Mediterranean Sea response to climate change in an ensemble of twenty first century scenarios

2.1 The regional ocean climate model (ORCM)

The numerical simulations are performed with the NEMOMED8 model (e.g. Beuvier et al. 2010; Herrmann et al. 2010) which is a Mediterranean configuration of the NEMO ocean model (Madec 2008). NEMOMED8 is a state-of-the-art eddy-permitting regional ocean model for climate-scale studies. Running the 6-member ensemble (760 years including historical simulations and spin-up) would not have been possible with a higher resolution in a reasonable timing. For the horizontal discretization, an orthogonal curvilinear Arakawa C-grid is used (Arakawa and Lamb 1977), the horizontal resolution is \(1/8^\circ \times 1/8^\circ \hbox { cos}(\phi ), \phi\) being the latitude (9–12 km). Thus, the resolution decreases northward. The grid is tilted and stretched at the Gibraltar Strait so that the local resolution is increased up to 6 km. We use a 43-level z-coordinates system, with 10 levels in the upper 100 m. The first layer is centered at 3 m. The time step is 20 min. The partial steps definition of the bottom layer is used, and the surface is parameterized with the free surface configuration.

The horizontal eddy diffusivity is set to \(125\,\hbox {m}^{2}\,\hbox {s}^{-1}\) and the horizontal viscosity coefficient is set to \(1\,\times \,10^{10}\,\hbox {m}^{4}\,\hbox {s}^{-2}\). Convective adjustment is treated by an enhanced vertical coefficient to \(50\,\hbox {m}^{2}\,\hbox {s}^{-1}\) for tracers.

The model grid covers the entire Mediterranean including a small area of the Atlantic, west of Gibraltar up to 11°W. The bathymetry is based on the ETOPO \(5^\prime \times 5^\prime\) database and the Gibraltar Strait is represented by two grid points.

We use exactly the same settings as in Herrmann et al. (2010) and Beuvier et al. (2010). The largest difference with the study by Somot et al. (2006) is that they used the OPAMED8 regional ocean model and ran only one 140-year long simulation.

2.2 Forcings

The setup used here is similar to the one used in the study by Somot et al. (2006) (Fig. 1).

Open image in new window

2.3 Simulations set

In the scenario sensitivity experiments, the origin of the boundary conditions F, R and A are alternatively changed. These different combinations of forcings allow to analyse the sensitivity of the future evolution of Mediterranean water masses to boundary conditions. The sensitivity to river runoff will be assessed through the comparison of A2-F and A2-RF, the sensitivity to the Atlantic hydrography can be analysed by comparing A2-RF and A2-ARF, and the sensitivity to surface fluxes is evaluated confronting A2 and A2-F. Besides, the sensitivity to the choice of the scenario can also be assessed through the comparison of A2-ARF, A1B-ARF and B1-ARF.

HIS is the baseline of the scenario simulation A2, both get surface fluxes from the old version RCM3. The historical simulation HIS-F uses the surface fluxes from the newer version RCM4 and is the baseline of the scenario simulations A2-F, A2-RF, A2-ARF, A1B-ARF and B1-ARF.

2.4 Rivers and Atlantic forcing in scenarios

For the projections, the regional ocean scenario simulations A2-F, A2-RF, A2-ARF, B1-ARF and A1B-ARF (2001–2099 period) use the T/S anomaly field from the global model (GCM2 or GCM3) for the prescription of Atlantic hydrographic conditions at the western margin (Fig. 1). We compute the anomalies by subtracting the mean of the 1961–2000 period from the scenario simulation results. The anomaly profile is then added to the Reynaud et al. (1998) climatology used in the historical simulations. Concerning the rivers, we proceed in the same way as for the Atlantic by adding the runoff anomalies from the stretched ARPEGE-Climate model to the rivers climatology. It must be noted that ARPEGE-Climat includes the land-surface scheme ISBA which collects the runoffs.

In the A2 simulation, we proceed as Somot et al. (2006) using 10-year filtered anomalies for both rivers and Atlantic boundary forcings. This may induce very small changes.

2.5 Initial conditions, spin-up and runs

In the Mediterranean basin, we start the simulations from an initial hydrographic state given by the MEDATLAS-II climatology (MEDAR-Group 2002) for the month of August. In the Atlantic part of the model, we start from the hydrography given by Reynaud et al. (1998). We first performed a 5-year spin-up with a 3-D restoring to the climatologies, so that the velocities can start. Then, we run the model for 15 more years for the historical period 1960–1974. We are aware of the relatively too short spin-up period, but the hydrography of the surface and intermediate waters becomes stable enough after these 20 years with the help of the water flux correction (see Sect. 2.2.1). The stability of the historical simulations is analysed in Sect. 3.4.

After this spin-up, the historical simulations HIS and HIS-F start from 1961 until 2000. The scenario simulations start from 2001 to 2099 from HIS-F, except for A2 which starts in year 2000 from HIS. To make this current study comparable to former studies and to maximize the signal-to-noise ratio and minimize the uncertainty due to natural variability, we compare the period 2070–2099 for the scenario simulations with the period 1961–1990 of the historical simulations. These both time slices are commonly used to evaluate present and future climates.

We are aware that the use of a single model may be a limitation but it also allows a clearer identification of the impact of the different uncertainty sources. The uncertainty related to the choice of the GCM or ARCM or ORCM has been partially tackled in CIRCE (Gualdi et al. 2013; Dubois et al. 2012) and a larger multi-model exercice is currently being organised under the umbrella of the Hymex, Med-CORDEX and Med-CLIVAR programs.

3.1 Surface fluxes

Table 2 compiles the mean value of the freshwater budget components for the simulations and for observational data. The precipitation term of the atmospheric forcing is rather high in comparison with observations (see Sanchez-Gomez et al. 2011), however the observed precipitation over the Mediterranean Sea is very uncertain, especially near the coast where the different observational dataset strongly disagree. Concerning the evaporation, the mean value of our forcing fields is high, but this correlates well with the high precipitation rate displayed by the atmospheric model. ARPEGE-Climat is known to have an enhanced hydrological cycle as analysed in Sanchez-Gomez et al. (2011) and Dubois et al. (2012). This explains the overestimation of P and E.

Mean river runoff is weak, but the ”runoff“ from the Black Sea into the Mediterranean falls into the observations range. The ”E–P–R–B + corr“ value is strictly compensated by the net water inflow through the Strait of Gibraltar due to model setup (see Sect. 2.2.1). The net water flux, the inflow and the outflow through Gibraltar are relatively consistent with the range of observations recorded by Soto-Navarro et al. (2010, 2014). Soto-Navarro et al. (2014) have also recently shown that NEMOMED8 shows good results in representing the volume transport at the Gibraltar Strait despite a resolution which is not sufficient to resolve the small scale processes taking place in the narrow strait. In any case, it must be kept in mind that mixing across the strait is underestimated in Mediterranean climate models due to low resolution and the absence of tidal forcing (Naranjo et al. 2014). Unfortunately, running regional climate models with enough resolution to solve the hydraulic regimes in the Strait (<500 m; Sannino et al. 2014) is currently not feasible.

3.2 Heat, salt and density contents

3.3 Vertical stratification

Figure 3 represents \(IS\) of the MEDATLAS-II climatology and the \(IS\) anomalies of HIS and HIS-F with respect to the climatology. We found HIS-F to present a more realistic vertical stratification than HIS on a global average (Table 5). However, focusing on the western basin, HIS-F is too vertically stratified, especially in the Alboran and Balearic regions (Fig. 3) and HIS performs better, despite a slight lack of vertical stratification in the Gulf of Lions region. The positive biases of HIS-F in the Alboran region could be related to a deficient mixing of the surface Atlantic water passing through the Strait of Gibraltar, which may create a strong surface stratification. In the eastern basin, HIS -F is much closer to the MEDATLAS-II vertical stratification than HIS, the latter presents strong negative biases reflecting the low vertical stratification in disagreement with MEDATLAS-II (Fig. 3).

Open image in new window

3.4 Stability of historical simulations

The achievement of water masses stability in the ocean model is the biggest pre-requisite before starting scenario simulations to make sure that future water masses evolution is exclusively connected to climate change. For computer-time reasons, we did not perform a long spin-up before starting the historical simulations (see Sect. 2.5) neither a control simulation. However, despite a short spin-up, the model trend remains acceptable. Table 3 displays the trend of the Mediterranean heat content \(\varDelta HC\) in \(\hbox {W}\,\hbox {m}^{-2}\). The weak trend modelled for the historical period 1961–1990 is mainly related with a temperature increase occuring during the 1980’s. The trend can also be expressed in  \(^\circ {\mathrm{C}}\,\hbox {y}^{-1}\) which gives the values of \(0.001\,^\circ {\mathrm{C}}\,\hbox {y}^{-1}\) in HIS and \(0.002\,^\circ {\mathrm{C}}\,\hbox {y}^{-1}\) in HIS-F. These trends are three orders of magnitude lower than the expected trend related to climate change (see Somot et al. (2006) and Sect. 4.2).

3.5 Surface circulation

The general dynamic feature of the Mediterranean Sea for the 1961–1990 period is illustrated in the Fig. 4 by the near-surface currents field at 34 m depth. In the model, lower density Atlantic waters enter at Gibraltar, are trapped into gyres in the Alboran Sea (more intense in HIS-F) and then exit, either sticking to the North-African coast (HIS), or in the direction of the Balearic Islands (HIS-F). This latter path is likely unrealistic according to observations (Millot and Taupier-Letage 2005). However, a new study by Pinardi et al. (2013) shows results from a retrospective reanalysis which displays a similar northward flowing segment. Thus the realism of this ”northward“ branching remains under debate. After mixing with the western Mediterranean surface waters, the modified surface Atlantic waters penetrate into the eastern basin through the Strait of Sicily. The main path of the surface current remains along the southern coast but there is a branch into the mid-Ionian after passing the Sicily Strait; in HIS, this upper branch flows more northern than in HIS-F. The main path follows then the eastern coast and split into two branches: one circulates into the Aegean counterclockwise, the other travels westward below Crete. In HIS-F, the Rhodes gyre is well-noticed. Westward from Crete, the surface water penetrates then into the Adriatic with a cyclonic circulation. In the western basin, a wide cyclonic gyre, including the liguro-provencal-catalan current is present in the Gulf of Lions, surrounding the main area of deep water formation. To summarise, we found the general structure of the modelled subsurface circulation to be relatively consistent with the scheme proposed by Millot and Taupier-Letage (2005), except for the circulation in the Alboran Sea in the simulation HIS-F which displays a northward current toward the Balearic Islands, whose existence is debated.

Open image in new window

3.6 THC characteristics

The Mediterranean thermohaline circulation (MTHC) is driven by salinity and temperature differences that induce a vertical circulation in the basin. This vertical circulation cell is triggered by deep and intermediate water formation, which occurs in the Gulf of Lions for the Western Mediterranean Deep Water (WMDW), in the Adriatic and Aegean basins for the Eastern Mediterranean Deep Water (EMDW), depending if we are in a classical configuration or in a EMT-like circulation, and finally in the Levantine basin for the Levantine Intermediate Water (LIW). Note that the shelf cascading is probably not well reproduced due to low resolution and z-level vertical discretization.

There are no direct observations of the intensity of the MTHC. However, some approximation of the dense water formation rate in specific regions can be found in the literature. The winter mixed layer depth (MLD) can also be used as an index to assess the convective areas and the intensity of the convection. From the model outputs, another way to assess the MTHC is to compute the overturning stream functions in \(Sv\) for different basins of the Mediterranean. Whereas the meridional overturning stream function (MOF) is commonly used to assess the strength of the Atlantic thermohaline circulation, the zonal overturning stream function (ZOF) is rather calculated for the Mediterranean. The ZOF has been used by Myers and Haines (2002) in their study to assess the stability of the MTHC, and later by Somot et al. (2006) to study the changes of the MTHC under climate change scenarios. Following this method, we calculate the zonal overturning stream function of the Mediterranean. The ZOF is a meridional integration from the South (\(y_S\)) to the North (\(y_N\)) of the zonal velocity \(u(x,y,z)\), which is then integrated from the bottom of the sea \(h_{bot}(x,y)\) upward. This stream function is given by the following equation:

$$\begin{aligned} ZOF(x,z)\,= -\,{\int _{h_{bot}}^{z}\,\int _{y_{S}}^{y_{N}} \,u(x,y,z) \,\,\,\,dy \,\,dz} \end{aligned}$$

(2)

Figure 5 shows the ZOF for the two historical experiments. The MTHC is characterised by two main circulation cells. The first cell covers both the western and the eastern basins, and has a clockwise circulation with a main path of Atlantic surface water travelling toward the east, and Levantine intermediate water travelling from the east to the west. In HIS, this cell has a maximum of \(1.0\,Sv\) at subsurface depth, between 100 and 200 m depth. In HIS-F, the clockwise cell is weaker (maximum of \(0.9\,Sv\)) and flatter (Fig. 5). These values are generally weaker than in the modelling studies by Somot (2005) and Myers and Haines (2002) (comparison shown in Table 6).

Open image in new window

Table 6

Comparison of the maximum strength of different MTHC cells (in Sv) for the historical simulations and other simulations from the literature

Experiment	Western basin maxima	Eastern basin maxima	Eastern basin minima (Ionian)	Eastern basin minima (Levantine)
HIS	1.0	1.1	−0.6 (2,000 m)	−0.7 (1,500 m)
HIS-F	0.9	0.9		−0.2 (1,000 m)
Somot et al. (2006)	1.5	1.2	−0.6 (1,800 m)	−0.8 (1,300 m)
Myers and Haines (2002)	1.5	0.9	−0.3 (1,000 m)	−0.1 (500 m)

Located in the eastern basin, the second cell shows a deep counterclockwise vertical circulation. It corresponds to the circulation of EMDW toward the extreme east of the basin, this being compensated by a subsurface flow of LIW from the northern Levantine toward the west. In HIS, this cell has a negative maximum of \(-0.5\,Sv\) in both Ionian and Levantine basins. In HIS-F, the deep counterclockwise cell is restricted to the Levantine basin, east of 25°E, and is \(-0.2\,Sv\). In general, HIS-F displays a rather weak deep vertical circulation cell in comparison with former numerical studies (Table 6), whereas HIS has a strong vertical circulation in the east. Despite their large differences, it is quite interesting to have two historical simulations with such contrasted vertical stratification states because it can give some hints about the impact of the initial state on the Mediterranean response to climate change. However, it is worth mentionning that we can not establish a direct link between the vertical stratification and the deep water formation in the Ionian Sea because the EMDW is not formed locally.

The MTHC also has meridional components. It is for example useful to compute the MOF to assess the formation of Adriatic deep water (ADW) and its cascading in the Ionian basin south of the Otranto Strait. This quantity is displayed in Fig. 6 and shows the formation of ADW in the South Adriatic (between 40°N and 42°N). The export of dense water through the strait of Otranto is of \(0.3\,Sv\) in HIS and \(0.2\,Sv\) in HIS-F. South of the Strait of Otranto (south of 40°N), the deepening of the vertical cell represents the cascading of ADW in the Ionian basin where it mixes and becomes EMDW. For this feature, HIS is more realistic that HIS-F where the ADW does cascade a little but does not reach the bottom of the Ionian Sea. Another local MOF can be calculated for the Aegean Sea to define the formation of dense water in this region. Figure 7 displays the MOF for the Aegean Sea. The formation of intermediate water is modeled in the Aegean basin with a transport of \(0.4\, Sv\) in HIS and \(0.3\,Sv\) in HIS-F in the southern part. In HIS-F, the vertical circulation cell is extended until 39°N.

Open image in new window

Open image in new window

The ZOF and MOF features presented before represent a mean state of the deep circulation. However the Mediterranean Sea is characterised by a strong interannual variability in terms of deep water convection, which does not reach the bottom every single year. To assess this variability in the model, Fig. 8 shows time series of the yearly maximum in space and time of daily MLD calculated with a density criterion of \(0.01\,\hbox {kg}\,\hbox {m}^{-3}\) in the sub-basins where dense water forms. This figure shows that the deep convection is more active in HIS than in HIS-F, which is consistent with the IS diagnostics (Fig. 3). When compared with the little amount of observations of the convection in the Gulf of Lions during the past decades, the modelled convection seems too weak in HIS-F and too strong in HIS. In the eastern basin, especially in the Levantine, HIS-F seems to behave better in terms of dense water formation. Besides, both simulations appear to be relatively stable for the period 1961–1999 and show a realistic interannual variability.

Open image in new window

A spatial representation of the areas of dense water formation is represented in Fig. 9 which displays the winter (JFM) MLD averaged over the years 1961–1990. Figure 10 displays the comparison with the climatology by Houpert et al. (2014) and shows that HIS overestimates the winter mean dense water formation at all the main locations, whereas HIS-F compares better with the climatological data in both eastern and western basins.

Open image in new window

Open image in new window

To illustrate the future changes in MTHC, we set an EMT-index to qualify the type of circulation we may face in the future, especially in the eastern basin found to be the most sensitive. The EMT-index characterizes the EMT-situation, when the Aegean is the main contributor for eastern dense water production. The EMT-index corresponds to the difference between the maximums of the Levantine deep negative circulation cell and the Ionian deep positive circulation cell (Fig. 5), and applies only when the Ionian presents an intermediate/deep positive circulation cell, characteristic of the EMT. The highest the index the more we simulate an EMT-like circulation. When the index is 0, the vertical circulation corresponds to the “standard” situation, with a negative intermediate/deep circulation cell in the Ionian. The EMT-index will be used in Sect. 4.2.4 to asses the changes in water masses dynamics in the scenario simulations.

3.7 Model evaluation synthesis

This section aims to discuss the realism of both historical simulations presented in this study and compare our results with other modelling studies. Concerning the heat budget, both simulations are realistic with values of \(-6\) and \(-4\hbox { W}\,\hbox {m}^{-2 }\) and compare well with the value of \(-7\hbox { W}\,\hbox {m}^{-2 }\) by Pettenuzzo et al. (2010) and the value of \(-1\pm 8\) gathered by Sanchez-Gomez et al. (2011). For the net water budget, our simulations perform better than the historical simulation in Somot et al. (2006) and most of the historical simulations by coupled models presented in the study by Dubois et al. (2012) over the same period. Moreover, our values are very consistent with the compilation of observation estimates by Sanchez-Gomez et al. (2011). However the hydrological cycle is too strong, due to an overestimation of P and E by ARPEGE-Climat, both values being far out of the range of the observational estimates and worse than the values in other models presented in Dubois et al. (2012). With respect to heat and salt content, we improve the results compared to Somot et al. (2006), especially for the heat content over the full water column. The averaged SST and SSS have also been improved, especially in the HIS-F simulation. Compared to the historical simulation E20C in Carillo et al. (2012), our both simulations HIS and HIS-F display more consistent heat and salt content values, considering MEDATLAS-II as reference. Concerning the surface circulation, HIS and HIS-F are quite consistent with the circulation pattern proposed by Millot and Taupier-Letage (2005), except for the northward branch from the Alboran Sea to the Balearic Islands displayed in HIS-F. However, this pattern is found in the reanalysis by Pinardi et al. (2013). The subsurface circulation in HIS-F has similar qualities and drawbacks as the stand-alone ocean simulation of Beuvier et al. (2010). In addition, the formation of deep water and the associated MTHC is quite different in our two historical simulations. HIS has a too strong vertical circulation whereas HIS-F has a too weak vertical circulation, especially in the western basin. Because of the lack of long-term observations, the evaluation of the MTHC and of the dense water formation in Mediterranean ocean models is non-trivial. However, we do not claim that our simulations display the most-realistic MTHC but they represent a range of “possible situations”. In summary, the model configuration used here is able to reproduce the main features of the Mediterranean Sea hydrography and circulation with at least similar skills than state-of-the-art models. Therefore, it is expected that the model will be able to reflect the impact of climate change on the functioning of the Mediterranean hydrography and dynamics.

4.1 The SRES scenarios

The socio-economic scenarios used in the present study are based on the Special Report on Emissions Scenarios (SRES). Among the three scenarios which have been performed, B1 is the most optimistic, A2 is the most pessimistic and A1B is intermediate in terms of gases emissions.

The A2 scenario has been developed in the frame of the Third Assessment Report (TAR) of the Intergovermental Panel on Climate Change (IPCC), and is one of the worst scenarios in terms of greenhouse gas emissions. It is based on the following asumptions: a continuously increasing population, a regionally oriented economic development and a world of self-reliant nations.

4.2 Future Mediterranean climate

4.2.1 Surface fluxes

As depicted in Table 2, E–P–R–B increases for all the projections for the period 2070–2099 compared with the historical period 1961–1990. Precipitation, runoff and Black Sea freshwater input tend to decrease whereas the evaporation increases, consistently with the studies by Mariotti et al. (2008), Sanchez-Gomez et al. (2009), Elguindi et al. (2011) and Dubois et al. (2012). To compensate the resulting enhanced water loss, the net water inflow at Gibraltar is intensified. These changes are strongly influenced by the choice of the scenario with the strongest effect under the A2 scenario, and the weakest effect under the moderate scenario B1, A1B being intermediate. As in Somot et al. (2006) and Dubois et al. (2012), we found a decrease of the surface net heat loss (Table 3) in all projections. Note that none of the scenarios shows a surface heat gain in 2070–2099 what means that the atmosphere still extracts heat from the Mediterranean Sea at the end of the twenty-first century. This change varies according to the scenario used, but also strongly depends on the chosen boundary conditions. The latest point will be investigated in Sect. 4.3.

We found a general decrease of the Mediterranean surface potential density (Fig. 11; Table 5) because the increase of the SSTs (Fig. 12) prevails on the increase of SSSs (Fig. 13). This is not true for B1-ARF where the surface density increases because the density gain from the saltening prevails on the density decrease from the warming. In the scenarios, the SSTs become warmer with a range between +1.7 and +3 °C (2070–2099 vs. 1961–1990, averaged over the Mediterranean) and this large spread reflects the sensitivity to the chosen scenario. The SST term has a low sensitivity to the boundary conditions due to the use of the SST relaxation term (Table 4; Fig. 12). Figures 14 and 15 represent the composites of the SST and SSS anomalies maximums and minimums: the largest or smaller anomaly out of the 6 scenario simulations is taken into account at each grid point. These figures show that the warming is not homogenous: the region of the Balearic Islands, the Northwest Ionian, the Aegean and Levantine Seas get warmer than the average (Fig. 14). Concerning the SSS, it generally increases with a range between +0.48 and +0.89 (Fig. 13) and the interannual standard deviation of the historical simulations is 0.07 in HIS and 0.08 in HIS-F. As for the SST, the SSS anomaly signal is not homogenous: the Aegean basin is getting saltier than the rest of the Mediterranean and both Balearic region and North Ionian display a weaker response, even a freshening in some simulations (Fig. 15). The choice of the scenario (experiments A2-ARF, B1-ARF and A1B-ARF) does not induce too strong variations in the SSS anomalies but the choice of the boundary conditions does (experiments A2, A2-F, A2-RF and A2-ARF), especially the choice of the Atlantic salinity (Fig. 13). The warming and saltening of the Aegean are mainly explained by the strong decrease of the net water input from the Black Sea, whereas the other strong patterns mainly refer to subsurface circulation changes (Sect. 4.2.3). In Figs. 14 and 15, it is worth noting that in most of the cases, the SST MAX map corresponds to the scenario A2-ARF, the SST MIN to B1-ARF, SSS MAX to A2-ARF and SSS MIN to A2.

Open image in new window

Open image in new window

Open image in new window

Open image in new window

Open image in new window

4.2.2 Heat, salt and density contents

A warming and saltening are modelled in all the layers of the water column (Table 4), but this signal decreases through depth from an average of +0.7 and +2.6 °C at the surface to +0.4 and +1.0 °C in the 600 m-bottom layer. The penetration of the heat and salt anomalies from surface to depth varies according to the simulation, depending on the changes in convective areas, themselves influenced by the historical state of the vertical stratification and the associated MTHC. The changes of MTHC will be described in Sect. 4.2.4.

The density averaged over the whole water column increases in all the scenario simulations (Table 5). This also applies for the mean density of the layers 150–600 m and 600-bottom. However, the layer 0–150 m displays a density decrease in A2, A2-F and A2-RF: this signal is directly depending on the properties of Atlantic water entering in the Mediterranean which are lighter in these simulations. These density changes lead to changes in the vertical stratification of the basin in the future as displayed by the \(IS\) anomalies (Table 5; Fig. 16). Overall, the simulations with lighter Atlantic waters from GCM2 (A2, A2-F and A2-RF) represent a strong increase of the vertical stratification whereas A1B-ARF, B1-ARF and A2-ARF display a large decrease of \(IS\) in the eastern basin.

Open image in new window

4.2.3 Surface circulation

Figure 17 displays anomalies of near-surface circulation at 34 m depth and shows a clear change in the Balearic region with the penetration of the Atlantic surface water toward the North, along the spanish coast, westward of the Balearic Islands. This feature is new if compared to HIS, or reinforced, when compared to HIS-F. However this change should be interpreted cautiously since the circulation of the historical simulation is not very realistic in this small region. In the Ionian basin, the “F” family simulations display a substantial modification of the trajectory of modified Atlantic waters into the eastern basin, which forms a “North Ionian jet”, similar to the surface circulation changes which occur during the EMT. These two large changes explain both surface warming and freshening (or weaker saltening) of the Balearic region and the North Ionian mentioned in Sect. 4.2.1.

Open image in new window

4.2.4 THC characteristics and water masses dynamics

Figure 5 presents the picture of the MTHC in all the simulations. The MTHC future evolution in the scenarios is strongly related with the state of the vertical stratification in the historical simulations which is weaker in HIS than in HIS-F. The large differences between both historical states provide an interesting diversity for the analysis. We first analyse the results of the “F” simulations, whose atmospheric forcing comes from RCM4.

In the “F” family simulations, despite the general increase of the MTHC, there are strong changes in the areas where dense water is formed: the eastern basin becomes generally more productive and faces a similar situation as during the Eastern Mediterranean Transient. This EMT-like situation happens because the Aegean basin experiences a buoyancy loss in comparison with the historical period, this loss is related to both intensified winds and very low net water inflow from the Black Sea, which reduce the Aegean buoyancy. This situation is clearly depicted in Fig. 5 with the increase of the negative intermediate/deep cell between 27°E and 35°E, which represents the strong outflow of Aegean deep water into the Levantine basin through the straits of Kassos and Karpathos (not shown) as proposed by Beuvier et al. (2010). The intermediate/deep positive cell between 15°E and 27°E represents the outflow of Aegean deep water into the Ionian basin through the Strait of Antikythera (not shown).

To illustrate the EMT situation, we analyse the EMT-indexes (described in Sect. 3.6) averaged over the period 2070–2099 for each simulation (Fig. 18). The highest the index the more we simulate an EMT-like circulation. As expected, the historical simulations have an EMT-index set to 0 Sv, since both simulations depict the expected “standard” circulation, with no positive intermediate/deep vertical circulation cell in the Ionian basin. The “standard situation“ refers to a MTHC situation with a single source of deep water located in the Adriatic for the eastern Mediterranean. In the “F” family scenarios, the index increases with a range between 1.01 and 2.86 Sv. In the simulation A2, the index stays at 0 Sv with a vertical circulation which weakens but keeps the standard configuration. We aim to relate the EMT-index to the Aegean buoyancy. Figure 18 represents the relation between the EMT-index anomalies (scenario vs. historical period) and the Aegean buoyancy fluxes anomalies. From HIS to A2, the buoyancy fluxes of the Aegean basin increase, whereas they decrease from HIS-F to all the “F” scenario simulations. An Aegean full buoyancy decrease clearly pushes toward an EMT situation and is mainly expected in the future Mediterranean ocean climate (as seen in A2-F, A2-RF, A2-ARF, A1B-ARF and B1-ARF).

Open image in new window

In Fig. 5, the negative very deep cell in the Ionian basin (between 15°E and 27°E) represents the deep circulation of the ADW which flows through Otranto, and cascades southwestward into the Ionian. The scenario simulations generally face a situation where the density difference between the ADW and the northwest Ionian water increases (in comparison with the historical period), therefore, the cascading depth of the ADW is increased from 1,265 m in HIS-F to a range between 1,861 and 2926 m in the “F” scenarios (Fig. 6). However, Fig. 6 also shows that the intensity of the circulation cell from the Adriatic into the northern Ionian weakens (but in B1-ARF) and tends to be restricted to the extreme North of the Ionian basin in A2-ARF and A1B-ARF.

From the “F” simulations analysis, we can state that we head towards an EMT-circulation in the future, but the ADW production is still active: the ADW is produced with a slightly lower intensity but the cascading goes deeper and is confined in the high northern Ionian due to the presence of enhanced Aegean outflow (negative cell in Fig. 6 for A2-ARF and A1B-ARF).

Concerning the changes occurring in the western basin, the dense water production in the Gulf of Lions decreases by the end of the century in all the scenario simulations “A2“ but remains similar to the historical state in A1B-ARF and increases in B1-ARF whose vertical stratification diminishes in this area (Figs. 16, 19). For the western Mediterranean, we can establish a direct link between the vertical stratification in the Gulf of Lions and the deep water production since the WMDW is formed locally.

Open image in new window

If we focus on the differences between HIS and A2 (atmospheric forcing comes from RCM3), we find a decreasing MTHC in both eastern and western basins (Fig. 5), confirming results obtained by Somot et al. (2006) who used the same forcing but another ORCM and found a decrease in the surface density associated with a weakening of the thermohaline circulation, except for the Adriatic. This significative difference of future changes is related to the fact that the HIS simulation has a low vertical stratification and is very convective (especially in the eastern basin), due to very strong winter heat loss compared to HIS-F in the historical period. However, despite a decrease of the MTHC in A2 compared to its historical period, the high-convective initial state of HIS allows to bring the surface hydrographic anomalies to deeper layers from the beginning of the A2 simulation on. This explains the large anomalies found in the deepest layer despite moderate surface anomalies compared to the ”F“ family of scenario simulations.

4.2.5 Sea level changes

The model configuration does not consider changes in the Atlantic sea level, so from this ensemble we cannot estimate future changes of total sea level. However, the thermosteric component of sea level (the one related with expansion/contraction of the water column) can be estimated from the ensemble. In the former sections, we focused on changes in the formation of the water masses and their hydrographic properties in the future. These large changes have a strong impact on the mean thermosteric sea level (Fig. 20) which results from the heat content changes over the full water column. For the 2070–2099 period, a mean increase between 34 and 49 cm is simulated. This large spread is explained by differences in the convective areas and in the intensity of the convection, leading to differences in the transfer of the surface heat to the deepest part of the basin, thus impacting the basin heat content and the thermosteric sea level. It is important to highlight that this range of values (34–49 cm) only refers to the thermosteric component of sea level. To recover total sea level other components should be included (e.g. mass addition due to land ice melting, changes in the Atlantic circulation,...). From the recent 5th IPCC report (IPCC 2013), it seems that those components will be positive adding between 15 and 30 cm to the thermosteric component.

Open image in new window

4.3 Sensitivity to boundary conditions

This section aims to identify and quantify the impact of the different boundary conditions on the obtained results for a given scenario. We examine the sensitivity of the future Mediterranean Sea to the prescribed hydrography of the Atlantic water, the river runoff and the surface fluxes. For that purpose, we only focus on A2 scenario simulations. In each of the analysed simulations, the combination of boundary forcing changes in such a way that we can isolate the sensitivity to each type of boundary forcing, by comparing simulations one to one. The uncertainty range explored in the present study can be small compared to the large spread displays among other regional and global models regarding the future changes in Atlantic hydrography, river runoffs and atmospheric fluxes. This point is discussed in Sect. 4.6.

4.3.3 Choice of the Atlantic hydrography: A2-RF versus A2-ARF

The impact of the hydrographic properties of the Atlantic water on the oceanic state of the Mediterranean is assessed through the comparison of the simulations A2-ARF and A2-RF, whose Atlantic hydrographic conditions come from different GCM versions, GCM2 and GCM3 respectively. Figure 21 displays the anomalies of the Atlantic hydrographic properties in both simulations. Future Atlantic waters which enter in the Mediterranean basin are warmer and saltier than in the historical period. This signal is nevertheless larger in A2-ARF which displays a warming and saltening respectively 0.5 °C and 0.40 larger than in A2-RF for the 0–150 m layer of Atlantic waters for 2070–2099. This results in a smaller density decrease in A2-ARF (about \(-0.1\,\hbox {kg}\,\hbox {m}^{-3}\)) than in A2-RF (about \(-0.25\,\hbox {kg}\,\hbox {m}^{-3}\)), considering the same ocean layer (0–150 m) and the same time period (2070–2099 vs. 1961–1990). The hydrographic changes of the Atlantic buffer zone are advected through the Mediterranean, which displays mean SSS changes of +0.89 in A2-ARF and +0.64 in A2-RF for 2070–2099 (Table 4).

Open image in new window

A2-RF and A2-ARF display quite different vertical circulation changes in the eastern Mediterranean. A2-ARF projects a more intense EMT situation for the future than A2-RF (Figs. 5, 18) because A2-RF has a larger surface density decrease from the incoming Atlantic water, thus a stronger vertical stratification than A2-ARF (Fig. 16). For the same reasons, the meridional overturning stream function of the Adriatic basin displays a deeper cascading of the ADW into the north Ionian in A2-ARF than in A2-RF (Fig. 6) and the reduction of dense water production in the Gulf of Lions is quite larger in A2-RF (Table 7; Fig. 19). This leads to large differences in the propagation of warming from the surface toward depth. For the same surface warming anomaly of 2.97 °C, A2-ARF has a more active transfer than A2-RF, as displayed by the heat content anomalies in the 600 m-bottom layer with the values of +1.14 °C for A2-ARF versus +0.68 °C in A2-RF (Table 4).

The hydrography of the Atlantic water is thus substantially impacting the new organization of water masses production. The changes are weaker in A2-RF whose lower surface density allows a stronger vertical stratification than in the simulation A2-ARF.

4.4 Sensitivity to the socio-economic scenario

The comparison between the simulations A2-ARF, A1B-ARF and B1-ARF provides some answers on the sensitivity to the choice of the socio-economic scenario. Table 4 shows that the surface salinity and temperature changes averaged over the Mediterranean basin are larger in A2-ARF than in A1B-ARF and B1-ARF, consistently with the greenhouse gases concentrations in each scenario. However, both heat and salt content changes over the full water column are found to be larger in A1B-ARF than in A2-ARF, reflecting a larger propagation of the surface heat anomaly to deeper layers in A1B-ARF. A1B-ARF is indeed having the strongest EMT-index among the whole family of scenario simulations (Fig. 18) but also keeps a deep cascading of ADW (Figs. 6, 18) and its maximum of MLD in the Gulf of Lion is not decreasing (unlike A2-ARF). All these features make that the simulation A1B reproduces the strongest vertical circulation at the end of the twenty-first century (Fig. 5), inducing thus the strongest advection of the surface changes toward deep layers. B1-ARF, as the moderate scenario, displays the weakest surface hydrographic changes, and the largest decrease of vertical stratification compared to its historical period. It shows a small increase of the dense water formation in the Gulf of Lions (Fig. 19), its EMT signal is the weakest among the 3 scenarios, and it maintains a strong ADW formation with an intense cell and a deep cascading into the Ionian basin (Figs. 5, 6).

4.5 Impact of the socio-economic scenario versus boundary conditions

Concerning the changes in thermosteric mean sea level for the 2070–2099 period (Fig. 20), \(\varDelta \hbox {Atl}\) displays a spread of 11 cm, with an increase between 34 and 45 cm. This large difference is driven by a weaker penetration of the heat anomaly in the A2-RF simulation, due to a weaker eastern convection compared to A2-ARF. \(\varDelta \hbox {Scen}\) has a spread of 9 cm with a range between 40 and 49 cm. \(\varDelta \hbox {Riv}\) and \(\varDelta \hbox {Atm}\) show respectively a spread of 4 and 3 cm.

This results highlight the relevance of the choice of boundary conditions for Mediterranean ocean projections. The choice of Atlantic conditions seems to be of highest importance and prevails on the choice of the socio-economic scenario.

4.6 Discussion on uncertainties among other models

In order to place our results in a wider context, we compare the uncertainty ranges explored in our study for each of the three boundary conditions with the large spread among global and/or regional models. From the study by Marcos and Tsimplis (2008), 10 GCMs used in CMIP3 show a large spread of changes in Atlantic hydrography computed for the buffer zone of NEMOMED8 model for the layer 0–150 m. The changes vary from −0.35 to +0.66 for salinity, and from −0.06 to +2.56 °C for temperature. This gives a range of 1.01 in salinity and 2.62 °C in temperature, which is larger than the uncertainty range used in our study (0.40 for salinity and 0.5 °C for temperature). Sanchez-Gomez et al. (2009) analysed Mediterranean runoff changes for the 2070–2099 period among various regional climate models and found a runoff decrease ranging from −5 to −43 %, and changes in Black Sea fresh water input from +25 to −102 %. In our study, the uncertainty ranges from −5 to −32 % for rivers and −25 to −61 % for the Black Sea input, thus is smaller than the ”possible“ range of uncertainty. For the total fresh water budget, Sanchez-Gomez et al. (2009) found changes from +20 to +60 % whereas our study investigates changes from +35 to +60 %. Finally, the study by Dubois et al. (2012) compiles model data from the CIRCE project with net heat budget changes ranging from +2 to +\(5.5\,\hbox {W}\,\hbox {m}^{-2}\). In our study we propose a smaller range of uncertainty which is between +1 to +\(3\,\hbox {W}\,\hbox {m}^{-2}\). The uncertainty range explored in our study is about 2–3 times smaller than the spread displayed among other regional and global models regarding the projected changes in Atlantic hydrography, river runoffs and atmospheric fluxes. Therefore, the spread of our ensemble must be considered as a lower bound for the actual range of uncertainties in Mediterranean Sea projections.