THE SEA: THE SCIENCE OF OCEAN PREDICTIONJournal of Marine Research, 75, 877–921, 2017Coupled ocean-atmosphere forecastingat short and medium time scalesby Julie Pullen1,2, Richard Allard3, Hyodae Seo4, Arthur J. Miller5, Shuyi Chen6,Luciano Ponzi Pezzi7, Travis Smith3, Philip Chu8, José Alves9, and Rui Caldeira9ABSTRACTRecent technological advances over the past few decades have enabled the development of fullycoupled atmosphere-ocean modeling prediction systems that are used today to support short-term(days to weeks) and medium-term (10–21 days) needs for both the operational and research com-munities. We overview the coupling framework, including model components and grid resolutionconsiderations, as well as the coupling physics by examining heat fluxes between atmosphere andocean, momentum transfer, and freshwater fluxes. These modeling systems can be run as fully cou-pled atmosphere-ocean and atmosphere-ocean-wave configurations. Examples of several modelingsystems applied to complex coastal regions including Madeira Island, Adriatic Sea, Coastal Califor-nia, Gulf of Mexico, Brazil, and the Maritime Continent are presented. In many of these studies, avariety of field campaigns have contributed to a better understanding of the underlying physics associ-ated with the atmosphere-ocean feedbacks. Examples of improvements in predictive skill when run incoupled mode versus standalone are shown. Coupled model challenges such as model initialization,data assimilation, and earth system prediction are discussed.Keywords: Coupled air-sea modeling1. IntroductionCoupled air-sea forecasting systems have been planned and architected over severaldecades (Hodur 1997; Hodur et al. 2002; Bender and Ginis 2000; Bender et al. 2007).Whereas global coupled models typically have resolutions of approximately 1◦, regional1. Stevens Institute of Technology, Hoboken, New Jersey 07030, USA2. Corresponding author: e-mail: julie.pullen@stevens.edu3. Naval Research Laboratory, Stennis Space Center, Mississippi 39529, USA4. Woods Hole Oceanographic Institution, Woods Hole, Massachusetts 02543, USA5. Scripps Institution of Oceanography, La Jolla, California 92037, USA6. University of Washington Seattle, WA 981957. National Institute for Space Research, São José dos Campos, São Paulo 12227-010, Brazil8. NOAA Great Lakes Environmental Research Laboratory, Ann Arbor, Michigan 48108, USA9. Oceanic Observatory of Madeira, Agência Regional para o Desenvolvimento da Investigacão Tecnologia eInovacão, Funchal, Portugal, and Instituto Dom Luiz, University of Lisbon, Portugal© 2017 Julie Pullen, Richard Allard, Hyodae Seo, Arthur J. Miller, Shuyi Chen, Luciano Ponzi Pezzi,Travis Smith, Philip Chu, José Alves, and Rui Caldeira.877878 The Sea: The Science of Ocean Prediction [75, 6coupled models enable high resolution (10 km or less) for local areas. The accumulationof studies employing coupled air-sea models and the opportunity to synthesize their resultshas encouraged the evolution of coupled modeling systems and their application to manyareas around the globe. Coupled systems have been shown to yield benefits across multi-ple forecast horizons: from short (multiple days) to medium (multiple weeks) timescales.Significant factors and examples in the path of development and application of regionalcoupled models are described in this contribution.A growing recognition of the benefits to both realms of air-sea coupled prediction hascome with the expanded demonstration by observations that oceanic fronts can modify theoverlying atmosphere (Chelton et al. 2004; Pezzi et al. 2004; Small et al. 2008; Cheltonand Xie 2010). Coupled models have demonstrated skill in capturing key aspects of coastalfrontal dynamics (Perlin et al. 2007; Seo et al. 2008b, 2016) in hurricane prediction (Chenet al. 2007; Seo and Xie 2013) and across numerous regional applications. The positiveimpacts of coupling can extend through the entire atmospheric boundary layer, includingchanges to cloud dynamics (Fallmann et al. 2017). In this Introduction we utilize twothemes to introduce topics that will be explored in the remainder of this overview paper.These themes are “research to operations” and “multimodel approaches.”The advent of two-way coupled ocean-atmosphere regional modeling has enabled newcapabilities in high-resolution prediction. The U.S. Navy’s Coupled Ocean/AtmosphereMesoscale Prediction System (COAMPS) has been employed and recognized for its skillin representing complex coastal regions and for its utility in guiding field campaigns (Doyleet al. 2009; Chen et al. 2010; Pullen et al. 2011; Shinoda et al. 2013). Among its strengthsare its origin at the interface of oceanography and marine meteorology research groups,its ease in relocating to new domains (Hodur et al. 2002), and its operational deploymentthroughout the world.Open-source alternatives exist in the form of coupled Weather Research and ForecastingModel (WRF) and Regional Ocean Modeling System (ROMS) frameworks, such as theCoupled-Ocean-Atmosphere-Wave-Sediment-Transport System (COAWST) (Warner et al.2010). COAWST has been applied to Adriatic cold air bora events (Ricchi et al. 2016), firststudied with two-way coupled COAMPS by Pullen, Doyle, and Signell (2006). We give anupdate on this Mediterranean domain in the context of “research to operations.” Then weprobe the results of these two models, COAMPS and COAWST, applied contemporaneouslyto the same island domain (Section 1b).a. From research to operations: an evolutionAccelerated by a series of international field and modeling campaigns over the pastdecade, coupled air-sea modeling systems are transitioning from research to operations. Forexample, in early prototype two-way coupled studies for the Adriatic Circulation Experi-ment (ACE), Pullen, Doyle, and Signell (2006) and Pullen et al. (2007) examined downslopewindstorms, or bora, that occur in the topographic mountains of the Dinaric Alps of Croatia2017] Pullen et al.: Coupled ocean-atmosphere forecasting 879during late fall and winter months. The circulation patterns in the northern Adriatic Seaare influenced by the bora wind jet as the sea transfers heat to the colder overlying air. Inthat COAMPS study, the heat flux root-mean-square error (RMSE) was reduced by 30%on average at four over-water instrumentation sites in the northern Adriatic using two-waycoupling. Although the original study utilized coupling at a 6-hour coupling interval, recentoperationally-oriented Earth System Modeling Framework (ESMF)-based simulations witha coupling interval on the order of minutes (Allard et al. 2010) for that same region, timeperiod, and locations found a total heat flux RMSE reduction of 44% when comparing thecoupled versus uncoupled runs.In both studies, the coupled model clearly outperforms the uncoupled model, a result thathelped motivate the operational deployment of coupled systems. The two studies furthersuggest that the coupling frequency influences the dynamics of coupled model simulations—a topic that we will revisit further. But first we probe coupled modeling approaches for acommon island setting.b. A multimodel approach: islands as a coupled model challengeIslands represent a challenging and important setting for coupled modeling systems(Caldeira et al. 2016). Ocean features initiated by islands (such as mesoscale eddies, fila-ments, and warm wakes) generate distinct sea surface temperature (SST) signatures (Sangràet al. 2007). The strong near-surface wind and ocean surface heat flux gradients occurringin the flanks and lee of mountain terrain are loci of vigorous air-sea interaction (Grubišic´,Sachsperger, and Caldeira 2015). The dynamics create an island-modulated circulation thatis, fundamentally, a two-way coupled environment, and is therefore a key setting for realisticcoupled models to emulate.Of particular interest in this regard are local impacts generated by submesoscale andmesoscale SST variations that can alter heat fluxes and winds. Warm SSTs are commonlyfound in the lee of islands (Caldeira et al. 2002). Caldeira and Tomé (2013) speculated thatlee-side ocean warming results from the enhanced solar heating of cloud-free atmosphericwakes. Caldeira et al. (2002) and Caldeira (pers. comm.) measured this effect on the island ofMadeira to be 2 to 3◦C through in the upper 10 to 20 m. During summertime, the Madeirawarm wake extends for over 100 km. Similar characteristics of warm wakes have beenmeasured in the lee of the Canary Islands (Barton et al. 2000). And wake regions can haveeven longer-ranging and more-pronounced effects (>300 km) (Smith and Grubišic 1993;Hafner and Xie 2003).A recent deployment of multiple coupled models, at high resolution, to examine a com-mon island domain yields insight on model performance in this challenging geographicalsetting. The COAMPS model was configured for the island of Madeira (Figure 1) usinga triply-nested domain from 18/6/2-km resolution (Pullen et al. 2017). The ocean modelis activated on one domain at 2-km resolution under the third (innermost) atmosphere880 The Sea: The Science of Ocean Prediction [75, 6Figure 1. Terrain and bathymetry (m) of the Madeira archipelago in the eastern Atlantic, consistingof the main island of Madeira and the flat islands of Porto Santo to the northeast, and Desertas tothe southeast.model domain. The ocean and atmosphere have 50 and 60 vertical levels, respectively. Theatmosphere and ocean models exchange flux information at a time interval of 6 minutesusing the ESMF. The COAWST system was applied to Madeira doubly nested with resolu-tions of 3 and 1 km (see Appendix b for more details). There are 40 vertical levels in WRFand 32 in ROMS. The variables are exchanged every 5 minutes. The coupling is performedusing the Model Coupling Toolkit (MCT) (Larson, Jacob, and Ong 2005; Jacob, Larson,and Ong 2005). The runs were performed for the month of June 2011.In the simulations both models produce a diurnal warm wake south of Madeira havinga magnitude of approximately 2–3◦C above surrounding waters (Figure 2). However thewarming in COAWST is more pronounced and shows stronger warming windward of theisland (the “antiwake”) compared with COAMPS. Daily satellite SST composites suggestthe pattern and amplitude (1–2◦C) of warming but do not permit an assessment of thetemporal evolution of the feature. Both models show strong flank “tip jet” winds and aweak wind region in the immediate lee of Madeira (Figure 3). Another distinction is thatCOAWST produces overall stronger winds whereas COAMPS exhibits terrain-modulatedflow from drainage winds (06UTC on 15 June).The variation between the models is indicative of differences in model configurationsand parameterizations. Section 2 describes the general modeling frameworks and param-eterization options, and Section 3 details the underlying physics of the coupling choices.Then in Section 4 we examine several case studies of short-term (∼days to week) cou-pled prediction that highlight modeling configuration choices such as the effects of wave2017] Pullen et al.: Coupled ocean-atmosphere forecasting 881Figure 2. Sea surface temperature surrounding Madeira Island, North Atlantic for 14–15 June2011 using two different coupled models: Coupled Ocean/Atmosphere Mesoscale Prediction Sys-tem (COAMPS), and Coupled-Ocean-Atmosphere-Wave-Sediment-Transport System (COAWST)(top), as well as multisensor daily composite from Group for High Resolution Sea Surface Tem-perature (GHRSST) (1 km) (bottom).coupling (Section 4a), coupling frequency and complex land-sea gradients (Section 4b),and sensitivity to grid resolution (4c). Section 5 follows with illustrations of medium-term(∼multiple weeks) sensitivities including the impact of temporal frequency of exchange(Section 5a) and role of ocean current feedback (Section 5b). Throughout, we focus onchallenging geographical settings and processes where energetic air-sea coupling is mani-fest. Our examples span novel hurricane genesis off Brazil and tropical weather patterns inthe Maritime Continent. We employ both operationally-oriented (COAMPS) and commu-nity model (ROMS/WRF) implementations. We arrive in Section 6 at a discussion of someof the challenges facing coupled modeling systems as they strive for enhanced realism intheir predictions. Section 7 contains the summary.882 The Sea: The Science of Ocean Prediction [75, 6Figure 3. 10-m winds around Madeira Island, North Atlantic for 14–15 June 2011 using two differ-ent coupled models: Coupled Ocean/Atmosphere Mesoscale Prediction System (COAMPS), andCoupled-Ocean-Atmosphere-Wave-Sediment-Transport System (COAWST).2. Coupling frameworka. ComponentsCoupled modeling systems comprise multiple components consisting of numerical atmo-spheric, oceanographic, and static databases and data assimilation systems for both theatmosphere and ocean (Figure 4). They can also incorporate hydrology and wave models.Typical systems have the ability to configure and cycle regional/analysis models in single-model (atmosphere, ocean, or wave) or coupled-model (atmosphere-ocean, ocean-wave,atmosphere-wave, and atmosphere-ocean-wave) modes.The systems consist of databases that provide terrain, bathymetry, land-sea masks, andsources of variability such as tides, rivers, and boundary forcing and initialization fields.Data assimilation approaches for the ocean or atmosphere, or a combination, are importantdrivers of enhanced predictability. They are commonly utilized in operational systems butmay not appear in research simulations. In the physical modules, the air and sea componentsexchange key parameters: 10-m surface winds, atmospheric pressure, relative humidity,atmospheric and oceanic surface temperature, precipitation, cloud fraction, surface windstress, heat fluxes, and net shortwave and longwave fluxes. The wind and wave componentsshare near-surface wind and Charnock parameter, and the ocean and wave components sharesea surface height, surface current, Stokes drift current, wave radiation stress gradient, andbottom orbital wave current. These variables and their sources within the physical modulecalculations are discussed as follows and in Section 3.2017] Pullen et al.: Coupled ocean-atmosphere forecasting 883Figure 4. Schematic of a typical coupled modeling system.In the Appendix, we use the example of two models, COAMPS and COAWST, to illustraterepresentative configurations of coupled systems. Details for other coupled model compo-nents may vary, but the underlying coupling strategies and functional modules remain thesame. Other coupled models referenced in this paper are SCOAR (Scripps Coupled Ocean-Atmosphere Regional Model with ROMS/WRF) utilizing an independent flux coupler, andUnified Wave INterface Coupled Model (UWIN-CM) consisting of WRF/Hybrid Coordi-nate Ocean Model (HYCOM) employing ESMF. Next we examine temporal and spatialconsiderations in coupled modeling configurations.b. Temporal exchangeFluxes may be computed sequentially at a specified coupling interval or interactivelyat each time step. The former case requires time-averaging the fields of interest (either thesurface variables or the internally computed fluxes) for consistency changes in the dynamics884 The Sea: The Science of Ocean Prediction [75, 6could occur if the coupling interval is larger than an hour or so. The atmosphere and oceanmodels alternate their time-stepping procedures in between the coupling intervals. Long-term (climate-scale) simulations often use 3-hourly, 6-hourly, or daily coupling frequency,whereas shorter than 1-hour coupling frequency is preferred for short-term simulations (e.g.,for simulation of tropical cyclones and weather forecasting). In Sections 4 and 5, we studythe effects of coupling frequency on short- and medium-term forecasting.c. Grid resolutionCare must be taken at land-sea boundaries to map only quantities and fluxes influencedby underlying ocean variables rather than land variables, or excessive high fluxes can resultfrom interpolation across the boundary.The regridding error can be large near steep orography, complex coastlines, and smallislands. To eliminate this issue, many models employ identical horizontal resolutions in theocean and atmosphere with matching grids and land-sea mask (e.g., Seo et al. 2008b, 2014,2016). Despite the higher computing cost of increasing the resolution of the atmosphereto match that of the ocean, one obvious benefit for the study of air-sea interaction is thatthe effect of air-sea coupling is maximized at the scale of the SSTs simulated by the oceanmodel. It also helps to lessen, somewhat, the computing burden associated with regridding,especially for large grids. In Section 4 we study coupled modeling sensitivity to resolutionand land-sea gradient effects.3. Coupling physicsa. Heat fluxesBesides employing high-fidelity models of the atmosphere and ocean, the computationof the surface fluxes of heat, momentum, and freshwater is the most critical aspect of theregional and global coupling system.The surface latent and sensible heat fluxes are very sensitive to the SST, both for itsdirect effect on the air-sea temperature and the saturation humidity (Cayan 1992) and forits influence on the stability of the atmospheric boundary layer (see review by Small et al.2008) This is manifest directly in the boundary layer (BL) model of the atmosphere and incomputations of latent and sensible heat fluxes computed from bulk formulae (e.g., Fairallet al. 1996). Latent heat flux, LH = ρaLew′q ′, and sensible heat flux, SH = ρacpaw′T ′,can be calculated using the standard bulk expressions LH = ρaLeCe(qs − qa)|Ua − Uo|and SH = ρacpaCh(Ts −Ta)|Ua − Uo|, respectively. The exchange coefficients, Ce and Ch,are larger when the air overlies warm SST compared to cold SST (increasing the heat flux),although their contributions are known to be small when compared with the increased windspeeds and temperature (humidity) differences (e.g., Liu et al. 1979).There are generally two ways to construct the flux coupler, using either bulk formulaor using fluxes computed from the atmospheric model. Both methods use the atmosphericfields as the input: short- and long-wave radiation fluxes from the radiation scheme, cloud2017] Pullen et al.: Coupled ocean-atmosphere forecasting 885Figure 5. Two estimates of summertime (July–September) surface latent heat flux (LH) (shading,[Wm−2], negative cooling the ocean) averaged for 6 years (2005–2010) obtained from the 7-kmScripps Coupled Ocean-Atmosphere Regional Model (SCOAR) model (Regional Ocean ModelingSystem-Weather Research and Forecasting [WRF-ROMS]) simulations for the U.S. West Coast. (a)LH estimated from the WRF surface layer scheme based on MM5 Monin–Obukhov similarity inconjunction with the Yonsei University (YSU) nonlocal-K scheme (Hong, Noh, and Dudhia 2006).(b) LH estimated from the Coupled Ocean–Atmosphere Response Experiment (COARE) bulk air-sea flux algorithm (Fairall, Kepert, and Holland 1994; Fairall et al. 2003). Black contours, identicalin both panels, denote the climatological summertime sea surface temperature (SST) simulated bythe model.fraction from the microphysics scheme, and precipitation rate from the convection andmicrophysics scheme. Both methods also need the near-surface meteorological fields, suchas air temperature and humidity at 2 m and vector and scalar wind at 10 m, that are calculatedin the planetary boundary layer (PBL) scheme, which is model dependent. These fieldsprovided by the atmospheric model are combined with the SST predicted by the ocean modeland the wave fields (if coupled to a wave model) to compute the surface latent and sensibleheat, freshwater, and momentum fluxes. The two methods differ in their simplification ofthe surface layer process to compute the surface fluxes. The bulk formulae are based on alocal closure-scheme, whereas the boundary layer model of the atmosphere invokes oftencomplicated nonlocal PBL parameterization schemes over the land and the water (e.g.,Hong and Pan 1996), which interact with the surface layer scheme to compute the surfacefluxes.The resultant surface flux estimates from the two methods can differ. Figure 5 illustratesthis by comparing the estimates of the summer mean surface latent heat flux from the 7-km SCOAR (WRF-ROMS) coupled model for the U.S. West Coast (Seo et al. 2016). One(left) is calculated from the WRF surface layer scheme based on the MM5 Monin–Obukhovsimilarity in conjunction with the Yonsei University (YSU) nonlocal-K scheme (Hong, Noh,and Dudhia 2006), and the other (right) is from the Coupled Ocean–Atmosphere Response886 The Sea: The Science of Ocean Prediction [75, 6Experiment (COARE) bulk air-sea flux algorithm (Fairall, Kepert, and Holland 1994; Fairallet al. 2003). The near-shore upwelling zone coincides with the region of weak latent heatflux (negative meaning cooling of the ocean), where the difference in the two latent heatflux estimates is small. However, for offshore regions with strong wind speed and warmSST, the difference can be quite large (up to 40 Wm−2).These differences can be due to several factors. Bulk formulae are empirical methodsoriginally developed for the tropical oceans. The bulk formulae thus include parameteri-zations for marine boundary processes that are not represented in many atmospheric BLmodels, including sea spray effects (which influence heat fluxes), skin temperature effects,or rainfall effects on surface fluxes, modification of drag coefficients under high- and low-wind regimes, etc. (Fairall, Kepert, and Holland 1994; Fairall et al. 2003). In oceanic regionswhere such effects might be important to improve the accuracy of surface flux estimates,bulk formulae might be preferred. Additional advantages of using bulk formulae may derivefrom the flux being calculated on the ocean grid, which usually has higher horizontal res-olution than the atmospheric model. The resultant surface fluxes would thus better reflectthe fine-scale SST patterns, which are otherwise smoothed in the coarser atmospheric gridsat which the BL models compute the surface fluxes. Bulk formulae are simple to modify inorder to be used in a variety of oceanic settings, including the polar oceans (e.g., Hunk andHolland 2007). However, the surface flux directly calculated from the BL models mightprovide more internally consistent surface fluxes to the system (including land and ocean),thereby conserving total enthalpy and mass.Computationally, the bulk formulaic method is more straightforward to calculate fluxesand apply to the coupler, because atmospheric variables from the BL models can be interpo-lated (if needed) to a common grid and a simple computation of flux components obtainedwith a standard scheme. The BL method instead invokes complicated nonlocal parameter-izations in the atmospheric model (e.g., Hong and Pan 1996), with SST, surface velocity,and roughness (wave heights) specified as inputs at the surface from the ocean model. Theresulting fluxes of momentum, sensible heat, and latent heat are then mapped to the oceangrid (if needed) as forcing terms.Long-wave radiative fluxes are determined directly from the black body formulationfor SST, whereas short-wave (incoming) radiative fluxes are computed based on the solarangle and cloudiness of the atmospheric model from the radiation and cloud microphysicsschemes.b. Momentum transferTurbulence is generated by the vertical shear of near-surface wind over water and landand is typically expressed as the Reynolds stress, τx = −ρu′w′ and τy = −ρv′w′, whereprime denotes the turbulent fluctuation, overbar is the time average, and ρ is the air density.A velocity scale u∗, friction velocity, is derived from u∗ = 1ρ√τ2x + τ2y . Using u∗, semiem-pirical distribution of wind speed at height z within the surface layer can be determined by2017] Pullen et al.: Coupled ocean-atmosphere forecasting 887u(z) = u∗k[ln(z−dz0) − ψm(z/L)], where L is the Monin–Obukhov length-scale and ψm isan empirical function for stability correction. The roughness length, z0, is a parameter thataccounts for the effect of the overlying wind on surface roughness due to waves (swells,gravity and capillary waves) (Bourassa, Vincent, and Wood 1999) and is often determinedby the steepest waves (Taylor and Yelland 2001). This information can be obtained from awave model (if coupled), which provides the wave-induced roughness length to the surfacelayer scheme to calculate the surface fluxes. If not, z0 is prescribed with a typical value of2×10−4 m over water.Surface layer schemes, whether from the atmospheric model scheme or bulk formulae,are based on the surface layer similarity theory and solve for u∗ to derive the momentumfluxes. In this formulation, local SST anomalies can modify the stability parameter, ψm,and thus the well-known logarithmic wind profiles within the surface layer, yielding theincreased (decreased) wind speeds over warm (cold) SST (e.g., Sweet et al. 1981; Busingerand Shaw 1984; Hayes, McPhaden, and Wallace 1989; Wallace, Mitchell, and Deser 1989;Chelton et al. 2004; Xie 2004).For very fast ocean waves and transient SST conditions (e.g., mixed-layer eddies andfronts), the stability function profiles described using Monin–Obukhov similarity theorymay not apply in these conditions, and the relationship between SST and surface windmay not be valid by the current empirical formulation in the surface layer schemes. More-over, different empirical formula in BL and bulk models would also mean that there existsuncertainty in estimates of surface flux, which can impact the result of the simulation. Forexample, different roughness length schemes (Bao et al. 2000) produce diverse results inthe intensity of Hurricane Opal (1995).The momentum fluxes (stresses) at the air-sea interface have traditionally been computedfrom wind speed, while assuming the ocean is at rest (Bye 1986). However, modern compu-tations of the stress can easily accommodate a changing ocean surface current and waves inboth the BL formulation and the bulk formulaic calculation of fluxes. In the former case, cou-pling of the atmospheric model to a wave model can provide the important surface roughnesslength scale associated with the wave heights. Surface currents, by creating a relative motionbetween air and sea surface, modify the total wind stress, τ = ρCd |Ua − Uo|(Ua − Uo),where Ua (Uo) refers to the wind (surface current) speed. Although in most cases surfacecurrent speed is an order of magnitude smaller than the wind speed, the influence of theocean currents associated with the mesoscale eddies and swift surface currents on the windstress have been noted in multiple studies (e.g., Kelly et al. 2001; Cornillon and Park 2001;Park, Cornillon, and Codiga 2006). Several of these studies showed that an inclusion ofsurface currents in the wind stress formulation leads to a decrease in the mean and eddykinetic energy (Pacanowski 1987) due in part to increased surface drag (Seo et al. 2007;Small et al. 2009; Eden and Dietze 2009; Seo et al. 2015; Seo 2017). Furthermore, Ekmanpumping is sensitive to surface vorticity and its gradient (Stern 1965; Gaube et al. 2015),implying some significant impact on the energetics of the eddy and circulation (Seo et al.2015) and biological productivity (Chelton et al. 2011; Gaube et al. 2014).888 The Sea: The Science of Ocean Prediction [75, 6c. Freshwater fluxesThe net freshwater fluxes are the sum of evaporation (E), precipitation (P), and runoff(R) and are important for the atmospheric moisture budget and convective response. Thefreshwater fluxes also affect the upper ocean salinity and density, which then affect deepocean convection and subduction. Evaporative fluxes to the atmosphere are directly derivedfrom the latent heat computation in the BL model or the bulk formulaic parameterization.Precipitation is obtained from the atmospheric model based on the cumulus scheme forconvective rain and the microphysics scheme for nonconvective rain.In addition, the river discharge is often an important driver of the regional ocean circu-lation and the hydrological cycle such as in the Bay of Bengal in the Indian Ocean (Seoet al. 2009) and the western equatorial Atlantic near the Amazon River mouth (Hu et al.2004). River runoff has been relatively crudely treated in regional ocean and coupled mod-els, often prescribed as the local source of freshwater (such as precipitation as in Howdenand Murtugudde 2001), as the virtual runoff (e.g., Yu and McCreary 2004), or by prescribedsalinity stratification (e.g., Vialard and Delecluse 1998). However, the runoff informationcan now be provided from the land model of the atmospheric component at a specifiedlocation of the domain. This approach ensures the conservation of the freshwater content inthe system. Although the freshwater fluxes affect density, the mass changes in precipitationand evaporation in the oceanic part of regional coupled models are very small and thereforeoften ignored in the model formulation.A new development in coupled models is to introduce a short-term hydrology fore-cast capability into the coupled modeling suite. For example, hydrological (WRF-Hydro)and enhanced soil treatment (NASA Land Information System [LIS]) are being added toCOAMPS to improve prediction of coupled air-sea-hydrological processes. The benefitof representing inland precipitation via an atmospheric model accrues to the hydrologymodel, which then affords the forecasting of river discharge at the coast. Significant buoy-ancy effects from river discharge are apparent in many coastal regions, and prediction ofthe local impacts of flood hazards and storm surge are a growing and significant applicationarea of coupled models (Saleh et al. 2016).4. Short-term forecasting (days to week) applicationsa. Effects of wave couplingIn short-term forecasting applications, a key consideration is whether wave coupling canincrease the fidelity of the simulation. In Section 3b, the role of momentum transfer in thephysics of coupling was emphasized. Here we examine U.S. east coast circulation duringa hurricane and evaluate the impact of waves in the context of air-sea coupled modeling.Such an evaluation approach can also be employed to assess the value of coupling othercomponents of the earth system (e.g., hydrology, ice) for the particular region of interest.Feedback from the ocean model component to the wave model component typicallyincludes the input of ocean model surface currents and water levels to the wave component.2017] Pullen et al.: Coupled ocean-atmosphere forecasting 889Water levels can modify the water depth used within the wave model physics calculations,though this effect is only significant if the water depth is sufficiently shallow that the wavesfeel the ocean bottom. Surface currents input into the wave model alter the effective windspeed (i.e., the wind speed relative to a frame of reference moving with the currents) andthe horizontal shear of the currents produce changes in the length, height, and direction ofthe waves.Wave model feedback to the ocean model component typically includes the Stokes’drift current (SDC), the wave radiation stress gradients, and the characteristic velocity andfrequency of the wave orbital motion near the ocean bottom. Wave motion near the bottomis used to enhance bottom drag in shallow water, and the radiation stress gradients that arisefrom breaking waves and their associated momentum in shallow water are applied to theocean model as a surface stress. The SDC is the average current velocity when followinga specific fluid parcel as it travels with the fluid flow. For instance, a particle floating atthe free surface of water waves, experiences a net Stokes drift current in the direction ofwave propagation. The SDC is calculated within the wave model component and passed tothe ocean model component to advect the ocean model fields. The SDC may also be usedin the enhancement of vertical mixing within the ocean model by Langmuir turbulence(Kantha and Clayson 2004). Additional shear production terms are added to the turbulentkinetic energy equations within the ocean model component to enhance vertical mixing. Thefollowing are some examples of air-ocean-wave model interactions that improve model skill.For the northeastern Gulf of Mexico outer continental shelf and slope domain consideredhere, observational wave and ocean data sets included an array of 14 Acoustic DopplerCurrent Profilers (ADCP) deployed by the Naval Research Laboratory (Teague et al. 2007)in May 2004 (Figure 6). These data were utilized by Smith et al. (2013) in relation toatmosphere-ocean-wave coupling in Hurricane Ivan using COAMPS-TC without a wavecomponent.The COAMPS-TC forecast track and intensity exhibited low errors allowing for a rea-sonable comparison of the ocean and wave fields. Table 1 depicts summary statistics for sixshallow-water ADCPs (M1–M6) shown in Figure 6 averaged over a 72-hr forecast periodinitiated on 14 September 2014, 1200 UTC. The complex correlation coefficient (see Smithet al. 2013) accounts for both the current speed and direction. The table shows highly cor-related speeds and directions for all 6 ADCP’s with the mean directional error lowest forthe 3 northern-most locations with ranges of approximately 10 to 15◦ for M4–M6 to thesouth. The modeled speeds agree well with observations. Figure 7 shows a comparison ofNavy Coastal Ocean Model (NCOM) current magnitude versus observation at the locationof ADCP M1. The NCOM results were adjusted temporally by six hours to more closelymatch the timing of the hurricane track. The COAMPS-TC simulation was approximatelysix hours too fast.In order to address these discrepancies, a fully coupled atmosphere-ocean-wave hindcaststudy was performed for a 2-month period beginning 1 September 2013. Figure 8 (top)depicts the modeling domain with the location of two NDBC (National Data Buoy Center)890 The Sea: The Science of Ocean Prediction [75, 6Figure 6. Acoustic Doppler Current Profilers (ADCP) array in the northern Gulf of Mexico (adaptedfrom Teague et al. 2007). Bathymetry contours (m) and the best track of Hurricane Ivan is shown.Inset: Infrared satellite image of Hurricane Ivan as it approaches the northern Gulf of Mexico coaston 15 September 2004.Table 1. Acoustic doppler current profiles (ADCP) versus NCOM current statistical comparisonsfor ADCPs M1-M6 averaged over the 72-hr 1200 UTC 14 September 2004 forecast (Smith et al.2013). CCC, complex correlation coefficient (takes in account speed and direction [see Kuzmicet al. 2006]); MDE, mean directional error; MB, mean bias; OBS, observed current speed.Top bin MDE Speed Max Speed Max SpeedADCP depth (m) CCC (degrees) MB (ms−1) NCOM (ms−1) OBS (ms−1)M1 6 0.83 1.57 0.046 2.04 2.14M2 4 0.77 8.45 0.082 1.75 1.87M3 6 0.79 12.29 0.110 1.61 1.73M4 10 0.78 10.44 0.121 1.77 1.96M5 11 0.82 10.37 0.071 1.82 1.91M6 9 0.83 15.54 0.145 1.69 1.82buoys used to evaluate the system in coastal waters. The ocean and wave models had agrid resolution of 2 km and were fully coupled to a 6-km atmospheric model. Initial andboundary conditions were provided by the models described in the Appendix. The modelwas spun-up for a 14-day period and run from 1 September to 27 October 2013. The wavemodel had a directional resolution of 7.5◦ (48 bins) and 25 frequencies. NCOM simulationsincluded tides and data assimilation. In this study, we will only present results for the CapeHenry, VA buoy location. Model results for Duck, NC (not shown) were very similar.2017] Pullen et al.: Coupled ocean-atmosphere forecasting 891Figure 7. Current velocity vertical profile for Navy Coastal Ocean Model (NCOM) (left) and observa-tions (right) at the location of acoustic Doppler current profiler (ADCP) M1, modified from Smithet al. 2013.Figure 8 (bottom) depicts a model-data comparison of significant wave heights at CapeHenry, VA (44099) during September/October 2013. The red line shows the significantwave heights with the fully coupled model, the blue line represents the wave-only results,and the black dashed line depicts the observed wave height. Several events are identifiedwhere the observed wave height exceeds 2 meters. The highest wave height was recorded onOctober 11 when it surpassed 4 m. This is associated with a low-pressure system located justoff the North Carolina coast with a minimum central pressure of 1,007 hPa. This featurewas resolved well by the COAMPS atmospheric model (not shown). Table 2 presentssummary statistics for the comparisons between the uncoupled and coupled wave model atthis location.Overall, the wave results from the coupled model simulations perform better than thewave-only runs. This is evident during periods of peak wave events (e.g., 11 October storm)where the uncoupled model wave heights gradually decrease, whereas the coupled modelportrays a more rapid decline in wave heights and strong agreement with observations. Thewave period biases were lower in October and within the range of instrument accuracy forboth sets of hindcasts. The wave directions for the coupled run show improved skill witha bias of −6.3◦ versus −11.5◦ and within the directional resolution of 7.5◦ for SimulatingWAves Nearshore (SWAN) (and instrument accuracy of 10◦). In summary, the inclusion ofcurrents (wave-current interaction) provides a more accurate wave-height field and is welljustified in significant sea states.b. Coupling frequency and complex land-sea gradientsOf key importance for coupled model configurations is the choice of frequency of modelcoupling (in time), as mentioned in Sections 1a and 2b. Furthermore, in complex coastal892 The Sea: The Science of Ocean Prediction [75, 6Figure 8. (top) Bounding box indicates the east coast U. S. domain of the fully coupled atmosphere-ocean-wave COAMPS (Coupled Ocean/Atmosphere Mesoscale Prediction System). The atmo-spheric model was run at 6-km resolution and the ocean and wave models were run at 2-kmresolution. National Data Buoy Center (NDBC) buoy 44099 (Cape Henry) is shown near themouth of Chesapeake Bay. (bottom) Comparison of significant wave height between coupled (red),uncoupled (blue), and observed (dashed black) wave height at Cape Henry, VA for the period of 1September to 27 October 2013.regions with steep terrain, the diurnal cycle depends on land-sea gradients and can be achallenge to simulate accurately. One example of such a setting is the Maritime Continent(MC) that bridges the Indian and Pacific Oceans.The MC is characterized by strong multiscale (diurnal to intraseasonal) and multicompo-nent (land, atmosphere, and ocean) interactions that are not properly represented in manyglobal prediction models. The subdaily variability in rainfall, and wind initiated over steeporography interacts with the adjoining coastal oceans to influence a region’s intraseasonalrainfall variability, and the Madden–Julian Oscillation (MJO). Prediction models show that2017] Pullen et al.: Coupled ocean-atmosphere forecasting 893Table 2. Statistics showing comparison of wave height, period, and direction for September and Octo-ber 2013 between NDBC (National Data Buoy Center) Buoy 44099 versus SWAN for uncoupled(left) and coupled air-ocean-wave (right) hindcasts. Shaded boxes indicate improved results overthe coupled or uncoupled runs. RMSE, root-mean squared error; MB, mean bias; CC, correlationcoefficient.Uncoupled CoupledMB RMSE CC MB RMSE CCSeptember 2013Wave height (m) −0.15 0.28 0.75 −0.12 0.20 0.90Wave period (s) −1.64 3.5 0.47 −1.46 2.80 0.55Wave direction (◦) −7.51 0.45 −9.08 0.49October 2013Wave height (m) 0.11 0.33 0.90 −0.04 0.22 0.95Wave period (s) −0.35 0.60 0.80 −0.71 0.97 0.81Wave direction (◦) −11.54 0.59 −6.28 0.68extended-range MJO prediction skill drops sharply as the MC is approached, suggestingthat some aspects of the deep convection over the MC are underrepresented. On the MC,thermally forced diurnal convection over the land is a primary factor that entails the deepconvection over the ocean, often leading to a formation of mesoscale convective systems.The propagation and intensification of diurnal rainfall across the land-sea boundaries is asalient feature of the rainfall variability in the MC, which is known to explain the majorityof the intraseasonal rainfall variability (Peatman, Matthews, and Stevens 2014).The primary driver of diurnal rainfall is the diurnal land heating (e.g., Mapes, Warner, andXu 2003). When averaged over the MC regions, typical amplitudes of diurnal land surfacetemperature (LST) and 2-m air temperatures (T2) are about 10◦C and 4◦C, respectively,whereas diurnal SST amplitude is much less, about 0.5◦C (Figure 9). Even this relativelyweak diurnal SST, however, should modulate the land-sea thermal contrast and the land-sea breezes, to which the propagation and intensification process of the rainfall is sensitive(Houze et al. 1981). Diurnal SST is not well-represented in many prediction models, and theextent to which the diurnal SST influences the diurnal amplitude and phase of the rainfallis an active area of research.To quantify this, a comparison is made from two regional model simulations target-ing the one-month (14 November–13 December 2011) period of the Dynamics of theMJO/Cooperative Indian Ocean Experiment on Intraseasonal Variability in Year 2011(DYNAMO/CINDY) experiment. One run is based on the multiensemble SCOAR (WRF-ROMS) fully coupled model simulations with 1-hourly coupling frequency (hereafter CF1)to represent the realistic diurnal SST and the air-sea interaction. The other run is based on aWRF-only simulation that is identical to WRF of CF1 but forced with the daily mean SSTfrom CF1 (CF1DM). Having the identical daily mean SST, the difference in atmosphericprocesses between CF1 and CF1DM indicates the effect of SST fluctuations on a diurnal894 The Sea: The Science of Ocean Prediction [75, 6Figure 9. Hourly composites of (a) rainfall (mmhr−1) averaged over the land only and (b) rainfallaveraged over the ocean only for the area of 95◦E–150◦E 10◦S–7◦N for 14 November to 13December 2011. (c) as in (a) but showing the land surface temperature (LST) (◦C, line with filledcircles) and 2-m air temperature (T2; ◦C, line with crosses), and (d) Sea surface temperature(SST). Rainfall is from the Tropical Rainfall Measuring Mission (TRMM)-3B42 satellite analysisproduct. CF1 is the fully coupled Weather Research and Forecasting-Regional Ocean ModelingSystem (WRF-ROMS) model simulation with diurnal sea surface temperature (SST). CF1DM isthe uncoupled WRF simulation forced with the daily mean SST.basis. Diurnal SST also raises the daily mean SST (Shinoda 2005), affecting initiationand intensity of the MJO convection (Seo et al. 2014), but this process is not considered.Because the ocean-atmosphere resolution of the model is 40 km, it relies on parameterizedconvection; that is, it does not capture the gravity wave dynamics that partially explain theoffshore migration of diurnal rainfall (Love, Matthews, and Lister 2011). Diurnal SST inthe present model affects diurnal rainfall by altering the land-sea thermal contrasts.Figure 9(a) and (b) compares the hourly composite anomalies of rainfall averaged sep-arately over the land grid points and the ocean grids within the MC domain (95–150◦E,10◦S–7◦N). The TRMM-3B42 precipitation estimate (gray) shows that the land rainfall isinitiated in the afternoon and peaks in the evening hours (19 Local Time [LT]), and thedaily minimum rainfall is during the morning hours at 10LT (Mori et al. 2004). The diurnalrainfall over the ocean has weaker diurnal amplitude and roughly the opposite phase, with2017] Pullen et al.: Coupled ocean-atmosphere forecasting 895the maximum in early morning hours (9LT) and the minimum in late night (21LT). Thetwo simulations capture reasonably well the observed amplitude and phase of the over-landand over-water diurnal rainfall variability. A close examination indicates that CF1 produceslower rainfall in the early morning hours (e.g., 6–10LT) compared to CF1DM both overthe land and ocean, while the difference in daily maximum (minimum) rainfall over land(ocean) at 19LT is negligible. The effect of diurnal SST is most clearly seen in the earlymorning hours, which is when the over-water diurnal rainfall is at a maximum and over-landrainfall is at a minimum.The difference in diurnal rainfall is attributed to the altered diurnal land-sea thermalcontrast. Figure 9(c) shows the diurnal composite of LST/T2, which are nearly identicalbetween CF1 and CF1DM. The inclusion of diurnal SST in CF1 (Figure 9[d]) producesthe lower SST by 0.2◦C in the morning/night hours but the higher SST by 0.3◦C duringthe daytime. Because the early morning hours correspond to the land breeze condition,the cooler SST in CF1 implies reduced land-sea thermal contrast and the weakened landbreeze. This would lead to weaker propagation of the land rainfall and the weaker rainfallover the ocean in CF1. Likewise, under the sea breeze condition during the daytime, CF1SST is higher, weakening the sea breeze and the intrusion of relatively cool marine airmassinland. Though this coincides with the slightly stronger over-land rainfall maximum (Figure9[a]), this effect appears to be small, as the diurnal rainfall at this time is dominated by thevigorous thermal convection.The top panel of Figure 10 shows the spatial patterns of the diurnal amplitude of rainfall(Δr), calculated following Love, Matthews, and Lister (2011), by fitting the hourly com-posite rainfall to a curve of the form, r = r¯ + Δrcos 2π(t − tmax)/24. Here, r is the totalrainfall, r¯ the daily mean rainfall, t the time (in hours), Δr the amplitude of the diurnalharmonic, and tmax the time of maximum rainfall. The same method is applied to T2. In theobservations and CF1 (Figure 10[a] and [b]), Δr is strong, reaching 1 mmhr−1 over landwith high elevation, whereas, in the open ocean, Δr is small and the diurnal cycle is notwell-defined. Δr remains relatively strong in the coastal regions that are affected by themigration of diurnal rainfall. The percent change in Δr due to diurnal SST (Figure 10[c])suggests that Δr increases moderately by 5% to 10% in Borneo and northern Sumatra, butit strongly decreases by more than 10% to 20% in southern Sumatra, Java, northern Borneo,and New Guinea. The change of Δr is accompanied by noticeable shifts in tmax. Figure10(d) and (e) shows that TRMM and CF1 have the tmax at night and morning hours overland and at early morning over the coastal seas. The shift in tmax between CF1 and CF1DM(Figure 10[f]) can be up to ±3 hours. Diurnal SST clearly modulates the amplitude andtiming of the diurnal rainfall over land.The areas of reduced Δr coincide well with the regions of reduced diurnal temperaturerange. Figure 10(g) and (h) show that ΔT2 (ΔLST as well, not shown) is large in thehighlands reaching 2.5◦C in the ERA-Interim and exceeding 3◦C in CF1. In contrast, thelow-lying coastal regions facing the ocean exhibit considerably weakerΔT2.ΔT2 increasesweakly (∼5%) in regions of the moderate increase in Δr, whereas it strongly decreases in896 The Sea: The Science of Ocean Prediction [75, 6Figure 10. (top): Diurnal amplitude of rainfall (Δr, mmhr−1) from (a) Tropical Rainfall MeasuringMission (TRMM) and (b) CF1. (c) Percent difference over land between CF1 and CF1DM, i.e.,(CF1–CF1DM)/CF1 * 100. (middle) Peak diurnal rainfall hours (Tmax) from (d) TRMM, (e) CF1,and (f) the difference between CF1 and CF1DM. (bottom): Diurnal amplitude of 2-m air temper-ature T2 (ΔT2, ◦C) from (g) ERA-Interim, (h) CR1, and (i) percent difference between CF1 andCF1DM.places where Δr is also decreased. The reduction in Δr and ΔT2 are most pronounced inJava and southern New Guinea, implying that the diurnal rainfall and the diurnal temperatureranges in these regions are subject to a strong marine influence.In summary, the analysis demonstrates some sensitivity of the amplitude and phase of thediurnal rainfall over the MC islands to diurnal SST. Incorrect phasing and amplitude of theMC diurnal rainfall simulated in prediction models can be thus partially attributable to lack,or underrepresentation, of diurnal SST. However, the result comes with an important caveatin that the model does not resolve the essential gravity wave response to convective andstratiform heating (Houze 1997). A more rigorous analysis is needed in a numerical modelthat explicitly resolves the diabatic heating and the gravity wave response. Nevertheless,it is evident that properly resolving with sufficient temporal resolution the land and oceandiurnal processes in the vicinity of land-sea gradients is a prime consideration in the MC.Diurnal air-sea coupling would be important for skillful simulation of SST and rainfall in2017] Pullen et al.: Coupled ocean-atmosphere forecasting 897other oceans surrounded by landmasses such as the Bay of Bengal. The strong salinity-driven stratification is known to amplify the diurnal SST (Mujumdar et al. 2011), which inturn affects the air-sea flux and diurnal rainfall variability (Weller et al. 2016; Kilpatrick,Xie, and Nasuno 2017).c. Sensitivity to grid resolutionThe role of grid resolution (Section 2c) is illustrated with an example of a novel phenom-ena of an intense extra-tropical cyclone that evolved into a hurricane that impacted Brazil.The choice of resolution impacts the quality of the fluxes (Section 3a) and propagatesthrough the numerics of the simulation to manifest in the simulated hurricane track.Brazil had its first documented experience with a hurricane during March 2004 when anextratropical storm transitioned into a hurricane (reaching category 1), named HurricaneCatarina. The diagnosis and understanding of the life cycle of this phenomenon was quitecomplex as an extratropical cold core low pressure system located southeast of Brazil shiftedoffshore over the open sea (Mctaggart-Cowan et al. 2006). It developed, gained strength, andtransitioned into a tropical cyclone due to various physical factors. Beyond the dynamicsand thermodynamics of the atmosphere, one of the suggested causes was the displacementof Catarina over the relatively warmer waters of the Southwest Atlantic that were presentalong the hurricane’s trajectory (Davis and Bosart 2004; Mctaggart-Cowan et al. 2006;Vianna et al. 2010).The oceanic surface feedback was crucial in determining the life cycle of this system.Hurricane Catarina interacted with four warm core rings, according to the oceanic obser-vational study by Vianna et al. (2010). From that analysis it is possible to deduce that thedestabilization of the marine atmospheric boundary layer, which promoted low verticalwind shear in the lower troposphere, may have played a key role in the intensification ofthis system, because the temperatures were higher (24◦C) in the warm rings than the surfaceair temperature (14◦C). This is an ideal opportunity for the transfer of heat from the sea tothe atmosphere, as shown in Pezzi et al. (2005, 2009) in the Brazil-Malvinas Confluence(BMC) region. This mechanism is also reported in several studies of nearby ocean regionswith strong thermal gradients in oceanic fronts (Acevedo et al. 2010; Camargo et al. 2013).In order to gain understanding of the physics of Hurricane Catarina’s transition from anextratropical to tropical cyclone, a total of four numerical simulations were performed. Theywere divided into a) atmospheric simulations with WRF-only where prescribed SST is usedand b) coupled (COA), which used fully two-way coupled WRF and ROMS. Furthermore,these simulations were divided according to their horizontal resolution (20 and 6 km). Allsimulations started on 18 September 2004 at 00 GMT and ran for 10 days without beingreinitialized and without any flux correction.Both WRF and COA simulate well the beginning of the phenomenon, in both 6- and 20-km resolutions (Figure 11). For the high-resolution experiments, Catarina’s lowest sea levelpressure reached 998 (WRF) and 1001 hPa (COA). For the very-high-resolution simulations,898 The Sea: The Science of Ocean Prediction [75, 6Figure 11. Sea level pressure (hPa) for the Hurricane Catarina on 23 March 2004. (a) WeatherResearch and Forecasting (WRF)-only high resolution. (b) Coupled ocean-atmosphere (COA) highresolution. (c) WRF-only very-high resolution. (d) COA very-high resolution.they were 1000 hPa (WRF) and 1003 hPa (COA). Those values found for the 23 Septembersimulations are in agreement with Mctaggart-Cowan et al. (2006), in which they estimatedvalues ranging from 1002 to 990 hPa during that day. However, analyzing the subsequentdays of the simulations the resolution turns out to be crucial, because, in both experiments2017] Pullen et al.: Coupled ocean-atmosphere forecasting 899Figure 12. Sea level pressure (hPa) for the Hurricane Catarina on 27 March 2004. (a) WeatherResearch and Forecasting (WRF)-only high resolution. (b) Coupled ocean-atmosphere (COA) highresolution. (c) WRF-only very-high resolution. (d) COA very-high resolution.using 20-km horizontal resolution, the cyclone intensity is decreased and it dissipates beforemaking landfall on the continent (Figure 12, upper panels).On the other hand, the simulations indicate that the phenomenon is relatively well rep-resented in the 6-km case (Figure 12, lower panels). Nevertheless, the very-high resolution900 The Sea: The Science of Ocean Prediction [75, 6simulations for both WRF and COA do not lower the pressure center as much as the oneestimated by Mctaggart-Cowan et al. (2006), being 1004 hPa (WRF) and 1002 hPa (COA),respectively, whereas the estimate was near 974 hPa.The differences found in the simulations may have several causes, but, as discussed inSections 2c and 3a, the grid resolution may be playing a determinant role in calculating thesurface fluxes by the atmospheric model bulk BL scheme, which is used in both WRF andCOA experiments. The resulting fluxes found in the case of very-high resolution simula-tions better represent the smaller-scale features in the SST, whereas in the lower resolutionsimulations these features are smoothed (Figure 13). This smoothing occurs both when theoceanic model is used at lower resolution and when WRF-only is used with prescribed SST.Together with the conditional instability of the second kind (CISK) mechanism, the heattransferred from ocean to atmosphere in cyclones in their mature phase is the mechanismthat maintains these phenomena (Rotunno and Emanuel 1987). In this way the cyclones arealso maintained by the energy transferred from the ocean to atmosphere via sensible andmainly latent heat flows in tropical regions. This is the case seen in the very-high resolutionexperiments (Figure 13[c] and [d]) where the latent heat fluxes are still higher undeneath thehurricane, compared to high-resolution experiments where Catarina has vanished (Figure13[a] and [b]).In addition to the resolution issue, the coupled simulation (COA), shown in Figure 14,can simulate the Hurricane Catarina eye closer to the place where it touched the continent,compared to the WRF-only experiment. This COA result is in very close agreement to thetrack estimation by McTaggart-Cowan et al. (2006). Thus the value of high-resolution gridsfor capturing dynamic processes in strongly coupled regimes is reinforced. This study addsto a growing body of literature documenting the superior skill that coupled (air-sea andair-sea-wave) models produce for hurricane prediction (e.g., Lee and Chen 2012; Chen andCurcic 2016).5. Medium-term (multiple weeks) applicationsa. Impact of temporal frequency of exchangeIn the arena of medium-term (several-week horizon) forecasts, the demands of managingcomputational resources can outweigh the considerations that would enhance model fidelity.This is because, for medium-range simulations, the domains are necessarily larger. To thatend, we examine the gains enabled by more-frequent coupling exchange between the airand sea in the context of medium-range projections.A key target of medium-range prediction is the MJO (Zhang 2005). The MJO is thedominant form of intraseasonal variability in the tropical atmospheric system, characterizedby large-scale, eastward-propagating, equatorially trapped, baroclinic oscillations in thetropical wind field at periods of 30–90 days (Madden and Julian 1971, 1994). MJO haspredictability time scales of 10–30 days, which exceeds classical weather prediction time-scales (e.g., Hendon et al. 2000; Waliser, Murtugudde, and Lucas 2003). Despite recent2017] Pullen et al.: Coupled ocean-atmosphere forecasting 901Figure 13. Latent heat flux (W/m2) and surface wind vectors (m/s) for the Hurricane Catarina on27 March 2004. (a) Weather Research and Forecasting (WRF)-only high resolution. (b) Coupledocean-atmosphere (COA) high resolution. (c) WRF-only very-high resolution. (d) COA very-highresolution.advancements in MJO simulation and prediction in the realistic full-physics global climatemodels and numerical weather prediction models (e.g., Woolnough, Vitart, and Balmaseda2007; Neale at al. 2008; Vitart and Molteni 2010; Subramanian et al. 2011), many aspects ofthe generation and maintenance of MJO remain elusive. These models lack many important902 The Sea: The Science of Ocean Prediction [75, 6Figure 14. Estimated track for Hurricane Catarina extracted from the very-high resolution experi-ments using Weather Research and Forecasting (WRF)-only (green line) and fully coupled oceanand atmosphere (COA) (red line).features of MJO because the multiscale interactions involving clouds and the oceans duringinitiation, growth, and propagation (Zhang 2005; Lin et al. 2006; Subramanian et al. 2011;Hung et al. 2013) are not well represented at coarse-grid resolutions.Regional climate models, which are driven by boundary conditions from reanalysis prod-ucts or free-running global climate models, could provide a unique opportunity to identifykey processes affecting MJO, such as initiation and growth of the MJO and the role ofair-sea interactions (Flatau et al. 1997). Upper-ocean processes determine mixed-layer heatcontent, stratification, and mixing, which alter the SST and the surface heat flux. The ensu-ing air-sea interactions are an important element for initiation and evolution of the MJO(Waliser, Lau, and Kim 1999; Zhang et al. 2006). The convection and the associated windsand incident solar radiation, in turn, generate intraseasonal variations in SST, feeding backto the intraseasonal variations in tropical atmospheric circulations associated with the MJO.Regional coupled models, however, have been only recently used for MJO simulation.Shinoda et al. (2013), for example, used the COAMPS regional coupled model with multi-nested grids and data assimilation in the atmosphere to demonstrate that the upper oceanresponse to the westerly wind bursts associated with the March–April 2009 MJO event was2017] Pullen et al.: Coupled ocean-atmosphere forecasting 903sensitive to the horizontal resolution of the atmospheric model. The equatorial jet of upto 1 m/s and the response in the upper ocean salinity and temperatures were significantlyunderestimated if the atmospheric resolution was reduced from 9 km to 27 km. This studyalso demonstrates the pronounced diurnal SST warming during the suppressed phase ofthe MJO, but their impact on the MJO convection has not been evaluated. And utilizing ahigh-resolution two-way coupled model, Pullen et al. (2015) demonstrated the influence ofMJO on the rainfall patterns in the mountains of the MC.A recent study by Seo et al. (2014) adopted a high-resolution, tropical-channel versionof the SCOAR model (WRF-ROMS) to evaluate the impact of preconvection SST warmingassociated with diurnal temperature variability in the upper ocean on the initiation andmaintenance of the MJO. A tropical channel configuration provides several benefits for thestudy of the MJO. Without the east-west lateral boundary conditions, it allows the circum-equatorial atmospheric disturbances to freely evolve (e.g., Ray et al. 2009, 2011) and interactwith the ocean at high resolution. For this reason, a tropical channel model approach hasbeen often used for the MJO simulations in the previous atmosphere-only regional models.For example, using the tropical-channel WRF model, Ray et al. (2009) demonstrated thecritical role of time-varying lateral boundary conditions for initiation of MJO by allowingmeridional advection of the westerly momentum from the extratropics.Using a series of the tropical-channel SCOAR coupled model simulations employingvarying coupling frequencies (CF) from one-hour to daily, the impact of the preconvectionSST on the MJO onset and intensification was quantified (Seo et al. 2014), as shown inFigure 15. The SCOAR in the tropical channel configuration represents reasonably wellthe observed eastward propagating precipitation anomalies associated with the observedMJO in November 2011 (Figure 15[a–d]). Although some predictability of the large-scalecirculation associated with the MJO might have been provided by the initial and lateralboundary conditions, sensitivity experiments with different CFs (Figure 15[e]) show a clearsensitivity of the initiation and intensity of the MJO deep convection to the preconvec-tion diurnal SST variability. The simulated precipitation amount associated with the MJObecomes much smaller when the model is coupled hourly to once daily (CF24). This isbecause the inclusion of resolved, interactively-generated diurnal SST variability nonlin-early rectifies and raises the time-mean SST and latent heat flux, preconditioning the localatmosphere for the deep convection. Diurnal SST variability also strengthens the diurnalmoistening of the troposphere by collocating the diurnal peak in latent heat flux with thatof SST. Both effects enhance the deep convection such that the total precipitation amountscales quasilinearly with preconvection diurnal SST amplitude and time-mean SST. How-ever, the timing of peak convection is not strongly influenced by the diurnal SST variability,suggesting that it may be more sensitive to the initial and lateral boundary conditions. Acomplementary WRF-only simulation forced with the persistent initial SST (CF1PS), lack-ing the enhanced preconvection warming and moistening, produces a weaker and delayedconvection. The fact that time-evolving SST with a diurnal cycle strongly influences theonset and intensity of the MJO deep convection suggests that improved representation of904 The Sea: The Science of Ocean Prediction [75, 6Figure 15. (a–e) Time-longitude diagrams of daily mean precipitation rate anomaly (mmhr−1) aver-aged in 10◦S–10◦N from (a) Tropical Rainfall Measuring Mission (TRMM)-3B42, (b) 1-hourlycoupled Scripps Coupled Ocean-Atmosphere Regional (SCOAR) model simulation (CF1), (c) dailycoupled (CF24), and (d) Weather Research and Forecasting (WRF)-only simulation forced withdaily mean SST obtained from CF1 (CF1DM). Two magenta diagonal lines, common to all thepanels, denote the 8-ms−1 phase lines derived from (a). (e) Time-series of daily mean total pre-cipitation rates (mmhr−1) averaged over the central Indian Ocean (73–80.5◦E and 0.7◦S–7◦N) forTRMM, CF1, CF24, and WRF only simulation forced with persistent initial sea surface temperature(SST) (CF1PS).diurnal SST might help to produce more reliable forecasts of MJO onset and growth forenhancing medium-range predictability (DeMott, Klingaman, and Woolnough 2015).b. Role of ocean current feedbackThe energetic effects of surface currents on regional atmospheric forecasts for themedium-range are a manifestation of the momentum transfer between models (Section3b). This aspect of coupling is frequently not fully appreciated as a factor in the formulationof the coupling interface. Therefore, we close with an explicit examination of the role ofsurface current feedback in the context of a coupled model.Tropical instability waves (TIWs) occur in the eastern tropical Pacific and tropical AtlanticOceans along the SST fronts associated with the equatorial cold tongues. These SST undu-lations are observed to have horizontal wavelengths of 1,000 to 2,000 km and periods of20 to 30 days, propagating westward at approximately 0.5 m/s (Qiao and Weisberg 1995).2017] Pullen et al.: Coupled ocean-atmosphere forecasting 905Observations and numerical studies have shown that TIWs are generated through instabilityof the equatorial current and frontal system (e.g., Philander 1976, 1978; Cox 1980; Hansenand Paul 1984; Yu, McCreary Jr., and Proehl 1995), and play a leading role in the upperocean heat balance in the equatorial cold tongue (e.g., Jochum, Malanotte-Rizzoli, andBusalacchi 2004; Pezzi et al. 2004, Jochum and Murtugudde 2006; Menkes et al. 2006).These intraseasonal undulations of SST fronts by TIWs induce surprisingly coherentresponses in the surface and lower-tropospheric fields (e.g., Hashizume et al. 2002; Chel-ton et al. 2004; Pezzi et al. 2004). This covariability in ocean and atmospheric fields onintraseasonal time-scales driven by the ocean dynamics has been a topic of several regionalcoupled modeling studies, given their role in the upper ocean heat balance. The dynamics ofcoupling and impact on the large-scale fields by TIWs were examined in a series of SCOARmodel simulations by Seo et al. (2006; 2008a) and Seo, Miller, and Roads (2007), and Seoand Xie (2011). They reveal that TIW-induced SST and surface current anomalies exert sig-nificant damping effect on the thermodynamic and energetics of the TIWs, influencing thepropagation characteristics and the upper-ocean heat budget and the overlying atmosphericprecipitation variations. Using the International Pacific Research Center (IPRC) regionalcoupled model (IROAM), a series of papers (Xie et al. 2007; de Szoeke et al. 2007; Small etal. 2009) also diagnose the effect of TIWs on the heat budget, SST, and eddy kinetic energy inthe eastern equatorial Pacific. These studies demonstrate that including the feedback effectby TIWs on the atmosphere via SST and surface currents in the coupled model influencespredictability aspects of both the atmospheric and ocean flows in this region. This effect willneed to be included in regional forecast experiments to quantify their impact, especially inoceanic regions with strong eddy activities, such as in the California Current System (Seoet al. 2016; Renault et al. 2016b) and in western boundary current regions. In these regions,the current feedback may influence the low-level baroclinicity/storm track variability (e.g.,Renault 2016a) and horizontal moisture transport (Seo 2017) in the atmosphere.6. Coupling challenges and advancesa. Coupled model initializationAs limited area models, regional coupled systems require high-fidelity inputs to maximizeforecast skill. When setting up a new region for a coupled modeling system, considerationmust be given to the length of time required to spin up the individual components of thesystem. It is usually difficult for basin-scale and global general circulation models to reach anequilibrium state, as it can take hundreds of years. Typically, an ocean model is initializedwith the present ocean state and is integrated forward until the circulation is consistentwith the prescribed water mass structure. In other words, it adjusts geostrophically to itsinitial state. However, the ocean initial state is imperfect due to the sparseness of dataat depth. In contrast, the atmospheric model, if run uncoupled, can typically be spun-upover a period of several days to resolve the synoptic features in the region. Similarly, aminimum period of a week is required to spin-up a regional wave model to adequately906 The Sea: The Science of Ocean Prediction [75, 6resolve swell propagating into the model domain from distant locations, depending on thegeographic region of interest. When a regional coupled model system is initialized from acold start, each component is initialized from a global atmosphere, ocean, and wave model.These same models are used to provide boundary conditions for the respective model. Theready availability of global fields and all requisite atmosphere and ocean observations isessential. These requirements make demands on the storage and computational systems athand.b. Coupled model data assimilationAssimilation techniques are used to make modeled fields resemble atmospheric andoceanic circulation patterns observed (and measured) at a given time and place, and tomake predictions of future oceanic and atmospheric circulations from the initial state. Dataassimilation is also useful to understand previous patterns of variability. The above timespanto achieve the initial state assume that both the atmosphere and ocean models (both globaland regional) are being run in a data assimilative mode.There are two types of assimilation schemes: direct insertion and incremental insertion.In the first scheme, data is inserted whenever it becomes available (global models typicallylook back for observations at least 72 hours). This has some advantages for the ocean, wherein situ data is sparse and doesn’t necessarily arrive on a regular schedule. The second type,incremental insertion, uses new observations by introducing the data gradually over time inorder to minimize the imbalances that result from mismatches between the model and thedata. Often, a hindcast with data assimilation is performed before a forecast is generated inorder to utilize observations to provide the best possible initial state for the pending forecast.An important aspect to data assimilation is quality control of the observational data.Quality control checks of the data are performed to flag data outliers that fall outside of 1to 3 standard deviations of the data, usually based on climatology.In general, the update cycle chosen in a coupled system is based upon the availableobservational data for each medium, as well as the dynamics of the fluid involved. Becauseatmospheric observations are much greater in number than oceanic observations, a six-hourly-update cycle with data assimilation may be adequate to capture the initial state andprovide an accurate forecast. On the other hand, the ocean model may require a 24-hour oronce-per-day update cycle due to the limited number of observations.Increasingly, consistent assimilation across the air-sea boundary is viewed as the mostreliable means of generating a coherent initial and updated state. In particular, “strongly-coupled” data assimilation, in which observations within one realm affect the other con-stituents (with some type of constraints), may represent a significant path to enhancedpredictability of coupled systems (e.g., Frolov et al. 2015; Lu et al. 2015).Related advances include the incorporation of data assimilation into other coupled modelcomponents. For example, the COAMPS effort includes incorporation of a wave data assimi-lation method (Veeramony et al. 2014) under development. And COAMPS is being extended2017] Pullen et al.: Coupled ocean-atmosphere forecasting 907for Arctic domains by coupling to the Community Ice CodE (CICE) sea ice model for arelocatable COAMPS-CICE modeling system. ROMS is also being coupled to CICE usingthe Model Coupling Toolkit (MCT). This is an active area of research and will certainly berepresented in more coupled models of the future.As more components of the earth system are drawn into coupled model interfaces, atten-tion needs to be given to the underlying assumptions of the individual models. For instance,many components were tuned to run in a standalone manner. These issues are exacerbatedby data assimilation schemes (and hence motivated the need for coupled data assimilation),but these challenges exist even in the absence of assimilation. Such drivers necessitate aperpetual examination of the component model physics and associated underlying param-eterizations to confirm that their validity is upheld in the context of fully two-way coupledmodeling systems.7. SummaryAs the studies mentioned and presented herein attest, there is a growing body of researchand operational experience with coupled models to draw on in advancing the field. Thispaper has overviewed multiple ways in which coupled forecasting improves the predictiveskill of both retrospective and operational simulations.Coupled air-sea modeling advances have relied on a variety of different field cam-paigns over the past decade to refine and validate coupled modeling systems. These includeCoupled Boundary Layers Air-Sea Transfer (CBLAST, 2001–2003), Adriatic CirculationExperiment (ACE, 2002–2003), Ligurian Sea Air-Sea Interaction Experiment (LASIE07,2007), VAMOS Ocean Cloud Atmosphere Land Study (VOCALS, 2008), Philippine StraitsDynamics Experiment (PhilEx, 2007–2009), CYNDY/DYNAMO (2011–2012), and Air-Sea Interaction Regional Initiative in the Northern Indian Ocean (ASIRI, 2014–2016).The application of coupled models is progressing from documenting when, where, andhow coupling matters to the holistic adoption of coupled systems in research and oper-ations. The Earth System Prediction Capability (ESPC) is a U.S. initiative to coordinatemultirealm coupled prediction across federal agencies (Carman et al. 2017) utilizing acommon infrastructure and ESMF-based coupling (Theurich et al. 2016). A particularlycompelling development is the unified model approach whereby a single modeling suiteis employed targeting subseasonal to seasonal timescales that bridge weather and climate(National Research Council 2016). Yet challenges remain in seamless initialization, in theintegration of the models tuned to individual realms, and in the further development of dataassimilation, especially coupled data assimilation.Short time-scale forecasting clearly benefits from including highly resolved physicalprocesses that currently require regional coupled models to properly implement. Futureincreases in computational efficiency will allow ever more finely-resolved grids and evermore complicated representations of physical processes in these models, potentially extend-ing to global grids. Carefully designed sensitivity studies and skill assessments, with908 The Sea: The Science of Ocean Prediction [75, 6accompanying observational verification, need to support these types of simulations andforecasts in order to verify that the mechanisms leading to skillful predictions are cor-rectly represented in the models. Such studies and assessments will also facilitate emergingadvances in data assimilation and their incorporation and validation in coupled air-sea pre-diction systems. These steps will also be carried out across more components of the earthsystem in emerging coupled model frameworks.On the research to operations front, operational coupled systems continue to proliferatein international environmental prediction centers such as the European Center for Medium-range Weather Forecasting (ECMWF), Japan Agency for Marine-Earth Science and Tech-nology (JAMSTEC), the UK Met Office, and the Canadian Meteorological Center, amongothers (Smith, Roy, and Brasnett 2013; Fallmann et al. 2017). And increasingly, interna-tional projects are linking modeling and observational capabilities. An upcoming program,the Years of the Maritime Continent (2017–2019; http://www.jamstec.go.jp/ymc/), is anexample of a multicountry initiative to model and measure coupled processes related toMJO propagation across the land region between the Indian and Pacific Oceans. It offers anunprecedented opportunity to examine configuration and parameterization choices and theirassociated outcomes across multiple coupled models. Further advances will be achievedin this way by exercising coupled models side-by-side in operational and field campaignsettings.Appendixa. COAMPSCoupled Ocean/Atmosphere Mesoscale Prediction System (COAMPS), developed bythe Naval Research Laboratory is used to study and predict short-term (3 to 7 days) atmo-spheric and oceanic phenomena for research, operations, and real-time experiment support.The atmospheric component of COAMPS (Hodur 1997; Chen et al. 2003) is a nonhydro-static, terrain-following, and cloud-resolving nested model that includes parameterizationsfor subgrid-scale mixing, cumulus cloud parameterizations, and moist physics. A tropicalcyclone version of the model (COAMPS-TC) (Doyle et al. 2014) has been developed forthe prediction of cyclone track, intensity, and structure.The ocean component is the Navy Coastal Ocean Model (NCOM) developed by theNaval Research Laboratory for regional and coastal ocean modeling (Martin 2000; Barron2006; Allard et al. 2014; Rowley et al. 2014) of ocean temperature, salinity, currents, andsea surface height. The vertical grid levels in NCOM consist of a set number of terrain-following sigma levels from the surface down to a user-specified depth, with z-levels below.The model has numerous mixing parameterizations, includes tides, and has an option torun with real-time river discharge. Wave forcing terms including the Stokes drift current,gradients of radiation stress, and enhancement of vertical mixing by Langmuir turbulenceas described by Kantha and Clayson (2004) have been incorporated into NCOM.The wave component options in COAMPS are the Simulating Waves Nearshore (SWAN)(Booij, Ris, and Holthuijsen 1999; SWAN Team 2010) and WAVEWATCH III (Tolman2017] Pullen et al.: Coupled ocean-atmosphere forecasting 9092009, Rogers, Dykes, and Wittmann 2014). SWAN is designed to run efficiently at highresolutions (e.g., 2 km or finer) through the utilization of implicit propagation methods.SWAN can be run in nested mode receiving boundary conditions from a global or regionalWAVEWATCH III. Recent modifications to the SWAN wave model include a new wind-to-wave energy input and wave-breaking energy dissipation source functions (Babanin etal. 2010) and a new ocean surface drag formulation for high-wind conditions. The newfeatures showed significant improvement in forecasting wave properties during tropicalcyclone conditions. In COAMPS, WAVEWATCH III is typically run as a regional wavemodel while SWAN is used for nearshore domains.The system is coupled using the Earth System Modeling Framework (ESMF) (Campbellet al. 2010; Chen et al. 2010) and supports air-ocean (Small et al. 2011), air-wave, ocean-wave, and air-ocean-wave coupling. The wind–wave coupling consists of the atmosphericmodel providing wind forcing to the wave model and the wave model providing feedbackvia a sea-state-dependent Charnock parameter, which is used to compute the atmosphericmomentum drag at the air–sea interface. The exchange of fields can be set by the user (e.g.,hourly) but typically ranges between 5 and 10 minutes.The atmospheric data assimilation system in COAMPS is the Navy Atmospheric Vari-ational Data Assimilation System (NAVDAS); Daley and Barker 2000) that assimilatesconventional observations such as surface, upper-air, and satellite data. NCOM assimilatesreal-time ocean observational data including remotely-sensed SST, sea surface height, insitu surface and subsurface observations of temperature and salinity, and measurementsfrom ships, buoys, expendable thermographs, and floats using the Navy Coupled OceanData Assimilation (NCODA) system (Cummings 2005; Cummings and Smedstad 2013).NCODA is a fully three-dimensional, multivariate, optimal-interpolation (MVOI) oceandata assimilation system.Each of the model components described receives initial/boundary (I/B) conditionsfrom a global model. The NAVy Global Environmental Model (NAVGEM) (Hogan etal. 2014) is the Navy’s operational global weather prediction model. NAVGEM combinesa semi-Lagrangian/semi-implicit dynamical core with parameterizations for subgrid scalemoist processes, convection, and radiation. This model is used for I/B conditions for theregional COAMPS atmospheric model. The U.S. Navy’s operational global ocean now-cast/forecast system is comprised of the HYbrid Coordinate Ocean Model (HYCOM)and the Navy Coupled Ocean Data Assimilation (NCODA) system (Metzger et al.2014).Global HYCOM is used to provide I/B conditions for regional NCOM forecasts. WAVE-WATCH III (Tolman 2009; Rogers, Dykes, and Wittmann 2014) is an operational globaland regional model that is run at the NOAA National Center for Environmental Prediction(NCEP), the Fleet Numerical Meteorology and Oceanography Command, and the NavalOceanographic Office. I/B conditions for this model are used for regional SWAN now-cast/forecast systems.910 The Sea: The Science of Ocean Prediction [75, 6The coupled system is typically spun-up for several weeks prior to generating forecastproducts. Typically, a 72-hr ocean forecast is generated twice daily (00, 12 UTC) in acoupled mode, with an atmospheric update cycle performed at 06 and 18 UTC.Numerous validation studies for the system components have been performed foratmosphere-ocean (Allard et al., 2010; Jensen et al., 2015), ocean-wave (Allard et al.,2012), and atmosphere-wave coupling (Chen et al. 2014).b. COAWSTThe Coupled-Ocean-Atmosphere-Wave-Sediment-Transport System (COAWST) cou-pling is performed using the Model Coupling Toolkit. Some model specifications used inthe Madeira archipelago simulation of Section 1 are described here, but data sources andmodel settings may vary depending on the region for which the open source code is con-figured. The initial and boundary atmospheric conditions are retrieved from the NationalCenter for Environmental Prediction (NCEP) Final Analyses with a 1◦ spatial resolutionevery 6 hours. WRF topography was retrieved from SRTM 3-second database, and, for landuse type, the Corine Land Cover dataset is employed. For WRF, the following options werechosen: cloud microphysics use the WRF Single-Moment 6-class scheme (Hong, Noh, andDudhia 2006), longwave radiation uses the Rapid Radiative Transfer Model (Mlawer andClough 1997), shortwave radiation uses the Dudhia scheme (Dudhia 1989), the surfacelayer follows the Eta similarity scheme (Janjic 1996, 2002) and the planetary boundarylayer follows the Mellor–Yamada–Janjic scheme (Janjic 1990, 1994).Oceanic initial and boundary conditions are based on 5-day intervals of HYCOM (HybridCoordinate Ocean Model) + NCODA (Navy Coupled Ocean Data Assimilation) global1/12◦ analysis data. Raw bathymetry data were taken from the 30-second resolution,GEBCO (General Bathymetric Chart of the Oceans) bathymetry. In ROMS (Regional OceanModeling System), radiation and nudging boundary conditions are imposed for tracers (tem-perature and salinity) and baroclinic velocities (Marchesiello, McWilliams, and Shchepetkin2001), a Flather condition for barotropic velocities (Flather 1976), and a Chapman boundarycondition for free-surface height (Chapman 1985). Vertical turbulent mixing is parameter-ized with the Generic Length Scale (GLS) scheme, using a k–kl closure, where k is theturbulent kinetic energy and l is the length scale (Warner et al. 2005). The model statevariables are staggered on an Arakawa C-grid.Acknowledgments. JP acknowledges support from Office of Naval Research (ONR) grant N00014-10-1-0300. RAA and TAS were supported through the 6.2 NRL Core Project “Coupled Ocean–WavePrediction System” Program Element #0602435N. HS acknowledges support from ONR (N00014-15-1-2588), NSF (OCE-f 419235), and NOAA (NA15OAR4310176). AJM was supported by the NSFEarth System Modeling Program (OCE1419306) and the NOAA Climate Variability and PredictionProgram (NA14OAR4310276). LPP is supported by CNPq’s fellowships on scientific productiv-ity (CNPq 304009/2016-4). This is also one of his contribution to the project Advanced Studies inMedium and High Latitudes Oceanography (CAPES 23038.004304/ 2014-28). JA and RC were2017] Pullen et al.: Coupled ocean-atmosphere forecasting 911financially supported by the Oceanic Observatory of Madeira Project (M1420-01-0145-FEDER-000001-Observatório Oceânico da Madeira-OOM). This is NOAA GLERL Contribution Number1854.REFERENCESAcevedo, O. C., L. P. Pezzi, R. B. Souza, V. Anabor, and G. Degrazia. 2010. Atmospheric boundarylayer adjustment to the synoptic cycle at the Brazil-Malvinas Confluence South Atlantic Ocean.J. Geophys. Res., 115, D22107. doi: 10.1029/2009JD013785Allard, R. A., T. J. Campbell, T. A. Smith, T. G. Jensen, S. Chen, J. A. Cummings, J. Doyle et al.2010. Validation Test Report for the Coupled Ocean Atmosphere Mesoscale Prediction System(COAMPS) Version 5.0 NRL/MR/7322–10-9283. Stennis Space Center, MS: Oceanography Divi-sion, Naval Research Laboratory, 172 pp.Allard, R., E. Rogers, P. Martin, T. Jensen, P. Chu, T. Campbell, J. Dykes et al. 2014. The US Navy cou-pled ocean-wave prediction system. Oceanography, 27(3), 92–103. doi: 10.5670/oceanog.2014.71Allard, R. A., T. A. Smith, T. G. Jensen, P. Y Chu, E. Rogers, T. J. Campbell, U. M. Gravois et al.2012. Validation Test Report for the Coupled Ocean/Atmosphere Mesoscale Prediction System(COAMPS) Version 5.0: Ocean/Wave Component Validation. NRL/MR/7320-12-9423, 91 pp.Babanin, A. V., K. N. Tsagareli, I. R. Young, and D. J. Walker. 2010. Numerical investigation ofspectral evolution of wind waves. Part II: Dissipation term and evolution tests. J. Phys. Oceanogr.,40, 667–683. doi: 10.1175/2009JPO4370.1Bao, J. W., J. M. Wilczak, J. K. Choi, and L. H. Kantha. 2000. Numerical simulations of air-sea inter-action under high wind conditions using a coupled model: A study of hurricane development. Mon.Weather Rev., 128(7), 2190–2210. doi: 10.1175/1520-0493(2000)128<2190:NSOASI>2.0.CO;2Barron, C. N., A. B. Kara, P. J. Martin, R. C. Rhodes, and L. F. Smedstad. 2006. Formulation,implementation and examination of vertical coordinate choices in the global Navy Coastal OceanModel (NCOM). Ocean Model., 11, 347–375. doi: 10.1016/j.ocemod.2005.01.004Barton, E. D., G. Basterretxea, P. Flament, E. G. Mitchelson-Jacob, B. Jones, J. Arístegui, and F.Herrera. 2000. Lee region of Gran Canaria. J. Geophys. Res.: Oceans, 105(C7), 17173–17193. doi:10.1029/2000JC900010Bender, M. A., I. Ginis, R. Tuleya, B. Thomas, and T. Marchok. 2007. The operational GFDL coupledhurricane–ocean prediction system and a summary of its performance. Mon. Weather Rev., 135,3965–3989. doi: 10.1175/2007MWR2032.1Bender, M. A. and I. Ginis. 2000. Real-case simulations of hurricane–ocean interaction using a high-resolution coupled model: Effects on hurricane intensity. Mon. Weather Rev., 128, 917–946. doi:10.1175/1520-0493(2000)128<0917:RCSOHO>2.0.CO;2Booij, N., R. C. Ris, and L. H. Holthuijsen. 1999. A third-generation wave model for coastal regions:Part 1. Model description and validation. J. Geophys. Res.: Oceans, 104(C4), 7649–7666. doi:10.1029/98JC02622Bourassa, M. A., D. G. Vincent, and W. L. Wood. 1999. A flux parameterization including theeffects of capillary waves and sea state. J. Atmos. Sci., 56(9), 1123–1139. doi: 10.1175/1520-0469(1999)056<1123:AFPITE>2.0.CO;2Businger, J. A. and W. J. Shaw. 1984. The response of the marine boundary layer to mesoscale variationin sea-surface temperature. Dyn. Atmos. Oceans, 8, 267–281. doi: 10.1016/0377-0265(84)90012-5Bye, J. A. T. 1986. Momentum exchange at the sea surface by wind stress and understress. Q. J. R.Met. Soc., 112, 501–510. doi: 10.1002/qj.49711247212912 The Sea: The Science of Ocean Prediction [75, 6Caldeira, R., X. Couvelard, R. Vieira, C. Lucas, I. Sala, and I. Vallès Casanova. 2016. Challengesof building an operational ocean forecasting system for small island regions: Regional to local.J. Oper. Oceanogr., 9(1), 1–12. doi: 10.1080/1755876X.2016.1205304Caldeira, R. M. A., S. Groom, P. Miller, D. Pilgrim, and N. P. Nezlin. 2002. Sea-surface signaturesof the island mass effect phenomena around Madeira Island, Northeast Atlantic. Remote Sens.Environ., 80(2), 336–360. doi: 10.1016/S0034-4257(01)00316-9Caldeira, R. M. and R. Tomé. 2013. Wake response to an ocean-feedback mechanism: Madeira Islandcase study. Boundary Layer Meteorol., 148(2), 419–436. doi: 10.1007/s10546-013-9817-yCamargo, R., E. Todesco, L. P. Pezzi, and R. B. Souza. 2013. Modulation mechanisms of marineatmospheric boundary layer at the Brazil-Malvinas Confluence region. J. Geophys, Res.: Atmos.,118, 6266–6280. doi: 10.1002/jgrd.50492Campbell, T., R. Allard, R. Preller, L. Smedstad, A. Wallcraft, S. Chen, H. Jin et al. 2010. Inte-grated modeling of the battlespace environment. Comp. Sci. Eng., 12(5), 36–45. doi: 10.1109/MCSE.2010.78Carman, J. C., D. P. Eleuterio, T. C. Gallaudet, G. L. Geernaert, P. A. Harr, J. A. Kaye, D. H. McCarrenet al. 2017. The National Earth System prediction capability: Coordinating the giant. Bull. Am.Meteorol. Soc., 98(2), 239–252. doi: 10.1175/BAMS-D-16-0002.1Cayan, D. R. 1992. Latent and sensible heat-flux anomalies over the Northern Oceans: Drivingthe sea surface temperature. J. Phys. Oceanogr., 22, 859–881. doi: 10.1175/1520-0485(1992)022<0859:LASHFA>2.0.CO;2Chapman, D. C. 1985. Numerical treatment of cross-shelf open boundaries in a barotropiccoastal ocean model. J. Phys. Oceanogr., 15, 1060–1075. doi: 10.1175/1520-0485(1985)015<1060:NTOCSO>2.0.CO;2Chelton, D. B., P. Gaube, M. G. Schlax, J. J. Early, and R. M. Samelson. 2011. The influence ofnonlinear mesoscale eddies on near-surface chlorophyll. Science, 334, 328–332. doi: 10.1126/sci-ence.1208897Chelton, D. B., M. Schlax, M. H. Freilich, and R. F. Milliff. 2004. Satellite measurements reveal per-sistent small-scale features in ocean winds. Science, 303, 978–983. doi: 10.1126/science.1091901Chelton, D. B. and S.-P. Xie. 2010. Coupled ocean-atmosphere interaction at oceanic mesoscales.Oceanography, 23, 52–69. doi: 10.5670/oceanog.2010.05Chen, S., T. Campbell, S. Gaberšek, R. Hodur, and P. J. Martin. 2010. Effect of two-way air–seacoupling in high and low wind speed regimes. Mon. Weather Rev., 138(9), 3579–3602. doi:10.1175/2009MWR3119.1Chen, S., J. Cummings, J. Doyle, R. H. Hodur, T. Holt, C. Liou, M. Liu et al. 2003. COAMPS Version3 model description: General theory and equations. NRL/PU/7500–03-448, 145 pp.Chen, S. S. and M. Curcic. 2016. Ocean surface waves in Hurricane Ike (2008) and SuperstormSandy (2012): Coupled model predictions and observations. Ocean Model., 103, 161–176. doi:10.1016/j.ocemod.2015.08.005Chen, S., S. Gaberšek, J. D. Doyle, J. Cook, P. Y. Chu, R. A. Allard, P. A. Wittmann, and W. A. Nuss.2014. Validation test report for the Coupled Ocean/Atmosphere Mesoscale Prediction System(COAMPS) Version 5.0: Air/Wave Component Validation. NRL/MR/7533–9526, 68 pp.Chen, S. S., J. F. Price, W. Zhao, M. A. Donelan, and E. J. Walsh. 2007. The CBLAST- Hurricaneprogram and the next-generation fully coupled atmosphere–wave–ocean models for hurricaneresearch and prediction. Bull. Amer. Meteor. Soc., 88, 311–317. doi: 10.1175/BAMS-88-3-311Cornillon, P. and K.-A. Park. 2001. Warm core ring velocity inferred from NSCAT. Geophys. Res.Lett., 28, 575–578. doi: 10.1029/2000GL0114872017] Pullen et al.: Coupled ocean-atmosphere forecasting 913Cox, M. D. 1980. Generation and propagation of 30-day waves in a numerical model of thePacific. J. Phys. Oceanogr., 10, 1168–1186 doi: 10.1175/1520-0485(1980)010<1168:GAPODW>2.0.CO;2Cummings, J. A. 2005. Operational multivariate ocean data assimilation. Q. J. R. Meteorol. Soc., 131,3583–3604. doi: 10.1256/qj.05.105Cummings, J. A. and O. M. Smedstad. 2013. Variational data assimilation for the global ocean, inData Assimilation for Atmospheric, Oceanic and Hydrologic Applications, vol. II. S. K. Park andL. Xu, eds. Berlin: Springer-Verlag, pp. 303–343Daley, R. and E. Barker. 2000. The NAVDAS Source Book, Publication NRL/PJ/7530-01-441. NRL,Monterey, CA: Naval Research Laboratory, 163 pp.Davis, C. A., and L. F. Bosart. 2004. The TT problem: Forecasting the tropical transition of cyclones.Bull. Amer. Meteor. Soc., 85(11), 1657–1662. doi: 10.1175/BAMS-85-11-1657DeMott, C. A., N. P. Klingaman, and S. J. Woolnough. 2015. Atmosphere-ocean coupled processesin the Madden-Julian oscillation. Rev. Geophys., 53(4), 1099–1154. doi: 10.1002/2014RG000478de Szoeke, S. P., S.-P. Xie, T. Miyama, K. J. Richards, and R. J. O. Small. 2007. What maintains theSST front north of the equatorial cold tongue? J. Clim., 20, 2500–2514. doi: 10.1175/JCLI4173.1Doyle, J. D., R. M. Hodur, S. Chen, Y. Jin, J. R. Moskaitis, S. Wang, E. A. Hendricks et al. 2014.Tropical cyclone prediction using COAMPS-TC. Oceanography 27(3), 104–115. doi: 10.5670/oceanog.2014.72Doyle, J. D., Q. Jiang, Y. Chao, and J. Farrara. 2009. High-resolution real-time modeling of the marineatmospheric boundary layer in support of the AOSN-II field campaign. Deep Sea Res. Part II Top.Stud. Oceanogr., 56(3), 87–99. doi: 10.1016/j.dsr2.2008.08.009Dudhia, J. 1989. Numerical study of convection observed during the winter monsoon experimentusing a mesoscale two-dimensional model. J. Atmos. Sci., 46, 3077–3107. doi: 10.1175/1520-0469(1989)046<3077:NSOCOD>2.0.CO;2Eden, C. and H. Dietze. 2009. Effects of mesoscale eddy/wind interactions on biological new pro-duction and eddy kinetic energy, J. Geophys. Res., 114, C05023. doi: 10.1029/2008JC005129Fairall, C. W., E. F. Bradley, J. E. Hare, A. A. Grachev, and J. B. Edson. 2003. Bulk parameterizationof air–sea fluxes: Updates and verification for the COARE algorithm. J. Clim., 16, 571–591 doi:10.1175/1520-0442(2003)016<0571:BPOASF>2.0.CO;2Fairall, C. W., E. F. Bradley, D. P. Rogers, J. B. Edson, and G. S. Young. 1996. Bulk parameteriza-tion of air–sea fluxes for tropical ocean-global atmosphere Coupled-Ocean Atmosphere ResponseExperiment. J. Geophys. Res., 101, 3747–3764. doi: 10.1029/95JC03205Fairall, C. W., J. D. Kepert, and G. J. Holland. 1994. The effect of sea spray on surface energytransports over the ocean. Global Atmos. Ocean Syst., 2(2–3), 121–142.Fallmann, J., H. Lewis, J. Castillo, A. Arnold, and S. Ramsdale. 2017. Impact of sea surface tempera-ture on stratiform cloud formation over the North Sea. Geophys. Res. Lett., 44(9) 4296–4303. doi:10.1002/2017GL073105Flather, R. A. 1976. A tidal model of the north-west European continental shelf. Memoires de laSociete Royale des Sciences de Liege, 6(10), 141–164.Flatau, M., P. J. Flatau, P. Phoebus, and P. P. Niiler. 1997. The feedback between equatorial convec-tion and local radiative and evaporative processes: The implication for intraseasonal oscillations.J. Atmos. Sci., 54, 2373–2386. doi: 10.1175/1520-0469(1997)054<2373:TFBECA>2.0.CO;2Frolov, D. S., C. H. Bishop, D. T. Holt, D. J. Cummings, and D. D. Kuhl. 2015. Facilitating strongly-coupled ocean-atmosphere data assimilation with an interface solver. Mon. Weather Rev., 144,3–20. doi: 10.1175/MWR-D-15-0041.1914 The Sea: The Science of Ocean Prediction [75, 6Gaube, P., D. B. Chelton, R. M. Samelson, M. G. Schlax, and L. W. O’Neill. 2015. Satelliteobservations of mesoscale eddy-induced Ekman pumping. J. Phys. Oceanogr., 45, 104–132. doi:10.1175/JPO-D-14-0032.1Gaube, P., D. J. McGillicuddy, Jr., D. B. Chelton, M. J. Behrenfeld, and P. G. Strutton. 2014. Regionalvariations in the influence of mesoscale eddies on near-surface chlorophyll. J. Geophys. Res.:Oceans, 119, 8195–8220. doi: 10.1002/2014JC010111Grubišic´, V., J. Sachsperger, and R. M. Caldeira. 2015. Atmospheric wake of Madeira: First aerialobservations and numerical simulations. J. Atmos. Sci., 72(12), 4755. doi: 10.1175/JAS-D-14-0251.1Hafner, J. and S. P. Xie. 2003. Far-field simulation of the Hawaiian wake: Sea surface tem-perature and orographic effects. J. Atmos. Sci., 60(24), 3021–3032. doi: 10.1175/1520-0469(2003)060<3021:FSOTHW>2.0.CO;2Hansen, D. and C. A. Paul. 1984. Genesis and effects of long waves in the equatorial Pacific. J. Geo-phys. Res., 89, 10431–10440. doi: 10.1029/JC089iC06p10431Hashizume, H., S.-P. Xie, M. Fujiwara, M. Shiotani, T. Watanabe, Y. Tanimoto, W. T. Liu, and K.Takeuchi. 2002. Direct observations of atmospheric boundary layer response to SST variationsassociated with tropical instability waves over the eastern equatorial Pacific. J. Clim., 15, 3379–3393 doi: 10.1175/1520-0442(2002)015<3379:DOOABL>2.0.CO;2Hayes, S. P., M. J. McPhaden, and J. M. Wallace. 1989. The influence of sea surface temperature onsurface wind in the eastern equatorial Pacific: Weekly to monthly variability. J. Clim., 2, 1500–1506doi: 10.1175/1520-0442(1989)002<1500:TIOSST>2.0.CO;2Hendon, H. H., B. Liebmann, M. Newman, J. D. Glick, and J. Schemm. 2000. Medium range forecasterrors associated with active episodes of the Madden–Julian oscillation. Mon. Weather Rev., 128,69–86. doi: 10.1175/1520-0493(2000)128<0069:MRFEAW>2.0.CO;2Hodur, R. M. 1997. The Naval Research Laboratory’s Coupled Ocean/Atmospheric MesoscalePrediction System (COAMPS). Mon. Weather Rev., 125, 414–430. doi: 10.1175/1520-0493(1997)125<1414:TNRLSC>2.0.CO;2Hodur, R., J. Pullen, J. Cummings, X. Hong, J. D. Doyle, P. Martin, and M. Rennick. 2002. TheCoupled Ocean/Atmosphere Mesoscale Prediction System (COAMPS). Oceanography, 15(1), 88–98. doi: 10.5670/oceanog.2002.39Hogan, T. F., M. Liu, J. A. Ridout, M. S. Peng, T. R. Whitcomb, B.C. Ruston, C.A. Reynoldset al. 2014. The Navy global environmental model. Oceanography, 27(3), 116–125. doi:10.5670/oceanog.2014.73Hong, S.-Y., Y. Noh, and J. Dudhia. 2006. A new vertical diffusion package with explicit treatmentof entrainment processes. Mon. Weather Rev., 134, 2318–2341. doi: 10.1175/MWR3199.1Hong, S.-Y and H.-L. Pan. 1996. Nonlocal boundary layer vertical diffusion in a medium-range forecast model. Mon. Weather Rev., 124, 1215–1238. doi: 10.1175/1520-0493(1996)124<2322:NBLVDI>2.0.CO;2Houze, R. A. 1997. Stratiform precipitation in regions of convection: A meteorological paradox?Bull. Amer. Meteor. Soc., 78, 2179–2196. doi: 10.1175/1520-0477(1997)078<2179:SPIROC>2.0.CO;2Houze, R. A., S. G. Geotis, F. D. Marks, Jr., and A. K. West. 1981. Winter monsoon convection in thevicinity of North Borneo. Part I: Structure and time variation of the clouds and precipitation. Mon.Weather Rev., 109, 1595–1614. doi: 10.1175/1520-0493(1981)109<1595:WMCITV>2.0.CO;2Howden, S. D. and R. Murtugudde. 2001. Effects of river inputs into the Bay of Bengal. J. Geophys.Res., 106(C9), 19825–19843. doi: 10.1029/2000JC0006562017] Pullen et al.: Coupled ocean-atmosphere forecasting 915Hu, C., E. T. Montgomery, R. W, Schmitt, and F.E. Muller-Karger. 2004. The dispersal of the Ama-zon and Orinoco River water in the tropical Atlantic and Caribbean Sea: Observation from spaceand S-PALACE floats. Deep Sea Res. Part II Top. Stud. Oceanogr., 51(10), 1151–1171. doi:10.1016/j.dsr2.2004.04.001Hung, M.-P., J.-L. Lin, W. Wang, D. Kim, T. Shinoda, and S. J. Weaver. 2013. MJO and convectivelycoupled equatorial waves simulated by CMIP5 climate models. J. Clim., 26, 6185–6214. doi:10.1175/JCLI-D-12-00541.1Hunke, E. C. and M. M. Holland. 2007. Global atmospheric forcing data for Arctic ice-ocean modeling,J. Geophys. Res., 112, C04S14. doi: 10.1029/2006JC003640Jacob, R., J. Larson, and E. Ong. 2005. MxN communication and parallel interpolation in CCSM3using the model coupling toolkit. Int. J. High Perform. Comput. Appl., 19(3), 293–307. doi:10.1177/1094342005056116Janjic, Z. I. 1990. The step-mountain coordinate: Physical package. Mon. Weather Rev., 118(7),1429–1443. doi: 10.1175/1520-0493(1990)118<1429:TSMCPP>2.0.CO;2Janjic, Z. I. 1994. The step-mountain eta coordinate model: Further developments of the convec-tion, viscous sublayer, and turbulence closure schemes. Mon. Weather Rev., 122, 927–945, doi:10.1175/1520-0493(1994)122<0927:TSMECM>2.0.CO;2Janjic, Z. I. 1996. The surface layer in the NCEP Eta model, in Proceedings of the 11th Conferenceon Numerical Weather Prediction, Norfolk, VA, 19–23 August 1996; Boston: American Meteoro-logical Society, pp. 354–355.Janjic, Z. I. 2002. Nonsingular implementation of the Mellor-Yamada level 2.5 scheme in the NCEPMeso Model. NCEP Office Note 437. Camp Springs, MD: National Centers for EnvironmentalPrediction, 61 pp.Jensen, T. G., T. Shinoda, S. Chen, and M. Flatau. 2015. Ocean response to CINDY/DYNAMO MJOsin air-sea-coupled COAMPS. J. Meteor. Soc. Japan, 93A, 157–178, doi: 10.2151/jmsj.2015-049Jochum, M., P. Malanotte-Rizzoli, and A. Busalacchi. 2004. Tropical instability waves in the AtlanticOcean. Ocean. Model., 7, 145–163. doi: 10.1016/S1463-5003(03)00042-8Jochum, M. and R. Murtugudde. 2006. Temperature advection by tropical instability waves. J. Phys.Oceanogr., 36, 592–605. doi: 10.1175/JPO2870.1Kantha, L. H. and C. A. Clayson. 2004. On the effect of surface gravity waves on mixing in the oceanicmixed layer. Ocean Model., 6, 101–124. doi: 10.1016/S1463-5003(02)00062-8Kelly, K. A., S. Dickinson, M. J. McPhaden, and G. C. Johnson. 2001. Ocean currents evident insatellite wind data. Geophys. Res. Lett., 28, 2469–2472. doi: 10.1029/2000GL012610Kilpatrick, T., S.-P. Xie, and T. Nasuno. 2017. Diurnal convection-wind coupling in the Bay of Bengal.J. Geophys. Res.: Atmos., Submitted. doi: 10.1002/2017JD027271Kuzmic´, M., I. Janekovic´, J. W. Book, P. J. Martin, and J. D. Doyle. 2006. Modeling the northernAdriatic doublegyre response to intense Bora wind: A revisit. Journal of Geophysical Research:Oceans, 111(C3).Larson, J., R. Jacob, and E. Ong. 2005. The Model Coupling Toolkit: A new Fortran90 toolkit forbuilding multiphysics parallel coupled models. Int. J. High Perform. Comput. Appl., 19(3), 277–292. doi: 10.1177/1094342005056115Lee, C. Y. and S. S. Chen. 2012. Symmetric and asymmetric structures of hurricane boundary layerin coupled atmosphere–wave–ocean models and observations. J. Atmos. Sci., 69(12), 3576–3594.doi: 10.1175/JAS-D-12-046.1916 The Sea: The Science of Ocean Prediction [75, 6Lin, J.-L., G. N. Kiladis, B. E. Mapes, K. M. Weickmann, K. R. Sperber, W. Lin, M. C. Wheeleret al. 2006. Tropical intraseasonal variability in 14 IPCC AR4 climate models. Part I: Convectivesignals. J. Clim., 19, 2665–2690. doi: 10.1175/JCLI3735.1Liu, W. T, K. B. Katsaros, and J. A. Businger, 1979. Bulk parameterization of the air-sea exchangeof heat and water vapor including the molecular constraints at the interface. J. Atmos. Sci., 36,1722–1735.Love, B. S., A. J. Matthews, and G. M. S. Lister. 2011. The diurnal cycle of precipitation over themaritime continent in a high-resolution atmospheric model. Q. J. R. Meteorol. Soc., 137, 934–947.doi: 10.1002/qj.809Lu, F., Z. Liu, S. Zhang, and Y. Liu. 2015. Strongly Coupled Data Assimilation Using LeadingAveraged Coupled Covariance (LACC). Part I: Simple model study. Mon. Weather Rev., 143,3823–3837. doi: 10.1175/MWR-D-14-00322.1Madden, R. A. and P. R. Julian. 1971. Detection of a 40–50 day oscillation in the zonal wind inthe tropical Pacific. J. Atmos. Sci., 28(5), 702–708. doi: 10.1175/1520-0469(1971)028<0702:DOADOI>2.0.CO;2Madden, R. A. and P. R. Julian. 1994. Observations of the 40–50-day tropical oscillation—A review. Mon. Weather Rev., 122(5), 814–837. doi: 10.1175/1520-0493(1994)122<0814:OOTDTO>2.0.CO;2Mapes, B. E., T. T. Warner, and M. Xu. 2003. Diurnal patterns of rainfall in northwestern SouthAmerica. Part III: Diurnal gravity waves and nocturnal convection offshore. Mon. Weather Rev.,131, 830–844. doi: 10.1175/1520-0493(2003)131<0830:DPORIN>2.0.CO;2Marchesiello, P., J. C. McWilliams, and A. Shchepetkin. 2001. Open boundary conditions for long-term integration of regional oceanic models. Ocean Model., 3(1), 1–20. doi: 10.1016/S1463-5003(00)00013-5Martin, P. J., 2000. A Description of the Navy Coastal Ocean Model Version 1.0. NRL ReportNRL/FR/7322–00-9962. Stennis Space Center, MS: Oceanography Division, Naval Research Lab-oratory, 42 pp.McTaggart-Cowan, R., L. F. Bosart, C. A. Davis, E. H. Atallah, J. R. Gyakum, and K. A. Emanuel.2006. Analyses of Hurricane Catarina. Mon. Weather Rev., 134, 3029–3053. doi: 10.1175/MWR3330.1Menkes, C. E., J. G. Vialard, S. C. Kennan, J. P. Boulanger, and G. V. Madec. 2006. A modeling ofthe impact of tropical instability waves on the heat budget of the eastern equatorial Pacific. J. Phys.Oceanogr., 36, 847–865. doi: 10.1175/JPO2904.1Metzger, E. J., O. M. Smedstad, P. G. Thoppil, H. E. Hurlburt, J. A. Cummings, A. J. Wallcraft,L. Zamudio et al. 2014. US Navy operational global ocean and Arctic ice prediction systems.Oceanography, 27(3), 32–43. doi: 10.5670/oceanog.2014.66Mlawer, E. J. and S. A. Clough. 1997. On the extension of rapid radiative transfer model to theshortwave region, in Proceedings of the 6th Atmospheric Radiation Measurement (ARM) ScienceTeam Meeting, US Department of Energy, CONF-9603149. Washington, DC: US Department ofEnergy.Mori S., J. I. Hamada, Y. I. Tauhid, M. D. Yamanaka, N. Okamato, F. Murata, N. Sakurai et al. 2004.Diurnal land-sea rainfall peak migration over Sumatera Island, Indonesian Maritime Continent,observed by TRMM satellite and intensive rawinsonde soundings. Mon. Weather Rev., 132, 2021–2039. doi: 10.1175/1520-0493(2004)132<2021:DLRPMO>2.0.CO;2Mujumdar, M., K. Salunke, S. A. Rao, M. Ravichandran, and B. N. Goswami. 2011. Diurnalcycle induced amplification of sea surface temperature intraseasonal oscillations over the Bay2017] Pullen et al.: Coupled ocean-atmosphere forecasting 917of Bengal in summer monsoon season. IEEE Trans. Geosci. Remote Sens., 8(2), 206–210. doi:10.1109/LGRS.2010.2060183National Research Council. 2016. Developing a U.S. research agenda to advance subseasonal toseasonal forecasting. Washington, DC: National Academies Press, 372 pp.Neale, R. B., J. H. Richter, and M. Jochum. 2008. The impact of convection on ENSO: From a delayedoscillator to a series of events. J. Clim., 21, 5904–5924. doi: 10.1175/2008JCLI2244.1Pacanowski, R. C. 1987. Effect of equatorial currents on surface stress. J. Phys. Oceanogr., 17(6),833–838. doi: 10.1175/1520-0485(1987)017<0833:EOECOS>2.0.CO;2Park, K.-A., P. Cornillon, and D. L. Codiga. 2006. Modification of surface winds near ocean fronts:Effects of Gulf Stream rings on scatterometer (QuikSCAT, NSCAT) wind observations. J. Geophys.Res., 111, C03021. doi: 10.1029/2005JC003016Peatman, S. C., A. J. Matthews, and D. P. Stevens. 2014. Propagation of the Madden–Julian Oscillationthrough the Maritime Continent and scale interaction with the diurnal cycle of precipitation. Q. J.R. Meteorol. Soc., 140, 814–825. doi: 10.1002/qj.2161Perlin, N., E. D. Skyllingstad, R. M., Samelson, and P. L. Barbour. 2007. Numerical simula-tion of air-sea coupling during coastal upwelling. J. Phys. Oceanogr., 37(8), 2081–2093. doi:10.1175/JPO3104.1Pezzi, L. P., R. B. de Souza, O. Acevedo, I. Wainer, M. M. Mata, C. A. Garcia, and R. de Camargo. 2009.Multiyear measurements of the oceanic and atmospheric boundary layers at the Brazil-Malvinasconfluence region. J. Geophys. Res.: Atmos., 114(D19), D19103. doi: 10.1029/2008JD011379Pezzi, L. P., R. B. Souza, M. S. Dourado, C. A. E. Garcia, M. M. Mata, and M. A. F. Silva-Dias. 2005.Ocean-atmosphere in situ observations at the Brazil-Malvinas confluence region. Geophys. Res.Lett., 32, L22603. doi: 10.1029/2005GL023866Pezzi, L. P., J. Vialard, K. J. Richards, C. Menkes, and D. Anderson. 2004. Influence of ocean atmo-sphere coupling on the properties of tropical instability waves. Geophys. Res. Lett., 31, L16306.doi: 10.1029/2004GL019995Philander, S. G. H. 1976. Instabilities of zonal equatorial currents. J. Geophys. Res., 81, 3725–3735.doi: 10.1029/JC081i021p03725Philander, S. G. H. 1978. Instabilities of zonal equatorial currents, 2. J. Geophys. Res., 83, 3679–3682doi: 10.1029/JC083iC07p03679Pullen, J., R. Caldeira, J. D. Doyle, P. May, and R. Tomé. 2017. Modeling the air-sea feedback systemof Madeira Island. J. Adv. Model. Earth. Syst., in revision. doi: 10.1002/2016MS000861Pullen, J., J.D. Doyle, T. Haack, C. Dorman, R. Signell, and C. M. Lee. 2007. Bora event variabilityand the role of air-sea feedback. J. Geophys. Res., 112, C03S18. doi: 10.1029/2006JC003726Pullen, J., J. D. Doyle, and R.P. Signell. 2006. Two-way air-sea coupling: A study of the Adriatic.Mon. Weather Rev., 134(5), 1465–1483. doi: 10.1175/MWR3137.1Pullen, J. A., L. Gordon, J. Sprintall, C. M. Lee, M. A. Alford, J. D. Doyle, and P. W. May. 2011.Atmospheric and oceanic processes in the vicinity of an island strait. Oceanography, 24(1), 112–121. doi: 10.5670/oceanog.2011.08Qiao, L. and R. H. Weisberg. 1995. Tropical instability wave kinematics: Observations from the Trop-ical Instability Wave Experiment. J. Phys. Oceanogr., 100, 8677–8693 doi: 10.1029/95JC00305Ray, P., C. Zhang, J. Dudhia, and S. S. Chen. 2009. A numerical case study on the initiation of theMadden–Julian oscillation. J. Atmos. Sci., 66, 310–331. doi: 10.1175/2008JAS2701.1Ray, P., C. Zhang, M. W. Moncrieff, J. Dudhia, J. M. Caron, L. R. Leung, and C. Bruyere. 2011.Role of the atmospheric mean state on the initiation of the Madden-Julian Oscillation in a tropicalchannel model. Clim. Dyn., 36, 161–184. doi: 10.1007/s00382-010-0859-2918 The Sea: The Science of Ocean Prediction [75, 6Renault, L., M. J. Molemaker, J. Gula, S. Masson, and J. C. McWilliams. 2016a. Control and sta-bilization of the Gulf Stream by oceanic current interaction with the atmosphere. J. Clim., 46,3439–3453. doi: 10.1175/JPO-D-16-0115.1Renault, L., M. J. Molemaker, J. C. McWilliams, A. F. Shchepetkin, F. Lemaré, D. Chelton, S. Illig,and A. Hall. 2016b. Modulation of wind work by oceanic current interaction with the atmosphere.J. Phys. Oceanogr., 46, 1685–1704. doi: 10.1175/JPO-D-15-0232.1Ricchi, A., M. M. Miglietta, P. P. Falco, A. Benetazzo, D. Bonaldo, A. Bergamasco, M. Sclavo, andS. Carniel. 2016. On the use of a coupled ocean–atmosphere–wave model during an extreme coldair outbreak over the Adriatic Sea. Atmos. Res. 172, 48–65. doi: 10.1016/j.atmosres.2015.12.023Rogers, W. E., J. D. Dykes, and P. A. Wittmann. 2014. US Navy global and regional wave modeling.Oceanography, 27(3), 56–67. doi: 10.5670/oceanog.2014.68Rotunno, R. and K. Emanuel. 1987. An air–sea interaction theory for tropical cyclones. Part II:Evolutionary study using a nonhydrostatic axisymmetric numerical model. J. Atmos. Sci., 44,542–561. doi: 10.1175/1520-0469(1987)044<0542:AAITFT>2.0.CO;2Rowley, C. and A. Mask. 2014. Regional and Coastal Prediction with the Relocat-able Ocean Nowcast/Forecast System. Oceanography, 27(3), 44–55. JSTOR, JSTOR,www.jstor.org/stable/24862188Saleh, F., V. Ramaswamy, N. Georgas, A. F. Blumberg, and J. Pullen. 2016. A retrospective streamflowensemble forecast for an extreme hydrologic event: a case study of Hurricane Irene and on theHudson River basin. Hydrology and Earth System Sciences, 20(7), 2649–2667.Sangrà, P., M. Auladell, A. Marrero-Díaz, J. L. Pelegrí, E. Fraile-Nuez, A. Rodríguez-Santana, J. M.Martín et al. 2007: On the nature of oceanic eddies shed by the Island of Gran Canaria. Deep SeaRes. Part II Top. Stud. Oceanogr., 54(5), 687–709. doi: 10.1016/j.dsr.2007.02.004Seo, H. 2017. Distinct influence of air-sea interactions mediated by mesoscale sea surface temperatureand surface current in the Arabian Sea. J. Clim., 30, 8061–8080. doi: 10.1175/JCLI-D-16-0834.1Seo, H., M. Jochum, R. Murtugudde, and A. J. Miller. 2006. Effect of ocean mesoscale vari-ability on the mean state of tropical Atlantic climate. Geoyphs. Res. Lett., 33, L09606. doi:10.1029/2005GL025651Seo, H., M. Jochum, R. Murtugudde, A. J. Miller, and J. O. Roads. 2007. Feedback of tropicalinstability wave-induced atmospheric variability onto the ocean. J. Clim., 20, 5842–5855. doi:10.1175/JCLI4330.1Seo, H., M. Jochum, R. Murtugudde, A. J. Miller, and J. O. Roads. 2008a. Precipitation fromAfrican easterly waves in a coupled model of the tropical Atlantic. J. Clim., 21, 1417–1431.doi: 10.1175/2007JCLI1906.1Seo, H., A. J. Miller, and J. R. Norris. 2016. Eddy-wind interaction in the California current system:Dynamics and impacts. J. Phys. Oceanogr., 46, 439–459. doi: 10.1175/JPO-D-15-0086.1Seo, H., A. J. Miller, and J. O. Roads. 2007. The Scripps Coupled Ocean-Atmosphere Regional(SCOAR) model, with applications in the eastern Pacific sector. J. Clim., 20, 381–402. doi:10.1175/JCLI4016.1Seo, H., R. Murtugudde, M. Jochum, and A. J. Miller. 2008b. Modeling of mesoscale coupled ocean-atmosphere interaction and its feedback to ocean in the western Arabian Sea. Ocean Model., 25,120–131. doi: 10.1016/j.ocemod.2008.07.003Seo, H., A. C. Subramanian, A. J. Miller, and N. R. Cavanaugh. 2014. Coupled impacts of the diurnalcycle of sea surface temperature on the Madden-Julian oscillation. J. Clim., 27, 8422–8443 doi:10.1175/JCLI-D-14-00141.12017] Pullen et al.: Coupled ocean-atmosphere forecasting 919Seo, H. and S.-P. Xie, 2011. Response and impact of equatorial ocean dynamics and tropical insta-bility waves in the tropical Atlantic under global warming: A regional coupled downscaling study.J. Geophys. Res.: Oceans, 116, C03026. doi: 10.1029/2010JC006670Seo, H. and S.-P. Xie. 2013. Impact of ocean warm layer thickness on the intensity of HurricaneKatrina in a regional coupled model. Meteorol. Atmos. Phys., 122, 19–32. doi: 10.1007/s00703-013-0275-3Seo, H., S.-P. Xie, R. Murtugudde, M. Jochum, and A. J. Miller. 2009. Seasonal effects ofIndian Ocean freshwater forcing in a regional coupled model. J. Clim., 22, 6577–6596. doi:10.1175/2009JCLI2990.1Shinoda, T. 2005. Impact of diurnal cycle of solar radiation on intraseasonal SST variability in thewestern equatorial Pacific. J. Clim., 18, 2628–2636. doi: 10.1175/JCLI3432.1Shinoda, T., T. Jensen, M. Flatau, and S. Chen. 2013. Surface wind and upper ocean variability asso-ciated with the Madden-Julian oscillation simulated by the coupled ocean/atmosphere mesoscaleprediction system (COAMPS). Mon. Weather Rev., 141, 2290–2307. doi: 10.1175/MWR-D-12-00273.1Small, R. J., T. Campbell, J. Teixeira, T. A. Smith, J. Dykes, S. Chen, and R. Allard, 2011. Air-seainteraction in the Ligurian Sea: Assessment of a coupled ocean-atmosphere model using in situdata from LASIE07. Mon. Weather Rev., 139, 1785–1808. doi: 10.1175/2010MWR3431.1Small, R. J., K. J. Richards, S.-P. Xie, P. Dutrieux, and T. Miyama. 2009. Damping of tropicalinstability waves caused by the action of surface currents on stress. J. Geophys. Res., 114, C04009.doi: 10.1029/2008JC005147Small, R. J., S. P. de Szoeke, S. P. Xie, L. O’Neill, H. Seo, Q. Song, P. Cornillon et al. 2008.Air–sea interaction over ocean fronts and eddies. Dyn. Atmos. Oceans, 45, 274–319. doi:10.1016/j.dynatmoce.2008.01.001Smith, T. A., S. Chen, T. Campbell, P. Martin, W. E. Rogers, S. Gaberšek, D. Wang et al. 2013. Ocean-wave coupled modeling in COAMPS-TC: A study of Hurricane Ivan (2004). Ocean Model., 69,181–194. doi: 10.1016/j.ocemod.2013.06.003Smith, R. B. and V. Grubišic. 1993. Aerial observations of Hawaii’s wake. J. Atmos. Sci., 50(22),3728–3750. doi: 10.1175/1520-0469(1993)050<3728:AOOHW>2.0.CO;2Smith, G. C., F. Roy, and B. Brasnett. 2013. Evaluation of an operational ice–ocean analysis andforecasting system for the Gulf of St Lawrence. Q. J. R. Meteorol. Soc., 139(671), 419–433. doi:10.1002/qj.1982Stern, M. 1965. Interaction of a uniform wind stress with a geostrophic vortex. Deep-Sea Res.Oceanogr. Abstr., 12, 355–367. doi: 10.1016/0011-7471(65)90007-0Subramanian, A. C., M. Jochum, A. J. Miller, R. Murtugudde, R. B. Neale, and D. E. Waliser. 2011. TheMadden Julian oscillation in CCSM4. J. Clim., 24, 6261–6282. doi: 10.1175/JCLI-D-11-00031.1SWAN Team. 2010. SWAN Scientific and Technical Documentation, SWAN Cycle III Version40.81. Delft, the Netherlands: Delft University of Technology, 118 pp. Available: http://www.swan.tudelft.nlSweet, W., R. Fett, J. Kerling, and P. La Violette. 1981. Air–sea interaction effects in the lowertroposphere across the north wall of the Gulf Stream. Mon. Weather Rev., 109, 1042–1052. doi:10.1175/1520-0493(1981)109<1042:ASIEIT>2.0.CO;2Taylor, P. K. and M. J. Yelland. 2001. The dependence of sea surface roughness on theheight and steepness of the waves. J. Phys. Oceanogr., 31(2), 572–590. doi: 10.1175/1520-0485(2001)031<0572:TDOSSR>2.0.CO;2920 The Sea: The Science of Ocean Prediction [75, 6Teague, W. J., E. Larosz, D. W. Wang, and D. A. Mitchell. 2007. Observed oceanic response over theupper continental slope and outer shelf during Hurricane Ivan. J. Phys. Oceanogr., 37, 2181–2206.doi: 10.1175/JPO3115.1Theurich, G., C. DeLuca, T. Campbell, F. Liu, K. Saint, M. Vertenstein, J. Chen et al. 2016. The EarthSystem Prediction Suite: Toward a coordinated U.S. modeling capability. Bull. Amer. Meteorol.Soc., 97(7), 1229–1247. doi: 10.1175/BAMS-D-14-00164.1Tolman, H. L. 2009. User Manual and System Documentation of WAVEWATCH IIITM, Version3.14. Technical Note, NOAA/NWS/NCEP/MMAB. Camp Springs, MD: U.S. Dept. of Commerce,220 pp.Veeramony, J., M. D. Orzech, K. L. Edwards, M. Gilligan, J. Choi, E. Terrill, and T. DePaolo. 2014. Navy nearshore ocean prediction systems. Oceanography, 27(3), 80–91. doi:10.5670/oceanog.2014.70Vialard, J. and P. Delecluse. 1998. An OGCM study for the TOGA decade. Part II: Barrier-layerformation and variability. J. Phys. Oceanogr., 28, 1089–1106. doi: 10.1175/1520-0485(1998)028<1089:AOSFTT>2.0.CO;2Vianna, M. L., V. V. Menezes, A. B. Pezza, and I. Simmonds. 2010. Interactions between HurricaneCatarina (2004) and warm core rings in the South Atlantic Ocean. J. Geophys. Res., 115, C07002.doi: 10.1029/2009JC005974Vitart, F. and F. Molteni. 2010. Simulation of the Madden–Julian oscillation and its teleconnectionsin the ECMWF forecast system. Q. J. R. Meteorol. Soc., 136, 842–855.Waliser, D. E., R. Murtugudde, and L. E. Lucas. 2003. Indo-Pacific Ocean response to atmosphericintraseasonal variability: 1. Austral summer and the Madden-Julian Oscillation. J. Geophys. Res.,108, 3160. doi: 10.1029/2002JC001620Waliser, D. E., K. Lau, and J. H. Kim. 1999. The influence of coupled sea surface temperatures onthe Madden-Julian oscillation: A model perturbation experiment. J. Atmos. Sci., 56, 333–358. doi:10.1175/1520-0469(1999)056<0333:TIOCSS>2.0.CO;2Wallace, J. M., T. P. Mitchell, and C. Deser. 1989. The influence of sea surface temperature on surfacewind in the eastern equatorial Pacific: Seasonal and interannual variability. J. Clim., 2, 1492–1499.doi: 10.1175/1520-0442(1989)002<1492:TIOSST>2.0.CO;2Warner, J. C., B. Armstrong, R. He, and J. B. Zambon. 2010. Development of a Coupled Ocean-Atmosphere-Wave-Sediment Transport (COAWST) modeling system. Ocean Model., 35(3), 230–244. doi: 10.1016/j.ocemod.2010.07.010Warner, J. C., C. R. Sherwood, H. G. Arango, and R. P. Signell. 2005. Performance of four turbulenceclosure models implemented using a generic length scale method. Ocean Model., 8(1), 81–113.doi: 10.1016/j.ocemod.2003.12.003Weller, R. A., J. T. Farrar, J. Buckley, S. Mathew, R. Venkatesan, J. Sree Lekha, D. Chaudhurietal. 2016. Air-sea interaction in the Bay of Bengal. Oceanography, 29(2), 28–37. doi: 10.5670/oceanog.2016.36Woolnough, S. J., F. Vitart, and M. Balmaseda. 2007. The role of the ocean in the Madden–Julian Oscil-lation: Implications for MJO prediction. Q. J. R. Meteorol. Soc., 133, 117–128. doi: 10.1002/qj.4Xie, S.-P. 2004. Satellite observations of cool ocean–atmosphere interaction. Bull. Amer. Meteor.Soc., 85, 195–209. doi: 10.1175/BAMS-85-2-195Xie, S.-P., T. Miyama, Y. Wang, H. Xu, S. P. de Szoeke, R. J. Small, K. J. Richards et al. 2007.A regional ocean-atmosphere model for eastern Pacific climate: Towards reducing tropical biases.J. Clim., 20, 1504–1522. doi: 10.1175/JCLI4080.12017] Pullen et al.: Coupled ocean-atmosphere forecasting 921Yu, Z. and J. P. McCreary. 2004. Assessing precipitation products in the Indian Ocean using an oceanmodel. J. Geophys. Res.: Oceans, 109, C05013. doi: 10.1029/2003JC002106Yu, Z., J. P. McCreary, Jr., and J. A. Proehl. 1995. Meridional asymmetry and energetics of tropicalinstability waves. J. Phys. Oceanogr., 25, 2997–300 doi: 10.1175/1520-0485(1995)025<2997:MAAEOT>2.0.CO;2Zhang, C., 2005. Madden–Julian oscillation. Rev. Geophys., 43, RG2003. doi: 10.1029/2004RG000158Zhang, C., M. Dong, S. Gualdi, H. H. Hendon, E. D. Maloney, A. Marshall, K. R. Sperber, and W.Wang. 2006. Simulations of the Madden–Julian oscillation by four pairs of coupled and uncoupledglobal models. Clim. Dyn., 27, 573–592. doi: 10.1007/s00382-006-0148-2Received: 5 October 2015; revised: 20 June 2017.Editor’s note: Contributions to The Sea: The Science of Ocean Prediction are being published sepa-rately in special issues of Journal of Marine Research and will be made available in a forthcomingsupplement as Volume 17 of the series.

The Coupled Ocean/Atmosphere Mesoscale

Abstract

Extracted data

The Coupled Ocean/Atmosphere Mesoscale

Abstract

Extracted data

Related items

Related items