*2.3. Hydrological Simulations*

Hydrological impacts are simulated using sub-models extracted from the European scale hydrological model E-HYPE v2.5 [31]. E-HYPE is an application of the Hydrological Predictions for the Environment (HYPE) model developed in the Swedish Meteorological and Hydrological Institute [32]. The model is process-based, semi-distributed and designed for hydrological modeling at different spatial scales also in ungauged regions. The source code for HYPE (v4.10) is available at [33]. The model calibration and evaluation details together with a list of used topographical and land-use data sets can be found from [31]. E-HYPE was chosen due to its large spatial coverage, distributed nature and also to see how a hydrological model commonly used in impact assessments [34] responds to the bias adjustment step. Because the model is distributed (over sub-basins of median size 215 km2), spatial variability in biases can also be assessed. Hydrological simulations were first conducted using the full 32-year periods. The first two years served to spin-up the hydrological model and have not been included in further analysis.

To sample the varying hydrological conditions in the European region, the model was run with bias adjusted daily mean temperature and precipitation within four sub-models selected from the catchments shown in Figure 1. These catchments have predominantly natural flow conditions which the hydrological model was capable of capturing well in the present-day climate. The northernmost model domain is located in the upper parts of Tornio river catchment, where water stored in snow pack and variations in it strongly regulate the annual cycle of surface hydrology, leading to peak river discharges in late spring and early summer. Two domains with maritime mild climatic conditions, which cover parts of Trent and Ems catchments, are less affected by snow processes and river discharges reach their maximum values in winter months. The southernmost study region is located in the Sava tributary of the Danube river and is characterized by a mixture of Alpine and Mediterranean climates. The seasonal cycle of river discharge has two distinctive peaks here, the first one caused by snow melt in Alpine regions in spring and the latter one by heavy rainfalls in autumn.

**Figure 1.** Geographical locations of the four sub-models selected for the hydrological simulation tests. The sub-models cover parts of (**A**) Tornio, (**B**) Ems, (**C**) Trent and (**D**) Sava river catchments. White dots denote the locations of the discharge gauging stations for which the reference period statistics are shown in Table 2.

**Table 2.** Statistics for daily time-series of simulated river discharge calculated against the observed discharge at the mouth of each hydrological sub-model (cf. Figure 1). The first column shows the catchment name, the second column Nash-Sutcliffe efficiency coefficient and the third column the relative volume error.


The ability of HYPE to simulate river flows in the reference period (1981–2010), when WFDEI is used directly as forcing, is briefly illustrated in Table 2, which shows the Nash-Sutcliffe efficiency coefficient (NSE) and relative volume error (RE) in simulated river discharge for the four gauging stations located at the outlets of the selected sub-models. The NSE values vary from 0.83 in Ems to 0.52 in the Sava region. These values are reasonable considering that E-HYPE has been calibrated uniformly for all of Europe to optimize predictions in ungauged regions. The RE values range from −17.0% to 6.0% with largest deviations seen in the Tornio sub-model, where the model tends to underestimate river discharge volume, particularly during the spring season. These differences are at least partially explained by the limitations of the WFDEI data set; the representation of daily precipitation variability is not sufficient in regions with large topographical variations, and subject to gauge undercatch for which the corrections are particularly uncertain in windy, snow dominated regions. Also temperature discontinuities might have a role in explaining the differences to the observations. In addition, the inherent limitations in the HYPE formulation and parameterisation likely explain part of this discrepancy (as would any other hydrological model).
