Regional earth system modeling: review and future directions

Abstract The authors review recent advances in the development of coupled Regional Earth System Models (RESMs), a field that is still in its early stages. To date, coupled regional atmosphere-ocean-sea ice, atmosphere-aerosol and atmosphere-biosphere models have been developed, but they have been applied only to limited regional settings. Much more work is thus needed to assess their transferability to a wide range of settings. Future challenges in regional climate modeling are identified, including the development of fully coupled RESMs encompassing not only atmosphere, ocean, cryosphere, biosphere, chemosphere, but also the human component in a fully interactive way.


Introduction
Since the pioneering work of Dickinson et al. (1989), Giorgi and Bates (1989) and Giorgi (1990), the field of regional climate modeling has tremendously grown. Today a number of regional climate models (RCMs) from laboratories around the world are used for a wide variety of applications, from process studies at the seasonal to interannual scale, to multicentennial climate projections, with resolutions varying from ~50 km to convection permitting (<5 km) (Giorgi and Gutowski 2015). A number of intercomparison projects, culminating into the international Coordinated Regional Downscaling EXperiment (CORDEX, Giorgi, Jones, and Asrar 2009;Jones, Giorgi, and Asrar 2011;Gutowski et al. 2016), have enabled the RCM community to explore key issues in regional modeling and assess the potential of this technique to produce climate information relevant for impact and adaptation studies. The evolution of regional modeling is discussed in a number of review papers (See Giorgi and Gutowski 2015, and papers cited therein) to which the reader is referred to have an overview of the status of this field of research.
Here we focus on what is considered to be one of the main future directions in RCM research, namely the development of interactively coupled Regional Earth System Models (RESM) (Giorgi and Gutowski 2015). The great opportunity represented by RCMs in describing the regional interactions across different components of the climate system was already recognized in the early stages of RCM development (Giorgi 1995). This is because many of these interactions take place at spatial scales that are not resolved by global models, and are more closely captured at the high resolution achievable with RCMs. Typical examples are the presence of complex vegetation structures, small river and ocean basins, lakes, tropospheric aerosols, mesoscale atmosphere, and ocean circulation features, etc.
Recognizing this opportunity, the RCM modeling community has actively engaged in the development of RESM systems usable in different regional contexts. Today several coupled RESMs exist, including varying sets of components, which have been applied to a wide range of different regions (e.g. Peng et al. 2012;Giorgi and Gutowski 2015;Schrum 2017). Some models, in particular, include multiple Earth system components, such as atmosphere, oceans, sea ice, hydrology, and land and/or marine biogeochemistry (Drobinski et al. 2012;Zou and Zhou 2012;Lorenz and Jacob 2014;Sevault et al. 2014;Sein et al. 2015;Sitz et al. 2017).
Two technical aspects make coupled modeling particularly difficult. The first is that most often, the different components are run on different spatial grids, resolutions and domains. For example, in order to resolve mesoscale ocean eddies, often the ocean models are run at higher horizontal resolution than the atmospheric counterparts. Therefore variables exchanged at their interfaces (for example, energy fluxes and wind stress at the ocean-atmosphere interface) need to be interpolated from one domain grid to another. This is usually achieved either using ad hoc procedures or general interpolator packages that can actually be quite complex. It should be remembered, however, that there might be errors or excessive approximations in the spatial interpolation introduced as a result of the different scales characteristic of different phenomena. For example, given interface fields may have to be disaggregated or upscaled in the exchange process. These may result in an imbalance across the models, which calls for careful interpolation procedures or suitable correction techniques (e.g. flux correction).
The second difficulty is related to the different temporal scales of evolution across components. For example, the atmosphere evolves and responds to external forcings much more rapidly than the oceans, which have a higher thermal and dynamical inertia, and thus longer evolution times. In fact, some chemical reactions can occur at extremely fast temporal scales. On the one hand, this implies that each component runs with different time steps, and thus some temporal interpolation is necessary in the information exchange. In the meantime, relatively long simulation times are needed to equilibrate the atmosphere and ocean components which, for a high resolution model, can prove to be a daunting computational task. This problem has been often bypassed through the so-called asynchronous coupling, in which ocean and atmosphere components are not coupled at each time step, but at different temporal scales.
Without going into greater technical detail, it is clear that the development of a fully coupled comprehensive RESM system is a formidable modeling task, which has been so far approached by incremental steps in which various components have been incrementally added to a base modeling framework. This has resulted in studies addressing different interactions across the components, as will be reviewed in the next section.

Atmosphere-ocean coupling
The development of the first coupled atmosphere-ocean RCMs (AORCMs) is relatively recent, late 2000s and early 2010s, however a number of coupled AORCMs are already In this paper we thus review the status and perspectives of RESM development by first describing the basic modeling structure of a coupled RESM (Section 2), then presenting illustrative examples of studies addressing the regional interactions across different climate system components (Section 3), and finally discussing future directions. Figure 1 illustrates the basic structure of a coupled RESM. The main components of the model are the atmosphere, ocean, land surface and hydrology, cryosphere, chemosphere (gaseous compounds and aerosols), and biosphere. The atmosphere interacts with the oceans, land, and cryosphere through exchanges of energy and mass; with chemistry/aerosols through processes of emission, transport, and removal; and with the biosphere through its effects on the biogeochemical cycles. Similar interactions are found for the oceans and cryospheric components, with the addition of a direct coupling of oceans and land through river discharge. Atmospheric constituents and aerosols can have strong interactions with the atmosphere and regional climate, as well as the biosphere, both terrestrial and marine, through radiative, microphysical, and chemical processes. It is also important to emphasize that many of the interactions described above are highly non-linear due to the presence of strong feedbacks, such as the ice-albedo one.

The basic structure of a coupled RESM
The main difference between the RESM structure of Figure 1 and the structure of global Earth System Models (ESM) is that RESMs, being run only over a limited area domain, require time-dependent lateral boundary conditions for those components regulated by three-dimensional dynamical equations, such as the atmosphere, oceans, and chemosphere. These have to be provided either by global ESMs or by analyses of observations. Figure 1. schematic depiction of a coupled regional earth system model framework and the interactions across it's different components and Gcm drivers (ctm: chemical transport model). Arrows indicate the flow of information. Blue arrow: the interaction with driving global models; red arrow: the interaction inside the resm.
available (e.g. Schrum 2017). An issue central to regional ocean-atmosphere coupling is weather the coupling itself improves the simulation of climate features compared to the uncoupled atmospheric model. This has been clearly shown for extreme weather phenomena, such as tropical cyclones (Bao et al. 2000;Bender and Ginis 2000;Bender et al. 2010), or in highly convective regions, such as the maritime continent (Aldrian et al. 2005;Seo, Miller, and Roads 2007;Wei et al. 2014), or the South Atlantic (Byrne et al. 2015;Ratnam et al. 2015).
A region where air-sea coupling is particularly important is the Indian ocean, and for this reason several coupled RCM development efforts focused on the Indian ocean basin. Among them, Krishna, Hoerling, and Rajagopalan (2005), Seo et al. (2008Seo et al. ( , 2009, Ratnam et al. (2009), Samala et al. (2013, Samson et al. (2014), and Di Sante (2017) built AORCMs coupling different atmospheric and ocean regional model components. Their studies indeed found that, compared to the stand alone atmospheric RCMs, the coupled models improved the simulation of the patterns and intraseasonal variability of the South Asia monsoon precipitation, as well as the simulation of tropical storms, due to the effect of air-sea coupling. It thus appears that coupled AORCMs can be especially useful in regional tropical climate settings.
A similar result was also found for the East Asia monsoon, where air-sea coupling significantly improved the simulation of the monsoon's evolution and variability Zhou 2013, 2016). For example as shown in Figure 2, the temporal correlation coefficients of precipitation anomaly between observed and simulated over the Western North Pacific and South China Sea are 0.14 and 0.37 before the coupling, and 0.50 and 0.55 after the coupling, respectively (Zou and Zhou 2013). Several other efforts have indeed been devoted to the development of coupled atmosphere-ocean regional models for the East Asia region (Lin, Qian, and Zhang 2006;Fang et al. 2009;Yao and Zhang 2009;Li and Zhou 2010).
In addition, Seo, Miller, and Roads (2007) found that air-sea mesoscale interactions improved the simulation of frontal systems and associated precipitation patterns over the Eastern Pacific Sector, while different models have shown good performance in reproducing air-sea mesoscale interactions in coastal upwelling regions (Ribeiro, Soares, and de Oliveira 2011;Putrasahan, Miller, and Seo 2013;Li et al. 2014aLi et al. , 2014b. Another region for which coupled regional modeling has been particularly active is the Mediterranean basin, where several AORCMs have been developed as part of the European project CIRCE (Gualdi et al. 2013) as well as the MED-CORDEX initiative (Ruti et al. 2016 Sevault et al. (2014). The Mediterranean basin is an optimal setting for regional coupling because many air sea interactions occur at the mesoscale, for example in the generation of deep water and extreme meteorological events (e.g. Lebeaupin Brossier et al. 2015), and because it is a semi-enclosed sea, thus reducing the importance of the provision of ocean lateral boundary conditions. Coupled AORCMs for the Mediterranean have been run for twenty-first century projections, and have indicated that the coupling can indeed modulate significantly the climate change signal from the uncoupled models (e.g. Somot et al. 2008;Dubois et al. 2012).
Finally, since the early years of RCM development, a number of studies have investigated the interactions between atmosphere and lakes via the use of coupled RCM-lake models. This has been done for lakes in the continental U.S. (e.g. Hostetler, Bates, and Giorgi 1993;Bates, Hostetler, and Giorgi 1995;Martynov et al. 2012;Notaro et al. 2013), Central Asia (e.g. Small et al. 1999;Turuncoglu et al. 2013), Africa (e.g. Thiery et al. 2015;Diallo, Giorgi, and Stordal 2017), and Europe (e.g. Mironov et al. 2010). non-linear interactions between the climate and the cryosphere in order to enhance the reliability of climate projections in cold-climate regions.

Atmosphere-chemistry/aerosol coupling
One of the areas that has received considerable attention is the interactive coupling of regional climate and chemistry/ aerosol models. This is because tropospheric aerosols can exert a radiative forcing sufficient to significantly affect regional climates, especially in tropical regions. The first simplified aerosol model coupled to an RCM was developed by Qian and Giorgi (1999), who found that the direct radiative forcing of anthropogenic sulfate aerosols over East Asia is sufficient to produce a statistically significant cooling and decrease of precipitation over the region. This pioneering work was followed by a series of further RCMbased studies of direct and indirect effects of sulfate and organic aerosols over East Asia (e.g. Qian et al. 2001Qian et al. , 2003Giorgi, Bi, and Qian 2003;Li et al. 2009;Ji et al. 2011Ji et al. , 2015Wang et al. 2015b), as well as mineral dust emitted from the Gobi desert (e.g. Zhang et al. 2009Zhang et al. , 2016and  Similar experiments have been conducted also over the Africa continent, where biomass burning and Saharan dust are important environmental and climate concerns. A series of coupled RCM-aerosol based studies investigated the role of biomass burning aerosols and mineral dust emissions on the West Africa monsoon precipitation (Konare et al. 2008;Solmon et al. 2008;Ji et al. 2016a). They found that the aerosol can significantly affect the monsoon development through the competing effect of surface cooling and elevated heat pumping, the former leading to an inhibition of the inland penetration of the monsoon rain band, and the latter leading to a contrary effect. Depending on the aerosol optical properties, one effect was dominant over the other, demonstrating the used in RCMs have included interactive snow modules, so we will not focus on snow, but on specific models of cryospheric systems, such as sea ice and glacier models. Coupled regional atmosphere-ocean-sea ice models have been developed for cold climate regions, such as the Arctic (e.g. Curry and Lynch 2002;Roberts et al. 2008), the Antarctic (Bailey and Lynch 2000), the North Sea and Baltic Sea basins (e.g. Gustafsson, Nyberg, and Omstedt 1998;Hagedorn, Lehmann, and Jacob 2000;Doscher et al. 2002;Schrum et al. 2003;Lehmann, Lorenz, and Jacob 2004;Van Pham et al. 2014;Su et al. 2014;Wang et al. 2015a), and the Caspian Sea (e.g. Turuncoglu et al. 2013).
Many of these developments have occurred within the context of multi-model intercomparison projects (e.g. Raschke et al. 2001;Curry and Lynch 2002;Rinke et al. 2006), and the models have been applied to a variety of studies, from analyses of atmosphere-ocean-sea ice coupling (Rinke et al. 2003;Mikolajewicz et al. 2005;Döscher et al. 2010;Döscher and Koenigk 2013;Gröger et al. 2015) to future climate projections (Kjellström, Döscher, and Meier 2005;Meier 2006;Bülow et al. 2014;Meier 2015;Koenigk, Do¨Scher, and Nikulin 2011;Schrum et al. 2016). All these studies clearly indicated that the interactive sea ice component substantially increases both the model varibility and the inter-model spread in the simulated response to future warming. This is mostly because of the model representation of sea ice thermodynamical processes and their response to external forcings and air-sea-ice interactions (Schrum 2017).
The development of interactive land glaciers, to date, has received less attention, mostly because of the scale mismatch between glacier processes and the resolution of RCMs (Kotlarski et al. 2010). This will likely require either the use of sub-grd scale parameterizations of land surface processes, or the move to very high resolution convection-permitting models.
More generally, increased development and testing of coupled atmosphere-ocean-cryosphere RCMs will be necessary to better describe the complex and highly in the next years. It presents some important technical difficulties, such as the interface of models running on different grid and having a wide range of characteristics times of evolution, and the provision of lateral boundary conditions for the different components of the system. The former problem can be best addressed with the use of general-purpose interpolators, a few of which are available as community tools. The latter may require some more ad hoc solutions, especially concerning the ocean and chemistry components. In both cases, however, attention should be given to the implications of interpolating variables associated with phenomena characterized by different spatial and temporal scales.
Several coupled atmosphere-ocean-sea ice-aerosol/ chemistry RESMs have been developed, but they have been tested mostly over specific domains. Much more work and intercomparison studies are needed to assess the transferability of these coupled models in different regional settings and to obtain general conclusions concerning the importance of the representation of coupled processes. In particular, the inclusion of interactive biosphere has been limited so far, but the interest in this aspect of coupled modeling is indeed growing in view of the role it will play especially within the context of future climate change, which may lead to pronounced changes in natural ecosystems.
In our opinion, the next challenge in RESM modeling is the inclusion of the human factor. Human activities such as land-use change and greenhouse gas and aerosol emissions, are currently considered in most model experiments as external players in the climate system, either as forcings or as receptors (e.g. impacts). However, there is a two-way interaction between human societies and the natural environment, whereby on the one hand human populations may migrate in response to climatic, environmental and socio-economic stresses, and on the other hand adaptation policies to respond to climate change may in turn affect climate. In an era where humans are now a key component of the climate system, these processes will have to be included in the next generation of earth system models. RESMs can be an optimal test-bed for this model development because it can focus on specific regional interactions, and the lessons learned from this exercise can eventually be extended and generalized to global models.

Disclosure statement
No potential conflict of interest was reported by the authors. importance of accurately characterizing aerosol characteristics, both microphysical and optical, in order to best represent aerosol effects on regional climate.

Funding
Coupled RCM-based simulations of aerosol-climate interactions were also conducted over the European region, for example by Nabat et al. (2014Nabat et al. ( , 2015 and Zanis et al. (2012), who showed that direct and semi-direct aerosol effects are key ingredients necessary to explain the spatio-temporal structure of solar radiation and temperature over Europe.
Within the context of coupled RCM-based aerosol modeling, it should be mentioned that highly simplified aerosol models have been mostly used in climate-type applications due to the computational requirements of running more complex microphysical and chemical schemes. However, efforts are under way to interactively couple within RESM systems full three-dimensional chemistry modules which will allow a more refined representation of chemistry-climate interactions (e.g. Shalaby et al. 2012).

Atmosphere-biosphere coupling
Coupling of biosphere and atmosphere components in RESMs is important because the surface characteristics can substantially affect regional and local climates. Numerous studies have addressed the effect of land surface changes in RCMs (e.g. Giorgi and Gutowski 2015), but these were imposed and thus no two-way interactions between climate and land use was allowed. Some advanced land surface modules, such as CLM (Oleson et al. 2008), include the capability of describing vegetation dynamics in response to climate input, as well as the presence of crops with annexed technology options. These land modules have been incorporated in different RCMs, and thus would allow the simulation of two-way biosphere-atmosphere interactions. Alternatively, dynamical vegetation/bioeochemistry models can be coupled to RCMs, as done for example by Smith et al. (2011)  Some early studies of regional biophysical feedbacks have been conducted (e.g. Zhang et al. 2014;Wang et al. 2017), but generally these studies present the difficulty of needing long spin-up times to bring the biosphere model to equilibrium with the RCM's climate. This is clearly an area of research that will receive increasing attention, as anthropogenic climate change and other human activities continue to substantially alter the natural biogeochemical cycle of the Earth.

Future directions in RESM development
Clearly, the development of RESMs is still in its beginning stages, and will undoubtedly receive increasing attention