Future projections of the surface heat and water budgets of the Mediterranean Sea in an ensemble of coupled atmosphere–ocean regional climate models

Abstract

Within the CIRCE project “Climate change and Impact Research: the Mediterranean Environment”, an ensemble of high resolution coupled atmosphere–ocean regional climate models (AORCMs) are used to simulate the Mediterranean climate for the period 1950–2050. For the first time, realistic net surface air-sea fluxes are obtained. The sea surface temperature (SST) variability is consistent with the atmospheric forcing above it and oceanic constraints. The surface fluxes respond to external forcing under a warming climate and show an equivalent trend in all models. This study focuses on the present day and on the evolution of the heat and water budget over the Mediterranean Sea under the SRES-A1B scenario. On the contrary to previous studies, the net total heat budget is negative over the present period in all AORCMs and satisfies the heat closure budget controlled by a net positive heat gain at the strait of Gibraltar in the present climate. Under climate change scenario, some models predict a warming of the Mediterranean Sea from the ocean surface (positive net heat flux) in addition to the positive flux at the strait of Gibraltar for the 2021–2050 period. The shortwave and latent flux are increasing and the longwave and sensible fluxes are decreasing compared to the 1961–1990 period due to a reduction of the cloud cover and an increase in greenhouse gases (GHGs) and SSTs over the 2021–2050 period. The AORCMs provide a good estimates of the water budget with a drying of the region during the twenty-first century. For the ensemble mean, he decrease in precipitation and runoff is about 10 and 15% respectively and the increase in evaporation is much weaker, about 2% compared to the 1961–1990 period which confirm results obtained in recent studies. Despite a clear consistency in the trends and results between the models, this study also underlines important differences in the model set-ups, methodology and choices of some physical parameters inducing some difference in the various air-sea fluxes. An evaluation of the uncertainty sources and possible improvement for future generation of AORCMs highlights the importance of the parameterisation of the ocean albedo, rivers and cloud cover.

1 Introduction

The Mediterranean Sea basin encompasses a large number of regions over much of western and central Europe and North Africa. Its climate is influenced by mid-latitude and sub-tropical regimes and by the surrounding complex orography. The Mediterranean region is one of the regions described as a “hot spot” under global warming (Giorgi 2006; Intergovernmental Panel on Climate Chang—IPCC 2007). This region is also characterized by intense human activities and an increasing interest in the local climate, and the effect of climate change is fundamental from a socio-economic point of view.

Previous studies have tried to quantify the impacts and changes under climate change scenarios over the Mediterranean and European regions in the framework of the EU Projects: PRUDENCE (Christensen and Christensen 2007) and ENSEMBLES (Christensen et al. 2007). Those studies are based on global atmosphere–ocean general circulation models (AOGCMs) or on atmospheric only regional climate models (RCMs). The AOGCMs remain too expensive to run long term simulations at very high resolution. They omit small-scale features and processes that characterize the region, thus the use of RCMs provides a more realistic representation of regional scales features (e.g. Giorgi and Bates 1989; Bärring and Laprise 2005; Elguindi et al. 2009). The spatial structures of the Mediterranean region are better represented using high resolution regional models. Indeed, the low resolution or the atmospheric only models are unable to well represent the local thermal and dynamical processes over the Mediterranean Sea. The higher resolution in RCMs directly impacts the air-sea exchanges with a better representation of the temperature, humidity, wind and other hydrological parameters of potential relevance in impact studies (Sotillo et al. 2005; Herrmann and Somot 2008; Herrmann et al. 2011; Dell’Aquila et al. 2012). However, the use of standalone RCMs inhibits the representation of processes linked to high frequency and high resolution interactions between the atmosphere and the presence of the Mediterranean Sea. The climate projections done with RCMs use sea surface temperatures (SST) provided by a low resolution AOGCM and do not reproduce small scale spatial structures over the Mediterranean Sea. They are imposed as the lower boundary conditions of the RCMs thus do not respond to the surface air temperature above it and thus do not take into account any air-sea feedbacks. In particular, this set-up can have SSTs warming the surface air temperature which is incoherent over the Mediterranean Sea (Sanchez, personal communication).

The Mediterranean Sea is an active semi-enclosed marginal sea and communicates with the Atlantic Ocean through the narrow strait of Gibraltar. Its unique configuration makes it an ideal basin to investigate hydrological and heat budgets. It has the particularity of having some deep convection zones and a steady state thermohaline circulation (Wüst 1961; Robinson et al. 2001). Atlantic waters (AW) entering through the strait of Gibraltar in the upper layer have a low salinity and a warm temperature (36.2 PSU, 15.4°C on average) and are coming from the Atlantic. Whereas waters leaving the Mediterranean Sea are relatively more saline and cooler (38.4 PSU, 13°C on average) and outflow at depth through the Strait into the Atlantic Ocean (Bryden et al. 1994; Tsimplis and Bryden 2000). Estimations from observations of the net water transport through the strait of Gibraltar into the Mediterranean Sea range from 0.038 to 0.09 Sv (1 Sv = 10⁶ m³/s) (Bryden and Klinder 1991; Bryden et al. 1994; Tsimplis and Bryden 2000; Candela 2001; Baschek et al. 2001; García Lafuente et al. 2007; Soto-Navarro et al. 2010).

Estimation of the net heat transport through the strait of Gibraltar into the Mediterranean Sea has been obtained by using mooring-based measurements and range from 3 to 10 W/m² (Béthoux 1979; Bunker et al. 1982; McDonald et al. 1994). Saline waters outflowing from the strait of Gibraltar are then found in the Atlantic ocean at a depth of 1,000 m and might have an impact on the global ocean circulation (Reid 1979; Béthoux et al. 1999; Potter and Lozier 2004; Artale et al. 2005; Millot et al. 2006). The characteristics of the Atlantic surface water are changing as the water is traveling to the eastern side of the sea. Overall, the evaporation over the Mediterranean basin is stronger than precipitation and runoff combined. The heat loss is greater than the solar uptake at its surface, leading to a cooling and an increase in the salinity of the surface waters. The deep or intermediate convection of the seawater occurs in selected areas: the Gulf of Lions, the Levantine, Aegean and Adriatic basins, where favorable oceanic conditions and air-sea interactions are present (Tsimplis et al. 2005; Robinson et al. 2001).

Recent observations over the Mediterranean Sea show an increase in the evaporation and a decrease of the precipitation rate leading to an enhanced freshwater deficit (Mariotti 2010; Xoplaki et al. 2006). The unique characteristics of the Mediterranean Sea and the recent trend highlight the importance of including a realistic Mediterranean Sea in the climate system. This could have important consequences on the quality and reliability of the climate change projections over the region. Giorgi and Lionello (2008) show a substantial precipitation reduction and a warming of the Mediterranean region enhanced over the summer months at the end of the twenty-first century using climate projections from AOGCMs and RCMs. Mariotti et al. (2008) show that the average climate projections from the World Climate Research Program Coupled Model Intercomparaison Project Phase 3 (CMIP3) have an increase in freshwater loss over the Mediterranean Sea. Sanchez-Gomez et al. (2009) found similar results using RCMs from the ENSEMBLES project. Both studies found a freshwater decrease for the period 2070–2099 related to the 1950–1999 period, equal to a mean reduction of 40% with similar trends but the magnitude of the change between individual models differs significantly. Thus uncertainties still remain on the projection of the water budget. Recently, two studies used coupled AORCMs to evaluate the Mediterranean climate. The first study, Somot et al. (2008) compared a coupled and an uncoupled experiment and showed that in the coupled simulation the climate change signal was generally amplified over large parts of Europe. The second study by Artale et al. (2009) simulated the ERA40 period using an AORCM. Their AORCM was able to simulate a variability of the Mediterranean SST, winds as well as surface fields in line with observations. All those models are constrained by a net positive heat transport through the strait of Gibraltar implying a net negative compensating sea surface heat flux inside the Mediterranean Sea. For example, Somot et al. (2008) obtained a negative net surface heat flux of −7.1 W/m² and a deficit for the water budget of 2.1 mm/day (Li et al. 2006, 2011). AORCMs are expected to improve the description of SST patterns with respect to the coarse resolution global driver (Dell’Aquila et al. 2012) and thus to better simulate ocean–atmosphere feedbacks thereby providing more confidence in present and future climate changes over the Mediterranean region. Within the CIRCE project (Gualdi et al. 2011, in revision) five AORCMs have been developed in order to obtain a realistic representation of the climatology and evaluation of the signal of future climate change in the region. In this study, for the first time, the heat and water budgets components of high resolution AORCMs is evaluated over the Mediterranean Sea in present day climate. Results will be compared with a large set of observations over the sea. The importance of the choice of the model parameters, the design and the modeling methodologies adopted In the considered AORCMs are discussed and the uncertainties evaluated. Then the evolution of the heat and water budgets in the future climate over the Mediterranean Sea is assessed. Those fluxes are consistent between the SSTs and the surface air temperature and are constrained by the net entering flux at the strait of Gibraltar. They rise many important issues and some recommendations are to be addressed for future projects such as HyMeX (Hydrological cycle in the Mediterranean Experiment, Drobinsky and Ducrocq 2008) and MedCORDEX (MEDiterranean COordinated Regional climate Downscaling Experiment, Ruti et al., EOS, submitted).

Section 2 describes the coupled models used within the CIRCE project. Section 3 gives the methodology and a validation of the present day fluxes over the Mediterranean Sea in comparison with available datasets. Section 4 looks at the evolution of the heat and water budgets over the Mediterranean Sea under the climate scenario. Summary and conclusions are discussed in Sect. 5.

2 Description of the models and experiments

2.1 Models

Within the framework of CIRCE, five AORCMs have been developed. The five AORCMs are set up by different institutes and in the rest of the paper referred to by the name of that institute: CNRM (Météo-France, Centre National de Recherches Météorologiques), LMD (Laboratoire de Météorologie Dynamique), MPI (Max Planck Institute for Meteorology), ENEA (Italian National Agency for New Technologies, Energy and the Environment) and INGV (Istituto Nazionale di Geofisica e Vulcanologia). Table 1 summarizes some characteristics of the different components and resolutions used in the different coupled models. They have a relatively high resolution atmosphere component over the Mediterranean region using either a stretched/global atmospheric high resolution model or limited-area regional model over this region. They are coupled interactively with a high resolution ocean model over the Mediterranean Sea with the OASIS3 coupler (Valcke 2006). When the atmospheric model is global, the rest of the global ocean is represented by a global low resolution ocean model. The heat and salt fluxes and water masses can be exchanged through the strait of Gibraltar between the two ocean models. When the atmospheric model is regional, boundary conditions from a global atmospheric model are needed as well as fluxes and water masses at the strait of Gibraltar. The exchanges with the Atlantic Ocean evolve interactively with the global ocean for the CNRM, INGV and LMD models. The others (MPI and ENEA models) are using characteristics from a global low-resolution coupled model (IPCC 2007).

Table 1 Summary of the different model components and characteristics for each of the CIRCE model

Reproducing a physically reasonable exchange for the strait of Gibraltar via numerical models is not straightforward. As suggested by the recent numerical modeling studies of Sannino et al. (2004, 2007, 2009a) and Sanchez-Garrido et al. (2011), who successfully reproduced most of the aspects of the exchange including hydraulic control, the minimum requirements necessary for a model to simulate properly the small scale nature of the exchange are: a horizontal resolution of about 0.5 km and a vertical resolution of about 10 m, the inclusion of explicit tidal forcing, and physical parameterizations taking into account diapycnal mixing and entrainment. However, at present the explicit representation of such small-scale processes in regional Mediterranean models for climate studies is beyond the capabilities of current available computer resources. In particular, in the five CIRCE regional models the strait of Gibraltar is represented by few grid points and vertical levels, and relatively simple parameterizations for entrainment and mixing processes are adopted. In particular: ENEA uses two 1/8°◦ grid boxes in latitude, corresponding to about 27 kms horizontal resolution, 11 vertical levels and a flat bathymetry limited to 280 m. MPI has 2 grid points in the narrowest area of the Strait, at a resolution of 1/10°km, 14 vertical levels and considers as minimum depth of the Strait of 270 m. CNRM and LMD use a stretched grid in the region of the Strait that is represented in the narrowest region by two grid boxes in latitude, corresponding to about 17 km, 10 vertical levels and considers as minimum depth 103 m. INGV uses 3 grid points in the narrowest region at a resolution of 1/16° corresponding to about 20 km, 36 vertical levels and 188 m as a minimum depth. Concerning specific physical parameterization for entrainment and mixing processes applied at the strait: INGV uses an up-stream advection scheme for active tracers, whilst in the rest of the basin a MUSCL (Monotonic Upwind Scheme for Conservation Laws, Van Leer 1979) scheme is adopted. The vertical diffusivity in this area and below 30 m depth is increased in order to parameterise unresolved processes induced by vertical mixing. In the same geographical area the bottom friction drag coefficient is linear and five times larger than in the other parts of the model. ENEA increases the vertical diffusion 10 times with respect to the rest of the model in the Strait area, and replaces the horizontal bi-Laplacian viscosity with a Laplacian viscosity. MPI, CNRM, and LMD do not apply any specific parameterization.

Another important component of the coupled system, in particular for the representation of the water cycle, is the river runoff. Its representation in the coupled system is done by a river routage or imposed climatology at the rivers’ mouths as a simple river. Another water input into the Mediterranean Sea is the Black Sea. It is considered either as a simple river calculated from the evaporation, precipitation and river runoff over the Black Sea or as a climatology (Stanev et al. 2000) or as an explicit Black Sea with a flow through the Dardanelles strait, the rivers from the Black Sea catchment basin are outflowing in the Black Sea.

2.2 Experiments

Before starting the simulation, the coupled system has to be in equilibrium. The different components of the AORCMs coupled system are initialized with observations. The global and regional atmospheric models are initialized with the reanalyses. The Mediterranean Sea models have their spin-up carried out using Levitus or MedAtlas (MEDAR/MEDATLAS Group 2002) as initial conditions. The forcing at the strait of Gibraltar is provided by the output of a global ocean model or by observations. Then, all the different components of the AORCMs are coupled together and a spin-up integration is carried out over some decades using the concentrations and distribution of atmospheric greenhouse gases (GHGs) and aerosols conditions of 1950. After, the coupled system has reached a reasonable stable state, the present climate simulation is carried out over the period 1950 to 2050. During the twentieth century, the simulations use observed GHGs and aerosols up to the year 2000. During the twenty-first century, the simulations use the concentrations and distribution of GHG and aerosols specified by the A1B hypothesis provided by the IPCC-SRES (Nakicenovic et al. 2000). The multi-model approach is preferred as the climate change signal in all scenarios is still weak by the year 2050 and the uncertainty associated to the model is stronger than the signal.

3 Validation

3.1 Observations datasets

Different sets of observations are used to validate the mean characteristics of the present day climate over the sea in order to validate the different components of the water and heat budget over the Mediterranean Sea (Dubois et al. 2010; Sanchez-Gomez et al. 2011). The NOCS (National Oceanography Centre, Southampton) dataset adapted for the Mediterranean Sea provides surface fields for the shortwave, longwave, sensible, latent, evaporation fluxes, specific humidity and sea surface and air temperatures at a 1° × 1° resolution from January 1980 to December 2004 (Berry and Kent 2009; Sanchez-Gomez et al. 2011). The HOAPS (Hamburg Ocean Atmosphere Parameters and fluxes from Satellite) dataset provides surface fields for the sensible, latent and evaporation fluxes and precipitation rate at a 0.5° × 0.5° resolution from September 1987 to December 2005 (Andersson et al. 2007). The OAFlux (Objectively Analyzed air-sea Flux) dataset provides data for the sensible, latent and evaporation fluxes at a 1° × 1° resolution from January 1958 to December 2008 (Yu et al. 2008). The CMAP (CPC Merged Analysis of Precipitation) and GPCP (Global Precipitation Climatology Project) are precipitation rate datasets at a 2.5° × 2.5° resolution from January 1979 to July 2008 or April 2009 respectively (Xie and Arkin 1997; Adler et al. 2003). The runoff for the Mediterranean and Black Seas are provided over the period 1960–2000 and 1923–1997 respectively (Ludwig et al. 2009; Stanev et al. 2000). The albedo is evaluated from satellite observations collected by ISCCP (Zerefos et al. 2009). The winds dataset used is the QuikSCAT level 3 with a 25 km resolution (ftp://podaac.jpl.nasa.gov/pub/ocean_wind/quikscat/L3, Perry 2001). The different observation datasets cover different periods of time and have not necessarily the same length as the model outputs. The error estimate associated with the observation dataset is calculated by t_n−1,α/2*σ/${\sqrt { ( {\text{n}} - 1 )} }$, where the t value is used in a 95% confidence interval given in a table of the t-student distribution depending of the sampling size, σ is the inter-annual standard deviation and n the number of years covered by the dataset.

3.2 Calculation of heat and water budgets components

The different components of the Mediterranean Sea heat and water budgets are computed for the 5 CIRCE simulations. Over the period 1961–1990 the global warming is neglected over this period, the heat and water budgets over the Mediterranean Sea should be compensated by the net positive inflow at the strait of Gibraltar (Bethoux and Gentili 1999). All models were integrated over multi year integration to reach a quasi-state equilibrium. This equilibrium of the heat and water budgets should be satisfied over the Mediterranean Sea with the coupled AORCMs. Thus the heat and water budgets over the Mediterranean Sea can be assessed as follow.

The heat budget (HB) can be decomposed at the sea surface into its different terms:

$$ {\text{HB}} = {\text{netSW}} + {\text{netLW}} + {\text{LH}} + {\text{SH}} \approx - {\text{HG}} $$

(1)

where the net shortwave is: netSW = SW_down + SW_up and the net longwave is: netLW = LW_down + LW_up), LH is the Latent Heat and SH is the Sensible Heat. By convention, the downward fluxes are positive and the upward fluxes negative as they are considered as heat gain and loss for the Mediterranean Sea respectively. HG is the net positive heat flux at the Gibraltar strait. The heat content of the Mediterranean Sea is assumed to be constant. Over the 1950–2000 period, it has increased as little as 0.4 W/m² according to Rixen et al. (2005).

Each component is spatially averaged over the Mediterranean Sea domain using the original grid and land-sea mask for each model taking into account the surface of each mesh. It should be pointed out that the land-sea mask have mixed grid points close to the coast with a fraction of sea and a fraction of land for the LMD, MPI and INGV. This particularity has been very carefully considered for each term to ensure that fluxes received by the ocean are in facto the sum of each individual term. The grid points have only 100% sea points or 100% land points for the CNRM and ENEA models.

The freshwater budget (WB) at the sea surface can be estimated by:

$$ {\text{WB}} = {\text{P}} - {\text{E}} + {\text{R}} + {\text{B}} \approx - {\text{WG}} $$

(2)

where P is the precipitation, E is the evaporation, R is the river runoff from the Mediterranean catchment basin and B is the input from the Black Sea. WG is the freshwater input at the Gibraltar strait assuming that the water content over the Mediterranean Sea stays constant. Downward fluxes are considered positive.

3.3 Description of the model differences concerning the processing of the river runoff

3.3.1 The catchment basins

This paragraph describes the uncertainty associated with the design of the river routing scheme used in the RCMs. To quantify the freshwater discharge from the rivers into the ocean, different river routing schemes with their associated catchment basin are defined. The river network from the major, minor and coastal rivers are outflowing the water from precipitation and evaporation over land at the actual coastal mouth and ocean basin. The most commonly used river routine scheme in GCMs is TRIP (Oki and Sud 1998). Another definition of the different catchment basins for the Mediterranean Sea is provided by Ludwig et al. (2009). Figure 1 shows the catchment basin of the Mediterranean and Black Sea region for the two different river routine schemes. The Nile river catchment basin is excluded as it is heavily dammed and extensively exploited for irrigation purposes, infiltration in swamps (Nixon 2003). The surface area of the Mediterranean Sea is 2.45 × 10¹² m² and its associated catchment basin is of similar size. The surface area of the Black Sea is 4.36 × 10¹¹ m² and its associated catchment basin is 2.4 × 10¹² m² using the Ludwig catchment basins definition. The Mediterranean Sea river catchment is very narrow along its coastline as many mountainous regions surround it. The Black Sea river catchment extends over a large area of central and eastern Europe, it is several times larger than the Black Sea where it discharges. The domains of the catchment basins provided by the two river routing schemes have a lot of similarities but also many important differences that need to be highlighted. Whereas with the TRIP river routine, the Mediterranean catchment basin does not cover a small part of the coast in the south west of France as well as along the eastern shore of the Adriatic Sea, those regions are covered and included in catchment basin with the Ludwig river catchment definition. Those two coastal regions are very narrow, but are subject to very heavy precipitation events. For example, the Tet river which is known to generate strong flooding event discharges in the Mediterranean Sea in Southern France. This river discharges in the Atlantic with the TRIP river routine scheme but in the Mediterranean Sea with the Ludwig routine scheme. Another important difference is in the western part of Turkey near Greece, the catchment region there outflow in the Black Sea for the TRIP routine scheme and in the Mediterranean Sea for the Ludwig river routine scheme. The catchment basin over the Libyan region is also different but this region has very little precipitation and the associated river runoff is quasi null.

Fig. 1

Catchment basin for the Mediterranean (purple) and Black (red) Seas with the TRIP routine scheme (left) and Ludwig (right). The catchment from the Nile has been removed

In the CNRM and LMD models, the river runoff is not interactive to the coupled system and it is imposed as a climatology, however over their land the river runoff is still calculated. To evaluate the impact of this non-conservative coupling in those two models, the river discharge into the Mediterranean and Black Seas are calculated using the surface runoff with those two available catchment basins. The uncertainty associated with the two possible catchment basins is evaluated (Table 2). Over the Mediterranean Sea, the river discharge in both models is 20% weaker with the TRIP catchment basin than with the Ludwig catchment basin over the period 1961–1990. On the other side, over the Black Sea, is it 5–15% larger with the TRIP catchment basin than with the Ludwig catchment basin over the period 1961–1990. When the catchment basin used is TRIP instead of Ludwig less water flows into the Adriatic Sea and more in the Black Sea and vice versa. The Ludwig catchment basin is more realist than the TRIP catchment basin. Those differences should be taken into account in the new generation of coupled regional models.

Table 2 Mean and standard deviation river discharge for the LMD and CNRM models into the Mediterranean and Black Seas with the TRIP and Ludwig catchment basins over the 1961–1990 period (the Nile catchment has been excluded)

3.3.2 Nile River

In the previous paragraph, the Nile River was not integrated in the Mediterranean runoff, because of its particularity. Spatially, the Nile catchment basin extends south of the Victoria lake, it has a surface area of about 2.9 × 10⁶ km² and its waters are highly used. Before 1968, the year when the Aswan dam was completed, the Nile river was one of the major freshwater input from river runoff into the Mediterranean Sea (Nixon 2003). Nowadays, its flow is highly anthropogenised and strongly reduced. The mean flow from observations is evaluated at around 2,800 m³/s before 1968 and at around 1,150 or 1,800 m³/s afterwards depending of the dataset (Vörösmarty et al. 1996; Ludwig et al. 2009; Dümenil Gates et al. 2000). Other estimates give an annual discharge over the 1984–2001 period of about 415 m³/s (Egyptian Ministry of Water Resources and Irrigation 2002).

Within the CIRCE models, the catchment of the Nile river is either fully represented such as in the INGV, LMD and CNRM models (however, the last two do not have interactive rivers), or the catchment basin is only represented over a smaller domain covered by the regional atmospheric models (up to 20°N for the ENEA model and 24°N for the MPI model). As the INGV model is the only model which covers the whole catchment basin and has an interactive river routing scheme, the consequences of the size of the catchment basin on the runoff are evaluated in this model. Using the INGV model, the river runoff is calculated over a catchment basin extending up to 24°N, 20°N or over the whole catchment basin. The river flow found is about two orders of magnitude smaller for the reduced catchment basin (Table 3). The strong differences in the runoff are due to the fact that the region of heavy precipitations and strong runoff is not included in the smaller domains. Table 3 summarised the Nile flow calculated from the river runoff over the catchment basin for each model and the actual runoff received by the ocean.

Table 3 Mean and standard deviation of the Nile River runoff calculated over the Nile catchment basin and the runoff sent to the ocean for the different models over the 1961–1990 period

Figure 2 shows the Nile runoff receive into the ocean by the different models and by the observations (Vörösmarty et al. 1996; Ludwig et al. 2009). The runoff from observations decreases from 1964 when the dam construction started until 1968, when the construction ended. As the INGV model is the only model which covers the whole catchment basin the damming effect has to be considered. Before 1968, the runoff of the Nile River is about 3,000 m³/s, afterwards, a correction based on the observations is applied on the calculated runoff to outflow at a rate of about 1,500 m³/s. The exceeding water is distributed equally over the all Earth surface as precipitation. The decline in the runoff is slightly delayed compared with the observations. The CNRM and LMD models do not have interactive rivers thus the outflow into the ocean is constant over the period 1950–2050 taken from the observations after the construction of the dam. The inter-annual variability and trends under climate change forcing are not take into account.

Fig. 2

River runoff of the Nile for the different CIRCE models and from observations (Vörösmarty et al. 1996; Ludwig et al. 2009; Dümenil Gates et al. 2000)

Different studies suggested that the drastic changes in the Nile runoff have impacted the salt content in the Eastern Mediterranean Sea and its water masses (Skliris and Lascaratos 2004, 2007). The different models are not reproducing correctly the inter-annual variability of the observed Nile river runoff, either because of the reduced size of their catchment basin or because they do not have interactive river runoff. Only the INGV model is taking into account the anthropogenic use of the river using a correction of the flow over the twentieth century. All the other models have to weak values or take a climatology after the construction of the dam.

3.3.3 Black Sea

Another important freshwater contribution into the Mediterranean Sea is the Black Sea. The variability of its freshwater input into the Aegean sea is considered as a possible cause of the changes in the Aegean water characteristics and thus as a driving mechanism of the Eastern Mediterranean transient (EMT) (Zervakis et al. 2004). The Black Sea discharge (generally positive) in the Mediterranean Sea results from the balance between the runoff (R) over its catchment basin, the evaporation (E) and the precipitation (P) over the Black Sea assuming no sea level change. This water flux outflow is then controlled through hydrological constraints at the narrow Dardanelles strait, Marmara Sea and Bosphorus straits.

Within the CIRCE models, different representations of the Black Sea flow are used (Table 1). The water budget is calculated from the P + R − E fields over the Black Sea region to obtain the Black Sea inflow into the Mediterranean Sea for the INGV and ENEA models (for the ENEA model, the calculated flow is corrected with the output of a simulation forced by the ERA40 reanalysis performed). For the CNRM and LMD models, the Black Sea outflow is calculated from observations (Stanev et al. 2000) and in the MPI model, by a fully represented Black Sea basin connected through a flow at the Dardanelles strait. The Black Sea outflow varies between 0.26 to 0.85 mm/day depending on the models relative to the Mediterranean Sea (Table 4). The MPI and ENEA models overestimate the Black Sea outflow compared with observations. The surface area of the Mediterranean Sea is equal in the different model to: 2.53 × 10¹² m² for the CNRM, 2.39 × 10¹² m² for the LMD, 2.59 × 10¹² m² for the ENEA, 2.47 × 10¹² m² for the MPI, 2.54 × 10¹² m² for the INGV.

Table 4 Mean and standard deviation of the Black Sea flow into the Mediterranean Sea (relative to the surface of the Mediterranean Sea) for the different models over the 1961–1990 period expressed in mm/day

3.4 Mean characteristics over present climate

The mean characteristics of the heat and water budgets simulated for each model are studied over the period 1961–1990, considered as the present day climate, and are compared with different sets of observations.

3.4.1 Heat budget

Over the period 1961–1990, the basin averaged seasonal cycle of the different terms of the heat budget is evaluated and compared with different sets of observations described in Sect. 3.1 (Dubois et al. 2010; Sanchez-Gomez et al. 2011). The net surface fluxes are very similar between the different models with the heat gain in the summer months directly link to the annual cycle of the shortwave and with the heat lost in the winter months link to the annual cycle of the latent and sensible fluxes (Fig. 3). The LMD and ENEA models have a shortwave stronger than the NOCS observations. The CNRM model has its shortwave close to the observations, whereas the MPI and INGV models have a weaker radiation flux than the observations. The longwave has the highest inter-model variability both in averaged values and in phase. ENEA and CNRM models are characterized by an annual cycle having an opposite behaviour than the observations. The heat transport through the Gibraltar strait is always positive in all models with a maximum in the summer months due to maximum surface Atlantic water temperature in that season. The LMD model has the strongest annual cycle (Fig. 3).

Fig. 3

Seasonal cycle for the total heat flux, the Gibraltar heat transport and different terms of the heat budget for the different simulations and in grey the spread given by the different observations datasets over the 1961–1990 period

The observations dataset for the latent flux are from NOCS, HOAPS and OAFLUX. NOCS and OAFLUX have a similar seasonal cycle close to that of most models. The maximum in August–September is only found in the HOAPS dataset. Those differences within the datasets can be explained partly by the differences in the spatial resolution and coverage over the Mediterranean Sea. The HOAPS dataset is omitting large regions over the Mediterranean Sea such as the Adriatic and Aegean Seas where strong winds occurs (Dubois et al. 2010). In the models, the latent flux is the weakest at the end of spring and beginning of the summer. The maximum is reached in November–December in all models when strong winds and weak specific humidity occur. Only the ENEA model has a second maximum in August (Romanou et al. 2010). The sensible flux obtained by the different models are in the range of the observations (NOCS, HOAPS and OAFLUX datasets).

The different fluxes of the heat budget show strong differences between the CIRCE models. Different thermodynamics quantities representing the inherent physics of each model can help to explain some of those differences. The radiative fluxes of the heat budget are composed of upward and downward fluxes, which can be decomposed as:

For the shortwave;

$$ {\text{Q}}_{\text{sw}} = {\text{Q}}_{\text{sw}} \left( {\text{down}} \right) + {\text{Q}}_{\text{sw}} \left( {\text{up}} \right) = {\text{Q}}_{\text{sw}} \left( {\text{down}} \right)\left( { 1-\alpha } \right) $$

(3)

where α is the surface effective albedo. Q_sw(down) depends mainly on the cloud cover. Using the Bulk formula (Bignami et al. 1995), the longwave flux can be decomposed as:

$$ \begin{aligned} {\text{Q}}_{\text{lw}} & = {\text{Q}}_{\text{lw}} \left( {\text{down}} \right) + {\text{Q}}_{\text{lw}} \left( {\text{up}} \right) \\ & = [\sigma {\text{T}}_{\text{A}}^{ 4} \left( {0. 6 5 3 { } + \, 0.00 5 3 5 {\text{ e}}_{\text{A}} } \right)]\left( { 1 { } + \, 0. 1 7 6 2 {\text{ C}}^{ 2} } \right)-\varepsilon \sigma {\text{T}}_{\text{S}}^{ 4} \\ \end{aligned} $$

(4)

where σ is the Stefan-Boltzmann constant, ε is the ocean emissivity usually equal to 1 in models. T_S is the sea surface temperature, T_A is the air temperature, e_A is the atmospheric vapour pressure and C is the cloud cover.

Then, the turbulent fluxes (latent and sensible fluxes) can be decomposed as:

$$ {\text{Q}}_{\text{latent}} = - \rho_{\text{A}} {\text{L}}_{\text{E}} {\text{C}}_{\text{E}} \left| {\overrightarrow {V} } \right|{(q_{S} - q_{A} )} $$

(5)

where $\left| {\overrightarrow {V} } \right|$ is the magnitude of the wind speed, ρ_A is the density of the moist air, L_E is the latent heat of vaporisation, C_E is the turbulent exchange coefficient for humidity (depends on the wind speed), q_A is the specific humidity of the air and q_S the specific humidity at saturation at sea surface temperature T_S. This last term depends on T_S and the dependence on air pressure is small and considered constant at 1013.25 mb (Buck 1981). The magnitude of the wind speed $\left| {\overrightarrow {V} } \right|$ is calculated from the 6 h outputs of the meridional and longitudinal components then temporal average is carried outand

$$ {\text{Q}}_{\text{sensible}} = - \rho_{\text{A}} {\text{C}}_{\text{P}} {\text{C}}_{\text{H}} |\mathop{V}\limits^\rightarrow| (T_s-T_A)$$

(6)

where C_P is the specific heat capacity and C_H is the turbulent exchange (depending on the wind speed) coefficient for temperature.

The radiative fluxes depend mainly on the albedo, cloud cover, T_A, T_S and e_A (Table 7). In all models the cloud cover is underestimated and the minimum cloud cover is during the summer months and the maximum in the winter months (Fig. 4, top left). Only the MPI model overestimates the cloud cover during the summer months. The ENEA model has a strong seasonal cycle in its cloud cover and overestimates the cloud cover in the winter months. The other variable that have an effect on the shortwave is the albedo. The seasonal cycle of the albedo is very different for each model (Fig. 4, top right). The MPI, INGV, CNRM and ENEA have a constant or nearly constant seasonal cycle. Only the LMD model has an seasonal cycle very close to the observations. The MPI model has the strongest albedo and the INGV model has the weakest. The net shortwave depends on the albedo and on the cloud cover. The combined effect of the strong cloud cover and strong albedo for the MPI model can explain the weak shortwave. In spring, the CNRM model has a good value for the albedo and low cloud cover thus the SW is overestimated.

Fig. 4

Annual cycle for the cloud cover, albedo, T_S and T_A for the different simulations over the 1961–1990 period over the Mediterranean Sea and in grey the observations

The longwave depends on the T_S, T_A and C. The T_S and T_A have an annual cycle reaching their maximums in the summer months and its minimum in the winter months (Fig. 4, bottom respectively left and right). All models are simulating temperatures (T_S and T_A) that are cooler than the observations. The LMD model has the coolest temperatures.

The formulation of the albedo is found to be very different from one model to another and is defined either as a constant over the ocean, or depending on the zenith angle or also including a diffusive albedo term. This diffusive albedo is able to take into account evolution of the cloud cover: its inter-annual variability and changes over the twenty-first century. The albedo between the different models varies between 0.03 and 0.1 (Table 5). The impact of the different value and annual cycle of the albedo can be assessed by comparing the SWnet obtained by the model and the calculated one using Eq. 3 with the mean albedo over the period 1961–1990 (Table 6). The impact of the albedo on the net shortwave is not negligible. The difference in the shortwave between an albedo at 0.03 (INGV model) and 0.1 (MPI model) is 15.5 W/m² on the annual average and 22 W/m² for the month of July. A difference in the albedo values between 0.065 and 0.1 can have an impact of 7 W/m² (respectively 12 W/m²) on average (respectively in July). The MPI model has the weakest netSW. Th choice of albedo formulation accounts for 25% of this low value.

Table 5 Albedo over the ocean for the different models(

Table 6 The SWnet obtained by the LMD model and the calculated SWnet with different albedo values using the mean SWdown over the 1961–1990 period (mean SWdown = 220.6 W/m² over period 1961–1990, maximum in July: 331.0 W/m²)

The turbulent fluxes depend on the wind speed, the temperature gradient and the specific humidity gradient between the sea surface and the air above it (Eqs. 3 and 4, Table 7). The mean annual cycle of the wind and gradients of temperature and humidity are plotted in Fig. 5 for different simulations over the 1961–1990 period. The wind field over the Mediterranean Sea reaches a maximum in the winter months when the local northerly winds (Mistral, Bora, Tramontane) are dominant (Perry 2001). Only the ENEA model has a small second peak in July as found in the observations. It could correspond to strong Etesian winds over the Aegean Sea at this time of the year. The specific humidity gradient between the air and the sea surface differs among the models. In the observations, the annual cycle reaches a maximum in September and a minimum in May. The LMD reaches its minimum in May–June and its maximum in October. The amplitude of its annual cycle is weak. The CNRM has its minimum in April–May and its maximum between July to October. The ENEA has its maximum in September. The temperature gradient between the surface and the ocean show very similar annual cycle in all the models except for the LMD model, which has a stronger annual cycle of the temperature gradient reaching a maximum in the winter months and a minimum in the summer months. In late spring-summer, the LMD, MPI and INGV models have a sea surface temperature cooler than the air surface temperature, although in the observations the sea remains warmer than the air during the whole year (well captured in the CNRM and ENEA models). The weak amplitude of the seasonal cycle of the wind and the difference in humidity in the ENEA model explains that the latent heat flux reaches a maximum in August–September. The CNRM model has a similar seasonal cycle for the difference in humidity but the seasonal cycle of its wind field is much stronger in winter inducing strong latent heat during this season. The seasonal cycle of the sensible heat flux is strongly controlled by the difference in T_S–T_A in all models.

Table 7 Mean wind speed, specific humidity, T_S, T_A and cloud cover for the different models over the period 1961–1990 compared with observations

Fig. 5

Annual cycle for the wind speed, the difference between QS–QA and between TS–TA for the different simulations over the 1961–1990 period and in grey the observations

To be able to explore the role of each individual variable (T_A, T_S, wind, albedo and cloud cover) in controlling the error in the annual cycle of the different component of the heat budget compare with the observation (Kothe and Ahrens 2010). The annual cycle of the error for the explaining variable and for each fluxes is computed. Thus, the squared correlation coefficient between the temporal error in the annual cycle between each parameter and their associated fluxes has been plotted (Fig. 6). The combine effect of the error of each predictor on the error of the flux is considered by taking into account the variance of each predictors (right group in Fig. 6). For most of the fluxes, 70% of their errors in the annual cycle are explained by the combined effect of the individual predictors. The errors in the annual cycle of the shortwave are linked to the error in the annual cycle of the cloud and of the albedo in the ENEA and MPI models and only to the error in the annual cycle of the albedo for the LMD model. The errors in the annual cycle of the longwave are correlated with the error in the annual cycle of the T_S, especially for the LMD, INGV and ENEA models. For the turbulent flux, the errors in the annual cycle of the latent flux are mainly correlated with the error in the annual cycle of the wind especially for the MPI, CNRM and ENEA models. The errors in the annual cycle of the sensible flux are linked to the error in the annual cycle of the difference between T_S and T_A. Each model has a different sensibility to the error of different predictors on the surface fluxes. A particular predictor cannot be isolated and this underlines the difference in the model physics and the associated errors also combine with the biais of the global driver.

Fig. 6

Temporal averaged explained variances of surface fluxes errors attributable to the cloud cover, albedo, Ta, Ts and humidity over the ocean (in black: MPI, in red: CNRM, in yellow: LMD, in green: INGV and in blue: ENEA)

Very recent studies evaluated the total heat budget over the Mediterranean Sea from observations (Sanchez-Gomez et al. 2011; Pettenuzzo et al. 2010). In both studies They gave estimates of the total net heat budget on the assumption that it should compensate the net positive inflow through the Gibraltar strait known as the closure hypothesis. Sanchez-Gomez et al. (2011) used the NOCS and HOAPS dataset and obtained a total net heat flux of –1 ± 8 W/m². Pettenuzzo et al. (2010) adopted a correction method on the ERA40 reanalysis (1958–2001 period) and obtained a flux of −7 W/m². Looking at the single components of the budget, while the above estimates agree very well for the LH and SH components, they differ for the SW and LW values. The heat budget is thus compared to this observations for each CIRCE model over the period 1961–1990 (Table 8). A net negative surface heat flux using an AORCM has already been found in the study by Somot et al. (2008). However, among all the ARCMs from ENSEMBLES project, only the PROMES and the KNMI models (Sanchez-Gomez et al. 2011) found a total net heat budget in the range of the observations, whereas the others vary between −40 and +20 W/m². Instead, within the CIRCE models, all models have a negative net surface heat flux varying between −2 and −6 W/m², well within the range of the observations.

Table 8 Averaged value ± the error associated for the period 1961–1990 for the components of the Mediterranean Sea heat budget in W/m²

Such a coherent description of the total heat budget among CIRCE models is a significant achievement, indicating ocean–atmosphere coupling as a necessary methodology for physically consistent climate simulation over the Mediterranean Sea. Compare to the ARCMs used in ENSEMBLES, all models are coupled with an ocean model which are constraint at the strait of Gibraltar. It is worth noting that the surface heat flux is almost balanced by the lateral heat flux at the strait of Gibraltar (Table 8). All models have a net positive heat flux entering into the Mediterranean Sea with are within the range of the observations. The flow at the strait is control by its geometry and also by the density gradient between the incoming and outgoing currents. The incoming transport is larger than the outgoing transport to compensate for the water loss over its surface. The incoming transport enters warm surface Atlantic water and a net positive heat flux through the strait of Gibraltar. The ENEA model has a strong incoming transport with warmer Atlantic water than observations (not shown) constraining the ocean model with a net heat flux of 6 W/m². The CNRM model has cold Atlantic water compare with observations (not shown) constraining the ocean model with a weak net heat flux of 2 W/m².

If the Mediterranean Sea is considered by two homogenous mixed layers filled with water of constant temperature (Calmanti et al. 2006). In a steady state, the heat flowing into and out of the surface layer is done by surface fluxes, transport through the strait of Gibraltar and by convection and/or diffusion with the deep ocean. The convection of dense water in the deep ocean in taking place at some localised regions: Gulf of Lion (MEDOC group 1970; Schott et al. 1996; Herrmann et al. 2009), the Southern Adriatic Sea (Artegiani et al. 1997a, b), Levantine basin (Lascaratos et al. 1993) and the Aegean Sea (Roether et al. 1996). On average the volume transport into the deep is equivalent to 1 Sv or about 2 W/m². In the steady state equilibrium, the deep ocean as a constant temperature implying that the convective flux is compensated by a diffusive flux. The surface layer has a ventilation time of about 10 years and it reaches its equilibrium through negative feedback of the surface temperature on the surface flux (longwave, latent and sensible fluxes). All models have positive net heat flux at the strait of Gibraltar imposing thus in all models a net negative heat flux at the surface over the Mediterranean Sea.

Such a closure mechanism is missing in ARCMs that are instead constrained to adjust towards prescribed SSTs by adjusting the fluxes according to the respective physical parameterisation. The importance of the strait of Gibraltar as a driver for the coupled model equilibrium was also pointed out in the study by Dell’Aquila et al. (2012) who were able to obtain realistic SST patterns inside the Mediterranean Sea with the same ENEA model considered here only after de-biasing the temperature and salinity applied in the Atlantic box of the ocean model.

In spite of the agreement of the net heat budget, the spread between individual terms of the heat budget highlights that the equilibrium is obtained by different compensations between the fluxes in each model. In particular, a spread of 54 W/m² for the SW, of 30 W/m² for the LW, of 9 W/m² for the SH and of 20 W/m² for the LH is found. None of the model agrees with observation estimates for all of the different terms. This highlights inherent differences between the model physics, which have thus a strong impact on the surface fluxes as already highlighted by Sanchez-Gomez et al. (2011). Reconciling such differences implies an improvement of physical parameterisations in the models.

3.4.2 Water budget

Similarly to the heat budget, the various components of the Mediterranean Sea water budget are studied. The seasonal cycle for the precipitation rate exhibits a maximum in the winter months (Fig. 7, top left). All models simulate a precipitation rate within the range of the observations. However, the precipitation rate obtained by the observations in the HOAPS dataset is half the precipitation rate obtained by GPCP or CMAP dataset. The CNRM model has the strongest precipitation rate up to 3 mm/day over the winter season (Jacob et al. 2007; Somot et al. 2008). Those strong values are in the range of the observations provided by the GPCP dataset (Adler et al. 2003), but those datasets are considered to overestimate the actual rate (Dubois et al. 2010). The evaporation rate is directly correlated to the latent heat flux (Fig. 7, top right). It reaches a maximum over the winter season and a minimum during summer months.

Fig. 7

Annual cycle for the total water flux, the Gibraltar water flux and the different terms of the water budget for the different simulations and in grey the spread given by the different observation datasets and their associated error bars in mm/day

The observed river runoff is characterized by a winter and spring maxima (in January and April) and a summer minimum (in September), (Ludwig et al. (2009) (Fig. 7, bottom left). The river runoff peaks during the spring, a few months later than the peak in precipitations. This delay is mainly due to the melting of the snow accumulated during the winter months in mountainous regions. The simulated river runoff in the different models peaks in March and in most models the runoff is very weak in the summer months. The runoff decreases from March probably due to an early snow melting or a small snow cover. The summer months are characterised by a weak river runoff probably due to the fact that the aquifers and water storage in the ground are not represented in the regional coupled model (Hagemann and Jacob 2007; Habets et al. 2008). The LMD model has a very strong runoff with a strong variability in the seasonal cycle. In this model as well as in the CNRM model the river runoff plotted here is not the one used for the ocean, which get a climatological runoff equal to: R = 0.3 mm/day (Ludwig et al. 2009). For the CNRM model, the annual mean climatological runoff is of the same order as the mean annual runoff estimated from the atmospheric model, so no imbalance is introduced. By contrast, in the LMD model, the mean annual runoff estimated from the atmospheric model is twice the prescribed mean annual runoff to the ocean. If the rivers runoff were coupled to the ocean in the LMD model. The Mediterranean Sea would have a weak salinity.

The Black Sea inflow into the Mediterranean Sea is obtained by two different observations datasets. Stanev et al. (2000) estimates this inflow either by the Black Sea water budget (P − E + R) with the runoff over the Black Sea catchment basin, the evaporation and the precipitation or by a direct measurement of the flow at the Dardanelles strait (Stanev et al. 2000; Stanev and Peneva 2002). The first estimation of the Black Sea inflow reaches a maximum in April–May and a minimum in August linked to the minimum in precipitation and runoff. The second estimate of the Black Sea direct inflow has a much weaker annual cycle. The difference is the annual cycle between the two estimates is due to the sea level changes of the Black Sea which is controlled through the hydrological constraint of the strait of Dardanelles. The strait of the Dardanelles acts as a bottle neck and only a certain volume can pass through the strait. The ENEA, LMD, INGV and CNRM models have a coherent annual cycle with the P − E + R estimate. They peak at the end of the winter season and reach a minimum in summer. They usually overestimates the mean freshwater inflow. The freshwater inflow simulated by the MPI model has a much reduced annual cycle. Its seasonal variability is similar to the flow estimated at the strait. The MPI model is the only model with an interactive Black Sea, the inflow from the Black Sea into the Mediterranean is found to be improved by the coupling with a Black Sea model. It allows to take into account the delayed seasonal cycle due to the strait physical constraint. It should be noted that the CNRM and LMD models have the Black Sea freshwater inflow imposed as a climatology.

Previous studies have investigated the water budget over the Mediterranean Sea from observations. Sanchez-Gomez et al. (2011) obtained a best estimate of −1.7 mm/day over the Mediterranean basin for the water budget choosing HOAPS for E and P and Ludwig and Stanev for R and B respectively. Pettenuzzo et al. (2010) obtains +1.4 mm/day for the precipitation and −3.2 mm/day for the evaporation, but they are not considering the runoff. Struglia et al. (2004) gives an estimate of 0.28 mm/day for the Mediterranean river runoff. Estimates from OBS1 and OBS2 agree very well for E but differ for the P value. Table 9 shows the values averaged over the period 1961–1990. The models have very different values for the net water budget, which would have different influence on the salinity over the Mediterranean Sea. It should be noted that the water flux transferred to the ocean is different for the CNRM and LMD models (the river runoff and Black Sea outflow are imposed to a climatology equals to 0.52 mm/day in total). Among the 5 CIRCE models the spread between the models is equal to 0.3 mm/day for P, 0.7 mm/day for E, 0.4 mm/day for R, 1.0 mm/day for B. Again this indicates that improvements concerning the physics of the model are still required.

Table 9 Averaged value (±standard deviation) over the period 1961–1990 for the components of the Mediterranean Sea water budget, in mm/day

For the Mediterranean basin another component of the water budget is represented by the water exchanged through the strait of Gibraltar. In Table 10 are presented the computed annual mean for the Atlantic water inflow, Mediterranean water outflow, and net flow for the period 1961–1990, as well as the most recent observed values from Soto-Navarro et al. 2010. By convention the inflow is positive and the outflow is negative. From Table 10 appears evident that the simulated inflow and outflow of ENEA and LMD models are overestimated respect to the observations; in particular ENEA model overestimate both flows of about 20%. However, it is noteworthy to remember that the representation of the Strait geometry has an effect on the magnitude of the exchanged flows (Sannino et al. 2009b), thus the coarse resolution adopted by ENEA model (1/8°) can explain such discrepancies. For the remaining models, (MPI, CNRM, and INGV models) the inflow and outflow is in very good agreement with observations. Examining the net flow emerges that all the models simulate a mean flow close to the observations, except for MPI that simulates a value too low. For all the models, the net water flow balances the water loss from the surface of the entire basin (except for the MPI model) (Table 9).

Table 10 Averaged value (±standard deviation) over the period 1961–1990 for the outflow (OUT), inflow (IN),and net (NET) water transport through the Strait of Gibraltar in Sverdrup (Sv)

In Fig. 8 is shown the climatological monthly mean, computed over the period 1961–1990 for the 5 models. In agreement with the observations, all the models, except for MPI model, display an evident seasonal behaviour for the net flow, with minimum values (positive, except for INGV model) during the period February–May and maximum values peaking around August–September. Differently from the observations, the climatological monthly mean inflow and outflow appear almost locked in phase in all the 5 models. The outflow, in agreement with observations, has a maximum value around April and minimum values in autumn in all the models except for LMD model that shows an opposite cycle. The inflow simulated by the models does not show the observed maximum during summer, except for LMD model, that however shows the same maximum also for the outflow.

Fig. 8

Annual cycle for the different terms of the water budget for the different simulations and in grey the spread given by the different observation datasets and their associated error bars in Sv

4 Future changes

4.1 Heat budget

The evolution of the different terms of the heat budget for the near future period 1991–2020 and the future period 2021–2050 are listed in Tables 11 and 12. Despite the spread in the different terms of the heat and water budget in present climate, all the models give a qualitatively similar response for the period 2021–2050 compared to the period 1961–1990. By 2021–2050, all the changes are statistically significant in all models whereas over the period 1991–2020 only the CNRM and LMD models have significant changes in all the different terms of the heat budget. Each term of the heat budget is changing under the climate change scenario. The SW is weakly increasing due to a small reduction of the cloud cover over the Mediterranean Sea. The LW loss is decreasing: the LWdown increases (under the greenhouse gas effect) is larger than the LWup increase due to increasing SSTs). The SH loss is decreasing due to the fact that the air-sea temperature gradient decreases as the atmosphere warms faster that the ocean, whereas the winds are not found to change over the Mediterranean. The LH loss is increasing because the atmosphere can contain more humidity in a warmer atmosphere under climate change and as the Mediterranean land climate becomes drier. The LMD model has the strongest response and sensitivity to the increase in greenhouse gases. The MPI model has the weakest sensitivity. This can be explained by the fact that the LMD model has the strongest increase in the SST and the MPI model has the weakest increase in the SST. Thus, the surface net heat loss decreases during the twenty-first century leading to a weaker cooling of the ocean by the atmosphere. With the ENEA and LMD models, the net heat flux even becomes positive, which implies that the atmosphere starts to warm the Mediterranean Sea. This gain in heat for the Mediterranean water is added to the positive heat gain from the Gibraltar strait which is also increasingly positively during the twenty-first century. Figure 9 shows that the evolution in the fluxes under climate changes forcing are continuous and have a strong interdecadal variability. The ENEA and LMD models have the strongest temporal variability. The LMD model has the fastest and strongest response. It is interesting to notice that the temporal variability of the MPI and the INGV models is very similar for all the fluxes. This is due to the fact that the boundary conditions for the MPI model are provided by the INGV model, which has a global model for the atmosphere and for the ocean. Using regional coupled climate model, coherent air-sea fluxes are obtained over the Mediterranean basin in all models. It is the first time that an ensemble of AORCMs shows that the SST evolution is consistent with the atmosphere evolution above it and the response under a warming climate is showing equivalent trend in all the models. Considering the independent of the biais of each individual term, the trend under the climate change scenario is always in the same direction for each individual term. The response only depend on the external forcing from the greenhouse gases.

Table 11 Anomalies of the heat budget, shortwave, longwave, latent and sensible fluxes over the Mediterranean Sea for the 1991–2020 period as compared to the 1961–1990 period for all the different models (in W/m2)

Table 12 Same as Table 11 but for the period 2021–2050

Fig. 9

Anomalies of the heat budget, shortwave, longwave, latent and sensible fluxes over the Mediterranean Sea compared to the 1961–1990 period (in W/m²). A centered 5-year running mean smoothing is applied

4.2 Water budget

Similarly to the heat budget, the changes in the water budget are calculated for the different terms in Tables 13 and 14 for the near future period (1991–2020) and the future period (2021–2050). The net water loss to the atmosphere increases during the twenty-first century for all the models. The changes are nearly all statistically significant for the period 2021–2050 with a decrease in P, R and B and an increase in E in all the models. The Black Sea runoff is decreasing except for the MPI model. The MPI model is the only model with an interactive Black Sea. The increase in the LMD model is larger in the near future probably due to a high decadal variability (Fig. 10). The decrease in the precipitation is a well known feature of the climate change impact over the Mediterranean region (Gibelin and Déqué 2003; Giorgi 2006; IPCC 2007; Somot et al. 2008). The increase in the evaporation is linked to the latent flux. The trend found in the different components is in agreement with previous studies over the Mediterranean Sea (Somot et al. 2006, 2008; Mariotti et al. 2008; Sanchez-Gomez et al. 2009). The river runoff over the Mediterranean basin is decreasing in all models but this trend can be amplified through damming and anthropogenic.

Table 13 Anomalies of the water budget, precipitation, evaporation, runoff and Black Sea over the Mediterranean Sea for the 1991–2020 period compared to the 1961–1990 period for all the different models (in mm/day)

Table 14 Same as Table 13 but for the period 2021–2050

Fig. 10

Anomalies of the water budget, precipitation, evaporation, runoff and Black Sea over the Mediterranean Sea compared to the 1961–1990 period (in mm/day). A centered 5-year running mean smoothing is applied

5 Discussion and conclusions

5.1 Discussion

Within the CIRCE project, an ensemble of coupled AORCMs is used to simulate the future climate change over the Mediterranean region. The impact of global warming on the Mediterranean heat and water budgets is investigated. For the first time, an inter comparison of the net surface heat and water budgets in high resolution coupled atmosphere–ocean models is analysed under an A1B climate scenario. So far very few studies have systematically examined the heat and water budgets of climate models over the Mediterranean Sea. Among these, Mariotti et al. (2008), Sanchez-Gomez et al. (2009) and Somot et al. (2008) agree In predicting increasing water loss in either GCMs, ARCMs and AORCMs respectively.

In this study, the ensemble mean net surface heat loss is −4 W/m² and the ensemble mean net surface water loss is –0.95 mm/day over the period 1961–1900. Those values are consistent with the positive heat and water transport through the strait of Gibraltar. Notice that for the first time in regional climate models, the SSTs can be modified both by the atmosphere above it and by the ocean below. Therefore physically consistent air-sea fluxes can be derived.

A striking result of this inter-comparison exercise is the consistent description of the total heat budget among models in spite of a substantial disagreement in some of the individual components. We argue that this consensus is a result of the explicit role of the heat and mass flux at the strait of Gibraltar as a driver of the model equilibrium. In particular, differently from atmospheric only RCM where SST are prescribed and ocean–atmosphere fluxes must adjust accordingly, in the case coupled models the surface flux at equilibrium is indirectly prescribed by imposing the conditions for the heat/mass exchange at Gibraltar.

A better consensus between the design of the coupled model could reduce the differences and match better with the observations. Some of those differences are highlighted and future recommendations can be suggested.

• The surface albedo over the ocean can be defined by a simple constant, and should be dependent on the zenith angle or the combined effect of a direct and diffuse albedo should be added to add a dependence on the zenith angle. The choice of the values calculated by the models can have an impact of more than 15 W/m² on the net shortwave. The error in the shortwave seasonal cycle is very correlated with the error in the albedo. Only the albedo of the LMD model has a seasonal cycle, which follows closely the observations. This should be a standard in future regional climate models.

• For the water budget, the river runoff is tested using the two different catchment basins: the TRIP (Oki and Sud 1998) and the Ludwig (Ludwig et al. 2009). When using the TRIP catchment basin, the Mediterranean runoff is 20% weaker than with the Ludwig catchment basin. This difference can be explained by the exclusion of some coastal rivers in south west of France or along the eastern side of the Adriatic Sea, which should be carefully included in future coupled AORCMs.

The spatial resolution of the RCMs is increasing and thus extreme events will be more accurately represented. The choice of the catchment basin is important, its impact is not negligible on the river outflow, especially in regions where extreme precipitation events occur. In a fully coupled system, those freshwater inputs influence the local salinity at the river mouth and the water content of the sea where it outflows. The variability and changes in the freshwater impact the salt content of the water masses, they are considered as the possible cause of the Eastern Mediterranean Transient (EMT) (Skliris and Lascaratos 2004, 2007; Zervakis et al. 2004; Beuvier et al. 2010).

• Another river input into the Mediterranean Sea is the Nile river, which is very anthropogenised through irrigation and damming (Nixon 2003). All the limited area regional climate models are not representing its entire catchment basin implying a runoff 100 times weaker than the observations when the domain stops at 24°N. Only the INGV model covers the full catchment basin of the Nile and thus applies a correction from the year 1968 on the river runoff to represent the impact of the Aswan dam. This anthropogenic effect should be applied in all regional climate models as its decreasing effect is equivalent to a freshwater reduction of 0.1 mm/day over the Mediterranean basin and this reduction should be gradual from 1964 to 1968 period when the dam has been constructed. Some studies suggested that the changes in the Nile runoff have impacted the salt content in the Eastern Mediterranean Sea and its water masses (Skliris and Lascaratos 2004, 2007).

• Another important freshwater input is the Black Sea, which is represented in different ways within the different CIRCE models. Only, the MPI model so far has an interactive Black Sea connected to the Mediterranean Sea through the Dardanelles strait. The other models are calculating the P − E + R budget over the Black Sea catchment basin or are using climatology at the location of the Dardanelles strait into the Mediterranean Sea. The inflow obtained with an interactive sea has much weaker fluctuations of its annual cycle, closer to the observations measured at the strait. The Dardanelles strait acts as an hydrological constraint on the Black Sea freshwater input to the Mediterranean Sea. For regional climate models, which are not representing the Black Sea interactively, the P − E + R budget should be corrected by estimating the hydrological constraint on the flow.

• The errors in the seasonal cycle of the different fluxes are strongly linked in most models to the error in the seasonal cycle of the SSTs. Even if the evolution of the SSTs under climate change are coherent with the atmosphere above it. All the simulated SSTs are too cold as compared to the observations. This bias can be linked with the different representation of the flow at the strait of Gibraltar or to bias in the atmospheric boundary conditions from global models or in the ocean global models.

5.2 Conclusions

Under the climate change scenario A1B, all simulations show a significant change for the period 2021–2050, the changes are not necessarily significant for the period 1991–2020. Even if the spread between each individual term can be high, all terms follows the same trend under the global warming. The total net heat flux loss is decreasing and even some models obtain for the period 2021–2050 a positive net surface heat flux which comes in addition to the positive heat flow at the strait of Gibraltar. The water budget has its water loss increasing; it is still in the same uncertainty range as found by Mariotti et al. (2008) in GCMs and Sanchez-Gomez et al. (2009) in RCMs. The spread between each term of the surface fluxes in present climate can be high; it underlines the different physics in the models or set up of the models. The MPI model has the weakest sensitivity in each individuals terms of the budget whereas the LMD model has the strongest. However, we argued that the consensus in the total surface heat flux between all the model is related to the strong constraint imposed at the strait of Gibraltar. This can only be achieved using AORCMs.

To conclude, the heat and water budgets over the Mediterranean Sea are both affected by the global warming. For the first time in regional climate over the Mediterranean region, coherent air-sea fluxes are obtained on this region considered as a “hot spot”; nevertheless large uncertainties still remain for each individual term. All the models predict a heat gain over the sea and an increase in the water loss through a drying of the region. Future changes in the Mediterranean water masses are expected; those should have strong implications on the deep water formation, thermohaline circulation and Mediterranean dense water outflowing into the Atlantic ocean (Somot et al. 2006). A better evaluation of the uncertainty in the long term changes is currently addressed in several other large research programs, such as MedClivar (Mediterranean CLImate VARiability and Predictability) (Lionello et al. 2006), HyMeX (Hydrological cycle in the Mediterranean Experiment) (Drobinsky and Ducrocq 2008) and MedCORDEX (MEDiterranean COordinated Regional climate Downscaling Experiment) (Ruti et al. EOS, in preparation).