Full Terms & Conditions of access and use can be found athttp://www.tandfonline.com/action/journalInformation?journalCode=tjoo20

Download by: [65.190.89.127] Date: 20 May 2017, At: 18:50

Journal of Operational Oceanography

ISSN: 1755-876X (Print) 1755-8778 (Online) Journal homepage: http://www.tandfonline.com/loi/tjoo20

Advancing coastal ocean modelling, analysis, andprediction for the US Integrated Ocean ObservingSystem

John Wilkin, Leslie Rosenfeld, Arthur Allen, Rebecca Baltes, Antonio Baptista,Ruoying He, Patrick Hogan, Alexander Kurapov, Avichal Mehra, JosieQuintrell, David Schwab, Richard Signell & Jane Smith

To cite this article: John Wilkin, Leslie Rosenfeld, Arthur Allen, Rebecca Baltes, Antonio Baptista,Ruoying He, Patrick Hogan, Alexander Kurapov, Avichal Mehra, Josie Quintrell, David Schwab,Richard Signell & Jane Smith (2017): Advancing coastal ocean modelling, analysis, and predictionfor the US Integrated Ocean Observing System, Journal of Operational Oceanography, DOI:10.1080/1755876X.2017.1322026

To link to this article: http://dx.doi.org/10.1080/1755876X.2017.1322026

Published online: 20 May 2017.

Submit your article to this journal

View related articles

View Crossmark data

Advancing coastal ocean modelling, analysis, and prediction for the US IntegratedOcean Observing SystemJohn Wilkin a, Leslie Rosenfeld b, Arthur Allen c, Rebecca Baltes d, Antonio Baptista e, Ruoying He f,Patrick Hogan g, Alexander Kurapovh, Avichal Mehrai, Josie Quintrellj, David Schwab k, Richard Signell l

and Jane Smithm

aMarine and Coastal Sciences, Rutgers, The State University of New Jersey, New Brunswick, NJ, USA; bCeNCOOS, MBARI, Moss Landing, CA, USA;cOffice of Search and Rescue, U.S. Coast Guard, New London, CT, USA; dU.S. IOOS Program Office, Silver Spring, MD, USA; eCenter for CoastalMargin Observation & Prediction, Oregon Health & Science University, Portland, OR, USA; fMarine, Earth and Atmospheric Sciences, NorthCarolina State University, Raleigh, NC, USA; gU.S. Naval Research Laboratory Stennis Space Center, Kiln, MS, USA; hCollege of Oceanic andAtmospheric Sciences, Oregon State University, Corvallis, OR, USA; iEnvironmental Modeling Center, National Centers for EnvironmentalPrediction, NOAA/National Weather Service, College Park, MD, USA; jIOOS Association, Harpswell, ME, USA; kUniversity of Michigan WaterCenter, Ann Arbor, MI, USA; lU.S. Geological Survey, Woods Hole, MA, USA; mEngineer Research and Development Center, U.S. Army Corps ofEngineers, Vicksburg, MS, USA

ABSTRACTThis paper outlines strategies that would advance coastal ocean modelling, analysis and predictionas a complement to the observing and data management activities of the coastal components ofthe US Integrated Ocean Observing System (IOOS®) and the Global Ocean Observing System(GOOS). The views presented are the consensus of a group of US-based researchers with a cross-section of coastal oceanography and ocean modelling expertise and community representationdrawn from Regional and US Federal partners in IOOS. Priorities for research and developmentare suggested that would enhance the value of IOOS observations through model-basedsynthesis, deliver better model-based information products, and assist the design, evaluation,and operation of the observing system itself. The proposed priorities are: model coupling, dataassimilation, nearshore processes, cyberinfrastructure and model skill assessment, modelling forobserving system design, evaluation and operation, ensemble prediction, and fast predictors.Approaches are suggested to accomplish substantial progress in a 3–8-year timeframe. Inaddition, the group proposes steps to promote collaboration between research and operationsgroups in Regional Associations, US Federal Agencies, and the international ocean researchcommunity in general that would foster coordination on scientific and technical issues, andstrengthen federal–academic partnerships benefiting IOOS stakeholders and end users.

ARTICLE HISTORYReceived 10 October 2016Accepted 19 April 2017

KEYWORDSCoastal ocean; modelling;forecasting; real-time;operational; dataassimilation;cyberinfrastructure; skillassessment; model coupling;observing system design;GOOS

1. Introduction

The US Integrated Ocean Observing System (IOOS®) is afederal, regional, and private-sector partnership workingto enhance the collection, delivery, use, and prediction ofocean information. The coastal component of IOOS(Coastal IOOS) involves 17 federal agencies and 11Regional Associations (RA) with the RAs having primaryresponsibility for non-federal observations within theirrespective regions, for integrating those assets with thefederal system, and for delivering timely and effectiveproducts to meet regional user needs in the GreatLakes, coastal ocean, and adjacent deep sea of the USEEZ (Price & Rosenfeld 2012).

Real-time observations by Coastal IOOS capture thestate of the ocean at particular locations and times, andlong-term monitoring enables the detection of climatevariability and trends. But measurements alone are not

enough. Numerical modelling allows for interpolation,interpretation, and prediction of the environment, andcombining data with models aids the conversion ofobservations into meaningful information products.Sustained development of modelling capabilities, theapplication of models to enhancing the design and oper-ation of observing systems, and effective data manage-ment and communication are vital components of atruly integrated system.

On basin and global scales, modelling research anddevelopment for IOOS is coordinated through collabora-tive agreements between federal agencies (notablyNOAA and US Navy) and partnerships between federal,academic, and international groups through initiativessuch as the Global Ocean Data Assimilation Experiment(GODAE) OceanView Science Team (GOVST 2014; Bellet al. 2015) and its specialist Task Teams. While RAs

© 2017 Institute of Marine Engineering, Science & Technology

CONTACT John Wilkin jwilkin@rutgers.edu

JOURNAL OF OPERATIONAL OCEANOGRAPHY, 2017https://doi.org/10.1080/1755876X.2017.1322026

already have regional modelling capabilities and areactive in coastal model development, overall coordi-nation of the Coastal IOOS modelling subsystem is lessmature. The call made by the IOOS Modeling andAnalysis Steering Team (Ocean.US 2008) for a highlevel of sustained coordination remains largely unmet.For example, while there has been progress on aspectsof coastal modelling through the IOOS Coastal andOcean Modeling Testbed (COMT), there have been nopan-regional efforts in which groups using differingmethodologies have analysed common data sets andinter-compared model-based coastal ocean state esti-mates using standardised metrics in the way thatGOVST has promoted such efforts for global systems.

The National Ocean Policy Implementation Plan(National Ocean Council 2013) echoed this call forcoordination, and in response the US InteragencyOcean Observation Committee convened a ModelingTask Team (MTT) and workshop in 2014 to proposestrategies and priorities for advancing coastal modellingcapabilities for IOOS. The workshop brought togetherexpertise and community representation from the RAsand federal partners in IOOS, including agencies forwhich applied coastal ocean modelling is vital to advan-cing their capacity to meet mandated responsibilities.

This paper presents the consensus of the MTT on pri-ority areas for coastal modelling research and develop-ment in the next 3–8 years, approaches to acceleratingthe integration of models with the observing and datamanagement subsystems of IOOS, and mechanisms forpromoting research and operational collaboration.

2. Background

2.1. Integration of IOOS subsystems and partnercoordination

IOOS is composed of three major subsystems:observations, modelling, and data management andcommunications (DMAC). Integration of these com-ponents is required to achieve an accurate representationof the ocean state because models without observationsgive at best a virtual representation of the ocean state,while without models the observations provide anincomplete picture due to their inevitable scarcity. Mod-elling provides the predictive capability that is vital tomany user requirements. DMAC infrastructure facili-tates this integration and dissemination of the outputto the user community.

In Coastal IOOS, the subsystems are integrated tovarying degree within each region, but at the nationallevel they operate largely separately. Growing coordi-nation between DMAC and the observing subsystem at

the national level is principally within individual observ-ing technologies, and not yet across technologies in waysthat centralise data access by ocean variable. This com-plicates discovery of data for model assimilation, forcing,and validation, and the implementation of re-locatableand inter-operable modelling systems. Additionally, itdivorces discussions on strategies for observationaldata acquisition, management, archiving, and reportingfrom those for modelling, which impedes the use ofmodels for improving the observing system.

Traditionally, federal agencies were the primaryorganisations implementing US operational models(Federal Backbone (FB) systems), while academic insti-tutions concentrated on process studies and modeldevelopment and experimentation. Now, many non-fed-eral agencies routinely run real-time modelling systems.Though these systems might not meet federal require-ments for operational robustness and reliability, never-theless many user communities find the immediateenvironmental information served by RA models to bevaluable. This may be because the systems are superiorin local skill, or because they offer regional products,higher resolution, or local expertise that are not matchedby FB systems. A need has grown to clarify the roles offederal and non-federal modelling groups, enhance thecommunication among them, and further explore waysto incorporate RA efforts into FB systems.

To better coordinate coastal modelling across the FBand RAs to make modelling research and developmentmore responsive to user requirements, the MTT deliber-ated on procedures that could address common needs,encourage efficiencies, and make two-way connectionsto end users and stakeholders. It was concluded thatthe US coastal modelling community should considerempanelling two consultative and advisory groups tothese ends, possible formats for which are presented inSection 5.

2.2. IOOS coastal modelling objectives

In a synthesis of RA build-out plans for the coming dec-ade, Price and Rosenfeld (2012) noted 27 products orservices desired to meet stakeholder needs in the areasof marine operations; coastal, beach, and nearshorehazards; water quality; ecosystems and fisheries; andlong-term change and decadal variability. Two-thirdsof these products and services required results frommodels. The synthesis identified, therefore, that it wasa core requirement across all regions that modellingcapabilities be developed to deliver analyses and fore-casts, on appropriate time and space scales, for ocean cir-culation, waves, inundation, weather, water quality, andecosystems.

2 J. WILKIN ET AL.

The principal goal of IOOS coastal modelling cantherefore be summarised as enhancing the value ofobservations through model-based synthesis and dataassimilation (DA) to provide robust and reliable past,present, and forecasted ocean conditions to underpinuser products. A second, important goal is to applymodels to observing system design and operation tohelp optimise the observational suite and thereby furtherenhance model-based outputs.

With these goals and requirements in mind, the MTTmembers and workshop participants applied their tech-nical expertise and regional experience to consider howto advance modelling capabilities for Coastal IOOS.The group was steered by a charge to the MTT to con-sider the full spectrum of model uses, emerging model-ling technologies, anticipated technical and scientificchallenges, and how to sustain continuous improvementin model skill and development of new and enhancedmodel-based products.

Guided by this charge, the MTT identified seven topicareas as priorities for concerted community effort inresearch and development over the next 3–8 years.

(1) Model coupling, emphasising improvements toocean state realism through coupling techniquedevelopments applicable to ocean circulation, ice,air, ecosystem, wave, and other components.

(2) Data assimilation (DA), including research anddevelopment on DA methods, and DA-systeminter-comparison frameworks emphasising use ofthe full suite of IOOS observations, including eco-logical data.

(3) Nearshore processes, linking ocean analyses withmodels of surface and groundwater flow, wetlands,estuaries, surf zone dynamics, coastal geomorphol-ogy and sediment transport, discharge and plumedispersion, pathogens, toxins, harmful algae, andbiogeochemistry.

(4) Cyberinfrastructure and model skill assessment,including development of a pan-regional IOOSdata portal built on standardised web services, andcomprehensive tools and benchmarks for interoper-ability, modelling metrics, and skill assessment.

(5) Modelling for observing system design, evaluation,and operation, using observing system simulations,network gap analyses, sensitivity analysis, and pro-totyping the cycle of designing, operating, and eval-uating a coastal observing system.

(6) Ensemble prediction, developing probabilistic pre-diction methods for weather, inundation, naviga-tion, and extreme events, and deliveringquantitative uncertainty estimates for models andproducts.

(7) Fast predictors, using dynamical models and obser-vations to train specialised models for targetedapplications.

All of these topics have emerging communities ofpractice within the field of coastal ocean modelling.The first three will accelerate progress on data assimila-tive and coupled physical–ecological models for estimat-ing ocean state conditions relevant to a variety of oceanusers. Areas 4 and 5 enhance the integration of model-ling with IOOS Observing and DMAC subsystems,while the last two topics address how modelling systemscan be used to analyse uncertainty and explore scenarios.These topics are expanded upon in Section 3.

2.3. Workforce development

It is difficult to find knowledgeable, experienced per-sonnel to fill all the positions available in the US forocean modellers, especially in the realms of modelcoupling and advanced applied DA. In their 10-yearbuild-out plans developed in 2011, the RAs estimatedthat in total, they would need the equivalent of about100 personnel to operate the modelling part of theregional IOOS enterprise (Price & Rosenfeld 2012).This includes operators, forecasters, product develo-pers, and research and development personnel. Thereis an unmet need to develop intellectual capacity inthis area.

Beyond coordinated, targeted research and develop-ment, it is therefore also important that students andearly career scientists be entrained into these efforts toensure the evolution of a skilled workforce that can sus-tain applied coastal modelling in the long term.

3. Scientific developments in coastalmodelling capabilities

The priorities for coastalmodelling research and develop-ment introduced above are not specific to a given model,but have relevance across a variety of models and appli-cations and should facilitate integration of models, obser-vations, and data management. They were chosen by theMTT not to address needs of specific user communities,but to deliver fundamental capabilities that will underpinexpanded, comprehensive use of the full suite of IOOSobservations to realise the objective of an integratedcoastal modelling and observation system.

Under each topic, we present the MTT consensus as aset of recommended actionable tasks that are tractableand, if pursued by the community, would lead to sub-stantive progress on expanding capabilities in theshort-to-medium term.

JOURNAL OF OPERATIONAL OCEANOGRAPHY 3

3.1. Model coupling

Greater dynamical complexity in the coastal ocean’sresponse to forcing can be achieved by directly couplingcomponent models for ocean, atmosphere, and waves,and several RA groups have demonstrated the emergenceof important feedbacks when two-way interactions areincluded (e.g. Olbarrietta et al. 2012). Resolving fasttime scales and short length scales can impact processesin coastal weather prediction (Chambers et al. 2014), andaccurate coupling requires attention to consistency andfrequency of exchange of fluxes of heat, momentum,and mass (e.g. Warner et al. 2010). Beyond ocean–atmosphere dynamics, there are important interactionswith the geosphere (sediment transport, shorelinemigration), biosphere (optically active ecosystem con-stituents; cloud condensation nuclei), and cryosphere(sea-ice and ice shelves).

In the federal agencies, there is some movementtoward standardisation of model coupling architecture,such as the National Unified Operational PredictionCapability (NUOPC) and Earth System Modeling Fra-mework (ESMF), whereas in academia approaches aremore diverse to accommodate active experimentationin coupling complexity.

3.1.1. Recommendations

(1) Experimentation is needed in coupling earth systemcomponent models of groundwater, wetlands, sur-face water hydrology, geomorphology, air-sheds,ecosystems, and biogeochemistry. Efforts shouldinclude human systems that impact water, energy,and ecosystem services.

(2) Limitations of existing toolkits for coupling coastalland–ocean–atmosphere processes should be ident-ified, and capabilities expanded accordingly.Where it does not compromise innovation, RAactivities should anticipate transition to operationsby working with toolkits supported by federal part-ners. Operational centres should make complemen-tary efforts to transfer expertise to academic units,and provide a simulated operational environmentfor research community experimentation. Suchactivities would be suited to the ‘Centre withoutWalls’ (Cw/oW) concept (Section 4), but could com-mence with workshops and personnel exchanges.

(3) Observational and experimental research pro-grammes should be developed that address scientificgaps in model dynamical fidelity highlighted bycoupled models.

(4) Enhanced cyberinfrastructure systems and tools,and added high-performance computing power are

required to allow experimentation with ways inwhich coupled systems can add to the IOOS coastalmodelling enterprise.

3.2. Data assimilation: improving ocean stateestimation through model/data synthesis

Well-configured contemporary coastal ocean modelsnow routinely achieve a useful degree of realism. How-ever, when run for extended time periods they may cap-ture mesoscale variability that is accurate only in astatistical sense, with events and features at the subme-soscale being significantly distorted due to the limits ofpredictability inherent in nonlinear dynamics. Othererrors stem from approximation and parameterisationof the governing equations, numerical discretisation,and insufficient numerical resolution.

While every effort might be taken to increase skill,models will never be error-free. Guided by recognisedsuccesses in Numerical Weather Prediction (NWP),improvements in forecast quality can be achieved utilis-ing DA to optimally combine observations and modelestimates to derive a ‘best estimate’ of the ocean statefrom which to launch a forecast. From the standpointof mathematical and practical implementation, coastalocean DA is challenged by the large problem size (thenumber of model variables to adjust), difficulties project-ing surface observations to the 3D ocean state, strongnonlinearities in the dynamics, the error of represen-tation associated with observed dynamics absent fromthe model, and limited understanding of how modelerrors evolve.

The theoretical underpinnings of DA and the manyapproaches taken in practice from simple nudgingthrough optimal interpolation to the ensemble Kalmanfilter and four-dimensional variational (4DVAR)methods need not be reviewed here. RAs have experi-mented with and contributed to research on manyapproaches, and have implemented pilot real-time DAsystems that have shown skill and found users. Theyhave also highlighted research challenges in severalimportant areas, such as joint assimilation in coupledocean–atmosphere–wave–ice or physical–biogeochem-ical models, how to better use sampling platforms likeautonomous underwater vehicles, and how to retainthe resolution of coastal fronts and jets amid detailedbathymetric and coastline constraints.

3.2.1. Recommendations

(1) Facilitate adoption of all available observations intoprototype DA systems by establishing a unified

4 J. WILKIN ET AL.

IOOS pan-regional data portal offering timely deliv-ery and geospatial search and sub-setting of quality-controlled observations from all platforms in UScoastal waters. Beyond near real-time operation,the service should include a deep archive of pastobservations for multi-year retrospective re-analyses.

(2) Initiate projects that compare differing DA frame-works when presented with a common analysisand prediction challenge or region, a comprehensiveunified data stream, and an agreed set of perform-ance metrics.

(3) Experts from the research community should beplaced into FB development environments to tran-sition progress on new methods and best practices,while also acquainting RA researchers with the con-straints of practical operational environments.

(4) Emphasising coastal ocean environments, collabora-tive research should be encouraged to build newcapacity in ensemble and variational algorithms,observation operators, computational efficiencyand scalability, the incorporation of new data types(e.g. bio-optics), and coupled systems (e.g. ocean–atmosphere–wave–ice).

(5) DA methods should be introduced to water qualityand ecosystem models and models of littoral andnearshore waters, including the assimilation of bio-geochemical observations.

(6) It should be recognised that a substantial user com-munity exists for long retrospective re-analyses ofthe ocean state in support of marine living resourcemanagement and the diagnosis of coastal climatetrends.

3.3. Nearshore processes

Circulation and water elevation in the nearshore zoneimpacts natural and built environments through coastalwater quality, dispersal of pathogens and pollutants,coastline erosion, wetland and estuary ecosystems, andfisheries. Understanding and predicting these processesare important for establishing resilient, sustainablecoastal communities.

Nearshore processes act on a range of time scales,from very short (wave run-up, dune over-topping)through weather time scales (storm surge, river plumes,littoral zone currents), to longer time scales that drivegeomorphological change (coastal erosion, sedimentdeposition), and global sea level rise and human-inducedchanges in the watershed. Biogeochemical and waterquality models depend upon skilful hydrodynamicmodels to determine physical transport and mixingacross all these scales in order to simulate

eutrophication, hypoxia, algal blooms, pathogens, toxins,and sediments. But water quality models themselves alsoneed development. Eutrophication models that simulatenutrients, biomass, and oxygen in the water column andbenthos may have dozens of empirical coefficients.Models of phytoplankton community dynamics,microbial pathogens, or community level responses totoxins may entail an even higher level of parameterisa-tion; rigorous calibration or even identification of domi-nant processes and sources of error is difficult. Aquaticecosystem models must also consider stresses that arisefrom the adjacent land and air, and should not stop atsome chemical endpoint but extend through flora andfauna to ecosystem services; thus coupled model devel-opments are key to progress in this area.

Models for predicting coastal hazards have typicallyevolved from hydrodynamic models with features andcapabilities added as required to capture key processes.Adding further sophistication will further expand com-putational demands and possibly render high fidelitymodels prohibitive for many applications. ‘Fast predic-tor’ models that are trained using data and/or complexmodels may be more amenable to computing probabilis-tic products for extreme events and exploring environ-mental scenarios.

3.3.1. Recommendations

(1) Nearshore water quality model development shouldconsider multiple stressors, interaction with coastalflora and fauna, and ecosystem services. Testing andevaluation in multiple regional settings should beaimed at progress toward robust and portable models.

(2) A pan-regional or national effort is required to coor-dinate the production of consistent physical and bio-geochemical ocean boundary conditions for regionalcoastal models.

(3) Circulation model enhancements are required forwave transformations and overland wave and waterpropagation (e.g. wave nonlinearities, growth anddecay, swash, rip currents, and representationof reefs).

(4) Improvements are required in modelling the trans-port of non-cohesive and mixed grain sedimentsfrom offshore bars to dunes, bluffs, and cliffs,including erosion and recovery; and in modellinglong-term morphological change of beaches, barrierislands, marshes, and estuary shorelines as landcover changes and sea level rises.

(5) Observing system simulation experiments (OSSEs)should be pursued to determine which biogeochem-ical and sediment observations, and observationstrategies, are more effective for constrainingmodel skill in nearshore processes.

JOURNAL OF OPERATIONAL OCEANOGRAPHY 5

3.4. Cyberinfrastructure and model skillassessment

Coastal IOOS requires cyberinfrastructure standards,services, and tools that enhance discovery and utilisationof observations and modelling system outputs. Evolvingcommunity metadata conventions and web services fordata access are complementing the development of stan-dardised tools that enhance the efficiency of scientificanalysis and the development of model-based products.But there is still significant work to be done in improvingthe scope and robustness of these tools, and training anddocumentation is needed to encourage and facilitatetheir widespread adoption.

To support their local stakeholders, RAs have devel-oped portals that serve their own models, observations,and regional satellite data subsets, but the portals differamong the 11 RAs, making it difficult to aggregate col-lections of unique data for larger regions. A centralisedcatalog and catalog services providing access to allobservations acquired by RAs and other IOOS and glo-bal observing systems on the continental shelf and inadjacent deep waters would enable greater communityengagement in coastal model skill assessment andwould provide a foundation for inter-comparisonstudies of DA models and observing system exper-iments (OSEs).

Metrics that characterise model performance pro-vide information on model strengths and weaknessesto spur research and development to improve modelskill and robustness. Such metrics are routine in theNWP community, and GODAE has formulatedmetrics for mesoscale forecast systems (Hernandezet al. 2009) that offer a useful starting framework forappraising coastal models.

The short time frame for which some model outputsare retained on data servers, their partial coverage inspace and time (e.g. serving only surface or daily averageconditions), and the provision of analyses but not the fullset of forecasts, all limit community efforts at modelinter-comparison and skill assessment.

Research to Operations transitions could be acceler-ated if non-federal researchers had access to an exper-imental operational environment – a ‘computationalsandbox’ – that mimicked data streams within FBcentres. Researchers could then evaluate how a prototypesystem performs in a setting that simulates the actualconstraints on data availability (latency and quality con-trol) in an operational centre, and experiment with theimpact of changes in dynamical parameterisations,algorithms, or configuration for open-source codesused in operation.

3.4.1. Recommendations

(1) Create documentation describing best practices formanaging model data using dynamic documentsthat are updated regularly and invite communityinput. Communicate these capabilities throughworkshops and training materials.

(2) Expand the development of standardised tools andlower-level utilities for popular scientific analysissoftware and communicate these capabilitiesthrough workshops and training materials.

(3) Establish a pan-regional data portal that aggregatescoastal ocean data from all RAs and national IOOSsystems, deposits metadata in a geospatial database,allows standardised queries of temporal and spatialextents, keywords, and variable names, and deliversdata seamlessly for both interactive scientific analy-sis and automated computing environments.

(4) Create a parallel testing environment (compu-tational sandbox) that enables researcher access todata streams and model configurations that simulatethose within FB operational centres.

(5) Engage the coastal modelling community in devel-oping a set of model skill metrics. Initiate the routinegeneration and reporting of these metrics across allCoastal IOOS modelling systems.

3.5. Modelling for observing system design,evaluation and operation

Though principally a user of IOOS observations, coastalmodelling also has complementary roles to play instrengthening the observing system itself. These includedemonstrating how observations add skill to model-based analyses and forecasts, and contributing to thedesign and efficient operation of observing systems.

Sampling density and accuracy directly impact dataassimilative analyses, so DA systems can be used toquantitatively appraise the information content of obser-vation networks. OSEs that selectively with-hold obser-vations can examine the sensitivity of forecast skill toobservation types or platforms and the density or fre-quency of observations. OSSEs that sample model outputto construct sets of hypothetical observations can be usedto identify gaps in a network, reveal vulnerabilities tooperational failures of observing elements, evaluate thepotential of instruments that do not yet exist (e.g. newsatellites), and to examine how analysis skill changeswith quality control standards. Array mode analyses(e.g. Bennett 1990; Le Hénaff 2009) are examples ofmodel-based approaches to identifying patterns of

6 J. WILKIN ET AL.

ocean variability that are not constrained by an observa-tional network and whose predictability might improvewith acquisition of new observations.

An extension of these methods is adaptive control ofobserving platforms such as autonomous underwater gli-ders. Model-based systems can suggest where re-locata-ble assets might most profitably be sent to acquireindependent data about under-observed regions. More-over, predictions of the ocean current regime in whichautonomous vehicles operate offer insight on a vehicle’s‘reachability area’ (Wang et al. 2013; Garau et al. 2014)given vehicle speed and power characteristics.

3.5.1. Recommendation

(1) DA modelling groups within the RAs should under-take OSE and observation impact analyses of theirregional observing systems to build a pan-regional,multi-model view of observing system strengthsand vulnerabilities.

(2) IOOS gap analyses of the adequacy of observing net-work density (e.g. the number and location of HF-radar sites; the frequency and location of repeat gli-der transects) should include model-based OSE,OSSE, and sensitivity analyses.

(3) A regional demonstration project using many moreobserving assets than is presently typical should testwhether quantitatively justified array designs can inpractice perform better than ad hoc ‘expert’ observ-ing strategies with respect to agreed skill metrics rel-evant to specific user requirements; and with a verydense data set test how well DA systems quantifiedtheir actual forecast skill, uncertainty, and obser-vation impact.

(4) A pilot project should test algorithms for glider pathplanning that integrate environmental awarenessfrom modelling systems. The NSF Ocean Observa-tories Initiative (OOI) Coastal Arrays are well posi-tioned to capitalise on these capabilities of IOOScoastal modelling.

(5) The coastal modelling community should assessfuture observing systems (e.g. swath altimetry,high-resolution geostationary satellite SST, and col-our) to gauge the value they could add to existingnetworks and their capacity to supplant existingtechnologies with superior capabilities.

3.6. Ensemble prediction

Small perturbations in the initial state, forcing or modelparameters lead to divergence of forecast end states,which limits the duration over which a single forecast

has useful predictive skill. Within an individual modelframework, ensembles are widely used to quantify thisspread in forecast trajectories. Multi-model ensemblemethods offer the added possibility to reduce forecasterrors that stem from errors within individual modelsdue to algorithmic and parameterisation choices, misre-presentation of dynamics, and other systematic factors.

IOOS partners and international colleagues operatenumerous regional models and basin or global modelsthat cover US coastal waters. Multiple models using dif-fering approaches operating in common geographicalareas provide the fundamental capacity for combiningmodel outputs as ensembles. The promise of ensemblemethods is that they provide ocean state estimates withlower expected error than any single dynamical forecast,while a challenge is selecting ensemble sets that effi-ciently and effectively capture forecast error statistics.

3.6.1. Recommendations

(1) Coastalmodellers shoulddevelopand test systems thatperturb their model forecasts in order to characteriseand quantify forecast spread, and establish a quantitat-ive basis for subsequent multi-model mergers.

(2) The coastal ocean modelling community shouldprototype a consensus forecast system based on amulti-model ensemble approach for a pilot regioncovered by several models and for which a densedata set exists.

(3) A working group or workshop should be convenedto foster community multi-model ensemble effortsby setting conventions for participation, addressingappropriate metrics for coastal model weighting,verification, and validation, and developing the pres-entation of probabilistic forecast information tostakeholders.

3.7. Fast predictors

The computational expense of high fidelity, high-resol-ution simulations of circulation and other coastal pro-cesses (sediments, biogeochemistry, ecosystems) areoften prohibitive for probabilistic ‘Monte Carlo’methodsin which a large number of long simulations are used tosample the model error probability space. This maydemand that lower resolution and lower fidelity modelsbe employed. Alternatively, fast and robust surrogatemodelling systems (e.g. van der Merwe et al. 2007; Tafla-nidis et al. 2013) offering adequate accuracy andenhanced computational efficiency can be developedbased on a database of high fidelity simulations or obser-vations. Surrogate models allow both deterministic and

JOURNAL OF OPERATIONAL OCEANOGRAPHY 7

probabilistic simulations with short turnaround times,and can be used in support of DA and network optimis-ation (e.g. Frolov et al. 2009).

3.7.1. Recommendations

(1) Coastal modelling groups should create and skill-assess decade-scale, high-resolution simulationdatabases of circulation, sediment transport, biogeo-chemistry, and other key processes for training fastpredictor modelling systems.

(2) Encourage development of coastal ‘fast predictor’systems with a view to deploying these for physicaland environmental stakeholder needs, and forobserving system design and operation, networkoptimisation, analysis of return periods for hazards,and integration into DA systems.

(3) Initiate a test-bed to coordinate surrogate modeldevelopment and application, and to undertake ret-rospective analyses of well-observed events to evalu-ate surrogate versus traditional forecasting methods.

4. Coordinating and sustaining a coastalmodelling strategy

Building a nationally coordinated coastal ocean analysisand prediction system that is responsive to user require-ments, exploits the best numerical codes and algorithms,and utilises the full spectrum of in situ and remotelysensed data requires a level of regional–federal partner-ship that has hitherto been absent in the US coastal mod-elling community. Accordingly, we recommend that thecommunity consider empaneling consultative and advi-sory groups to help shape a more coordinated nationalcollaboration.

The members of these groups would be knowledge-able of existing and emerging capabilities and userrequirements, and would be charged with advising onthe division and sharing of effort between the FB andthe RAs that would enable mutually beneficial partner-ships. Other key activities would be ensuring thatDMAC yields the necessary data access and interoper-ability of cyberinfrastructure elements to aid the partner-ship, and communicating to FB and RA modellinggroups the needs or model-based investigations ongaps, vulnerabilities, and efficiencies in observing assetdeployment. The MTT identified requirements for fos-tering interchange on scientific and technical experience,and strengthening federal–academic partnerships toencourage efficiencies and connections to end usersand stakeholders.

4.1. Recommendations

(1) Form a caucus comprised principally of modellersand model users from the RAs, with added involve-ment from federal counterparts in much the sameway that the GODAE OceanView communitymelds federal and academic participation in researchfor global and basin-scale modelling. The caucuscould foster interchange of research and develop-ment experience and needs within the US coastalocean research community through events such asfocused workshops, training events, test-beds, andthemed publication collections. Galvanised bythese efforts to promote coordination of coastalmodelling capacity growth in the short term, it willbe collaborative programmes and teams that evolvein the longer term that ultimately enable IOOS todeliver coastal ocean model-based products andinformation that meet user needs on a sustainedbasis.

(2) Form a Task Team to further prioritise and guideinitial action on the recommendations in thispaper, and take steps to establish collaborativeenvironments for Coastal IOOS modelling. Thiswould include facilitating RA exposure to FB oper-ational environments, establishing channels bywhich federal research and development needs,and user requirements, are communicated to theresearch community. In the longer term, a mechan-ism such as a community Steering Team that sus-tains coordination of FB and RA modellingactivities may be in order to keep pace with evolvingresearch priorities, an expanding observing system,and increasingly sophisticated downstreamapplications that use data-informed modellinginfrastructure.

Greater coordination and communication will ensurethat federal agencies reap the benefits that IOOS observ-ing and modelling can bring to their respective responsi-bilities for scientifically informed stewardship of thenation’s marine environment; and also that the fullsuite of expertise resident in the RAs and academic part-ners is brought to bear on delivering the technical andscientific solutions that agencies require. It should benoted that enhanced academic and operational coordi-nation, and accelerated research and development, willnot be accomplished by regional and coastal US IOOSalone, but will include relationships to internationalpartners engaged in global and basin scales modellingand analysis.

8 J. WILKIN ET AL.

5. Implementation: accomplishing progresson research priorities

The actions suggested in Section 3 are varied and wouldrequire quite different approaches to implement. Here,we present an overview of existing US programmesand organisational structures that have supported coastalmodelling in the past and comprise instruments thatcould help enact many of the recommendations.

A subset of the recommendations call for establishingclosely collaborating communities of practice formed ofnon-traditional groupings of ocean science professionals,and these do not match well to existing supporting fra-meworks for research and development in the USAccordingly, we advocate a novel ‘Cw/oW’ concept toprovide a home for these collaborative endeavours.

The remarks below are not intended to be exhaustive,but merely illustrative of the capacity of existing fundingavenues to support innovation and experimentation incoastal ocean modelling.

5.1. Coastal and Ocean Modeling Testbed

The mission of the IOOS COMT is to accelerate the tran-sition of advances from the research community to oper-ational ocean products and services. COMT teams (offederal, academic, and private industry members) haveaddressed projects related to coastal inundation, estuar-ine and shelf hypoxia, and contributed to creating cross-cutting cyberinfrastructure of benefit to the IOOS datainfrastructure in general.

COMT tasks have a defined beginning and end, anddeliverables. As such, it is a useful mechanism for assem-bling teams to tackle recommendations above on estab-lishing a DA inter-comparison test-bed, and supportingenhanced software tools and a pan-regional data portal.The call for pilot projects to explore model-based analy-sis of observing systems falls squarely in the COMT bai-liwick. The coordinating groups advocated in Section 4could provide valuable input suggesting COMTpriorities.

5.2. Federal funding opportunities

5.2.1. National Ocean Partnership ProgramThe National Ocean Partnership Program (NOPP) is acollaboration of federal agencies that support oceano-graphic research and technology, resource management,education, and outreach. A relevant example of previousNOPP sponsorship is the substantial effort (more than50 investigators) that demonstrated performance andapplication of the HYCOM model for eddy-resolving,real-time ocean prediction (Chassignet et al. 2009).

That project, which received the 2007 NOPP Excellencein Partnering Award, has since transitioned to oper-ations and is widely used as boundary conditions toRA real-time coastal models.

Recommendations in Section 3 on model coupling,ensemble prediction, and littoral and nearshore environ-ments outline activities where NOPP partnerships thatcross multiple federal agencies and bring together exist-ing and emerging capabilities could drive significantprogress.

5.2.2. National Science FoundationThere are topics in Section 3 that would meet NSF’s cri-teria for innovation and relevance, and progress could beachieved by individuals or by teams submitting NSF‘Collaborative Research’ proposals.

It is also within NSF’s mandate to encourage researchtargeted at specific national needs and community inter-ests. NSF formed Climate Process and Modeling Teamsto ‘speed development of global coupled climate models… by bringing together theoreticians, field observers,process modelers and the large modeling centres to con-centrate on the scientific problems facing climate modelstoday’. To foster collaboration exploring, for example,connections at the interface of wetlands, estuaries, thenearshore zone, and coastal ocean, NSF could establish‘Coastal and Nearshore Process and Modeling Teams’.

NSF’s investment in the OOI Coastal Endurance andCoastal Pioneer arrays provides opportunities to put intopractice efforts at inter-comparison of DA methods andobserving system assessment, gap analysis, and exper-iments with optimisation of operations.

NSF Science and Technology Centers (STCs) useteam science to address ‘grand challenges’, and to cata-lyse technology transfer, workforce development, andbroaden participation. NSF might call for an STC tofocus on one or more of the research categories wehave highlighted, while also contributing to neededworkforce training.

5.2.3. NASASatellites are a growing component of IOOS coastalobserving with the spectre of significant enhancementsin the advent of swath altimetry (SWOT), geostationarycoastal imaging (GEO-CAPE), and new SAR and hyper-spectral imaging technologies from other internationalagencies. At the ocean mesoscale, NASA project scien-tists have amply demonstrated the synergy betweenremote sensing, in situ ocean observation by Argo profil-ing floats, and advanced data assimilative modelling.With NASA encouragement, the coastal oceanographycommunity could make comparable advances in the syn-thesis of coastally focused satellite observations in

JOURNAL OF OPERATIONAL OCEANOGRAPHY 9

conjunction with IOOS in situ observations, and in sodoing make a sizeable reciprocal contribution toNASA’s missions.

Prior to launch, satellite mission design has long uti-lised rigorous methods for quantifying requirements forinstrument precision, orbital sampling patterns, errorbudgets, and resolution. During mission operationssimulation and modelling play a key role in adaptingto operational contingencies and instrument perform-ance. Bringing NASA expertise to bear on evaluating,enhancing, and operating IOOS coastal observatoriesthrough collaborative projects (e.g. Wang et al. 2013)would further the synergistic use of satellite and in situdata.

5.3. Core IOOS funding

RAs have differing levels of involvement in numericalmodelling. Some create products using results frommodels run by other organisations, while others config-ure and run models for their region and producemodel-derived products (Price & Rosenfeld 2012). RAsmight use IOOS funding to develop new, or expandexisting, model capabilities, but in few instances is it suf-ficient to sustain robust real-time operations or to bring amodelling system to full maturity for transition fromresearch to another entity that will operate it.

The RAs coordinate various elements of regionalobserving systems and play a key role in deliveringobservations to the models. They also play a part indirecting coastal ocean model development by helpingidentify user needs that would benefit from productsand services that incorporate model output, and mayhelp to design and distribute such products. Supportingongoing improvements in modelling systems for stake-holder information products needs to be an IOOS fund-ing priority in concert with sustaining the observatoriesthemselves.

RAs act largely independently in constructing andoperating portals and web services to deliver data andmodel-derived products. However, as we have notedalready, there is an increasing need for pan-regionalinter-operable services to access data and models. RAscould also be making greater use of models for observingsystem design. One of the community entities suggestedin Section 4 could spur IOOS to encourage greatercoordination and collaboration in these respects.

5.4. NOAA cooperative institutes

Through its Cooperative Institute framework NOAA cansupport non-federal organisations with outstandingresearch programmes in areas relevant to NOAA long-

term goals. A Cooperative Institute for applied coastalocean modelling collocated with a NOAA research oroperational laboratory could create a strong, long-termcollaboration between academic researchers and NOAAgroups at the forefront of operational implementationof coastal products. Experts from the RA research com-munity could contribute directly to the transition ofdevelopments and practices to operations, and theenvironmentwould also enable RA researchers to becomemore aware of practical operational constraints and emer-ging user requirements. Many cooperative agreementsbetween NOAA and academic partners provide for for-mal sponsorship of students through fellowships, andthus Cooperative Institutes would also help educate andtrain the next generation the nation’s scientific workforce.

5.5. ‘Centre without walls’

Some of the research priorities in Section 3 requirecollaborative activity on the part of non-traditionalgroupings of ocean and information science pro-fessionals. For example, creating a new generation offlexible and computationally more efficient models,and advanced earth system model coupling, are topicswhere experts with different skills need to work inclose collaboration and in conjunction with significantsupporting cyberinfrastructure.

The MTT expressed concern that some such group-ings may not align well with existing mechanisms forsupporting US research and development, and suggesteda new construct – a ‘Cw/oW’ – as a framework to fosterclose collaboration across a breadth of skills. The envi-sioned centre would bring together diverse expertise tomake rapid and significant progress on targeted projects,yet also provide a home for a professional core to sustainongoing development of tools and best practices forworking with ocean models and observational data.The centre would facilitate synergies with RA modellingwhere appropriate, but would not be focused on particu-lar coastal geographic regimes.

The centre could be virtual – hence, the ‘without walls’moniker –withmodest anchoring facilities at a universityor federal laboratory, though a physical home proximateto an oceanographic operational centre also has merit.Either way, the Cw/oW would provide infrastructureand protocols that enable experimentation within a vir-tual operational environment – what might be called a‘computational sandbox’ – to accelerate Research toOperations transitions. This would echo the successfulEuropean Centre for Medium-range Weather Forecast-ing (ECMWF) Fellowship Program by encouragingcoastal modelling researchers from universities andother agencies to spend extended periods of time working

10 J. WILKIN ET AL.

on problems directly related to improving operationalmodelling within federal agencies.

By formalising such a centre, infrastructure could bemade available to conduct training workshops anddevelop comprehensive cyberinfrastructure tools with adedicated technical workforce. Such an effort might rep-resent a maturing of software development efforts pio-neered under COMT. The centre would need to befunded primarily by new resources.

6. Summary and actions

The strategies and recommendations presented here seekto advance the coastal modelling subsystem of IOOS, andcoastal GOOS, through targeted research innovation andby establishing better links between federal and non-fed-eral modellers to communicate needs and developments.

The priority areas (model coupling, DA, nearshoreprocesses, cyberinfrastructure and model skill assess-ment, modelling in support of observing systems, ensem-ble prediction, and fast predictors) are aimed atdeveloping the capabilities necessary to make full useof the observational assets of IOOS through advancedDA, to use models to inform and improve the observa-tory, and to enhance the fidelity, scope and utility ofmodels to underpin the creation of model-derived pro-ducts that meet the needs of IOOS stakeholders.

To improve coordination between federal and non-federal modelling groups, and among the respectivemodelling, observing and data management subsystemsof IOOS, it is suggested that two groups be empanelled:(i) a caucus comprised of model developers would be aforum for interchange of research and developmentexperience that is responsive to needs of the US coastaloceanography and ocean modelling community, andthat would sustain the development cycle in the longerterm; and (ii) a Task Team that would guide initialimplementation of the actions this article describes andtake steps to facilitate collaborative environments condu-cive to coordinating federal efforts with activities in theIOOS regions, and globally.

Benefits that would flow from these initiatives are effi-ciencies through the coordination of efforts that addresscommon needs, demonstration of the value and skill ofintegrated coastal ocean modelling through robust vali-dation and assessment processes, and contributions tothe development of a workforce that can capitalise onthe nation’s investment in coastal observing.

Acknowledgements

We thank Eric Lindstrom, David Legler, and Bob Houtman ofthe US IOOC for initiating and supporting the activities of the

Modeling Task Team, and Nicholas Rome, Hannah Dean, andKruti Desai at the Consortium for Ocean Leadership for logis-tical and editorial assistance throughout the many meetingsand teleconferences. We also thank JC Feyen, JC Lehrter, EPMyers, HL Tolman, and H Townsend for their participationin the MTT deliberations.

Disclosure statement

No potential conflict of interest was reported by the authors.

ORCID

John Wilkin http://orcid.org/0000-0002-5444-9466Leslie Rosenfeld http://orcid.org/0000-0002-0768-819XArthur Allen http://orcid.org/0000-0002-6061-9396Rebecca Baltes http://orcid.org/0000-0003-3121-1495Antonio Baptista http://orcid.org/0000-0002-7641-5937Ruoying He http://orcid.org/0000-0001-6158-2292Patrick Hogan http://orcid.org/0000-0001-5931-3675David Schwab http://orcid.org/0000-0001-5138-5682Richard Signell http://orcid.org/0000-0003-0682-9613

References

Bell MJ, Schiller A, Le Traon PY, Smith NR, Dombrowsky E,Wilmer-Becker K. 2015. An introduction to GODAEOceanView. J Oper Oceanogr. 8(suppl 1):s2–s11. doi:10.1080/1755876X.2015.1022041.

Bennett AF. 1990. Inverse methods for assessing ship-of-opportunity networks and estimating circulation andwinds from tropical expendable bathythermograph data. JGeophys Res Oceans. 95:16111–16148.

Chambers CR, Brassington GB, Simmonds I, Walsh K. 2014.Precipitation changes due to the introduction of eddy-resolved sea surface temperatures into simulations of the“Pasha Bulker” Australian east coast low of June 2007.Meteorol Atmos Phys. 125(1–2):1–15.

Chassignet E, Hurlburt H, Metzger E, Smedstad OM,Cummings J, Halliwell G, Bleck R, Baraille R, Wallcraft A,Lozano C, et al. 2009. US GODAE: Global OceanPrediction with the HYbrid Coordinate Ocean Model(HYCOM). Oceanography. 22(2):64–75.

Frolov S, Baptista AM, Leen TK, Lu Z, van der Merwe R. 2009.Fast data assimilation using a nonlinear Kalman filter and amodel surrogate: An application to the Columbia Riverestuary. Dynam Atmos Oceans. 48(1):16–45.

Garau B, Bonet M, Alvarez A, Ruiz S, Pascual A. 2014. Pathplanning for autonomous underwater vehicles in realisticoceanic current fields: Application to gliders in the westernMediterranean Sea. J Marit Res. 6(2):5–22.

GOVST. 2014. GODAE Ocean View Strategic Plan 2015–2020, Report No. 2. GODAE International ProgramOffice. UK Met Office. Exeter, UK. 26 pp. + Annex I-III.

Hernandez F, Bertino L, Brassington G, Chassignet E,Cummings J, Davidson F, Drévillon M, Garric G,Kamachi M, Lellouche J-M, et al. 2009. Validation andintercomparison studies within GODAE. Oceanography.22:128–143.

JOURNAL OF OPERATIONAL OCEANOGRAPHY 11

Le Hénaff M, De Mey P, Marsaleix P. 2009. Assessment ofobservational networks with the Representer MatrixSpectra method—application to a 3D coastal model of theBay of Biscay. Ocean Dynam. 59(1):3–20.

NationalOceanCouncil. 2013. National ocean policy implemen-tation plan. Available from: https://obamawhitehouse.archives.gov/sites/default/files/national_ocean_policy_implementation_plan.pdf (last visited April 2017).

Ocean.US. 2008. The Integrated Ocean Observing System(IOOS) modeling and analysis workshop report, ArlingtonVA, June 22–28, 2008. Ocean.U.S. Publication No. 18,21 pp. Available from: http://www.iooc.us/wp-content/uploads/2010/12/18.pdf [last visited April 2017].

Olabarrieta M, Warner JC, Armstrong B, Zambon JB, He R.2012. Ocean–atmosphere dynamics during Hurricane Idaand Nor’Ida: An application of the coupled ocean–atmos-phere–wave–sediment transport (COAWST) modeling sys-tem. Ocean Model. 43–44:112–137.

Price H, Rosenfeld L. 2012. Synthesis of regional IOOS build-out plans for the next decade. IOOS Association, 59 pp.

Available from: http://www.ioosassociation.org/sites/nfra/files/documents/ioos_documents/regional/BOP%20Synthesis%20Final.pdf (last visited April 2017).

Taflanidis AA, Kennedy A, Westerink J, Smith J, Cheung K,Hope M, Tanaka S. 2013. Rapid assessment of wave andsurge risk during landfalling hurricanes: probabilisticapproach. J Waterway, Port, Coastal, Ocean Eng. 139(3):171–182.

van der Merwe R, Leen TK, Lu Z, Frolov S, Baptista AM. 2007.Fast neural network surrogates for very high dimensionalphysics-based models in computational oceanography.Neural Net. 20(4):462–478.

Wang X, Chao Y, Thompson DR, Chien SA, Farrara J, Li P, VuQ, Zhang H, Levin JC, Gangopadhyay A. 2013. Multi-modelensemble forecasting and glider path planning in the Mid-Atlantic Bight. Cont Shelf Res. 63:S223–S234.

Warner J, Armstrong B, He R, Zambon J. 2010. Developmentof a Coupled Ocean–Atmosphere–Wave–SedimentTransport (COAWST) modeling system. Ocean Model. 35(3):230–244.

12 J. WILKIN ET AL.