Show Summary Details

Page of

Printed from Oxford Research Encyclopedias, Climate Science. Under the terms of the licence agreement, an individual user may print out a single article for personal use (for details see Privacy Policy and Legal Notice).

date: 02 July 2022

Regional Climate Modeling and Air-Sea Couplingfree

Regional Climate Modeling and Air-Sea Couplingfree

  • Corinna SchrumCorinna SchrumHelmholtz-Zentrum Geesthacht, Institute of Coastal Research


Regional models were originally developed to serve weather forecasting and regional process studies. Typical simulations encompass time periods in the order of days or weeks. Thereafter regional models were also used more and more as regional climate models for longer integrations and climate change downscaling. Regional climate modeling or regional dynamic downscaling, which are used interchangeably, developed as its own branch in climate research since the end of the 1990s out of the need to bridge the obvious inconsistencies at the interface of global climate research and climate impact research. The primary aim of regional downscaling is to provide consistent regional climate change scenarios with relevant spatial resolution to serve detailed climate impact assessments.

Similar to global climate modeling, the early attempts at regional climate modeling were based on uncoupled atmospheric models or stand-alone ocean models, an approach that is still maintained as the most common on the regional scale. However, this approach has some fundamental limitations, since regional air-sea interaction remains unresolved and regional feedbacks are neglected. This is crucial when assessing climate change impacts in the coastal zone or the regional marine environment. To overcome these limitations, regional climate modeling is currently in a transition from uncoupled regional models into coupled atmosphere-ocean models, leading to fully integrated earth system models. Coupled ice-ocean-atmosphere models have been developed during the last decade and are currently robust and well established on the regional scale. Their added value has been demonstrated for regional climate modeling in marine regions, and the importance of regional air-sea interaction became obvious. Coupled atmosphere-ice-ocean models, but also coupled physical-biogeochemical modeling approaches are increasingly used for the marine realm. First attempts to couple these two approaches together with land surface models are underway. Physical coupled atmosphere-ocean modeling is also developing further and first model configurations resolving wave effects at the atmosphere-ocean interface are now available. These new developments now open up for improved regional assessment under broad consideration of local feedbacks and interactions between the regional atmosphere, cryosphere, hydrosphere, and biosphere.


  • Modeling

Coupled air-sea models are today state-of-the-art in global climate research and increasingly used for regional climate modeling. The general aim of this article is to provide an overview of the field of regional coupled air-sea modeling. After a general introduction, the historic development of the field is briefly summarized and model concepts are introduced. Thereafter, major methodological issues and recent advancements are discussed and key scientific challenges are highlighted. The relevance of air-sea interaction is briefly discussed for selected exemplary case studies and state-of-the-art in the field is explored. Regional coupled air-sea models are increasingly used for climate change downscaling to regional systems. The recent progress here is reviewed and key challenges and the added value of regional coupled downscaling vs. stand-alone atmosphere or ocean downscaling are discussed. Finally an outlook on remaining challenges and future developments is given.

Introduction to the Field

The primary aim of regional modeling was to study the dynamics and feedbacks in regional systems and the need for regional forecasts on timescales of days to weeks resolving finer details. The regional forecast models were thereafter also used in climate research on timescales of months or seasons to years to decades and regional climate modeling developed as its own research field in climate research since the end of the 1990s out of the need to bridge the obvious inconsistencies at the interface of global climate research and the requirement for sufficient details in climate impact research. Eventually, the field became crucial to serve societal needs in assessment of regional climate change impacts, and regional climate models were employed as standard tools to provide consistent regional scenarios with relevant spatial resolution for detailed climate change assessment.

Similar to global climate modeling, the early attempts at regional climate modeling were based on uncoupled atmospheric models and stand-alone ocean models. The uncoupled models were forced by atmospheric boundary conditions (ocean models) or sea surface boundary conditions (atmospheric models). Uncoupled stand-alone atmosphere or ocean modeling is still the most common approach on the regional scale. It forms, for example, the standard approach for the WRCP1-coordinated regional downscaling experiment (CORDEX, Giorgi et al., 2009), which aims at resolving climate change impacts for regional terrestrial systems utilizing uncoupled regional atmospheric models. Likewise also most climate change assessments for regional marine systems are still based on uncoupled regional ocean models (e.g., Chust et al., 2014; BACC2: Meier, 2015; NOSCCA3: Schrum et al., 2016).

Albeit the use of stand-alone regional atmospheric models can be considered appropriate for many terrestrial environments, as it was shown, for example, by Gröger et al. (2015) for Europe, this is not the case for marine environments. Air-sea interactions are particularly crucial to resolve regional coupled air-sea modes such as El Niño Southern Oscillation (ENSO) (Neelin et al., 1992), for regional monsoon systems (Ratnam et al., 2009; Zou & Zhou, 2013) and appear to be relevant at the sea ice edge (Rinke et al., 2003). They influence moreover local land-sea circulation and have a substantial potential to modify local coastal weather conditions (e.g., Gröger et al., 2015; Li et al., 2014a). Air-sea interactions are moreover central for extreme marine weather phenomena like hurricanes or typhoons (e.g., Bender et al., 1993) and improved understanding and modeling of air-sea interactions in these systems remains critical for short-term prediction and future climate change risk assessment (e.g., Bender et al., 2010; Knutson et al., 2013). Important regional and local air-sea feedbacks, which are not resolved in global climate models, are neglected in stand-alone ocean or atmospheric models of the coastal zone and future climate projections might be fundamentally flawed. To overcome these limitations coupled air-sea modeling is increasingly used on a regional scale. A substantial number of regional coupled ocean-atmosphere models were developed during the last decade, and coupled air-sea modeling is currently robust and well established on the regional scale. The added value of coupled air-sea models has been demonstrated in coastal and marine regions (e.g., Gröger et al., 2015; Seo et al., 2007a) and the importance to consider regional air-sea interaction became obvious. Albeit the recent progress in the field, the use of coupled air-sea modeling for regional climate impact studies is still not standard and first coupled assessments are available only for a few regions (Bülow et al., 2014; Gualdi et al., 2013; Klein et al., 2014; Koenigk et al., 2011; Meier, 2006, 2015; Somot et al., 2008). However, the cost-to-improvement ratio is not necessarily large and a future challenge in coupled regional modeling lies in identifying the cases of added value of regional coupled modeling.

Historic Development

The development of coupled air-sea climate modeling started on the global scale in the 1980s (e.g., Charnock & Philander, 1989; Meehl, 1990; Nihoul, 1985; Nihoul & Jamart, 1990). The early attempts typically suffered from inconsistencies in the heat, momentum, and freshwater fluxes and the drift of the coupled model systems appeared to be a problem, which was typically addressed through flux correction (Sausen et al., 1988). Different classes of models were developed, fully coupled General Circulation Models (GCMs) in both systems, atmosphere and ocean, or so-called hybrid models with a GCM coupled to a simplified model with reduced physics (see Neelin et al., 1992, for a representative summary and classification of these early attempts). A large part of these early attempts to model coupled air-sea interaction has focused on tropical air-sea interaction (Neelin et al., 1992). ENSO, the most prominent air-sea interaction in the tropics, served as a benchmark to test these early developments and Neelin et al. (1992) first presented a framework to classify the response of the climate models to the coupled air-sea interactions in the tropics.

First regional air-sea models were developed only in the end of the 1990s with the primary purpose to improve weather forecasting and the prediction of extreme events (e.g., Bender et al., 1993; Gustafsson et al., 1998; Hodur, 1997; Xue et al., 2000). These range from hybrid models involving simplified model components in atmosphere or ocean (e.g., Bender et al., 1993; Gustafsson et al., 1998) to 2D models (Xue et al., 2000) and fully complex 3D GCMs (Hodur, 1997). Bender et al. (1993) coupled a movable mesh atmospheric hurricane model to a 3D ocean model and modeled firstly air-sea interaction for a moving hurricane using an idealized setup. The first coupled regional 3-d atmosphere-ocean model was developed and applied for idealized studies of a tropical cyclone and for a realistic forecast application in connection to the America’s Cup races (Hodur, 1997). Hodur (1997) modeled realistically the temperature feedback and ocean surface cooling due to the development of a tropical cyclone, although the modeled cooling was somewhat larger than observed. Challenges in initialization of the models remained and Hodur (1997) recommended increased observations and the use of extended data assimilation as a prerequisite for realistic forecasts of tropical cyclones. Xue et al. (2000) presented a 2D coupled atmosphere-ocean model for the Gulf Stream region. They found significant atmospheric feedback to the ocean during cold air outbreaks. Li et al. (2002) extended the approach further into 3D and illustrated the model’s potential to study extra-tropical cyclone development.

Gustafsson et al. (1998) presented for the first time a coupling approach encompassing also sea ice. They coupled a low-resolution ice-ocean box model to a high-resolution atmospheric model for the Baltic Sea. The model needed control of the ocean state through data assimilation, and dynamic full coupling was not yet achieved. However, Gustafsson et al. (1998) could show that weather forecasts were significantly improved by using the coupled model approach. Improvements were in particular achieved for cold air outbreaks, mesoscale secondary atmospheric circulation systems, and convective snowbands.

The first coupled models were typically used for short-term integrations up to a few weeks, but Gustafsson et al. (1998) hypothesized that air-sea coupling on the regional scale is likely also important for regional climate studies and suggests to use coupled models to derive better estimation for water and energy balances at the regional scale. BALTEX,4 which was launched as a regional contribution to the WCRP GEWEX5 boosts thereafter the development in regional coupled air-sea modeling. In addition to the hybrid model presented by Gustafsson et al. (1998), full regional coupled 3D air-sea models became available for the Baltic region. Hagedorn et al. (2000) presented a 3D coupled atmosphere-ocean model. In this coupled system the air-sea fluxes were calculated from the atmospheric model and the sea surface temperature (SST) was provided from the ocean model. Special care was taken to conserve energy while interpolation of the fluxes to the higher resolution ocean grid taking into account also the different staggered grids and differences in the land/ocean mask. The model was successfully integrated over a couple of months and they succeeded in reproducing observed SSTs sufficiently well without flux correction and no obvious drift occurred during the autumn cooling period, which was used for a first test simulation. Hagedorn et al. (2000) speculated that the good performance of the coupled model was caused jointly by a well-performing stand-alone ocean model component together with the strong constraints at the lateral boundaries of the regional model. This model did not consider sea ice, so Hagedorn et al. (2000) were limited in simulation period and were unable to test the performance for a full seasonal cycle.

The first 3D two-way coupled regional atmosphere-ocean model, which also incorporated sea ice dynamics was presented by Schrum et al. (2001, 2003) for the North Sea and Baltic Sea region. For the coupling a full flux scheme for thermal and fresh water fluxes was chosen, that is, the air-sea fluxes were calculated by the atmospheric model and transferred to the ocean model every 6 hours. In return, the ocean model provides SST and sea ice boundary conditions to the atmospheric model using a daily time step. Flux corrections were not required, similar to the coupled model presented by Hagedorn et al. (2000). The model operated stably over a full seasonal cycle and a significant improvement of simulated surface ocean conditions were achieved compared to a one-way passive coupling using forcing from a regional atmospheric model forced by ocean boundary condition from reanalysis (Schrum et al., 2003). The simulated regional ocean model results in the coupled run appeared to be of similar quality as the ocean model simulations forced by reanalysis data and Schrum et al. (2003) concluded that the coupling is stabilizing the regional model and speculated that the coupled model provides an alternative to data assimilation in data poor conditions and thereby opens up for improved forecasts and regional climate change projections.

Shortly after a full resolution atmosphere-ice-ocean model for the purpose of regional climate studies was presented by Döscher et al. (2002). They used a smaller ocean region only comprising the Baltic Sea. A similar but more frequent air-sea coupling was used compared to Schrum et al. (2001, 2003), exchanging the fluxes and surface boundary conditions every 3 hours. Döscher et al. (2002) found that the coupled regional model runs stably over several years without drift. The performance of the coupled models was good, for the ocean variables it was comparable to the performance of the stand-alone ocean model simulations forced by observations. For most surface ocean quantities the performance of the ocean model in the coupled simulation was similar to the performance of the stand-alone ocean model simulation forced by observational-based atmospheric data, except for sea ice. The maximum sea ice extend was somewhat overestimated by the coupled model compared to the stand-alone ocean model, a somewhat contrasting result to Schrum et al. (2003). Döscher et al. (2002) suspected uncertainties in the latent heat flux to be responsible for the weaker performance in projecting sea ice.

The development of coupled regional models for the Arctic started shortly after. Rinke et al. (2003) presented a coupling of a regional atmospheric model and a regional ocean model for the Arctic. The regional model domains chosen for this model were of similar size in both atmosphere and ocean but not entirely overlapping. A different approach using a global ocean model equipped with a high-resolution zoom coupled only for the Arctic to a regional atmospheric model was chosen by Mikolajewicz et al. (2005) (see Fig. 1 for comparable configurations). The model was stable without flux corrections in momentum and heat fluxes during the multi-decadal simulations, but had to use a salinity restoring (Mikolajewicz et al., 2005). A similar configuration was used by Aldrian et al. (2005) to resolve the role of regional air-sea coupling in Indonesian rainfall. For the Mediterranean, a contrasting concept was developed, consisting of a regional ocean model and a global atmospheric model (Somot et al., 2008). In this coupled model a flux correction for heat fluxes was used to ensure model stability (Somot et al., 2008).

Parallel to the development in Europe and the United States, coupled regional modeling also developed in China (Peng et al., 2012). This started out with 2D approaches such as the one presented by Wang et al. (1995, in Chinese, cited after Peng et al., 2012) and from 2000 onward further developed into 3D coupled air-sea models (e.g., Lü et al., 2000; Ren & Qian, 2000; all in Chinese and cited after Peng et al., 2012). For a more detailed reading about the Chinese developments in regional coupled modeling, the reader is referred to the recent review by Peng et al. (2012).

Coupling to Wave Models

A particular challenge in coupled air-sea modeling relates to the influence of waves on the air-sea energy and mass fluxes. The coupled air-sea models discussed so far considered an instantaneous balance between momentum transfer from the atmosphere to the ocean and energy removed from the wind field is directly transferred into energy gain of the surface currents. However, mechanical energy removed from the wind field is partly used to create sea surface wind waves, which are not resolved in regional ocean models. The local wave state, that is, whether the waves are growing or breaking or are in equilibrium, modulates the energy transfer to ocean currents (see for further reading Breivik et al., 2015, and references therein). Moreover, wave motion also adds an additional term to the ocean momentum equation, the Coriolis-Stokes force or Hasselmann force (Hasselmann, 1970; Jenkins, 1987; Weber, 1983) and enhances turbulence and mixing in the ocean through wave breaking (Craig & Banner, 1994), wave-current interaction and non-breaking waves (e.g., Breivik et al., 2015; Fan & Griffies, 2014; and references therein). The debate how to formulate atmosphere-wave and wave-ocean interaction is still not settled and the theoretical framework is currently still under development (e.g., Aiki & Greatbatch, 2014; Chen et al., 2013; Moon et al., 2007). Spatial and temporal changes in the wave field, propagation of wave energy, and wave-current interaction remain usually unconsidered for momentum and enthalpy transfer across the sea surface in regional atmosphere and ocean models. Breaking waves furthermore produce sea spray, which modulates the sea surface enthalpy transfer and the near surface atmosphere. The latter is particularly relevant under high wind conditions (e.g., Andreas, 1992; Bao et al., 2000; Kepert et al., 1999). The dynamic coupling to wave models is particularly important near the coast and under extreme weather conditions (e.g., Staneva et al., 2016) but has substantial effects also on mean and offshore energy transfer (e.g., Breivik et al., 2015).

No attempt is so far made to consider a full ocean-atmosphere-wave coupling taking into account all mentioned effects together. First attempts to consider wave effects in air-sea coupling concentrated on the impacts of wave induced turbulence. Ly (1995) studied momentum transfer and turbulence at the air-sea interface using a consistent formulated model approach for atmosphere and ocean. Bao et al. (2000) presented a coupling approach using well-developed stand-alone atmosphere, ocean, and wave models. Here they considered the roughness length at the sea surface to be proportional to the total stress at the sea surface, but independent of the sea state. Bao et al. (2000) also considered the effect of sea spray modulated enthalpy flux and used the coupled model to study the surface wave impact on hurricane development. However, wave-current interaction through the Coriolis-Stokes force and wave-induced turbulence remain unconsidered. Fan et al. (2009) considered spatiotemporal sea state variations and wave-current interaction, but the Coriolis-Stokes force and the enthalpy flux modulations due to wave effects remain unconsidered. Fan and Griffies (2014) presented a coupled global ocean-wave-atmosphere model to study the impacts of parameterized turbulence effects due to Langmuir turbulence and non-breaking wave turbulence testing different parameterizations. Fan and Griffies did not finally conclude the importance of these effects for the climate; however, their results indicate that the interaction processes are complex and potentially important for the climate system. Curcic et al. (2016) studied the surface transport and dispersion during a hurricane situation using a coupled atmosphere-wave-current model, but neglected the Coriolis-Stokes force and vortex forces. A more comprehensive approach to modeling wave effects in the coupled global atmosphere-ocean system was recently presented by Breivik et al. (2015). They considered wave-induced modification of surface stress, sea-state dependent turbulent kinetic energy flux from breaking waves and the Coriolis-Stokes force, but non-breaking wave turbulence and sea spray effects were neglected. Breivik et al. (2015) found all three effects noticeable in the extratropics, but the sea-state dependent turbulent energy flux was by far the most important impact. They found moreover a significant reduction of the SST bias in the coupled system and a much better performance of the coupled model when compared to heat content estimates from an ocean reanalysis.

Methodological Challenges and Advancements

Technically three different methods have been used to interactively couple regional stand-alone state-of-the-art atmosphere and ocean models. One of the models can be directly integrated as a module or subroutine into the code of the other one (e.g., as done by Hagedorn et al., 2000). The advantage of this coupling method is a consistent simulation code, which is time efficient since communication is handled within one executable process. However, a disadvantage is the relatively large programming effort of this coupling method. Connecting to new developments of stand-alone models remains difficult, since consideration of an advanced version for the atmosphere or ocean model components requires again similar programming efforts. An alternative coupling method was used, for example, by Schrum et al. (2001); here both models run in parallel in a configuration similar to their stand-alone applications and communication between the atmospheric and ocean modules is realized via input and output and interprocess pipeline communication. While this method can be implemented relatively fast, it is less general and requires additional specific interpolation routines. Moreover, the coupling is less efficient, and the coupled models have to wait for each other and for completion of the slow and frequent in- and output operations. The latter is getting increasingly important with more frequent communication between the modules. Both methods were used in the beginning of regional coupled model development. However, their inefficiency and lack of flexibility made them less attractive. Today the most often used coupling method is the use of standard couplers, which provide efficient and flexible software code to optimize the coupling procedure and coordinate the different components of the coupled models. Versatile couplers have been developed from the 1990s onward and a number of different couplers are currently available and in use by the modeling community (see Peng et al., 2012, for an overview about commonly used couplers). The couplers are flexible due to modular design and have facilitated the coupling of atmosphere-ocean-sea ice models greatly. However, still they are more computationally demanding than the coupling via subroutine interfaces (Rockel, 2015).

Characteristic time and spatial scales in ocean and atmosphere differ largely and the ocean typically requires a much higher resolution to resolve similar mesoscale features (e.g., Peng et al., 2012) and ocean models therefore often have a much higher resolution than atmospheric models. Air-sea fluxes in the first coupled models were typically calculated by the atmospheric component on a coarse spatial resolution, while SST was provided in high resolution through the ocean model. A general challenge is arising from the different grids and grid resolution in atmosphere and ocean (e.g., as discussed by Hagedorn et al., 2000). Fluxes calculated in higher resolution do not match the course resolution fluxes calculated from averaged fields, which is most relevant for long-wave radiation heat losses, since these change with 4th order dependency on temperature as discussed by Döscher et al. (2002). Due to the different schemes and grids used in calculating fluxes, conservation of energy and mass might not be ensured (e.g., Döscher et al., 2002; Schrum et al., 2003). Energy conservation in state-of-the-art coupled models is today accomplished through modern couplers, which calculate fluxes preponderantly on both grids and make adjustments afterward to ensure energy conservation (e.g., Valcke, 2013; see more detailed discussion in Peng et al., 2012).

The very first attempts to regional coupled modeling had a challenge ensuring model skill, regardless whether they were so-called hybrid models involving simplified system components (Gustafsson et al., 1998) or employed full resolution 3D models in both atmosphere and ocean (Hodur, 1997). Low performance in the coupled systems gave reason for data assimilation to prevent a model to drift away from realistic conditions (Gustafsson et al., 1998; Hodur, 1997). Inconsistencies in flux formulations (Gustafsson et al., 1998) and challenges in initialization of the ocean model (Hodur, 1997) were suspected to be the cause for these drift problems. Later many fully coupled 3D regional models run stable without flux correction or data assimilation and initialization problems were solved (e.g., Döscher et al., 2002; Hagedorn et al., 2000; Schrum et al., 2003). Regional coupled modeling developed thereafter into a standard approach and a large number of 3D coupled air-sea models became available for various regions (Döscher et al., 2010; Fang et al., 2009; Ho et al., 2012; Ho-Hagemann et al., 2015; Li & Zhou, 2010; Lin et al., 2006; Pullen et al., 2006; Ren & Qian, 2005; Seo et al., 2007a). These studies have demonstrated the potential of coupled models to investigate air-sea interaction, and many studies found enhanced skill for the coupled models compared to uncoupled setups and drift problems remain minor in many regional coupled models (e.g., Pullen et al., 2006; Schrum et al., 2003; Seo et al., 2007a). However, this is not the case for all models and configurations presented (e.g., Ren & Qian, 2005; Sein et al., 2014; Small et al., 2011). More detailed investigations revealed that the setup of the coupled model is essential for its performance. Both the location and extension of the coupled region (Sein et al., 2014), the coupling frequency (Fang et al., 2009), and the quality of initialization and boundary forcing appeared to be critical (Wei et al., 2014).

If properly initialized and calibrated, regional models are strongly constrained through lateral boundary conditions, which seem to prevent regional models from unrealistic drift and internal free oscillations. Drift and stability problems are therefore typically less in coupled regional models compared to global climate models (e.g., many CMIP3 models still used flux correction; Flato et al., 2013). However, this is not the case if regional models extend over large domains or if global model components are involved in the coupling (e.g., Mikolajewicz et al., 2005; Somot et al., 2008). Mikolajewicz and colleagues (2005) discussed the challenges for the global mass and energy balance arising from coupling between a global model and a regional model. An inflow to the coupled region is calculated based on global forcing outside the coupling area and an outflow is calculated based on the regionally coupled model; the regional outflow might therefore be inconsistent with the global setup and a significant global climate signal might arise from the regionally coupled domain (Mikolajewicz et al., 2005). These inconsistencies can lead to a substantial drift and changes of the global model climate and often restoring or flux correction is used to handle these issues (e.g., Mikolajewicz et al., 2005; Somot et al., 2008). However, drift problems seem to be of minor relevance when the coupled region is smaller as it was demonstrated for a coupled regional-global model setup for the Indonesian region (Aldrian et al., 2005).

Mikolajewicz and colleagues (2005) found that their regionally coupled model was sensitive to small disturbances in forcing. While the averaged response to disturbances was only small compared to the main signal for most investigated parameters, this was not the case on a more local scale. Mikolajewicz and colleagues (2005) concluded therefore that ensemble simulations are required to provide reliable model results on regional and local scale. The importance and need for ensemble simulations is expected to increase with larger model domains, since these have increased degrees of freedom (Mikolajewicz et al., 2005). The related problem of changing relevance of internally vs. externally forced variability in regionally coupled modeling as a function of the size and location of the coupling domain was systematically investigated by Sein et al. (2014; Fig. 1).

Figure 1. MPIOM/REMO/HD Coupling (a) and coupled setups (b). Colored spherical rectangles indicate different regional atmospheric model domains and the coupling area for the coupled Arctic model configurations tested by Sein et al. (2014).

Figure by Sein et al. (2014), licensed under Creative Commons CC-BY 4.0.

They studied the role of the choice of the model domain for the regional atmospheric model for the performance of the coupled model in configurations similar to the one presented by Mikolajewicz et al. (2005). Sein et al. (2014) found that the regional model was able to stay in phase with observations if the leading climate mode is externally forced rather than subject to internal variability. Sein and colleagues (2014) found that this could only be ensured for their coupled Arctic model if the North Pacific was excluded from the coupling domain. They found that the inclusion of the North Pacific drastically changed the Arctic variability. The Arctic oscillation became an internal free mode of the regional model in this case and the coupled model was not anymore in phase with observed variability. The correlation to observations on a year-to-year base vanished and the skill of the model dropped. The findings of Sein et al. (2014) were consistent with earlier results published by Diaconescu and Laprise (2013), who found that the ratio between internal to external (and hence prescribed) variability changes with changing domain size in a regional atmospheric model. The careful and problem-based choice of the model domain remains therefore crucial. Sein and colleagues (2014) pointed out that the choice of the coupled model domain based on simple geographical arguments is not sufficient and decisions should be based on fundamental understanding of oceanographic and atmospheric processes and their feedbacks (Sein et al., 2014).

Relevance of Air-Sea Coupling

The progress in 3D coupled model development and their increased availability allows now for comprehensive and quantitative assessment of the degree of regional air-sea coupling and has also led to improved understanding about the role of air-sea interaction in different regional systems. It is impossible to provide a full overview about the topic in the frame of this contribution. Instead, some examples are presented in the following to illustrate the broad ranges of topics, which were addressed so far.

Forecasting of extreme weather events in the marine realm are among the first topics addressed with regional coupled models. Tropical cyclones and hurricanes are classical examples of weather systems involving strong air-sea interaction and considerable positive and negative feedbacks. During genesis positive feedbacks contribute to cyclone development: Increased wind speed leads to increased evaporation rates and increasing latent heat energy, which in turn leads to further increase in wind speeds (Bender et al., 1993). With further intensification of the storm, however, the system turns into a negative feedback system: Intensified turbulent mixing at the sea surface generates mixed layer deepening and surface cooling. This cooling in turn results into a reduction of surface heat and water fluxes and consequently, a reduction of storm intensity (for further reading, see Bender et al., 1993). Ocean-atmosphere feedbacks were modeled for idealized setups using coupled air-sea models since the beginning of the 1990s (e.g., Bender et al., 1993; Hodur, 1997). Later on, more realistic coupled model setups were used to investigate the dynamics of specific cyclones (e.g., Bao et al., 2000; Bender & Ginis, 2000; Seo & Xie, 2013). These studies jointly indicated substantial air-sea feedbacks, enhancing or decreasing the cyclone intensities. The intensity of the interaction strongly depends on the propagation speed and direction of the cyclones, which in turn is influenced by the regional atmospheric structure and only slightly influenced by the ocean (e.g., Bao et al., 2000; Hodur, 1997).

Modern hurricane research and forecasting models are typically coupled 3D air-sea models, which utilize data assimilation to further improve the models skill (e.g., Bernardet et al., 2015). Although a lot of progress has been made in modeling these high energetic systems, scientific challenges remain. Recent studies by Lynn et al. (2015) indicated that aerosols can play a significant role in reducing the hurricane’s intensity. Zambon and colleagues (2014) found significant improvements of modeling the intensity and moderate improvements in modeling the track of tropical cyclones when using full complexity of the coupled atmosphere-wave-ocean physics. They achieved a fairly accurate simulation of the coupled ocean-atmosphere system without additional benefit of data assimilation, but left unconsidered the effects of dissipative heating and sea spray, which earlier have been identified to play a significant role in hurricane intensities (Bao et al., 2011; Jin et al., 2007). The integration of all relevant effects with full coupling to wave dynamics remains as future task for research and forecasting models for tropical cyclones.

Air-sea interaction in the tropics forms another important area of application for coupled air-sea models. A prominent example is the large archipelago between the Indian and Pacific Oceans, the Maritime Continent. The Maritime Continent is an energetically important area of the climate system and regional convective activity contributes to the Walker Circulation (Aldrian et al., 2005; Neale & Slingo, 2003). Variations in monsoon regimes and convective activity are tightly related to large-scale climate variations such as ENSO and the Indian Ocean Dipole (IOD) (Chou et al., 2003; Neale & Slingo, 2003). The development of first coupled regional models for the Maritime Continent started during the last decade with the regionally coupled model by Aldrian et al. (2005). Aldrian and colleagues (2005) identified significant regional air-sea interaction and skill enhancement in simulating Indonesian rainfall through consideration of air-sea coupling. A number of other coupled models were developed for the Maritime Continent thereafter (Fang et al., 2009; Li & Zhou, 2010; Ren & Qian, 2005; and Zou & Zhou, 2012), but not all of them covered the entire Maritime Continent. Many of these models had a problem with a cold bias and climate drift; these problems were attributed to uncertainties in the heat fluxes and convective mixing in the atmospheric model (see discussion by Wei et al., 2014). A recent high-resolution coupled model for the entire Maritime Continent, presented by Wei et al. (2014), in contrast, was able to run stably without drift and showed significant improvement in SST evolution compared to uncoupled simulations. In contrast to an earlier study by Aldrian et al. (2005), they did not find any improvement in simulating precipitation through regional air-sea coupling. The reason for these somewhat contrasting results remains unclear and Wei et al. (2014) suggests further research on local air-sea feedbacks to assess the role of local air-sea coupling for rainfall. Such local feedbacks have been identified earlier to be important for the Indian monsoon (Krishna et al., 2005; Wang et al., 2005). These authors found significant improvement for statistical relation patterns between SST and rainfall compared to a tiered-two approach forcing the atmospheric model through observed SSTs presupposing the Indian monsoon as a purely forcing-response system. Their conclusion, that the Indian monsoon is a coupled atmosphere-ocean system called for reshaping of the seasonal monsoon prediction strategies and suggested the use of coupled monsoon prediction models (Wang et al., 2005). The degree of regional air-sea coupling in the Western North Pacific for a summer monsoon was investigated by Zou and Zhou (2013). They found skill enhancement when using the coupled model compared to uncoupled modeling. Moreover, they found, that the use of uncoupled modeling could lead to false conclusions about the driving role of the atmosphere vs. the ocean. While in coupled simulations the ocean appears to be a slave of the atmosphere, the opposite was the case in uncoupled simulations.

Air-sea interactions are substantial also in other marine and coastal regions and a large number of regional studies using coupled air-sea models have contributed to assess and quantify the role of air-sea interactions for various sites. Seo et al. (2007a, 2007b) used a coupled atmosphere-ocean model to study the role of air-sea interaction for tropical instabilities in the eastern tropical Pacific, for mesoscale eddies in the California current system and for gap winds along the central American coast. They identified thermodynamic and dynamic feedbacks in all three systems (Seo et al., 2007a). More detailed investigation of the air-sea interaction in the case of tropical instabilities identified the need to consider air-sea feedbacks to predict a realistic damping of the development of tropical instabilities through adjustment to local SSTs (Seo et al., 2007b). A particular strong feedback mechanism was induced to the wind stress through local ocean currents. This effect could alter the local wind stress by about ± 25%–30% (Seo et al., 2007b). However, in annual mean the effect was negligible due to oscillatory behaviour (Seo et al., 2007b).

Seo and colleagues (2007a) suggested systematic modeling studies comparing fully and partly coupled models to uncoupled models to develop optimal regional modeling strategies for different regional phenomena and research questions. In a number of further studies using the same coupled atmosphere-ocean models they contributed further understanding of regional air-sea interaction and highlighted the need for applying coupled regional atmosphere-ocean models to improve understanding and modeling of regional coastal and marine systems (e.g., Putrasahan et al., 2013; Seo et al., 2008, 2009, 2011, 2014; Seo & Xie, 2011, 2013). Atmosphere-ocean feedback was also identified for coastal upwelling (Ribeiro et al., 2011; Seo et al., 2008). The general pattern of upwelling could be reproduced with uncoupled models, but Ribeiro et al. (2011) found enhanced upwelling due to the sea-breeze circulation and more realistic profiles of ocean mixing ratio and vertical wind while using the coupled model. Seo and colleagues (2008) assigned a considerable contribution from the coupled air-sea feedback to Ekman pumping.

Other research groups complemented these studies for the Gulf Stream region (Li et al., 2002; Xue et al., 2000), the Mediterranean (Dell’Aquila et al., 2012; Herrmann et al., 2011; Lebeaupin Brossier et al., 2015; Sanna et al., 2013), and Adriatic Sea (Pullen et al., 2006), Southern Africa (Ratnam et al., 2015), the North Sea/Baltic Sea region (Gröger et al., 2015; Ho et al., 2012; Ho-Hagemann et al., 2015; Lorenz & Jacob, 2014; Sein et al., 2015; Tian et al., 2013; Van Pham et al., 2014) the Arctic region (Döscher et al., 2010; Mikolajewicz et al., 2005; Rinke et al., 2003; Sein et al., 2014) and for the Southern Ocean (Byrne et al., 2015). It is beyond the scope of this contribution to discuss all findings in great detail, and the reader is instead referred to the original papers for further reading. Most of these studies identified relevant air-sea interaction, and, if properly initialized and forced, skill could be improved while using coupled models (e.g., Gröger et al., 2015; Rinke et al., 2003; Sein et al., 2014).

Figure 2. Difference between interactively coupled simulation and passively coupled simulation for winter (upper, left) and summer (lower, left), all changes (left) and only changes significant at the 95% significance level (right).

Figure from Gröger et al. (2015), licensed under Creative Commons CC-BY 4.0.

The degree of air-sea coupling varies from system to system and also within a system regionally and seasonally. More detailed assessment for the Northern European region indicates that the impact of local air-sea interaction over the North Sea and Baltic Sea is mainly restricted to the marine and near coastal environment (Fig. 2). Only a small impact of interactive atmosphere-ocean coupling is seen over land, mostly restricted to extreme precipitation events (Figs. 2 and 3, Gröger et al., 2015; Ho-Hagemann et al., 2015; Tian et al., 2013). Contrasting results were achieved by Li et al. (2014a) for California; here they found rather far-reaching terrestrial impacts of local air-sea interaction, in particular during summer. Wang and colleagues (2015) pointed out that the potential of a coupled model to identify terrestrial impacts of regional air-sea interaction might be biased by too small coupling regions. This is potentially an issue for many coupled North and Baltic Sea models (Döscher et al., 2002; Gröger et al., 2015; Schrum et al., 2003; Tian et al., 2013; Wang et al., 2015), which typically disregard atmosphere-ocean interaction in the North Atlantic. The latter is certainly more influential than the North Sea and Baltic Sea, in particular for the Norwegian coast, which was unaffected by the coupling in the models by Gröger et al. (2015) and Wang et al. (2015) (Figs. 2 and 3).

Figure 3. Difference between interactively coupled and passively coupled simulations for strong precipitation above the 90% percentile (left all changes, right changes significant at the 95% confidence level).

Figure from Gröger et al. (2015), licensed under Creative Commons CC-BY 4.0.

Downscaling Climate Change Projections

The growing evidence of regional air-sea interaction and their relevance for the local coastal climate in various regions initiated increasing utilization of coupled atmosphere-ocean models also for downscaling the effects of climate change to regional marine and coastal systems. However, so far regional downscaling by coupled air-sea models still remains the exception in climate change downscaling and substantial progress is today only made for four regions: the North European Shelf including the North and Baltic Seas, the Mediterranean, the Tropical Ocean, and the Arctic.

The first use of 3D coupled regional models to downscale global climate change was reported for the Baltic Sea (Kjellström et al., 2005; Meier, 2006; Räisänen et al., 2004) well before regional coupled models were utilized systematically in climate change downscaling elsewhere. In recent years extended efforts were undertaken as part of a national German climate change assessment program to extend this approach into the North Sea region. Three different coupled atmosphere-ocean models have been developed: a global ocean model coupled to a regional atmospheric model (Sein et al., 2015), a regional North Sea model coupled to a regional atmospheric model (Su et al., 2014), and a regional North Sea/Baltic Sea model coupled to a regional atmospheric model (Gröger et al., 2015; Wang et al., 2015). These developments allow now for multimodel ensemble simulations using regionally coupled models with the first of these presented by Bülow et al. (2014) and Klein et al. (2014). The first coupled ensemble projections for the North Sea consists of downscalings from these three models, which all were forced by the same global climate model. This first ensemble was recently compared to a number of uncoupled downscaling studies using ocean-only models (Schrum et al., 2016). From these comparisons it was evident that the coupled ensemble had a somewhat larger spread in projected SST changes compared to intermodel spreads from different regional ocean models using the same forcing global models. However, the spread arising from the regional models is small compared to the spread arising from the forcing global models and the coupled projections lay within the uncertainty range from previous stand-alone downscaling to the North Sea marine system (Schrum et al., 2016). The coupled downscaling has so far not been compared to stand-alone regional atmospheric downscaling and the effect on regional atmospheric changes remains to be assessed.

The first atmosphere-regional-ocean climate change downscaling for the Mediterranean was presented by Somot et al. (2008). They showed, that Mediterranean coupled climate change projections are to the first order comparable to downscaling using uncoupled atmospheric models; however, they found a significant amplification of the climate change signal over Europe. This was attributed to the regionally poor resolution of the Mediterranean Sea in global climate models, which imprint their variability to the regional atmospheric model through the SST boundary condition in uncoupled models (Somot et al., 2008). The coupled approach was than further followed up and a number of coupled models were developed for the Mediterranean (Artale et al., 2009; Dubois et al., 2012; Nabat et al., 2014; Sevault et al., 2014). Ensemble projections were presented by Dubois et al. (2012) and further evaluated and reviewed by Gualdi et al. (2013) as part of a regional climate change assessment for the Mediterranean. The latter was based on an ensemble of five different regional coupled air-sea models. Further coordinated ensemble projections using CMIP5 scenarios are underway for the Mediterranean (Somot et al., 2014).

Despite the impressive effort in developing regional coupled models for the Mediterranean and despite the increased resolution of the Mediterranean region in these models compared to global climate models, still substantial errors remain and the intermodel spread is larger compared to uncoupled downscaling (Dubois et al., 2012; Gualdi et al., 2013). The regional coupled downscaling increases the degree of freedoms and adds uncertainties arising from different regional models to the projections rather than leading to converging projections (Gualdi et al., 2013). These findings by Gualdi et al. (2013) are consistent with the findings by Schrum et al. (2016) for the North Sea. While comparing to uncoupled downscaling with regional atmospheric models for the Mediterranean, Gualdi et al. (2013) found that the coupled regional projections are however in general agreement with those performed earlier with uncoupled models and lay and are within the uncertainty range from stand-alone projections. Gualdi and colleagues (2013) concluded therefore, that the regional coupling has no strong impact on the regional projected response to anthropogenic climate change for the Mediterranean. This conclusion is somewhat different to the first results by Somot et al. (2008) and Dell’Aquila et al. (2012) comparing coupled to uncoupled downscaling and global climate model results. Despite significant differences occur when comparing coupled to uncoupled projected future climate change from single models, these differences appear obviously to be within the range of intermodel differences arising from different forcing global and regional models.

Regionally coupled models have also been used to assess the climate change impact to various regions in the tropical ocean (e.g., Knutson et al., 2013; Li et al., 2014b; Seo & Xie, 2011). A central research question here is the future development of tropical cyclone intensity and frequency, which was earlier investigated by using both global climate models and regional atmosphere only models. The results from these studies were inconsistent and both more and fewer tropical cyclones were projected regionally (see review by Knutson et al., 2010). Downscaling climate change impacts to hurricane frequency and intensity is in later studies performed using regional coupled atmosphere-ocean models (Bender et al., 2010; Knutson et al., 2013). From the latter a robust and statistically significant decrease in tropical cyclone frequency in the Atlantic is evident.

A first coupled future climate change downscaling experiment was also carried out for the region of California and adjacent coastal regions (Li et al., 2014b). Li and colleagues (2014b) found that coupled and uncoupled downscaling project quite different regional temperature changes. Changes projected by the coupled model system were more moderate than changes projected with the uncoupled model, which appeared particularly relevant for the occurrence of extreme warm events. The differences in projected temperature change between coupled and uncoupled downscaling were significant with the exception of changes in wintertime. Projected differences in precipitation change in contrast were not significant, which was attributed to the fact that air-sea coupling in the region was strongest during summer time, which is the dry season in the California region. The experiment was short, only a 10-year-long integration was performed and only forcing by one global climate model was considered.

Several regionally coupled Arctic models have been developed so far (Döscher et al; 2010; Mikolajewicz et al., 2005; Sein et al., 2014) or are underway (Maslowski et al., 2012; Roberts, 2010), which are appropriate for regional downscaling of climate change. However, so far only few attempts have been undertaken to systematically use these coupled models to downscale climate change impact to the Arctic (Döscher & Koenigk, 2013; Koenigk et al., 2011). Systematic ensemble studies with coupled models need still to be carried out, and the added value of coupled vs. uncoupled downscaling still remains to be assessed for the Arctic.

Added Value and Future Challenges

Atmosphere and ocean exchange momentum, fresh water, and heat and their states and properties are tightly coupled on various scales. However, local air-sea interaction remains often unconsidered, when studying regional systems, despite the growing evidence on their importance for regional marine systems. Today, the typical regional climate model still consists of a regional atmospheric model and a land surface scheme, which is forced by sea surface boundary conditions over the ocean derived from observational-based analysis or global models (e.g., Giorgi et al., 2009; Rockel, 2015). Complementary, ocean hydrodynamics and their climate variations are typically modeled using stand-alone ocean models forced by often coarsely resolved atmospheric boundary conditions and river runoff. Value addition by regional models depends on the problem being studied (Feser et al., 2011; Rummukainen, 2016). Controversies exist about whether or not regional downscaling provides added value compared to coarser resolved reanalysis and global climate models, and there are a number of factors controlling the downscaling ability, among those the choice of boundary conditions and domain size (see for further reading Rummukainen et al., 2015; Xue et al., 2014). Coastal or marine applications furthermore require a careful case-to-case assessment to identify the suitability of stand-alone regional atmosphere or ocean models, which are computationally less expensive than coupled models and therefore often prioritized. The studies reviewed here suggest that using a stand-alone model might still be acceptable (e.g., Gröger et al., 2015; Wang et al., 2015) or even justified by better performance for some applications (e.g., Small et al., 2011). In particular, they appear sufficient to hindcast environmental conditions in atmosphere or ocean, insofar proper surface boundary forcing is ensured through high-quality observational-based products and under more moderate air-sea interaction (as demonstrated, e.g., by Döscher et al., 2002; Schrum et al., 2003; Sein et al., 2014; Small et al., 2011). However, they fail or perform less well if proper local boundary conditions are not available (e.g., Kjellström et al., 2005; Schrum et al., 2003; Yao & Zhang, 2009). The desired configuration of a regional model under such conditions requires therefore careful inspection and the degree of local air-sea coupling should be assessed (Seo et al., 2007a).

Coupled models typically outperform uncoupled models under extreme marine weather conditions and over the ocean (e.g., Aldrian et al., 2005; Bender & Ginis, 2000; Gröger et al., 2015; Lynn et al., 2015; Pullen et al., 2006; Seo et al. 2007a). For highly energetic phenomena, which are strongly controlled by air-sea interactions, the use of coupled atmosphere-ocean models appears therefore mandatory. For tropical cyclones, the use of coupled models is today already standard in research, forecast, and climate assessments (e.g., Bender et al., 2010; Bernadet et al., 2015; Knutson et al., 2013). The need for coupled regional models is less clear for seasonally or regionally weaker air-sea coupling and less energetic conditions (e.g., Döscher et al., 2002; Gröger et al., 2015; Small et al., 2011), and a careful choice of the modeling strategy is required to take into account the specific research question or application.

Couplings add additional degrees of freedom and the selection of the ideal model setup to balance internal variability and external constraints is not trivial and requires understanding of the regional dynamic system (Sein et al., 2014). If the model area is too large, internal modes of variability might get dominant and skill enhancement in coupled models could disappear or skill might even drop (Sein et al., 2014). On the other hand, the degree of regional air-sea coupling might be underestimated if the coupled area is too small, as discussed by Wang et al. (2015).

The choice of an ideal setup for a regional coupled system is to a certain degree also a philosophical question. The question to what extent a regional setup is allowed to overcome the inner domain dependency from the forcing parent global model and to generate its own upstream climate change signal needs to be further explored. Regional high-resolution GCMs (e.g., McClean et al., 2011) and coupling studies as the one by Sein et al. (2014) for different regions might bring new understanding to regional atmosphere-ocean coupling and help to carefully select the models regional setups to find an optimal configuration to allow regional air-sea interaction and on the other hand provide sufficient constrains to not distort the signal through uncontrolled and out of phase internal variability. Involving regionally coupled RCM-GCM models (e.g., Mikolajewicz et al., 2005; Sein et al., 2014, 2015) allows the study of regional atmosphere-ocean feedbacks to the global scale. However, it also makes computations expensive since long spin-up times of several hundreds of years and ensemble simulations might be inevitable (Mikolajewicz et al., 2005; Sein et al., 2014). Moreover, with these types of models it is difficult to accomplish regional multimodel ensembles resolving the full spectra of existing global climate projections. Their regular use in regional climate change assessments is therefore less sensible and the choice of coupled regional models with a regionally smaller setup (e.g., Wang et al., 2015) or even of uncoupled models (Chust et al., 2014; Pushpadas et al., 2015) might here be more appropriate and allows for resolving a broader range of global climate model scenarios.

In regional climate change assessments, the use of coupled atmosphere-ocean models is growing, but there is still not a general standard. Coupled high-resolution regional models remain computationally expensive and even with increased computational potentials, there is a need to carefully select an appropriate modeling strategy to balance the human and technical resources to solve given tasks in climate change assessment. Coupled systematic downscaling studies have been performed and compared to uncoupled climate change downscaling only for a few regions; among those are the North Sea and Baltic Sea region and the Mediterranean (e.g., Gualdi et al., 2013; Meier, 2015; Schrum et al., 2016). From studies comparing coupled and uncoupled regional downscaling using selected models it is evident that regional projections from regionally coupled models differ from downscaling using stand-alone models (Kjellström et al., 2005; Li et al., 2014b; Somot et al., 2008; Zou & Zhou, 2012). However when comparing ensemble studies performed with coupled models to huge ensembles performed with uncoupled models, the added value of coupled downscaling is less clear. The few regional climate change assessment utilizing regionally coupled models have neither shown generally different results, nor did they contribute to narrow down the large uncertainty range in projected regional future climate change (Gualdi et al., 2013; Meier, 2015; Schrum et al., 2016), except for the climate change studies of tropical cyclones, which appeared to be more consistent when using coupled models (Bender et al., 2010; Knutson et al., 2013). Forecasting and solid assessment of future climate change impacts for marine and coastal systems including the assessment of the added value from coupled modeling and identification of appropriate modeling strategies require mature coupled air-sea models. Despite recent progress regional downscaling in the marine realm is still in its infancy and an international community effort such as CORDEX for regional climate modeling of the land surface is required to enforce substantial progress and develop best practices and standards in coupled atmosphere-ocean downscaling and regional downscaling to the marine realm. Regional initiatives such as Med-CORDEX demonstrated capacity to advance regional coupled climate system models and led to substantial progress.

Figure 4. Exemplary flow chart for a regionally coupled air-sea model (Sein et al., 2015). In addition to the regional model for the atmosphere and the global model for ocean hydrodynamics and sea ice, further model components are incorporated for the marine biosphere and for surface and river runoff.

Figure from Sein et al. (2015), licensed under Creative Commons CC-BY 4.0.

So far regional coupled models include mainly the components atmosphere, land surface and hydrology, and the regional ocean (Ruti et al., 2014; Somot & Ruti, 2014). In the future, regional coupled atmosphere-ice-ocean models need to evolve into multicomponent regional system models, which also consider chemistry and biogeochemistry in atmosphere, ocean, and the land surface. Regional climate modeling is currently in a transition from uncoupled regional atmosphere or ocean modeling into fully coupled earth system modeling, replicating the earlier developments in global climate modeling (e.g., Arora et al., 2013; Breivik et al., 2015; Friedlingstein, 2006; Neelin et al., 1992). This process is currently only in initiation, but the first regional coupled atmosphere-ice-ocean-biogeochemistry models are already available (Drobinski et al., 2012; Park et al., 2014; Sein et al., 2015, Fig. 4). First attempts have also been undertaken to incorporate subscale spatiotemporal variations at the atmosphere-ocean interface through coupling with wave models (e.g., Bao et al., 2000; Breivik et al., 2015; Drobinski et al., 2012), but their consistent integration in regional climate change downscaling systems remain to be subject to future efforts. On the regional scale, coupling of atmosphere-ocean models through a wave model interface and consideration of wave induced fluxes of momentum, energy and enthalpy is not yet standard and remains unconsidered in most of the coupled atmosphere-ocean models discussed here.

For coastal ocean applications, also morphological and coastline changes (Le Conzannet et al., 2014; Woodroffe & Murray-Wallace, 2012) and groundwater seawater interaction (e.g., Alekseeva et al., 2009 have to be taken into account and incorporation of prognostic approaches to deal with interactions at the land-ocean transition is a coming challenge for regional modeling.

Further Reading