The climate change signal in the Mediterranean Sea in a regionally coupled atmosphere-ocean model

We analyze the climate change signal in the Mediterranean Sea using the regionally coupled model REMO-OASISMPIOM (ROM). The ROM oceanic component is global with regionally high horizontal resolution in the Mediterranean Sea, so that the water exchanges with the adjacent North Atlantic and Black Sea are explicitly simulated. Simulations forced by ERA-Interim show an accurate representation of the present Mediterranean climate. Our analysis of the RCP8.5 scenario using the Max-Planck Institute Earth System Model shows that the Mediterranean waters will be warmer and saltier throughout most 15 of the basin by the end of this century. In the upper ocean layer temperature is projected to have a mean increase of 2.7°C, while the mean salinity increases by 0.2 psu, presenting a decreasing trend in the Western Mediterranean, opposite to the rest of the basin. The warming initially takes place at the surface and propagates gradually to deeper layers. Hydrographic changes have an impact on intermediate water characteristics, potentially affecting the Mediterranean Thermohaline Circulation in the future. 20

resolution atmosphere-ocean simulations with the coupled ROM model, which has been used for some previous multimodel studies (see e.g. Darmaraki et al., 2019). Here we analyze the evolution of Mediterranean Sea under the RCP8.5 scenario with boundary conditions taken from CMIP5 simulation with the MPI-ESM global model. Especially, we focus on water masses properties such as SST and SSS and their evolution at the end of 21 st century.
The objectives of this study can be summarized as follows : 5 (i) Assess the skills of ROM in reproducing the observed regional climate over the Mediterranean Sea when driven by ERA-Interim reanalysis.
(ii) Examine the added value that high-resolution ROM brings with respect to the driving global model in the area of study, when forced by MPI-ESM.
(iii) Assess the projected climate change signal in the Mediterranean Sea in the RCP8.5 scenario. 10 This paper is organized in the following way: a general description of our coupled model and each of its components is given in section 2. In section 3, we present the results of the validation followed by the coupled model simulations for the Mediterranean region. Finally, section 4 contains the discussion and 5 the conclusions.

Atmosphere (REMO)
The atmospheric component of ROM is the REMO. The dynamic core of the model and the discretization in space and time are based on the Europa-Model of the Germany Weather service (Majewski, 1991). The physical parameterizations are taken from the global climate model ECHAM versions 4 and 5 (Roeckner et al., 1996(Roeckner et al., , 2003. The variables which exchange info between REMO and MPI-OM via OASIS are 10 m wind velocity, wind stress over water, wind stress over sea ice, liquid 25 precipitation, solid precipitation, net shortwave radiation, total heat flux over water, conductive heat flux and residual heat flux (Fig. 2a). To avoid the largely different extensions of the grid cells close to the poles, REMO uses a rotated grid, with the equator of the rotated system in the middle of the model domain. The horizontal discretization is done on the Arakawa Cgrid and the hybrid vertical coordinates are defined according to Simmons and Burridge (1981). More information about the parameterizations of atmospheric component can be found in Sein et al. (2015). 30

Ocean (MPI-OM)
The oceanic component of ROM is the MPI-OM developed at the Max Planck Institute for Meteorology (Hamburg, Germany). MPI-OM is a free surface, primitive equations ocean model, which uses the Boussinesq and incompressibility approximations. MPI-OM is formulated on an orthogonal curvilinear Arakawa C-grid (Arakawa and Lamb, 1977) with variable spatial resolution. This grid allows for the placement of the poles over land, thus removing the numerical singularity 5 associated with the convergence of meridians at the geographical North Pole. An additional advantage of the curvilinear grids is that a higher resolution in the region of interest can be reached, while maintaining a global domain. Using the global ocean model alleviates issues related to ocean open boundary conditions and provides an additional "degree of freedom" in the model setup and tuning, which can be helpful to adjust the ocean component for the better performance within the region of interest. The model parameterizations and setup are detailed in Sein et al. (2015). 10 with a resolution of about 25 km on a rotated grid. More information about the ROM coupled system is summarised in Table   1. The HD model (global domain) computes the river discharge at 0.5º resolution. The atmosphere and ocean exchange information each 60 minutes, while HD interacts with MPI-OM and REMO each 24 hours (Fig. 2a).

ROM experiment set-up
In this work, 30-year time series from three different experiments have been analyzed. The first simulation, thereafter ROM_P0, was forced by ERA-Interim for 1980-2012 and used to assess the skills of ROM in reproducing the observed 25 regional climate over the Mediterranean Sea. In order to offer an integrated vision of the impact introduced by the climate change in the Mediterranean Sea, we make a dynamical downscaled of present time simulation with MPI-ESM-LR which covers 1950-2005 period (for our analysis we take from 1976-2005, ROM_P1) and a climate change projection from 2006-2099 (for our analysis we take from 2070-2099, ROM_P2) under the Representative Concentration Pathways 8. 5 (RCP8.5) scenario. 30

Validation Methodology
The ROM present Mediterranean climate is analyzed in terms of mean state, seasonal cycle and interannual variability of the main atmospheric and oceanic variables. For the ROM atmospheric component REMO, three representative variables were chosen: Mean Sea Level Pressure (MSLP), near-surface temperature (T2m) and precipitation; while for the ocean component MPI-OM: Sea Surface Temperature (SST), Sea Surface Salinity (SSS), Sea Surface Height (SSH) and the velocity 5 components of the sub-surface current. These fields are compared to gridded data from different sources to evaluate the ability of ROM model to simulate the present Mediterranean climate. These data sets are derived from observations or reanalysis where appropriate (Table 2).
For MSLP and T2m we compare the output of ROM with ERA-Interim reanalysis. The ERA-Interim data assimilation system uses a 2006 release of the Integrated Forecasting System (IFC) developed jointly by ECMWF and Météo-France. The 10 spatial resolution of data set is approximately 80 km (T255 spectral) on 60 vertical levels from the surface up to 0.1 hPa (Dee et al., 2011); free access data can be found at https://www.ecmwf.int/en/research/climate-reanalysis/era-interim. Total precipitation was validated against the Tropical Rainfall Measuring Mission (TRMM; Huffman et al., 2014) dataset, a joint mission between NASA and the Japan Aerospace Exploration Agency (JAXA) to study rainfall for weather and climate research. 15 Three datasets were used for the evaluation of the SST: ERA-Interim, EN4 and OISST. For the development of EN4 data set Good et al. (2013) performed a 1-degree monthly objective analysis from ocean temperature and salinity bathythermograph profiles (MBT, XBT). The version EN4.1.1 used here includes the improvements on the estimation of MBTs and XBTs downward velocity developed by Gouretski and Reseghetti (2010). On the other side, the NOAA performed an analysis constructed by combining observation from different platforms (satellites, ships, buoys) on a regular global grid 1/4º x 1/4º, 20 known as Optimum Interpolation Sea Surface Temperature (OISST; Reynolds et al., 2007). Currently, the OISST dataset is considered the best-observed SST dataset available, in terms of spatial and temporal resolution.
To validate ROM SSS, we made comparisons with two climatologies: EN4 v.4.1.1 (Good et al., 2013) and MEDSEA_REANALYSIS_PHY_006_009 (Fratianni et al., 2015) implemented by Copernicus Marine Environment Monitoring Service (CMEMS). MEDSEA_REANALYSIS_PHY_006_009 is based on the NEMO code, the data 25 assimilation scheme is variational, and all historical  in-situ and satellite observations were used. The model is primitive equation in spherical coordinates implemented for the Mediterranean at 1/16º x 1/16º horizontal resolution (Fratianni et al., 2015). 2013) for ocean as well as JSBACH (Reick et al., 2013) for terrestrial biosphere and HAMOCC (Ilyina et al., 2013) for the ocean's biogeochemistry. The coupling of the atmosphere, ocean and land surface is made possible by the OASIS3 (Valcke, 2013) coupler. Depending on the resolution of the ECHAM6 or MPI-OM the MPI-ESM has different configurations (MPI-ESM-LR (low resolution), -MR (medium resolution); the -LR uses a bipolar grid with 1.5º resolution, while the -MR version doubles the number of levels in the atmosphere and decreases the horizontal grid spacing of the ocean to 0.4º (Giorgetta et al., 2013). 5

Results
In this section, a selection of key fields corresponding to the period 1980-2012 (ROM_P0) is presented. In a second step changes in the Mediterranean Sea state under RCP8.5 conditions are estimated from the analysis of differences between present climate (1976( -2005 and the climate projection (2070-2099, ROM_P2).

Atmosphere validation 10
Mean sea level pressure (MSLP) is a good indicator of large-scale circulation, which influences near-surface temperature (T2m) and precipitation distributions. Erroneous MSLP gradients lead to an erroneous regional wind circulation, and can also have a strong effect on ocean circulation (Sein et al., 2015). Figs. 3a and 3b display the biases of modeled MSLP with respect to ERA-Interim for the boreal winter (defined as December, January, and February; DJF) and summer (defined as June, July, and August; JJA) in the 1980-2012 period (ROM_P0). 15 According to Figs. 3a and 3b ROM provides a good agreement with ERA-Interim MSLP, showing maximum deviations smaller than 3 hPa over most of the domain for both seasons. The strongest departures can be found in DJF, due to an overestimation of the Azores high during the winter months. Those differences could be attributed partly to REMO parameterizations, but a more important role could be played by the deficiencies in the simulated ocean circulation in the North Atlantic, which are analyzed in a paper in preparation. Nonetheless, these relatively small deviations imply a small 20 change in terms of regional wind circulation. During summer months ( This disagreement can be attributed to differences in the resolution of orographic features. Winter months show the largest 25 T2m biases located close to the Mediterranean coastline, where atmospheric-ocean interactions could play a role. At first glance, ROM generally underestimates the simulated cumulative precipitation over most of the Mediterranean region, for both winter and summer seasons. The largest discrepancies for DJF can be located over the Black Sea, the Adriatic Sea, the Gulf of Lions and the southwest portion of Iberian Peninsula (Fig. 3e), where negative anomalies can reach 3mm/d. Moreover, it is worth stating that during the same period the total precipitation was overestimated in regions linked 30 to significant topographic reliefs (e.g. the Alps). Some coastal areas also showed positives anomalies that probably are related to the atmosphere-ocean coupling. In the very dry summer season, ROM shows a clear tendency to underestimate the precipitation (Fig. 3f). Over the ocean, this bias can be related to the cold SST bias, common to the most of the AORCMs simulations of the Mediterranean climate (see Darmaraki et al., 2019). The seasonal mean precipitation is reasonably well simulated by our coupled system along the most of the Mediterranean basin. However, the ROM simulation shows significant systematic errors (up to ±3.5 mm/d) remain substantial over the region in terms of precipitation. 5 The impact of interactive atmosphere-ocean coupling in REMO is shown in Fig. 4, presenting the climatology differences between ROM and stand-alone REMO in the simulations forced by ERA-Interim for MSLP, T2m, and precipitation. Over land the simulated fields have a larger dependency on the internal details of the atmospheric component, and the impact of the coupling is dependent on the large-scale circulation and land-sea contrasts. Therefore, we can expect the differences over land to be overly small, except for the regions where the large-scale circulation or the land-sea contrasts are important. 10 The winter MSLP over the Atlantic is higher in the coupled run ( Fig. 4a), causing an anomalous strong anticyclonic circulation that extends to land and the Mediterranean Sea, west of the Balearic Islands. The large-scale influence of the Atlantic anomalous circulation offsets the effect of the warmer SST here (see Fig. 5, where the SST biases are represented).
However, elsewhere over the Mediterranean Sea, where the ROM SST is colder (warmer) than ERA-Interim, a higher (lower) MSLP is simulated by ROM. In summer (Fig. 4b), the differences in MSLP seem to be determined mainly by the 15 colder SST in ROM, which leads to higher MSLP in the model than in the reanalysis.
The changes in T2m induced by the coupling over the Mediterranean (Figs. 4c and 4d) seem to be mainly determinate by the SST, through the turbulent heat fluxes. In both seasons the T2m differences induced by the coupling correspond very well with the SST biases with respect to ERA-Interim. However, in winter T2m seems to be also influenced by the transport of Atlantic air carried by the anomalous anticyclonic circulation simulated in the Atlantic. Over land the differences in winter 20 T2m are mainly determined by the changes induced in large scale circulation by the interactive SST in the Atlantic, while in summer the land-sea contrasts seem to be more important.
The differences between the SST from ERA-Interim and the simulated by ROM is also reflected in the rainfall simulated by REMO and ROM (Figs. 4e and 4f). In winter the Mediterranean Sea regions where the ROM SST is warmer have a higher precipitation, while colder ROM SST leads to a lower precipitation. The impact of the SST biases on the precipitation is 25 clearer in summer: the cold SST bias leads in ROM to a precipitation which is weaker than in the REMO simulation all over the Mediterranean Sea, especially in the northern part, where the reduction of precipitation in ROM with respect to REMO is comparable in magnitude to the ROM precipitation bias (Fig. 3f).

Seasonal cycle 30
The differences between ROM and observed SST climatologies for winter (DJF) and summer (JJA) in the 1980-2012 period are presented in Fig. 5. The SST seasonal cycle is well represented by the model, although its amplitude is reduced over most of the Mediterranean Sea. The deviations in absolute value do not exceed 3ºC, although ROM shows a cold bias, which is more significant in the eastern Mediterranean, especially in summer (Fig. 5).
In DJF ROM overestimates SST over the Mediterranean northern coasts and the whole western basin, showing positive biases reaching 2ºC (Figs. 5a, 5b and 5c). In summer, the negative SST extends over a large part of the Mediterranean domain (Figs. 5d, 5e, and 5f) indicating that the model is simulating colder temperatures than expected for JJA. 5 In order to assess the improvement that higher resolution in ROM brings to the simulation of the present Mediterranean climate, comparisons with MPI-ESM-LR and -MR have been done (Fig. 6): SST seasonal cycle amplitude is smaller in ROM than in MPI-ESMs, with warmer DJF and colder JJA. The SST bias is between ±3ºC in the whole Mediterranean basin. In winter, ROM shows warmer temperatures than MPI-ESM (-LR and -MR, Figs. 6a and 6b) with the exception of southeastern Mediterranean coasts where negative differences appear (near to -10 1ºC). In JJA a cold bias is presented into the western basin (-1.5ºC), southern coasts (-0.5ºC), Levantine and Aegean seas (-3ºC) while into the Tyrrhenian, Adriatic and Ionian seas low positive anomalies (up to +1ºC) are presented (Figs. 6c and 6d). The increasing trend of modeled SST is weaker than in the observed climatologies (Table 3) due to the absence of effects introduced by aerosols. Despite these small differences, the interannual variability and the trend is well reproduced by ROM 20 simulation during 1980-2012 period. An offset of SST is visible in Fig. 7, although it keeps constant during all period.

Interannual variability
A Taylor diagram (Fig. 8) was used to quantitatively evaluate ROM performance. ERA-Interim, EN4 and ROM are all well correlated (r>0.7) with the observation-based climatology (OISST). However, the SST standard deviation of ROM (0.27ºC) is lower than OISST (0.32ºC), while ERA-Interim, EN4 present closer values (0.34 and 0.33ºC, respectively). The corresponding root-mean-square-errors (RMSE, red contours) are enclosed by 0.07 and 0.22ºC, being ROM close to the 25 climatological uncertainty. Fig. 9 shows the differences between the SSS modeled by ROM and the selected reanalysis averaged for DJF and JJA during 1980-2012. In all comparisons a common pattern in spatial SSS bias distribution is observed, showing positive bias over the western basin and Adriatic Sea and negative bias through the Levantine Sea and north Aegean Sea. It is precisely at 30 northeast Adriatic Sea, by the Po Delta, where the largest positive differences occur (+3 psu), and to the north of the Aegean Sea where largest negative differences (-3 psu) are found. Nevertheless, the deviations do not exceed, in absolute value, 0.5 psu in a large part of the domain (Fig. 9).

SSS
Comparison with MPI-ESM-LR and -MR SSS is shown in Fig. 10. ROM is always saltier over the whole Mediterranean, with decreasing difference towards the southeast. In general, ROM SSS is closer to EN4 and CMEMS climatology than any of the MPI-ESM versions. 5

SSH and circulation
To conclude with the analysis of the ocean component of ROM, the SSH was analyzed. The time-averaged SSH and horizontal current velocity at 31 m depth simulated by ROM between 1980-2012 are shown in Fig. 11. It is clearly seen how  . Fig. 12 shows the yearly mean and the seasonal cycle of ROM SSH compared to altimetric data. The modeled SSH shows lower values than the observed (Fig. 12a); however, it 20 represents quite acceptably the behavior of AVISO time series. The amplitude of the mean seasonal cycle is 12 cm for the simulation, and 14.5 cm for AVISO (Fig. 12b). Thus, the model is able to reproduce a realistic interannual variability and seasonal cycle.
Finally, a mass balance was made to estimate the net transport of water throughout the Strait of Gibraltar and Dardanelles in order to compare the water flux modeled by ROM with the observations. Table 4 gives the water budget of ROM averaged 25 over the 1980-2012 period. The water loss by evaporation (E) is greater than the gain by precipitation (P) and river runoff (R) generating a deficit of 0.034 Sv into the basin. However, this deficit is partially compensated by the net water inflow through the Strait of Gibraltar (0.03 Sv) and the Dardanelles, where the inflow (0.132 Sv) exceeds the outflow (0.109 Sv).
ROM water budget is 0.026 Sv lower compared to RCSM4 model, but a significant part of the difference is due to difference in river runoff. 30

Projections under RCP8.5 scenario
Projections performed by the ensembles of ESMs collected by the IPCC (2013) under the conditions of different scenarios anticipate that the climate change will cause a generalized and perceptible global warming of the oceans, at surface and deeper layers at the end of 21 st century. Fig. 13 shows the mean SST and SSS fields for the present climate (1976( -2005 together with the differences with 5 respect to future projections under the RCP8.5 scenario (ROM_P2-ROM_P1). The averaged zonal SST gradient over most of the Mediterranean Sea increases from north to south. The western Mediterranean is colder than the eastern, especially in the Gulf of Lions and in the northern Adriatic Sea where the SST minima are located (Fig. 13a). The warmest area is found along the Levantine Sea coastline. The averaged Mediterranean SST is 18.61ºC (Fig. 13a) while at the end of the 21 st century under RCP8.5 scenario it is expected to have a mean increase of +2.73ºC, warming in a range from a maximum of 3.8ºC at 10 the Aegean Sea to a minimum of 0.9ºC at the Alboran Sea (Fig. 13b).
To verify that the trend remains stable and it is not affected by a strong ROM's SST bias, comparisons for DJF and JJA have been performed separately (see Supplementary Figures). Noting the results, the trend analysis is not affected by the SST bias ( Fig. 5).
As shown in Fig. 13c the Eastern Mediterranean is saltier than western, particularly in the Levantine Sea (39 psu). The 15 Western basin presents lower salinities (< 38.25 psu) influenced by the inflow of Atlantic freshwater through the Strait of Gibraltar (36.6 psu) along the African coasts up to the Ionian Sea. Another source of freshwater is located at the Dardanelles strait where the water from the Black Sea has salinities lower than 35 psu. The averaged Mediterranean SSS is 38.02 psu while under the RCP8.5 projection it will experience a mean increase of 0.17 psu. The differences between the mean SSS projection and present climate shows a dipolar structure through the Mediterranean Sea (Fig. 13d). Under the RCP8.5 20 scenario, the Western Mediterranean is expected to slightly freshen (from -0.5 to -1 psu), while the Eastern will become saltier. It is precisely at the north of the Aegean Sea where largest SSS increases (+4 psu) are found.
In general, MPI-ESM-LR and -MR projections under the RCP8.5 scenario at the end of the 21 st century are warmer than that of ROM over most of the Mediterranean Sea. Namely, the projected mean SST increases are 2.80 and 2.87ºC for MPI-ESM-LR and -MR, respectively (Table 5). Despite differences in horizontal resolution, MPI-ESM-LR and ROM show a similar 25 spatial distribution of the expected warming (Fig. 14a), contrary to MPI-ESM-MR where the SST warming is projected to be higher in the western basin and north of Adriatic Sea (Fig. 14b).
ROM projection is always saltier at surface than any MPI-ESM version over the whole Mediterranean basin. The mean SSS change for the 2070-2099 period with respect to 1976-2005 in MPI-ESMs shows the same spatial pattern than ROM model (Figs 14c and 14d). The MPI-ESM-LR forced by the RCP8.5 scenario shows a SSS mean increase of 0.10 psu while for the -30 MR of 0.12 psu. Fig. 15 shows the mean temporal evolution of temperature and salinity anomalies in the water column over the Western and Eastern Mediterranean along the 21 st century according to ROM projection for the RCP8.5 scenario. To calculate these anomalies in a given region we first average horizontally, as indicated in Fig. 15 insets, the temperature and salinity in each model level for the present time period  and the RCP8. 5 projection period (2006-2099). The anomalies are defined as the difference between the time series for RCP8. 5 scenario (2006-2099) and the time mean for the present climate period (ROM_P1). The Mediterranean Sea shows an increase of its temperature through the entire water column (Figs. 15a and 15c), which is more evident in surface layers. The warming that initially takes place in the upper ocean propagates 5 gradually to deeper layers along the 21st century. The behavior of the Eastern Mediterranean is similar to that of the western basin, but with warmer temperatures, especially in the surface layers. At the end of the 21 st century the eastern basin is expected to have a surface temperature increase up to 3.8ºC (Fig. 15a) and the western up to 3ºC (Fig. 15c). In deeper layers (1000 m) the water temperature will increase by 0.6ºC for both basins, which is a very significant warming at these depths.
The mean temporal evolution of salinity anomalies displays different patterns through the Mediterranean Sea. During the 10 21st century the upper layer (0-100 m) of the Western Mediterranean is projected to freshen (-0.5 psu) while the deeper layers tends to get saltier up to 0.5 psu. However, the Eastern Mediterranean will increase its salinity up to 0.5 psu in the entire water column along the current century.

Discussion
AORCMs are capable of improving the simulation of the climate system by the driving model through dynamical Also, ROM is able to give a good representation of the main characteristics of the ocean circulation in the Adriatic Sea, while in MPI-ESM-LR the ocean model is not able to represent these features due to the lack of the necessary horizontal 20 resolution. Compared to other state-of-art regional climate models, ROM introduces a remarkable innovation, which consist in the implementation of a global oceanic model with high horizontal resolution at regional scales. This approach allows to obtain information of the global ocean without losing spatial resolution in the coupling area. An important disadvantage of the proposed model, related previously in Sein et al. (2014), is that the bias and internal variability generated from the global domain can influence the results in the coupled domain, making it difficult to separate the source of bias. 25 ROM shows good skills in reproducing the main characteristics of the climate of the Mediterranean Sea. The biases of the main atmospheric and oceanic parameters are in the range shown by other state of art regional models (L'Hévéder et al., 2013;Sevault et al., 2014;Akhtar et al., 2018;Darmaraki et al., 2019).
The seasonal MSLP was validated against ERA-Interim, showing biases smaller than ±3 hPa over most the domain for DJF and JJA, a performance similar to other models (see e.g. Giorgi and Lionello, 2008). Positive MSLP biases over a large 30 extend of the domain during DJF (Fig. 3a) could generate anticyclonic conditions which lead a greater stability and lower storm generation; while in JJA (Fig. 3b) the biases are generally much lower. With respect to the seasonal cycle of near-surface atmospheric parameters such as near-surface (2m) temperature (T2m) and precipitations, the LMDz-NEMO-Med coupled model (L'Hévéder et al., 2013) gives a bias range of (-4; +4ºC/-2; +3 mm/d, respectively) which is comparable to the ROM estimates (Figs. 3c, 3d, 3e and 3f). Similarly, to most of Mediterranean regional models, ROM shows a higher than observed rainfall over areas with pronounced topography such as the Alps (Artale et al., 2010;L'Hévéder et al., 2013;Di Luca et al., 2014). More recently, Fantini et al. (2016) also reported a similar bias (±3 mm/d) in an ensemble of regional 5 coupled models forced by ERA-Interim. Panthou et al. (2018) observed that for heavy precipitation increasing resolution increases the wet biases. We totally agree with the final consideration of Fantini et al. (2016), which proposed that to improve the performance RCMs simulating precipitation it is necessary the availability of high quality and high-resolution observation for the assessment of the models.
The comparison of the ROM with standalone REMO shows that the changes in SST generated by the coupling in the 10 Atlantic Ocean influence the simulated Mediterranean climate, causing a spurious anticyclonic circulation in winter which impacts the surface temperature in the Western Mediterranean. In summer the modeled SST is significantly colder than observations, leading to colder T2m and less precipitation over the basin, as the colder SST reduces the evaporation.
Regarding the oceanic parameters, ROM shows biases within ±3ºC, correlation coefficients above 0.7 and RMSE below 0.25ºC when compared to ERA-Interim, EN4 and OISST data sets. ROM presents cold biases along the Eastern 15 Mediterranean that become stronger and extend to the whole basin in summer months. The summer biases are common to most of the Mediterranean regional coupled simulations (see for instance, Dubois et al., 2012;Li et al., 2012. Akhtar et al. (2018) studied the impact of resolution and coupling in modelling the climate of the Mediterranean Sea and concluded that coupling generates a negative bias in SST. Most recently, Darmaraki et al. (2019) assessed an ensemble of 17 simulations from six models, in which our ROM coupled system was included. Their results showed an averaged cold 20 bias ranging from (-0.29 to -1.01ºC) when regional models are compared to satellite data. When forced by ERA-Interim, ROM shows averaged Mediterranean SSTs that are colder than reference climatologies during 1980-2012 period (Fig. 7), a common trait with other RCSMs Ruti et al., 2015). Macias et al. (2018) showed that a simple spatiallyuniform bias correction improves the simulated surface oceanic conditions of the Mediterranean basin when forcing an oceanic model with atmospheric data from RCM realizations. These results show the summer biases could be related either 25 to a deficit of solar radiation by the atmospheric model or to shortcomings in the simulation of some processes in the ocean model, as vertical mixing or turbidity. As theses biases appear in the coupled runs, we could speculate that some coupling feedbacks are present. This topic deserves a separated study and will be tackled in a future paper. The SSS simulated by ROM shows seasonal biases in the -1; +1 psu range, with a similar magnitude and spatial distribution than those in RCSM4 . The biases are higher in some problematic locations such as the northern Adriatic Sea and Dardanelles 30 Strait (Fig. 9), a feature that also has been shown in previous studies (L'Hévéder et al., 2013;Di Luca et al., 2014;. The Mediterranean water fluxes simulated by ROM (Table 4) have been compared to available observations (Sanchez-Gomez et al., 2011;Soto-Navarro et al., 2014) and model  estimates, providing a physically consistent assessment in the straits. ROM water balance terms over the Mediterranean Sea are similar to those obtained by different authors (Table 4). The main difference is the exchange flows through the Strait of Gibraltar, where ROM presents estimates much lower than those presented by Soto-Navarro et al. (2014), although the net flow is in agreement with most of estimates.
The ROM SSH and surface (31m) circulation are able to reproduce the different quasi-permanent elevation/depression (anticyclonic/cyclonic) structures occurring in the Mediterranean Sea (Fig. 11). The cyclonic gyres (SSH depressions) 5 correspond properly to the water mass formation sites. The 31m depth level has been chosen to remove the high-frequency variability of the uppermost ocean, while retaining a characteristic upper ocean circulation pattern. Also, the choice of this level depth makes our result more comparable with other works such as L' Hévéder et al. (2013) and Sevault et al. (2014).
For 1980-2012 period the comparation between ROM and AVISO (SSALTO/DUACS, 2013) altimetry data (Fig. 12a) produced a quite satisfactory correlation of 0.61, close to the obtained by the RCSM4 (0.68). Finally, the ROM amplitude of 10 the mean seasonal cycle measured was 12 cm while for AVISO was 14.5 cm and for RCSM4 16.9 cm (Fig. 12b). Thus, the model is able to satisfactorily reproduce the seasonal cycle and interannual variability of different oceanic variables.
The model also demonstrated good skills in reproducing the area-averaged interannual standard deviations of SST for the Mediterranean Sea (Fig. 16d). According to Fig. 16, ROM coupled system presents yearly SST standard deviations close to the reference OISST dataset. In fact, ROM does not only improve the yearly spatial standard deviations with respect to the 15 MPI-ESMs (Figs. 16e and 16f) but also regarding to ERA-Interim and EN4 ( Fig. 16b and 16c). The MPI-ESM-LR and -MR are not able to reproduce those local patterns due to the absence of resolution, thus indicating that the dynamical downscaling from MPI-ESM improves the simulation of GCMs.
In our simulations, the Mediterranean Sea will be warmer and saltier at the end of 21 st century. Under the RCP8.5 scenario ROM provides integrated estimates of climate change similar to other models ( Table 5). The mean ∆SST projected by ROM 20 under RCP8.5 scenario is 2.73ºC, close to MPI-ESM simulations, which show an SST increase of 2.80ºC (-LR) and 2.87ºC (-MR). It is also close to the mean increase (+2.6ºC) projected by the CMIP5 ensemble of Shaltout and Omstedt (2014) ( Table 5). The SST warming estimates under the RCP8.5 scenario agree with those obtained by Adloff et al. (2015) and with the multi-model Mediterranean of Darmakari et al. (2019), which show a mean increase of +3.1ºC, with model increases ranging from +2.7 to +3.8ºC. Despite differences horizontal resolutions, ROM expected warming (Fig. 13b) shows a similar 25 spatial distribution to MPI-ESM-LR than MPI-ESM-MR (Figs. 14a and 14b). This is due to the ROM_P1 simulation used to computed the ∆SST (ROM_P1-ROM_P2) was forced by MPI-ESM-LR. Figs. 15a and 15b shown how the warming that initially takes place at the surface layer is transported gradually to deeper layers. The warming expected by the models and its intensity will be strongly linked with the choice of emissions scenario.
As shown in Table 5, the mean ∆SSS projected by ROM under RCP8.5 is lower than those estimated by other authors 30 (Somot et al., 2006;, Adloff et al., 2015. This seems to be related to the fact that the SSS filed in the ROM RCP8.5 projection shows a dipolar structure in the Mediterranean (Fig. 13d). So far, previous works (Somot et al., 2006, Adloff et al., 2015 denoted positive values of ∆SSS over the whole Mediterranean Sea by the end of the century, which differ to our results. The ROM SSS decrease in the western basin could be related to the influence of surface waters from the Northeast Atlantic. As we exposed in section 3.5, the MPI-ESM-LR and -MR models under RCP8.5 scenario also represent this dipolar pattern and the averaged increment of SSS shown by ROM into the Mediterranean Sea.

Conclusions
In this study, the regional atmosphere-ocean coupled model ROM (Sein et al., 2015) was described and validated for the Mediterranean region. The experiment in which our model is driven by ERA-Interim shows a good performance in 5 simulating the present climate. ROM is able to reproduce the main characteristics of the Mediterranean Sea, providing a physically consistent estimation of the average behavior, seasonal cycle and interannual variability of both atmospheric and oceanic parameters.
The model also demonstrated improvements of local processes such as the exchange of water through the Gibraltar and Dardanelles Straits or internal seas behaviors contrasted to ESMs. The dynamical downscaling from MPI-ESM implemented 10 in our AORCM offers high spatial resolutions, being capable of reproducing with a remarkable detail the main local and mesoscale processes that take place into the Mediterranean basin.
Our analysis of the simulations driven by the MPI-ESM RCP8.5 scenarios shows that by the end of the 21 st century the Mediterranean Sea will be warmer and saltier across most of the basin. The temperature in the upper ocean layer during 2070-2099 period will increase in 2.73ºC in comparison with the 1976-2005 control period, while the mean salinity will 15 increase by 0.17 psu. The warming, that initially takes place at the surface propagates gradually to the deeper layers.
Furthermore, it is very remarkable that the Western Mediterranean surface layer presents a salinity decreasing tendency, opposite to the rest of the Mediterranean.
Finally, we conclude that the ROM is a powerful model system that can be used to estimate possible impacts of climate change on regional scale. In the future, we plan to use our ROM coupled system to characterizing and analyzing the climate 20 variability of deep water formations in the Mediterranean Sea.