STOTEN-15601; No of Pages 12 Science of the Total Environment xxx (2013) xxx–xxx

Contents lists available at ScienceDirect

Science of the Total Environment journal homepage: www.elsevier.com/locate/scitotenv

Does internal climate variability overwhelm climate change signals in streamflow? The upper Po and Rhone basin case studies S. Fatichi ⁎, S. Rimkus, P. Burlando, R. Bordoy Institute of Environmental Engineering, ETH Zürich, Stefano Franscini-Platz 5, HIL D 23.2, 8093 Zürich, Switzerland

H I G H L I G H T S • • • •

Natural climate variability is a fundamental source of uncertainty. Stochastic variability is mostly larger than climate change signals in streamflow. Glacier retreat and ice melt reduction are considerably affecting streamflow. Stochastic approaches are required for decision making and design under change.

a r t i c l e

i n f o

Article history: Received 30 June 2013 Received in revised form 4 December 2013 Accepted 4 December 2013 Available online xxxx Keywords: Climate change Hydrological modeling Stochastic downscaling Uncertainty Water resources Alps

a b s t r a c t Projections of climate change effects in streamflow are increasingly required to plan water management strategies. These projections are however largely uncertain due to the spread among climate model realizations, internal climate variability, and difficulties in transferring climate model results at the spatial and temporal scales required by catchment hydrology. A combination of a stochastic downscaling methodology and distributed hydrological modeling was used in the ACQWA project to provide projections of future streamflow (up to year 2050) for the upper Po and Rhone basins, respectively located in northern Italy and south-western Switzerland. Results suggest that internal (stochastic) climate variability is a fundamental source of uncertainty, typically comparable or larger than the projected climate change signal. Therefore, climate change effects in streamflow mean, frequency, and seasonality can be masked by natural climatic fluctuations in large parts of the analyzed regions. An exception to the overwhelming role of stochastic variability is represented by high elevation catchments fed by glaciers where streamflow is expected to be considerably reduced due to glacier retreat, with consequences appreciable in the main downstream rivers in August and September. Simulations also identify regions (west upper Rhone and Toce, Ticino river basins) where a strong precipitation increase in the February to April period projects streamflow beyond the range of natural climate variability during the melting season. This study emphasizes the importance of including internal climate variability in climate change analyses, especially when compared to the limited uncertainty that would be accounted for by few deterministic projections. The presented results could be useful in guiding more specific impact studies, although design or management decisions should be better based on reliability and vulnerability criteria as suggested by recent literature. © 2013 Elsevier B.V. All rights reserved.

1. Introduction The scientific community is subjected to a rising pressure for providing projections of climate change impacts on water resources at spatial and temporal scales at which water managers take decisions (Hill Clarvis et al., 2014), a scale that is typically addressed neither by global and nor by regional scale climate simulations (Schmidli et al., 2006; Fowler et al., 2007; Maraun et al., 2010; Rajczak et al., 2013). This pressure is dictated by the fact that water management systems and flood protection measures were mostly designed and are still ⁎ Corresponding author. Tel.: +41 44 6324118; fax: +41 44 3331539. E-mail address: [email protected] (S. Fatichi).

operated under the assumption of stationarity, i.e., that the climate system fluctuates within an unchanging envelope of variability (Milly et al., 2008). The stationary concept is now strongly questioned by the fact that the Earth energy budget and consequently the Earth climate are modified by the increasing concentration of human emitted greenhouse gases (e.g., Huber and Knutti (2012)). Therefore, scientists are asked to quantify changes in hydrological budget components and, most importantly, in streamflow to guide engineering design and management under climate change conditions (Fatichi et al., 2012a). This is especially true for a crucial geographical area such as the European Alps that is expected to be highly sensitive to climate change (Fuhrer et al., 2006; Viviroli et al., 2011) and that represents the principal focus of the European project “ACQWA” (Assessing Climate

0048-9697/$ – see front matter © 2013 Elsevier B.V. All rights reserved. http://dx.doi.org/10.1016/j.scitotenv.2013.12.014

Please cite this article as: Fatichi S, et al, Does internal climate variability overwhelm climate change signals in streamflow? The upper Po and Rhone basin case studies, Sci Total Environ (2013), http://dx.doi.org/10.1016/j.scitotenv.2013.12.014

2

S. Fatichi et al. / Science of the Total Environment xxx (2013) xxx–xxx

change impacts on the Quantity and quality of WAter) (Beniston et al., 2011). European Alps cover a wide range of elevations and climates (from continental to Mediterranean) and are characterized by rather different hydrological regimes (glaciological, nival, pluvial), which make any assessment of climate change effects in streamflow particularly challenging (Schädler and Weingartner, 2010; Viviroli et al., 2011). In this study, we present a summary of the key results obtained in the ACQWA project in terms of projected changes in streamflow metrics. Projections are presented for the near future (up to year 2050) and for several locations distributed throughout the river networks of two among the major rivers originating from the European Alps, the upper Po and the upper Rhone, respectively located in northern Italy and south-western Switzerland. Results are obtained through a combination of stochastic downscaling of climate model realizations and distributed hydrological modeling (Burlando and Rosso, 2002; Manning et al., 2009; Sulis et al., 2012). Specifically, realizations from one General Circulation Model (GCM), ECHAM5 and two regional climate models (RCMs), RegCM3 and REMO for the emission scenario A1B were selected as drivers of climate change in the ACQWA project (Bordoy, 2013; Bordoy and Burlando, 2013b). The distributed hydrological model Topkapi-ETH was used to transfer the downscaled changes in the climate forcing (air temperature and precipitation) into changes of hydrological variables, e.g., streamflow, ice melt, soil water content, etc. (Liu and Todini, 2002; Fatichi et al., 2013b). The adoption of the Topkapi-ETH model was a pragmatic choice driven by the computational effort required to run multiple, long-term, distributed hydrological simulations. For this reason, it was preferred in comparison to more detailed hydrological models (e.g., Rigon et al., 2006; Ivanov et al., 2008; Fatichi et al., 2012b,c). Future projections of streamflow require a series of assumptions and methodological steps that generate a considerable amount of uncertainty. Uncertainty is originated from at least five different sources that can be listed as: (i) CO2 emission scenario, (ii) driving climate model, (iii) internal (stochastic) climate variability, (iv) downscaling methodology, and (v) hydrological model. Since the presented climate change impact analysis is limited to year 2050, the CO2 emission scenario represents a rather small source of uncertainty (Hawking and Sutton, 2009; Prein et al., 2011). Conversely, the spread among climate model predictions has been discussed to represent one of the principal sources of the overall uncertainty (Räisänen, 2007; Knutti, 2008). Therefore, there is the possibility that evaluating climate change signals using only one GCM and two RCMs can underestimate significantly the total uncertainty being other climate models neglected (Christensen et al., 2010; Knutti et al., 2010). We show that this limitation is mostly overcome by the adopted methodology, which allows including the effects of internal (stochastic) climate variability using a stochastic downscaling approach (Section 2). Accounting for stochastic variability has been demonstrated to alleviate the issue of adopting only one or few climate models with respect to the use of a model ensemble (Deser et al., 2012b; Fatichi et al., 2013a), because it represents the principal source of uncertainty for the near future (Hawking and Sutton, 2011). Finally, uncertainties related to the choice of a specific downscaling methodology and hydrological model were not addressed being considered of minor importance when compared to the other sources. The large uncertainty expected in climate change projections due to the above mentioned sources has recently led to question the added value of using climate change projections (Dessai et al., 2009). Approaches for robust adaptations to climate change were thus proposed following a bottom-up strategy based on vulnerability scenarios. In such an approach climate change projections at fine spatial and temporal scales were considered unnecessary since decision should be guided by system vulnerabilities identified using a large range of possible forcings (Wilby and Dessai, 2010; Pielke et al., 2012; Brown and Wilby, 2012). This alternative strategy is very interesting and promising for taking decision in uncertain conditions, however we

argue that it does not fully eliminate the need for downscaled meteorological variables, since many impact models cannot be forced with a theoretically “infinite” range of forcings, and because of the practical problems involved with the definition of the “large range” of forcings (Steinschneider and Brown, 2013). In this regard, the study we present here and the relevant methods could be very useful in delineating a range of plausible future projections, in which important uncertainty sources are accounted for. 2. Material and methods 2.1. Case studies Two major case studies were used within the ACQWA project to analyze climate change impacts on the hydrological regimes: the upper Rhone and the upper Po basins. These two river basins are located in the western side of the Alps and drain respectively an area of 5338 and 37,874 km2 at their outlets assumed to be Porte du Scex (close to the Lake Geneva) for the upper Rhone, and Ponte della Becca (after the confluence with the Ticino river) for the upper Po (Fig. 1). Climate in the upper Po basin is mainly Mediterranean with hot summers and relatively wet springs and autumns while the upper Rhone basin is influenced by both Mediterranean and Atlantic climate with larger precipitation during summer months (Fig. 2). Mean annual air temperature and precipitation in the observational periods are 1400 mm yr−1 and 1.8 °C for the upper Rhone (October 1990–December 2008) and 1123 mm yr−1 and 8.9 °C for the upper Po (October 1999–December 2010). The upper Rhone is a typical alpine catchment with an average elevation of about 2100 m a.s.l. characterized by steep slopes and lateral narrow valleys, as well as a long central valley. The upper Po basin has an average elevation of about 1130 m a.s.l. and a more heterogeneous topography, which is characterized by the Alpine mountain range in the north and west sides, the Apennines in the south side and a large plain in its central-east part (Fig. 1). 2.2. Stochastic climate forcing Future climate projections derived from global or even regional climate models cannot be directly used to force detailed hydrological models due to their coarse spatial resolution and bias with respect to the observed climate. For these reasons, methodologies called “downscaling” have been presented in the literature to transfer the climatic information obtained from climate models to the temporal and spatial scales required by local analysis, including hydrological modeling (Burlando and Rosso, 2002; Wilby et al., 2002; Fowler et al., 2007; Maraun et al., 2010; Fatichi et al., 2011). Time series of meteorological variables representing future scenarios were simulated for the case studies presented here using a stochastic downscaling methodology (Burlando and Rosso, 2002; Burton et al., 2010; Bordoy and Burlando, 2013b), which makes use of realizations of daily precipitation and air temperature from one GCM, ECHAM5, and two RCMs, REMO and RegCM3, for the emission scenario A1B (Roeckner et al., 2003; Jacob et al., 2007; Pal et al., 2007). The simulations with the RCMs, REMO and RegCM3, were driven by the same GCM, ECHAM5. Since the ACQWA project focused in the near future up to 2050, differences due to emission scenarios are considered of minor importance and therefore neglected (Prein et al., 2011). The methods use factors of change of various climatic statistics (e.g., mean, variance, no-rain probability, and skewness of precipitation at daily scale, mean and standard deviation of daily temperature) derived as ratios between two periods from bias corrected climate models. A non-linear parametric bias correction to the entire distribution of precipitation and air temperature was applied to remove the differences between station observations and climate model realizations in the control scenario (Bordoy and Burlando, 2013a). Specifically, after bias correction factors of change were derived comparing each decade in

Please cite this article as: Fatichi S, et al, Does internal climate variability overwhelm climate change signals in streamflow? The upper Po and Rhone basin case studies, Sci Total Environ (2013), http://dx.doi.org/10.1016/j.scitotenv.2013.12.014

S. Fatichi et al. / Science of the Total Environment xxx (2013) xxx–xxx

Fig. 1. Digital Terrain Model (DTM), catchment boundaries of the upper Po and Rhone basins, glaciated areas, major rivers and lakes for the case study area.

the period 2011 through 2050 with the control scenario, that is 1992–2010 for the case study of the upper Rhone and 2002–2010 for the upper Po. All of these periods were assumed to be climate stationary as required by the downscaling procedure. Factors of change were subsequently used to modify the parameters of two stochastic generators. The first generator is used to simulate multi-site hourly precipitation (Bordoy, 2013; Bordoy and Burlando, 2013c) and it is based on the Neyman–Scott Rectangular Pulses spatiotemporal model (Cowpertwait, 2006; Burton et al., 2008, 2010; Paschalis et al., 2013). The second generator is used to simulate multisite hourly air temperature and it is based on a multivariate autoregressive model once seasonality and daily cycle of air temperature are subtracted (Bordoy, 2013). Hourly precipitation and air temperature are assumed uncorrelated. The parameters of two stochastic generators were originally derived for the control scenario periods based on station observations. The modified parameters accounting for the factors of change are used for the simulations of future decades. For additional methodological details on the bias correction strategy, and description of the stochastic generators and of the downscaling methodology the reader is referred to Bordoy and Burlando (2013a,b, submitted for publication) and Bordoy (2013). By using stochastic generators for precipitation and air temperature, we were able to simulate ensembles of possible climate trajectories for all of the analyzed periods, i.e., the control scenarios of the Po and Rhone and each future decade up to 2050. For a given period, assumed as stationary, all the trajectories share the same baseline climate (dictated by the climate change signal) but represent possible different realizations as a result of the internal (stochastic) climate variability. This variability occurs in the absence of any external forcing and is mostly related to non-linear dynamical processes taking place in the atmosphere and to feedbacks between ocean, land and atmosphere (Schneider and Kinter, 1994; Deser et al., 2012a,b). In stochastic generators this variability is imposed by sampling random occurrences from probabilistic distributions within the generators and typically

3

slightly underestimates the natural variability (Katz and Parlange, 1998; Wilks, 1999) but see Fatichi et al. (2011) for an exception. In this study, 60 and 85 realizations for each period (control scenario and future decades) and for each climate model (ECHAM5, REMO, RegCM3) driving the stochastic downscaling were generated to analyze changes of the hydrological regime in the upper Rhone and Po basins, respectively. The number of ensemble members was obtained as a pragmatic compromise between a large stochastic ensemble and computational feasibility for hourly distributed hydrological simulations spanning over 50–60 years. Since climate change signals are derived from only one GCM and two RCMs, we compared the variability in climate changes obtained with the stochastic downscaling methodology applied in this study and using two popular ensembles of climate models. Specifically, we directly computed seasonal changes of precipitation and air temperature averaged over the river basins using realizations of 20 RCMs from the ENSEMBLES project (van der Linden and Mitchell, 2009; Fischer et al., 2012) and 12 GCMs from the CMIP3 dataset (Meehl et al., 2007), emission scenario A1B. Results are presented for the upper Po basin (Fig. 3) but they are rather similar for the upper Rhone (Fatichi et al., 2013b). For air temperature the range of variability provided by the stochastic downscaling is slightly smaller in most of the months than the climate change forcing derived from the two ensembles of climate models. However, the variability of changes in precipitation is typically larger in the stochastic downscaling rather than in the multi-model ensembles (Fig. 3). This is the result of the stronger natural variability of precipitation compared to air temperature that is neglected considering only deterministic trajectories in the GCM and RCM simulations. Note that in certain months the median of the stochastic ensemble deviates from the medians of the two multi-model ensembles. Such an outcome should not be surprising because it is the combined result of using a single driving GCM and of filtering the climate change signal through the bias correction and downscaling methodologies. 2.3. Hydrological simulations The hydrological simulations have been carried out with the distributed hydrological model Topkapi-ETH (Fatichi et al., 2013b), which is a substantial evolution of the rainfall-runoff Topkapi (TOpographic Kinematic APproximation and Integration) model, originally presented by Ciarapica and Todini (2002) and Liu and Todini (2002). The model explicitly resolves the topography of the catchment using a regular square grid and accounts for the principal hydrological processes such as evapotranspiration, snow accumulation and melting, ice melting, infiltration of surface water, soil drainage, and subsurface, overland, and channel lateral flows. The model is designed to account for hydraulic infrastructures such as reservoir and river diversions as well as for water withdrawals for irrigation and domestic/industrial water consumption. These characteristics make Topkapi-ETH suitable to deal with river basins presenting strong anthropogenic influences such as Po and Rhone. For a more detailed description of the model the reader is referred to Rimkus (2013) and Fatichi et al. (2013b). Topographic data were obtained from Digital Elevation Models, with a resolution of 250 × 250 m2 for the upper Rhone and 1000 × 1000 m2 for the upper Po. These resolutions were chosen as a compromise between preserving a detailed spatial representation of the basin topography and the computational cost of using finer grids. Nonetheless, we recognize that the chosen resolutions do not allow to resolve detailed topographic and glacier features. Land cover data were obtained from the Global Land Cover Product (GlobCover), glacier cover was assigned based on the GLIMS Glacier Database, and soil maps were made available by partners of the project (see Fatichi et al. (2013b) for more details). Topkapi-ETH was first used to simulate hourly streamflow throughout the river network for the periods where meteorological

Please cite this article as: Fatichi S, et al, Does internal climate variability overwhelm climate change signals in streamflow? The upper Po and Rhone basin case studies, Sci Total Environ (2013), http://dx.doi.org/10.1016/j.scitotenv.2013.12.014

4

S. Fatichi et al. / Science of the Total Environment xxx (2013) xxx–xxx

Upper Po 20

Temperature [°C]

Precipitation [mm]

150

15 100

10 5

50

0 0

1

2

3

4

5

6

7

8

9

10 11 12

−5

Month Upper Rhone 15

Temperature [°C]

Precipitation [mm]

150

10 100

5 0

50

−5 0

1

2

3

4

5

6

7

8

9

10 11 12

−10

Month Fig. 2. Monthly precipitation (bars) and air temperature (dotted lines) averaged over the upper Po (subplot a) and Rhone (subplot b) basins in the observational periods, 1992–2008 for Rhone and 2002–2010 for Po.

observations were available to force the model. Precipitation, air temperature, and cloud transmissivity are the meteorological input variables requested by Topkapi-ETH and were estimated using a combination of station based data and distributed gridded products (Fatichi et al., 2013b). The historical observation periods correspond to October 1990 through December 2008 for the upper Rhone and October 1999 through December 2010 for the upper Po. Simulations forced with observed meteorological data were used to confirm model

performances against measured discharge for the same periods. Streamflow simulations for the future decades (2011–2050) were obtained forcing Topkapi-ETH with the ensembles of time series of precipitation and air temperature simulated using the two stochastic generators parameterized for the future climates. Climate change effects were finally analyzed comparing streamflow statistics in the future decades with streamflow statistics obtained from additional hydrological simulations representative of the control scenario climate (1992–2010 for the upper Rhone and 2002–2010 for the upper Po); these are forced with stochastically generated time series. Such a procedure allow us to account for the internal climate variability of the control scenario and to compare present and future hydrological simulations both forced with generated meteorological time series, thus filtering possible biases introduced by the stochastic generators. Streamflow regimes in both river basins are highly influenced by the presence of reservoirs, river diversions, irrigation channels and withdrawals for domestic and industrial uses. These disturbances are explicitly accounted for in the hydrological modeling, simulating the effect of 14 reservoirs in the upper Rhone and 57 in the upper Po. River diversions and water withdrawals were implemented with a greater detail in the upper Rhone due to the adopted finer spatial resolution (Fatichi et al., 2013b). Irrigation was simulated subdividing the irrigated areas in districts and accounting for soil water content conditions and irrigation channel capacity (for further methodological details the reader is referred to Fatichi et al. (2013b)). Results and model performance for the historical period are widely discussed in Fatichi et al. (2013b) for the upper Rhone. Briefly, in this catchment Topkapi-ETH was shown to provide very good results in reproducing the streamflow regime at the hourly, daily, monthly, and annual time scale in 15 points where discharge measurements were available. The simulation of distributed snow cover was also satisfactory once compared with snow cover from remote sensing products of the Moderate Resolution Imaging Spectroradiometer (MODIS). Confirmation metrics in terms of determination coefficient (R2) and Nash–Sutcliffe efficiency (NS) for daily discharge in the period October

a) Changes in mean air temperature 2041−2050 vs. 2002−2010 4 3

[°C]

2 1 0 −1

b) Changes in mean precipitation 2041−2050 vs. 2002−2010 2.5 2

[−]

1.5 1 0.5 0

Jan

Feb

Mar

Apr

May

Jun

Jul

Aug

Sep

Oct

Nov

Dec

Fig. 3. Boxplots representing the variability of changes in climate forcing for each month over the upper Po basin computed as differences between the future period 2041–2050 and control scenario 2002–2010. The left boxplots (red) are constructed with changes derived from the 20 RCMs used in the ENSEMBLES project (van der Linden and Mitchell, 2009), the central boxplots (magenta) are constructed with changes derived from 12 GCMs used in the CMIP3 dataset (Meehl et al., 2007), and the right boxplots (black) are constructed using the stochastic downscaling presented in this study (stochastic ensembles driven by ECHAM5, RegCM3, and REMO are combined for this purpose). The changes are for mean air temperature (delta-change) (a) and mean precipitation (ratio between 2041–2050 and 2002–2010) (b). On each box, the central mark is the median, the edges of the box are the 25th and 75th percentiles, and the whiskers extend to the most extreme data points not considered as outliers, outliers are plotted individually. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

Please cite this article as: Fatichi S, et al, Does internal climate variability overwhelm climate change signals in streamflow? The upper Po and Rhone basin case studies, Sci Total Environ (2013), http://dx.doi.org/10.1016/j.scitotenv.2013.12.014

S. Fatichi et al. / Science of the Total Environment xxx (2013) xxx–xxx

1990 through December 2008 were ranging from R2 above 0.85 in the smaller upstream catchments to 0.7–0.8 for stations in the main reach of Rhone, down to 0.6–0.7 in the most disturbed catchments. A confirmation of the seasonality of discharge for natural streamflow conditions (“pre-dam” period) was also very satisfactory confirming that the model simulations are capable to well reproduce the effect of human operations on streamflow (Fatichi et al., 2013b). The comparison between simulated and observed streamflow in the upper Po basin was carried out for 43 stream gauges where observed discharge was available, even though with significant data gaps and low quality for most of the stream gauges in the Italian territory. Results are worse than in the case of the upper Rhone but still satisfactory. The medians of the determination coefficient (R2) for the 43 check points are 0.54, 0.63, 0.77, and 0.86 and the medians of the Nash–Sutcliffe efficiency (NS) are 0.46, 0.53, 0.59, and 0.46, for hourly, daily, monthly, and annual discharge, respectively. There is no correlation between the obtained metrics and basin size, elevation, and geographical distribution, most probably due to the interplay of precipitation accuracy, river network and horizontal grid resolution in the model, effects of reservoir management, and uncertainty in the discharge observations. All these factors are contributing to create such a diverse performance. The best results are obtained for the Toce river at Candoglia and for the Bormida river at Cassine with R2 at the hourly (daily) scale of 0.81 (0.85) and 0.79 (0.80) respectively. The worst results are obtained for the Ticino at Piotta and Maira at Racconigi with R2 at the hourly (daily) scale of 0.29 (0.42) and 0.38 (0.44), respectively. For Ticino at Piotta the unsatisfactory performance is related to effects of reservoir management and river diversions which are only partially reproduced in the model. For Maira at Racconigi the explanation for such a performance rather relies on a poor spatial

5

representation of the floodplain area south of Torino that could have led to a misinterpretation of the upstream contributing area. The seasonality of discharge is well reproduced by the model in most of the analyzed river sections with the exception of the Ticino at Piotta, Varaita at Polonghera, Stura di Demonte at Fossano and Dora Riparia at Susa. Differences between observed and simulated discharge amounts are typically in the order of ± 25%, a significantly worse performance when compared to the upper Rhone (Fatichi et al., 2013b); nonetheless, biases tend to decrease with increasing basin size. The parameters of Topkapi-ETH for the upper Po and Rhone adopted for the observational periods were successively used to simulate the hydrological response for 60 (Rhone) and 85 (Po) stochastic climate trajectories (for each forcing climate model) obtained as explained in Section 2. The single core computational time of all the hydrological simulations was 1268 days for the Rhone and 658 days for the Po, highlighting the relatively large computational effort required for stochastic, hourly, distributed hydrological simulations with a model such as Topkapi-ETH. 3. Results and discussion Simulations driven by the three climate models and for the four future decades give different changes in mean discharge for both the upper Rhone and Po basins. Specifically, changes in annual discharge at the basin outlets averaged across the stochastic ensembles range between −139 and +156 mm yr−1 (equivalent to −13%, +14%) for the upper Rhone and between − 60 and + 207 mm yr−1 (equivalent to −9.7%,+33%) for the upper Po when different decades or driving climate models are considered. These changes need to be compared with the standard deviation of the stochastic ensemble (natural variability)

a) UPPER PO Difference 2041 − 2050 vs 2002 − 2010

[mm yr−1]

500

0

−500

−1000

25

50

75

100

125

Station

b) UPPER RHONE Difference 2041 − 2050 vs 1992 − 2010

[mm yr−1]

500

0

−500

−1000 40

80

120

160

200

240

280

Station Fig. 4. Boxplots representing the difference in simulated streamflow [mm yr−1] for 297 and 145 points (Rhone and Po respectively) distributed throughout the river networks and ranked according to the upstream contributing area. The differences are between the future period 2041–2050 and the control scenario, 1992–2010 for the upper Rhone and 2002–2010 for the upper Po. Results are obtained combining the three ensembles of simulations forced with the stochastic downscaling driven by ECHAM5, RegCM3 and REMO.

Please cite this article as: Fatichi S, et al, Does internal climate variability overwhelm climate change signals in streamflow? The upper Po and Rhone basin case studies, Sci Total Environ (2013), http://dx.doi.org/10.1016/j.scitotenv.2013.12.014

6

S. Fatichi et al. / Science of the Total Environment xxx (2013) xxx–xxx

in a decadal period. Standard deviations are 50–100 mm yr− 1 and 30–45 mm yr− 1 for the upper Rhone and Po respectively. The stochastic variability is mostly related to the variability of precipitation being natural variability in ten-year evapotranspiration and ice melt in the two basins rather small (b 10 mm yr−1). The minor stochastic variability in the Po is due to its larger area, which has an enhanced dampening effect of local climatic fluctuations. The net result is that robust projections of changes in mean discharge at the outlet of the upper Rhone and Po basins remain substantially elusive because of the small signal to noise ratio, with a tendency toward an absence of change when all the members of the stochastic ensembles are grouped together. In other words, even climate change signals of ± 20% in streamflow can be neither identified nor rejected due to differences in the scenarios driven by the three climate models and most importantly due to the internal climate variability that it is reflected in the stochastic ensemble and in the variability among decades. The large uncertainty in detecting any possible change is not confined to the average discharge at the basin outlets but is consistent for most of the sub-basins with few exceptions that are described in the following. Changes in the average discharge between the decade 2041–2050 and the control period were sorted according to the upstream catchment area for 297 and 145 selected points (Rhone and Po respectively) distributed throughout the river networks (Fig. 4). The selected points cover in a rather uniform way the river networks and span catchment sizes as small as few square kilometers up to the entire basins (Fig. 5). Given the large number of selected river sections the arbitrary choice of the points has no effect on the presented results and conclusions. Boxplots are obtained combining together all of the stochastic simulations produced with the three different driving climate models. For most of the analyzed river sections both a positive and a negative change in streamflow can occur with a central tendency toward unmodified streamflow mean. There is also a trend toward a decrease in uncertainty in the largest sub-basins, especially within the upper Po (Fig. 4). A clear result is that the combination of internal climate variability and change signals coming from the three climate models is providing very uncertain predictions even on the sign of the change in most of the stations of both the upper Rhone and Po, with few exceptions. An exception is represented by stations presenting a significant decrease of discharge regardless of stochastic variability. This is more evident in the upper Rhone basin where some stations show a decrease for all of the members of the stochastic ensemble. Another exception is represented by a group of stations that show a robust tendency toward an increase in streamflow although with large uncertainty. Such a behavior can be found in both the upper Rhone and Po and it is related to consistent projections of increasing precipitation in two distinct geographical regions. On the basis of these results, we identified three principal regimes of projections in mean and seasonality of streamflow that are further analyzed in the following sections. Changes in mean discharge and seasonality is only one way to look at climate change effects in streamflow being changes in higher order statistics very important. Analyzing changes in these statistics is however even more difficult than analyzing changes in the mean due to their larger natural variability. A detailed analysis of changes in maxima and minima at different time scales for selected points in the upper Rhone is presented in Fatichi et al. (2013b). This shows how, despite noticeable uncertainties, maximum flow at the hourly and daily time scale have a tendency to increase throughout the basin. Changes in the minimum flow are instead more uncertain and did not present any clear trend. The same finding holds true for the river sections of the Po river. Changes in both minimum and maximum flows and at different aggregation periods (from hourly to 30 days) did not show any trend (not shown). While the central tendency is pointing toward a lack of change, changes of ±50% are simulated as possible, with internal climate variability that precludes any robust conclusion for these metrics.

Fig. 5. River network of the upper Po and Rhone basins, the 297 and 145 river sections (Rhone and Po respectively) analyzed in the simulations are indicated and the 17 river sections used to highlight specific regimes are marked with larger symbols.

3.1. Projection regime I: dominance of stochastic variability The first regime includes the large majority of the selected river sections and is characterized by projected changes that are mostly within the stochastic variability of the control scenario. This does not only refer to the mean streamflow but also to its seasonality (Fig. 6) and its frequency distribution (Fig. 7). Projected seasonality of streamflow for the decade 2041–2050 in six exemplary subcatchments in the Po and three in the Rhone basins is largely within the range imposed by the stochastic variability of the control scenario. Catchments are selected to represent different geographical locations and contributing areas (Fig. 5). This result is mostly independent from the climate model driving the downscaling scenario and from the month of the year. A reduction in streamflow below the range of the control scenario is appreciable only for the summer months in some of the river sections (e.g., Po at Isola San Antonio, Rhone at Fully) and it is related to the considerable reduction of the ice melt contribution that represents an important percentage of summer runoff for these stations (see further discussion below). The frequency analysis of discharge represented in terms of flow duration curves, i.e., the Cumulative Frequency Distributions, CDFs, of streamflow with the probability expressed in days confirms the dominant role of stochastic variability. Present climate variability is very similar to the projections obtained with the three stochastic ensembles driven by different climate models for most of the stations and durations (Fig. 7). The stochastic ensemble driven by the GCM ECHAM5 tends to produce the drier simulations with a decrease in medium and low flows for many stations. However, significant departures from the variability of the control scenario are appreciable only for duration larger than 250 days (low flows) in the Po at Isola

Please cite this article as: Fatichi S, et al, Does internal climate variability overwhelm climate change signals in streamflow? The upper Po and Rhone basin case studies, Sci Total Environ (2013), http://dx.doi.org/10.1016/j.scitotenv.2013.12.014

S. Fatichi et al. / Science of the Total Environment xxx (2013) xxx–xxx

Po − Isola San Antonio (26214 km2)

[m3 s−1]

800

Bormida − Alessandria (2613 km2)

100

100

50

50

400 200 0

2

4

6

8

10

12

0

Belbo − Castelnuovo (376 km2)

2

4

6

8

10

12

0

2

Stura di Lanzo − Lanzo (647 km2)

4

6

8

10

12

Gesso − Cuneo (578 km2)

40

10

[m3 s−1]

Po − Torino (5126 km2) 150

150

600

7

40

30 20

5

20 10

0

2

4

6

8

10

12

0

2

Rhone − Fully (3774 km2)

4

6

8

0

2

4

6

8

10

12

Morge − Conthey (78 km2)

1992/2002−2010 2041−2050 ECHAM 10 2041−2050 RegCM3 2041−2050 REMO

15

[m3 s−1]

12

Vieze − Monthey (138 km2)

300

200 10

5

100

0

10

5

2

4

6

8

10

12

0

2

4

Month

6

8

Month

10

12

0

2

4

6

8

10

12

Month

Fig. 6. A comparison among simulated annual cycles of streamflow [m3 s−1] for 9 selected river sections belonging to regime I. The comparison is between the control scenario period (1992–2010 for the upper Rhone and 2002–2010 for the upper Po), and future scenarios 2041–2050 for the stochastic downscaling driven by RegCM3, ECHAM5, and REMO. The colored bands include simulations within the 10 and 90 percentiles of the stochastic ensembles. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

San Antonio and Stura di Lanzo at Lanzo river sections reflecting an average decrease in low flows during summer. 3.2. Projection regime II: glacierized catchments The second regime characterizes all of the river sections for which the contribution from melting glaciers represents a significant fraction of the annual runoff. In all of these catchments streamflow is projected to decrease below the natural climatic variability of the control scenario because of the progressively reduced contribution of ice melt. This is shown using four exemplary catchments, two in the Po and two in the Rhone (Fig. 5), where June to September discharge in the decade 2041–2050 is considerably lower than for the control scenario, independently of the climate model driving the stochastic downscaling (Fig. 8). Discharge in the other months is within the uncertainty of the control scenario but it only represents a minor fraction of the annual runoff. Simulated glaciers are projected to retreat and to be confined at very high elevations thus determining a decrease of the ice melt contribution. As discussed by Fatichi et al. (2013b) ice melt from glaciers at relatively low elevation is not a renewable resource even without additional climate warming. The persistence of climatic conditions similar to the control scenario is already sufficient to considerably modify the streamflow regime of highly glacierized catchments. Note that changes in ice melt are relatively independent from changes in

snow melt. The contribution of snow melt is expected to remain similar or even increase in the projected future with just a time shift toward an earlier onset of melting. The strong effect of retreating and disappearing glaciers is considerably dampened with increasing contributing area and as the catchment glacierized fraction decreases. It is obviously absent in all the catchments without glaciers that represent the majority of the selected river sections in the upper Po. The relatively small contribution of ice melt to total streamflow with decreasing glacierized area explains why most of the catchments show a dominance of stochastic climate variability and belong to the projection regime I also in the upper Rhone. However, the effects on summer streamflow, especially during August and September, can be significant also when the fraction of glacierized area is small, as pointed out by the results for the river section of Po at Isola San Antonio and Rhone at Fully (Fig. 6). In order to highlight this point, we show the seasonality of the ratio between simulated streamflow and ice melt during the observational period over the entire upper Po and Rhone basins (Fig. 9). This analysis does not account for the routing time in the river network but it suggests that in August streamflow is composed of about 35–40% ice melt in the upper Rhone. This fraction reduces to 10–15% in the upper Po. These contributions are produced by a fraction of glacierized catchment area at the beginning of the simulation equal to 11.5% and 0.5% for the upper Rhone and Po, respectively. The very large contribution

Please cite this article as: Fatichi S, et al, Does internal climate variability overwhelm climate change signals in streamflow? The upper Po and Rhone basin case studies, Sci Total Environ (2013), http://dx.doi.org/10.1016/j.scitotenv.2013.12.014

8

S. Fatichi et al. / Science of the Total Environment xxx (2013) xxx–xxx

Po − Isola San Antonio (26214 km2) Bormida − Alessandria (2613 km2)

[m3 s−1]

10

10

10

4

Po − Torino (5126 km2)

3

10

3

2

10

2

10

2

1

10

1

10 10

1

0

0

100

200

300

Belbo − Castelnuovo (376 km2) 10

10

0

100

100

200

300

10

Rhone − Fully (3774 km2)

200

300

Gesso − Cuneo (578 km2)

10

0

0

100

1

1

10

0

0

1992/2002−2010 2041−2050 ECHAM 2 2041−2050 RegCM310 2041−2050 REMO

2 2

[m3 s−1]

300

Stura di Lanzo − Lanzo (647 km2)

10

10

200

0

100

200

300

10

0

0

Vieze − Monthey (138 km2)

100

200

300

Morge − Conthey (78 km2)

2

10

[m3 s−1]

103

1

1

10

0

10

10 102

0

10 101

-1

-1

0

100

200

300

10

0

100

[Days]

200

300

[Days]

10

0

100

200

300

[Days]

Fig. 7. A comparison among simulated flow duration curves for 9 selected river sections belonging to regime I. The comparison is between the control scenario period (1992–2010 for the upper Rhone and 2002–2010 for the upper Po), and future scenarios 2041–2050 for the stochastic downscaling driven by RegCM3, ECHAM5, and REMO. The colored bands include simulations within the 10 and 90 percentiles of the stochastic ensembles. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

relative to the glacierized area in the Po basin is related to the fact that in August only few additional sources of runoff exist in addition to ice melt due to the relatively dry soil conditions in most of the catchment. These results are consistent with the estimates of ice melt contribution given by Huss (2011), which were obtained with a very different methodology based on glacier mass balance and streamflow observations. Simulations suggest that about 70% of the ice melt contribution for the Rhone and 90% for the Po could disappear by 2050. This has an impact in August streamflow that can be quantified in about − 25% and − 10% for the entire upper Rhone and Po, respectively. Note that this result is substantially independent on the projected climate change and stochastic downscaling. However, the strong uncertainties related to the initialization of ice thickness and to the adopted spatial resolution, which is insufficient to describe small features of glaciers (see discussion in Fatichi et al., 2013b), could lead to a poor prediction on the timing of the glacier melt and retreat. Most probably, the presented simulations tend to anticipate glacier disappearance when compared to other glacier focused studies (Gabbi et al., 2012). 3.3. Projection regime III: increasing precipitation Projected changes in precipitation are mostly within the range of natural variability of the control scenario leading to uncertain consequences in streamflow in most of the catchments where glaciers have no significant impact (regime I). However, there are specific geographical areas where there is a convergence among the simulations driven by the RCMs toward an increase in February to April and

sometimes October to November precipitation. These areas include the north-east of the upper Po basin, i.e., the basins of Toce and Ticino rivers and the north-west of the upper Rhone, including several Rhone tributaries such as Sionne, Morge, Lizerne, Grand Eau, Vieze, and Trient. In these areas a significant increase of precipitation is projected by the stochastic ensembles driven by the two RCMs, REMO and RegCM3, while the stochastic ensemble driven by the GCM ECHAM5 provides results similar to regime I. The projected increase in February to April precipitation has direct consequences in the seasonality of streamflow with simulated values in the 2041–2050 decade during the melting season that are significantly larger than for the control scenario(Fig. 10). Four exemplary stations located in the Toce, Ticino and Sionne basins (Fig. 5) highlight such a behavior with simulated streamflow, in the period April to June, that are 50 to 100% larger than in the control scenario (Fig. 10). A considerable increase in October and November is also apparent in the Toce and Ticino rivers. The plausibility of these changes obtained using the downscaled RCM climate projections is thus related to the capacity of RCMs to resolve regional scale precipitation changes, and to the downscaling procedure to transfer these changes at the local scale, which are both highly uncertain (Themßl et al., 2011; Bordoy and Burlando, 2013a,b). 4. Conclusions We analyzed possible changes in streamflow regime considering climate change signals and climate internal variability. The latter is accounted for by using two stochastic generators. Streamflow natural variability was found to be likely larger than climate change signal expected

Please cite this article as: Fatichi S, et al, Does internal climate variability overwhelm climate change signals in streamflow? The upper Po and Rhone basin case studies, Sci Total Environ (2013), http://dx.doi.org/10.1016/j.scitotenv.2013.12.014

S. Fatichi et al. / Science of the Total Environment xxx (2013) xxx–xxx

Dora Baltea − Tavagnasco (3391 km2)

Dora di Valsavaranche − Fenille (139 km2)

200

10 8

150

[m3 s−1]

9

6 100 4 50 2 0

2

4

6

8

10

12

0

2

4

6

8

10

12

10

12

Matter Visp − Tasch (287 km2)

Rhone – Ulrichen (159 km2) 30

1992/2002−2010 25 2041−2050 ECHAM 2041−205020RegCM3 2041−2050 REMO

25

[m3 s−1]

20

15

15 10

10

5

5

0

2

4

6

8

10

12

Month

0

2

4

6

8

Month

Fig. 8. A comparison among simulated annual cycles of streamflow [m3 s−1] for 4 selected river sections belonging to regime II. The comparison is between the control scenario period (1992–2010 for the upper Rhone and 2002–2010 for the upper Po), and future scenarios 2041–2050 for the stochastic downscaling driven by RegCM3, ECHAM5, and REMO. The colored bands include simulations within the 10 and 90 percentiles of the stochastic ensembles. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

Ice Melt/Streamflow [−]

by the middle of 21st century in most of the river network of the upper Po and Rhone. This holds true for several streamflow metrics (mean, seasonality, annual maximum and minimum, etc.). In other words, changes (or

Rhone − Porte du Scex

0.8 0.6 0.4 0.2 0

0

100

200

300

Ice Melt/Streamflow [−]

DoY Po − Ponte della Becca

0.2 0.15 0.1 0.05 0

0

100

200

300

DoY Fig. 9. Seasonality of simulated contribution of ice melt, averaged over the upper Rhone (a) and Po (b) basins, to the total streamflow for the observational periods (1992–2008 for Rhone and 2002–2010 for Po).

the lack of) can be induced or masked by natural climate fluctuations to such an extent that these cannot be identified. This is true for changes up to ±20% for the mean flow and ±50% for annual streamflow extremes. Despite the fact that we used only three driving climate models, the presented results are accounting for a range of uncertainty similar (for air temperature) and superior (for precipitation) to the uncertainty interval obtainable from ensembles of multiple climate models (Fig. 3). This is happening because stochastic variability can cover (or can go beyond) the spread of multimodel ensembles, at least for a variable such as precipitation (Deser et al., 2012b; Fatichi et al., 2013a; Fischer et al., 2013), which is key for any hydrological analysis. Robust climate change signals were found in catchments where ice melt represents a significant fraction of the total runoff. Ice melt is a diminishing, not renewable resource, not only in a warming climate but also in the present one (Fatichi et al., 2013b). Reduced ice melt as a consequence of glacier retreat is mostly affecting basins at high elevations. For these basins the decrease in the total amount of available water and the modification of streamflow seasonality were found to be significantly larger than natural variability. Ice melt reduction has also the possibility to affect discharge at the entire upper Rhone and Po scale during the late summer season. This result can be explained by the fact that the fraction of streamflow contributed by ice melt is relatively large for both basins in August and September (Fig. 9). While in the upper Rhone this is mostly a consequence of the large extent of the glaciated area, in the upper Po it is a consequence of the relatively dry and warm Mediterranean summer where even small contributions from glacier sources can represent an important fraction of the total runoff for large parts of the river network. Another exception to the overwhelming role of climate internal variability is represented by regions where projected future

Please cite this article as: Fatichi S, et al, Does internal climate variability overwhelm climate change signals in streamflow? The upper Po and Rhone basin case studies, Sci Total Environ (2013), http://dx.doi.org/10.1016/j.scitotenv.2013.12.014

10

S. Fatichi et al. / Science of the Total Environment xxx (2013) xxx–xxx

Toce − Candoglia (1595 km2)

Ticino − Bellinzona (1612 km2) 200

150

[m3s−1]

150 100 100 50

0

50

2

4

6

8

10

12

0

2

Ticino − Turbigo (7039 km2)

4

6

10

12

Sionne − Sion (25 km2) 2

600

1992/2002−2010 2041−2050 ECHAM 2041−2050 RegCM3 2041−2050 REMO

1.5

[m3s−1]

8

400 1 200 0.5 0

2

4

6

8

10

12

Month

0

2

4

6

8

10

12

Month

Fig. 10. A comparison among simulated annual cycles of streamflow [m3 s−1] for 4 selected river sections belonging to regime III. The comparison is between the control scenario period (1992–2010 for the upper Rhone and 2002–2010 for the upper Po), and future scenarios 2041–2050 for the stochastic downscaling driven by RegCM3, ECHAM5, and REMO. The colored bands include simulations within the 10 and 90 percentiles of the stochastic ensembles. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

precipitation, especially in the February to April period, increases beyond the expected range attributable to natural climate fluctuations. These results are strongly related to the climate model used in the stochastic downscaling and therefore less robust than the projected change in ice melt contribution. However, according to RegCM3 and REMO, this precipitation increase seems to consistently happen in two subregions, the west upper Rhone and the Toce/Ticino river basins (Fig. 10). We can safely conclude that precise projections of streamflow changes up to the year 2050 would remain elusive and generally projections will not be much informative in most of the upper Po and Rhone river networks simply because the stochastic variability of precipitation would maintain the signal to noise ratio very low. In this context, it is important to clearly communicate to water managers, engineers, public authorities and stakeholders that adaptation measures to climate change have to be designed without the availability of precise information (Dessai et al., 2009), i.e., in conditions of large uncertainty. Arguably, the fact that climate internal variability overwhelms climate change signals in most of the river networks (with the exceptions of a limited number of river sections in regimes II and III) can be regarded as a rather positive news, because infrastructures already appropriately designed to account for stochastic variability are very likely to still serve their purposes in the next 40 years. At the same time, if climate variability was not accounted for properly, they can fail their requirements even without a change in climate. In this regard, this study highlights once more the fundamental role of stochastic analysis in engineering design, an analysis that is rarely pursued in practice. An obvious exception is represented by the operation of those reservoirs which are mostly filled with water of glacier origin. In the latter case, the effect of climate change can be rather severe leading to considerable impacts on hydropower production (Finger et al., 2012; Fatichi et al., 2013b). In light of an increasing rather than decreasing uncertainty in climate change predictions (Rowlands et al., 2012; Maslin and Austin,

2012), the “bottom-up” approach based on vulnerability analyses discussed recently (Wilby and Dessai, 2010; Pielke et al., 2012; Brown and Wilby, 2012) represents an appealing alternative and a way forward for handling climate change issues. This alternative approach requires (i) the determination of essential weak points and threats in a system, not only strictly climate related (Pielke et al., 2009); (ii) the use of stress tests to evaluate the system under a wide range of forcings; and (iii) an eventual comparison with climate projections. We argue that, since the vulnerability assessment requires a probabilistic quantification of the forcing distribution, in order to simply run the impact models with a finite number of possibilities (e.g., Steinschneider and Brown, 2013), studies such as the one presented here can be very useful in guiding such a type of analysis. We especially consider fundamental the generation of forcing through stochastic ensembles, possibly including all the known sources of uncertainty with a relative quantification. Using a single or few deterministic model driven scenarios, the typical approach in climate change impact studies found in literature, could be instead very misleading because it conveys a too certain information to final users and neglect natural climate variability. We finally conclude that a systematic adoption of “stochastic design” criteria, eventually enforced by legislation, should represent an important solution for engineering design and decision making under a changing climate. Conflict of interest The authors declare no conflict of interests for the submitted manuscript. Acknowledgments The present study is part of the ACQWA project (Assessing Climate change impacts on the Quantity and quality of WAter), funded within the seventh Framework Program of the European Union, contract 212250, www.acqwa.ch. Meteorological data were provided by

Please cite this article as: Fatichi S, et al, Does internal climate variability overwhelm climate change signals in streamflow? The upper Po and Rhone basin case studies, Sci Total Environ (2013), http://dx.doi.org/10.1016/j.scitotenv.2013.12.014

S. Fatichi et al. / Science of the Total Environment xxx (2013) xxx–xxx

MeteoSwiss, the Federal Office of Meteorology and Climatology, Arpa Piemonte, Compagnie Valdôtaine des Eaux (CVA), and by ARPA Lombardia. We thank all the people who collaborated in the project making available information useful for the hydrological simulations.

References Beniston M, Stoffel M, Hill M. Impacts of climatic change on water and natural hazards in the Alps: can current water governance cope with future challenges? Examples from the European “ACQWA” project. Environ Sci Pol 2011;14(7):734–43. http://dx.doi.org/10.1016/j.envsci.2010.12.009. Bordoy R. Spatiotemporal downscaling of climate scenarios in regions of complex orography. (Ph.D. thesis Nr. 21201) Switzerland: ETH Zurich; 2013. Bordoy R, Burlando P. Bias correction of regional climate model simulations in a region of complex orography. J Appl Meteorol Climatol 2013a;52:82–101. http://dx.doi.org/ 10.1175/JAMC-D-11-0149.1. Bordoy R, Burlando P. Stochastic downscaling of climate model precipitation outputs in orographically complex regions. part 2: downscaling methodology. Water Resour Res 2013b. http://dx.doi.org/10.1002/wrcr.20443. Bordoy R, Burlando P. Stochastic downscaling of precipitation to high-resolution scenarios in orographically complex regions. Part 1: Model evaluation. Water Resour Res 2013c. http://dx.doi.org/10.1002/2012WR013289. Brown C, Wilby RL. An alternate approach to assessing climate risks. Eos Trans AGU 2012;93(41):401–2. Burlando P, Rosso R. Effects of transient climate change on basin hydrology. 1. Precipitation scenarios for the Arno river, central Italy. Hydrol Process 2002;16: 1151–75. Burton A, Kilsby CG, Fowler HJ, Cowpertwait PSP, O'Connell P. RainSim: a spatial–temporal stochastic rainfall modelling system. Environ Model Software 2008;23:1356–69. http://dx.doi.org/10.1016/j.envsoft.2008.04.003. Burton A, Fowler HJ, Blenkinsop S, Kilsby CG. Downscaling transient climate change using a Neyman–Scott Rectangular Pulses stochastic rainfall model. J Hydrol 2010;381: 18–32. http://dx.doi.org/10.1016/j.jhydrol.2009.10.031. Christensen JH, Kjellström E, Giorgi F, Lenderink G, Rummukainen M. Weight assignment in regional climate models. Climate Res 2010;44:179–94. http://dx.doi.org/10.3354/ cr00916. Ciarapica L, Todini E. TOPKAPI: a model for the representation of the rainfall-runoff process at different scales. Hydrol Process 2002;16:207–29. Cowpertwait PSP. A spatial–temporal point process model of rainfall for the Thames catchment. J Hydrol 2006;330:586–95. Deser C, Knutti R, Solomon S, Phillips AS. Communication of the role of natural variability in future north American climate. Nat Clim Chang 2012a;2:775–9. Deser C, Phillips A, Bourdette V, Teng H. Uncertainty in climate change projections: the role of internal variability. Climate Dynam 2012b;38:527–46. http://dx.doi.org/ 10.1007/s00382-010-0977-x. Dessai S, Hulme M, Lempert R, Pielke Jr R. Do we need better predictions to adapt to a changing climate? Eos Trans AGU 2009;90(13):111. http://dx.doi.org/10.1029/ 2009EO130003. Fatichi S, Ivanov VY, Caporali E. Simulation of future climate scenarios with a weather generator. Adv Water Resour 2011;34:448–67. http://dx.doi.org/10.1016/ j.advwatres.2010.12.013. Fatichi S, Ivanov VY, Caporali E. On the search for vital details of climate change. In: Molnar TEP, Burlando P, editors. Proceedings of the 9th International Workshop on Precipitation in Urban Areas. Switzerland: Sankt Moritz; 2012a. p. 250–8. Fatichi S, Ivanov VY, Caporali E. A mechanistic ecohydrological model to investigate complex interactions in cold and warm water-controlled environments. 1. Theoretical framework and plot-scale analysis. J Adv Model Earth Syst 2012b;4: M05002. http://dx.doi.org/10.1029/2011MS000086. Fatichi S, Ivanov VY, Caporali E. A mechanistic ecohydrological model to investigate complex interactions in cold and warm water-controlled environments. 2. Spatiotemporal analyses. J Adv Model Earth Syst 2012c;4:M05003. http://dx.doi.org/ 10.1029/2011MS000087. Fatichi S, Ivanov VY, Caporali E. Assessment of a stochastic downscaling methodology in generating an ensemble of hourly future climate time series. Climate Dynam 2013a;40:1841–61. http://dx.doi.org/10.1007/s00382-012-1627-2. Fatichi S, Rimkus S, Burlando P, Bordoy R, Molnar P. Elevational dependence of climate change impacts on water resources in an alpine catchment. Hydrol Earth Syst Sci Discuss 2013b;10:3743–94. http://dx.doi.org/10.5194/hessd-10-3743-2013. Finger D, Heinrich G, Gobiet A, Bauder A. Projections of future water resources and their uncertainty in a glacierized catchment in the Swiss Alps and the subsequent effects on hydropower production during the 21st century. Water Resour Res 2012;48(W02521). http://dx.doi.org/10.1029/2011WR010733. Fischer AM, Weigel AP, Buser CM, Knutti R, Künsch HR, Liniger MA, et al. Climate change projections for Switzerland based on a Bayesian multi-model approach. Int J Climatol 2012;32(15):2348–71. Fischer EM, Beyerle U, Knutti R. Robust spatially aggregated projections of climate extremes. Nat. Clim. Chang 2013. http://dx.doi.org/10.1038/nclimate2051. Fowler HJ, Blenkinsop S, Tebaldi C. Linking climate change modelling to impacts studies: recent advances in downscaling techniques for hydrological modelling. Int J Climatol 2007;27:1547–78. http://dx.doi.org/10.1002/joc.1556. Fuhrer J, Beniston M, Fischlin A, Frei C, Goyette S, Jasper K, et al. Climate risks and their impact on agriculture and forests in Switzerland. Clim Change 2006;79:79–102. http://dx.doi.org/10.1007/s10584-006-9106-6.

11

Gabbi J, Farinotti D, Bauder A, Maurer H. Ice volume distribution and implications on runoff projections in a glacierized catchment. Hydrol Earth Syst Sci 2012;16:4543–56. Hawking E, Sutton R. The potential to narrow uncertainty in regional climate predictions. Bull Am Meteorol Soc 2009;90:1095–107. http://dx.doi.org/10.1175/ 2009BAMS2607.1. Hawking E, Sutton R. The potential to narrow uncertainty in projections of regional precipitation change. Climate Dynam 2011;37:407–18. http://dx.doi.org/10.1007/ s00382-010-0810-6. Hill Clarvis M, Fatichi S, Allan A, Fuhrer J, Stoffel M, Romerio F, et al. Governing and managing water resources under changing hydro-climatic contexts: the case of the Upper Rhone basin. Environ Sci Pol 2014. http://dx.doi.org/10.1016/j.envsci.2013.11.005. Huber M, Knutti R. Anthropogenic and natural warming inferred from changes in Earth's energy balance. Nat Geosci 2012;5:31–6. http://dx.doi.org/10.1038/ngeo1327. Huss M. Present and future contribution of glacier storage change to runoff from macroscale drainage basins in Europe. Water Resour Res 2011;47(W07511). http://dx.doi.org/ 10.1029/2010WR010299. Ivanov VY, Bras RL, Vivoni ER. Vegetation-hydrology dynamics in complex terrain of semiarid areas: 1. A mechanistic approach to modeling dynamic feedbacks. Water Resour Res 2008;44:W03429. http://dx.doi.org/10.1029/2006WR005588. Jacob D, Bärring L, Christensen OB, Christensen JH, deCastro M, Déqué M, et al. An inter-comparison of regional climate models for Europe: model performance in present-day climate. Clim Change 2007;81:31–52. http://dx.doi.org/10.1007/ s10584-006-9213-4. Katz RW, Parlange MB. Overdispersion phenomenon in stochastic modeling of precipitation. J Climate 1998;11:591–601. Knutti R. Should we believe model predictions of future climate change? Phil. Trans. R. Soc. A 2008;366:4647–64. http://dx.doi.org/10.1098/rsta.2008.0169. Knutti R, Furrer R, Tebaldi C, Cermak J, Meehl GA. Challenges in combining projections from multiple climate models. J Climate 2010;23:2739–58. Liu Z, Todini E. Towards a comprehensive physically-based rainfall-runoff model. Hydrol Earth Syst Sci 2002;6(5):859–81. Manning LJ, Hall JW, Fowler HJ, Kilsby CG, Tebaldi C. Using probabilistic climate change information from a multimodel ensemble for water resources assessment. Water Resour Res 2009;45(W11411). http://dx.doi.org/10.1029/2007WR006674. Maraun D, Wetterhall F, Ireson AM, Chandler RE, Kendon EJ, Widmann M, et al. Precipitation downscaling under climate change: recent developments to bridge the gap between dynamical models and the end user. Rev Geophys 2010;48:RG3003. Maslin M, Austin P. Climate models at their limit? Nature 2012;486:183–4. Meehl GA, Covey C, Delworth T, Latif M, McAvaney B, Mitchell JFB, et al. The WCRP CMIP3 multimodel dataset: a new era in climate change research. Bull Am Meteorol Soc 2007;88:1383–94. http://dx.doi.org/10.1175/BAMS-88-9-1383. Milly PCD, Betancourt J, Falkenmark M, Hirsch RM, Kundzewicz ZW, Lettenmaier DP, et al. Stationarity is dead: whither water management? Science 2008;319:573–4. http://dx.doi.org/10.1126/science.1151915. Pal JS, Giorgi F, Bi X, Elguindi N, Solomon F, Gao X, et al. Regional climate modeling for the developing world: the ICTP RegCM3 and RegCNET. Bull Am Meteorol Soc 2007;88: 1395–409. http://dx.doi.org/10.1175/BAMS-88-9-1395. Paschalis A, Molnar P, Fatichi S, Burlando P. A stochastic model for high-resolution space-time precipitation simulation. Water Resour Res 2013;49. http://dx.doi.org/ 10.1002/2013WR014437. Pielke Sr Roger A, Beven K, Brasseur G, Calvert J, Chahine M, Dickerson RR, et al. Climate change: the need to consider human forcings besides greenhouse gases. Eos Trans AGU 2009;90(45):413. http://dx.doi.org/10.1029/2009EO450008. Pielke S, Roger A. Sr, Wilby R, Niyogi D, Hossain F, Dairuku K, et al. Extreme events and natural hazards: the complexity perspective, 196. Geophys. Monogr. Ser.; 2012. p. 345–59. (chap. Dealing with complexity and extreme events using a bottom-up, resource-based vulnerability perspective, AGU). Prein AF, Gobiet A, Truhetz H. Analysis of uncertainty in large scale climate change projections over Europe. Meteorol Z 2011;20(4):383–95. http://dx.doi.org/10.1127/ 0941-2948/2011/0286. Räisänen J. How reliable are climate models? Tellus A 2007;59:2–29. http://dx.doi.org/ 10.1111/j.1600-0870.2006.00211.x. Rajczak J, Pall P, Schär C. Projections of extreme precipitation events in regional climate simulations for Europe and the alpine region. J Geophys Res 2013;118:3610–26. http://dx.doi.org/10.1002/jgrd.50297. Rigon R, Bertoldi G, Over TM. GEOtop: a distributed hydrological model with coupled water and energy budgets. J Hydrometeorol 2006;7(3):371–88. Rimkus S. TOPKAPI-ETH documentation and user guide. Technical. Zurich: ETH; 2013. [112 pp.]. Roeckner E, Bäuml G, Bonaventura L, Brokopf R, Esch M, Giorgetta M, et al. The atmospheric general circulation model ECHAM5. part i: model description, Tech. Rep. 349. Hamburg, Germany: Max Planck Institute for Meteorology; 2003. Rowlands DJ, Frame DJ, Ackerley D, Aina T, Booth BBB, Christensen C, et al. Broad range of 2050 warming from an observationally constrained large climate model ensemble. Nat Geosci 2012;5:256–60. Schädler B, Weingartner R. Alpine waters, chap. Impact of climate change on water resources in the Alpine regions of Switzerland, 6. Springer-Verlag Berlin Heidelberg; 201059–69. Schmidli J, Frei C, Vidale PL. Downscaling from GCM precipitation: a benchmark for dynamical and statistical downscaling methods. Int J Climatol 2006;26: 679–89. Schneider EK, Kinter III JL. An examination of internally generated variability in long climate simulations. Climate Dynam 1994;10:181–204. Steinschneider S, Brown C. A semiparametric multivariate, multisite weather generator with low-frequency variability for use in climate risk assessments. Water Resour Res 2013;49. http://dx.doi.org/10.1002/wrcr.20528.

Please cite this article as: Fatichi S, et al, Does internal climate variability overwhelm climate change signals in streamflow? The upper Po and Rhone basin case studies, Sci Total Environ (2013), http://dx.doi.org/10.1016/j.scitotenv.2013.12.014

12

S. Fatichi et al. / Science of the Total Environment xxx (2013) xxx–xxx

Sulis M, Paniconi C, Marrocu M, Huard D, Chaumont D. Hydrologic response to multimodel climate output using a physically based model of groundwater/surface water interactions. Water Resour Res 2012;48(W12510). http://dx.doi.org/10.1029/2012WR012,304. Themßl MJ, Gobiet A, Leuprecht A. Empirical–statistical downscaling and error correction of daily precipitation from regional climate models. Int J Climatol 2011;31:1530–44. http://dx.doi.org/10.1002/joc.2168. van der Linden P, Mitchell JFB. ENSEMBLES: climate change and its impacts: summary of research and results from the ENSEMBLES project. Tech. rep. Exeter, UK: Met Office Hadley Centre; 2009.

Viviroli D, Archer DR, Buytaert W, Fowler HJ, Greenwood GB, Hamlet AF, et al. Climate change and mountain water resources: overview and recommendations for research, management and policy. Hydrol Earth Syst Sci 2011;15:471–504. Wilby RL, Dessai S. Robust adaptation to climate change. Weather 2010;65(7):180–5. Wilby RL, Dawson CW, Barrow EM. SDSM—a decision support tool for the assessment of regional climate change impacts. Environ Model Software 2002;17(2): 145–57. Wilks DS. Interannual variability and extreme-value characteristics of several stochastic daily precipitation models. Agr Forest Meteorol 1999;93:153–69.

Please cite this article as: Fatichi S, et al, Does internal climate variability overwhelm climate change signals in streamflow? The upper Po and Rhone basin case studies, Sci Total Environ (2013), http://dx.doi.org/10.1016/j.scitotenv.2013.12.014