A methodology adapted to the Mediterranean context and adaptation strategies to climate change were set up (Fig. 1). This method was applied over the current period in order to define the current state of pressure on the Mediterranean water resources. It was also applied by the 2050 horizon in order to address the impacts of climatic and anthropogenic changes on water resources variability.

Freshwater availability was estimated as a long-term mean annual value over the reference climate period 1971–1990 and over the future climate period 2041–2060 (2050 horizon). The reference period was chosen for its representativeness of the Mediterranean climate, covering dry and wet periods. Projections of climatic changes were based on the A2 greenhouse gas emission scenario (IPCC, 2007), which corresponds to the largest increase in greenhouse gas emissions, and thus to very substantial changes in climate.

Current water withdrawals were estimated over the recent period 2001–2009, because homogeneous data were not available for all 73 catchments over the 1971–1990 period. Estimates of future withdrawals at the 2050 horizon consider improvements in the efficiency of transport and distribution networks and of irrigated plots. In 2005, the 21 Mediterranean countries adopted the Mediterranean Strategy for Sustainable Development (MSSD), in order to support a dynamic regional process to include environmental concerns in the economic development of these countries. Under this framework, they gave commitments to reduce agricultural and domestic networks water losses and to increase irrigation efficiency at the plot level (UNEP-MAP, 2006). This water-use scenario was explored over the medium term.

To assess the impacts of global changes on water resources, it was necessary to first evaluate freshwater availability, defined as the annual renewable water resources, or discharge, within a given catchment.

The conceptual rainfall-runoff Water Balance Model (WBM; Yates, 1997) was used to evaluate it. The model relies on a one-dimensional reservoir that represents both the root and upper soil layers. The dynamics of groundwater resources are considered in terms of drainage from soil moisture. The model uses continuous functions of relative storage to compute the water balance of each 0.5° grid cell at a monthly time-step. Contributions of the grid cells of each catchment are summed to give estimates of total discharge. Theoretically, the hydrological model requires no calibration, since its parameters are predetermined by a bioclimatic diagram (Holdridge, 1947; Yates, 1997). Nonetheless, a sensitivity analysis was carried out for ten large Mediterranean catchments (> 10 000 km²). Systematic runs were conducted and aimed at minimizing the difference between simulated and observed runoff volumes using datasets that covered as many consecutive years as possible of the reference period. Since modifying the parameter values provided better agreement between the simulated and observed runoff volumes for all catchment, the new set of parameters was retained. The hydrological model, thus adjusted against historical data, was run over the future period using these optimized parameter values. It was assumed that the bias in the freshwater simulations over the reference period was reproduced in the simulations of future conditions and thus that the error related to freshwater resources modeling on water vulnerability assessment was the same over the reference and future periods.

The model requires monthly precipitation (P) and potential evapotranspiration (PE) input data on a 0.5° grid. Over the reference period, P and temperature data were obtained from the CRU TS 3.0 World database (Mitchell and Jones, 2005). Although this database relies on spatial interpolation of observed data in areas where data are scarce or incomplete, it is currently recognized as one of the most reliable databases for large-scale investigations. For future estimation, outputs from the CSIRO-Mk3.0 (Gordon et al., 2002), HadCM3 (Pope et al., 2000), ECHAM5 (Jungclaus et al., 2005), and CNRM-CM3 (Salas-Mélia et al., 2005) global climatic models (GCMs) were extracted from the IPCC's Data Distribution Centre. These models were selected because:

• • they are approved by the IPCC for its fourth and fifth assessment reports;
• • they have available data over the periods of time covered in this study;
• • they have a low resolution (< 3°×3°) in comparison to other GCMs;
• • and they have often been used in Mediterranean case studies.

Climate scenarios under changing levels of greenhouse gas emissions, as specified by the 20C3 M scenario (20th century climate–350 ppm CO₂) and A2 scenario (2050 horizon climate–CO₂ stabilization at 850 ppm) were generated using the perturbation method (Ruelland et al., 2012). PE was then computed using a simple formula based on extraterrestrial radiation and mean temperature (Oudin et al., 2005).

Water withdrawals were evaluated for irrigated agriculture, currently the most water-demanding sector (65.9 km³/year), and for the domestic sector (19.5 km³/year). The latter includes urban and rural water, drinking water supplies and touristic activities which draw on municipal water networks, and are often given priority for water supply. Water withdrawals for industries that are not connected to the municipal water networks and for energy production were not taken into account because they represent less than 2 km³/year within each Mediterranean catchment, except for those in Spain, France and Italy where they amount, respectively, to 4.2 km³/year, 13.2 km³/year and 14 km³/year (Margat, 2004). Water withdrawals were therefore defined as the annual volume of water withdrawn from rivers and groundwater reservoirs that is directly available for irrigated agriculture or domestic purposes.

Due to the limited availability of data on water withdrawals at the catchment scale for the current period, unit water demand (UWD) was evaluated at the country level for each sector. UWD was defined as the ratio of the water withdrawn from the water system plus the unconventional water resources used, to the irrigated surface area in the case of agricultural water use or to national population in the case of domestic water use. Two assumptions were made: (i) the UWD computed at the country scale was applied to each catchment of the country considered; and (ii) for unconventional water resources, reused wastewater was assumed to be dedicated to irrigated agriculture, while desalinated water was considered to be for domestic use. For the 2050 horizon, current agricultural UWD was adjusted according to variations in irrigation water requirements as computed by the irrigation management tool CROPWAT (Allen et al., 1998). Within this model, the various irrigated crops and the expansion of irrigated area – i.e. new agricultural areas or currently non-irrigated areas that are subsequently irrigated – were considered. The proportion of irrigated crops in the current irrigated fields was assumed to remain unchanged by the 2050 horizon. The expansion of irrigated area was taken from national reports (Plan Bleu, 2011) produced between 2005 and 2009 by scientists and policy-makers from the 21 Mediterranean countries for the regional activity center Plan Bleu (Plan Bleu, 2006). Future domestic UWD was also obtained from these reports, which considered an increase in population and in touristic activities (Plan Bleu, 2011).

The UWD of each sector was then multiplied by the catchments population and irrigated area and corrected by an efficiency rate in order to obtain the total water demand (TWD). The efficiency rate was defined as the ratio of current efficiency (obtained from the national reports) to the efficiency objectives set by the MSSD. The latter aimed at reducing agricultural and domestic networks water losses and improving irrigation techniques for a larger consumption of distributed water, thus rising water-use efficiency to 72% for agriculture and 80% for domestic use. However, these objectives have been modified in some countries (e.g. Italy, Croatia) in accordance with the national reports assessment of their feasibility. Finally, unconventional water resources were subtracted from TWD in order to obtain the water withdrawals of each sector at the catchment scale.

For the current period: water withdrawals; unconventional water production; and irrigated area and crop data were collected, respectively, from the Aquastat (FAO, 2010), FAOStat (2010), and MIRCA 2000 (Portmann et al., 2010) databases. Population data at the national and catchment levels were drawn from the most recent census. For the 2050 horizon, trends in irrigated area and efficiency were taken from the national reports. The climatic data (P and PE) used as inputs to CROPWAT were the same as for the hydrological model. Finally, in order to be consistent with the assumptions of the A2 greenhouse gases emission scenario used for the climatic scenario, the A2 demographic scenario published by the United Nations (UNPD, 2001) was used for future population estimates.

Once these main variables were obtained, it was possible to compute the water stress index (WSI) adopted in the MSSD to monitor the impacts of climatic and anthropogenic changes on water resources. The index is based on the ratio of annual water withdrawals to annual renewable water resources (Shiklomanov, 1991; equation (1)).

This index expresses the intensity of anthropogenic pressure on available water resources: the higher the index, the stronger the pressure. It indicates the margin between renewable water resources and water demand, and consequently the scope of action available to water managers. It has previously been applied to the Mediterranean region (Plan Bleu, 2005) but without taking into account the impacts of climate change on water use and availability. For our study, this index was computed over the Mediterranean basin to estimate the current state and evolution of water stress. Possible trends in water stress were explored first under climate changes, then under anthropogenic changes, and finally under both climatic and anthropogenic changes.

Normally, the use and presentation of different climatic scenarios is recommended in order to give a range of possible futures and to express uncertainties for water planning (Le Treut et al., 2008; Menzel and Matovelle, 2010). However, as the four selected GCMs simulated very similar climatic trends, individual runs of the method gave very similar results in terms of water stress shifts. Uncertainty linked to the choice of GCM is thus negligible here, and for the sake of concision, the following section presents average trends for water resources and water stress evolution.

As can be seen from Fig. 4a, the current water resources of the Mediterranean basin are unequally distributed, being particularly scarce on the southern rim. The four selected GCMs agree that annual air temperature should increase by 1.5 to 2.5 °C (Fig. 2a) and annual precipitation should decrease by 5–20% (Fig. 2b) over the Mediterranean basin as a whole by the 2050 horizon. The strongest decreases are projected for southern Spain, Morocco, Algeria and the Middle East (20–40%), while an increase is projected for Libya (40–60%), although this would still leave the latter with low absolute values of precipitation. The combined effects of increasing temperatures and decreasing precipitation should reduce freshwater availability and exacerbate the disparities between rims.

In climate change simulations under the A2 greenhouse gas emission scenario at the 2050 horizon, the hydrological model simulates a significant decrease (30–50%) in freshwater resources over the whole Mediterranean basin (Fig. 4b). Catchments in southern Spain, Morocco, Algeria and the southeastern Mediterranean should be the most affected, with a reduction of over half of their current freshwater resources. The smallest decreases (15%) are projected in catchments in France, northern Italy, and the Balkans, as precipitation is expected to remain constant in the mountainous parts of these catchments. Only Libya and southern Tunisia should experience an increase (10%), maintaining their current level of renewable freshwater resources, between 0 and 15 mm/year (Fig. 4a and b).

Climate change should thus have a significant impact on freshwater availability by the 2050 horizon, with the already arid to semi-arid catchments being the most affected. Moreover, the latter should be the most prone to anthropogenic changes.

According to the national reports, irrigated area should stay at their current level in Spain and Slovenia (+2%); decrease in France (4%), Italy (18%) and Malta (27%); and increase over the rest of the Mediterranean basin (Fig. 3a). The largest increases should be observed in catchments in the Balkans, Algeria, Libya, Israel, and Lebanon, where irrigated area is expected to double (Fig. 3a). Population should stabilize or decline in northern catchments, increase by 85–90% in catchments in Turkey and Lebanon, and at least double in the southeastern Mediterranean and on the southern rim (Fig. 3b). These trends will inevitably lead to changes in water withdrawals. These changes are presented in Fig. 4d.

Efficiency improvements should hold down water withdrawals over the whole Mediterranean basin. On the northern rim, total water withdrawals should increase only in Greece, Albania, the Veneto region of northern Italy, and the Ebro catchment in Spain (Fig. 4d). This can be attributed to an increase in agricultural water withdrawals (Fig. 3a and b) as a result of the 42% expansion of irrigated lands in Greece and Albania (Fig. 3a) and of warmer and drier conditions in Spain and Italy. Otherwise, total water withdrawals over the northern catchments should remain at their current level or decrease by 10 to 40% at the 2050 horizon, notably in Italy, Slovenia, and Croatia (Fig. 4d). This trend should be due mostly to a decrease in domestic water withdrawals, as access to water supply is already adequate and population is projected to stabilize in these areas (Fig. 3b). In the Balkans, in contrast, this evolution can be attributed to considerable progress in efficiency, which should rise from 50 to 80% on average. In catchments of the Maghreb and the Middle East, total water withdrawals should at least double (Fig. 4d), owing to both the expansion of irrigated land and high population growth (Fig. 3a and b). Total water withdrawals should remain rather close to their current level (+10–20%) only in Turkey, Syria, and the Moulouya catchment in Morocco as a consequence of strong improvements in agricultural efficiency, which should rise from 45 to 72%.

Analysis of Fig. 4e shows that the Mediterranean basin is currently under high water stress. Hot-spots include catchments where renewable water resources are scarce and water demand is high, i.e. in southern Spain, Tunisia, Libya, and the southeastern Mediterranean (Syria, Lebanon, Israel and the Palestinian Territories). Catchments in northern Italy, western Greece and the Ebro in Spain are shown to experience moderate water stress, whereas catchments in France and the Balkans seem not to suffer from any stress. As stated above, however, water withdrawals for industrial use were not taken into account. As such withdrawals amount to around 13.2 km³/year in Mediterranean catchments in France alone (Margat, 2004), human pressure might be higher than illustrated in these catchments. The combined effects of freshwater availability and water withdrawal trends shown above suggest that the Mediterranean water stress state will deteriorate by the 2050 horizon, especially on the southern and eastern rims.

If the medium-term impacts of climate change alone are considered, the reduction in water resources on the northern rim should cause catchments with low to moderate water stress to experience respectively moderate to high stress, and even severe stress in southern Italy (Fig. 4f). Only catchments in France and the Balkans should continue to experience no stress. On the southern and eastern rims and in southern Spain, water tensions could be severe (Fig. 4f).

If water resources are projected to remain at their current volumes and water withdrawals to change, water stress should remain at its current state on the northern rim, except in Italy and Greece. In western Greece, the current low to moderate state of water stress should deteriorate to moderate to high water stress respectively. In Italy, the projected 30 to 35% decrease in water withdrawals should lead catchments currently under moderate to high water stress to experience low to moderate water stress respectively in the medium term (Fig. 4g). On the southern and eastern rims, water stress should certainly increase, remaining close to its current level only in Turkey and in the Moulouya catchment in Morocco (Fig. 4g).

Therefore, in the northern Mediterranean, although moderate to high water stress could occur under climate change conditions alone, the current water stress state should be maintained owing to a projected decrease in population and irrigated areas, and increased efficiency (Fig. 4h). Water stress should worsen only in western Greece and in the Ebro catchment in northern Spain due to the projected increase in total water withdrawals. Catchments in France and the Balkans should be the only ones remaining under no- to low-stress conditions (Fig. 4h). On the southern and eastern rims, the combined effect of decreasing freshwater availability and increasing water withdrawals should lead to severe water stress (Fig. 4h). Moreover, despite the projected progress in efficiency, Turkey, Syria and the Moulouya catchment in Morocco should still experience severe water stress owing to the significant decrease in their freshwater resources (Fig. 4b) and high population growth (Fig. 3b).