Hydrological modelling and future runoff of the Damma 1 Glacier CZO watershed using SWAT . Validation of the model 2 in the greater area of the Göscheneralpsee , Switzerland 3

In this study, we investigated the application of the Soil Water and Assessment Tool (SWAT) for the 10 simulation of runoff in the partly glacierised watershed of the Damma glacier Critical Zone Observatory (CZO), 11 Switzerland. The model was calibrated using daily time steps for the period of 2009–2011, while two different 12 approaches were used for its validation. Initially the model was validated using daily data for the years 2012–2013. 13 Subsequently, the calibrated version of the model was applied on the greater area that drains to the hydropower 14 reservoir of the Göscheneralpsee and includes the Damma glacier watershed, using inflow data. This validation 15 approach can help in assessing model uncertainty under changing land use and climate forcing. Model performance 16 was evaluated both visually and statistically and it was found that even though SWAT has rarely been used in high 17 alpine and glacierised areas and despite the complexity of simulating the extreme conditions of Damma glacier 18 watershed; its performance was very satisfactory. Our novel validation approach proved to be successful since the 19 performance of the model was similarly good when applied for the greater catchment feeding Göscheneralpsee. 20 Finally, we investigated the response of the two regions, Damma glacier and its greater area, to climate change 21 using SWAT and results were compared to a previous study using the models PREVAH and ALPINE 3D. It 22 confirmed that SWAT can predict changes in future runoff and peak flow in alpine areas with the same accuracy 23 as more demanding models such as ALPINE 3D and PREVAH. This study demonstrates the applicability of 24 SWAT in high elevation, snow and glacier dominated watersheds and in quantifying the effects of climate change 25 on water resources. 26


Introduction 27
The use of calibrated watershed models enables researchers and stakeholders to assess the impact of natural and 28 management induced environmental changes and, as many studies have pointed out, is of high importance (Arnold 29 et al., 1998;Abbaspour et al., 2007). However watershed modelling in alpine areas is challenging due to the rough 30 terrain, heterogeneous land cover, extreme climatic conditions and glacier dynamics (Viviroli and Weingartner,31 2004; Rahman et al., 2013). These can also be the factors that increase the inherent uncertainties in watershed 32 models (Kobierska et al., 2013). 33

34
Modelling the impact of climate change in future runoff provides crucial information for the assessment of water 35 resources, water quality, and aquatic ecosystems. However, the uncertainties at this stage of modelling include 36 uncertainties related to climate change scenarios and therefore, in order to reduce the uncertainty in the climate 37 Hydrol. Earth Syst. Sci. Discuss., https://doi.org/10.5194/hess-2018-493 Manuscript under review for journal Hydrol. Earth Syst. Sci. Discussion started: 17 October 2018 c Author(s) 2018. CC BY 4.0 License.

Soil and landuse map 113
In order to better describe the glacier forefield and to reduce the uncertainty of the calibration for the Damma 114 glacier watershed, we created a more detailed soil and landuse map based on the observations, field and 115 experimental data from the Biglink and SoilTrEC projects (Bernasconi et  For the landuse, we used the Corine land cover dataset 2006 (version 16, 100m resolution) produced by the 120 European Environmental Agency (http://www.eea.europa.eu/data-and-maps/data/corine-land-cover-2006-raster-121 2). 122

Climate data 123
Meteorological data from one local weather station and one station of the ANETZ network were used in this study. Climate change scenarios -. The climate change predictions were provided by the EU regional climate modelling 135 initiative ENSEMBLES (van der Linden and Mitchell, 2009) and were based on the emission scenario A1B. The 136 model chains produced by the ENSEMBLES project are a combination of a general circulation model (GCM) with 137 a regional climate model (RCM). In Switzerland, model chain data were interpolated to the locations of the 138 MeteoSwiss stations and the Swiss Climate Change Scenarios CH2011 were created (CH2011, 2011). The method 139 used for the creation of the datasets is the "delta-based approach". For this method the temperature and 140 precipitation predictions are calculated using daily temperature changes (ΔT), and precipitation scaling factors 141 (ΔP). Incoming short-wave irradiation, wind speed and relative humidity were left unchanged. In Switzerland it is 142 predicted that the mean temperature will increase 2.7-4.1°C and the precipitation during the summer months will 143 decrease 18%-24% by the end of the century, in the case when no actions for the mitigation of climate change are 144 taken (CH2011, 2011). 145

146
In this study, three climate scenarios with interpolated data for Gütsch weather station are used. These scenarios 147 are: the CNRM ARPEGE ALADIN scenario, the ETHZ HadCM3Q0 CLM scenario, which predicts the highest 148 (ΔT) and (ΔP) in comparison to the other two, and the SHMI BCM RCA scenario, which predicts the lowest (ΔT) 149 and (ΔP), referred to as CNRM, ETHZ and SHMI respectively. Data resulted from the work of (Bosshard et al., In agreement with the predictions for Switzerland, the scenarios for Gütsch weather station predict warmer and 156 dryer summers and slightly increased precipitation in autumn. The highest ΔT for the near future period is 1.5°C 157 in the mid-summer, 2.5°C in late spring, and below 1.0°C in early summer for the CNRM, ETHZ and SHMI 158 respectively and for the far future period is approximately 5°C in the mid-summer, 4°C along the whole summer 159 and 3°C in early summer respectively. The biggest temperature increase is predicted at the end of the century when  The first step of the calibration was the manual calibration. The manual calibration was followed by an automatic 185 calibration and uncertainty analysis using the SWAT-CUP software with the Sequential Uncertainty Fitting ver. 2 186 (SUFI-2) algorithm for inverse modelling (Abbaspour et al., 2007). Starting with some initial parameter values, 187 SUFI-2 is iterated until (i) the 95% prediction uncertainty (95PPU) between the 2.5th and 97.5th percentiles 188 include more than 90% of the measured data and (ii) the average distance between the 2.5th and 97. Pearson's product moment correlation is indicated with R 2 . The NS shows the relationship between the measured 194 and the simulated runoff (Eq. 1). R 2 (Eq. 2) represents the proportions of total variance of measured data that can 195 be explained by simulated data. Better model performance is considered when both criteria are close to 1. NS 196 coefficients greater than 0.75 are considered ''good,'' whereas values between 0.75 and 0.36 as ''satisfactory'' 197 (Wang and Melesse, 2006). 198 (2) 202 203 Where is the individual measured value, ̂ for the individual simulated value and ̅ the mean measured value. 204 205 After calibration, we applied SWAT for the greater area that feeds the Göscheneralpsee, using the parameters that 206 resulted from the calibration of the model for Damma watershed. This means that in this case SWAT was set up 207 using the input data (DEM, soil, landuse and meteorological data) described above but was not recalibrated. The 208 parameters were adjusted according to the calibration of Damma watershed. Results of the model were then 209 compared to data of the input of the Göscheneralpsee reservoir that were provided by the energy company that 210 manages the reservoir. We consider that this is a validation step that helps to assess the performance of the model 211 for an area with similar characteristics to that of the Damma watershed. This approach can prove to be useful in 212 cases where there is scarcity in meteorological data or upscaling of the model is needed. 213 4 Results and Discussion 214

Model Calibration 215
The most sensitive parameters during manual calibration are the ones related to snow melt such as: i) TIMP, which 216 is the parameter for the snow pack temperature lag factor, ii) SMFMX, the snow melt factor on the 21st of June 217 (mmH2O / o C-day), iii) SMFMN, the snow melt factor on the 21st of December (mmH2O / o C-day), CN_FROZ, 218 which was set to active in order for the model to consider the frozen soil and finally the snow fall and snow melt 219 temperature SFTMP and SMTMP respectively. Groundwater flow parameters such as the GW_DELAY, the 220 groundwater delay time, ALPHA_BF, the base flow alpha factor and the SURLAG, the surface runoff lag 221 coefficient, were also found to be sensitive.

Model Validation 252
The model was validated using two different approaches. For the first approach, we validated the model, using the 253 meteorological data for 2012 and 2013 and results are presented in Fig. 4(a). The accumulative graph in Fig. 4

(b) 254
reveals that there is the same trend in the validation period as the one observed in the calibration period, with best 255 fit during spring and early summer while in July and August the estimated runoff is slightly lower. A small 256 exemption to that is in late spring of 2012, when estimated runoff is underestimated, probably due to the extremely 257 wet May in that year that cannot be efficiently simulated. Overall, model performance during validation is very 258 satisfactory and almost identical to that of calibration, with Nash-Sutcliffe efficiency of 0.85 and R 2 of 0.86. 259 Therefore, the model is considered to be validated. The small seasonal differences in model performance are due 260 to the fact that runoff in spring and early summer, that is from May till June, originates mainly from snowmelt 261 while in July and August originates from glacier melt. Although there are two different water sources during the 262 two different periods, we can only assign one set of parameters. Nevertheless, differences are very small and 263 therefore, it is confirmed that SWAT can be successfully applied for a partly glacierised watershed. 264

265
The results of the model for the greater area that feeds the Göscheneralpsee, are presented in Fig. 5(a) satisfactory results. The efficiency of inflow predictions (NS) dropped to 0.49 and the R 2 to 0.72, which are 268 however satisfactory. The observed and predictive accumulative flow is presented in Fig. 5(b). 269 270 This successful validation approach shows that the model can be applied, without recalibration, for a greater area 271 with similar climatic conditions. This methodology can be an ideal way of model validation in studies that include 272 climate change scenarios, since it helps in assessing the uncertainties that occur due to climate change or evolution 273 of land use. It can also be useful in cases where there is scarcity of runoff data but good quality of GIS input data. satisfactory. Its performance is comparable to that of PREVAH and ALPINE3D, however its advantage over the 288 other two models is that is a model widely used around the globe for different areas and projects, with easily 289 available input data. This makes it an ideal choice for water managers and policy makers. ALPINE3D and 290 PREVAH models have been used mainly in mountainous areas and have high requirements in meteorological data 291 and computational time. The results of SWAT model are presented as the average runoff for each different scenario along the whole period 300 in Fig. 7(a) for the near and in Fig. 7(b) for the far future periods. The results show many similarities regarding 301 the seasonality of runoff for the three different scenarios for all the simulation periods. 302 303 During the reference period, which describes the current situation, runoff peaks in early July when snowmelt is 304 combined with glacier melt. For the near future period T1, the main difference is noted from July to September 305 when runoff is dominated by glacier melt. During this period, predicted runoff for all scenarios, and in particular 306 for the warmer ETHZ scenario, is lower than the reference period, indicating that the glacier melt cannot 307 Hydrol. Earth Syst. Sci. Discuss., https://doi.org/10.5194/hess-2018-493 Manuscript under review for journal Hydrol. Earth Syst. Sci. Discussion started: 17 October 2018 c Author(s) 2018. CC BY 4.0 License. compensate the decrease in the precipitation, estimated by the future climatic data. From September until the end 308 of the season, simulated stream flow of all scenarios is higher than the one of the reference period, which is 309 explained by the higher predicted precipitation during autumn. Moreover, the predicted prolonged warm period 310 leads to increased glacier melt and therefore higher runoff. The annual peak continues to be observed in early July, 311 since the glacier hasn't melted away yet providing a significant source of water through glacier melt. 312 313 For the far future period T2, runoff from spring to mid-June is predicted to be significantly higher for all three 314 scenarios than that of the reference period. This can be explained by considering that the warming climate leads 315 to faster rates of snow melt and increased runoff in snowmelt dominated period. In addition, higher precipitation 316 is predicted by the climatic data for this period. In 2070 the total volume of the Damma glacier is estimated to be 317 reduced to almost half, resulting in significant drop of the volume of water that comes from glacier melt between 318 July and late August. For this reason, and in combination with the significant decrease in precipitation predicted 319 by all scenarios for this period, the simulated runoff is lower than that of the reference. Finally, the snow free 320 period of the watershed is prolonged until December instead of September. 321 322 At the end of the far future period, the average temperature increase in our site will be 3.35 o C and only a small 323 part of the Damma glacier will be left in high elevation. The peak of the highest runoff is significantly shifted and 324 is projected to be in the beginning of June. The main volume of runoff is expected to be observed in spring and 325 early summer while during the glacier melt period, streamflow is significantly lower than that of the reference 326 period. Overall the total water yield for the far future scenario is significantly decreased. These findings are of 327 great significance for water resources management. 328

329
In order to better observe the seasonal changes of estimated runoff, Fig. 8 shows the average runoff for a) May-330 June, b) July-August and c) September-October for the T1 and T2 future periods divided by the average of the 331 reference period of the same months for all the three scenarios. In May and June, as mentioned above, runoff is 332 mainly dominated by snowmelt. The three climate change scenarios predict increased temperatures and higher 333 precipitation during May and June which result in faster snowmelt and therefore in the increased predicted runoff, 334 as observed in Fig. 8(a). The increase is higher in the far future period due to the higher temperatures. The only 335 exemption to that is the SHMI scenario for the near future period, since it is the colder scenario that predicts the 336 lowest temperature and precipitation changes. In July and August climate scenarios predict a significant decrease 337 in precipitation, which is also depicted in the predicted runoff. As seen in Fig. 8(b), predicted to reference runoff 338 ratio is considerably below one, especially in the far future period. The scenario that has the most drastic effect is 339 the ETHZ because it is the warmer scenario that predicts the highest increase in the temperature and decrease in 340 the precipitation. Finally, for September and October, results do not show a clear trend for the warmer ETHZ 341 scenario, however for the CNRM and SHMI scenarios, future runoff is lower than the reference. The big shifts in 342 the ratio especially in the far future period T2 can indicate the increase in extreme events. Climate change predictions showed that daily and total annual runoff will change significantly in the future 362 especially towards the end of the century. Daily runoff during spring and early summer, in May and June, is 363 predicted to increase because of faster snow melt and the predicted wetter springs. Projected runoff from July to 364 October for the far future period, when the major part of Damma glacier will have already disappeared, but also 365 for the near future, is significantly decreased. These results proved that SWAT shows sensitivity for the modelling 366 of glacier melt, which is crucial for the climate change assessment and therefore can be a useful tool for water 367 managers and policy makers. 368

Author Contributions 369
Maria Andrianaki applied SWAT model, analysed data and prepared the manuscript with contributions from all 370 co-authors. Juna Shrestha reviewed the manuscript and assisted in the modelling procedure. Florian Kobierska 371 provided meteorological and runoff data. Nikolaos P. Nikolaidis provided guidance for the research goals. Stefano 372 M. Bernasconi was the supervisor of the research project and provided the funding that lead to this publication. 373

Competing interests 374
The authors declare that they have no conflict of interest. 375