Impacts of climate change on stream flow and hydro power generation in the Alpine region

In order to provide meteorological input (temperature and precipitation amount) for the hydrological model in the Greater Alpine Region (GAR; see Fig. 1), four carefully selected (see below) regional climate model (RCM) simulations from the ENSEMBLES project (van der Linden and Mitchell 2009) were used. Within ENSEMBLES, 23 RCM simulations are available based on 8 different global climate model (GCM) simulations. They are operated at a horizontal resolution of 25 km, which was the highest resolution available covering the entire region and period of interest at the time when this study was conducted. Recently, a new set of regional climate simulations for Europe became available from the EURO-CORDEX initiative (Jacob et al. 2013). These regional climate simulations are downscaling the new CMIP5 global climate projections (Taylor et al. 2012) and the new representative concentration pathways (RCPs) (van Vuuren et al. 2011). Comparisons of EURO-CORDEX results given in Jacob et al. (2013) and Kotlarski et al. (2014) suggest that the model performance for large-scale patterns of changes in mean temperature, precipitation, and related indices in the EURO-CORDEX simulations is similar to those of ENSEMBLES.

The ENSEMBLES RCMs provide daily time series from 1951 to 2050 and are based on the SRES greenhouse gas emission scenario A1B (Nakicenovic et al. 2000). A1B is a moderate emission scenario, but features no large differences to other emission scenarios up to 2050 in its consequences to climate change (e.g. Prein et al. 2011). However, after 2050 these differences become important and a single emission scenario might not be appropriate to cover the bandwidth of possible long-term climate change in the Alpine area. In total, 19 out of 23 ENSMBLES models entered the scenario selection process; four were disregarded due to technical and availability reasons at the time of the study. Four representative climate simulations were chosen that cover the bandwidth of climate change of the entire ENSEMBLES multi-model data set (see Bachner et al. 2013, p. 7–14 for details), and range from hot-dry over moderate to humid-warm—thus, it is expected that relevant climate change conditions for impacts on the hydrological and power generation systems are considered in this study. Although changes in wet or dry spells or similar extreme indicators were not taken into account for model selection, the spread of the selected scenarios can be regarded as representative for model uncertainty within the A1B scenario as demonstrated by Heinrich et al. (2014). For a more detailed description of the selection process, see Bachner et al. (2013). The selected model #1 (Meteo-HC HadRM3Q0) features a very warm and dry climate change signal in summer, and mild conditions in winter. This scenario is considered to result in overall positive effects on the electricity system, due to an expected decline of residual load in winter, although the residual load will slightly increase in summer. However, for hydro power generation, these very warm and dry conditions should yield rather negative effects. Thus, model #1 will be termed “warm-dry” scenario from now on. Model #2 (C4IRCA3) features a very wet and warm climate change signal (CCS) in both summer and winter. It will be termed “humid-warm” scenario. Model #3 (CNRM-RM4.5) features the special case of a stronger summer than winter warming. According to its expected effect on the electricity system, model #3 will be termed “warm-summer” scenario. Model #4 (KNMI-RACMO2) represents more or less an average realization of all 19 considered RCMs. Precipitation is shifted towards winter season, the overall precipitation during the year changes insignificantly. This scenario was chosen as an average scenario, because it does not comprise any outstanding precipitation CCS. It will be termed “moderate” scenario in the following, because the expected effects on the electricity system are moderate in total.

Temperature and precipitation of the selected RCMs have then been bias-corrected independently on daily basis, applying quantile mapping (QM) as described in Themeßl et al. (2011, 2012) and Wilcke et al. (2013). For bias correction, the E-OBS European gridded observational data set (Haylock et al. 2008; van den Besselaar et al. 2011) was used. Wilcke et al. (2013) showed that these corrections (applied in this study) retain the inter-variable dependencies of RCMs and therefore do not add additional uncertainty for further impact assessments in this respect. E-OBS contains gridded daily time series of mean, minimum, maximum air temperature (°C), and precipitation amount (mm/day) on a 25-km grid and is based on the most complete freely available collection of station data over wider Europe (Klok and Klein Tank 2009). A similar data set for global radiation and wind speed does not yet exist for entire Europe and is—amongst other things discussed later in the text—a reason why a simple hydrological model has been chosen that needs only temperature and precipitation as meteorological input.

The purpose of the hydrological model within the model chain is to estimate the impacts of climate change on the discharge characteristics of rivers within the Alpine region at a monthly time step. In other words, a simple yet effective model has been chosen that converts precipitation to stream flow in catchments of variable size and topography by means of temporal storage and delay of precipitation. For model calibration and validation, discharge time series of 101 gauging stations in Austria, Germany, Slovenia, France, Italy, and Switzerland, which were provided by various organizations (see “Acknowledgements”), were compiled. The length of the time series ranges from 10 to 60 years, and the size of the catchments related to the gauging stations ranges from 150 km² to more than 100,000 km² (see Fig. 1). Some of the data were available as monthly runoffs only, some also as daily runoffs. As meteorological input for calibration and validation, the E-OBS data set described in “Climate scenarios and observations” section was used.

Since the hydrological model needs to be applicable for a wide region as well as various catchment sizes and should operate with only precipitation and air temperature as input data, the parsimonious lumped-parameter model GR2M (Perrin et al. 2001, 2003; Mouelhi et al. 2006; Okkan and Fistikoglu 2013, Wagner et al. 2013) has been selected and extended by a snow model that accounts for snow storage and melt based on temperature only (as proposed by Xu et al. 1996). Potential evapotranspiration (PET) is computed based on the formulation of Oudin et al. (2005), as only temperature is available as input parameter here. In addition to temperature, extraterrestrial solar radiation is needed for the computation of the PET. Following Oudin et al. (2005), extraterrestrial radiation depends on the geographical position of the catchment and the time of the year as it exhibits a seasonal variation but is assumed to remain constant over the years (data taken from http://​www.​cgiar-csi.​org).

A schema of the model structure is shown in Fig. 2. The approach is based on a spatial, temporal, and conceptual lumping which is considered to be a suitable model structure for the purpose of monthly rainfall-runoff prediction due to its parsimony and the limited information content of discharge time series available for calibration and validation (Edijatno et al. 1999; Perrin et al. 2001; Gupta et al. 2005). A monthly time step has been considered most appropriate because of (1) computational efficiency, (2) an accepted capability of monthly rainfall-runoff models to simulate runoff with relatively few parameters (e.g. Mouelhi et al. 2006), and (3) the fact that long historical time series of flow data are consistently available on a monthly time step; generally, daily time step flow data are likely related to more uncertainty and often include larger data gaps (e.g. Smakhtin 2000). Moreover, monthly runoff data provide sufficient input for the subsequent step in the model chain, as for most of the hydro power plants considered in the hydro power model (except for the Austrian hydro power plants) only monthly data were available.

Processes accounted for are snow storage and snow melt, evapotranspiration, soil water storage represented by a soil moisture accounting store (=production store as used in Mouelhi et al. 2006), groundwater storage (=routing store as used in Mouelhi et al. 2006), and water exchange with neighbouring catchments (intercatchment flow; Wagner et al. 2013). There are in total four free (to be calibrated) parameters in the model (in the following called GR2M+): a critical temperature T _S below which snowfall starts and adds to the snow store; a critical temperature T _M above which snow melt starts and decreases the snow store; the maximum capacity X ₁ of the production store; and a groundwater exchange term X ₅, which accounts for water flow to (X ₅ > 1) or from (X ₅ < 1) neighbouring catchments. Such a parameter has been shown to be important, as groundwater flow from or to a neighbouring catchment is often observed and the discharge measured at stream gauges does not take into account groundwater flow within the alluvial aquifer accompanying the stream (Mouelhi et al. 2006; Le Moine et al. 2007; Wagner et al. 2013). However, the routing store (accounting for groundwater storage) is held constant (60 mm; representing a quadratic reservoir), which is a necessary simplification due to the unavailability of detailed local data of groundwater flow, but was shown by Mouelhi et al. (2006) to be effective.

As was pointed out by Gupta et al. (2005), the information content of a runoff time series is limited and a model with only a few free parameters avoids the issue of equifinality. Although incorporating additional processes into the hydrological model is appealing, it bears the risk that the model is overparameterized (Loague and Freeze 1985; Beven 1989; Perrin et al. 2003; Wagner et al. 2013). To demonstrate that the above-described model is able to adequately reproduce the runoff behaviour observed in the past, historical area-weighted average monthly temperature and precipitation data (based on the E-OBS v4 data set) are computed for the individual catchments and used as forcing input in the model. The simulated runoff is compared to the observed runoff from the individual stations, and the four free parameters are optimized to yield a minimum difference between observed and simulated runoff based on three objective functions: (1) the classical Nash–Sutcliffe efficiency criterion (Nash and Sutcliffe 1970), (2) the Nash–Sutcliffe criterion calculated on the square-root-transformed stream flow, and (3) the Nash–Sutcliffe criterion on the logarithm-transformed stream flow (Perrin et al. 2003; Wagner et al. 2013, 2016). The classical Nash–Sutcliffe criterion focuses on optimizing peak flows, the one based on logarithm-transformed stream flow focuses on low flows and the one based on square-root-transformed stream flow gives an intermediate picture of the overall hydrograph fit (e.g. Oudin et al. 2006). In the following, the combined efficiency criteria will be termed \(\overline{\text{NSE}}\) as their value is given as an average number of the three criteria (equal weights) to have the same value range as the individual criterion (−∞ to 1). The optimization of GR2M+ is performed using the Excel Solver.

To evaluate the predictive capability of the model, the model was calibrated using the first half of the data set and validated using the second half and vice versa (split sample test; Klemes 1986). At least one year of warm-up period was assigned to minimize the effects of the initial conditions. An example of a calibrated hydrograph including a split sample test is shown in Fig. 3a using the example of the gauging station Kienstock along the River Danube (catchment size of 95,970 km²; see Fig. 1a for location of the catchment). It is noteworthy that if the model is calibrated using the first half of the time series (1661–1985; \(\overline{\text{NSE}}\) of 74.9%), it is able to simulate the runoff for the second half (1986–2010) rather good (\(\overline{\text{NSE}}\) of 79.1%), even the low flow during the drought period in 2003.

Hence, model validation is based on different efficiency criteria (trade-off in single efficiency criteria to have an overall consistency; the “closeness” of simulated and observed stream flow; Krause et al. 2005), visual inspections of the hydrographs, and split sample tests. This evaluation was done for the whole data set of 101 measurement stations. The average of the three efficiency criteria (\(\overline{\text{NSE}}\)) for all the stations yielded 78.9 ± 6.8% when calibrated for the whole time period, and when validated on one half of the time series after calibrating it on the other half of the time series 75.1 ± 8.5 and 76.1 ± 9.0%, respectively (Fig. 3b). Mouelhi et al. (2006) reported an average Nash–Sutcliffe efficiency criterion of 64.2% for the 410 catchments they analysed, indicating that our results are comparatively good. According to the classification of Nash–Sutcliffe efficiency criteria proposed by Moriasi et al. (2007), nearly all of the calibrated models are rated as “good” (>65%) or “very good” (>75%). Moreover, there is no correlation of model efficiency and catchment size, which suggests that the simple model is able to reproduce stream flow for a wide range of catchment sizes (Fig. 3c).

However, seasonality in discharge time series can produce relatively high NSE values; therefore, a benchmark approach as proposed by Garrick et al. (1978) and later by Schaefli and Gupta (2007) was followed for some of the catchments analysed. The applied model performs superior to the benchmark model (assuming interannual mean monthly runoff values; Schaefli and Gupta 2007); however, for individual catchments, the improvement is small, indicating that seasonality plays an important role in the discharge behaviour of these catchments (e.g. station Golling along the Salzach River). Nevertheless, this is an expected fact of monthly runoff data of alpine catchments.

In addition to the acceptable efficiencies discussed above, the model results are compared to published results of other models for historical data (e.g. Kranzl et al. 2010; Pöhler et al. 2010; Stanzel and Nachtnebel 2010; Klein et al. 2011; Kling et al. 2011; ZAMG/TU-Wien Studie 2011). As an example, the model performance to simulate observed runoff for the two gauging stations Passau-Ingling (26,084 km²) and Golling (3555.7 km²) at the Rivers Inn and Salzach, respectively, is shown in Fig. 4 compared to the simulations by Pöhler et al. (2010), which used the physically based model WaSiM-ETH at a daily time step. A general agreement between observed data, the published simulated data from the distributed model WaSiM-ETH (reduced to a monthly time step), and the simulations from the parsimonious lumped-parameter model used herein (GR2M+) increases the confidence in the appropriateness of the simple model structure to generate monthly runoff data. In addition, the benchmark approach described in Schaefli and Gupta (2007) based on interannual mean monthly runoff values yields NSE values of 0.681 for Passau-Ingling and 0.808 for Golling. Actual NSE values for the runoff simulation of the GR2M+ for the two gauging stations are higher (0.874 and 0.848). On the one hand, this demonstrates an expected strong seasonality of the runoff and, on the other hand, that the model is superior to a simple interannual monthly mean.

The direct impacts of long-term average climate changes can be estimated by investigating changes in the average annual electricity generation of run-of-river plants, which is defined in this paper as the long-term average annual net electrical energy output of every single plant within the model. To convert the simulated river runoffs to the energy output of run-of-river plants, conversion models (e.g. Kishor et al. 2007; Acakpovi et al. 2014) were used. For the hydro power plants, the corresponding runoff gauging station was assigned by using the nearest station along the river. The relationship between the changes in runoff and the changes in the average annual electricity generation of run-of-river plants depends on technical parameters of the individual hydro power plant as well as on possible differences in size of the catchment areas of the gauging station and the hydro power plant.

For the hydro power plants in Austria, the known technical parameters, such as hydraulic head, maximum inflow, and maximum capacity, as well as historical runoff data are used as model input to calculate hydro power generation on a daily basis for each power plant (Schüppel 2010). The model then is calibrated to the historical power generation data by adjusting the unknown efficiency factor of the power plant, assuming as an approximation that the known average head height is constant.where W _el,d (Ws) is the daily electrical energy produced by the power plant at day d, Q _PP,d (m³) is the water volume flowing into the power plant at day d (see below), \(\bar{H}\) (m) is the average hydraulic head, g (m/s²) is the gravitational acceleration. ρ (kg/m³) is the density of water and η _el (–) is the total efficiency of the power plant, used to calibrate the model.

Another variable factor accounting for the aforementioned difference between the catchment areas of the stream gauge and the power plant is calibrated using the average annual electricity generation.where Q _mod (m³) is the historical runoff volume for day d, modified by the results of the hydrological model output, Q _max (m³) is the maximum daily inflow capacity of the power plant, and f (–) is the catchment size correction factor, ratio of model catchment size and plant catchment size.

Based on the monthly simulation results from the hydrological model, changes in the monthly flow duration curves of the nearest gauging stations are computed and used to alter the monthly duration curves from the hydro power plants (derived from historical daily inflow data, delta-sigma approach). Changes in the characteristics (e.g. increasing frequency of high water) are approximated by altering the historical inflow data considering the simulated changes in the standard deviations in the monthly runoff data simulated by the hydrological model. The modified flow duration curves are then fed into the power plant model. Although this is a simple approach of converting monthly to daily inflow data, it is considered a reasonable first approximation. More complex stream flow conversions exist (e.g. Rebora et al. 2016), but are beyond the scope of this paper. The resulting changes in average annual electricity generation and monthly generation coefficients (monthly share of the annual energy yield) are the final results discussed here (and can be used for further economic considerations, as was done within the EL.ADAPT project: Bachner et al. 2013).

Due to a lack of technical information and temporal resolution of runoff data on individual hydro power plants, a less detailed linear approach was used for the hydro power plants in other countries (Italy, Switzerland, Germany, and France), mapping the monthly change signals of runoff directly to the monthly average amounts of electricity generation and neglecting the nonlinear effects of changing runoff characteristics and the limiting inflow capacity (delta approach).where \(W_{{{\text{el}},{\text{m}}}}^{\text{CC}}\) (Ws) is the monthly average electricity generation in the future affected by climate change, \(W_{{{\text{el}},{\text{m}}}}^{\text{hist}}\) (Ws) is the historical monthly average generation, and \(\Delta q_{\% }\) is the relative change in monthly runoff of the corresponding gauge (climate change signal).

Calibration of the individual hydro power stations in Austria results in efficiencies on average of 0.83 (with a standard deviation of 0.12; see Eq. (1) and Bachner et al. 2013, p. 92–94 for details). These values are within a reasonable range (e.g. Giesecke et al. 2014, p. 32). Due to the lack of technical data, efficiencies were not calculated for the stations outside Austria.

At the end of their operational period, hydro power plants will be renewed (refurbished). These refurbishments are assumed to result in an increasing capacity (rated power) by 5% and are correspondingly considered in the average annual electricity generations modified by the hydrological model results. Moreover, an increase in the number of hydro power plants is expected in the future. This increase has been estimated based on various sources, e.g. World Energy Outlook 2010 and the National Renewable Energy Action Plans of EU member states (see Bachner et al. 2013 for more details), and is considered by incorporating additional hydro power plants in the model for the future periods. Yet within this paper, all results are referred to a base scenario that includes the same changes (including additional generation capacities) in the hydro power generation system such that the estimated change in electricity production only results from the hydrological change.

Regarding meteorological forcing, four representative regional climate scenarios have been selected to cover a large uncertainty range of expected climate change. The selected RCMs show different characteristics: Meteo-HC HadRM3Q0 being a hot and dry realization, C4IRCA3 being a warm and wet realization, KNMI-RACMO2 being a moderate realization, and CNRM-RM4.5 representing a special case, which shows stronger summer than winter warming. The climate change signals for all the considered 101 catchments between 1961–1990 and 2011–2030 and between 1961–1990 and 2031–2050 of the selected RCMs are summarized in Fig. 5 in a cumulative frequency plot. For the individual catchments of the gauging stations Kienstock (95,970 km²) and Passau-Ingling (26,084 km²) along the rivers Danube and Inn (which will be used as examples in detail also in the next sections), a change in temperature from +0.7 to +2.5 °C and from +0.7 to +2.7 °C, respectively, and in precipitation from +5 mm/year to +106 mm/year and from −11 mm/year to +111 mm/year, respectively, is estimated (Fig. 5; the triangles and crosses, respectively). The cumulative frequency plots of absolute temperature change and absolute precipitation change illustrate the differences in the climate scenarios. Although the trend in temperature change is consistently positive, the magnitudes are different and of importance concerning future changes in evapotranspiration. Interestingly, the cumulative frequency plots of the individual scenarios are rather steep, indicating a low variability of the projected temperature increase within the Alpine region. The picture of precipitation change is more diverse, on the one hand regarding the differences between the individual scenarios, and on the other hand, regarding the variability within the Alpine region, as most scenarios display a range of positive as well as negative changes of more than 50 mm/year. This is reflecting the complex meteorological behaviour of the Alpine region.

Using temperature and precipitation input from the four climate scenarios individually, the calibrated and validated hydrological model is used to simulate runoff for the two periods 2011–2030 and 2031–2050. The results are compared to the reference period 1961–1990 for all 101 catchments. As representative examples, three gauging stations related to a range of catchment sizes are shown in Fig. 6 (their locations are indicated in Fig. 1a). Figure 6a shows the predicted seasonal change in the mean monthly runoff of the River Danube at the station Kienstock for each of the four climate change scenarios. Figure 6b, c shows the same but for the station Passau-Ingling and Imst of the river Inn. Subtracting the runoff simulated for the reference period 1961–1990 from the predicted mean monthly runoff of the respective time period yields the expected change in monthly runoff for each of the four climate scenarios (Fig. 6d–f). The overall annual changes for the two periods 2011–2030 and 2031–2050 related to the reference period 1961–1990 are shown on the very left of Fig. 6d–e. Note that small absolute increases in the runoff during periods of low flow (e.g. in winter) might yield large relative changes (up to 100% and more, e.g. Fig. 6f); however, these might not have a great influence on the difference in the overall annual runoff.

In general, especially for the smaller (and mostly mountainous) catchments considered, a shift towards earlier runoff (especially months of February to May) and a decrease in the summer months (July to August/September) are observable. Monthly variations up to +190% are observed due to seasonal changes and a shift towards higher runoff especially in March and April related to a warming trend observed in all four scenarios. The strongest decrease of up to −70% is observed in the summer months of July and August (most pronounced in the warm-dry scenario for the period 2031–2050 compared to 1961–1990). An example of both effects being observable is shown for the River Inn at the gauging station Imst in Fig. 6c, f.

In addition to the mean monthly runoffs (μ) as shown in Fig. 6, the standard deviations (σ) of the runoff for each catchment, each month, and each scenario for both time periods 2011–2030 and 2031–2050 versus the reference period 1961–1990 were computed to get a bandwidth of possible runoff changes in the near future (as an example see Fig. 7). The monthly changes in runoff computed for the different climate scenarios were further used as input for the hydro power model (“Changes in the hydro power generation” section).

Depending on the climate scenario used, there is a certain variation in the change in runoff for all the catchments analysed and moreover for the periods 2011–2030 and 2031–2050 compared to the reference period 1961–1990 (Fig. 8). Interestingly, in the warm-dry scenario, a positive trend in runoff change is observed for the period 2011–2030, whereas a strong decrease is estimated for the period 2031–2050. However, this is not unexpected, as the precipitation patterns already demonstrate such a behaviour (Fig. 5). Nevertheless, the consistently positive changes in temperature (Fig. 5a) generally result in more negative changes in runoff (Fig. 8) compared to the changes in precipitation (Fig. 5b). The warm-dry as well as the moderate scenarios for the periods 2031–2050 suggest a consistent decrease in runoff for all the catchments analysed compared to 1961–1990. In contrast, only the warm-dry scenario for 2011–2030 results in a generally positive runoff change compared to 1961–1990. Interestingly, the cumulative frequency plots show clearly that the maximum and minimum changes result from the two time periods considered within a single scenario (the warm-dry one).

As Fig. 8 allows no spatial differentiation between individual regions, Fig. 9a–h shows the relative changes in the mean annual runoff of all catchments considered for the four scenarios and the two time periods. Besides the mentioned differences between individual scenarios as shown in Fig. 8, Fig. 9 reveals very diverse spatial distributions with rather inconsistent runoff changes. However, the south-western part of the study area, namely the Rhone River and its tributaries, exhibits a consistent negative trend for the period 2031–2050. The Rhine River (north-western part), the Danube River (north-eastern part), and also the Po River (south-eastern part) and their respective tributaries do not show a consistent trend.

Since hydro power plants are not able to convert the total available river runoff into electricity due to discharge capacities, environmental measures, and power plant efficiencies, changes in runoff might have diverse effects on the actual power generation as will be discussed in the following. Figure 10 depicts an example of the influence of runoff changes for monthly hydro power production. On the one hand, the power plant generation capacity is limited to a certain amount of discharge due to cost-optimized operation, so that certain changes in stream flow at higher levels do not affect the energy output, because the additional water is passing the hydro power plant via a spillway. On the other hand, runoff shifts, e.g. to earlier months of the year (due to a general warming trend resulting in less snow storage and earlier snow melt), can lead to an increase in power output of the plant if these water volumes were previously lost via the spillway and can now be used to generate electrical energy. As noted in “Hydro power model for run-of-river plants” section, these effects were considered only for power plants in Austria due to the lack of data on the individual power plants in other parts of the Alpine region.

Within the project framework of this study (see Bachner et al. 2013), changes in power plant structures have been considered in the future, which does not allow a direct comparison of the future periods (2011–2030 and 2031–2050) with the historical period (1961–1990). Consequently, a scenario neglecting climate change (i.e. using the unmodified historical flow duration curves for the future time periods), but considering the changes in the power plant structure, i.e. refurbishment or additional capacities, had to be computed. Thus, a comparison between the climate scenarios and this base case (here called “BASE” scenario) was made.

For the two time periods considered, relative changes in average annual electricity generation compared to the BASE scenario (without consideration of climate change) are shown for the entire GAR and for Austria in particular (Fig. 11a, b).

Considering all simulated run-of-river hydro power plants (390 in total), Fig. 11a, b shows that the projected climate change leads to an increase in total average annual electricity generation by 5% in the GAR in the most positive case (warm-dry scenario) for the time period 2011–2030 and to a decrease by 7% in the most negative case (warm-dry scenario as well) for the time period 2031–2050. This is a progression of what was already indicated in the runoff data (see Fig. 8). The bandwidth of projected changes in Austria’s average annual electricity generation is slightly lower compared to those of the GAR, ranging from an increase by almost 4% in the humid-warm scenario for the time period 2031–2050 to a decrease by about 4% in the moderate scenario for the same time period in the most negative case.

Besides the changing annual production, seasonal shifts are very important for the hydro power generation. Hydro power in Austria reaches its maximum production in spring, due to melting snow in the Alps, and its lowest energy output usually in February. However, due to a general warming trend, a seasonal shift is estimated in the future (Fig. 11c–f).

Monthly generation characteristics reflect the seasonal changes in runoff characteristics (see, e.g. Fig. 6): A shift towards the winter and spring (particularly, February to May) and a decrease in the summer months (particularly, July to August) are observed for all climate scenarios in Austria and for all except the warm-summer scenario in the GAR (where the decrease in summer months is not observed). The observed trends are noticeable already in the time period 2011–2030, but are more pronounced in 2030–2050.

The trends are similar in Austria and the GAR in general (Fig. 11c–f), pointing to an increased generation in winter and a decrease in summer. For the GAR having its peak load in winter, this is a positive development, because the additional energy yield supports and potentially stabilizes the energy system.