand is incurred The global forecastmust be run first with an additional suite of surface atmospheric statevariables and fluxes saved at daily or finer resolution and used to forcethe regional model Operationally this results in a longer forecastproduction time additional data storage for non-standard global vari-ables as well as regional model output and demands on data sharing asglobal and regional models are typically not run on the same archi-tecture or by the same people

Beyond serving as a restriction to operational forecasting effortsthese computational challenges present a difficulty to the researchersoutside of operational centers and federal laboratories Running re-levant experiments in the S2I prediction architecture and performingdata transfer and analysis on large datasets often requires computa-tional resources inaccessible to researchers not affiliated with largeinstitutions Funds from research grants are often diverted to purchaselocal storage and analysis systems or researchers compete for in-creasingly oversubscribed federal high-performance computing re-sources In this respect the marine forecasting community benefitsimmensely from efforts to make forecast outputs publicly availableboth by individual researchers and centers and through collaborativeprojects like the NMME LC-LRFMME and SubX

55 Observational limitations

Predictive skill is often reduced by the limitations of the predictionsystem which are in turn related to limitations of the observationsthat are key in the development initialization and verification ofmodels used for prediction (Capotondi et al 2019a) While satellitesensors provide regular global coverage for certain surface oceanproperties (eg SST and SSH) co-located surface and subsurfacemeasurements are important for producing consistent atmospheric andoceanic initial conditions verifying coupled feedbacks and in the caseof coupled ocean-atmosphere reanalyses constraining the coupled errorcovariance matrix However such measurements are sparse Argo floatshave considerably extended our ability to sample the subsurface oceanbut it is challenging for Argo and drifting buoys to provide sustained

measurements in strong currents or near the equator Likewise seasonalevents such as hurricanes or sea-ice formation and melt can prohibitsustained deployment of in situ observing platforms Biogeochemicaldata available for model initialization are even more sparse oftenleading to long spin-up times for biogeochemical variables (Wunschand Heimbach 2008) and a possible mismatch between biogeochem-istry and physics (Seferian et al 2016) The limited spatiotemporalcoverage of biogeochemical data combined with the variable timingand often ephemeral nature of biogeochemical anomalies are limitingfactors in model verification as short observational records may not berepresentative of the model climatology

The long-term stability of the observing system and the availabilityof long and homogeneous records are critical for error minimizationand verification of ocean and atmosphere reanalyses (Balmaseda et al2015 Xue et al 2017) Similarly long and homogeneous records arecritical for the development and verification of statistical predictionsystems whose performance relies on the robustness of statistical re-lationships between predictands and predictors On regional scalesforecasting activities rely heavily on the availability of observationsthat target the spatial and temporal scales of interest and come from adiverse set of observing assets including moorings and floats autono-mous underwater vehicles research cruises and satellite sensors (egSiedlecki et al 2016 Ortiz et al 2016) Preservation and extension ofphysical biogeochemical and ecological data records (ie throughsustained or expanded support for the platforms listed above andothers) are fundamental requirements for developing initializing andvalidating marine ecosystem forecasts Additionally efforts to expandthe suite of observations available in marine forecasting would benefitfrom incorporation of cross-disciplinary technology (eg animal-bornetelemetry tags) and opportunistic research platforms (eg fishing andtransportation vessels)

56 Forecast user engagement

Improving predictability and developing skillful forecasts of marineecosystems can be highly useful to a variety of end users includingmanagers and industry representatives However the utility of forecastsdepends on the forecast skill in time and space the needs of the end-user (eg spatial or temporal resolution and lead time necessary formanagement decisions) and effective dissemination of information(Hobday et al 2016) While there are many examples of predictivemodels developed with management or industry applications in mindfew have developed readily-available forecasts or nowcasts or havebeen integrated into decision making processes for businesses or man-agers (though see examples in Howell et al 2008 Hazen et al 2017Hobday et al 2016 Tommasi et al 2017) Studies that have suc-cessfully applied forecasts to inform decision-making processes high-light the importance of communication between scientists and man-agers or industry stakeholders and emphasize that forecast delivery iscritical to the utility and implementation of forecasts (Hobday et al2016 Tommasi et al 2017 Dunstan et al 2018) User engagementand feedback is necessary to refine the format and visualization of theforecast product in order to meet user needs and allow for widespreaduse and integration of the product into decision-making processes Thetransfer of forecasts to end users often takes the form of automated webproducts providing daily weekly or monthly forecasts posted allowingwide dissemination to managers and stakeholders

6 Conclusions and summary recommendations

The ability to anticipate changing environmental and ecologicalconditions provides opportunities to mitigate adapt to and plan fortheir impacts a great benefit to end-users including marine resourcemanagers industry representatives and the general public (Clark et al2001) S2I forecasts enable proactive decision making on timescalesthat are well aligned with marine ecosystem management (Hobday

et al 2018) and constitute a key component of a ldquoclimate-readyrdquofisheries management strategy that leverages information on multipletimescales (Pinsky and Mantua 2014) The need for skillful forecastshas motivated a growing body of research on marine ecosystem pre-dictability and numerous predictive models have been developed withmanagement or industry applications in mind To facilitate the devel-opment and uptake of marine ecosystem forecasts we provide summaryrecommendations for future work in four related areas model devel-opment skill and uncertainty assessment forecasting infrastructureand communication with end-users

Model development

bull Improve representation of key processes via model resolution andor complexity as appropriate and quantify predictability associatedwith these improvements

bull Quantify the sensitivity of regional ecosystem responses to physicaland biogeochemical formulations at model boundaries includingmethods for matching biogeochemical fields between global andregional domains and investigating the degree to which signals thatare poorly resolved in global models (eg coastal trapped wavespoleward undercurrents buoyancy-driven coastal currents) can betransmitted through model boundaries into the regional domainwhere they are better resolved

bull Explore empirical models (eg LIM) for both physical and ecolo-gical forecasts

bull Better integrate numerical and statistical methods (eg throughhybrid statisticaldynamical forecasts and statistical prepost-pro-cessing of dynamical forecasts) to maximize predictive skill

bull Develop statistical models for higher trophic levels based on a priorimechanistic understanding or hypotheses

Skill and uncertainty assessment

bull Quantify predictability and forecast skill for a broader range ofsurface and subsurface variables as well as suites of variables thatmay be co-stressors for living marine resources or coastal commu-nities

bull Quantify predictability and forecast skill for ecological forecaststhemselves building on physical forecast skill assessment and his-torical assessments of ecological models

bull Quantify the dependence of predictability and forecast skill onspatiotemporal scale and the degree to which spatiotemporal scalesof physical and ecological predictability are consistent with eachother

bull Establish protocols (eg temporal extent skill metrics) for refor-ecast evaluations to ensure they are adequate for building con-fidence in forecasts

bull Quantify forecast uncertainty including error propagation fromphysics to biogeochemistry and higher trophic levels

bull Assess skill not only for properties of interest but also for the me-chanisms that drive variability in those properties

Forecasting Infrastructure

bull Make forecast fields readily available at spatiotemporal resolutionsneeded for analysis and application to marine resource manage-ment and (in the case of global forecasts) to force downscaled re-gional forecasts

bull Preserve and extend physical biogeochemical and ecological datarecords needed to develop initialize and validate marine ecosystemforecasts

Communication

bull Establish and maintain communication between scientists and end-users early in the forecast development process to ensure that

forecasts align with the needs of the end-user and in the im-plementation and delivery phase to refine the forecast product andfacilitate uptake in decision-making processes

bull Develop methods to bring uncertainty in the mechanisms of pre-dictability into the communication of forecast uncertainty (eg incommunication of conditional forecast skill)

In this paper we have provided a review of existing forecastingapproaches described environmental and ecological dynamics thathave been (or could be) exploited in predictability studies and forecastdevelopment and recommended priorities for future work While ourfocus is on coastal ecosystems surrounding North America and the USin particular the content (including forecasting approaches physicaland biological mechanisms and priority developments) apply globallyOur aim is to support and motivate continued advances in the field ofmarine ecosystem forecasting To that end we suggest that if marineforecasts are to effectively support ocean decision making particularattention should be paid to the mechanisms underlying their skill Thismechanistic understanding of marine ecosystem predictability willallow the research community to target specific areas for forecast skillimprovement quantify and communicate forecast uncertainties ensurepredictive relationships hold up over time and provide a basis for end-user confidence in marine forecast products

Declaration of Competing Interest

The authors declare that they have no known competing financialinterests or personal relationships that could have appeared to influ-ence the work reported in this paper

Acknowledgments

This study was supported by the NOAA Climate Program OfficersquosModeling Analysis Predictions and Projections (MAPP) programthrough grants NA17OAR4310108 NA17OAR4310112NA17OAR4310111 NA17OAR4310110 NA17OAR4310109NA17OAR4310104 NA17OAR4310106 and NA17OAR4310113 Thispaper is a product of the NOAAMAPP Marine Prediction Task Force

References

Abrahms B Welch H Brodie S Jacox MG Becker EA Bograd SJ Irvine LMPalacios DM Mate BR Hazen EL 2019 Dynamic ensemble models to predictdistributions and anthropogenic risk exposure for highly mobile species DiversDistrib httpsdoiorg101111ddi12940

Agnew DJ Beddington JR Hill SL 2002 The potential use of environmental in-formation to manage squid stocks Can J Fish Aquat Sci 59 1851ndash1857

Alexander MA Deser C 1995 A mechanism for the recurrence of midlatitude SSTanomalies during winter J Phys Oceanogr 25 122ndash137

Alexander MA Deser C Timlin MS 1999 The re-emergence of SST anomalies in theNorth Pacific Ocean J Climate 12 2419ndash2431

Alexander MA Timlin MS Scott JD 2001 Winter-to-Winter recurrence of seasurface temperature salinity and mixed layer depth anomalies Prog Oceanogr 4941ndash61

Alexander MA Bladeacute I Newman M Lanzante JR Lau NC Scott JD 2002 Theatmospheric bridge the influence of ENSO teleconnections on airndashsea interactionover the global oceans J Clim 15 (16) 2205ndash2231

Alexander MA Scott JD 2002 The influence of ENSO on air-sea interaction in theAtlantic Geophys Res Lett 29 (14) httpsdoiorg1010292001GL014347

Alexander MA Matrosova L Penland C Scott JD Chang P 2008 Forecastingpacific SSTs linear inverse model prediction of the PDO J Climate 21 385ndash402

Alexander MA Scott JD 2008 The role of Ekman ocean heat transport in theNorthern Hemisphere Response to ENSO J Climate 21 5688ndash5707

Alin SR Feely RA Dickson AG Hernaacutendez-Ayoacuten JM Juranek LW OhmanMD Goericke R 2012 Robust empirical relationships for estimating the carbonatesystem in the southern California Current System and application to CalCOFI hy-drographic cruise data (2005ndash2011) J Geophys Res Oceans 117 (C5)

Anderson CR Kudela RM Kahru M Chao Y Rosenfeld LK Bahr FL et al 2016Initial skill assessment of the California harmful algae risk mapping (C-HARM)system Harmful Algae 59 1ndash18

Anderson DM Keafer BA Kleindinst JL McGillicuddy DJ Martin JL Norton KPilskaln CH Smith JL Sherwood CR Butman B 2014 Alexandrium fundyensecysts in the Gulf of Maine long-term time series of abundance and distribution and

linkages to past and future blooms Deep-Sea Res II 103 6ndash26Andres M Kwon Y-O Yang J 2011 Observations of the Kuroshiorsquos barotropic and

baroclinic responses to basin-wide wind forcing J Geophys Res 116 C04011httpsdoiorg1010292010JC006863

Annamalai H Hafner J Kumar A Wang H 2014 A framework for dynamical sea-sonal prediction of precipitation over the Pacific Islands J Climate 27 (9)3272ndash3297 httpsdoiorg101175JCLI-D-13-003791

Auad G Miller A Di Lorenzo E 2006 Long-term forecast of oceanic conditions offCalifornia and their biological implications J Geophys Res Oceans 111 C09008httpsdoiorg1010292005JC003219

Balmaseda MA et al 2015 The ocean reanalysis intercomparison project (ORA-IP) JOper Oceanogr 8 httpsdoiorg1010801755876X20151022329

Banzon V et al 2016 A long-term record of blended satellite and in situ sea-surfacetemperature for climate monitoring modeling and environmental studies Earth SystSci Data 8 (1) 165ndash176

Barth JA 2003 Anomalous southward advection during 2002 in the northernCalifornia Current evidence from Lagrangian surface drifters Geophys Res Lett doi1010292003GL017511

Blanchard-Wrigglesworth E Armour KC Bitz CM DeWeaver E 2011 Persistenceand inherent predictability of Arctic sea ice in a GCM ensemble and observations JClim 24 (1) 231ndash250

Bograd SJ Lynn RJ 2003 Anomalous subarctic influence in the southern CaliforniaCurrent during 2002 Geophys Res Lett 2003 httpsdoiorg1010292003GL017446

Bograd SJ Castro CG Di Lorenzo E Palacios DM Bailey H Gilly W ChavezFP 2008 Oxygen declines and the shoaling of the hypoxic boundary in theCalifornia Current Geophys Res Lett 35 (12)

Brander K 2003 What kinds of fish stock predictions do we need and what kinds ofinformation will help us make better predictions Scientia Marina 67 21ndash33

Brodie S Hobday AJ Smith JA Everett JD Taylor MD Gray CA Suthers IM2015 Modelling the oceanic habitats of two pelagic species using recreational fish-eries data Fish Oceanogr 24 (5) 463ndash477

Brodie S Hobday AJ Smith JA Spillman CM Hartog JR Everett JD et al2017 Seasonal forecasting of dolphinfish distribution in eastern Australia to aid re-creational fishers and managers Deep Sea Res Part II 140 222ndash229

Brodie S Jacox MG Bograd SJ Welch H Dewar H Scales KL et al 2018Integrating dynamic subsurface habitat metrics into species distribution modelsFront Mar Sci 5 219

Brown ZW Arrigo KR 2013 Sea ice impacts on spring bloom dynamics and netprimary production in the Eastern Bering Sea J Geophys Res Oceans 118 43ndash62httpsdoiorg1010292012JC008034

Buckley LB Kingsolver JG 2012 Functional and phylogenetic approaches to fore-casting species responses to climate change Annu Rev Ecol Evol Syst 43205ndash226

Bushuk M Giannakis D Majda AJ 2014 Reemergence mechanisms for North Pacificsea ice revealed through nonlinear Laplacian spectral analysis J Clim 276265ndash6287

Bushuk M Msadek R Winton M Vecchi GA Gudgel R Rosati A Yang X 2017Skillful regional prediction of Arctic sea ice on seasonal timescales Geo-phys ResLett 44 4953ndash4964 httpsdoiorg1010022017GL073155

Byju P Dommenget D Alexander MA 2018 Widespread reemergence of sea surfacetemperature anomalies in the global oceans including tropical regions forced byreemerging winds Geophys Res Lett 45 7683ndash7691 httpsdoiorg1010292018GL079137

Calafat FM Wahl T Lindsten F Williams J Frajka-Williams E 2018 Coherentmodulation of the sea-level annual cycle in the United States by Atlantic Rossbywaves Nat Commun 9 (1) 2571

Capotondi A Alexander MA 2001 Rossby waves in the tropical North Pacific andtheir role in decadal thermocline variability J Phys Oceanogr 31 3496ndash3515

Capotondi A Sardeshmukh PD 2015 Optimal precursors of different types of ENSOevents Geophys Res Lett 42 9952ndash9960

Capotondi A Sardeshmukh PD 2017 Is El Nino really changing Geophys Res Lett44 httpsdoiorg1010022017GL074515

Capotondi A et al 2019a Observational needs supporting marine ecosystem modelingand forecasting Insights from US Coastal Applications Front Mar Sci httpsdoiorg103389fmars201900623

Capotondi A Sardeshmukh PD Di Lorenzo E Subramanian A Miller AJ 2019bPredictability of US west coast ocean temperatures is not solely due to ENSO SciRep 9 10993 httpsdoiorg101038s41598-019-47400-4

Chapman DC Beardsley RC 1989 On the origin of shelf water in the middle atlanticbight J Phys Oceanogr 19 384ndash391 httpsdoiorg1011751520-0485(1989)019lt0384OTOOSWgt20CO2

Chelton DB Davis RE 1982 Monthly mean sea-level variability along the West Coastof North America J Phys Oceanogr 12 757ndash784 httpsdoiorg1011751520-0485(1982) 012lt0757MMSLVAgt20CO2

Chelton Schlax 1996 Global observations of oceanic Rossby waves ScienceChen K He R 2015 Mean circulation in the coastal ocean off northeastern North

America from a regional-scale ocean model Ocean Sci 11 (4) 503ndash517 httpsdoiorg105194os-11-1-2015

Chen K Kwon Y-O 2018 Does pacific variability influence the northwest Atlanticshelf temperature J Geophys Res Oceans 123 4110ndash4131

Chen K Kwon Y-O Gawarkiewicz G 2016 Interannual variability of winter-springtemperature in the Middle Atlantic Bight relative contributions of atmospheric andoceanic processes J Geophys Res Oceans 121 4209ndash4227

Chen K He R Powell BS Gawarkiewicz GG Moore AM Arango HG 2014 Dataassimilative modeling investigation of Gulf Stream Warm Core Ring interaction with

continental shelf and slope circulation J Geophys Res Oceans 119 5968ndash5991Cheng W Curchitser E Ladd C Stabeno P Wang M 2014 Influences of sea ice on

the Eastern Bering Sea NCAR CESM simulations and comparison with observationsDeep Sea Res Part II Top Stud Oceanogr 109 27ndash38

Chiang JC Vimont DJ 2004 Analogous Pacific and Atlantic meridional modes oftropical atmospherendashocean variability J Clim 17 (21) 4143ndash4158

Chikamoto MO Timmermann A Chikamoto Y Tokinaga H Harada N 2015Mechanisms and predictability of multiyear ecosystem variability in the NorthPacific Global Biogeochem Cycles 29 2001ndash2019 httpsdoiorg1010022015GB005096

Clark JS Carpenter SR Barber M Collins S Dobson A Foley JA Lodge DMPascual M Pielke R Pizer W Pringle C Reid WV Rose KA Sala OSchlesinger WH Wall DH Wear D 2001 Ecological forecasts an emergingimperative Science 293 657ndash660

Clarke AJ Van Gorder S 1994 On ENSO coastal currents and sea levels J PhysOceanogr 24 (3) 661ndash680

Clarke AJ Dottori M 2008 Planetary wave propagation off california and its effect onzooplankton J Phys Oceanogr 38 702ndash714 httpsdoiorg1011752007JPO36911

Connolly TP Hickey BM Shulman I Thomson RE 2014 Coastal trapped wavesalongshore pressure gradients and the California Undercurrent J Phys Oceanogr44 (1) 319ndash342

Cuddington K Fortin MJ Gerber LR Hastings A Liebhold A Oconnor M RayC 2013 Process-based models are required to manage ecological systems in achanging world Ecosphere 4 (2) 1ndash12

Dai Y Feldstein SB Tan B Lee S 2017 Formation mechanisms of the Pacific-NorthAmerican teleconnection with and without its canonical tropical convection patternJ Climate 30 3139ndash3155 httpsdoiorg101175JCLI-D-16-04111

Davis RE 1976 Predictability of sea surface temperature and sea level pressureanomalies over the North Pacific Ocean J Phys Oceanogr 6 (3) 249ndash266

Davis KA Banas NS Giddings SN Siedlecki SA MacCready P Lessard EJet al 2014 Estuary-enhanced upwelling of marine nutrients fuels coastal pro-ductivity in the US Pacific Northwest J Geophys Res Oceans 119 (12) 8778ndash8799

Davis XJ Joyce TM Kwon Y-O 2017 Prediction of silver hake distribution on theNortheast US shelf based on the Gulf Stream path index Cont Shelf Res 13851ndash64

Demargne J Brown J Liu YQ Seo DJ Wu LM Toth Z Zhu YJ 2010Diagnostic verification of hydrometeorological and hydrologic ensembles Atmos SciLett 11 114ndash122

Deser C Simpson IR McKinnon KA Phillips AS 2017 The Northern Hemisphereextratropical atmospheric circulation response to ENSO How well do we know it andhow do we evaluate models accordingly J Clim 30 (13) 5059ndash5082

Di Lorenzo E Ohman MD 2013 A double-integration hypothesis to explain oceanecosystem response to climate forcing Proc Natl Acad Sci USA 110 (7)2496ndash2499 httpsdoiorg101073pnas1218022110

Di Lorenzo E Liguori G Schneider N Furtado JC Anderson BT Alexander MA2015 ENSO and meridional modes a null hypothesis for Pacific climate variabilityGeophys Res Lett 42 (21) 9440ndash9448

Di Lorenzo E Mantua N 2016 Multi-year persistence of the 201415 North Pacificmarine heatwave Nat Clim Change 6 (11) 1042

Dietze MC Fox A Beck-Johnson LM Betancourt JL Hooten MB Jarnevich CSet al 2018 Iterative near-term ecological forecasting needs opportunities andchallenges Proc Natl Acad Sci 115 (7) 1424ndash1432

Doblas-Reyes FJ Garciacutea-Serrano J Lienert F Biescas AP Rodrigues LR 2013Seasonal climate predictability and forecasting status and prospects WileyInterdiscip Rev Clim Change 4 (4) 245ndash268

Doi T Behera S Yamagata T 2019 Merits of a 108-member ensemble system inENSO and IOD predictions J Climate httpsdoiorg101175JCLI-D-18-01931

Dormann CF Schymanski SJ Cabral J Chuine I Graham C Hartig F et al2012 Correlation and process in species distribution models bridging a dichotomyJ Biogeogr 39 (12) 2119ndash2131

Duffy-Anderson JT Stabeno PJ Siddon EC Andrews AG Cooper DC EisnerLB et al 2017 Return of warm conditions in the southeastern Bering SeaPhytoplankton - Fish PLoS ONE 12 (6) e0178955 httpsdoiorg101371journalpone0178955

Dunstan PK Moore BR Bell JD Holbrook NJ Oliver EC Risbey J et al 2018How can climate predictions improve sustainability of coastal fisheries in PacificSmall-Island Developing States Mar Pol 88 295ndash302

Durski SM Barth JA McWilliams JC Frenzel H Deutsch C 2017 The influenceof variable slope-water characteristics on dissolved oxygen levels in the northernCalifornia Current System J Geophys Res Oceans 122 (9) 7674ndash7697

Enfield DB Allen JS 1980 On the structure and dynamics of monthly mean sea levelanomalies along the Pacific coast of North and South America J Phys Oceanogr 10(4) 557ndash578

Eveson JP Hobday AJ Hartog JR Spillman CM Rough KM 2015 Seasonalforecasting of tuna habitat in the Great Australian Bight Fish Res 170 39ndash49

Ezer T 2016 Can the Gulf Stream induce coherent short-term fluctuations in sea levelalong the US East Coast A modeling study Ocean Dyn 66 207 httpsdoiorg101007s10236-016-0928-0

Faggiani Dias D Subramanian A Zanna L Miller AJ 2018 Remote and local in-fluences in forecasting Pacific SST a linear inverse model and a multimodel ensemblestudy Clim Dyn doi 101007s00382-018-4323-z

Fennel K Wilkin J Levin J Moisan J OReilly J Haidvogel D 2006 Nitrogencycling in the Middle Atlantic Bight Results from a three-dimensional model andimplications for the North Atlantic nitrogen budget Global Biogeochem Cycles20 (3)

Firing YL Merrifield MA Schroeder TA Qiu B 2004 Interdecadal sea levelfluctuations at Hawaii J Phys Oceanogr 34 2514ndash2524

Forsyth JST Andres M Gawarkiewicz GG 2015 Recent accelerated warming of thecontinental shelf off New Jersey observations from the CMV Oleander expendablebathythermograph line J Geophys Res Oceans 120 2370ndash2384

Fourcade Y Besnard AG Secondi J 2018 Paintings predict the distribution of spe-cies or the challenge of selecting environmental predictors and evaluation statisticsGlob Ecol Biogeogr 27 (2) 245ndash256

Fowler HJ Blenkinsop S Tebaldi C 2007 Linking climate change modelling toimpacts studies recent advances in downscaling techniques for hydrological mod-elling Int J Climatol 27 1547ndash1578 httpsdoiorg101002joc1556

Frankignoul C de Coetlogon G Joyce TM Dong SF 2001 Gulf stream variabilityand ocean-atmosphere interactions J Phys Oceanogr 31 3516ndash3529

Freeland HJ Gatien G Huyer A Smith RL 2003 Cold halocline in the northernCalifornia Current An invasion of subarctic water Geophys Res Lett 30 (3) 1141httpsdoiorg1010292002GL016663

Frischknecht M Muumlnnich M Gruber N 2015 Remote versus local influence of ENSOon the California Current System J Geophys Res Oceans 120 (2) 1353ndash1374

Fu L-L Qiu B 2002 Low-frequency variability of the North Pacific Ocean The role ofboundary- and wind-driven baroclinic Rossby waves J Geophys Res 107 3220httpsdoiorg1010292001JC001131

Gaitan CF Hsieh WW Cannon AJ 2014 Comparison of statistically downscaledprecipitation in terms of future climate indices and daily variability for southernOntario and Quebec Canada Clim Dyn 43 (12) 3201ndash3217

Gill AE 1982 Atmosphere‐Ocean Dynamics Academic San Diego Calif 662 pGoddard L Mason SJ Zebiak SE Ropelewski CF Basher R Cane MA 2001

Current approaches to seasonal-to-interannual climate predictions Int J Climatol21 1111ndash1152

Goubanova K Echevin V Dewitte B Codron F Takahashi K Terray P Vrac M2011 Statistical downscaling of sea-surface wind over the Peru-Chile upwelling re-gion diagnosing the impact of climate change from the IPSL-CM4 model Clim Dyn36 (7ndash8) 1365ndash1378

Graham RJ Yun W-T Kim J Kumar A Jones D Bettio L Gagnon N Kolli RKSmith D 2011 Long-range forecasting and the Global Framework for ClimateServices Climate Res 47 (1ndash2) 47ndash55

Hamill TM Bates GT Whitaker JS Murray DR Fiorino M Galarneau TJ ZhuY Lapenta W 2013 NOAAs second-generation global medium-range ensemblereforecast dataset Bull Amer Meteor Soc 94 1553ndash1565 httpsdoiorg101175BAMS-D-12-000141

Hanawa K Sugimoto S 2004 lsquoReemergencersquo areas of winter sea surface temperatureanomalies in the worldrsquos oceans Geophys Res Lett 31 L10303 httpsdoiorg1010292004GL019904

Hare JA Cowen RK 1996 Transport mechanisms of larval and pelagic juvenilebluefish (Pomatomus saltatrix) from South Atlantic Bight spawning grounds toMiddle Atlantic Bight nursery habitats Limnol Oceanogr 41 1264ndash1280

Hazen EL Palacios DM Forney KA Howell EA Becker E Hoover AL Irvine LDeAngelis M Bograd SJ Mate BR Bailey H 2017 WhaleWatch a dynamicmanagement tool for predicting blue whale density in the California Current J ApplEcol 54 (5) 1415ndash1428

Hazen EL Scales KL Maxwell SM Briscoe DK Welch H Bograd SJ Bailey HBenson SR Eguchi T Dewar H Kohin S 2018 A dynamic ocean managementtool to reduce bycatch and support sustainable fisheries Sci Adv 4 (5) peaar3001

Henderson M Fiechter J Huff DD Wells BK 2018 Spatial variability in ocean-mediated growth potential is linked to Chinook salmon survival Fish Oceanogrhttpsdoiorg101111fog12415

Hermann A Aydin K 2014 Preliminary 9 month ecosystem forecast for the easternBering Sea In Zador S 2014 Ecosystem Considerations Stock Assessment andFishery Evaluation Report North Pacific Fishery Management Council 605 W 4thAve Suite 306 Anchorage AK 99501

Hermann AJ Curchitser EN Haidvogel DB Dobbins EL 2009 A comparison ofremote vs local influence of El Nino on the coastal circulation of the northeastPacific Deep Sea Res Part II 56 (24) 2427ndash2443

Hermann AJ Gibson GA Bond NA Curchitser EN Hedstrom K Cheng WWang M Stabeno PJ Eisner L Cieciel KD 2013 A multivariate analysis ofobserved and modeled biophysical variability on the Bering Sea shelf multidecadalhindcasts (1970ndash2009) and forecasts (2010ndash2040) Deep-Sea Res II 94 121ndash139httpsdoiorg101016jdsr2201304007

Hermann AJ Gibson GA Bond NA Curchitser EN Hedstrom K Cheng W et al2016 Projected future biophysical states of the Bering Sea Deep Sea Res Part II 13430ndash47

Hermann AJ Gibson GA Cheng W Ortiz I Aydin K Wang M Hollowed ABHolsman KK 2019 Projected biophysical conditions of the Bering Sea to 2100under multiple emission Scenarios ICES J Mar Sci httpsdoiorg101093icesjmsfsz043 fsz043

Hervieux G Alexander MA Stock CA Jacox MG Pegion K Becker E CastruccioF Tommasi D 2017 More reliable coastal SST forecasts from the North Americanmultimodel ensemble Clim Dyn 1ndash16 httpsdoiorg101007s00382-017-3652-7

Hickey BM 1979 The California current systemmdashhypotheses and facts ProgOceanogr 8 (4) 191ndash279

Hickey B Geier S Kachel N Ramp S Kosro PM Connolly T 2016 Alongcoaststructure and interannual variability of seasonal midshelf water properties and ve-locity in the Northern California Current System J Geophys Res Oceans 121 (10)7408ndash7430

Hill KT Crone PR Zwolinski JP 2017 Assessment of the Pacific sardine resource in2017 for US management in 2017-18 Pacific Fishery Management Council April

2017 Briefing Book Agenda Item G5a Portland Oregon 146 pHobday AJ Smith ADM Stobutzki IC Bulman C Daley R Dambacher JM

Deng RA Dowdney J Fuller M Furlani D Griffiths SP 2011 Ecological riskassessment for the effects of fishing Fish Res 108 (2ndash3) 372ndash384

Hobday AJ Spillman CM Paige Eveson J Hartog JR 2016 Seasonal forecastingfor decision support in marine fisheries and aquaculture Fish Oceanogr 25 45ndash56

Hobday AJ Spillman CM Eveson P Hartog J Zhang X Brodie S 2018 A fra-mework for combining seasonal forecasts and climate projections to aid risk man-agement for fisheries and aquaculture Front Mar Sci 5 137

Holbrook NJ Scannell HA Gupta AS Benthuysen JA Feng M Oliver EC et al2019 A global assessment of marine heatwaves and their drivers Nat Commun 10(1) 2624

Holsman KK Ianelli J Aydin KY Punt AE Moffit EA 2016 A comparison offisheries biological reference points estimated from temperature-specific multi-spe-cies and single-species climate-enhanced stock assessment models Deep Sea Res II134 360ndash378

Holsman KK Ianelli J Aydin KY 2017 Multi-species stock assessment for walleyepollock Pacific cod and arrowtooth flounder in the eastern Bering Sea StockAssessment and Fishery Evaluation Report for the Groundfish Resources of the BeringSeaAleutian Islands Region Anchorage Alaska North Pacific Fisheries ManagementCouncil

Hong BG Sturges W Clarke AJ 2000 Sea level on the US east coast decadalvariability caused by open ocean wind-curl forcing J Phys Oceanogr 302088ndash2098

Howell EA Kobayashi DR Parker DM Balazs GH Polovina JJ 2008TurtleWatch a tool to aid in the bycatch reduction of loggerhead turtles Carettacaretta in the Hawaii-based pelagic longline fishery Endangered Species Res 5 (2ndash3)267ndash278

Hu Z-Z Kumar A Huang B Wang W Zhu J 2013 Prediction skill of monthly SSTin the North Atlantic Ocean in NCEP Climate Forecast System Version 2 Clim Dyn40 2745ndash2756

Huang J Mariotti A Kinter J Kumar A 2014 Preface to CFSv2 topical collectionClim Dyn 43 (7ndash8) httpsdoiorg101007s00382-014-2316-0

Huth R 2002 Statistical downscaling of daily temperature in central Europe J Clim15 1731ndash1742

ISC albacore working group 2017 Stock assessment of albacore tuna in the North PacificOcean in 2017 httpiscfragojppdfISC17ISC17_Annex12-Stock_Assessment_of_Albacore_Tuna_in_the_North_Pacific_Ocean_in_2017pdf

ISC shark working group 2018 Stock assessment of shortfin mako shark in the NorthPacific Ocean through 2016 httpiscfragojppdfISC18ISC_18_ANNEX_15_Shortfin_Mako_Shark_Stock_Assessment_FINALpdf

Ito T Deutsch C 2010 A conceptual model for the temporal spectrum of oceanicoxygen variability Geophys Res Lett 37 L03601 httpsdoiorg1010292009GL041595

Jacobs GA Hurlburt HE Kindle JC Metzger EJ Mitchell JL Teague WJWallcraft AG 1994 Decade-scale trans-Pacific propagation and warming effects ofan El Nino anomaly Nature 370 360ndash363

Jacox MG Fiechter J Moore AM Edwards CA 2015a ENSO and the CaliforniaCurrent coastal upwelling response J Geophys Res Oceans 120 (3) 1691ndash1702httpsdoiorg1010022014JC010650

Jacox MG Bograd SJ Hazen EL Fiechter J 2015b Sensitivity of the CaliforniaCurrent nutrient supply to wind heat and remote ocean forcing Geophys Res Lett42 (14) 5950ndash5957 httpsdoiorg1010022015GL065147

Jacox MG Alexander MA Stock CA Hervieux G 2017 On the skill of seasonal seasurface temperature forecasts in the California Current System and its connection toENSO variability Clim Dyn 1ndash15 httpsdoiorg101007s00382-017-3608-y

Jacox MG Edwards CA Hazen EL Bograd SJ 2018 Coastal upwelling revisitedEkman Bakun and improved upwelling indices for the US west coast J GeophysRes Oceans 123 (10) 7332ndash7350 httpsdoiorg1010292018JC014187

Jacox M Tommasi D Alexander M Hervieux G Stock C 2019 Predicting theevolution of the 2014ndash16 California Current System marine heatwave from an en-semble of coupled global climate forecasts Front Mar Sci 6 497 httpsdoiorg103389fmars201900497

Joh Y Di Lorenzo E 2017 Increasing coupling between NPGO and PDO leads toprolonged marine heatwaves in the Northeast Pacific Geophys Res Lett 44 (22)11663ndash11671 httpsdoiorg1010022017gl075930

Johnson HL Marshall DP 2004 Global teleconnections of meridional overturningcirculation anomalies J Phys Oceanogr 34 1702ndash1722

Joyce TM Deser C Spall MA 2000 The relation between decadal variability ofsubtropical mode water and the North Atlantic Oscillation J Climate 13 2550ndash2569

Joyce TM Bishop JKB Brown OB 1992 Observations of offshore shelf-watertransport induced by a warm-core ring Deep-Sea Res 39 S97ndashS113

Kaplan IC Williams GD Bond NA Hermann AJ Siedlecki SA 2016 Cloudywith a chance of sardines forecasting sardine distributions using regional climatemodels Fish Oceanogr 25 (1) 15ndash27

Kawase M 1987 Establishment of deep circulation driven by deep water production JPhys Oceanogr 17 2294ndash2316

Kearney K Hermann A Cheng W Ortiz I Aydin K 2019 A coupled pelagic-benthic-sympagic biogeochemical model for the Bering Sea documentation and va-lidation of the BESTNPZ model (v20190823) within a high-resolution regionalocean model in review Geosci Model Dev Discuss httpsdoiorg105194gmd-2019-239

Kilpatrick T Schneider N Di Lorenzo E 2011 Generation of low-frequency spicinessvariability in the thermocline J Phys Oceanogr 41 (2) 365ndash377 httpsdoiorg1011752010jpo44431

Kirtman BP et al 2014 THE NORTH AMERICAN MULTIMODEL ENSEMBLE Phase-1

seasonal-to-interannual prediction phase-2 toward developing intraseasonalPrediction B Am Meteorol Soc 95 585ndash601 httpsdoiorg101175Bams-D-12-000501

Kirtman BP Pegion K DelSole T Tippett M Robertson AW Bell M Green BW2017 The subseasonal experiment (SubX) IRI Data Library

Kousky VE Higgins RW 2007 An alert classification system for monitoring and as-sessing the ENSO cycle Weather Forecast 22 (2) 353ndash371

Kumar A Hoerling MP 2000 Analysis of a conceptual model of seasonal climatevariability and implications for seasonal predictions Bull Amer Meteor Soc 81255ndash264

Kurapov AL Erofeeva SY Myers E 2017 Coastal sea level variability in the US WestCoast Ocean forecast system (WCOFS) Ocean Dyn 67 (1) 23ndash36

Kushnir Y et al 2019 Towards operational predictions of the near-term climate NatClim Change 9 94ndash101

Kwon Y-O Alexander MA Bond NA Frankignoul C Nakamura H Qiu BThompson L 2010 Role of Gulf Stream and Kuroshio-Oyashio systems in large-scaleatmosphere-ocean interaction a review J Climate 23 3249ndash3281

Lee CC Sheridan SC 2015 Synoptic climatology an overview In Reference Modulein Earth Systems and Environmental Sciences Elsevier doi 101016B978-0-12-409548-909421-5

Lee CC Sheridan SC Barnes BB Hu C Pirhalla DE Ransibrahmanakul V SheinK 2017 The development of a non-linear autoregressive model with exogenousinput (NARX) to model climate-water clarity relationships reconstructing a historicalwater clarity index for the coastal waters of the southeastern USA Theor ApplClimatol 130 (1ndash2) 557ndash569

Levin PS Fogarty MJ Murawski SA Fluharty D 2009 Integrated ecosystem as-sessments developing the scientific basis for ecosystem-based management of theocean PLoS Biol 7 (1) e1000014

Levin PS Schwing FB 2011 Technical background for an integrated ecosystem as-sessment of the California Current Groundfish salmon green sturgeon and eco-system health US National Oceanic and Atmospheric Administration Tech MemoNMFS-NWFSC-109

Leslie HM McLeod KL 2007 Confronting the challenges of implementing marineecosystem-based management Front Ecol Environ 5 (10) 540ndash548

Li Y Lau N-C 2012 Impact of ENSO on the atmospheric variability over the NorthAtlantic in late winter - role of transient eddies J Climate 25 320ndash342

Li Y Ji R Fratantoni PS Chen C Hare JA Davis CS Beardsley RC 2014Wind-induced interannual variability of sea level slope along-shelf flow and surfacesalinity on the Northwest Atlantic shelf J Geophys Res Oceans 119 2462ndash2479

Liguori G Di Lorenzo E 2018 Meridional modes and increasing pacific decadalvariability under anthropogenic forcing Geophys Res Lett 45 (2) 983ndash991 httpsdoiorg1010022017gl076548

Liu G Eakin CM Chen M Kumar A De La Cour JL Heron SF et al 2018Predicting heat stress to inform reef management NOAA Coral Reef Watchs 4-monthcoral bleaching outlook Front Mar Sci 5 57

Liu Z Di Lorenzo E 2018 Mechanisms and predictability of Pacific decadal variabilityCurrent Climate Change Reports 4 (2) 128ndash144

Long X Widlansky MJ Schloesser F Thompson PR Annamalai H MerrifieldMA Yoon H 2020 Higher sea levels at Hawaii caused by strong El Nintildeo and weaktrade winds J Clim httpsdoiorg101175JCLI-D-19-02211

Lorenz EN 1982 Atmospheric predictability experiments with a large numericalmodel Tellus 34 505ndash513

Lucey Nye 2010 Shifting species assemblages in the northeast US continental shelf largemarine ecosystem Mar Ecol Prog Ser

MacFadyen A Hickey BM Cochlan WP 2008 Influences of the Juan de Fuca Eddyon circulation nutrients and phytoplankton production in the northern CaliforniaCurrent System J Geophys Res Oceans 113 (C8)

Machu E Goubanova K Le Vu B Gutknecht E Garcon V 2015 Downscalingbiogeochemistry in the Benguela eastern boundary current Ocean Model 90 57ndash71

Mahadevan A Campbell JW 1926 Biogeochemical patchiness at the sea surfaceGeophys Res Lett 29 (19) 2002 httpsdoiorg1010292001GL014116

Mariotti A Ruti PM Rixen M 2018 Progress in subseasonal to seasonal predictionthrough a joint weather and climate community effort npj Climate Atmos Sci 1httpsdoiorg101038s41612-018-0014-z

Marmion M Parviainen M Luoto M Heikkinen RK Thuiller W 2009 Evaluationof consensus methods in predictive species distribution modelling Divers Distrib 15(1) 59ndash69

McKinnon KA Rhines A Tingley MP Huybers P 2016 Long-lead predictions ofeastern United States hot days from Pacific sea surface temperatures Nat Geosci 9389ndash394 httpsdoiorg101038ngeo2687

Meinvielle M Johnson GC 2013 Decadal water-property trends in the CaliforniaUndercurrent with implications for ocean acidification J Geophys Res Oceans 118(12) 6687ndash6703

Merryfield WJ et al 2013 The Canadian seasonal to interannual prediction systemPart I models and initialization Mon Weather Rev 141 2910ndash2945

Meyers SD Johnson MA Liu M OBrien JJ 1996 Interdecadal variability in anumerical model of the northeast Pacific Ocean 1970ndash1989 J Phy Oceanogr 262635ndash2652

Miller AJ White WB Cayan DR 1997 North Pacific thermocline variations onENSO timescales J Clim 27 2023ndash2039

Miller TJ Hare JA Alade LA 2016 A state-space approach to incorporating en-vironmental effects on recruitment in an age-structured assessment model with anapplication to Southern New England yellowtail flounder Can J Fish Aquat Sci 731261ndash1270

Mills KE Pershing AJ Hernaacutendez CM 2017 Forecasting the seasonal timing ofMaines lobster fishery Front Mar Sci 4 337

Monim M 2017 Seasonal and Inter-annual Variability of Gulf Stream Warm Core Ringsfrom 2000 to 2016 University of Massachusetts-Dartmouth pp 113

Mountain DG 2012 Labrador slope water entering the Gulf of Mainemdashresponse to theNorth Atlantic Oscillation Cont Shelf Res 47 150ndash155 httpsdoiorg101016jcsr201207008

Moss JH Farley Jr EV Feldmann AM Ianelli JN 2009 Spatial distribution en-ergetic status and food habits of eastern Bering Sea age-0 walleye pollock Trans AmFish Soc 138 (3) 497ndash505

Mueter FJ Litzow MA 2008 Sea ice retreat alters the biogeography of the Bering Seacontinental shelf Ecol Appl 18 309ndash320

Murphree T Bograd SJ Schwing FB Ford B 2003 (2003) Large scale atmosphere-ocean anomalies in the northeast Pacific during 2002 Geophys Res Lett 30 8026httpsdoiorg1010292003GL017303

Myers RA 1998 When do environmentndashrecruitment correlations work Rev Fish BiolFish 8 (3) 285ndash305

Nam S Takeshita Y Frieder CA Martz T Ballard J 2015 Seasonal advection ofPacific Equatorial Water alters oxygen and pH in the Southern California Bight JGeophys Res Oceans 120 (8) 5387ndash5399

Namias J Born RM 1970 Temporal coherence in North Pacific sea-surface tem-perature patterns J Geophys Res 75 5952ndash5955

Namias J Born RM 1974 Further studies of temporal coherence in North Pacific seasurface temperatures J Geophys Res 79 797ndash798

Neveu E Moore AM Edwards CA Fiechter J Drake P Crawford WJ JacoxMG Nuss E 2016 An historical analysis of the California Current circulation usingROMS 4D-Var System configuration and diagnostics Ocean Model 99 133ndash151

Newman M Sardeshmukh PD Penland C 2009 How important is air-sea coupling inENSO and MJO evolution J Clim 22 2958ndash2977

Newman M Shin S-I Alexander MA 2011 Natural variation in ENSO flavorsGeophys Res Lett 38 L14705 httpsdoiorg1010292011GL047658

Newman M Sardeshmukh PD 2017 Are we near the predictability limit of tropicalIndo-Pacific sea surface temperatures Geophys Res Lett 44 8520ndash8529

Nye JA Joyce TM Kwon YO Link JS 2011 Silver hake tracks changes inNorthwest Atlantic circulation Nat Commun 2 412

Ortiz I Aydin K Hermann AJ Gibson GA Punt AE Wiese FK Eisner LBFerm N Buckley TW Moffitt EA Ianelli JN Dalton M Cheng W Wang MHedstrom K Bond NA Curchitser EN Boyd C 2016 Climate to fishSynthesizing field work data and models in a 39-year retrospective analysis of sea-sonal processes on the eastern Bering Sea shelf and slope Deep-Sea Res II 134390ndash412

Osychny V Cornillon P 2004 Properties of rossby waves in the North Atlantic esti-mated from satellite data J Phys Oceanogr 34 61ndash76 httpsdoiorg1011751520-0485(2004) 034lt0061PORWITgt20CO2

Park S Leovy CB 2004 Marine low-cloud anomalies associated with ENSO J Clim17 (17) 3448ndash3469

Payne MR Hobday AJ MacKenzie BR Tommasi D Dempsey DP Faumlssler SMet al 2017 Lessons from the first generation of marine ecological forecast productsFront Mar Sci 4 289

Pegion K Kirtman BP Becker E Collins DC LaJoie E Burgman R et al 2019The Subseasonal Experiment (SubX) A multi-model subseasonal prediction experi-ment Bull Am Meteorol Soc httpsdoiorg101175BAMS-D-18-02701

Pentildea-Molino B Joyce TM 2008 Variability in the Slope Water and its relation to theGulf Stream path Geophys Res Lett 35 L03606 httpsdoiorg1010292007GL032183

Peng P Kumar A Halpert MS Barnston A 2012 An analysis of CPCrsquos operational05 month lead seasonal outlooks Weather Forecast 27 898ndash917

Penland C Sardeshmukh PD 1995 The optimal growth of tropical sea surface tem-perature anomalies J Clim 8 1999ndash2024

Peterson WT Fisher JL Peterson JO Morgan CA Burke BJ Fresh KL 2014Applied fisheries oceanography ecosystem indicators of ocean conditions informfisheries management in the California Current Oceanography 27 80ndash89

Pierce SD Smith RL Kosro PM Barth JA Wilson CD 2000 Continuity of thepoleward undercurrent along the eastern boundary of the mid-latitude north PacificDeep Sea Res Part II 47 (5ndash6) 811ndash829

Pinsky ML Mantua NJ 2014 Emerging adaptation approaches for climate-readyfisheries management Oceanography 27 146ndash159

Pirhalla DE Sheridan SC Ransi V Lee CC 2015 Assessing cold-snap and mortalityevents in South Florida coastal ecosystems development of a biological cold stressindex using satellite SST and weather pattern forcing Estuaries Coasts 382310ndash2322

Potts JM 2011 Basic concepts In Jolliffe IT Stephenson DB (Eds) ForecastVerification Wiley Chichester pp 11ndash29

Pozo Buil M Di Lorenzo E 2015 Decadal changes in Gulf of Alaska upwelling sourcewaters Geophys Res Lett 42 1488ndash1495 httpsdoiorg1010022015GL063191

Pozo Buil M Di Lorenzo E 2017 Decadal dynamics and predictability of oxygen andsubsurface tracers in the California Current System Geophys Res Lett 444204ndash4213 httpsdoiorg1010022017GL072931

Quan X Hoerling M Whitaker J Bates G Xu T 2006 Diagnosing sources of USseasonal forecast skill J Clim 19 (13) 3279ndash3293

Reed RK Halpern D 1976 Observations of the California undercurrent offWashington and Vancouver Island Limnol Oceanogr 21 (3) 389ndash398

Reynolds RW Smith TM Liu C Chelton DB Casey KS Schlax MG 2007 Dailyhigh-resolution-blended analyses for sea surface temperature J Clim 205473ndash5496

Richaud B Kwon Y-O Joyce TM Fratantoni PS Lentz SJ 2016 Surface andbottom temperature and salinity climatology along the continental shelf off the

Canadian and US East Coasts Cont Shelf Res 124 165ndash181 httpsdoiorg101016jcsr201606005

Roberts DR Bahn V Ciuti S Boyce MS Elith J Guillera-Arroita G et al 2017Cross-validation strategies for data with temporal spatial hierarchical or phyloge-netic structure Ecography 40 (8) 913ndash929

Rossby T 1996 The North Atlantic current and surrounding waters at the crossroadsRev Geophys 34 463ndash481 httpsdoiorg10102996RG02214

Rossby T Benway RL 2000 Slow variations in mean path of the Gulf Stream east ofCape Hatteras Geophys Res Lett 27 117ndash120 httpsdoiorg1010291999GL002356

Sanchez-Franks A Hameed S Wilson RE 2016 The icelandic low as a predictor ofthe gulf stream north wall position J Phys Oceanogr 46 817ndash826 httpsdoiorg101175JPO-D-14-02441

Sasaki YN Schneider N Maximenko N Lebedev K 2010 Observational evidencefor propagation of decadal spiciness anomalies in the North Pacific Geophys ResLett 37 L07708 httpsdoiorg1010292010GL042716

Schoof JT 2013 Statistical downscaling in climatology Geography Compass 7 (4)249ndash265

Schulte JA Georgas N Saba V Howell P 2018 North Pacific influences on longisland sound temperature variability J Climate 31 2745ndash2769 httpsdoiorg101175JCLI-D-17-01351

Seferian R et al 2016 Inconsistent strategies to spin-up models in CMIP5 implicationsfor ocean biogeochemical model performance assessment Geosci Model Dev 91927ndash11851

Shearman RK Lentz SJ 2010 Long-term sea surface temperature variability along theUS East Coast J Phys Oceanogr 40 1004ndash1017 httpsdoiorg1011752009JPO43001

Sheridan SC Pirhalla DE Lee CC Ransibrahmanakul V 2013 Evaluating linkagesof weather patterns and water quality responses in South Florida using a synopticclimatological approach J Appl Meteorol Climatol 52 (2) 425ndash438 httpsdoiorg101175JAMC-D-12-01261

Sheridan SC Pirhalla DE Lee CC Ransibrahmanakul V 2017 Atmospheric driversof sea-level fluctuations and nuisance floods along the mid-Atlantic coast of the USAReg Environ Change 17 (6) 1853ndash1861

Siddon E Zador S 2018 Ecosystem Status Report 2018 Status of the Eastern BeringSea Marine Ecosystem Stock Assessment and Fishery Evaluation Report NorthPacific Fishery Management Council 605 W 4th Ave Suite 306 Anchorage AK99501

Siedlecki SA Banas NS Davis KA Giddings S Hickey BM MacCready PConnolly T Geier S 2015 Seasonal and interannual oxygen variability on theWashington and Oregon continental shelves J Geophys Res Oceans 120 (2)608ndash633

Siedlecki SA Kaplan IC Hermann AJ Nguyen TT Bond NA Newton JA et al2016 Experiments with Seasonal Forecasts of ocean conditions for the Northernregion of the California Current upwelling system Sci Rep 6 27203

Sigler MF Stabeno PJ Eisner LB Napp JM Mueter FM 2014 Spring and fallphytoplankton blooms in a productive subarctic ecosystem the eastern Bering Seaduring 1995ndash2011 Deep-Sea Res II 109 71ndash83

Stabeno PJ Kachel NB Moore SE Napp JM Sigler M Yamaguchi A ZerbiniAN 2012 Comparison of warm and cold years on the southeastern Bering Sea shelfand some implications for the ecosystem Deep Sea Res Part II 65ndash70 31ndash45

Stabeno PJ Duffy-Anderson JT Eisner LB Farley EV Heintz RA Mordy CW2017 Return of warm conditions in the southeastern Bering Sea physics to fluor-escence PLoS ONE 12 (9) e0185464 httpsdoiorg101371journalpone0185464

Stergiou KI Christou ED 1996 Modelling and forecasting annual fisheries catchescomparison of regression univariate and multivariate time series methods Fish Res25 (2) 105ndash138

Stergiou KI Christou ED Petrakis G 1997 Modelling and forecasting monthlyfisheries catches comparison of regression univariate and multivariate time seriesmethods Fish Res 29 (1) 55ndash95

Stock CA Pegion K Vecchi GA Alexander MA Tommasi D Bond NAFratantoni PS Gudgel RG Kristiansen T OrsquoBrien TD Xue Y Yang X 2015Seasonal sea surface temperature anomaly prediction for coastal ecosystems ProgOceanogr 137 219ndash236

Stumpf RP Tomlinson MC Calkins JA Kirkpatrick B Fisher K Nierenberg Ket al 2009 Skill assessment for an operational algal bloom forecast system J MarSyst 76 (1ndash2) 151ndash161

Sturges W Hong BG 2001 Gulf stream transport variability at periods of decades JPhys Oceanogr 31 (5) 1304ndash1312

Sturges W Hong BG 1995 Wind forcing of the atlantic thermocline along 32degN at lowfrequencies J Phys Oceanogr 25 1706ndash1715

Sun C Feng M Matear RJ Chamberlain MA Craig P Ridgway KR Schiller A2012 Marine downscaling of a future climate scenario for Australian boundarycurrents J Clim 25 (8) 2947ndash2962

Sweet WV Park J Marra JJ Zervas C Gill S 2014 Sea Level Rise and NuisanceFlood Frequency Changes across the United States NOAA Technical Report NOS CO-OPS 073 Silver Spring Maryland

Taguchi B Schneider N 2014 Origin of decadal-scale eastward-propagating heatcontent anomalies in the North Pacific J Clim 27 (20) 7568ndash7586 httpsdoiorg101175Jcli-D-13-001021

Taguchi B Schneider N Nonaka M Sasaki H 2017 Decadal variability of upper-ocean heat content associated with meridional shifts of western boundary currentextensions in the North Pacific J Climate 30 6247ndash6264 httpsdoiorg101175Jcli-D-16-07791

Thomson RE Krassovski MV 2010 Poleward reach of the California Undercurrent

extension J Geophys Res Oceans 115 C09027 httpsdoiorg1010292010JC006280

Thorne LH Baird RW Webster DL Stepanuk JE Read AJ 2019 Predictingfisheries bycatch a case study and field test for pilot whales in a pelagic longlinefishery Divers Distrib httpsdoiorg101111ddi12912

Tommasi D Stock CA Hobday AJ Methot R Kaplan IC Eveson JP et al2017a Managing living marine resources in a dynamic environment the role ofseasonal to decadal climate forecasts Prog Oceanogr 152 15ndash49

Tommasi D Stock C Pegion K Vecchi GA Methot RD Alexander M CheckleyD 2017b Improved management of small pelagic fisheries through seasonal climateprediction Ecol Appl 27 378ndash388

Trenberth EK Branstrator GW Karoly D Kumar A Lau N-C Ropelewski C1998 Progress during TOGA in understanding and modeling global teleconnectionsassociated with tropical sea surface temperatures J Geophys Res 107 (C7)14291ndash14324

Turi G Lachkar Z Gruber N Muumlnnich M 2016 Climatic modulation of recent trendsin ocean acidification in the California Current System Environ Res Lett 11 (1)014007

Turi G Alexander M Lovenduski NS Capotondi A Scott J Stock C et al 2018Response of O2 and pH to ENSO in the California Current System in a high-resolutionglobal climate model Ocean Science (OS) 14 (1) 69ndash86 httpsdoiorg105194os-14-69-2018

Turner SM Hare JA Manderson JP Richardson DE Hoey JJ 2017 Evaluationof species distribution forecasts a potential predictive tool for reducing incidentalcatch in pelagic fisheries Can J Fish Aquat Sci 74 (11) 1717ndash1731

Van Hooidonk R Maynard JA Liu Y Lee SK 2015 Downscaled projections ofCaribbean coral bleaching that can inform conservation planning Glob Change Biol21 (9) 3389ndash3401

Van Oostende N Dussin R Stock CA Barton AD Curchitser E Dunne JP WardBB 2018 Simulating the oceanrsquos chlorophyll dynamic range from coastal upwellingto oligotrophy Prog Oceanogr 168 232ndash247

Veneziani M Edwards CA Doyle JD Foley D 2009 A central California coastalocean modeling study 1 Forward model and the influence of realistic versus cli-matological forcing J Geophys Res Oceans 114 C04015 httpsdoiorg1010292008JC004774

Vimont DJ Wallace JM Battisti DS 2003 The seasonal footprinting mechanism inthe Pacific implications for ENSO J Clim 16 (16) 2668ndash2675

Vitart F Ardilouze C Bonet A Brookshaw A Chen M Codorean C Deacutequeacute MFerranti L Fucile E Fuentes M Hendon H 2017 The subseasonal to seasonal(S2S) prediction project database Bull Am Meteorol Soc 98 (1) 163ndash173

Walters CJ Martell JD 2004 Fisheries Ecology and Management PrincetonUniversity Press Princeton New Jersey

Ward EJ Holmes EE Thorson JT Collen B 2014 Complexity is costly a meta-analysis of parametric and non-parametric methods for short-term population fore-casting Oikos 123 (6) 652ndash661

Welch H Hazen EL Bograd SJ Jacox MG Brodie S Robinson D et al 2019aPractical considerations for operationalizing dynamic management tools J ApplEcol 56 (2) 459ndash469

Welch H Hazen EL Briscoe DK Bograd SJ Jacox MG Eguchi T et al 2019bEnvironmental indicators to reduce loggerhead turtle bycatch offshore of SouthernCalifornia Ecol Ind 98 657ndash664

Wells BK Santora JA Henderson MJ Warzybok P Jahncke J Bradley RWet al 2017 Environmental conditions and prey-switching by a seabird predatorimpact juvenile salmon survival J Mar Syst 174 54ndash63

Wenger SJ Olden JD 2012 Assessing transferability of ecological models an un-derappreciated aspect of statistical validation Methods Ecol Evol 3 260ndash267

Wetterhall F Baacuterdossy A Chen D Halldin S Xu CY 2009 Statistical downscalingof daily precipitation over Sweden using GCM output Theor Appl Climatol 96(1ndash2) 95

Widlansky MJ Marra JJ Chowdhury MR Stephens SA Miles ER FauchereauN Spillman CM Smith G Beard G Wells J 2017 Multi-model ensemble sealevel forecasts for tropical Pacific islands J Appl Meteorol 56 849ndash862

Wilby RL Charles SP Zorita E Timbal B Whetton P Mearns LO 2004Guidelines for use of climate scenarios developed from statistical downscalingmethods Supporting material of the Intergovernmental Panel on Climate Changeavailable from the DDC of IPCC TGCIA 27

Wilks DS 2011 Statistical Methods in the Atmospheric Sciences third ed AcademicAmsterdam pp 704

Winship AJ OrsquoFarrell MR Satterthwaite WH Wells BK Mohr MS 2015Expected future performance of salmon abundance forecast models with varyingcomplexity Can J Fish Aquat Sci 72 (4) 557ndash569

Wunsch C Heimbach P 2008 How long to ocean tracer and proxy equilibrium QuatSci Rev 27 637ndash651

Xiu P Chai F Curchitser EN Castruccio FS 2018 Future changes in coastal up-welling ecosystems with global warming the case of the California Current SystemSci Rep 8 (1) 2866

Xu H Kim H-M Nye JA Hameed S 2015 Impacts of the North Atlantic Oscillationon sea surface temperature on the Northeast US Continental Shelf Cont Shelf Res105 60ndash66 httpsdoiorg101016jcsr201506005

Xue Y et al 2017 A real-time ocean reanalysis intercomparison project in the contextof tropical Pacific observing system and ocean monitoring Clim Dyn 493647ndash3672

Yang JY 1999 A linkage between decadal climate variations in the Labrador Sea andthe tropical Atlantic Ocean Geophys Res Lett 26 1023ndash1026 httpsdoiorg1010291999GL900181

Yarnal B 1993 Synoptic Climatology in Environmental Analysis A Primer Belhavenpress London

Yates KL Bouchet PJ Caley MJ Mengersen K Randin CF Parnell S et al2018 Outstanding challenges in the transferability of ecological models Trends EcolEvol 33 (10) 790ndash802 httpsdoiorg101016jtree201808001

Yeager S Danabasoglu G Rosenbloom N Strand W Bates S Meehl G KarspeckA Lindsay K Long M Teng H Lovenduski N 2018 Predicting near-termchanges in the Earth System a large ensemble of initialized decadal prediction si-mulations using the Community Earth System Model Bull Amer Meteor Soc 991867ndash1886 httpsdoiorg101175BAMS-D-17-00981

Zador SG Holsman KK Aydin KY Gaichas SK 2016 Ecosystem consideration sinAlaska the value of qualitative assessments ICES J Mar Sci 74 421ndash430

Zhang HH Wu LX 2010 Predicting North Atlantic sea surface temperature variabilityon the basis of the first-mode baroclinic Rossby wave model J GeophysResmdashOceans 115 C09030 httpsdoiorg1010292009jc006017

Zhang WG Gawarkiewicz GG 2015 Dynamics of the direct intrusion of Gulf Streamring water onto the Mid-Atlantic Bight shelf Geophys Res Lett 42 7687ndash7695

Seasonal-to-interannual prediction of North American coastal marine ecosystems Forecast methods mechanisms of predictability and priority developments

Introduction

Scope of marine ecosystem forecasts

Marine ecosystem forecasting methods

Physical forecasting methods

Coupled global climate forecasts

Dynamically downscaled regional forecasts

Statistical downscaling and forecasting

Hybrid statisticaldynamical approaches

Biogeochemical and ecological forecasting methods


Statistical forecasts

Mechanisms of predictability in marine ecosystems

Mechanisms of physical predictability

Oceanic processes

Persistence

Re-emergence

Ocean waves

Coastal waves

Baroclinic Rossby waves

Advection

Tropical-extratopical connections

Sea-ice related processes in subarctic regions

Mechanisms of biogeochemical and ecological predictability

Biogeochemical response to physical forcing

Species responses to environmental change

Speciesrsquo life history

Challenges and priority developments

Better resolution of key processes in general circulation models

Reforecast skill assessments

Forecast uncertainty and ensembles

Computing resources

Observational limitations

Forecast user engagement

Conclusions and summary recommendations

mkH1_35

Acknowledgments

References

Large marine ecosystem properties ranging from coastal sea level to marine top predator distributions We then describe known me-chanisms governing marine ecosystem predictability and how they have been used in forecasts to date Thesemechanisms include physical atmospheric and oceanic processes biogeochemical and ecological responses tophysical forcing and intrinsic characteristics of species themselves In reviewing the state of the knowledge onforecasting techniques and mechanisms underlying marine ecosystem predictability we aim to facilitate forecastdevelopment and uptake by (i) identifying methods and processes that can be exploited for development ofskillful regional forecasts (ii) informing priorities for forecast development and verification and (iii) improvingunderstanding of conditional forecast skill (ie a priori knowledge of whether a forecast is likely to be skillful)While we focus primarily on coastal marine ecosystems surrounding North America (and the US in particular)we detail forecast methods physical and biological mechanisms and priority developments that are globallyrelevant

1 Introduction

Businesses government agencies and the public rely on weatherforecasts for planning and responding to changing environmental con-ditions While these forecasts are useful to ocean-related sectors thereis also high and increasing demand for ocean forecasts to supportproactive decision making in the marine environment (eg for fish-eries aquaculture shipping energy tourism and public health)Forewarning of ocean conditions can help decision makers reduce ne-gative impacts in unfavorable conditions and maximize opportunities infavorable conditions (Tommasi et al 2017a Hobday et al 20162018) The rise of marine ecological forecasting research in recent yearsbuilds upon a substantial knowledge base and infrastructure in climateforecasting (Doblas-Reyes et al 2013) and has been driven largely byincreasing demand for decision-support tools to help marine stake-holders prepare for and adapt to ecosystem variability and change(Payne et al 2017 Tommasi et al 2017a) including extreme oceanconditions such as marine heatwaves (Holbrook et al 2019) ecosystemhealth stressors (Siedlecki et al 2016) and the lack of sea ice in sub-arcticarctic waters (Stabeno et al 2017)

Marine policy and resource planning decisions are made based onforecasts and outlook information on various timescales In contrast toclimate projections which are used to explore long-term trends basedon various scenarios and mostly inform policy level decisions and long-term planning forecasts for days to years aim to provide prognosticinformation of ocean conditions on the timescales relevant to marineresource managers and other end-users A number of marine ecosystemforecasts have been cited in the literature (eg Payne et al 2017)though many are in fact nowcasts or target short lead times (ie days)For example nowcast tools that predict marine species distributionsbased on observed environmental conditions have been developed tobetter manage fisheries bycatch and ship-strike risks (Howell et al2008 Hazen et al 2017 2018 Welch et al 2019a Thorne et al2019) and three-day probability forecasts for blooms of the harmful

algal bloom (HAB) forming phytoplankton Pseudo-nitzchia alert man-agers to potential shellfish toxicity that could necessitate fishery clo-sures (Anderson et al 2016) Similarly coastal high water levels havehistorically been a more suitable target for short-lead prediction with acrucial focus on inundation impacts on timescales from hours to daysHowever in many cases longer lead times (ie 1ndash24 months) arebetter aligned with management and industry decisions Forecast sys-tems for southern bluefin tuna in Australia with several months leadtime have been useful products for fishers to plan the start of theirfishing season (Eveson et al 2015) and for managers to set seasonalarea closures based on impending ocean conditions (Hobday et al2011) In the eastern Bering Sea preliminary seasonal predictions ofthe ldquocold poolrdquo (ie the area where bottom temperature islt 2 degC) wereincluded in Ecosystem Status Reports from 2015 to 2018 to allowmanagers to consider fishing quotas in light of expected conditions(Siddon and Zador 2018) Indeed marine ecosystems are well suited toseasonal-to-interannual (S2I 1ndash24 month) prediction given the con-vergence of ecosystem variability physical drivers and managementand industry decisions that take place on these timescales A number ofefforts are underway to develop marine ecosystem forecasts for S2Itimescales along the coasts of the United States (US) leveraging im-proved technological capabilities model development and scientificunderstanding (Fig 1 Table 1) The S2I timescale is the focus of thispaper though we briefly discuss shorter and longer timescales as ap-propriate

The forecasts outlined above differ greatly in the models observa-tions and assumptions upon which they rely and the specific end-userneeds they are meant to address Of particular note for this paper theyalso exploit a wide range of physical biogeochemical and ecologicalprocesses that give rise to predictability on different spatial and tem-poral scales However documentation of an apparently predictableresponse in the marine ecosystem often precedes a thorough under-standing of the mechanisms underlying that predictability For ex-ample while recent studies have suggested multiple potential S2I

Fig 1 Locations of active S2I marine ecosystem forecasting efforts along US coastlines Label numbers correspond to entries in Table 1 Ocean bottom depth (m) isindicated in bluewhite shading (For interpretation of the references to colour in this figure legend the reader is referred to the web version of this article)

Table1

Summaryof

season

al-to-interann

forecastsalon

oastlin

esinclud

ymecha

ictability

Correspon

location

sareshow

initalicsun

derldquoForecast

rdquoindicate

rrespo

ndingsubsection

Section3

ldquoMarineEc

osystem

ForecastingMetho

dsrdquo

Forecast

ictability

ration

alstatus

Referen

sternBe

Summer

(July1)

xsea-ice

amically

downscaledph

ysical

forecast

Persistenc

esea-iceproc

throug

hUnimak

talprod

uctfor

Herman

in(201

4)httpsaccessafscno

cificNorthwest

physicsan

emistry

sardine

distribu

amically

downscaledph

ysical

emical

distribu

el(3123

21322)

Persistenc

SOteleco

nnection

sspeciesrespon

environm

ebprod

arSied

leckie

tal(20

16)Kap

oosorg

uctsj-sco

homeph

physics

speciesdistribu

s(swordfi

shleathe

rbackturtles

Californiasea

lionsb

luesharksb

luewha

amically

downscaledph

ysical

forecast

els(312

Persistenc

SOteleco

nnection

sspeciesrespon

environm

velopm

eziani

etal(20

09)forph

ysical

dAbrah

Seasurfacetempe

ratureseasurfacehe

Statisticalforecast

(linearinve

rsemod

Persistenc

ical-extratrop

teleco

nnection

etal(20

ggiani

Diaset

Pacific

Island

sealeve

lGloba

namical

forecaststatistical

forecasts(3113

Persistenc

eanicRossbywav

ebprod

tedmon

Widlansky

etal(20

httpsuh

slcsoesth

awaiie

dusea-le

forecasts

6Coterminou

coasts

sealeve

tworkmod

Persistenc

eatmosph

forcing

opticatmosph

patterns)

velopm

eridan

etal(20

7Northeast

Shelftempe

ratures

pecies

distribu

interan

llowtailflou

nderAtlan

dsilver

butterfi

ultipletype

s)(313

Persistenc

eGulfStream

lfinteraction

vection

atmosph

teleco

nnection

velopm

etal(20

15)Dav

8Northeast

Seasurfacetempe

ilotwha

leriver

herring)

Statistically

downscaledclim

ateforecast

(delta

d)statistical

speciesdistribu

els(3143

Persistenc

especiesrespon

environm

entalch

velopm

Thorne

etal(20

9Gulfof

Lobsterland

Statisticalforecast

basedon

rature

Speciesrespon

environm

perimen

Millset

wwgmriorgo

research

projectsg

ulf-maine

10Gulfof

toxicdino

ellate

ndrium

Statisticalforecast

basedon

observed

sembleof

itions

history

speciesrespon

environm

entalch

ualforecastspa

ersonet

httpswwww

ebsiten

ortheast-

forecasting

Model development

Communication

Acknowledgments

References

Introduction













Oceanic processes

Persistence

Re-emergence

Ocean waves

Coastal waves


Advection











Computing resources




mkH1_35

Acknowledgments

References

Table1

Summaryof

season

al-to-interann

forecastsalon

oastlin

esinclud

ymecha

ictability

Correspon

location

sareshow

initalicsun

derldquoForecast

rdquoindicate

rrespo

ndingsubsection

Section3

ldquoMarineEc

osystem

ForecastingMetho

dsrdquo

Forecast

ictability

ration

alstatus

Referen

sternBe

Summer

(July1)

xsea-ice

amically

downscaledph

ysical

forecast

Persistenc

esea-iceproc

throug

hUnimak

talprod

uctfor

Herman

in(201

4)httpsaccessafscno

cificNorthwest

physicsan

emistry

sardine

distribu

amically

downscaledph

ysical

emical

distribu

el(3123

21322)

Persistenc

SOteleco

nnection

sspeciesrespon

environm

ebprod

arSied

leckie

tal(20

16)Kap

oosorg

uctsj-sco

homeph

physics

speciesdistribu

s(swordfi

shleathe

rbackturtles

Californiasea

lionsb

luesharksb

luewha

amically

downscaledph

ysical

forecast

els(312

Persistenc

SOteleco

nnection

sspeciesrespon

environm

velopm

eziani

etal(20

09)forph

ysical

dAbrah

Seasurfacetempe

ratureseasurfacehe

Statisticalforecast

(linearinve

rsemod

Persistenc

ical-extratrop

teleco

nnection

etal(20

ggiani

Diaset

Pacific

Island

sealeve

lGloba

namical

forecaststatistical

forecasts(3113

Persistenc

eanicRossbywav

ebprod

tedmon

Widlansky

etal(20

httpsuh

slcsoesth

awaiie

dusea-le

forecasts

6Coterminou

coasts

sealeve

tworkmod

Persistenc

eatmosph

forcing

opticatmosph

patterns)

velopm

eridan

etal(20

7Northeast

Shelftempe

ratures

pecies

distribu

interan

llowtailflou

nderAtlan

dsilver

butterfi

ultipletype

s)(313

Persistenc

eGulfStream

lfinteraction

vection

atmosph

teleco

nnection

velopm

etal(20

15)Dav

8Northeast

Seasurfacetempe

ilotwha

leriver

herring)

Statistically

downscaledclim

ateforecast

(delta

d)statistical

speciesdistribu

els(3143

Persistenc

especiesrespon

environm

entalch

velopm

Thorne

etal(20

9Gulfof

Lobsterland

Statisticalforecast

basedon

rature

Speciesrespon

environm

perimen

Millset

wwgmriorgo

research

projectsg

ulf-maine

10Gulfof

toxicdino

ellate

ndrium

Statisticalforecast

basedon

observed

sembleof

itions

history

speciesrespon

environm

entalch

ualforecastspa

ersonet

httpswwww

ebsiten

ortheast-

forecasting

Model development

Communication

Acknowledgments

References

Introduction













Oceanic processes

Persistence

Re-emergence

Ocean waves

Coastal waves


Advection











Computing resources




mkH1_35

Acknowledgments

References

Model development

Communication

Acknowledgments

References

Introduction













Oceanic processes

Persistence

Re-emergence

Ocean waves

Coastal waves


Advection











Computing resources




mkH1_35

Acknowledgments

References

Model development

Communication

Acknowledgments

References

Introduction













Oceanic processes

Persistence

Re-emergence

Ocean waves

Coastal waves


Advection











Computing resources




mkH1_35

Acknowledgments

References

Model development

Communication

Acknowledgments

References

Introduction













Oceanic processes

Persistence

Re-emergence

Ocean waves

Coastal waves


Advection











Computing resources




mkH1_35

Acknowledgments

References

Model development

Communication

Acknowledgments

References

Introduction













Oceanic processes

Persistence

Re-emergence

Ocean waves

Coastal waves


Advection











Computing resources




mkH1_35

Acknowledgments

References

Model development

Communication

Acknowledgments

References

Introduction













Oceanic processes

Persistence

Re-emergence

Ocean waves

Coastal waves


Advection











Computing resources




mkH1_35

Acknowledgments

References

Model development

Communication

Acknowledgments

References

Introduction













Oceanic processes

Persistence

Re-emergence

Ocean waves

Coastal waves


Advection











Computing resources




mkH1_35

Acknowledgments

References

Model development

Communication

Acknowledgments

References

Introduction













Oceanic processes

Persistence

Re-emergence

Ocean waves

Coastal waves


Advection











Computing resources




mkH1_35

Acknowledgments

References

Model development

Communication

Acknowledgments

References

Introduction













Oceanic processes

Persistence

Re-emergence

Ocean waves

Coastal waves


Advection











Computing resources




mkH1_35

Acknowledgments

References

Model development

Communication

Acknowledgments

References

Introduction













Oceanic processes

Persistence

Re-emergence

Ocean waves

Coastal waves


Advection











Computing resources




mkH1_35

Acknowledgments

References

Model development

Communication

Acknowledgments

References

Introduction













Oceanic processes

Persistence

Re-emergence

Ocean waves

Coastal waves


Advection











Computing resources




mkH1_35

Acknowledgments

References

Model development

Communication

Acknowledgments

References

Introduction













Oceanic processes

Persistence

Re-emergence

Ocean waves

Coastal waves


Advection











Computing resources




mkH1_35

Acknowledgments

References

Model development

Communication

Acknowledgments

References

Introduction













Oceanic processes

Persistence

Re-emergence

Ocean waves

Coastal waves


Advection











Computing resources




mkH1_35

Acknowledgments

References

Model development

Communication

Acknowledgments

References

Introduction













Oceanic processes

Persistence

Re-emergence

Ocean waves

Coastal waves


Advection











Computing resources




mkH1_35

Acknowledgments

References

Model development

Communication

Acknowledgments

References

Introduction













Oceanic processes

Persistence

Re-emergence

Ocean waves

Coastal waves


Advection











Computing resources




mkH1_35

Acknowledgments

References

Model development

Communication

Acknowledgments

References

Introduction













Oceanic processes

Persistence

Re-emergence

Ocean waves

Coastal waves


Advection











Computing resources




mkH1_35

Acknowledgments

References

Acknowledgments

References

Introduction













Oceanic processes

Persistence

Re-emergence

Ocean waves

Coastal waves


Advection











Computing resources




mkH1_35

Acknowledgments

References

Documents

Progress in Oceanography - cpb-us-e1.wpmucdn.com€¦ · Progress in Oceanography ... Seasonal-to-interannual prediction of North American coastal marine ecosystems: Forecast methods,