HYDROLOGICAL PROCESSESHydrol. Process. 20, 2091–2109 (2006)Published online in Wiley InterScience (www.interscience.wiley.com). DOI: 10.1002/hyp.6197

Assessment of climate-change impacts on alpine dischargeregimes with climate model uncertainty

Pascal Horton, Bettina Schaefli, Abdelkader Mezghani, Benoıt Hingray* and Andre MusyEcole Polytechnique Federale de Lausanne, Hydrology and Land Improvement Laboratory, CH-1015 Lausanne, Switzerland

Abstract:

This study analyses the uncertainty induced by the use of different state-of-the-art climate models on the predictionof climate-change impacts on the runoff regimes of 11 mountainous catchments in the Swiss Alps having currentproportions of glacier cover between 0 and 50%. The climate-change scenarios analysed are the result of 19 regionalclimate model (RCM) runs obtained for the period 2070–2099 based on two different greenhouse-gas emissionscenarios (the A2 and B2 scenarios defined by the Intergovernmental Panel on Climate Change) and on threedifferent coupled atmosphere-ocean general circulation models (AOGCMs), namely HadCM3, ECHAM4/OPYC3 andARPEGE/OPA. The hydrological response of the study catchments to the climate scenarios is simulated througha conceptual reservoir-based precipitation-runoff transformation model called GSM-SOCONT. For the glacierizedcatchments, the glacier surface corresponding to these future scenarios is updated through a conceptual glacier surfaceevolution model.

The results obtained show that all climate-change scenarios induce, in all catchments, an earlier start of the snowmeltperiod, leading to a shift of the hydrological regimes and of the maximum monthly discharges. The mean annual runoffdecreases significantly in most cases. For the glacierized catchments, the simulated regime modifications are mainlydue to an increase of the mean temperature and the corresponding impacts on the snow accumulation and meltingprocesses. The hydrological regime of the catchments located at lower altitudes is more strongly affected by the changesof the seasonal precipitation. For a given emission scenario, the simulated regime modifications of all catchments arehighly variable for the different RCM runs. This variability is induced by the driving AOGCM, but also in large partby the inter-RCM variability. The differences between the different RCM runs are so important that the predictedclimate-change impacts for the two emission scenarios A2 and B2 are overlapping. Copyright 2006 John Wiley &Sons, Ltd.

KEY WORDS climate change; regional climate models; hydrological modelling; snowmelt modelling; glacier hydrology;Alps

INTRODUCTION

The Alps are a key element of the hydrological regime of the Swiss river network and they are the source ofmany of Europe’s major river systems. The specific mean annual discharge in mountainous areas is generallyhigher than in catchments located at lower altitudes in the same climatic region. This results essentiallyfrom higher precipitation amounts induced by orographic effects, but also from low evapotranspiration ratesinduced in large part by the relatively low mean temperatures. Additionally, the hydrological regime of suchenvironments is strongly influenced by water accumulation in the form of snow and ice and the correspondingmelt processes, resulting in a pronounced annual cycle of the discharge. Therefore, a modification of theprevalent climate, and especially of the temperature, can considerably affect the hydrological regime andinduce important impacts on the water management (e.g. Burlando et al., 2002; Jasper et al., 2004; Schaefli,2005).

* Correspondence to: Benoıt Hingray, Hydram-ISTE, Station 2, EPFL, 1015 Lausanne, Switzerland. E-mail: benoit.hingray@epfl.ch

Received 18 May 2005Copyright 2006 John Wiley & Sons, Ltd. Accepted 17 November 2005

2092 P. HORTON ET AL.

The quantification of potential climate-change impacts on a water resources system is conditioned by theavailability of local or regional climate-change predictions of different key meteorological variables, such asprecipitation and temperature. Currently available climate-change projections are mainly based on the resultsof coupled atmosphere–ocean general circulation models (AOGCMs) or on the results of regional climatemodels (RCMs) driven by outputs of the former. RCMs have a higher spatial resolution than AOGCMs andare usually expected to describe the regional-scale climate better.

Regional climate-change projections based on such climate model outputs are, however, highly uncertain,mainly due to the unknown future greenhouse-gas emissions, but also due to the highly simplified representa-tion of reality encoded in these models. As a consequence, different AOGCM experiments, or different RCMexperiments driven by the same AOGCM outputs, generally yield different climate evolutions for the sameemission scenario (e.g. Arnell and Hulme, 2000; Frei et al., 2003; Raisanen et al., 2004). The uncertaintyintroduced by the RCM is generally considered to be substantially smaller than the one inherited by the drivingAOGCM (Jenkins and Lowe, 2003). An analysis of the data of the EU project PRUDENCE (Prediction ofregional scenarios and uncertainties for defining European climate change risks and effects; Christensen et al.,2002) suggests, however, that RCM inter-model variability cannot be neglected (Hingray et al., 2006a).

Different studies have assessed climate-change impacts on hydrological processes in the Swiss Alps (e.g.Braun et al., 2000; Etchevers et al., 2002; Jasper et al., 2004; Zierl and Bugmann, 2005). All of these studiesshow that the temperature increase strongly affects the temporal evolution of the snowpack and, accordingly,the runoff regimes of the catchments studied. Few studies have analysed different greenhouse-gas emissionscenarios or have taken into account the prediction uncertainty due to the climate model used. For alpinecatchments without glaciers, Jasper et al. (2004) and Zierl and Bugmann (2005) show, however, that thesetwo sources of climate-change prediction uncertainty lead to a wide range of possible future states.

This study focuses on the assessment of climate-change impacts on the discharge regime of 11 catchmentsin the Swiss Alps, considering the climate predictions of multiple state-of-the-art regional climate models.Some of the catchments selected have glacier-covered areas, decreases of which are expected to enhancethe climate-change-induced modifications of the hydrological regime (Braun et al., 2000; Schaefli, 2005).For each catchment, the climate-change impact analysis is conducted through the simulation for an observedcontrol period (1961–1990) and for a future period (2070–2099) having a variety of modified (predicted)climates. These climate predictions were derived from a suite of RCM experiments undertaken within theframework of the EU PRUDENCE project (Christensen et al., 2002), for the two greenhouse-gas emissionscenarios A2 and B2, as defined by the Special Report on Emission Scenarios (SRES; Nakicenovic and Swart,2000), of the Intergovernmental Panel on Climate Change.

A description of the case studies, the corresponding datasets, and the models used, both for the dischargesimulation and for the production of small-scale meteorological time-series scenarios, is presented next. Thenthe climate-change impact prediction results are discussed, followed by the main conclusions of this study.

CASE STUDIES: DATA AND MODELS

Catchment characteristics

The catchments analysed were selected to represent the seven different hydro-climatic zones of the SwissAlps (Laternser and Schneebeli, 2002) and the different mountainous hydrological regimes occurring in thesezones (Figure 1). The choice of catchments was also conditioned by data availability. Coincident series ofobserved discharge and meteorological data are necessary for model calibration and validation, and observedmeteorological data are required for the simulation of the 30-year reference (control) period (1961–1990).

The catchments selected have different proportions of glacier cover (between 0 and 50%) and altitude ranges(Table I). Their total areas vary between 39 and 185 km2, and their mean altitudes vary from 1340 m a.s.l.,for the Minster catchment, to 2940 m a.s.l., for the Drance de Bagnes catchment. The catchments representa large range of reference hydrological regimes, as defined by Aschwanden and Weingartner (1985) for the

Copyright 2006 John Wiley & Sons, Ltd. Hydrol. Process. 20, 2091–2109 (2006)

CLIMATE-CHANGE IMPACTS ON ALPINE DISCHARGE REGIME 2093

Figure 1. Location of the 11 case study catchments in the seven hydro-climatic regions of the Swiss Alps (SwissTopo, 1997)

Table I. Characteristics of the case study catchments and corresponding discharge regimes according to the regime definitionsgiven by Aschwanden and Weingartner (1985) for the Swiss Alps (proportion of glacier cover, annual precipitation and

hydrological regime correspond to the control period)

Catchment Area(km2)

Glacier(%)

Meanaltitude

(m a.s.l.)

Altituderange

(m a.s.l.)

Lapserate

(°C/100 m)

Annualprecipitation

(mm)

Hydrologicalregimea

Drance de Bagnes 166Ð6 39Ð0 2940 1960–4310 �0Ð50 1620 ‘a-glacial’Saaser Vispa 65Ð2 33Ð2 2840 1710–4190 �0Ð46 1627 ‘b-glacial’Lonza 77Ð8 32Ð8 2600 1520–3900 �0Ð57 2187 ‘a-glacial’Rhone at Gletsch 38Ð9 50Ð0 2710 1760–3610 �0Ð55 2262 ‘a-glacial’Weisse Lutschine 164Ð0 16Ð4 2150 650–4170 �0Ð60 1774 ‘a-glacio-nival’Minster 59Ð2 — 1340 880–2290 �0Ð54 2197 ‘nival de transition’Tamina 147Ð0 1Ð4 1810 550–3220 �0Ð61 1554 ‘nival alpin’Vorderrhein 158Ð0 2Ð8 2150 750–3310 �0Ð58 1748 ‘nivo-glacial’Dischmabach 43Ð3 2Ð6 2360 1670–3130 �0Ð56 1460 ‘b-glacio-nival’Rosegbach 66Ð5 27Ð2 2710 1760–4010 �0Ð43 1412 ‘a-glacial’Verzasca 185Ð2 — 1650 480–2880 �0Ð57 2175 ‘nivo-pluvial meridional’

a Aschwanden and Weingartner (1985).

Swiss Alps (Table I). These were determined on the basis of dimensionless Parde coefficients (Parde, 1933),which relate the mean monthly discharge of a given month to the total annual discharge.

All catchments selected (except the Verzasca catchment) show typical discharge patterns characterized bya single peak occurring between May and August. This peak is due to snowmelt-induced high flows. For

Copyright 2006 John Wiley & Sons, Ltd. Hydrol. Process. 20, 2091–2109 (2006)

2094 P. HORTON ET AL.

catchments with glaciers, the peak is sustained by usually significant ice-melt-induced flows occurring later inthe melt season. The amplitude of the monthly flows is high between the low flows during the winter seasonand the high flows in late spring or summer. The amplitude, as well as the occurrence date of the dischargepeak, depends strongly on the elevation range covered by the catchment (Aschwanden and Weingartner, 1985).The higher the mean elevation of the catchment is, the later the peak occurs and the higher the amplitude is (seethe regimes simulated for the control period, Figure 3). Meanwhile, the Verzasca River has a typical regimeof the southern Alps, with a second discharge peak occurring in the autumn, due to the heavy precipitationevents that are usually observed in this period.

System models

The simulation of catchment behaviour for the different time periods was carried out using a hydrologicalmodel, for the precipitation–runoff transformation, and a land-cover evolution model. The only land-coverchange that is explicitly accounted for is the modification of the glacier-covered surface. Land-use changesrelated to the vegetation could presumably have a significant impact on future hydrological regimes, butthey are not considered in this study. Zierl and Bugmann (2005), however, found that, for five Swiss alpinecatchments, such land-use changes only have a small impact on discharges under similar climate-changescenarios to the ones used here.

Hydrological model. The hydrological discharge simulation is carried out at a daily time-step through aconceptual reservoir-based model called GSM-SOCONT (Schaefli et al., 2005). The model has two levelsof discretization, with the ice-covered part of the catchment first separated from the not ice-covered part,and then both parts subdivided into 10 elevation bands of same surface. Each of the resulting spatial unitsis characterized by its surface and its mean elevation, and is assumed to have a homogeneous hydrologicalbehaviour.

For each spatial unit, the meteorological data series are computed from data observed at neighbouringmeteorological stations. For temperature, regional altitudinal gradients are estimated based on the observedtemperature series (Table I). Based on these gradients, the temperature time-series for a given spatial unit isinterpolated according to its mean elevation. To account for the enhancement of precipitation with altitude, amesoscale annual precipitation gradient of 80 mm per 100 m, estimated by Kirchhofer and Sevruk (1991) forthe entire Swiss alpine region, is applied. The resulting area-average daily precipitation amounts are correctedby a constant multiplicative factor in order to respect the mean annual precipitation amounts estimated foreach catchment by Schadler and Bigler (2002) based on a long-term water balance analysis.

Based on the meteorological time-series, the temporal evolutions of the snowpack and of the glacier meltare computed. The aggregation state of precipitation (liquid, solid or mixed) is determined through a fuzzytemperature threshold approach (Klok et al., 2001). Snow and ice melt are simulated through a simple degree-day approach (Rango and Martinec, 1995). Ice melt occurs only when the glacier surface of the spatial unitbeing considered is free of snow.

For each hydrological unit, a reservoir-based modelling approach is used to simulate the rainfall and melt-water–runoff transformation. For ice-covered spatial units, this transformation is completed through two linearreservoirs, one for ice melt and one for the equivalent rainfall, defined as the sum of snowmelt and rainfall.For spatial units that have no ice, the equivalent rainfall is transformed into runoff through the conceptualmodel SOCONT (Consuegra and Vez, 1996). It is composed of two reservoirs, i.e. a linear reservoir for theslow contribution (accounting for soil infiltration processes) and a non-linear reservoir for direct, or quick,runoff. Actual evapotranspiration is estimated as a function of the potential evapotranspiration (PET) and thefilling rate of the slow component reservoir.

The model has seven parameters to calibrate (four for catchments without a glacier): two degree-day factorsfor the ice and snowmelt computation, four time constants of the reservoirs, and the maximum storage capacityof the slow reservoir.

Copyright 2006 John Wiley & Sons, Ltd. Hydrol. Process. 20, 2091–2109 (2006)

CLIMATE-CHANGE IMPACTS ON ALPINE DISCHARGE REGIME 2095

The calibration procedure is based on the multicriteria calibration procedure presented by Schaefli et al.(2005). It aims to minimize the Nash–Sutcliffe model efficiency (NSE) coefficient (Equation (1); Nash andSutcliffe, 1970) and the bias, estimated between observed and simulated daily discharges. Special attention isalso paid to the reproduction of the observed mean annual and the mean monthly discharges over a calibrationperiod. Calibration and validation were undertaken for two consecutive 10-year periods (1971–1980 and1981–1990 respectively, for the majority of the catchments).

The calibrated model performs well for all catchments for both the calibration and the validation period.NSE coefficients of 0Ð85 or higher were obtained for the modelled daily discharge for most catchments. Onlythe two ice-free catchments of Minster and Verzasca had lower NSE values, both being close to 0Ð75. Forboth calibration and validation periods, NSE values for monthly discharges exceeded 0Ð9 for all catchments.

NSE D 1 �

n∑

iD1

�Qo,i � Qs,i�2

n∑

iD1

�Qo,i � Qo�2

�1�

where Qo,i and Qs,i are the observed and simulated discharges of the ith time step, Qo is the mean observeddischarge over the calibration or validation period, and n is the number of time steps simulated. An NSEcoefficient value of unity indicates a perfect model.

Glacier surface evolution model. The glacier area was considered to be constant during the control simulationperiod. For this period, the glacier area was obtained from vector-based topographic maps (SwissTopo, 1997).It corresponds approximately to the middle of the control period, the exact year depending on the date of thefield surveys. The glacier area was also assumed to be constant during the future simulation period, but usinga different (simulated) constant value that accounts for the glacier retreat.

The response of a glacier system to a climate modification could be simulated through a dynamical ice-flowmodel (e.g. Oerlemans et al., 1998). Such a physical modelling approach is, however, highly data intenseand can only be applied to well-investigated glacier systems. In this study, the glacier surface for future timeperiods is updated through a conceptual modelling approach based on the so-called accumulation area ratio(AAR) method.

The AAR is defined as the ratio between the accumulation area of the glacier and its entire surface(Anonymous, 1969). The AAR method assumes that the steady-state accumulation area of a glacier occupiessome fixed proportion of the total glacier area (e.g. Meier and Post, 1962; Gross et al., 1976). The concept isclassically used in the palaeoclimatic reconstruction of glacier surfaces (e.g. Porter, 1975; Benn and Lehmkuhl,2000). For a given glacier, these studies assume the existence of a constant steady-state AAR value that canbe estimated and used to predict the steady-state glacier area for any period according to

Aice,s D Aacc,s

AARs�2�

where Aice,s�km2� is the total steady-state glacier area, AARs is the steady-state AAR and Aacc,s�km2� is theestimated steady-state accumulation area for the climatic conditions of the period considered.

A glacier is defined as being in a steady state if its dimensions remain constant over the period considered.This is a theoretical concept that is rarely or never encountered in practice (Paterson, 1994) and the steady-state AARs value is unknown a priori. In this study, the mean annual AAR value AARm was used, on theassumption it is a good estimator of the relationship between the climate and the glacier area if estimated overa long enough time period. It was further assumed that this relationship remains constant for future periods.In other words, the mean annual AAR value for the present situation was assumed to remain constant for thefuture period.

Copyright 2006 John Wiley & Sons, Ltd. Hydrol. Process. 20, 2091–2109 (2006)

2096 P. HORTON ET AL.

The annual AAR value can be estimated based on the hydrological model outputs for the control period:for a given hydrological year (starting on the 1 October), the AAR is computed as the sum of spatial unitsthat experience net snow accumulation divided by the total (known) glacier surface.

Based on AARm,cont, the estimated AARm value for the 1961–1990 control period, the future glacier surfaceAice,fut �km2� can be simulated according to

Aice,fut D Aacc,fut

AARm,cont�3�

where Aacc,fut �km2� is the mean annual accumulation area of the future period as simulated by the hydrologicalmodel under future climate conditions (the annual accumulation area is estimated as the sum of spatial unitsthat experience net snow accumulation for a given hydrological year under the future climate conditions). Fora further discussion of the glacier surface evolution model, refer to Schaefli (2005).

Data

Observed periods. The glacio-hydrological model needs three input time-series, namely daily precipitation,daily PET and mean daily temperature. For all catchments, the precipitation and temperature data are obtainedfrom nearby meteorological stations of the Swiss Meteorological Institute. The PET data are derived fromthe monthly PET series estimated by New et al. (2000) for the 1961–1990 control period, according to thePenman–Monteith version given by Burman and Pochop (1994). The series of mean daily discharges formodel calibration and validation are provided by the Swiss Federal Office for Water and Geology.

The spatial discretization of the catchment is completed based on a 25 m resolution digital elevation model(SwissTopo, 1995) and on 1 : 25 000 digital, vector-based, topographic maps (SwissTopo, 1997).

Future periods. Each RCM experiment consists of a simulation for the period 1961–1990 (control run)and a simulation for the period 2070–2099 (future run). For each RCM experiment, the boundary conditionswere obtained from one of the three AOGCMs used in PRUDENCE (Christensen et al., 2002): ARPEGE/OPA(Royer et al., 2002), HadCM3 (Gordon et al., 2000; Pope et al., 2000) and ECHAM4/OPYC3 (Roeckner et al.,1999). In the case of HadCM3, a global model of the atmosphere alone (HadAM3H) was used between theglobal coupled model and the RCMs. The use of this intermediate model resulted in a much better simulationof the present-day climate (Hulme et al., 2002). The AOGCM experiments were completed for the two SRESemission scenarios A2 and B2 (Nakicenovic and Swart, 2000). The A2 scenario assumes a heterogeneousworld where regional cultural identities are strengthened, with a high population growth but with less concernfor rapid economic development (Nakicenovic, 2000). The B2 scenario also assumes a heterogeneous world,but with local solutions to enhance economic, social and environmental sustainability. Compared with A2,the technological progress in B2 is less rapid, and the greenhouse-gas emissions are expected to be smaller(Nakicenovic, 2000). Hence, a smaller global temperature increase is associated with B2 than A2.

In this study, nine RCMs are included (Table II). One of these, ARPEGE, is not a regional model; rather, it isa global atmospheric model with variable horizontal resolution, from 50 km in the centre of the MediterraneanSea to 450 km in the southern Pacific Ocean (Gibelin and Deque, 2003). In PRUDENCE, the boundaryconditions for this model (sea-surface temperatures, sea ice) were obtained from ARPEGE/OPA and HadCM3(Deque, personal communication). For the 11 catchments analysed in this study, the horizontal resolution ofARPEGE is comparable to the other RCMs and, for the purposes of the study, ARPEGE was considered asan RCM. Outputs from 19 RCM experiments are available in total: 12 for the A2 scenario and seven for B2.A synthesis of these experiments is given in Table III.

The RCMs are reported to reproduce well the regional features of meteorological surface variables suchas precipitation and temperature (e.g. Frei et al., 2003). Their reliability, however, is not the same for eachindividual RCM grid cell, particularly in mountainous regions. As a result, instead of considering only thegrid cell containing a single given case study, it was decided to use data from several grid cells. In this study,

Copyright 2006 John Wiley & Sons, Ltd. Hydrol. Process. 20, 2091–2109 (2006)

CLIMATE-CHANGE IMPACTS ON ALPINE DISCHARGE REGIME 2097

Table II. The three AOGCMs and the nine RCMs used in the PRUDENCE project (Christensen et al., 2002) and thecorresponding institutions

Institution Reference

AOGCMARPEGE/OPA Centre National de Recherches Meteorologiques, Toulouse, France Royer et al. (2002)HadCM3 Hadley Centre for Climate Prediction and Research, Bracknell, UK Gordon et al. (2000), Pope

et al. (2000)ECHAM4/OPYC3 Max-Planck-Institut fur Meteorologie, Hamburg, Germany Roeckner et al. (1999)

RCMARPEGE Centre National de Recherches Meteorologiques, Toulouse, France Gibelin and Deque (2003)HIRHAM Danish Meteorological Institute, Copenhagen, Denmark Christensen et al. (2001)CHRM Institute for Atmospheric and Climate Science, Zurich, Switzerland Vidale et al. (2003)CLM Institute for Coastal Research, Geesthacht, Germany Doms and Schattler (1999)HadRM3H Hadley Centre for Climate Prediction and Research, UK Hulme et al. (2002)RegCM International Centre for Theoretical Physics, Trieste, Italy Giorgi et al. (1993a,b)REMO Max-Planck-Institut fur Meteorologie, Hamburg, Germany Jacob (2001)RCAO Swedish Meteorological and Hydrological Institute, Norrkoping, Sweden Raisanen et al. (2004)PROMES Universidad Complutense de Madrid, Toledo, Spain Arribas et al. (2003)

Table III. RCM experiments according to the driving AOGCM and to theemission scenarios (all RCM experiments were conducted in the framework

of PRUDENCE)

Scenario name AOGCM name RCM name No.

A2 HADCM3 CHRM 1HADCM3 CLM 2HADCM3 HadRM3H 3HADCM3 HIRHAM 4HADCM3 PROMES 5HADCM3 RCAO(H) 6HADCM3 RegCM 7HADCM3 REMO 8HADCM3 ARPEGE 9ARPEGE/OPA ARPEGE 10ECHAM4/OPYC3 HIRHAM 11ECHAM4/OPYC3 RCAO(E) 12

B2 HADCM3 HadRM3H 13HADCM3 PROMES 14HADCM3 RCAO(H) 15HADCM3 ARPEGE 16ARPEGE/OPA ARPEGE 17ECHAM4/OPYC3 HIRHAM 18ECHAM4/OPYC3 RCAO(E) 19

for each catchment analysed, the regional changes predicted by a given RCM are thus averaged over ninegrid cells encompassing the catchment of interest. The uncertainty in the regional climate-change scenariosassociated with the number of aggregation grids (9, 16 or 25) has been estimated for the Rhone catchment atGletsch. It is much smaller than the uncertainty resulting from the inter-model differences and, therefore, isneglected.

The local-scale time-series of temperature and precipitation for a given future scenario are obtained byperturbing the observed time-series for a control period, based on the regional climate changes predicted by

Copyright 2006 John Wiley & Sons, Ltd. Hydrol. Process. 20, 2091–2109 (2006)

2098 P. HORTON ET AL.

the climate models. The method used for temperature perturbation is the one presented by Shabalova et al.(2003). It preserves the changes in mean and variability given by the climate-change scenario for each season.For a given RCM experiment, the climate-change scenario is defined in terms of the absolute changes of themean temperature XMTs between the control and the future runs and in terms of the relative changes of thestandard deviation of daily temperature XSDTs (defined as the difference between the values obtained fromthe RCM output for the future period and the control period, divided by the value for the control period),where s refers to the season (s D 1: December–February (DJF); s D 2: March–May; s D 3: June–August(JJA); s D 4: September–November). The future scenario temperature Tscen is calculated according to

Tscen,s�t� D [Tobs,s�t� � MTobs,s]�XSDTs C 1� C MTobs,s C XMTs �4�

where Tscen,s�t� (°C) is the local-scale scenario temperature on day t in the season s, Tobs,s�t� is the localtemperature on day t observed during the control period in the season s, and MTobs,s is the correspondingmean daily temperature.

Note that the standard deviations of daily temperatures SDT are not available from PRUDENCE experi-ments. Instead, PRUDENCE has produced grids of 90-day temperature variances for each season. The changein the daily standard deviations was thus set equal to the change in the 90-day standard deviations producedin PRUDENCE. It can be easily shown that, in this case, the perturbation of the observed daily series basedon Equation (4) gives a future daily series that preserves the relative change in the 90-day standard devi-ations. Note also that these 90-day statistics have been estimated in PRUDENCE directly from the RCMoutput series. A significant temperature trend has been simulated for the 30-year period of 2070–2099 thatcould potentially influence the variance estimation. PRUDENCE data were thus preprocessed according tothe method presented in Hingray et al. (2006b).

For precipitation, the use of a comparable perturbation method is not possible because this would generatenegative rainfall values. The future local-scale precipitation time-series are estimated based on the simpleproportional relationship

Pscen,s�t� D Pobs,s�t�MPfut,s

MPcont,s�5�

where Pscen,s�t� (mm) is the local-scale scenario precipitation on day t of the season s (s D 1–4), Pobs,s�t� isthe precipitation observed on day t during the control period, and MPfut,s and MPcont,s are respectively themean seasonal precipitations for the future and the control run for a given RCM experiment.

Note that, for the future scenarios, the PET is estimated as a function of the perturbed temperature based onthe observed linear relationship for the control period, assuming that this relationship remains constant in thefuture. In fact, Ekstrom et al. (2006) have shown that PET values calculated from the outputs of HadRM3Husing a Pennman–Monteith method reach unrealistically high values for both the control and the future period.

RESULTS

Climate changes

The ARPEGE/OPA, HadCM3 and ECHAM4/OPYC3 AOGCMs predict global-mean warmings for scenarioB2 of C2Ð4 °C, C2Ð4 °C and C2Ð8 °C respectively, and of C3Ð0 °C, C3Ð3 °C and C3Ð6 °C for scenario A2(Roeckner et al., 1999; Gordon et al., 2000; Gibelin and Deque, 2003). The regional mean annual temperatureincreases predicted by the 19 PRUDENCE RCM experiments are similar for all catchments and are higherthan the global-mean warming (Table IV), at around C3Ð0 °C for B2 and C4Ð0 °C for A2. This enhancedregional warming is consistent with the temperature increases that have already been observed in the SwissAlps over the 20th century (e.g. Beniston et al., 1994; Weber et al., 1997). The predicted regional warmingfor the summer is higher than for the other seasons (e.g. the maximum summer warming predicted for A2for the Rhone catchment is as large as C8Ð1 °C, being around C5Ð2 to C6 °C for the other seasons).

Copyright 2006 John Wiley & Sons, Ltd. Hydrol. Process. 20, 2091–2109 (2006)

CLIMATE-CHANGE IMPACTS ON ALPINE DISCHARGE REGIME 2099

Tabl

eIV

.M

inim

um,m

edia

nan

dm

axim

umre

gion

alcl

imat

ech

ange

spr

edic

ted

byth

ePR

UD

EN

CE

RC

Mex

peri

men

tsfo

r20

70–

2099

for

B2

scen

ario

(sev

enex

peri

men

ts)

and

A2

scen

ario

(12

expe

rim

ents

):a)

annu

alch

ange

sof

mea

nte

mpe

ratu

rean

dpr

ecip

itatio

n,b)

seas

onal

chan

ges

ofm

ean

prec

ipita

tion

amou

nts

for

win

ter

(DJF

)an

dsu

mm

er(J

JA)

Cat

chm

ent

Ann

ual

chan

gein

tem

pera

ture

(°C

)A

nnua

lch

ange

inpr

ecip

itatio

n(%

)Se

ason

alch

ange

inpr

ecip

itatio

n(%

)

DJF

JJA

B2

scen

ario

A2

scen

ario

B2

scen

ario

A2

scen

ario

B2

scen

ario

A2

scen

ario

B2

scen

ario

A2

scen

ario

Min

.M

ed.

Max

.M

inM

edM

axM

in.

Med

.M

ax.

Min

.M

ed.

Max

.M

in.

Med

.M

ax.

Min

.M

ed.

Max

.M

in.

Med

.M

ax.

Min

.M

ed.

Max

.

Dra

nce

deB

agne

s2Ð5

3Ð04Ð7

3Ð34Ð2

6Ð1�1

1�6

0�2

1�8

07

1023

014

28�4

6�2

3�3

�52

�27

�10

Saas

erV

ispa

2Ð53Ð0

4Ð63Ð3

4Ð06Ð1

�12

�60

�22

�8�1

312

23�1

1528

�43

�26

�2�4

8�2

7�9

Lon

za2Ð5

3Ð04Ð6

3Ð34Ð0

6Ð1�9

�51

�19

�62

312

23�1

1528

�43

�26

�2�4

8�2

7�9

Rho

neat

Gle

tsch

2Ð43Ð0

4Ð63Ð2

3Ð96Ð2

�10

�40

�20

�62

�115

20�6

1628

�39

�32

�1�5

1�2

5�9

Wei

sse

Lut

schi

ne2Ð5

3Ð14Ð7

3Ð23Ð9

6Ð2�1

3�6

�1�2

5�7

�31

1518

�514

26�3

9�3

1�2

�49

�27

�12

Min

ster

2Ð43Ð1

4Ð73Ð2

3Ð86Ð3

�13

�50

�25

�5�2

016

22�4

1424

�37

�28

0�5

5�2

2�8

Tam

ina

2Ð43Ð0

4Ð63Ð2

3Ð96Ð2

�13

�81

�24

�7�1

215

20�1

2127

�39

�31

1�5

0�2

2�6

Vor

derr

hein

2Ð43Ð0

4Ð63Ð2

3Ð96Ð1

�12

�50

�22

�60

015

21�4

1928

�40

�32

0�5

2�2

3�7

Dis

chm

abac

h2Ð4

2Ð94Ð7

3Ð24Ð0

6Ð2�1

3�9

2�2

4�7

03

1521

122

29�3

7�2

63

�45

�21

�3R

oseg

bach

2Ð43Ð0

4Ð63Ð2

4Ð06Ð1

�15

�82

�26

�11

�28

1724

1127

37�4

2�3

00

�50

�23

�5V

erza

sca

2Ð43Ð1

4Ð53Ð2

4Ð06Ð1

�16

�81

�26

�11

�23

1329

624

30�4

6�3

4�1

�54

�26

�6M

edia

n2Ð4

3Ð04Ð6

3Ð24Ð0

6Ð1�1

3�6

0�2

4�7

�13

1522

�116

28�4

0�3

0�1

�50

�25

�8

Copyright 2006 John Wiley & Sons, Ltd. Hydrol. Process. 20, 2091–2109 (2006)

2100 P. HORTON ET AL.

0123456789

10

0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0

Global ∆T [°C]

XM

T [°

C]

−60

−40

−20

0

20

40

60

0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0

Global ∆T [°C]

XM

P [%

]

0123456789

10

0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0

Global ∆T [°C]

XM

T [°

C]

Winter (DJF) Summer (JJA)

−60

−40

−20

0

20

40

60

XM

P [%

]

0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0

Global ∆T [°C]

Winter (DJF) Summer (JJA)

HADCM3-A2

HADCM3-B2

ARPEGE/OPA-A2

ARPEGE/OPA-B2

ECHAM4/OPYC3-A2

ECHAM4/OPYC3-B2

Figure 2. Regional seasonal changes of precipitation (XMP) and temperature (XMT) as a function of the annual global mean warming Tas predicted by the PRUDENCE RCM experiments for the Rhone catchment for the period 2070–2099. Top: regional temperature changes;bottom: regional precipitation changes. Each symbol corresponds to one RCM experiment; large symbols identify the RCAO experiments

There is a clear gradation of the projected regional warming between the B2 scenario and the A2 scenario:every percentile of the estimated temperature-change distribution based on the A2 RCM experiments hasa higher value than that estimated for B2. The spread of regional changes obtained from the differentAOGCM/RCM experiments for a given emission scenario is, however, considerable (Table IV and Figure 2).This variability is induced by the driving AOGCMs and by the RCMs. Considering a given RCM, the drivingAOGCM induces a variability of the same order of magnitude as the one induced by the emission scenario(Figure 2). Considering a given AOGCM, the inter-RCM variability represents an important additional sourceof prediction uncertainty. Whereas for each AOGCM the predicted global-mean warming is clearly distinctfor the B2 and A2 scenarios, the ranges of regional changes predicted by the different RCMs for a givenAOGCM are often overlapping.

Except for a few experiments and catchments, a decrease of the annual precipitation is predicted (Table IV);there is no clear gradation between the projected changes of the annual precipitation for B2 and A2. MostRCMs predict an increase of winter precipitation and a decrease of summer precipitation for both scenariosand for all catchments (Table IV and Figure 2). For spring, the predictions of the different RCMs do nothave a clear tendency towards an increase or a decrease (for all catchments and both scenarios). For autumn,precipitation is predicted to decrease for the A2 scenario, but no tendency is detectable for B2. The differentlevels of prediction variability highlighted for temperature changes are also observed for precipitation changes(Figure 2).

Glacier surface decrease

The predicted decrease of glacier surfaces between the control and the future period is considerable(Table V), with almost no ice-covered areas left in the future. For the highest catchment, the Drance catchment,

Copyright 2006 John Wiley & Sons, Ltd. Hydrol. Process. 20, 2091–2109 (2006)

CLIMATE-CHANGE IMPACTS ON ALPINE DISCHARGE REGIME 2101

Table V. Predicted change of the annual runoff and predicted glaciation rate for the B2 and A2 scenarios (2070–2099): theminimum, median and maximum values of the 7 and 12 simulations based on the PRUDENCE RCM experiments available

for the scenario B2 and A2

Catchment Discharge modification (%) Glaciation rate (% of total catchment area)

B2 scenario A2 scenario B2 scenario A2 scenario

Min. Med. Max. Min. Med. Max. Min. Med. Max. Min. Med. Max.

Drance de Bagnes �36 �26 �19 �49 �30 �24 0.8 1.8 6.7 0.1 1.0 1.7Saaser Vispa �21 �13 �5 �34 �17 �9 0.6 3.7 7.1 0.0 1.4 3.4Lonza �19 �14 �7 �32 �16 �8 1.0 2.6 5.1 0.1 1.6 2.3Rhone at Gletsch �28 �20 �15 �40 �23 �16 0.0 0.3 6.7 0.0 0.0 0.0Weisse Lutschine �33 �23 �12 �47 �25 �20 0.5 2.2 6.2 0.1 0.9 2.6Minster �26 �14 �6 �42 �16 �11 — — — — — —Tamina �24 �16 �5 �39 �16 �10 0.0 0.0 0.1 0.0 0.0 0.0Vorderrhein �22 �12 �6 �35 �14 �8 0.0 0.0 0.0 0.0 0.0 0.0Dischmabach �24 �17 �4 �39 �16 �10 0.0 0.0 0.0 0.0 0.0 0.0Rosegbach �28 �18 �6 �41 �22 �13 0.0 1.4 2.3 0.0 0.3 1.3Verzasca �27 �16 �3 �40 �19 �10 — — — — — —Median �26 �16 �6 �40 �17 �10 — — — — — —

the median future predicted glacier-covered area corresponds to 2% of the total catchment area for B2 (1%for A2), compared with the 39% glacier cover during the control period. This result is primarily due to thelarge predicted temperature increases. The periods during which precipitation falls in the form of snow arealso predicted to be shorter in the future. Despite the expected additional winter precipitation, this leads to lesssnow accumulation on the glacier surface. Owing to the predicted high summer temperatures, the accumulatedwinter snow disappears during the following melt season at much higher altitudes than for the control period.The accumulation area of glaciers, thus, decreases considerably, which leads to significant glacier retreat. Forthe Rhone catchment, the predicted median increase of mean temperature of C3Ð9 °C for A2 corresponds toan upward shift of around 710 m of the 0 °C isotherm. This shift cannot be directly translated into a reductionof the glacier surface because there is no linear relationship between the glacier surface and the 0 °C isothermelevation. Nevertheless, it is worth noting that the mean altitude of the glacierized area for the control periodis around 2940 m a.s.l. and the highest point of the glacier is only about 670 m higher, at 3610 m a.s.l. Thesimulated complete disappearance of the Rhone glacier is in line with the results obtained by Wallinga andvan de Wal (1998) with a physically based one-dimensional flow model. They concluded that the glacierwould vanish by 2100 under a 4Ð4 °C temperature increase (without precipitation modification).

Annual discharge

The simulated annual discharges show a significant decrease compared with the control period, for allcatchments and for all climate experiments (Table V). The median decrease ranges between �26% (Drancede Bagnes) and �12% (Vorderrhein) under the B2 scenario, and between �30% (Drance de Bagnes) and�14% (Vorderrhein) under scenario A2.

It should be noted that the simulated mean discharge decrease may be overestimated as a result of thehighly simplified simulation of the glacier retreat. The glacier surface estimate for a future climate scenariois based on the assumption that the glacier has reached a stationary state for the future climate situationconsidered. This will presumably not be the case due to the reaction time of the glaciers (from several yearsto a few decades depending on the glacier size and configuration). The glacier area for the future period thatwould result from a given climate scenario will thus presumably be slightly larger than the one simulatedwith our model. The glacier retreat for the A2 and B2 scenarios, however, is predicted to be so large that thecontribution of glacier melt to the future hydrological regime will be negligible.

Copyright 2006 John Wiley & Sons, Ltd. Hydrol. Process. 20, 2091–2109 (2006)

2102 P. HORTON ET AL.

Table VI. Simulated actual evapotranspiration for the control and the future period for the B2 and A2 scenarios; the valuesare specific values for the ice-free part of the catchment. Note: the area of the ice-free catchment part varies between the

climate experiments (see Tables I and V)

Catchment Evapotranspiration (mm)

Control B2 scenario A2 scenario

Min. Med. Max. Min. Med. Max.

Drance de Bagnes 296 360 377 445 385 416 491Saaser Vispa 243 299 305 333 308 328 353Lonza 269 331 338 375 349 371 408Rhone at Gletsch 294 365 373 426 385 414 455Weisse Lutschine 391 478 502 551 501 533 574Minster 507 593 610 662 614 646 689Tamina 274 333 341 388 342 376 418Vorderrhein 274 323 333 374 331 363 400Dischmabach 255 308 313 349 313 341 378Rosegbach 229 266 276 303 276 298 328Verzasca 380 447 462 511 461 494 551

Median increase — 62 69 114 80 102 144

The considerable decrease of the mean annual discharges has several reasons, the most evident being thedecrease in mean annual precipitation. Another significant part of the discharge reduction is due to the predictedincrease of evapotranspiration. For the currently glacierized catchments, this increase is partly a consequenceof the significant glacier retreat, which results in an increase of the ice-free portion of the catchment whereevaporation is subtracted from precipitation (in the ice-covered areas, evaporation is fed by ice and firn that ispre-existent to the simulation period considered). It is noteworthy that the increase in total evapotranspirationin the simulation is not linked to a potential recolonization by plants of the newly exposed glacier-free surface.It is induced by the increase of areas covered by bare soils and rocks that are also affected by evaporation. Theresults obtained, however, rely on the important assumption that the relationship between evapotranspirationand the hydrological model parameters governing it remains constant in the future. The model parameters arecalibrated for a given land-cover configuration, which for the high mountainous catchments studied includeslarge surfaces comparable to those that would appear through glacier retreat. A significant land-cover changewould, nevertheless, require a recalibration. However, this is not possible in the context of the present study.

Additionally, the substantial temperature increase throughout the seasons enforces the total evapotran-spiration on ice-free areas; accordingly, all catchments show a strong increase of total evapotranspiration(Table VI). This increase is expected to be less important in summers than in winters: even if there isa pronounced temperature increase in summer, summer evapotranspiration will be limited due to reducedprecipitation.

The important variability of annual precipitation changes simulated by the different climate models(Table IV) and the significant variability of simulated changes in evapotranspiration (induced by the variabilityof changes in seasonal temperatures, and thus in PET) explain the different magnitudes of mean annual runoffdecrease. Note that if the maximum predicted discharge decrease is significantly higher for the A2 scenariothan for the B2 scenario, then the median and minimum decreases obtained for both scenarios are quitesimilar. The variability highlighted is mainly due to inter-model differences.

Hydrological regime modifications

All simulations under the future regional climate scenarios show the same significant changes of thehydrological regimes and the same trends of these changes. Summer discharge is found to reduce significantly,

Copyright 2006 John Wiley & Sons, Ltd. Hydrol. Process. 20, 2091–2109 (2006)

CLIMATE-CHANGE IMPACTS ON ALPINE DISCHARGE REGIME 2103

winter discharge increases, and the snowmelt-induced peak is shifted to earlier in the year and decreases inmost cases (Figure 3). Owing to the substantial reduction of the glacierized areas, pure glacial dischargeregimes tend to disappear.

The changes are very similar for catchments having the same present regime, independently of theirgeographical situation. A gradation of changes is observed between the B2 and the A2 scenario, and thechanges predicted for B2 tend to be enhanced for A2. The median simulated shift of the maximum monthlyor semi-monthly discharge is, for example, around half a month for B2 and an entire month for A2. Theresults are, however, highly variable for the different RCM experiments.

For each future climate experiment, an attempt was made to identify the corresponding discharge regime onthe basis of the official Swiss classification (Aschwanden and Weingartner, 1985). However, future regimesappear to differ from the regimes currently observed in Switzerland, and for many catchments the maindifference compared with the present regime types was the appearance of a small secondary peak in autumnthat does not exist in present intra-alpine regimes, but only in south-alpine regimes.

Figure 3 illustrates the simulated regime modifications for the Rhone catchment (‘a-glacial’ regime), theWeisse Lutschine (‘a-glacio-nival’ regime), the Verzasca (‘nivo-pluvial meridional’ regime) and the Minstercatchment (‘nival of transition’ regime). For both scenarios, the Rhone catchment and the Weisse Lutschinecatchment tend to have a future regime essentially driven by spring snowmelt. The variability of the resultsobtained for the different climate models is, however, high and the corresponding regimes can be quitedifferent. The snowmelt-induced peak occurs, for example, between late May and July. The attenuation ofthis maximum monthly discharge is also highly variable from one scenario to another. The changes obtainedfor the Verzasca catchment are slightly less variable. The shift of snowmelt is about half a month for allexperiments. The rainfall response peak in autumn does not shift in time, but does vary in amount. Thiscatchment is, thus, still expected to exhibit the same ‘nivo-pluvial meridional’ regime, as it does presently.For the Minster catchment, the peak observed for the control run in May–June shifts to April–May under B2and to even earlier for A2. The seasonality of discharges is still significant, but much less pronounced thanfor the control period; for A2, the seasonality tends to disappear. For both emission scenarios, the predictedchanges vary significantly between the RCM experiments. For the two extreme climate experiments (RCMsdriven by ECHAM4/OPYC3), the Minster regime even becomes exclusively driven by rainfall.

It should be noted that the shift of a glacier- or snowmelt-driven hydrological regime to a future regimemore driven by snowmelt and/or rainfall will be associated with a modification of the year-to-year variabilityof the discharges (Figure 4). For high-elevation catchments (catchments with current proportions of glaciercover higher than 15%), the variation coefficient of annual discharges is, for example, predicted to increase byaround 85% for scenario A2 (from 45% to 90% for the Weisse Lutchine catchment and from 115% to 150%for the Saaser Vispa catchment, depending on the RCM). The interannual variability of seasonal dischargesalso changes significantly due to an important modification of the temporary water storage properties of thesemountainous catchments.

The large range of climate-change impacts predicted by the 19 RCM experiments is worthy of furtheranalysis. First, the way this variability acts on the discharge regime prediction has to be highlighted.The sensitivity of the hydrological regimes to temperature and precipitation variability in the PRUDENCERCM experiments was analysed. For catchments with high elevations, the regime modifications are highlyconditioned by the changes of the seasonal temperatures (Figure 5). The simulated maximum monthlydischarge is, for example, highly correlated to the change in mean spring temperature (Figure 6). The mainreason is twofold. First, the seasonal temperature changes are highly correlated. Consequently, the higher thefuture spring temperature is, the higher is the future winter temperature. Accordingly, the period during whichsnow accumulates is shorter and the snowpack at the beginning of the snowmelt period is smaller. Second, thefuture snowmelt peak is still predicted to occur in early summer; but, with higher mean spring temperatures,the volume of snowmelt during spring is higher and, as a consequence, the volume of the snowmelt at thebeginning of the summer is reduced. The high variability of the regional temperature prediction, therefore,has a strong influence on the range of predicted future regimes (Figure 5). For catchments at lower altitudes,

Copyright 2006 John Wiley & Sons, Ltd. Hydrol. Process. 20, 2091–2109 (2006)

2104 P. HORTON ET AL.

J F M A M J J A S O N D0

2

4

6

8

10

12

14

16

18

20M

onth

ly m

ean

runo

ff [m

m/d

]

J F M A M J J A S O N D0

2

4

6

8

10

12

14

16

18

20

Mon

thly

mea

n ru

noff

[mm

/d]

J F M A M J J A S O N D0

2

4

6

8

10

12

Mon

thly

mea

n ru

noff

[mm

/d]

J F M A M J J A S O N D0

2

4

6

8

10

12

Mon

thly

mea

n ru

noff

[mm

/d]

J F M A M J J A S O N D0

2

4

6

8

10

12

Mon

thly

mea

n ru

noff

[mm

/d]

J F M A M J J A S O N D0

2

4

6

8

10

12

Mon

thly

mea

n ru

noff

[mm

/d]

J F M A M J J A S O N D0

1

2

3

4

5

6

7

8

9

10

11

Mon

thly

mea

n ru

noff

[mm

/d]

J F M A M J J A S O N D0

1

2

3

4

5

6

7

8

9

10

11

Mon

thly

mea

n ru

noff

[mm

/d]

ControlMedian of RCMsB2 HadRM3(H)B2 PROMES(H)B2 RCAO(H)

B2 ARPEGE(H)B2 ARPEGE(A)B2 HIRHAM(E)B2 RCAO(E)

ControlMedian of RCMsA2 HadRM3(H)A2 CLM(H)A2 RCAO(H)

A2 REMO(H)A2 HIRHAM(H)A2 CHRM(H)A2 PROMES(H)A2 RegCM(H)

A2 ARPEGE(H)A2 ARPEGE(A)A2 HIRHAM(E)A2 RCAO(E)

Rhone at Gletsch

Weisse Lütschine

Verzasca

Minster

Figure 3. Changes of the hydrological regimes obtained by the 19 PRUDENCE RCM experiments for the period 2070–2099. Left:B2 scenario (seven experiments); right: A2 scenario (12 experiments); regime simulated for the control period (1961–1990) in boldcontinuous line (results obtained for the other catchments are presented in Horton et al. (2005)). This figure is available in colour online at

http://www.interscience.wiley.com/hyp

Copyright 2006 John Wiley & Sons, Ltd. Hydrol. Process. 20, 2091–2109 (2006)

CLIMATE-CHANGE IMPACTS ON ALPINE DISCHARGE REGIME 2105

J F M A M J J A S O N D0

5

10

15

20

25

30M

onth

ly m

ean

runo

ff [m

m/d

]

J F M A M J J A S O N D0

5

10

15

20

25

30

Mon

thly

mea

n ru

noff

[mm

/d]

Annual monthly discharge

Median of the monthly discharge

25th percentile of the monthly discharge

75th percentile of the monthly discharge

Figure 4. Interannual variability of monthly discharges of the Rhone catchment for the control period (left) and the period 2070–2099 (A2emission scenario) with RCAO(E) climate experiment (right); each line corresponds to 1 year of the 30-year simulation period. This figure

is available in colour online at http://www.interscience.wiley.com/hyp

the influence of precipitation modification is more pronounced and the variability of the climate-changeimpacts is mainly due to the large range of predicted regional precipitation changes (Figure 5). For a rainfall-driven regime (the Verzasca catchment in Figure 6), the relationship (previously highlighted for high-elevationcatchments, between the maximum monthly discharge and the increase of spring temperature) is still significantbut much weaker. Conversely, the changes of spring precipitation become a more relevant explanatory variable(Figure 6).

In terms of impact on the hydrological regime, this substantial contribution of the inter-RCM variability tothe total prediction variability can also be pointed out, as seen in the plots of the maximum monthly dischargesversus regional changes in spring temperature and precipitation (Figure 6). The results suggest that the drivingAOGCM may induce more impact prediction variability than the inter-RCM variability for a given AOGCM.For example, for the RCAO regional climate model (Table II; indicated with large symbols in Figures 2and 6), the maximum monthly discharge predicted for the Verzasca catchment for A2 with HadCM3 as adriving model is around twice as high as with ECHAM4/OPYC3 as a driving model. Differences of resultsobtained for two RCMs driven by the same AOGCM under the same emission scenario are often smaller.Nevertheless, the results show that the inter-RCM variability is far from being negligible, e.g. as illustrated inFigure 6 by the spread of the predicted maximum monthly discharges for HadCM3 under scenario A2. Notethat a quantitative evaluation of the uncertainty induced by each of the two climate modelling levels wouldrequire more AOGCM and RCM experiments.

CONCLUSIONS

The climate-change impact study presented on mountainous discharge regimes is based on a classicalsimulation approach using the outputs of a climate model to drive a hydrological model and a glacier-cover evolution model. The analysis focuses on the climate-change prediction uncertainty induced by the useof different climate models. Nineteen experiments were considered. They are based on two greenhouse-gasemission scenarios (SRES A2 and B2), three coupled AOGCMs and nine different RCMs. The catchmentsstudied are representative of the different hydro-climatic regions of the Swiss Alps; accordingly, some generalconclusions can be drawn for the alpine area studied.

Considering the various hydrological regimes and the predictions associated with the different climatemodels, there is a general trend in the simulated climate-change impact for both emission scenarios: the

Copyright 2006 John Wiley & Sons, Ltd. Hydrol. Process. 20, 2091–2109 (2006)

2106 P. HORTON ET AL.

J F M A M J J A S O N D0

2

4

6

8

10

12

14

16

18

20M

onth

ly m

ean

runo

ff [m

m/d

]

J F M A M J J A S O N D0

2

4

6

8

10

12

14

16

18

20

Mon

thly

mea

n ru

noff

[mm

/d]

J F M A M J J A S O N D0

2

4

6

8

10

12

Mon

thly

mea

n ru

noff

[mm

/d]

J F M A M J J A S O N D0

2

4

6

8

10

12

Mon

thly

mea

n ru

noff

[mm

/d]

Rhone at Gletsch

Verzasca

ControlMedian of RCMsA2 HadRM3(H)A2 CLM(H)

A2 RCAO(H)

A2 REMO(H)A2 HIRHAM(H)A2 CHRM(H)A2 PROMES(H)

A2 RegCM(H)

A2 ARPEGE(H)A2 ARPEGE(A)

A2 RCAO(E)A2 HIRHAM(E)

Figure 5. Sensitivity of the hydrological regimes to temperature and precipitation variability of the PRUDENCE RCM experiments (forthe Rhone and the Verzasca catchments, scenario A2, period 2070–2099). Left: regime variability induced by the precipitation variabilityunder median predicted seasonal temperature increase; right: regime variability induced by temperature variability under median predicted

precipitation change. This figure is available in colour online at http://www.interscience.wiley.com/hyp

predicted climate change results in a significant decrease of the total annual discharge and in a shift of themonthly maximum discharge to earlier periods of the year due to the temperature increase and the resultingimpact on the snow melting processes. An increase of the discharge year-to-year variability is also expected.For a given RCM experiment, the hydrological regimes associated with each of the two emission scenariosare clearly distinct. Considering all RCM experiments, the resulting ranges of hydrological regimes for bothemission scenarios are overlapping. Accordingly, an unambiguous prediction of the climate-change impactsfor a given greenhouse-gas emission scenario is not possible. The large prediction variability induced by the19 RCM experiments is partly due to the underlying driving AOGCMs. This source of climate-change impactuncertainty has already been highlighted by Zierl and Bugmann (2005) for alpine catchments. The resultsof the present study clearly show that the use of several RCM experiments with the same driving AOGCMcan result in impact differences that are comparable to the results from the same RCM driven by differentAOGCMs. This leads to the conclusion that the quantitative assessment of hydrological changes based on asmall number of regional climate-change scenarios may yield misleading results.

Copyright 2006 John Wiley & Sons, Ltd. Hydrol. Process. 20, 2091–2109 (2006)

CLIMATE-CHANGE IMPACTS ON ALPINE DISCHARGE REGIME 2107

R2 = 0.829

0

2

4

6

8

10

12

0 1 2 3 4 5 6 7 8XMT [°C]

Q m

ax [m

m/d

] .

R2 = 0.902

0

2

4

6

8

10

12

14

16

18

20

0 1 2 3 4 5 6 7 8XMT [°C]

Q m

ax [m

m/d

] .

R2 = 0.579

0

2

4

6

8

10

12

14

0 1 2 3 4 5 6 7 8XMT [°C]

Q m

ax [m

m/d

] .

R2 = 0.857

0

2

4

6

8

10

12

14

−50 −30 −10 10 30 50XMP [%]

Q m

ax [m

m/d

] .

Rhone at Gletsch Weisse Lütschine

Verzasca Verzasca

HADCM3-A2

HADCM3-B2

ARPEGE/OPA-A2

ARPEGE/OPA-B2

ECHAM4/OPYC3-A2

ECHAM4/OPYC3-B2

Control period R2 Coefficient of determination of the linear regression

Figure 6. Maximum mean monthly discharge Qmax as a function of regional changes of either mean spring (March, April, May) temperature(XMT) or mean spring precipitation (XMP) (large symbols identify the RCAO experiments)

It is also noteworthy that additional prediction uncertainty may arise from the method used for thedevelopment of local-scale scenarios. In this study, a simple perturbation methodology was used, but itwould be interesting to compare the present results with those obtained through local-scale climate modelsor statistical downscaling models that connect the local meteorological variables to synoptic-scale variables(e.g. see Wilby and Wigley (1997) and Xu (1999) for a review of downscaling approaches).

The present analysis does not consider the modelling uncertainties associated with the hydrological model.Schaefli (2005) showed that the uncertainty induced by the parameter uncertainty and the total modelling errorfor a given model structure is far less important than the uncertainty due to the climate predictions. However,as for the climate models, it is not possible to design a unique best hydrological model for a given catchment.The one that has been retained for this study represents just one possible model structure. The analysisof several equivalent hydrological models would complete the current results. The multi-model approach

Copyright 2006 John Wiley & Sons, Ltd. Hydrol. Process. 20, 2091–2109 (2006)

2108 P. HORTON ET AL.

presented by Schaefli (2005) suggests that different hydrological models could increase the uncertainty of thefuture discharge predictions considerably. An important additional source of uncertainty is the estimation ofthe PET and the simulation of the corresponding actual evapotranspiration through the hydrological model.Further research in this area is crucial for a detailed analysis of the climate-change impact uncertainty inherentin the system response modelling.

ACKNOWLEDGEMENTS

Part of this work has been funded by the Swiss Federal Office for Energy. We also wish to thank the SwissFederal Office of Hydrology and Geology, for providing the discharge data, and the national weather service,MeteoSwiss, for providing the meteorological time-series. We are grateful to the two anonymous reviewersand the co-editors for their comments on the manuscript.

REFERENCES

Anonymous. 1969. Mass-balance terms. Journal of Glaciology 8(52): 3–7.Arnell NW, Hulme M. 2000. Implications of climate changes for large dams and their management . Thematic Review II.2 prepared as an

input to the World Commission on Dams, Cape Town.Arribas A, Gallardo C, Gaertner M, Castro M. 2003. Sensitivity of the Iberian Peninsula climate to a land degradation. Climate Dynamics

20: 477–489.Aschwanden H, Weingartner R. 1985. Die Abflussregimes der Schweiz . Publikation Gewasserkunde Nr. 65, Geographisches Institut der

Universitat Bern.Beniston M, Rebetez M, Giorgi F, Marinucci MR. 1994. An analysis of regional climate-change in Switzerland. Theoretical and Applied

Climatology 49(3): 135–159.Benn DI, Lehmkuhl F. 2000. Mass balance and equilibrium-line altitudes of glaciers in high-mountain environments. Quaternary Interna-

tional 65–66: 15–29.Braun LN, Weber M, Schulz M. 2000. Consequences of climate change for runoff from alpine regions. Annals of Glaciology 31: 19–25.Burlando P, Pellicciotti F, Strasser U. 2002. Modelling mountainous water systems between learning and speculating looking for challenges.

Nordic Hydrology 33(1): 47–74.Burman R, Pochop LO. 1994. Evaporation, Evapotranspiration and Climatic Data. Elsevier: Amsterdam.Christensen JH, Raisanen J, Iversen T, Bjørge D, Christensen OB, Rummukainen M. 2001. A synthesis of regional climate change

simulations—a Scandinavian perspective. Geophysical Research Letters 28: 1003–1006.Christensen JH, Carter TR, Giorgi F. 2002. PRUDENCE employs new methods to assess European climate change. Eos, Transactions of

the AGU 83: 147.Consuegra D, Vez E. 1996. AMIE: Analyse et Modelisation Integrees du cheminement des eaux en zones habitees, modelisation hydrologique,

Application au bassin versant de la Haute Broye. IATE/HYDRAM, Ecole Polytechnique Federale de Lausanne, Lausanne, Switzerland.Doms G, Schattler U. 1999. The nonhydrostatic limited-area model LM (Lokal-Modell) of DWD, part I: scientific documentation. Deutscher

Wetterdienst, Meteorologische Analyse und Modellierung, Offenbach, Germany.Ekstrom M, Jones PD, Fowler HJ, Lenderink G, Buishand TA, Conway D. 2006. Regional climate model data used within the SWURVE

project 1: projected changes in seasonal patterns and estimation of PET. Hydrology and Earth System Sciences accepted for publication.Etchevers P, Golaz C, Habets F, Noilhan J. 2002. Impact of a climate change on the Rhone river catchment hydrology. Journal of Geophysical

Research: Atmospheres 107(D16): 4293, DOI: 10.1029/2001JD000490.Frei C, Christensen JH, Deque M, Jacob D, Jones RG, Vidale PL. 2003. Daily precipitation statistics in regional climate models: evaluation

and intercomparison for the European Alps. Journal of Geophysical Research: Atmospheres 108(D3): 4124, DOI: 10.1029/2002JD002287.Gibelin AL, Deque M. 2003. Anthropogenic climate change over the Mediterranean region simulated by a global variable resolution model.

Climate Dynamics 20(4): 327–339.Giorgi F, Marinucci MR, Bates GT. 1993a. Development of a second generation regional climate model (REGCM2). Part I: boundary layer

and radiative transfer processes. Monthly Weather Review 121: 2794–2813.Giorgi F, Marinucci MR, Bates GT, DeCanio G. 1993b. Development of a second generation regional climate model (REGCM2). Part II:

cumulus cloud and assimilation of lateral boundary conditions. Monthly Weather Review 121: 2814–2832.Gordon C, Cooper C, Senior CA, Banks H, Gregory JM, Johns TC, Mitchell JFB, Wood RA. 2000. The simulation of SST, sea ice extents

and ocean heat transports in a version of the Hadley Centre coupled model without flux adjustments. Climate Dynamics 16(2–3): 147–168.Gross G, Kerschner H, Patzelt G. 1976. Methodische Untersuchungen uber die Schneegrenze in alpinen Gletschergbieten. Zeitschrift fur

Gletscherkunde und Glazialgeologie 12: 223–251.Hingray B, Mezghani A, Buishand TA. 2006a. Development of probability distributions for regional climate change from uncertain global

mean warming and an uncertain scaling relationships. Hydrology and Earth System Sciences accepted for publication.Hingray B, Mouhous N, Mezghani A, Bogner K, Schaefli B, Musy A. 2006b. Accounting for global warming and scaling uncertainties in

climate change impact studies: application to a regulated lakes system. Hydrology and Earth System Sciences accepted for publication.

Copyright 2006 John Wiley & Sons, Ltd. Hydrol. Process. 20, 2091–2109 (2006)

CLIMATE-CHANGE IMPACTS ON ALPINE DISCHARGE REGIME 2109

Horton P, Schaefli B, Mezghani A, Hingray B, Musy A. 2005. Prediction of climate change impacts on alpine discharge regimes under A2and B2 SRES emission scenarios for two future time periods (2020–2049, 2070–2099). HYDRAM Technical Report, Swiss Federal Officefor Energy. Bern (available from HYDRAM, EPFL, CH-1015 Lausanne).

Hulme M, Jenkins GJ, Lu X, Turnpenny JR, Mitchell TD, Jones RG, Lowe J, Murphy JM, Hassell D, Boorman P, McDonald R, Hill S.2002. Climate change scenarios for the United Kingdom: the UKCIP02 scientific report . Tyndall Centre for Climate Change Research,School of Environmental Sciences, University of East Anglia, Norwich, UK.

Jacob D. 2001. A note to the simulation of the annual and inter-annual variability of the water budget over the Baltic Sea drainage basin.Meteorology and Atmospheric Physics 77(1–4): 61–73.

Jasper K, Calanca P, Gyalistras D, Fuhrer J. 2004. Differential impacts of climate change on the hydrology of two alpine river basins.Climate Research 26(2): 113–129.

Jenkins G, Lowe J. 2003. Handling uncertainties in the UKCIP02 scenarios of climate change. Technical note 44, Hadley Centre, Exeter,UK.

Kirchhofer W, Sevruk B. 1991. Mean annual corrected precipitation depths 1951–1980. In Atlas Hydrologique de la Suisse. ServiceHydrologique et Geologique National: Bern.

Klok EJ, Jasper K, Roelofsma KP, Gurtz J, Badoux A. 2001. Distributed hydrological modelling of a heavily glaciated alpine river basin.Hydrological Sciences Journal 46(4): 553–570.

Laternser M, Schneebeli M. 2002. Spatial grouping of snow stations, snow and avalanche climatology of Switzerland . ETH Zurich.Meier MF, Post AS. 1962. Recent variations in mass net budgets of glaciers in western North America. In Symposium of Obergurgl, 10–18

September 1962, Variations of the Regime of Existing Glaciers . IAHS Publication No. 58. IAHS Press: Wallingford; 63–77.Nakicenovic N. 2000. Greenhouse gas emissions scenarios. Technological Forecasting and Social Change 65: 149–166.Nakicenovic N, Swart R (eds). 2000. Special Report on Emissions Scenarios (Intergovernmental Panel on Climate Change). Cambridge

University Press: Cambridge, UK.Nash JE, Sutcliffe JV. 1970. River flow forecasting through conceptual models. Part I, a discussion of principles. Journal of Hydrology

10(3): 282–290.New M, Hulme M, Jones P. 2000. Representing twentieth-century space–time climate variability. Part II: development of 1901–96 monthly

grids of terrestrial surface climate. Journal of Climate 13(13): 2217–2238.Oerlemans J, Anderson B, Hubbard A, Huybrechts P, Johannesson T, Knap WH, Schmeits M, Stroeven AP, van de Wal SW, Wallinga J,

Zuo Z. 1998. Modelling the response of glaciers to climate warming, Climate Dynamics 14(4): 267–274.Parde M. 1933. Le regime des cours d’eau de l’Europe orientale (en collaboration avec St. Kolupaila). Revue Geographie Alpine T.XXI(IV):

651–748.Paterson WSB. 1994. The Physics of Glaciers . Pergamon: Oxford.Pope VD, Gallani ML, Rowntree PR, Stratton RA. 2000. The impact of new physical parametrizations in the Hadley Centre climate

model—HadAM3. Climate Dynamics 16: 123–146.Porter SC. 1975. Equilibrium-line altitudes of late Quaternary glaciers in Southern Alps, New Zealand. Quaternary Research 5(1): 27–47.Raisanen J, Hansson U, Ullerstig A, Doscher R, Graham LP, Jones C, Meier HEM, Samuelsson P, Willen U. 2004. European climate in

the late 21st century: regional simulations with two driving global models and two forcing scenarios. Climate Dynamics 22(1): 13–31.Rango A, Martinec J. 1995. Revisiting the degree-day method for snowmelt computations. Water Resources Bulletin 31(4): 657–669.Roeckner E, Bengtsson L, Feichter J, Lelieveld J, Rodhe H. 1999. Transient climate change simulations with a coupled atmosphere–ocean

GCM including the tropospheric sulfur cycle. Journal of Climate 12(10): 3004–3032.Royer JF, Cariolle D, Chauvin F, Deque M, Douville H, Planton S, Rascol A, Ricard JL, Salas y Melia D, Sevault F, Simon P, Somot S,

Tyteca S, Terray L, Valcke S. 2002. Simulation of climate changes during the 21st century including stratospheric ozone. Comptes RendusGeoscience 334: 147–154.

Schadler B, Bigler R. 2002. Components of the natural water balance 1961–1990. In Atlas Hydrologique de la Suisse. Service Hydrologiqueet Geologique National: Bern.

Schaefli B. 2005. Quantification of modelling uncertainties in climate change impact studies on water resources: application to a glacier-fedhydropower production system in the Swiss Alps . Doctoral thesis, Ecole Polytechnique Federale de Lausanne.

Schaefli B, Hingray B, Niggli M, Musy A. 2005. A conceptual glacio-hydrological model for high mountainous catchments. Hydrology andEarth System Sciences 9: 95–109.

Shabalova MV, Van Deursen WPA, Buishand TA. 2003. Assessing future discharge of the River Rhine using regional climate modelintegrations and a hydrological model. Climate Research 23: 223–246.

SwissTopo. 1995. Digital height model of Switzerland—DHM25 . Swiss Federal Office of Topography, Wabern, Switzerland.SwissTopo. 1997. Digital national maps of Switzerland—PM25 . Swiss Federal Office of Topography, Wabern, Switzerland.Vidale PL, Luthi D, Frei C, Seneviratne S, Schar C. 2003. Predictability and uncertainty in a regional climate model. Journal of Geophysical

Research: Atmospheres 108(D18): 4586. DOI: 10.1029/2002JD002810.Wallinga J, van de Wal RSW. 1998. Sensitivity of Rhonegletscher, Switzerland, to climate change: experiments with a one-dimensional

flowline model. Journal of Glaciology 44(147): 383–393.Weber RO, Talkner P, Auer I, Bohm R, GajicCapka M, Zaninovic K, Brazdil R, Fasko P. 1997. 20th-century changes of temperature in the

mountain regions of central Europe. Climatic Change 36(3–4): 327–344.Wilby RL, Wigley TML. 1997. Downscaling general circulation model output: a review of methods and limitations. Progress in Physical

Geography 21(4): 530–548.Xu CY. 1999. From GCMs to river flow: a review of downscaling methods and hydrologic modelling approaches. Progress in Physical

Geography 23(2): 229–249.Zierl B, Bugmann H. 2005. Global change impacts on hydrological processes in Alpine catchments. Water Resources Research 41(2):

W02028. DOI: 10.1029/2004WR003447.

Copyright 2006 John Wiley & Sons, Ltd. Hydrol. Process. 20, 2091–2109 (2006)