Future of land surface water availability over the Mediterranean basin and North Africa: Analysis and synthesis from the CMIP6 exercise

The Mediterranean basin and Northern Africa are projected to be among the most vulnerable areas to climate change. This research documents, analyzes, and synthesizes the projected changes in precipitation P, evapotranspiration E, net water supply from the atmosphere to the surface P–E, and surface soil moisture over these regions as simulated by 17 global climate models from the sixth exercise of the Coupled Model Intercomparison Project (CMIP6) under two Shared Socioeconomic Pathways, SSP2‐4.5, and SSP5‐8.5. It also explores the sensitivity of the results to the chosen climate scenario and model resolution and assesses how the projections have evolved from the fifth exercise (CMIP5). Models project a statistically robust drying over the entire Mediterranean and coastal North Africa. Over the Northern Mediterranean sector, a significant precipitation decrease reaching −0.4 ∓ 0.1 mm  day−1 is projected during the 21st century under the SSP5‐8.5 scenario. Conversely, a significant increase in precipitation of +0.05 to 0.3 ∓ 0.1 mm day−1 is projected over South‐Eastern Sahara under the same scenario. Evapotranspiration and soil moisture exhibit decreasing trends over the Mediterranean basin and an increase over the Sahara for both SSPs, with a notable acceleration from the 2020s. As a result, P‐E is projected to decrease at a rate of about −0.3 mm day−1 under the high‐end scenario SSP5‐8.5 over the Mediterranean whilst no significant changes are expected over the Sahara due to evapotranspiration compensation effects. CMIP6 and CMIP5 models project qualitatively similar patterns of changes but CMIP6 models exhibit more intense changes over the Mediterranean basin and South‐Eastern Sahara, especially during winter.

Climate projections indicate that the Mediterranean and adjacent Northern Africa are among the most vulnerable areas to global warming (Almazroui et al., 2020;Lee et al., 2020), specifically regarding the hydrological cycle (Hanel et al., 2018) and heat extremes (Seneviratne et al. 2021).The intensity of climate change in the Mediterranean region and the associated high socioeconomical risks raised the necessity of establishing the First Mediterranean Assessment Report (Cherif et al., 2020).Moreover, the recent sixth Assessment Report of the Intergovernmental Panel on Climate Change (IPCC) stresses that "there is high confidence that anthropogenic forcings are causing increased aridity and drought severity in the Mediterranean region" (Douville & John, 2021).Subsequently, potential stress on water resources are projected over the Southern Mediterranean region (Prudhomme et al., 2014), consistently with a projected decrease in surface water resources in North Africa under various climatic scenarios owing to precipitation reduction and evapotranspiration increase (Balhane et al., 2021;Driouech et al., 2010;Tramblay et al., 2018).Therefore, assessing the water cycle and surface hydrology response to global warming in waterstressed areas such as North Africa and the Southern Mediterranean region is crucial for anticipating consequences on water supplies, agriculture, and ecosystems (Marchane et al., 2017;Tramblay et al., 2018).
Global climate models (GCMs), and in particular those included in the Coupled Model Intercomparison Projects (CMIP; IPCC 2021), are commonly-used tools to assess the future effects of climate change on the water cycle and particularly on the different terms of the surface water budgets.In a CMIP5 (fifth CMIP phase)-based study, Mariotti et al., (2015) found a significant decrease in mean precipitation over southern areas of the Mediterranean during winter under a moderate greenhouse gas emission scenario.Lionello & Scarascia (2018) report a decrease of 4% per degree of warming in annual precipitation over the Mediterranean area according to CMIP5 models under the high-emission RCP8.5 scenario.Mariotti et al. (2015) found a precipitation reduction during summer over Spain, western Northern Africa, and Turkey along with a projected increase in winter evapotranspiration over Northern Mediterranean land by the end of the 21stcentury relative to the 1980-2005 reference period.
Precipitation (P) minus evapotranspiration (E) (P-E hereafter) represents the net freshwater flux from the atmosphere to the surface as defined by Byrne and O'Gorman (2015), and is also called "water availability."P-E drives the water percolation and soil moisture (SM), the run-off, and ultimately the river flow and groundwater recharge.The climatological P-E is close to zero over Mediterranean (MED hereafter) lands except in Northern areas during winter where P surpasses E (Mariotti et al., 2015).CMIP5 projections indicate a general decreasing trend for P-E across MED lands, specifically during summer due to the precipitation decrease over Northern areas (Mariotti et al. 2015).Assessing future changes in surface SM is also critical to anticipate the impact of changes of the hydrological cycle on several socio-economical activities such as agriculture (Ruosteenoja et al., 2018;Seneviratne et al., 2010), especially in arid and semi-arid areas.Over the Mediterranean basin, CMIP5 models simulate a continuous reduction in total soil moisture during the twentieth century, and it is expected to persist throughout the 21stcentury (Mariotti et al., 2015).
In addition to GCMs, regional climate models (RCMs) laterally forced with GCMs are also employed to provide scenarios at higher resolutions over a specific area.Climate change impacts across the MED area have been estimated from RCM simulations in the framework of the Coordinated Regional Climate Downscaling Experiment database (Drobinski et al., 2020;McSweeney et al., 2015;Tebaldi et al., 2005).Tuel et al. (2021) underlined the necessity of carefully selecting the GCMs used to force the RCM to properly project the climate over the MED region.Hence the need to select carefully and analyze the surface water projections and related uncertainties across the Mediterranean and North Africa from recent GCM simulations.
In this paper, we provide an in-depth analysis and synthesis of the surface water supply projections from the newly available CMIP6 GCMs-which show considerable improvements in terms of spatial resolution and physics with respect to CMIP5 (Eyring et al. 2016)-over MED and Saharan (SAH hereafter) subregions.For this purpose, we analyze the quantitative evolution of precipitation, evapotranspiration, P-E, and surface SM over those two regions for a carefully-selected set of simulations and thoroughly discuss the ensemble statistics.Note that evaluating RCM scenarios over MED and SAH is beyond the scope of the present paper, but our study will help better assess the suitability of future projections from the last generation global projections as boundary conditions for RCM runs.

| DATA AND METHODS
We considered 17 GCMs involved in the CMIP6 database (https://esgf-node.llnl.gov/search/cmip6/)and analyzed the "historical" (1850-2014) simulations (Eyring et al., 2016) and the low and high-end forcing Shared Socio-economic Pathways "SSP2-4.5" and "SSP5-8.5"(2015-2100) simulations from ScenarioMIP (O'Neill et al., 2016).The models' selection was based on the availability of the following output variables at the time of the study: 2-m temperature, precipitation rate, evapotranspiration, and surface SM content (top 10 cm) over the period 1850-2100 (historical + scenarios) (Table 1).Three models considering enclosed marginal seas as "land" have further been excluded to avoid inconsistencies in the multi-model land evapotranspiration statistics.Following Agosta et al. (2015) we have briefly assessed the performance of the selected model to simulate the large-scale climate patterns driving the MED and SAH climate (see Figure S8 in Section B of the supplement).None of the models considered here outperforms or lies outside the others but models with a higher horizontal resolution generally show better performance in terms of large-scale circulation metrics.For a thorough evaluation of CMIP6 models over the Mediterranean and African regions in terms of near-surface climate, we also refer the readers to Babaousmail et al. (2021);Ba gçaci et al. (2021); Seker & Gumus (2022).For the sake of fairness in our comparative study, we selected the available set of climate models from CMIP5 and its updated version of CMIP6 (Table S1).We will also briefly comment on the behavior of the total SM content.SM contents will be expressed as relative rather than absolute quantities since the soil schemes of the different models and especially their depths vary (Cook et al., 2020).Analyses are conducted over the MED and Sahara (SAH) subregions as defined for the IPCC's sixth assessment (Iturbide et al., 2020), as shown in Figure 1a.
Northern and Southern shores of the Mediterranean Sea are expected to exhibit quite different quantitative responses to climate change (Tuel et al., 2021), We, therefore, subdivide the MED sector into two sub-regions for our analyses: the Northern (N_MED, northern than 35 N) and Southern (S_MED) Mediterranean (see Figure 1a).
In order to assess uncertainties and identify robust signals in future model projections, we calculated the multimodel ensemble mean and variance across models.As the large-scale circulation fields over the MED and SAH regions seem to be resolution dependent (Section B of the supplement), we will separately consider two groups of models depending on their resolution (Table 1): one group for which the native horizontal resolution is close to $100 km and another one for which it is close to $250 km.Given that our work focuses on land, we masked out grid boxes where the land area fraction is less than 90%.This later has been chosen as a trade-off to conserve the largest possible study area avoiding retaining meshes at the continent margins for which the evapotranspiration flux is not representative for land surfaces (Figures S1 and S2).For assessing the statistical robustness of future changes (or non-changes), we use the nonparametric Wilcoxon-Mann-Whitney test comparing future (2071-2100) and historical (1981-2010 Consistently with P and E responses, an overall decrease in SM (Figures 1d, 2d) is simulated.The projected changes in SM reach À12 (À8)% in parts of Spain, Northern Morocco, and Algeria for the SSP5-8.5 (SSP2-4.5)scenario.However, large parts of SAH show nonrobust changes especially under the SSP2-4.5 scenario, while about +10% changes can be pointed out in the far south of SAH under the SSP5-8.5 scenario.The total SM shows similar evolution (Figure S3), with changes from À6 to À12 (respectively, À4 and À8) % over MED, and from +6 to +20 (respectively, +4 to +14) % over Southern SAH under the SSP5-8.5 (respectively, SSP2-4.5)scenario, with the strongest change much more localized in the IP, Italy, Northern Morocco, and Algeria.While P-E is close to zero, the relative SM changes in North Africa are significant.
F I G U R E 2 Same as Figure 1 but for the SSP2-4.5 scenario.
Our results thus show a robust decrease in future water availability over MED due to the decreasing precipitation over the region.However, despite the robust wetting signal for precipitation in South-Eastern Sahara, a near-zero change in the freshwater flux is projected owing to the response of evapotranspiration.Model ensembles exhibit a general south-to-north soil drying, and the inspection of the individual models behavior reveals that most models reproduce this general pattern (Figure S4), with the exception of the GISS-E2-1-G model that projects extensive drying over the entire study area.

| Seasonal aspects
Since the rainy season in the Mediterranean starts from October and extends to approximately April  (Driouech, 2010;Xoplaki, 2002), separate analyses are conducted for extended boreal winter (ONDJFMA) and summer (MJJAS) for the SSP5.8-5scenario.The projected 21st-century precipitation decrease is especially intense for the N_MED in MJJAS and for S_MED in ONDJFMA.Increases in precipitation over SAH are twice more intense in summer than in winter (Figure 3a, b).In the Iberian Peninsula, Turkey, and northwestern Maghreb, the precipitation reduction under SSP5-8.5 scenario (Figure 3a, b) reaches À0.2 mm day À1 to À0.4 mm day À1 during wintertime.While summertime results point to À0.2 mm day À1 over Northern Maghreb, and exceed À4 mm day À1 in N_MED.This significant decrease in wet-season precipitation throughout the S_MED sector is consistent with previous studies (Diffenbaugh & Giorgi, 2012;Tuel & Eltahir, 2020).The increase in annual precipitation over SAH found in Section 3.1 is most pronounced during summer which concurs with the study of Almazroui et al. (2020).Summertime changes over SAH vary between +0.3 and + 4 mm day À1 .Summertime evapotranspiration signal (Figure 3c, d) follows the precipitation behavior with a projected decrease reaching À0.4 mm day À1 over MED and an increase of about +0.4 mm day À1 over South-Eastern Sahara.However, wintertime changes for S_MED are about +0.2 mm day À1 , À0.2 mm day À1 in Northern Maghreb, and around +0.1 mm day À1 in Southern SAH. P-E (Figure 3e, f) shows an opposite sign between summer and winter over almost the study area, with decreasing changes over the Mediterranean ($ À0.3 mm day À1 ) and Southern Sahara ($ À0.1 mm day À1 ) during winter, and increasing changes during summer ($ + 0.2 mm day À1 ) over MED and Southern SAH. Surface SM (Figure 3g, h) follows the same behavior to a large extent.The projected changes reach À16 (À8)% during the extended winter (respectively, summer) over Northern Africa, and À8 (À18)% over the IP, Italy, Greece, and Turkey.Changes across SAH vary from +10% during winter to +14% during summer.
Results from SSP2-4.5 are shown in Figure S5 in the supplement.Seasonal changes exhibit similar behavior mainly during summer, with reduced intensity, compared to the SSP5-8.5 during winter.
3.3 | Regional assessment and impact of spatial resolution

| Time evolution by region
We now examine more in detail the time evolution of P, E, P-E, and SM, spatially averaged over the N_MED, S_MED, and SAH sectors for the extreme SSP5-8.5 scenario.Figure 4 shows the evolution of the mean annual anomalies and evidences the acceleration of the Mediterranean drying and saharan moistening from the 2020s.The amplitude of change is generally greater in N_MED than in S_MED, mainly due to the higher mean annual precipitation in the Mediterranean, which conditions the evaporation and P-E changes.Figure 4 also shows a gradual decrease in SM during the study period over the entire MED where projected anomalies are around À15% over both N_MED and S_MED, while the evolution over SAH shows a near-zero trend until the 2000s when a slight increase of +2.5% emerges.
Results corresponding to the SSP2-4.5 scenario (Figure S6) show similar behavior with less intermodel spread, particularly for the surface SM.

| Spatial resolution impact
Given the sharp topographic and land surface-type contrasts of the MED regions, regional-scale climate change projections in this region are expected to depend on the model resolution (Ashfaq et al., 2016).We conduct a preliminary analysis by separating a $ 100-km resolution CMIP6 simulation ensemble from that with a $ 250-km resolution.
While the 250-km resolution and the 100-km resolution ensembles agree over the "historical" period (Figure 4), their future projections show statistically significant differences in precipitation, exceeding 0.1 mm day À1 over MED.The lower resolution generally yields more pronounced changes and intermodel variability over MED, while this behavior is less marked over SAH.For almost all the studied variables, the intermodel variability increases with time and decreases with resolution for both shores of the Mediterranean Sea.The differences between the high and lowresolution ensembles also increased toward the end of the century.Unlike MED, the lowest resolution exhibits less intermodel variability over SAH.SM signal shows that less intermodel spread characterizes the highest resolution for almost the entire area.A quite similar behavior is also shown in total SM (Figure S7), except for N_MED, where the 100-km ensemble intermodel variability exceeds that of 250 km.This can be explained by the higher magnitude of total SM which reaches $2000 kg m À2 , while it varies from 500 to 800 kg m À2 in the rest of the study area.

| Temperature-versus scenariodependent Mediterranean drying and Saharan moistening
Our results and previous studies project a strong drying trend over the Mediterranean region and a substantial moistening trend over the South-Eastern Sahara for the next decades.To what extent precipitation and evapotranspiration changes may be explained by circulation changes and/or by the direct response of the hydrological cycle to the increase in global temperature and how this will vary in time and space remain open questions.Tuel et al., (2021) show that CMIP5 GCMs project the development of a strong winter anticyclone centered on the Mediterranean Sea that results from a combination of large-scale shifts in winter planetary waves and to the reduced warming of the sea with respect to the surrounding lands.This anticyclonic anomaly and the associated dry advection effect over North Africa and S_MED may explain most of the decrease in precipitation in this F I G U R E 4 Spatially-averaged mean annual changes in precipitation (mm day À1 ), evapotranspiration (mm day À1 ), P-E (mm day À1 ), and surface soil moisture (%) for the Northern and Southern Mediterranean (N_MED and S_MED) and Sahara (SAH) regions under the SSP5-8.5 scenario.Curves are constructed by subtracting, for each year in the period 1850-2100, the mean value for the baseline 1981-2010.Then, a 10-year running mean is applied to smooth the resulting time series.Bold curves represent the multi-ensemble mean anomaly of the 100 km (orange) and 250 km (blue) resolutions, color shadings show the envelope between the maximum and minimum values (light shading) and between the 25th and 90th percentiles (dark shading).In order to assess the significance of the discrepancies between the two different resolutions, we performed the Wilcoxon-Mann-Whitney test to compare 100-and 250-km ensemble time series by dividing the future period into four sub-periods.The horizontal dotted red curves indicate the sub-periods where the test is significant at a 95% threshold.region.Drobinski et al., (2020) also underlined that the aridification over the Sahara and IP cannot be solely explained by the increase in temperature and that the development of local circulations associated with the faster increase in temperature over land plays a significant role.
To explore the behavior of the CMIP6 simulations concerning this aspect, we compare the evolution of the surface water budget under a fastly-warming and a slowly-warming scenario for a given increment in mean global temperature.The increase in mean global temperature between the 2041-2070 period and the 1981-2010 period under the SSP5-8.5 in the CMIP6 model ensemble ($2.89 K) is very close to the increase in mean global land temperature between the 2075-2100 and 1981-2010 under the SSP2-4.5 scenario ($ 2.90 K).The future change by the mid-century (2041-2070 vs. 1981-2010) for the high-end scenario (left panels in Figure 5), compared to the end of the century (2071-2100 vs. 1981-2100) for the SSP2-4.5 scenario (right panels in Figure 5) is, therefore, assessed for P, E and the mean surface air temperature for the 100-km resolution ensemble.The overall near-surface temperature change over MED and SAH is very close between the two ensembles.Over SAH, both ensembles project a moistening, P and E do not significantly change and the two ensembles do not show statistically different results except in some parts in the southeast.This suggests that changes in E and P over SAH are slightly or not scenario-dependent and rather respond to the overall temperature increase.Regarding the Mediterranean basin, results show statistically different projections of P and E even if the geographical patterns are qualitatively alike.Changes in P and E are therefore scenario dependent and cannot be directly related to the net increment of global near-surface temperature.In line with Drobinski et al. (2020) and Tuel et al. (2021), CMIP6 models thus suggest local land heating differences-such as over Morocco (see Figure 5a, b)-and changes in circulation likely superimpose on the overall temperature increase signal to explain the future Mediterranean drying.Further investigation into those physical and dynamical mechanisms is needed, considering for instance seasonal aspects (Brogli et al., 2019) but it is beyond the scope of the present paper.

| Projected changes in CMIP6 compared to CMIP5
Prior to concluding this letter, it is worth examining to what extent our CMIP6-based results differ or concur with CMIP5 projections.We thus compare CMIP6 and CMIP5 results under high-end forcing over the MED and SAH regions, with a particular focus on precipitation. Figure 6a shows CMIP6 and CMIP5 differences in future projected precipitation (CMIP6 minus CMIP5).Results show more intense changes than in CMIP5.This former shows less intense future precipitation increase over SAH with differences ranging to +0.3 mm day À1 .This concurs with Almazroui et al., (2020) who found intense mean annual precipitation in CMIP6 compared to CMIP5 over Sahara.Differences in precipitation extends up to À0.2 mm day À1 across the Northern parts of Morocco and Algeria, Mauritanian coasts, and N_MED.Indeed, Figure 6b shows that CMIP6 and CMIP5 precipitation changes generally agree on the sign of the change.However, the projected drying is more intense in CMIP6 over western N_MED, Northern Morocco and Libya, and Atlantic coasts of Mauritanians, the Saharan moistening is also more intense in CMIP6.Similar results have also been found in previous studies (e.g., Chen et al., 2020;Lee et al., 2021).CMIP6 models project intense globallyaveraged surface air temperature (GSAT) increase in comparison with CMIP5 (Lee et al., 2021)-in particular, due to a higher climate sensitivity of models (Meehl et al. 2020)-the change in several climate quantities such as large-scale precipitation scales with GSAT signal (Lee et al., 2021).

| SUMMARY AND CONCLUSIONS
CMIP6 models project a robust decrease in precipitation over MED by the end of the century with larger uncertainty across Southern Morocco and Mauritania.An projected by CMIP6 models compare with those from CMIP5 models.Dark red (respectively, dark blue) regions are where CMIP5 models predict a decrease or drying (respectively, increase or moistening) and for which CMIP6 models predict an even stronger decrease (respectively, increase) by the end of the century.Light red (respectively, light blue) regions are where the CMIP5 models ensemble projects a decrease or drying (respectively, increase or moistening) and where the CMIP6 models ensemble projects a less intense decrease (respectively, increase).Green pixels are regions where CMIP5 models project a decrease or drying and where CMIP6 models project an increase.
opposite behavior is projected in South-Eastern Sahara with a relative increase in precipitation in all models.Evapotranspiration behavior follows the precipitation pattern with weaker intensity over N_MED, resulting in a negative projected freshwater flux over this region and a near-zero projected P-E over S_MED and Sahara.The future negative freshwater flux over MED may exacerbate existing water scarcity issues, particularly in areas heavily reliant on agriculture (García-Ruiz et al., 2011).These changes may also have implications in terms of wildfire which is already a significant risk in the Mediterranean region (Rovithakis et al., 2022;Turco et al., 2018).Interestingly, the projected increase in precipitation over South-Eastern Sahara does not result in a climatological increase in P-E which means that the net surface water availability is not projected to increase.This may have important implications for the future of water resources in this region but this result should be interpreted with caution.In fact, the increase in evapotranspiration that balances precipitation does not necessarily translate into a net loss of water, it may have the potential to support increased plant transpiration leading to a boost in vegetation growth in the region.The seasonal analysis also reveals a contrasted evolution with a relative increase in P-E in summer and a decrease in winter, the main rainy season, over North Africa.Such an evolution has potential implications regarding the adaptation to the future of the water resources in the region, for instance, rainwater harvesting, water storage facilities, and efficient water management and irrigation systems.
A preliminary analysis has revealed that the 100-km resolution models project less intense drying over MED with a less intermodel spread in comparison with the 250-km resolution models.However, those results should be interpreted with caution since the differences between the two ensembles are not only explained by the difference in resolution.Further work focusing on scenarios performed with one given model at various resolutions would help assess in a more robust way the effective sensitivity to resolution.
A brief complementary analysis suggests a higher scenario-dependency of the hydrological cycle changes over the Mediterranean than over the Sahara where changes seem more directly related to the overall global increase in temperature.Moreover, while the CMIP6 model ensemble qualitatively agrees with the CMIP5 ensemble in terms of precipitation projection, the magnitude of the projected changes is generally higher in CMIP6, with a more intense drying over MED and more intense moistening over the Sahara.
Our results allow us to identify robust signals in future model projections over the Mediterranean and North Africa even though on a regional scale, climate change depends on the ability of models to properly simulate the main atmospheric circulation patterns.Further studies making use of empirical bias-corrections methods (Krinner et al., 2020) would make it possible to further strengthen the reliability of GCM-based regional projections over the MED and SAH regions.Future scenarios over North Africa would also benefit from fostering the development and evaluation of model parameterizations and configurations specifically adapted for Africa, as models are generally insufficiently developed and evaluated for this continent (James et al., 2018).

F
I G U R E 3 As Figure 1 but for the extended boreal winter (left panel) and summer (right panel) seasons under the SSP5-8.5 scenario.

F
I G U R E 5 Mean future changes from the 100-km ensemble of annual surface air temperature ( C), precipitation (mm day À1 ) and evapotranspiration (mm day À1 ).Panels in the left column show the differences between the 2041-2070 and 1981-2010 periods for the SSP5-8.5 while those in the right column show the differences between the 2071-2100 and 1981-2010 periods for the SSP2-4.5.Hatches indicate grid boxes where the high-end and low-end changes are not statistically different based on the nonparametric Wilcoxon-Mann-Whitney with a 95% threshold.

F
I G U R E 6 Panel (a): CMIP6 minus CMIP5 model ensemble projection of mean annual precipitation change between 2071-2100 and 1981-2010.Backslashes indicate the grid boxes corresponding to statistically insignificant differences based on the non-parametric Wilcoxon-Mann-Whitney with a 95% threshold.The color shading in panel (b) qualitatively shows how the changes of precipitation