Blue Water in Europe: Estimates of Current and Future Availability and Analysis of Uncertainty

: This study presents a regional assessment of future blue water availability in Europe under different assumptions. The baseline period (1960 to 1999) is compared to the near future (2020 to 2059) and the long-term future (2060 to 2099). Blue water availability is estimated as the maximum amount of water supplied at a certain point of the river network that satisﬁes a deﬁned demand, taking into account speciﬁed reliability requirements. Water availability is computed with the geospatial high-resolution Water Availability and Adaptation Policy Assessment (WAAPA) model. The WAAPA model deﬁnition for this study extends over 6 million km 2 in Europe and considers almost 4000 sub-basins in Europe. The model takes into account 2300 reservoirs larger than 5 hm 3 , and the dataset of Hydro 1k with 1700 sub-basins. Hydrological scenarios for this study were taken from the Inter-Sectoral Impact Model Inter-Comparison Project and included simulations of ﬁve global climate models under different Representative Concentration Pathways scenarios. The choice of method is useful for evaluating large area regional studies that include high resolution on the systems´ characterization. The results highlight large uncertainties associated with a set of local water availability estimates across Europe. Climate model uncertainties for mean annual runoff and potential water availability were found to be higher than scenario uncertainties. Furthermore, the existing hydraulic infrastructure and its management have played an important role by decoupling water availability from hydrologic variability. This is observed for all climate models, the emissions scenarios considered, and for near and long-term future. The balance between water availability and withdrawals is threatened in some regions, such as the Mediterranean region. The results of this study contribute to deﬁning potential challenges in water resource systems and regional risk areas.


Introduction
Water management is challenged by climate change. By the 2070s, the percentage of the surface area under conditions of severe water stress is expected to increase from the current 19% to 35% in central and southern Europe [1]. Populations living under water stress conditions in regions from 17 countries of Western Europe are projected to increase by between 16 and 44 million [2]. It is also predicted that the runoff of certain rivers may diminish by up to 80% during the summers. Reservoirs may lose resources due to a decrease in rainfall and the frequency of droughts will increase. The consensus is that the effect of climate change will also exacerbate precipitation extremes with more pronounced drought and flood periods [3][4][5]. At the same time, future water demand is increasing due to climate and social changes. Higher temperatures lead to increased water demand for irrigation and urban supply, hydroelectric potential of Europe may decrease 6% on average, and between 20 and 50% in the Mediterranean region. Advances in technology efficiency may only affect industrial demand [2]. In the Mediterranean region, impacts of climate change on water will certainly have a large influence on human water security and biodiversity [6]. There are several hundred local studies on the potential impacts of climate change on water resources in the Mediterranean, which apply many different approaches. Although the results are diverse and sometimes contradictory, a common element is that one of the primary impacts of climate change will be a reduction of water availability in the Mediterranean Region [1,2]. Furthermore, several authors showed that Global Climate Models (GCMs) were the main source of uncertainty when assessing the impacts of climate change on hydrologic processes [7,8]. Meanwhile, uncertainty associated with streamflow appeared to be more consistent with precipitation than temperature and showed higher sensitivity to the selection of GCMs than to the Regional Climate Models (RCPs) [9,10].
Water availability focuses on blue water, which is defined as water that runs off the landscape into streams, rivers, reservoirs, and groundwater [11]. However, the term "water availability" includes multiple aspects. A multitude of studies consider water availability to be directly linked to changes in average runoff, estimated as the net difference between precipitation and evapotranspiration [12,13]. In non-altered basins, water availability would be either null or extremely low because it would be determined by long term minimum values of flow. It is clear that hydraulic infrastructure plays an important role in making water available for users, mainly by the regulation and transportation of water resources. Even though the storage-based strategy proved to be very successful in the past [14], expanding infrastructure is not an option to increase availability in many regions due to social and environmental constraints [15]. As a result, increasing demand relies heavily on management. The emphasis is currently being placed on how to improve management of existing infrastructure and on socio-economic measures through demand management and water use efficiency [16,17]. The main factors to be considered in regulated water resource systems are the stream flow variability, storage capacity, and yield reliability. In this study, we define blue water availability as the amount of water that can be supplied at a certain point of the river network to satisfy a regular demand under specified reliability requirements [18,19]. Therefore, water availability is the combined result of natural processes, existing infrastructure, and policy. A wide range of techniques have been proposed to analyse water availability, from relatively simple stochastic processes relating these variables to highly complex models solving the water allocation problem [20][21][22][23][24], even including social and economic considerations [25]. In the water sector, institutions, users, technology, and the economy cooperate to achieve equilibrium between water supply and demand in water resource systems. In order to understand the process of reaching future goals for water under climate change, science has developed a set of tools to understand uncertainty [26][27][28][29], assess future impacts [30,31], and facilitate policy development [1,16,18,32]. However, most studies were developed using detailed water management and planning models, and were applied at the local scale. In systems and situations where limited information is available and regional or continental-scale studies are needed, it is generally better to obtain a global overview of the water supply systems' performance under different climate and policy scenarios, using simplified regional models rather than carrying out very detailed simulations with conventional models, which require very specific information on water demands and infrastructure [18,33,34]. These continental scale-models are conceived to estimate the maximum water availability and to provide technical and quantitative support to possible water policies in the short and long term. Then, these models and detailed water management and planning models should be considered as complementary tools.
Over forty percent of the total water withdrawal in Europe is used for agriculture. Southern countries use the largest percentages of abstracted water for agriculture. This generally accounts for more than two thirds of total abstraction. In northern member States, levels of water use in agriculture are much smaller, with irrigation being less important but still accounting for more than 30% in some areas [35]. Moreover, if the climate in a given region gets drier and warmer, water availability will decrease, and the issue will be exacerbated by increasing water demand [36]. For example, it is expected that areas of maize grain cultivation will expand up to 30-50% in Europe [37][38][39][40] with increases of up to 50% in net primary productivity in northern European ecosystems, as a result of a longer growing season and higher CO 2 concentrations [37]. As the projected impacts on productivity of crops and ecosystems included the direct effects of increased CO 2 concentration on photosynthesis, the variation in simulated results attributed to differences between the climate models were, in all cases, smaller than the variation attributed to emissions scenarios [37]. The objective of this study is to estimate future potential blue water availability in Europe and its associated uncertainty, which is induced by emissions scenarios and climate change models. This study first proposes a methodology to conduct climate change analyses in water resource systems, which is based on a high-resolution geospatial model and the use of information available in public databases. Second, the study evaluates distributed mean annual runoff and its uncertainty in main rivers within Western Europe in the baseline period and in two future periods. Third, the study analyses water availability changes and its uncertainty across Western Europe under different climate change scenarios and climate models. Finally, the study analyses the geographically distributed relationships at a continental-scale among the mean annual runoff, water availability, and water withdrawals under the baseline and future periods.

Materials and Methods
The methodological approach is detailed in Figure 1. The methodology is based on a high-resolution GIS-based model, named "Water Availability and Adaptation Policy Assessment (WAAPA)" which enables the estimation of water availability under many climate scenarios to produce a global picture of the situation [33]. The model assimilates climate and geospatial information seamlessly, accounts for reservoir storage (from an individual reservoir or from a system of reservoirs), and produces blue water availability estimates. The model computes net blue water availability for consumptive use of a river basin, taking into account the regulation capacity of its water supply system, and a set of management standards defined by water policy. The model estimates the water availability not only at the outlet of sub-basins (e.g., river intersections), but also at any desired point of the defined river network (e.g., each dam location), by accounting for the entire system of dams in the upstream basin. Basic components of WAAPA are reservoirs, inflows, and demands and they are linked to nodes of the river network. The joint reservoir operation model simulates the behaviour of a set of reservoirs that supply water for a set of prioritized demands, complying with specified ecological flows and accounting for evaporation losses. increases of up to 50% in net primary productivity in northern European ecosystems, as a result of a longer growing season and higher CO2 concentrations [37]. As the projected impacts on productivity of crops and ecosystems included the direct effects of increased CO2 concentration on photosynthesis, the variation in simulated results attributed to differences between the climate models were, in all cases, smaller than the variation attributed to emissions scenarios [37]. The objective of this study is to estimate future potential blue water availability in Europe and its associated uncertainty, which is induced by emissions scenarios and climate change models. This study first proposes a methodology to conduct climate change analyses in water resource systems, which is based on a high-resolution geospatial model and the use of information available in public databases. Second, the study evaluates distributed mean annual runoff and its uncertainty in main rivers within Western Europe in the baseline period and in two future periods. Third, the study analyses water availability changes and its uncertainty across Western Europe under different climate change scenarios and climate models. Finally, the study analyses the geographically distributed relationships at a continental-scale among the mean annual runoff, water availability, and water withdrawals under the baseline and future periods.

Materials and Methods
The methodological approach is detailed in Figure 1. The methodology is based on a highresolution GIS-based model, named "Water Availability and Adaptation Policy Assessment (WAAPA)" which enables the estimation of water availability under many climate scenarios to produce a global picture of the situation [33]. The model assimilates climate and geospatial information seamlessly, accounts for reservoir storage (from an individual reservoir or from a system of reservoirs), and produces blue water availability estimates. The model computes net blue water availability for consumptive use of a river basin, taking into account the regulation capacity of its water supply system, and a set of management standards defined by water policy. The model estimates the water availability not only at the outlet of sub-basins (e.g., river intersections), but also at any desired point of the defined river network (e.g., each dam location), by accounting for the entire system of dams in the upstream basin. Basic components of WAAPA are reservoirs, inflows, and demands and they are linked to nodes of the river network. The joint reservoir operation model simulates the behaviour of a set of reservoirs that supply water for a set of prioritized demands, complying with specified ecological flows and accounting for evaporation losses. In this study, we evaluated the water availability of the joint reservoir operation model following a high resolution and global management scheme ( Figure 2). For each selected sub-basin (derived

Hydrologic Model
Input assimilated by WAAPA

R meets specified reliability?
Iteration: pre-set tolerance Figure 1. General scheme of the applied methodology. The displayed procedure was applied to each defined sub-basin. Grey areas indicate the first path carried out, from the selection of the emissions scenario to the estimation of the water availability. In this study, we evaluated the water availability of the joint reservoir operation model following a high resolution and global management scheme ( Figure 2). For each selected sub-basin (derived from dam locations and river confluences), this scheme considers each reservoir individually and all reservoirs are jointly operated to supply a set of prioritized demands. It is assumed that any demand at a given point in the stream network can be supplied by any reservoir located upstream from it. It corresponds to a situation where there is little development of system interconnections, but a large development of water distribution networks, which are managed globally to supply all demands present in the analysed system. Water is first released (to satisfy demands) from the reservoirs located at low areas of the basin. If these reservoirs are full and receive more contributions, uncontrolled spills are released and water falls out of the system. On the other hand, if upstream reservoirs are full and receive more inflows, the extra water is collected by the downstream reservoirs. This management criterion is not totally real, because real systems usually are managed taking into account more conditions and constraints. The joint reservoir operation model maximizes water availability because it minimizes the excess storage. In each time step, the model performs the following operations:

1.
It satisfies the environmental flow requirement in every reservoir with the available inflow. Environmental flows are passed to downstream reservoirs and added to their inflows.

2.
It computes evaporation in every reservoir and reduces storage accordingly.

3.
It computes excess storage (storage above maximum capacity) in every reservoir (if there is an increment of storage with the remaining inflow).

4.
It satisfies demands ordered by priority, if possible. It uses excess storage first, then available storage starting from higher priority reservoirs.

5.
If excess storage remains in any reservoir, it computes uncontrolled spills.
Water 2019, 11, x FOR PEER REVIEW 4 of 18 from dam locations and river confluences), this scheme considers each reservoir individually and all reservoirs are jointly operated to supply a set of prioritized demands. It is assumed that any demand at a given point in the stream network can be supplied by any reservoir located upstream from it. It corresponds to a situation where there is little development of system interconnections, but a large development of water distribution networks, which are managed globally to supply all demands present in the analysed system. Water is first released (to satisfy demands) from the reservoirs located at low areas of the basin. If these reservoirs are full and receive more contributions, uncontrolled spills are released and water falls out of the system. On the other hand, if upstream reservoirs are full and receive more inflows, the extra water is collected by the downstream reservoirs. This management criterion is not totally real, because real systems usually are managed taking into account more conditions and constraints. The joint reservoir operation model maximizes water availability because it minimizes the excess storage. In each time step, the model performs the following operations: 1. It satisfies the environmental flow requirement in every reservoir with the available inflow. Environmental flows are passed to downstream reservoirs and added to their inflows. 2. It computes evaporation in every reservoir and reduces storage accordingly. 3. It computes excess storage (storage above maximum capacity) in every reservoir (if there is an increment of storage with the remaining inflow). 4. It satisfies demands ordered by priority, if possible. It uses excess storage first, then available storage starting from higher priority reservoirs. 5. If excess storage remains in any reservoir, it computes uncontrolled spills.
The result of the joint reservoir operation model is a set of time series of monthly volumes supplied to each demand, monthly storage values, monthly values of spills, environmental flows, and evaporation losses in every reservoir. Finally, we calculated the system performance by applying the Gross Volume Reliability performance index. This index is the ratio of total volume supplied to demand in the system and the total volume demanded by the system, during the analysed period [33]. In this study, water availability is estimated by considering only one demand present in the system under the hypothesis of 90% reliability. To define the maximum amount of water that can be supplied at a certain point of the river network to satisfy a regular demand, a bipartition method is applied: Excessive values of demands are set (for example, similar to mean monthly runoff) and the simulation is carried out. The deficits are obtained and specified reliability requirements are checked. If the specified reliability The result of the joint reservoir operation model is a set of time series of monthly volumes supplied to each demand, monthly storage values, monthly values of spills, environmental flows, and evaporation losses in every reservoir. Finally, we calculated the system performance by applying the Gross Volume Reliability performance index. This index is the ratio of total volume supplied to demand in the system and the total volume demanded by the system, during the analysed period [33]. In this study, water availability is estimated by considering only one demand present in the system under the hypothesis of 90% reliability.
To define the maximum amount of water that can be supplied at a certain point of the river network to satisfy a regular demand, a bipartition method is applied: Excessive values of demands are set (for example, similar to mean monthly runoff) and the simulation is carried out. The deficits are obtained and specified reliability requirements are checked. If the specified reliability requirements are not fulfilled, the demand is reduced by half and simulated again. If the specified reliability requirements are satisfied, half of the difference is added and simulated again, and so on until the deficit (or gain) is smaller than a pre-set tolerance (e.g., 0.1 hm 3 /year).

Case Study
The area under analysis is composed of the major river basins in Western Europe. WAAPA model data are geographically referenced ( Figure 3). Following, we present the data used to build the WAAPA model. We determined the topology of the model by dividing the area under study into a number of units of analysis, which are homogeneous sub-basins from the water management perspective. The sub-basins are related through the "drain to" relationship, and the analysis is applied to all possible basins, from the small headwater sub-basins to the largest basin draining to the sea. In this work, we divided western Europe into sub-basins (3839), based on the Hydro1k data set (1.538 sub-basins [41]), and the derived-from dam locations (2.301 sub-basins), which belong to 621 large basins draining to the sea. The total area under study is over 6,000,000 km 2 .   Figure 4 shows the comparison of streamflow change from reference  to climate change RCP4.5 scenarios, both for short (2020-2059) and long term (2060-2099), and over the five climate models. Figure 4 is dimensionless (percentage), and the values were obtained by applying Equation 1. The red shading represents a decrease (negative values) and green shading an increase (positive values) of the future mean annual runoff. The yellow shading represents no changes of mean annual runoff for future periods compared to the reference scenario. Naturalized streamflow was obtained from the results of the application of the PCRGLOBWB model [42] to the Inter-Sectoral Impact Model Inter-Comparison Project [43]. The PCRGLOBWB model was run for the entire globe at 0.5 • resolution, using forcing from five global climate models under historical conditions and climate change projections, corresponding to four Representative Concentration Pathways scenarios: RCP2.6, RCP4.5, RCP6.0, and RCP8.5. The following climate models were used as input: GFDL-ESM2NM (GFDL), HadGEM2-ES (HadGEM2), IPSL-CM5A-LR (IPSL), MIROC-ESM-CHEM (MIROC), and NorESM1-M (NorESM1).

Results
Three time periods were considered: Reference (1960-1999), short term (ST, 2020-2059), and long term (LT, 2060-2099). Since runoff obtained from climate model input usually presents significant bias, average runoff values were corrected for bias using the UNH/GRDC (University of New Hampshire/Global Runoff Data Centre) composite runoff field, which combines observed river discharges with a water balance model [44], and is a reference of the current global surface runoff [34,44,45]. Following González-Zeas [45], we applied the bias-correction methodology based on the determination of a monthly correction factor. We calculated the monthly mean runoff series for the control scenario to obtain twelve representative statistical parameters: The ratios between the UNH/GRDC values (observed) and the simulated runoff. These multiplying factors were used to correct bias in the control and the projected series. The reservoir storage volume available for regulation in every sub-basin was obtained from the ICOLD World Register of Dams [46]. Dams in the register with more than 5 hm 3 of storage capacity were georeferenced and linked to the corresponding storage capacity and flooded area (2.301 dams). Environmental flows were computed through a hydrologic method. The minimum environmental flow was set to the 10% percentile of the marginal monthly distribution, according to Spanish legislation. In the absence of more advanced methods, the Spanish regulation for river basin plans establishes several hydrologic methods to define minimum environmental flows [31]. One of them is based on the percentile of the marginal distribution of monthly flows, defining a range between 5 and 15%.
In this study, we estimated current, short-, and long-term geographically distributed water withdrawals. Country-based data on current freshwater withdrawal were taken from the World Bank database. These data were spatially distributed using proxy variables: Population density for urban and industrial withdrawals and irrigated area for agricultural withdrawals. The population density was obtained from the Gridded Population of the World product of the Global Rural-Urban Mapping Project (GRUMP), available at the Centre for International Earth Science Information Network (Figure 3b) [47]. The area potentially under irrigation was estimated from the Global Map of Irrigated Area dataset [48]. Future withdrawals were estimated using the projections of population and gross domestic product (GDP) provided by IIASA. These projections were estimated following RCP scenario assumptions [38,39]. Projections of total freshwater withdrawal and industrial withdrawal were estimated from regressions based on World Bank data using per capita GDP projections [40]. Figure 4 shows the comparison of streamflow change from reference  to climate change RCP4.5 scenarios, both for short (2020-2059) and long term (2060-2099), and over the five climate models. Figure 4 is dimensionless (percentage), and the values were obtained by applying Equation (1). The red shading represents a decrease (negative values) and green shading an increase (positive values) of the future mean annual runoff. The yellow shading represents no changes of mean annual runoff for future periods compared to the reference scenario.

Results
Mean annual runoff change = Mean annual runoff at future scenario Mean annual runoff at reference scenario − 1 × 100 Overall, the models produce a smooth picture of mean annual runoff change in Europe, with decreases in the South. Severe negative changes are projected in the Iberian Peninsula, from the Black Sea in the South almost to the Baltic Sea in the North, and predominantly positive changes are projected in western to central Europe and in northern Europe. A mixed pattern with higher variability in mean annual runoff is shown across central Europe and the Carpathians. The climate models that produce more annual runoff reduction are HadGEM2 and NorEsM1. However, it can be seen that the values and spatial extent of the regions with reduced streamflow (in brownish colours) vary significantly from one climate model to another. This is remarkable considering that all simulations were performed with the same hydrologic model. As expected, in general, the changes are more intense in the long-term period. The region of neutral changes (represented in yellow) moves toward the north from low carbon (RCP2.6) to high carbon (RCP8.5) emissions scenarios (not shown).   The spatial distribution of changes (between the long term and reference periods) of potential water availability along the major rivers in Europe is presented in Figure 6, for all emissions scenarios and climate models analysed. Figure 6 is dimensionless, and the values were obtained by applying    The spatial distribution of changes (between the long term and reference periods) of potential water availability along the major rivers in Europe is presented in Figure 6, for all emissions scenarios and climate models analysed. Figure 6 is dimensionless, and the values were obtained by applying The spatial distribution of changes (between the long term and reference periods) of potential water availability along the major rivers in Europe is presented in Figure 6, for all emissions scenarios and climate models analysed. Figure 6 is dimensionless, and the values were obtained by applying an equation similar to Equation 1, but using potential water availability instead of runoff. Red shading represents a decrease (negative values) of the future potential water availability and green shading an increase (positive values). The yellow shading represents no changes of potential water availability compared to the reference scenario. Although, in general, the climate models show a gradient of potential water availability changes with larger reductions in South Western Europe and larger increases in Northern Europe, values show important differences by comparing the results among climate models (same emissions scenario). By comparing the maps within each column ( Figure 6), we visualize important differences in the results from one climate model to another, and by keeping each emissions scenario unaltered. The models that produce the most potential water availability reduction are HadGEM2 and NorEsM1, while IPSL and MIROC produce the least reductions. On the other hand, by comparing the maps within each row (Figure 6), we observe the different results obtained for the same climate model and different emissions scenarios. It can be seen that, in general, differences among the emissions scenarios (for each climate model) are smaller than those among different models (for each emissions scenario). The driest scenario is RCP8.5 for all analysed climate models. Figure 7 shows, for each analysed sub-basin, the changes in the potential water availability in the long-term period with respect to the reference period (y axis), as a function of changes in the mean annual runoff in the long-term period with respect to the reference period (x axis), for all emissions scenarios and the climate model GFDL. The equations used to plot the results are similar to the proposed Equation 1 for the runoff variable (see Figure 4) and the proposed for the water availability variable (see Figure 6). Quadrant 1 (q.I) shows sub-basins where runoff decreases in the future and water availability increases. Both runoff and water availability increase in q.II, runoff increases and water availability decreases in q.III, and both runoff and water availability decrease in q.IV. In addition, basins with the same reduction of runoff experience different reductions in availability as a result of changes in the hydrologic variability and their different regulation capacity. Figure 8 shows the spatial distribution of the ratio of the runoff, water availability, and water withdrawal for the model GFDL in emissions scenario RCP4.5 for the reference (1960-1999) and long-term period (2060-2099). The bottom row shows potential water availability as a fraction of runoff, the central one shows water withdrawal, also as a fraction of runoff, and the upper row shows the water withdrawal as a fraction of water availability.  Figure 7 shows, for each analysed sub-basin, the changes in the potential water availability in the long-term period with respect to the reference period (y axis), as a function of changes in the mean annual runoff in the long-term period with respect to the reference period (x axis), for all emissions scenarios and the climate model GFDL. The equations used to plot the results are similar to the proposed Equation 1 for the runoff variable (see Figure 4) and the proposed for the water availability variable (see Figure 6). Quadrant 1 (q.I) shows sub-basins where runoff decreases in the future and water availability increases. Both runoff and water availability increase in q.II, runoff increases and water availability decreases in q.III, and both runoff and water availability decrease in q.IV. In addition, basins with the same reduction of runoff experience different reductions in availability as a result of changes in the hydrologic variability and their different regulation capacity.  Figure 8 shows the spatial distribution of the ratio of the runoff, water availability, and water withdrawal for the model GFDL in emissions scenario RCP4.5 for the reference (1960-1999) and longterm period (2060-2099). The bottom row shows potential water availability as a fraction of runoff, the central one shows water withdrawal, also as a fraction of runoff, and the upper row shows the water withdrawal as a fraction of water availability.  Figure 9 shows the uncertainty associated with the climate models and emissions scenarios, both for mean annual runoff and mean water availability, by calculating the coefficient of variation (CV, standard deviation divided by mean) in each calculation point, for each climate model (five), for each emissions scenario (four) and for the short term (ST) and the long term (LT). We represented the probability distribution function (Pdf) of the CVs in each case. Continuous lines represent the uncertainty for each climate model, obtained by comparing the four emissions scenarios for each climate model. The dashed lines represent uncertainty for each emissions scenario, obtained by comparing the CV of the five models for each emissions scenario. Figure 9a,c shows the uncertainty associated with runoff for the ST and LT, respectively. Comparatively, Figure 9b,d shows the uncertainty associated with availability for the ST and LT, respectively. Water 2019, 11, x FOR PEER REVIEW 11 of 18  Figure 9 shows the uncertainty associated with the climate models and emissions scenarios, both for mean annual runoff and mean water availability, by calculating the coefficient of variation (CV, standard deviation divided by mean) in each calculation point, for each climate model (five), for each probability distribution function (Pdf) of the CVs in each case. Continuous lines represent the uncertainty for each climate model, obtained by comparing the four emissions scenarios for each climate model. The dashed lines represent uncertainty for each emissions scenario, obtained by comparing the CV of the five models for each emissions scenario. Figure 9a,c shows the uncertainty associated with runoff for the ST and LT, respectively. Comparatively, Figure 9b,d shows the uncertainty associated with availability for the ST and LT, respectively.

Discussion
Both for short-and long-term periods, the models show similar spatial patterns of mean annual runoff changes in Europe, with decreases in the south, especially in the south-west and increases in the north. Our results agree with global and continental-scale studies that reported mean annual runoff projections [1,49,50]. These studies provide a coherent pattern of change in annual runoff, predicting with a high degree of confidence severe decreases (up to 40%) of surface runoff in areas already affected by water scarcity, like the Mediterranean region, and are consistent with the projected runoff increases in northern Europe (5-30%). However, it can be seen that the values and spatial extent of the regions with reduced streamflow vary significantly from one climate model to another. It suggests that there is an important climate model uncertainty, being the changes of mean annual runoff among emissions scenarios (and the same climate model) smaller than those among the different climate models (same emissions scenario).
In the short-term runoff, Figure 9a clearly shows that higher CV values are more frequent (amplitude of each Pdf curve) by comparing the results among models (and the same emissions scenario, dashed lines) than among emissions scenarios (and the same climate model, continuous

Discussion
Both for short-and long-term periods, the models show similar spatial patterns of mean annual runoff changes in Europe, with decreases in the south, especially in the south-west and increases in the north. Our results agree with global and continental-scale studies that reported mean annual runoff projections [1,49,50]. These studies provide a coherent pattern of change in annual runoff, predicting with a high degree of confidence severe decreases (up to 40%) of surface runoff in areas already affected by water scarcity, like the Mediterranean region, and are consistent with the projected runoff increases in northern Europe (5-30%). However, it can be seen that the values and spatial extent of the regions with reduced streamflow vary significantly from one climate model to another. It suggests that there is an important climate model uncertainty, being the changes of mean annual runoff among emissions scenarios (and the same climate model) smaller than those among the different climate models (same emissions scenario).
In the short-term runoff, Figure 9a clearly shows that higher CV values are more frequent (amplitude of each Pdf curve) by comparing the results among models (and the same emissions scenario, dashed lines) than among emissions scenarios (and the same climate model, continuous lines). In addition, the uncertainties associated with the emissions scenarios are also similar among them (differences among continuous lines for the y axis). Although climate change models are the most robust tools available to generate consistent climate change projections, they are still a source of considerable uncertainties [10,51]. In this regard, Garrote [18] highlighted that the uncertainty has not been reduced with the progressive improvement of modelling tools; on the contrary, it seems to be increasing as a result of the evolving approach to generating emissions scenarios. On the other hand, results suggest that, because of the number of variables and complexity involved in the estimation of the future climate, its estimation has an implicit uncertainty that should be acknowledged for the development of climate adaptation plans. In the long-term runoff (Figure 9c), the uncertainty increases (increasing the amplitude of the Pdf curves) for both climate models and emissions scenarios, although climate models remain more uncertain than emissions scenarios. Also, greater dispersion of uncertainty is found among models than among emissions scenarios. It could be partially explained by the increase of the differences between emissions scenarios for the long-term analysis. These results are consistent with several inter-comparison studies that also show considerable variability in the magnitude and timing of the projected runoff [9,49,50,52,53]. At this point, it is remarkable that all simulations in this study were performed with the same hydrologic model. Databases of climate scenarios are available from different research projects [54,55], including surface runoff among their output variables. As the characterization of the water cycle in the models used in these types of studies usually is very simple and results provide a low signal-to-noise ratio (especially in arid and semi-arid regions), varying the large-scale hydrological models incorporates an additional source of uncertainty [18,50,52]. Some authors state that hydrologic model uncertainties are less significant than those originating from climate change models [9,56].
Changes in potential water availability in short-and long-term scenarios according to all climate models and emissions scenarios were analysed. High resolution results showed similar future spatial patterns to mean annual runoff, with the differences among the emissions scenarios (for each climate model) being smaller than those among different models (for each emissions scenario). Figure 9b shows that the uncertainty associated with the emissions scenarios increases and their values draw near to the climate model uncertainties. Furthermore, the Pdfs of the uncertainty associated with the climate models for water availability remain similar to that for runoff. Similar behaviour is observed for the long-term period (Figure 9d). It suggests that the management of hydraulic infrastructures (mainly reservoirs in this study) plays an important role by decoupling water availability from hydrologic variability. This is observed for all climate models and emissions scenarios considered. Svensson et al. [57] reinforced the importance of the installation of reservoirs in several river basins in Europe in the last century, by attenuating the basins' drought conditions. For quantifying and summarizing purposes, Table 1 shows the emissions scenarios' and climate models' uncertainty for the 50% probability of exceeding CV values. Several local and regional studies agree that the propagation of the uncertainties affects water resource system performances [26,[58][59][60]. Thus, the assessment (or projection) of the performance of a water resources system should be evaluated with extreme care. As previously stated, the reservoir operation model applied in WAAPA is highly simplified and was designed to maximize water availability. Thus, the reality of reservoir operation is much more complex. Usually, not all reservoirs in the basin are jointly managed to supply all demands. They are either managed individually to supply local demands or grouped in systems that are managed independently. Availability of storage volume for water conservation management is also variable according to local conditions, due to the need to allocate storage volume to flood control. Therefore, it is unlikely that upstream reservoirs are kept full to release space in downstream reservoirs. Normal operation would tend to balance storage in all reservoirs to prevent uncontrolled spills. In practice, the spatial pattern of water availability will differ from that obtained in WAAPA. WAAPA results should only be considered as an upper bound of the actual water availability that could be obtained in practice.
Results from the comparisons of the changes in potential water availability with changes in runoff clearly show how changes in the former are not proportional to changes in the latter, suggesting the inadequacy of methodologies that estimate availability as a fraction of mean annual runoff. As an example, in Figure 5, the red line shows the traditional value of 40% of the mean annual runoff adopted for water availability when no simulation of reservoir regulation is performed [61]. It can be seen that adopting this constant value as a proxy of water availability can be strongly misleading, since only those basins with very regular flow or very large reservoir storage can reach this value. In most basins, water availability is a smaller fraction of the mean annual runoff. As shown, availability and withdrawal are only a small fraction of runoff in most of Europe and their projected changes are small, except for the south-east and the south-west. However, the representation of the water withdrawal as a fraction of water availability (Figure 8, upper row) shows that these two variables have similar values in many regions of Europe, and that they are getting closer in the long-term scenario. It means that in many regions, water shortage struggles to satisfy the demand with a specific reliability could emerge or increase, both for the present and future periods. It can also be seen that the relationship between these variables is complex, and that it varies significantly among regions, depending on hydrologic regime, climate, reservoir storage, and socioeconomic factors.
Green water (not analysed in this study), similarly to blue water, is also expected to decrease in most of western Europe except for northern countries. However, changes in green water result from complex interplay of impacts on precipitation, temperature, and CO 2 concentration, which ultimately affects potential evapotranspiration, soil moisture conditions, and growing periods. Thus, patterns of expected changes differ for green and blue water [62]. Irrigation demands will also be affected, due to modified seasonal patterns and evapotranspiration demands [36,63].
Finally, some limitations of this study should be noted. We estimated the potential water availability (upper theory limit) by considering only one demand present in the system. System performance was evaluated as gross volume reliability. Potential water availability was obtained under the hypothesis of 90% reliability. The data used in this study were obtained from specific climate models and emissions scenarios, thus, the conclusions derived from this study are inextricably affected by the models' uncertainty. Additionally, we made a series of simplifying assumptions. We assumed variable geographic and temporal water withdrawals, both in the present and future climate, from indirect methods (GDP and population). We assumed that the reservoirs, whose sole purpose was hydropower generation, were not included in the systems to manage the water resources. We considered that the hydraulic infrastructure corresponding to each analysed sub-basin (determined from a given point in the stream network) was being jointly managed to supply global demands, while in some real cases it could have been divided in to several rather independent subsystems. Furthermore, in our model, there were no system interconnections nor a large-scale water distribution infrastructure. We did not consider other sources of uncertainty as, for instance, the observed climate data source or the hydrologic model applied and the inclusion of regional climate models (RCMs). It is expected that RCMs have less associated uncertainty than GCMs when a particular region is analysed, as they account for more detailed and specific regional characteristics.

Conclusions
This study presents the potential water availability changes under alternative climate change scenarios in western Europe. Results are geographically referenced at high resolution across the major European river basins. The study includes the estimation of the associated uncertainties, resulting from differences among climate change scenarios and climate models. The authors are not aware of similar studies conducted at such a high-resolution continental scale. In this study, we applied the WAAPA model on a high-resolution dataset to analyse water availability changes across western Europe. The proposed model and the applied methodology demonstrated their ability to perform regional studies covering extensive domains, while maintaining high resolution on the characterization of the systems. The climate models that produced the most reduction of mean annual runoff and potential water availability were HadGEM2 and NorEsM1, while IPSL and MIROC produced the least reduction. Overall, for both mean annual runoff and potential water availability, a gradually varying picture of change in Europe was observed, with a decrease in the south (especially in the south-west) and an increase in the north. Moreover, the region of neutral changes moves to the north, from low carbon (RCP2.6) to high carbon (RCP8.5) emissions scenarios. Climate model uncertainties for mean annual runoff and potential water availability were found to be higher than scenario uncertainties. This conclusion was derived by comparing the variability of the results obtained, while the PCRGLOBWB model was forced with different climate models under the same emissions scenario to that of the results from different emissions scenarios for the same climate model forcing. Thus, although climate change models are the most robust tools available to generate consistent climate change projections, they are still a source of considerable uncertainties and their results should be carefully used for operative purposes.
While potential water availability and water withdrawal are only a small fraction of runoff in most of Europe for current and future scenarios (except in the south-east and the south-west of Europe), water withdrawal and water availability are similar in many regions of Europe, and they are getting closer in the long-term scenario (2060-2099). Thus, the balance between water availability and withdrawals is threatened in some regions. Furthermore, social factors, like management of hydraulic infrastructure, play an important role by decoupling water availability from hydrologic variability. This is observed for all climate models and emissions scenarios considered. Finally, although this study presents significant progress in terms of spatial scale and detail compared to previous studies, it is still only indicative of the importance of regional change, due to the assumptions and uncertainties discussed. Nevertheless, the results are useful for envisioning potential water resource system conflicts and contributing to the identification of regions where an in-depth analysis may be necessary.