Each basin was divided into sub-basins accounting for the water supply to one or more demand nodes. The Herault and the Ebro basins were divided respectively into six and 20 sub-basins (see maps in Fig. 3a). The conceptualization of both hydrosystems accounted for climatic gradients and water use contrasts .

An integrative modelling framework (Fig. 1) was then applied in both basins. Anthropogenic water demands were simulated and aggregated at each demand node (see for more details). Three types of demand were simulated: urban water demand (UWD), agricultural water demand (AWD) and other water demands (OWD). Other water demands, comprising water demands for industries and energy production, were considered negligible in the Herault basin. Natural streamflow was simulated in each sub-basin, and a minimum environmental flow was computed at each water resource node. Streamflow regulations and storage were accounted for with a reservoir management model.

Water demand and availability were compared at the main water demand nodes. An order of priority was considered for water uses (Fig. 1). In case of insufficient water availability, a deficit was computed, defined as the percentage of water demand that could not be satisfied by available water resources. The modelling chain enabled us to simulate: (i) natural water resources and their availability considering water management rules and infrastructures; (ii) the ability to satisfy water demands throughout the basin; and (iii) the influenced streamflow resulting from hydro-climatic conditions and human influence (water withdrawals, return flows and dam management).

Three indicators were used to characterize the water supply capacity for the main water demands, i.e. urban and agricultural: (i) F: frequency of years with at least one significant deficit (greater than 50 % for AWD, greater than 5 % for UWD), (ii) Def10: average deficit at a 10-day time step and (iii) Def_AN: average annual deficit. For all three indicators values ranged from 0 (no deficit) to 1 (maximum frequency or intensity of deficit).

Details regarding the reconstruction of past water demand between 1971 and 2009 at the sub-basin scale in the Herault and the Ebro basins can be found in . UWD was estimated from time series of population and monthly unit water consumption data. AWD was computed from time series of irrigated area, irrigation efficiency, crop and soil data, and climate forcings. OWD in the Ebro basin was computed from industrial activity and allocations per employee and per added-value, and from water consumed by the open-air cooling systems of nuclear plants.

Natural streamflow was assessed in the six sub-basins in the Herault and the 20 sub-basins in the Ebro using GR4j , a conceptual hydrological model run at a daily time step and calibrated/validated at a 10-day time step. To assess natural runoff in each sub-basin, the model was calibrated only against runoff data that were considered natural, i.e. not influenced by withdrawals or dam management. Due to the lack of data, the influence of withdrawals on streamflow was considered to be negligible when simulated AWD was negligible .

River flow regulations were accounted for by developing a demand-driven reservoir management model, which was applied to the largest dam in the Herault basin and to 11 major dams in the Ebro basin. Inputs of the model are incoming streamflow, evaporation, the initial reservoir level and water demand at a 10-day time step. The model then calculates the volume of water released into associated canals and the river downstream from the dam during each time step, and the reservoir level at the end of each time step (see for more details).

The modelling chain was calibrated and validated over 1971–2009. This nearly 40-year period includes a warmer and drier period (1981–2009), used for calibration, and a colder and wetter decade (1971–1980), used for validation. Automatic calibration and validation of the hydrological model were performed against natural streamflow data using a three-step algorithm that minimized a multi-objective function aggregating three goodness-of-fit criteria: the Nash-Sutcliffe efficiency index (NSE), the cumulative volume error (VE) and the mean annual volume error (VEM). The simulation of water demand could not be thoroughly validated for lack of data. The simulation of influenced streamflow was validated against observed streamflow data at each resource node (see Sect. 3.1) by calculating the values of the NSE, VE and NSELF (NSE on low flows, i.e. from June to August) criteria per decade over 1971–2009.

The modelling chain was then applied under four combinations of climate and water use scenarios to distinguish the impacts of human- and climate-induced changes on water supply capacity at the 2050 horizon. Two water use scenarios were considered: water uses of the 2000s and a trend water use scenario at the 2050 horizon. Regarding climate scenarios, we considered a reference climate over 1976–2005 and 18 climate change scenarios at the 2050 horizon. Climate forcings over 1976–2005 were extracted from the 8 × 8 km grids presented in . Climate scenarios at the 2050 horizon (2036–2065) were then built based on climate change simulated by nine Global Climate Models (GCMs) from the last IPCC report, with the Representative Concentration Pathways (RCPs) 8.5 and 4.5. Using a change factor method , the reference climatic series were modified so as to reproduce the mean monthly variations obtained between the reference and future climatic simulations from GCMs.

The water use scenarios at the 2050 horizon were built based on projections from the local water management agencies and continuation of recent trends . The variations of the main variables between the 2000s and the 2050s are shown in Table 1.

Environmental water demand was also accounted for in the prospective part of the study. Since minimum flows have not yet been defined by local water agencies, they were considered as the value of natural streamflow exceeded 95 % of the time over the study period at the Herault sub-basin outlets, and as 10 % of the mean annual flow at the Ebro sub-basin ones (or 5 % of mean annual flow if it exceeded 80 m3 s-1).

Natural streamflow at the 2050 horizon was assessed by using the climate variables from the 18 scenarios described in Sect. 2.3.1 as inputs to the GR4j model while keeping the parameters calibrated over the reference period. Inputs to the dam management model varied depending on the climate and water use scenarios. Entering streamflow and evaporation were computed based on each climate scenario, while the water demand associated to each dam was dependent on the water use scenario. Future water use scenarios also included infrastructure and management projects under way in the Ebro, such as the doubling of the Yesa dam's capacity. In this case, the target reservoir levels were changed accordingly in the future water use scenario.

In the reference scenario, results showed that of the four areas presented here, only the Segre area in the Ebro basin had balanced water demand and availability (Fig. 4). In the Gignac and Agde areas in the Herault basin, restrictions on urban and agricultural water withdrawals were simulated to occur every two years, with an annual deficit of approximately 30 % on AWD in both areas. In the Ebro basin, of the large left bank systems only the Bardenas system had imbalanced demand and availability. Note that although restrictions on withdrawals were more frequent in Bardenas than in the Herault basin, average deficits at 10-day and annual time steps were lower.

Results showed that the relative impacts of climate and water use changes on water demand satisfaction could vary depending on the area (Fig. 4). For example in the Herault basin, the water use scenario impacted water demand satisfaction only in the Agde area. In the other areas, the climate change scenarios had a dominant impact on water demand satisfaction at the 2050 horizon. The results presented for the Ebro basin showed that sensitivity to climate change could vary according to the water use scenario taken into account: in the Segre area, climate change impacts could be acceptable without changes in water use, whereas with the trend water use scenario climate change could lead to frequent restrictions on agricultural water withdrawals. In the Bardenas area, the dam enlargement in the trend scenario could lead to an improvement in water supply, but this adaptation strategy may not be sufficient in scenarios of climate change.