Impact of Climate Change on the Hydrology of St. Lawrence Tributaries

Journal of Hydrology 384 (2010) 65–83

Contents lists available at ScienceDirect

Journal of Hydrology

journal homepage: www.elsevier.com/locate/jhydrol

Impact of climate change on the hydrology of St. Lawrence tributaries

Claudine Boyer a,*, Diane Chaumont b, Isabelle Chartier c, André G. Roy a a Département de géographie, Université de Montréal, C.P. 6128, Succursale Centre-ville, Montréal (Québec), Canada H3C 3J7 b Ouranos, 550, Sherbrooke Ouest, Tour Ouest, 19e étage, Montréal (Québec), Canada H3A 1B9 c Institut de recherche d’Hydro-Québec (IREQ), 1800, boul. Lionel-Boulet, Varennes (Québec), Canada J3X 1S1 article info summary

Article history: Changes in temperature and precipitation projected for the next century will induce important modifica- Received 17 April 2009 tions into the hydrological regimes of the St. Lawrence tributaries (Quebec, Canada). The temperature Received in revised form 11 January 2010 increase anticipated during the winter and spring seasons will affect precipitation phase and conse- Accepted 20 January 2010 quently the snow/precipitation ratio and the water volume stored into snow cover. The impact on northern river hydrology and geomorphology will be significant. In this study we aim to assess the magnitude This manuscript was handled by K. of the hydrological alteration associated with climate change; to model the projected temporal shift in Georgakakos, Editor-in-Chief, with the the occurrence of winter/spring center-volume date; to assess the sensitivity of the winter/spring cen- assistance of Ashish Sharma, Associate ter-volume date to changes in climatic variables and to examine the latitudinal component of the pro- Editor jected changes through the use of five watersheds on both shores of the St. Lawrence. The study emphasizes changes in the winter and spring seasons. Projected river discharges for the next century Keywords: were generated with the hydrological model HSAMI run with six climate series projections. Three Gen- River eral Circulation Models (HadCM3, CSIRO-Mk2 and ECHAM4) and two greenhouse gas emissions scenarios Hydrology (A2 and B2) were used to create a range of plausible scenarios. The projected daily climate series were Climate change produced using the historical data of a reference period (1961–1990) with a perturbation factor equiva- Stream flow lent to the monthly mean difference (temperature and precipitation) between a GCM in the future for Variability three 30 year horizons (2010–2039, 2040–2069; 2070–2099) and the reference period. These climate Modeling projections represent an uncertainty envelope for the projected hydrologic data. Despite the differences due mainly to the GCM used, most of the hydrological simulations projected an increase in winter discharges and a decrease in spring discharges. The center-volume date is expected to be in advance by 22–34 days depending on the latitude of the watershed. The increase in mean temperature with the simultaneous decrease of the snow/precipitation ratio during the winter and spring period explain a large part of the projected hydrological changes. The latitude of the river governed the timing of occurrence of the maximum change (sooner for tributaries located south) and the duration of the period affected by marked changes in the temporal distribution of discharge (longer time scale for rivers located at higher latitudes). Higher winter discharges are expected to have an important geomorphological impact mostly because they may occur under ice-cover conditions. Lower spring discharges may promote sedimentation into the tributary and at their confluence with the St. Lawrence River. The combined effects of modifications in river hydrology and geomorphological processes will likely impact riparian ecosystems. Ó 2010 Elsevier B.V. All rights reserved.

Introduction evaporation from lakes due to a rise in temperature explains a large part of the projected reduction (Croley, 2003). Shifts in the The hydrological regime of northern rivers could be severely timing and amount of input runoff are also expected to occur modiﬁed in response to the anticipated changes in temperature (Mortsch and Quinn, 1996). Downstream of Lake Ontario, seasonal and precipitation during the present century. For the Great- changes in the discharge of the St. Lawrence River may be accentu- Lakes-St. Lawrence watershed (USA and Canada), a reduction rang- ated or attenuated by the water regulation plan in order to modu- ing between 4% and 24% of the mean annual discharge is projected late the temporal variation in the Great Lakes levels. for the next 90 years as a consequence of current scenarios of cli- Winter and spring seasons are particularly vulnerable to mate change (Croley, 2003). For this large watershed, increased changes in air temperature. Warmer temperature during the winter season can increase the number of days with air temperature

* Corresponding author. Tel.: +1 514 343 8035; fax: +1 514 343 8008. above zero Celsius resulting in more frequent rain events. These E-mail address: [email protected] (C. Boyer). events will contribute to an increase of winter runoff and not to

0022-1694/$ - see front matter Ó 2010 Elsevier B.V. All rights reserved. doi:10.1016/j.jhydrol.2010.01.011 66 C. Boyer et al. / Journal of Hydrology 384 (2010) 65–83 the accumulation of a snowpack (Whitfield et al., 2003). The rela- shift in the occurrence of spring peak discharges (represented by tive amount of precipitation falling as rain or snow will directly af- the winter/spring center-volume date); (3) assess the sensitivity fect water supplied from the snowpack in the spring and the of the winter/spring center-volume date to changes in climatic amplitude and timing of river flows during the winter and spring variables and (4) identify similarities and differences among the (Hodgkins et al., 2003). Cooley (1990) has suggested that changing St. Lawrence tributaries. The paper presents historical (1932– mean air temperature by 2–4 °C may have a significant impact on 2005) and projected (2020s, 2050s and 2080s) climatic and hydro- the accumulation and melt of a snowpack depending on the origi- logical data for five tributaries with a particular focus on the winter nal temperature regime of the site. These changes will alter the and spring period. This research significantly adds to other recent hydrological regime. studies in Québec and Eastern Canada because the selected water- Modification of winter and spring streamflows has been ob- sheds cover a latitudinal range from 43.2°N to 49.1°N on both the served in southeastern Canada and northeastern USA during the south and north shores of the St. Lawrence. The selection of rivers course of 20th century (Hodgkins et al., 2003; Hodgkins and Dud- spans a current seasonal mean air temperature gradient for the ley, 2006; Whitfield and Cannon, 2000; Zhang et al., 2001). winter and spring period from 4 °Cto0°C. Changes toward earlier freshet (spring thaw resulting from snow and ice melt in rivers located in the northern latitudes) were found Methodology to be significant for most of these areas. Hodgkins and Dudley (2006) also found that river flows in January, February and March All located along the St. Lawrence fluvial corridor between show a tendency to increase from 1953 to 2002 in northeastern Montreal and Quebec City, the studied tributaries are the Richelieu USA. Conversely, river flows in April and May show a relative de- and St-François rivers on the south shore and the Yamachiche, St- crease during the same period. These changes are attributed to Maurice and Batiscan rivers on the north shore (Fig. 1). These long term changes in temperature and their impact on the phase watersheds are mostly located in low relief areas and cover close (snow or rain) of precipitation. Decline in the annual ratio of snow to 6° of latitude. The rivers differ in their hydrology, sedimentology to total precipitation has been reported from many climatologic and dynamics and they are representative of the diversity of tribu- stations in New-England from 1948 to 2000 with the most impor- taries along the St. Lawrence. Except for the Yamachiche River, all tant decrease occurring after 1975 (Burakowski et al., 2008; Hun- watersheds are exploited for hydro electricity or influenced by tington et al., 2004). This negative trend has been partly linked multiple dams used for flood control, water intake or recreational with positive North Atlantic Oscillation (NAO) anomalies index activities. Data concerning the management plan of these struc- which is associated with mild winters (in eastern USA) and low tures are, however, not available and cannot be accounted for in snowfall to rain ratio. Large-scale atmospheric and oceanic oscilla- the hydrological simulations. Exploitation of the rivers for hydro tions (e.g. North Atlantic Oscillation, Pacific Decadal Oscillation) electricity and flood control began before 1950 and around 1964. account for most of the climate natural variability by modulating The impact of these structures on the natural regime of the river precipitation and temperature regimes through the regulation of is low for the Batiscan and Richelieu and moderate for the St- the number and intensity of significant weather events particularly François. The natural hydrological regime of the St-Maurice River during the winter and early spring. These oscillations influence the has been substantially modified by water management for hydro snowpack variability and the timing and magnitude of flood peaks electricity. To reduce the impact of this hydrological control on at a decade scale (Cayan, 1996; Hartley and Keables, 1998; Jain and the St-Maurice, we have elected to study the response of a smaller Lall, 2000, 2001; Thompson and Wallace, 2001). Global warming watershed, LaGabelle, instead of the whole basin. The LaGabelle effects will be superimposed on NAO or other large-scale oscilla- watershed is used by Hydro-Quebec (Québec national hydro elec- tions and it remains uncertain whether global climate warming tricity company) to study the natural response of the drainage ba- will influence the variability of those oscillations (Hurrell et al., sin to meteorological variations. 2006; Visbeck et al., 2001). Changes in temperature and precipitation and the shift in winter precipitation from snow to rain will be crucial for the hydrolog- Hydrological modeling ical regime of St. Lawrence tributaries. For the future periods (2010–2039, 2040–2069 and 2070–2099), a few studies in Québec Hydrological simulations were performed with the HSAMI have already suggested an increase in winter flow and decrease in model (Bisson and Roberge, 1983; Chaumont and Chartier, spring flow compared to a historical reference period (Fortin et al., 2005; Fortin, 2000). This model is a lump rainfall (rain-and-snow) 2007; Minville et al., 2008; Quilbé et al., 2008). Higher winter runoff model. It is a discrete time conceptual model containing flows, lower spring flows and changes in the timing of spring run- three linear reservoirs (snow cover, surface water (surface runoff off can also have important impacts on fluvial processes, water and base discharge), unsaturated and saturated zones) in cascade management and on riparian and aquatic ecosystems. The ampli- which generate impulses filtered by two hydrograph units. Snow tude, duration and timing of spring floods play a critical role on accumulation (following a degree-day approach), snow melt, soil the structure and diversity of aquatic ecosystems (Toner and Ked- freezing and thawing, evapotranspiration (estimated from daily dy, 1997). Stronger winter floods can also substantially modify the maximum and minimum temperatures) and vertical and horizon- physical characteristics of habitats, enhance river channel erosion tal transit of water are simulated by the model with a system of and alter stream conditions for winter spawning fish species. These equations and empirical parameters which are adjusted during effects could be enhanced as winter floods could occur under ice- the calibration of the model. Simulations are carried out with a cover conditions. time step of one day. HSAMI model is simple and easy to use This study is part of a larger project that aims to model the mor- for the estimation of potential impacts of climate change on water phological and sedimentological response of St. Lawrence tributar- resources. This model has been used for more than twenty years ies to the anticipated environmental changes (hydrological and by Hydro-Quebec over the Québec province to predict runoff for base level drop changes). The paper focuses on future changes in their reservoirs. It has been largely tested and successfully applied the hydrological regimes of the St. Lawrence tributaries as a result over the same region and more northern watersheds (Chartier, of projected climate changes. The specific objectives of the study 2006; St-Hilaire et al., 2003). It requires a small amount of input are to: (1) analyze changes in the simulated river discharges at data and optimization of the parameters is done automatically the annual to monthly scales; (2) quantify the projected temporal using the shuffled complex evolution method (Duan et al., C. Boyer et al. / Journal of Hydrology 384 (2010) 65–83 67

Fig. 1. Location of the studied river basins. The GCM grids are represented HadCM3 (black solid lines), ECHAM4 (black dot lines) and CSIRO-Mk2 (black dash lines). The reference region, for which monthly projected changes were calculated, is represented by the red dash line.

1992). The model uses input data (minimum and maximum tem- calibrate the model and to build the simulated daily discharge data perature, rain and snow) that are averaged over the basin. The for the reference period. These time series were created by spatially model distinguishes the phase of precipitation (rain or snow). It averaging the daily observed data taken from several meteorological was thus necessary to carry out a partition between snow and stations. We have selected meteorological stations located within a rain according to the average temperature. A linear transfer was radius distance of 50 km from the tributary. Observed meteorologi- applied when the daily mean temperature was between 2 °C cal data were extracted from the Environment Canada network and and +2 °C. All precipitation was converted into snow at 2 °C the American NOAA network (for the Richelieu). Thiessen polygons and kept as rain at 2 °C. were used to calculate area-weighted average temperature and pre- Calibration and validation of HSAMI was carried on for each wa- cipitation for each watershed (Heywood et al., 2006). tershed over the reference period, 1961–1990 (Table 1). For this We have evaluated the quality of the simulated discharges with analysis, we have used discharges recorded by the Hydat network the Nash–Sutcliffe coefﬁcient, E (Nash and Sutcliffe, 1970). This (Environment Canada and provincial partnership) at the most coefﬁcient sums up the daily square differences between observed downstream gauging station. The period used for model calibration and simulated data over a year (Eq. (1)). on each river was dictated by the availability of data. For the P Yamachiche River, where no gauging station exists, we have used T ðQ t Q t Þ2 E ¼ 1 Pt¼1 obs sim ð1Þ data recorded on a larger neighboring watershed with a drainage T t 2 t¼1ðQ obs Q obsÞ area ratio to estimate historical mean daily discharges.

For each studied watershed, a unique mean time series for tem- where t is the day, Qobs and Qsim are observed and simulated dis- perature (maximum and minimum daily temperatures) and for pre- charge, respectively. The main priority during the calibration pro- cipitation (daily amounts of rain and snow) were used in HSAMI to cesses was to minimize the annual and spring peak volume 68 C. Boyer et al. / Journal of Hydrology 384 (2010) 65–83

Table 1 Drainage area of the studied rivers and period of calibration and validation of the hydrological data.

River Total area of the Area of Hydrological data Calibration perioda Validation period Nash coefﬁcient drainage basin (km2) calibration (km2) (observed and estimated) period Yamachiche 269 269 1932–2005 1960–1972 1980–1990 0.842 Batiscan 4,700 4,580 1932–2005 1950–1975 1976–1990 0.849 St-Maurice (LaGabelle) 43,250 716 1932–2005 1974–1987 1988–1998 0.827 St-François 10,180 9,610 1932–2005 1960–1975 1976–1990 0.831 Richelieu 23,720 22,000 1932–2005 1960–1975 1976–1990 0.789

a It is important to note the calibration and validation period may differ from one basin to the other depending of the availability of the data.

differences. We have considered the model to be well calibrated mate models themselves and the GHG scenarios used (Houghton when a threshold value of 0.75 was obtained or exceeded (Table 1). et al., 2001). Hydrological simulations for the future periods were generated with the calibrated model and the projected climate variable series. Projected climate variable time series: selection and justiﬁcations of the downscaling method In order to apply the GCMs at a regional scale and create future Modeling climate variables climate variable time series for local hydrological impact assessment, three main approaches can be used: (1) the dynamical Climate models and GHG scenarios downscaling method (Regional Climate Model) with bias correc- The output of three GCMs driven by two greenhouse gas emission tions; (2) the statistical downscaling method and (3) the perturba- (GHG) scenarios (A2 and B2) were used to construct climate variable tion (or delta) method. The advantages and disadvantages of each time series over three future periods (2010–2039, 2040–2069 and approach will not be discussed in details here. However, details 2070–2099). The three chosen climate models are: ECHAM4 (Roeck- regarding the justiﬁcation of our choice of the perturbation method ner et al., 1992), CSIRO-Mk2 (Hirst et al., 1996, 1999) and HadCM3 for this study are given. (Gordon et al., 2000; Pope et al., 2000)(Table 2). From the six climate In the perturbation approach, the observed climate variables models published by the IPCC Data Distribution Centre associated during the reference period are scaled by the monthly anomalies with the Third Assessment Report (Giorgi et al., 2001), these three calculated, for the temperature and precipitation amounts, be- models are the only ones that include a multilayer surface scheme tween the future and reference periods for a given GCM simulation. which allows for the minimization of the bias in the reconstruction The perturbation method assumes that: (1) the biases of the GCM and simulation of the surface processes (Crossley et al., 2000; Verse- are similar during the reference and the future periods and (2) ghy, 1996). This is of paramount importance for simulating climate temporal variability (daily to inter annual) of the observed climate change in regions subjected to rigorous winters. variables during the reference period is maintained for the simu- The selection of models covers a wide spectrum of temperature lated series. This method is simple and can be used to generate a and precipitation anomalies. In southern Québec, the ECHAM4 wide range of plausible climate scenarios from a group of global model is usually projecting very little changes in precipitation climate models which is an important aspect of this study. It is and a moderate increase in mean temperature compared to the used as an early phase method to assess the sensitivity of the rivers other models. The HadCM3 model projects the largest increase in to changes in the mean values (annual to monthly scale) of climate precipitation and the lowest increase in mean temperature. The variables (Diaz-Nieto and Wilby, 2005). It has the advantage of values obtained from the CSIRO-Mk2 model are usually in between being stable and robust (Graham et al., 2007) and it has been used those two models for precipitation and show the highest tempera- in many studies (Andréasson et al., 2004; Hay et al., 2000; Hewis- ture increase. The use of an intermediate model gives less weight ton, 2003; Merritt et al., 2006; Minville et al., 2008; Prudhomme to the two models that represents the extreme of the studied spec- et al., 2002, 2003). This provides a strong basis for comparison of trum for one or the other climate variables. However, the use of our results with previous studies. Despite the limitations of the more models would have given a better estimate of the uncertainty perturbation method (e.g. the analysis of extreme statistics, such associated with the projected climates variables (Johnson and as summer extreme runoff is not appropriate), it is a general meth- Sharma, 2009). The number of models and the variety of models od that can account for a wide range of GCMs and consequently al- chosen was partly dictated by the main objective of the larger low the assessment of the uncertainty linked with mean GCMs study which is to model river response to hydrological and base le- projected changes which represent a larger contribution to climate vel changes as a result of projected climate changes. This third le- projection uncertainties compared to the downscaling methods vel of modeling has constrained the number of models that could (Boé et al., 2009; Chiew et al., 2009). The hydro-indicators analyzed effectively be used in the study. in this study are mostly linked with snow accumulation. The per- The climate models and scenarios were selected to meet the turbation method is suitable to estimate these indices because objectives of producing a range of plausible climate scenarios in the preservation of the frequency distribution of the precipitation terms of the amplitude of change in precipitation and temperature data has only a small impact for snow accumulation. Also, the sta- and of capturing a part of the uncertainties associated with the cli- tistical analysis of the data will mainly focus on changes in the

Table 2 Global climate models and scenarios of greenhouse gas concentration used to generate climate scenarios.

GCM Research center and country Resolution (lat long) SGHG CSIRO-Mk2 Australia’s Commonwealth Scientific and Industrial Research Organization, Australia 3.2° 5.6° A2 and B2 ECHAM4 Max Planck Institute for Meteorology, Germany 2.8° 2.8° A2 and B2 HadCM3 UKMO United Kingdom Meteorological Office, United Kingdom 2.25° 3.75° A2b and B2b C. Boyer et al. / Journal of Hydrology 384 (2010) 65–83 69 mean values in order to be consistent with the perturbation Effect of the hydrological model: context of climate change studies method. Chartier (2006) has assessed the effect of the hydrological mod- Regional Climate Models were not chosen for two main reasons: el for studies of climate change impacts by comparing the results of (1) only one RCM simulation was available over the region of inter- the HSAMI model (averaging water volume over the watershed) est at the time of the study thus severely reducing the span of con- with those of the distributed model Hydrotel (Fortin et al., 1995) ditions needed for this research; (2) the relatively low relief of the both calibrated for a sub-basin (10 000 km2) of the Gatineau river studied regions and its continental position reduce the orographic (Quebec). The simulation results indicated that the effect of the and maritime effects on precipitation. The overall added value of hydrological model used is less important than the effect attrib- RCMs is small for areas weakly influenced by topographical forcing uted to the choice of the climatic model (Chartier, 2006). These re- (Denis et al., 2003; Feser, 2006) and concerns mostly the analysis of sults support other studies showing that the choice of a frequency distributions and of high-order statistics of climatic vari- hydrological model has a relatively minor impact on the results ables (Laprise, 2008). of hydrological simulations based on climatic projections (Bates As dynamical downscaling, statistical downscaling allows et al., 2008; Kay et al., 2006). However, the hydrological model examination of changes in the temporal structure of the future cli- may have a greater impact in contrasted watershed (e.g. alpine mate (e.g. occurrence of precipitation). The statistical downscaling foreland watersheds) as shown by Ludwig et al. (2009) for the approach involves the development of statistical relationships be- Southern Bavaria region in Germany. tween large-scale atmospheric variables and local variables (Wilby et al., 1998). Many types of statistical techniques can be used (Fow- Effect of calibration: observed vs. simulated hydrographs at a seasonal ler et al., 2007; Giorgi et al., 2001; Wilby et al., 2004) and different time scale methods have been compared over Europe (see for example Hay- Using the optimal combination of parameters, the calibrated lock et al., 2006). The statistical relations have to be carefully de- HSAMI model generally performs well to reproduce the observed fined for each specific site and require a large high-quality average annual runoff and the spring peak for the reference period dataset for calibration and validation (Diaz-Nieto and Wilby, (Fig. 2). However, the quality of the simulation is variable at a 2005). In view of the objectives of our study and of the recommen- monthly and seasonal scale. The comparison between the simu- dations from Diaz-Nieto and Wilby (2005), the added value of sta- lated mean seasonal discharges with the observed data show that tistical downscaling techniques for a global assessment (analysis of the simulated mean winter and spring discharges are lower than indices associated with snow accumulation, first phase impact the observed discharges for most rivers except for the Richelieu study and analysis of changes in mean values) of climate change and LaGabelle (winter only) (Table 3). The correlation between ob- impacts should be small compared to the perturbation method. served and simulated discharges are generally lower for January (r = 0.4–0.8). The simulated mean summer discharges are higher Projected climate variable time series: production of the daily series than the observed values while the simulated mean fall discharges In order to obtain projected climate variable time series, we are slightly higher than the observed ones for the Richelieu, Bati- have calculated the monthly differences (or anomalies) between scan and LaGabelle rivers and lower for the St-François and the future and the reference periods for temperature and precipita- Yamachiche rivers. The difference between mean seasonal simu- tion over a region covering at least four grid points on each GCM lated and observed discharges are generally lower than 35% except grid. The region is delimited by the latitudes 43°N and 49.5°N for the mean winter discharge on the Yamachiche river, where and by the longitudes 69°W and 77°W(Fig. 1). The monthly mean simulated discharges are much lower than the observed discharges regional differences are calculated using spatial averaging for each (77%), and for the mean summer discharge of the Richelieu where grid point (area-based weight method). Regional differences ob- simulated discharges are higher than the observed discharges tained with at least four grid points give more physically represen- (54%). For the Yamachiche, the larger differences during winter tative results than value calculated with the closest grid point may arise from the overestimation of observed winter discharge (Wilby et al., 2004). The monthly mean regional differences were using the drainage area-ratio method. The over- or underestima- calculated for the six selected GCMs projections between the refer- tion in seasonal (and monthly) discharges will also be present in ence period (1961–1990) and three future horizons: 2010–2039, the projected hydrological simulations for the next century. These 2040–2069 and 2070–2099 hereafter respectively referred to as percentages represent uncertainty zones under which interpreta- 2020s, 2050s and 2080s. tion of future changes will be hazardous. We have created the projected climate series in three steps: (1) the mean daily precipitation and the mean daily minimum and Effect of uncertainty in climate variable data sets maximum temperatures simulated by each GCM were averaged Uncertainty in modeling precipitation, evapotranspiration and over the region of interest at a monthly scale for the reference per- climate variability represents a large part of the error in hydrolog- iod (30 years) and for each future horizon; (2) the anomalies were ical projections in climate change studies (Bates et al., 2008). computed for minimum and maximum temperatures (in °C) and Therefore, changes in modeled river hydrology using projected cli- for precipitation (ratio in %) by comparing the future horizons with mate data sets have to be carefully appraised (Kay et al., 2006). the reference period; (3) for each watershed, the anomalies were Compared to the observations during the reference period, Had- added to the observed daily minimum and maximum temperature CM3 and ECHAM4 offer a better potential for the simulation of the during the reference period while the ratio was applied for the dai- variables of interest in Southern Quebec. CSIRO-Mk2 shows an ly precipitation. The 18 sets (six for each horizon) of projected cli- overestimation of precipitation especially during summer (data mate time series generated were integrated into the calibrated not shown). Although the quality in reproducing the present day hydrological model to produce hydrological scenarios for the fu- climate does not necessarily imply a more accurate simulation of ture at a daily time step. future climate change, it is expected from these results that the adequacy of the models to response to climate forcing is higher Quality of the hydrological simulations (Giorgi et al., 2001). The trend in precipitation change for the next century when The quality of the hydrological simulations for the future is sen- compared to the reference period shows a great seasonal variabil- sitive to the selected hydrological model, to the calibration of the ity (Fig. 3). For the winter, all simulations project an increase in model and most importantly to the climate variables projections. precipitation; for the spring, HadCM3 and CSIRO-Mk2 models 70 C. Boyer et al. / Journal of Hydrology 384 (2010) 65–83

800.0 (a) 900.0 (b) 700.0 800.0

/s) 700.0 3 600.0 /s) 3 600.0 500.0 500.0 400.0 400.0 300.0

Discharge (m 300.0 200.0 Discharge (m 200.0 100.0 100.0 0.0 0.0 1 51 101 151 201 251 301 351 1 51 101 151 201 251 301 351 Day of the year Day of the year

450.0 (c) 35.0 (d) 400.0 30.0 350.0 /s) /s)

3 25.0 3 300.0 250.0 20.0 200.0 15.0 150.0

Discharge (m 10.0 Discharge (m 100.0 50.0 5.0 0.0 0.0 1 51 101 151 201 251 301 351 1 51 101 151 201 251 301 351 Day of the year Day of the year

90.0 (e) 80.0 70.0

/s) 60.0 3 50.0 40.0 30.0 20.0 Discharge (m 10.0 0.0 1 51 101 151 201 251 301 351 Day of the year

Fig. 2. Mean hydrograph for observed and simulated data for the reference period. (a) St-François; (b) Richelieu; (c) Batiscan; (d) Yamachiche; (e) St-Maurice. The blue lines represent the simulated results and the red lines the observed data. (For interpretation of the references to colour in this ﬁgure legend, the reader is referred to the web version of this article.)

Table 3 Comparison of seasonal mean discharges observed and simulated for the reference period (1961–1990). River Qobs Qsim Q sim Q obs 100 Qobs Qsim Q sim Q obs 100 Q obs Q obs m3/s m3/s % m3/s m3/s % Winter Spring Richelieu 304.34 363.53 19.45 680.09 695.39 2.25 Batiscan 38.96 36.73 5.72 178.85 171.43 4.15 St-Françoisa 164.72 126.92 22.95 357.94 344.83 3.66 Yamachiche 3.26 0.76 76.66 15.41 13.40 13.05 St-Maurice (LaGabelle sub-basin) 4.33 4.75 9.81 29.11 28.79 1.10 Summer Autumn Richelieu 237.14 364.43 53.68 283.62 321.57 13.38 Batiscan 70.72 91.96 30.03 80.44 84.50 5.05 St-Françoisa 120.84 144.50 19.57 190.43 166.00 12.83 Yamachiche 3.51 3.86 10.05 5.84 4.59 21.47 St-Maurice (LaGabelle sub-basin) 6.79 9.79 44.23 8.98 9.34 4.01

a Period: 1974–1990. anticipate an increase in precipitation while ECHAM4 (A2 and B2) The temperature trend is clearer. All simulations project an in- curves project only small changes; for the summer and fall, small crease in seasonal maximum and minimum temperatures com- precipitation changes are projected. pared to the reference period (Fig. 3). Discrepancies between C. Boyer et al. / Journal of Hydrology 384 (2010) 65–83 71

Δ Tmax (ºC) ΔTmin (ºC) Δ P (%) 10 10 40

8 8 30 6 6 Winter (DJF) 20 4 4 10 2 2 0 0 0 1980 2020 2060 1980 2020 2060 1980 2020 2060 -10

10 10 40

8 8 30 6 6 Spring (MAM) 20 4 4 10 2 2 0 0 0 1980 2020 2060 1980 2020 2060 1980 2020 2060 -10

10 10 40

8 8 30 6 6 Summer (JJA) 20 4 4 10 2 2 0 0 0 1980 2020 2060 1980 2020 2060 1980 2020 2060 -10

10 10 40

8 8 30 HadCM3 A2b Fall (SON) 6 6 HadCM3 B2b 20 4 4 CSIRO -Mk2A2 10 2 2 CSIRO-Mk2 B2 0 0 0 ECHAM4 A2 1980 2020 2060 1980 2020 2060 1980 2020 2060 -10 ECHAM4 B2

Fig. 3. Seasonal projected changes in temperatures (minimum and maximum) and precipitations over the studied region for six climate projections. The changes are calculated compared to the reference 1961–1990. The 30 years moving average is represented.

simulations are larger for the winter and spring seasons mainly for that the monthly means runoff or discharge given by the pertur- the last period indicating that uncertainties in the projected bation method are generally comparable (amplitude and direc- changes are slightly more important for this period. For the sum- tion of change) with other methods like the quantile mapping, mer and fall, the difference between models stays constant weather typing and the daily scaling (Boé et al., 2009; Chiew through time. et al., 2009; Maurer and Hidalgo, 2008; Segui et al., 2009). For all seasons, most of the simulations used in this study are included within the 95% probability density surface limits defined for temperature and precipitation changes (Fig. 4). These limits Analysis were obtained by the more recent GCM simulations published in phase with the IPCC Fourth Assessment Report (Meehl et al., From the hydrological data (simulated for the reference and 2007). The CSIRO-Mk2 is occasionally outside this limit. The future periods), we have calculated the center-volume date for change projected during spring obtained from this model is always the winter/spring period (WS) for each year. The WS center-vol- higher than the 95% density probability because of its high temper- ume date (WS CV date) corresponds to the date at which half of ature change projections. the total water volume cumulated over the January 1 to May 31 For northern watersheds like those studied here, the state of period is reached. This approach is more robust than the identi- precipitation during late fall and winter is also a source of uncer- fication of peak discharge which can be associated with a single tainty for hydrological projected data. Therefore, the frequency of flood occurring before or after the bulk of high flows (Hodgkins rain events during winter is highly sensitive to both errors in tem- et al., 2003). It is also expected to be less sensitive to the modeling perature and precipitation projections. technique used to simulate the daily discharges. The center-volume The downscaling method is also a source of uncertainties. date is closely linked with the discharge temporal distribution However, some studies have found that the uncertainties related during the period of interest (Court, 1962). Since this value is fre- to the downscaling and bias correction methods is lower than quently used in other studies, comparison of the results will be those related to the GCMs (Boé et al., 2009; Seguí et al., 2009). facilitated. In this study, the uncertainty due to the downscaling method We have analyzed the long term trend of the annual WS CV was not studied as we have only applied the perturbation meth- date and WS mean temperature (WS Tmean) and WS snow/pre- od. Despite the limitations of this technique, studies have shown cipitation ratio (WS S/P) only for historical data. The slope of the 72 C. Boyer et al. / Journal of Hydrology 384 (2010) 65–83

Winter_2010-2039 Winter_2040-2069 Winter_2070-2099 40 40 40 20 20 20 0 0 0

P/P (%) P/P -20 -20 -20 -40 -40 -40 0 2 4 6 8 10 0 2 4 6 8 10 0 2 4 6 8 10

Spring_2010-2039 Spring_2040-2069 Spring_2070-2099 40 40 40 20 20 20 0 0 0

P/P (%) P/P -20 -20 -20 -40 -40 -40 0 2 4 6 8 10 0 2 4 6 8 10 0 2 4 6 8 10

Summer_2010-2039 Summer_2040-2069 Summer_2070-2099 40 40 40 20 20 20 0 0 0

P/P (%) P/P -20 -20 -20 -40 -40 -40 0 2 4 6 8 10 0 2 4 6 8 10 0 2 4 6 8 10

Fall_2010-2039 Fall_2040-2069 Fall_2070-2099 40 40 40 20 20 20 0 0 0

P/P (%) P/P -20 -20 -20 -40 -40 -40 0 2 4 6 8 10 0 2 4 6 8 10 0 2 4 6 8 10 Temperature (°C) Temperature (°C) Temperature (°C)

Fig. 4. Projected seasonal changes in temperature and precipitation in the region of interest. The results of 130 projections are represented in gray. The inner to the outer blue ellipses represent the surface with a probability density of 50%, 75% and 95% respectively; the surface is derived from the covariance matrix. The projected changes for HadCM3 (blue), CSIRO-Mk2 (red) and ECHAM4 (green) used in this study are indicated, circles and squares are respectively for A2 and B2 GHG emissions scenarios.

temporal trend was calculated with the Sen’s slope non-para- Results metric estimator. The Kendall-Tau non-parametric test was used to evaluate the significance of the trend. The Sen’s slope estima- Impact of climate changes on mean temperature, snow/precipitation tor and the Kendall-Tau test are not sensitive to the presence of ratio and total precipitation during the winter/spring period outliers and do not require any particular assumption regarding the data distribution (Kendall, 1938,1975; Sen, 1968). We also WS mean temperature employed the Kendall-Tau correlation coefficient (non-paramet- The temporal variation of the WS mean temperature (WS ric correlation coefficient) to examine the correlation between Tmean) for each tributary is analyzed to identify the decade at climatic variables and the WS CV date. Trend analysis was not which the 0 °C threshold is crossed. Fig. 5 shows that the WS performed for the projected climatic data since the time series Tmean crossed the 0 °C threshold at different decades depending inside each 30 year periods replicated the temporal variability on the latitude of the river. For the reference period, the 0 °C of the reference period. threshold is crossed in the mid-1970 (lowess curve) for the Riche- For the graphical representation of the series, we have drawn lieu which is the southernmost watershed. During that period, the the ‘‘locally-weighted scatterplot smoothing” (lowess technique) St-François is at ffi2 °C. For the St. Lawrence north shore tributar- curve. It is a robust locally weighted polynomial regression model ies, the WS Tmean is under 2 °C for all the reference period. The for which outliers have less influence (Cleveland and Devlin, 1988). trend in the WS Tmean is significant for the St-François and Riche- The smoothing window used in this study is 10 years. This window lieu rivers (Kendall-Tau coefficient = 0.35 and 0.33; p < 0.01).The size approximates the temporal frequency of large-scale oscilla- average increase in temperatures between 1970 and 1980 is con- tions (NOA) that have an impact on winter temperature and pre- sistent with observations made for the North-American East coast- cipitation and as a result on river hydrology. The central point of al region (Hayhoe et al., 2007; Huntington et al., 2004). the window has the largest weighting, and the points towards For the future, all models projected that WS Tmean for the the edge of the window have successively less influence on the Richelieu will be above 0 °C. For the other tributaries, the time at fit. The smoothed line follows peaks and troughs in the original which this threshold will be crossed differs. For the St-François, data series. WS Tmean will cross the threshold of 0 °C during the first horizon. In the presentation of the results, we frequently used the mean For the Yamachiche, St-Maurice and the Batiscan the threshold will value of the six sets of simulations with an indicator of the variabil- be crossed during the second horizon. For the last period, WS ity between these simulations. This choice simplifies the visual Tmean will be over 2 °C, for all five watersheds. The difference be- presentation of the data and is justified by the fact that all simula- tween simulations is larger for this period. tions generally point in the same direction despite the variability in the amplitude of the predicted changes. It also reduces the uncer- WS snow/precipitation ratio tainty associated with the projection obtained from one specific Huntington et al. (2004) have shown that the increase in the simulation. annual Tmean coincides with a reduction of the ratio snow/total C. Boyer et al. / Journal of Hydrology 384 (2010) 65–83 73

10.0 10.0 (a) (b) 8.0 8.0 6.0 6.0 4.0 4.0 2.0 2.0 0.0 0.0 1950 1980 2010 2040 2070 2100 1950 1980 2010 2040 2070 2100 - 2.0 - 2.0 - 4.0 - 4.0 WS Tmean (°C) -6.0 - 6.0 - 8.0 - 8.0 - 10.0 -10.0

10.0 10.0 (c) (d) 8.0 8.0 6.0 6.0 4.0 4.0 2.0 2.0 0.0 0.0 1950 1980 2010 2040 2070 2100 1950 1980 2010 2040 2070 2100 - 2.0 - 2.0

WS Tmean (°C) - 4.0 - 4.0 - 6.0 - 6.0 - 8.0 - 8.0 - 10.0 -10.0

10.0 (e) 8.0 6.0 4.0 2.0 0.0 1950 1980 2010 2040 2070 2100 - 2.0

WS Tmean (°C) - 4.0 - 6.0 - 8.0 - 10.0

Fig. 5. Variations of the WS mean temperature. (a) St-François; (b) Richelieu; (c) Batiscan; (d) Yamachiche; (e) St-Maurice. For the reference period and the three futures horizons, the dots represent the mean of all models. The robust lowess curve (with a 10 year smoothing) is represented with the red line. The variability between simulations is indicated by the error type dash line (2r). WS Tmean of each 30 years period is represented by the black line. (For interpretation of the references to colour in this ﬁgure legend, the reader is referred to the web version of this article.)

annual precipitation (S/Pan) for half of the climatic stations studied served on the Richelieu), 25% for the St-François and below 15% in New-England (USA) for the period from 1948 to 2000. The sta- for the south Richelieu. The differences between the first horizon tions located in the northern part of that region showed the stron- and the reference period and between each horizons are significant gest trend with a decrease of the S/Pan ratio (7%). For the St. for all rivers (p < 0.01). As in the case of WS Tmean, the variability Lawrence tributaries, Fig. 6 shows that the increase in WS Tmean between models is higher during the last horizon. from 1970 to 1980 was associated with a marked and rapid de- There is a clear latitudinal gradient in the natural variability ob- crease in the WS S/P ratio calculated for the WS period only. Before served in the WS S/P ratio during the reference period. The inter- 1980, the WS S/P ratio was about 50% except for the Richelieu annual variability in the WS S/P ratio is larger for the north shore where this ratio was close to 35%. From 1980 to 1999, this ratio rivers that have lower WS Tmean. This variability is expected to was around 30% for the Richelieu and 40% for the other tributaries. decrease in response to the projected increase in temperature dur- The decrease detected at the beginning of the 1980s is less clear for ing the winter/spring period. However, this change cannot be fully the Richelieu which has a higher mean temperature than other captured with the simulation technique used (perturbation meth- tributaries. The period from 1970 to 2001 is very likely influenced od) which preserves the inter-annual variability of temperatures by high winter NAO indices (positive anomalies). and precipitation amounts. For the projected series, the WS S/P ratio continues to decrease (Fig. 6). The overall mean annual decrease of the ratio is between WS total precipitation 0.15 to 0.2% per year. For the first time horizon (2020s), WS S/P Total precipitation during the winter/spring period (WS Ptotal) is approximately 40% for the St-François, St-Maurice, Yamachiche is characterized by substantial inter-annual variability (data not and Batiscan rivers. For the Richelieu, WS S/P is close to 25% during shown). For the reference period 1961–1990, the maximum pre- the course of the first horizon. When compared to the reference cipitation is registered in 1974–1975 and the minimum in 1964– period, the projected decrease in WS S/P is of 5% and of 10% for 1966 and after 1980. Although WS total precipitation is higher the north shore and south shore tributaries respectively. For the for the Richelieu and the St-François, the inter-annual variability second horizon, WS S/P decreases by about 4–5% compared to is similar for all rivers. Discrepancy between the WS total precipi- the previous horizon. During the last horizon, the rate of decrease tation observed in south and north shore watersheds is largest will be slightly higher (5 to 6%). The WS S/P ratio is around 30% for during the low precipitation period. Partly due to the high inter- the north shore rivers (a value similar to the current value ob- annual variability, historical data for each river do not show any 74 C. Boyer et al. / Journal of Hydrology 384 (2010) 65–83

80.0 80.0 (a) (b) 70.0 70.0 60.0 60.0 50.0 50.0 40.0 40.0 30.0 30.0 WS S/P ratio (%) 20.0 20.0 10.0 10.0 0.0 0.0 1950 1980 2010 2040 2070 2100 1950 1980 2010 2040 2070 2100

80.0 80.0 (c) (d) 70.0 70.0

60.0 60.0

50.0 50.0

40.0 40.0

30.0 30.0 WS S/P ratio (%) 20.0 20.0

10.0 10.0

0.0 0.0 1950 1980 2010 2040 2070 2100 1950 1980 2010 2040 2070 2100

80.0 (e) 70.0 60.0 50.0 40.0 30.0 WS S/P ratio (%) 20.0 10.0 0.0 1950 1980 2010 2040 2070 2100

Fig. 6. Variations of the WS S/P ratio. (a) St-François; (b) Richelieu; (c) Batiscan; (d) Yamachiche; (e) St-Maurice. For the reference period and the three futures horizons, the dots represent the mean of all models. The robust lowess curve (with a 10 year smoothing) is represented with the red line. The variability between simulations is indicated by the error type dash line (2r). The mean WS S/P ratio of each 30 years period is represented by the black line. (For interpretation of the references to colour in this ﬁgure legend, the reader is referred to the web version of this article.)

Table 4 Mean annual WS total precipitation over the reference period and the three simulated horizons.

All models Batiscan (cm) Yamachiche (cm) St-Maurice St-François Richelieu (cm) (LaGabelle) (cm) Reference period Mean 36.43 37.35 36.51 39.93 47.39 Minimum 9.61 9.53 9.59 7.45 8.77 Maximum 62.98 70.74 73.19 56.68 65.53 2010–2039 Mean 40.43 41.40 40.92 44.41 52.99 Minimum 27.45 27.10 28.18 26.34 29.87 Maximum 68.57 77.02 80.18 62.78 72.52 2040–2069 Mean 41.33 42.32 41.75 45.37 54.00 Minimum 28.21 26.85 28.85 26.98 30.54 Maximum 70.27 78.90 82.12 64.11 73.87 2070–2099 Mean 44.40 45.44 44.95 48.82 58.45 Minimum 30.05 30.11 31.06 28.85 33.13 Maximum 74.44 83.74 87.18 69.09 79.35

temporal trend (Kendall-Tau, p > 0.05). For the three simulated from one horizon to the other (Table 4). The difference between horizons, the annual WS total precipitation generally increases the mean of each horizon is however not signiﬁcant, (p > 0.05). C. Boyer et al. / Journal of Hydrology 384 (2010) 65–83 75

The small changes in WS total precipitation suggest that varia- within the uncertainty zone (±19%) associated with HSAMI calibra- tions in the WS S/P ratio are mainly explained by modification of tion process. the WS Tmean. The average rate of decrease of the WS S/P ratio with increasing temperature for the reference period and for the three future horizons is 5.2% per 1 °C. WS S/P ratios are projected Mean seasonal and monthly discharge to be lower than 25–30% for the Richelieu during all time horizons Unlike the mean annual discharge, changes in the temporal dis- and during the last horizon for the other rivers. This suggests that tribution of discharge (monthly and seasonal) can be important. As the potential for stocking water into the snow cover will be re- expected from the results of other studies in Québec (Minville duced, leading to important consequences for the winter and et al., 2008; Quilbé et al., 2008), the analysis of the simulated sea- spring discharges. sonal discharges indicates that mean winter and mean spring discharges are the most altered by changes in climatic variables. Results show that all models projected an increase in winter dis- Hydrologic impacts of climate change hydrological scenarios charge compared to the reference period (Fig. 8). Except for the Yamachiche, the mean winter discharge will increase by an aver- Mean annual discharge age of 52 % (ranging from 42% to 70%, which represent the upper For all five rivers and for all horizons, changes in mean annual and lower quartile) for the 2020s and by 133% (ranging from 92% discharge when compared to the reference period are generally to 163%) for 2080s. These differences are statistically significant lower than 15%. As noted in previous studies, there is a large var- (p < 0.01). The projected increase for the Yamachiche is larger than iability among the results from the different climate models due for the other rivers. However, in this case, the value of the simu- to the difference in the temperature and precipitation projections lated discharge (0.76 m3/s) for the reference period was fairly low- (Fig. 7). An increase in mean annual discharge is projected by the er than the observed value (3.26 m3/s). The underestimation for model HadCM3 (+6.6–+17.7%) and by the CSIRO-Mk2 model this river impinges on the analysis of the change projected for (1% to + 10 %). Conversely, the ECHAM4 model is projecting a de- the future horizons. It is postulated that the projected increase in crease in the mean annual discharge (4% to 12%). The decrease mean winter discharges are slightly underestimated for all tribu- in annual water volume estimated by this model is a result of the taries except for the Richelieu. This is due to a negative bias in projected low changes in total precipitation associated with a mod- mean winter discharge during calibration process (Table 3). erate increase in mean temperature (Fig. 3). These conditions con- For the spring season, the overall mean projected change is a de- tributed to an increase of water loss by evapotranspiration. All the crease of 8% (ranging from 18% to + 8%) for 2020s and of 26% (rang- projected values for changes in the mean annual discharge are ing from 16% to 40 %) for 2080s. Unlike the other models, the HadCM3 simulations projected a significant increase (16%) of the spring discharge for the Batiscan, Yamachiche, LaGabelle and Riche- lieu during the first horizon. The ECHAM4 and CSIRO-Mk2 project 20.00 2010 - 2039 that the mean discharges will decrease by 14% (ranging from 19% 15.00 to 7%) and by 33% for 2020s and 2080s, respectively. The projected 10.00 decrease in mean spring discharge may be slightly overestimated for 5.00 all tributaries except the Richelieu due to the negative bias in the spring mean discharge during the calibration (Table 3). For the first 0.00 horizon, the projected changes for the Batiscan, Yamachiche and -5.00 HadCM3 HadCM3 CSIRO CSIRO ECHAM4 ECHAM4mean all A2b B2b -Mk2 -Mk2 A2 B2 simulations LaGabelle watersheds are within the uncertainty zone. -10.00 A2 B2 The projected reduction of the mean discharge is lower in the Weighted difference (%) -15.00 summer than that expected for the spring. The mean reduction is under 20 % for the three time horizons. During the two first hori- 20.00 2040 - 2069 zons, some simulations (HadCM3 A2b and B2b and CSIRO-Mk2 15.00 B2) showed an increase in the summer discharge. The ECHAM4 (A2 and B2) model projected a significant decrease (15–30%) for 10.00 all rivers and horizons. For the fall, the trend is variable between 5.00 rivers and horizons. Changes in mean discharge are expected to 0.00 be lower than ±20%. HadCM3HadCM3CSIRO CSIRO ECHAM4ECHAM4 mean all The reduction of the differences in discharge magnitude be- -5.00 simulations A2b B2b -Mk2 -Mk2 A2 B2 tween winter and spring for south shore rivers (Richelieu and St- A2 B2 Weighted difference (%) -10.00 François) mainly occur during the first horizon and will be driven -15.00 by an increase in the winter months discharges (January, February and March) and a decrease in the May discharge (Fig. 8b). For the 20.00 2070 - 2099 Batiscan second horizon, mean monthly discharges decrease during April Yamachiche 15.00 Richelieu and May and continue to increase in January and February. March St-François mean monthly discharges change slightly compared to the horizon 10.00 St-Maurice 2020s. For the last horizon, mean January and February discharges 5.00 continue to increase and discharges from March to May do not 0.00 change compared to the previous horizon. HadCM3HadCM3 CSIRO CSIRO ECHAM4ECHAM4mean all A2 B2 simulations Changes for the north shore rivers (LaGabelle, Batiscan and -5.00 A2b B2b -Mk2 -Mk2 A2 B2 Yamachiche) appear to occur gradually throughout the three hori- -10.00 Weighted difference (%) zons (Fig. 8c–e). The first modifications will be that the mean -15.00 monthly discharges will increase in late winter and will decrease in May. For the second horizon, this trend is accentuated. Dis- Fig. 7. Relative mean annual discharges of each horizon in relation to the reference charges from January to March are shown to increase. For the last period (weighted difference (%) = (Qmean annual future Qmean annual reference period)/ Qmean annual reference period 100). horizon, mean monthly discharges continue to increase in January 76 C. Boyer et al. / Journal of Hydrology 384 (2010) 65–83

weighted difference CSIRO-Mk2 B2 ECHAM4 A2 HadCM3 B2b CSIRO-Mk2 A2 ECHAM4 B2 HadCM3 A2b 1961-1990 (a) (b) 250.0 2010-2039 600.0 100.0 2010-2039 1000.0 /s) 3 200.0 500.0 /s) 800.0 3 150.0 400.0 50.0 600.0 100.0 300.0 400.0 50.0 200.0 0.0 123456789101112 0.0 100.0 200.0

Weighted differenceWeighted (%) 123456789101112 Mean discharge (m Weighted differenceWeighted (%) - 50.0 0.0 - 50.0 0.0 Mean discharge (m

250.0 2040-2069 600.0 100.0 2040-2069 1000.0 /s)

200.0 500.0 3 800.0 /s) 3 150.0 400.0 50.0 600.0 100.0 300.0 400.0 50.0 200.0 0.0 123456789101112 0.0 100.0 200.0 123456789101112 Weighted differenceWeighted (%) Weighted difference (%) difference Weighted 0.0 Mean discharge (m - 50.0 0.0 - 50.0 Mean discharge (m

250.0 2070-2099 600.0 100.0 2070-2099 1000.0 /s) 200.0 500.0 3 800.0 /s) 3 150.0 400.0 50.0 600.0 100.0 300.0 400.0 50.0 200.0 0.0 123456789101112 0.0 100.0 200.0 Mean discharge (m Weighted differenceWeighted (%) 123456789101112 - 50.0 0.0 differenceWeighted (%) - 50.0 0.0 Mean discharge (m

(c) (d) 300.0 1200.0 2010-2039 25.0 /s) /s) 3 240.0 2010-2039 1000.0 20.0 3 800.0 180.0 600.0 15.0 120.0 400.0 10.0 60.0 200.0 5.0 0.0 0.0 123456789101112 123456789101112 Mean discharge (m 0.0 Mean discharge (m Weighted difference (%) - 60.0 Weighted difference- (%) 200.0

2040-2069 1200.0 2040-2069 25.0

300.0 /s) 3 /s)

3 1000.0 240.0 20.0 800.0 180.0 600.0 15.0 120.0 400.0 10.0 60.0 200.0 5.0 0.0 0.0 123456789101112

123456789101112 Mean discharge (m Weighted differenceWeighted (%) - 200.0 0.0 Weighted difference (%) Mean discharge (m - 60.0 1200.0 25.0

2070-2099 /s)

300.0 3

/s) 2070-2099 1000.0

3 20.0 240.0 800.0 180.0 600.0 15.0 120.0 400.0 10.0 60.0 200.0 5.0 0.0 0.0 123456789101112 123456789101112 Weighted differenceWeighted (%) - 60.0 - 200.0 0.0 Mean discharge (m Weighted difference (%) Mean discharge (m

Fig. 8. Mean monthly discharges (mean of each horizon) and relative mean monthly discharges in relation to the reference period. (a) St-François; (b) Richelieu; (c) Batiscan, (d) Yamachiche and (e) LaGabelle (LaGabelle). and February but changes in March and May are less marked than initiated in the historical period. Observed data from 1932 to what is anticipated for the first horizon. 2004 show a significant trend (Kendall-Tau, p < 0.1) toward an earlier date of the WS CV date (data not shown). A shift of 8–12 days Winter/spring center-volume date was observed during this period. The trend is generally stronger for Projected changes in the temporal distribution of discharges the period following 1960. However, for the Batiscan and LaGabelle indicate that the spring flood associated with snow melt occurs rivers, the trends are not significant (p > 0.1) for the period follow- at an earlier date in the future. We have used the winter/spring ing 1960. The WS CV date is earlier for the southern rivers (Riche- center-volume date (WS CV date) to evaluate the magnitude of lieu and St-François). From 1932 to 2004, two periods when the the shift in the timing of the spring flood. For all tributaries except WS CV date was later than the mean of the period can be distin- the Batiscan, the trend toward an earlier WS CV date was already guished (shown by the lowess curve). The first period is in the early C. Boyer et al. / Journal of Hydrology 384 (2010) 65–83 77

weighted difference CSIRO-Mk2 B2 1940s and the second period is in the early 1970s. These periods CSIRO-Mk2 A2 ECHAM4 B2 are known to have been characterized by low winter temperatures ECHAM4 A2 HadCM3 B2b with high total precipitation. For the studied tributaries, low WS HadCM3 A2b 1961-1990 (e) Tmean and high WS precipitation (snow and total) were observed 500.0 2010-2039 60.0 for these two periods. After 1950–1960, water regulation on rivers 400.0 50.0 for hydro electricity production may also have contributed to the 300.0 40.0 change in the WS CV date. 30.0 Simulated data for the next century show that the WS CV date 200.0 will occur at an earlier date (Table 5). The differences between the 20.0 100.0 mean WS CV dates for each horizon are signiﬁcant for all rivers 10.0 0.0 (Mann–Whitney test, p < 0.05). Transition towards an earlier WS Weighted difference (%) -50.0 1234567891011120.0 CV date will be more rapid for the rivers located at higher latitudes. 500.0 70.0 For the north shore tributaries, the WS CV date will occur 9– 2040-2069 12 days earlier at each horizon (Table 5). For the south shore rivers, 400.0 60.0 50.0 the changes will be a shift of 4–8 days per horizon. At the last hori- 300.0 40.0 zon, it is projected that the mean WS CV date will be 22 (Richelieu) 200.0 30.0 to 34 days (Batiscan) earlier than what was observed during the reference period. The temporal shift in the WS CV date is larger 100.0 20.0 for the Yamachiche (36 days) than for the Batiscan. However, 10.0 Weighted difference (%) 0.0 uncertainty of the hydrological simulations for January is high for -50.0 1923 4 5678 101112 0.0 the Yamachiche thus leading to potentially larger error in the pro- 500.0 70.0 jected value of WS CV date. These results are consistent with re-

2070-2099 /s) 400.0 60.0 3 sults reported by Minville et al. (2008) who indicated that the 50.0 300.0 time of peak discharge projected for a basin located in central Que- 40.0 bec will be about 23 days earlier in 2080 horizon with a large var- 200.0 30.0 iability between models (6–46 days). They are also consistent with 100.0 20.0 results obtained for the North-American Eastern coastal region 10.0 0.0 (Hayhoe et al., 2007). For this lower latitude region, the peak Mean discharge (m Weighted difference (%) - 50.0 12 345 67891011120.0 streamﬂow in spring is projected to be in advances of 10–15 days by the end of the century. Fig. 8 (continued)

Table 5 Change of the WS CV date for the reference period and the three simulated horizons. Date is deﬁned as DD-MM.

All models Batiscan Yamachiche St-Maurice (LaGabelle) St-François Richelieu Horizon 1 2010–2039 Mean 12–04 15–04 20–04 23–03 27–03 Lower quartile 05–04 07–04 15–04 15–03 21–03 Upper quartile 21–04 24–04 26–04 31–03 31–03 Horizon 2 2040–2069 Mean 01–04 04–04 10–04 15–03 21–03 Lower quartile 25–03 29–03 04–04 08–03 16–03 Upper quartile 10–04 14–04 17–04 23–03 27–03 Horizon 3 2070–2099 Mean 22–03 22–03 31–03 08–03 17–03 Lower quartile 14–03 14–03 24–03 28–02 12–03 Upper quartile 01–04 03–04 10–04 16–03 23–03 Reference period Mean 25–04 27–04 01–05 08–04 08–04 Lower quartile 24–04 19–04 05–04 31–03 06–04 Upper quartile 01–05 28–04 16–04 18–04 19–04

Difference between horizon 1 and reference period (days) 13 12 11 16 12 Difference between horizon 2 and 1 (days) 11 11 9 8 5 Difference between horizon 3 and 2 (days) 10 12 11 7 4 Difference between horizon 3 and reference period (days) 34 36 31 31 22 Slope of the temporal trend 2010–2100 0.328 0.368 0.324 0.262 0.161

Table 6 Kendall-Tau correlation coefﬁcients between the WS CV date and climatic variables (2010–2099). Signiﬁcant correlations (p < 0.05) are in bold.

Mean all models 2010–2099 Batiscan St-Maurice Yamachiche Richelieu St-Francois (LaGabelle)

WS Tmean 0.62 0.66 0.66 0.42 0.50 Tmean March 0.49 0.56 0.58 0.38 0.40 Tmean April 0.55 0.69 0.53 0.38 0.39 WS PTotal 0.03 0.02 0.04 0.02 0.04 WS snow 0.41 0.39 0.43 0.29 0.37 Snow March 0.37 0.34 0.37 0.28 0.45 WS S/P ratio 0.43 0.38 0.38 0.29 0.39 S/P ratio February 0.36 0.23 0.33 0.33 0.27 S/P ratio March 0.44 0.40 0.46 0.24 0.42 S/P ratio April 0.23 0.20 0.19 0.14 0.15 78 C. Boyer et al. / Journal of Hydrology 384 (2010) 65–83

Relation between WS CV date and climatic variables utaries are always stronger in late winter months but are generally An earlier WS CV date might be caused by an earlier snow melt weaker than for the north shore tributaries. For the Richelieu and and by changes in precipitation amounts or phase (Hodgkins et al., St-François, the WS CV date is positively correlated (p < 0.05) with 2003). If an earlier WS CV date is due to early snow melt, this date the March precipitation, which mainly consists of rain, for the hori- should be related to late winter and early spring temperatures zon 2080s. For this horizon, a weak negative correlation (p < 0.1) (Hodgkins et al., 2003). If an earlier WS CV date is linked to changes with precipitation is also observed in January for the Richelieu. in precipitation, it should be associated with winter and early For the north shore tributaries, the positive correlation between spring Ptotal and S/P ratio and with Tmean. For the northern WS CV date and total precipitation (snow) in February becomes New-England region (northwestern Maine and Northern New stronger for the last horizon. Hampshire, USA), Hodgkins et al. (2003) have found that changes We have plotted WS CV date in relation with WS Tmean (results in the WS CV date in historical data (1900–2000) were mainly from all models) for the five tributaries and for the reference and linked with January to April temperature and precipitation. For simulated periods (Fig. 9). This shows that WS CV date decreases the same region, Huntington et al. (2004) have shown that WS at a mean rate of 6 days with an increase of 1 °C of WS Tmean CV date was significantly and positively correlated with the annual but when WS Tmean is above 4 °C, variations in the WS CV date or winter S/P ratios. is no longer related to changes in WS Tmean. For a temperature For the entire simulated period (from 2010 to 2099), the WS CV of 4 °C, the WS S/P ratio is around 20%. For these conditions, the date of all tributaries was strongly negatively correlated with the mean value of WS CV date is March 14 with a standard deviation WS Tmean, as one might expect (Table 6). The correlation between of ±12 days. Results from the St-François frequently show lower WS CV date and WS Tmean is stronger for rivers located on the values of WS CV date contributing a large part of the variability ob- north shore of the St. Lawrence. For all tributaries, high correlation served in the WS CV dates. The threshold of WS Tmean = 4 °C will coefficients are observed with March and April temperatures. be reached during the last horizon only for the south shore rivers. There is no correlation between WS CV date and WS PTotal. How- For the north shore rivers, it is expected that WS CV date will con- ever, correlation coefficients between WS CV date and WS S/P ratio tinue to be affected by the temperature increase through its effect are significant and positive (+) for all tributaries. The correlation on the S/P ratio during the winter/spring period after 2100. coefficient between WS CV date and March S/P ratio is slightly stronger (+) than for other months for the LaGabelle, Batiscan, Yamachiche and St-François, For the Richelieu, the highest correla- Discussion tion value is with the February S/P ratio (data not shown). It is important to remember that the WS S/P ratio is negatively corre- Effect of latitude on hydrological change lated to both WS Tmean and WS PTotal. The correlation with WS Tmean is twice as much as that with WS PTotal for south shore All simulations give a consistent picture of changes where the rivers. latitude gradient of WS Tmean is likely the primary cause of The strength of the correlation between climate variables and changes in the WS S/P ratio, winter and spring snowmelt runoff WS CV date is variable through time (Table 7). For all tributaries, and the WS CV dates in the studied area. The timing and duration the strength of the correlation with Tmean in April and March of the maximum change period in the hydrological regime are also slightly decreases from the first to the last horizon. For the Bati- a function of the latitude of the watershed. Southern tributaries scan, St-François and Yamachiche, the relation with April Tmean will move quickly towards a new rain-based hydrological regime is no longer significant (p > 0.1) during the 2080s horizon. given that mean air temperatures are already relatively high for For the north shore tributaries the strength of the correlation these watersheds (1.4 °C for the StFrancois and +0.2 °C for the between the WS CV date and the monthly S/P ratio varies from Richelieu for 1990–1999). A strong link between WS CV dates one horizon to the other (Table 7). For the first horizon, correla- and mean air temperatures was observed in historical data tions are stronger for March and April than for the other months. (1942–2000) for rivers in the North-Eastern of USA and in the Wes- For the other two horizons, the correlation coefficient is stronger tern part of North America (Hodgkins and Dudley, 2006; Stewart for February and March. The correlations for the south shore trib- et al., 2004). The magnitude of the change towards an earlier date

Table 7 Signiﬁcant correlations (p < 0.05 or p < 0.1 ()) between WS CV date and monthly values of climatic variables (temperature, total precipitation and S/P ratio) for the three simulated horizons. Non-signiﬁcant correlation with any studied climate variables is indicated by NaN.

Horizon Batiscan St-Maurice (LaGabelle) Yamachiche St-François Richelieu 2010–2039 S/P : March (0.36) and S/P : March(0.30) and S/P : March (0.35) and S/P : February (0.24) and S/P : March (0.24) April (0.38) April (0.26) April (0.26) March (0.44) P : NaN P : February (0.23) and P : February (0.24) P : NaN P : March (0.26) March (0.22) Tmean : March (0.40) Tmean : March (0.39) Tmean : March (0.42) Tmean : March (0.30) Tmean : March (0.33) and April (0.52) and April (0.64) and April (0.60) and April (0.32) and April (0.26) 2040–2069 S/P : February (0.33) S/P : March (0.43) S/P : February (0.28) S/P : March (0.38) S/P : NaN and March (0.52) and March (0.54) P : NaN P : February (0.24) P : February (0.22) P : NaN P : March* (0.30) Tmean : March (0.34) Tmean : March (0.46) Tmean : March (0.39) Tmean : March (0.23) Tmean : March (0.30) and April (0.34) and April (0.57) and April (0.37) and April (0.23) and April (0.27) 2070–2099 S/P : February (0.29) S/P : February (0.26) S/P : February (0.32) S/P : March (0.23) S/P : February (0.23) and March (0.45) and March (0.39) and March (0.53) P : February (0.24) P : February (0.24) P : February (0.27) P : March (0.25) P : January (0.24) and March (0.32) Tmean : March (0.28) Tmean :March (0.43) Tmean : March (0.33) Tmean : March (0.23) Tmean : April (0.25) and April (-0.40) C. Boyer et al. / Journal of Hydrology 384 (2010) 65–83 79

Reference period (1961-1990) Batiscan Richelieu St-François 20-05 St-Maurice Yamachiche 30-04

10-04

21-03 March 14 01-03 WS CV date 10-02

21-01

01-01 -6.0 -4.0 -2.0 0.0 2.0 4.0 6.0 8.0 10.0 12.0 14.0 WS Tmean (°C)

Fig. 9. Relation between the WS CV date and WS Tmean for the reference period and 2010–2100 (results obtained with all the models and scenarios). The dash lines indicated the standard deviation (1r) around the mean for cases with WS Tmean > 4 °C.

with an increase of 1 °C obtained for many stations is similar than ment volumes that can be transported during that season (Boyer the one reported here. et al., 2009). Conversely, higher winter flows will increase the risk When the WS S/P ratio is lower than 20–25% (corresponding to of having events that will produce sediment transport with result- a value of WS Tmean of 4 °C), the contrast between mean winter ing geomorphological impacts especially when the rivers are cov- and spring discharges is reduced. The WS CV date remains almost ered with ice (for early winter months, Tmean will be under 0 °C constant when WS Tmean is increasing. During the last horizon, it for all horizons). Although, sediment transport processes are not is expected that both South shore rivers will have a low WS S/P ra- well documented under an ice cover, it is generally assumed that tio (lower than 25%), high mean WS temperature (higher than 4 °C) prevailing winter low flow conditions do not have a significant im- and a reduced number of days when Tmean below 0 °C. During this pact on river morphology. Such an assumption may need to be re- horizon, it is projected that Tmean will be above 0 °C early in vised in the future. For all rivers in this study, the frequency of days March (Julian day 60) for the Richelieu. These conditions lengthen with a discharge higher than the current mean sediment transport the period of moderate to high discharges. February, March and thresholds will increase during winter (Table 8). In comparison April will contribute almost equally to the total WS discharge with the reference period, the frequency of days with discharges (20% for each month). For the reference period, 40 % of the total higher than the sediment transport threshold during the winter WS discharge was observed during April for the St-François and may increase by 143 % for the St-François (horizon 2020s) and by during April (29 %) and May (28%) for the Richelieu. For the North 1693% for the Batiscan (horizon 2050s). For all tributaries, this fre- shore rivers, it is expected that the WS S/P ratio will be over 25 % quency would, however, remain under the value observed during and WS Tmean lower than 4 °C during the last horizon (2080s). It the spring for the reference period. A net increase of this frequency is presumed that changes in WS CV date will continue beyond for both winter and spring is expected only for the Richelieu (21%, the period considered in this study (2100). For these rivers, it is 28% and 31% for horizons 2020s–2080s). likely that most of the streamflow will be observed in March and April (>50 %) during the last horizon instead of April and May Rare flood events during winter (>50%) as it has been observed during the reference period. The magnitude and frequency of rare events, defined as discharges larger than three times the standard deviation, are impor- Potential impacts for sediment transport and river geomorphology tant for channel stability. Despite the fact that the perturbation method used in this study does not allow an exhaustive analysis Discharge higher than the sediment transport threshold of the frequency/magnitude of the simulated hydrological data, Hydrological changes projected for the St. Lawrence tributaries our results suggest that the magnitude of rare large events during can have significant impacts on the frequency and magnitude of winter would increase compared to the reference period. The high- sediment transport processes. It is likely that a reduction of up to est increase would be observed on rivers located at higher latitudes 32% of the maximum spring mean discharge and of the frequency (e.g. Batiscan). Although some of the simulations projected a max- of discharges higher than the current mean sediment transport imum winter discharge higher than the maximum spring discharge threshold discharge (discharge at which sediment transport oc- observed for the reference period, the magnitude of rare events curred at the majority of the cross-sections along a river reach) during winter are projected to be lower than for the reference per- during this period will reduce the size of particles and the sedi- iod spring value. Compared to the reference period, the frequency

Table 8 Frequency of days (mean per years) during the winter with discharge higher than the sediment transport threshold for three of the ﬁve studied tributaries. The minimum and maximum values are extracted from the six sets of simulation.

Richelieu St-François Batiscan Period Winter Freq. days with Q > 450 m3/s Freq. days with Q > 330 m3/s Freq. days with Q > 150 m3/s 1961–1990 16 7 1 2010–2039 Mean all simulations (min, max) 41 (32, 47) 16 (13, 19) 4 (3, 6) 2040–2069 Mean all simulations (min, max) 53 (46, 59) 22 (19, 26) 9 (6, 13) 2070–2099 Mean all simulations (min, max) 59 (51, 66) 29 (24, 35) 16 (11, 26) 80 C. Boyer et al. / Journal of Hydrology 384 (2010) 65–83 of these events during winter would increase and decrease for flooding. The timing of germination of the species currently found higher and lower latitude rivers, respectively. By the 2020s, the fre- in these rivers might also be out of phase with high flows which are quency of rare large events simulated during winter for the Bati- expected to occur at an earlier date. scan would be similar to the frequency of rare events observed Transformations of the wetlands located at the mouths of the for the St-François during the reference period. For this horizon, Batiscan, Yamachiche and St-François rivers are expected to be the winter mean temperature for the Batiscan watershed is pro- more important than changes anticipated along riparian corridors. jected to be similar to the winter mean temperature observed dur- These zones will be influenced by seasonal changes in tributaries ing the reference period in the St-François watershed (winter hydrology, by modifications of the St. Lawrence water levels, both Tmean ffi8 °C). This result suggests that the projected increase changes may amplify the temporal shift between high river dis- in the frequency of rare large hydrological events for rivers in high- charges and high St. Lawrence water levels (especially after hori- er latitudes is in line with what was observed in the St-François zon 2050s), and by changes in sedimentation and erosion watershed. This tributary may be used as an analogue to study dynamics. Deltas and tributary mouth bars along the St. Lawrence the behavior that is likely to occur in winter for rivers like the are important for its biodiversity because of their heterogeneous Batiscan. and contrasting physical characteristics (Desgranges and Jobin, 2003). Variable flooding conditions associated with seasonal flood- Potential impacts for aquatic and riparian ecosystems ing or daily tidal fluctuations create a complex and dynamic mo- saic of wetlands and aquatic habitats. Changes in hydrological Aquatic and riparian ecosystems are spatially and temporally characteristics and sediment dynamics of the tributaries are ex- dynamic and are largely shaped by fluvial processes (Hughes, pected to expand the areas of sediment deposition and to trans- 1997; Hupp, 1988; Malanson, 1993; McKenney et al., 1995; Nai- form their structures. Two-dimensional modeling of the flow and man et al., 1993; Naiman and Décamps, 1997). Flow velocity, water sediment dynamics of these zones will need to be conducted in or- depth and hydrological fluctuations (seasonal variability and dura- der to estimate the magnitude of this morphological development. tion, frequency and timing of floods) play a major role in the struc- It is, however, difficult to predict if diversity will increase or de- ture and diversity of ecosystems. Changes in these variables can crease under the projected new conditions. induce complex responses in riparian vegetation and ecosystem dynamics that are not well understood at present (Crowder et al., Potential impacts for water management (hydro electricity) 1996). Studies have shown that even if the mean annual discharge is not changed significantly, variation at a finer temporal scale (e.g. The projected shift in winter precipitation from snow to rain monthly) can cause important modifications in riparian vegetation. and its impact on the snowpack accumulation and on winter and spring discharges will influence water management at the wa- Effect of higher winter discharges tershed scale. For tributaries with dams and reservoirs used for hy- Hydrological changes projected during winter and early spring dro electricity, like the St-Maurice and St-François, managers may may affect riparian landscapes in the St. Lawrence and its tributar- have to adopt new operation strategies to account for higher win- ies. These changes are expected to be larger than what has been ter discharges and lower spring discharges and for the higher observed for the five studied St. Lawrence tributaries between uncertainty in runoff associated with rainfall compared to snow- 1964 and 1997 (Charron et al., 2008). Although winter flows are melt (Gleick and Adams, 2000). Although dam and reservoir man- rarely analyzed in ecological studies because vegetation is in dor- agers already adapt to the existing variable climatic and hydrologic mancy, their impacts are important for the stability of the river conditions from one year to the other, global climate change pro- channel and for preserving the substrate where seeds are stored. jections add a new level of uncertainty. A better knowledge of Higher winter flows (increase of 50–200% of the mean winter flow) the projected changes will help to develop adaptation and design may increase bed and bank erosion and consequently reduce the strategies in order to optimize hydroelectric production while min- surviving chance of plants and seeds stored into the bank sub- imizing risks. strate. Higher winter flows may also keep plants and seeds under water for a longer period. This may damage some plants or prevent Limitations due to hydrological and climate variability a successful germination of seeds. Changes in river hydrology and sedimentology during the winter may also modify fish habitat. Projected changes in discharge must be interpreted with care Higher winter flows may affect fish migratory behavior and en- due to the limitations link with the GCMs and with the perturba- hance the erosion and transport of fish eggs and larvae of winter tion method used to generate the climate variables. It is possible spawning species (Bergeron et al., 1998; Fortin et al., 1992). Flow that the larger differences observed between GCMs projections velocity will be a limiting factor for fish that have low swimming during the spring, and to a lesser extent during the winter, increase capacity like the Atlantic Tomcod (Microgadus tomcod)(East and the uncertainty in the estimate of the amplitude of the hydrologi- Magnan, 1989), a fish that is spawning into some of the St. Law- cal changes for these seasons, particularly for the last horizon. The rence tributaries (Sainte-Anne and Batiscan rivers) between the direction of change, however, is clear as indicated by the conver- end of December and February. gence of GCMs projections. The use of the perturbation method in this study does not allow us to fully appraise the potential Effect of earlier spring flood and of lower spring discharges changes that may occur in the inter-annual variability and in the Earlier spring floods (from 22 to 34 days) might be out of phase frequency of extreme meteorological events in response to climate with spring spawning species that is controlled by a combination changes. Climate variable extremes resulting from this approach of photoperiod, temperature and flow. Lower spring flows (down are those observed during the reference period enhanced or damp- by as much as 40%) will reduce the extent of areas available for ened according to the perturbation factors (Graham et al., 2007). species that are spawning into the St. Lawrence tributaries during Consequently, the method limits our ability to study the variability early spring, like the walleye (Sander vitreus) and the Northern pike at small temporal scales and restricts the analysis to basic statisti- (Esox lucius) for example. Concerning the vegetation, lower water cal properties of the hydrological data. The limitation due to the levels during the spring are likely to modify plants and seeds dis- assumption of the stationarity of the variance and other moments persion patterns. Depending of the form of the river cross-section, is less problematic in this study because we are mainly focusing on it can also reduce the areas of the riparian zones affected by annual hydrologic changes that are linked to the modification of snow C. Boyer et al. / Journal of Hydrology 384 (2010) 65–83 81 accumulation and snowmelt and consequently on the gradual quantify the uncertainty associated with the use of GCMs projec- shifting from a snow hydrological regime to a rain regime. Exten- tion at local scale, downscaling methods and hydrological model- sive study of changes in the frequency of high magnitude hydro- ing and to examine more closely the hydrological impact of logical events associated with extreme meteorological events and possible change in climate variability. Care should also be taken in the inter-annual variability on hydrological characteristics of in future simulations to represent discharge characteristics with the watersheds was also limited. a higher level of confidence in order to exhaustively assess changes Although it is generally presumed that the variability of climate in the frequency of discharge with active sediment transport (dis- variables will be modified during the current century, it remains charge higher than sediment transport thresholds), in the magni- uncertain how ongoing climate warming will influence this charac- tude of large events and in the temporal sequence of discharges. teristic. An important part of the temporal variability in the winter mean surface air temperature over the northern hemisphere is due to the NAO (Hurrell, 1996). However, the range of modeling results Acknowledgements obtained at the moment using various approaches indicates that substantial uncertainty still exists in the projection of the response This paper was funded by the Ouranos Consortium and the Nat- of the NAO to increasing GHG concentrations (Hurrell et al., 2006). ural Science and Engineering Research Council of Canada (NSERC). A complete investigation of the climate variability is then currently This research is part of the program of the Canada Research Chair limited even with the most sophisticated approach (Hurrell et al., in Fluvial Dynamics. The comments from three anonymous review- 2006; Hurrell and Deser, 2009). ers were very helpful to improve this paper.

Conclusion References

Andréasson, J., Bergström, S., Carlsson, B., Graham, L.P., Lindström, G., 2004. The results from this study clearly indicate that the climate Hydrological change–climate change impact simulations for Sweden. Ambio: A changes projected for the next century will induce important mod- Journal of the Human Environment 33 (4), 228–234. ifications of the St. Lawrence tributaries hydrological regime with a Bates, B.C., Kundzewicz, Z.W., Wu, S., Palutikof, J.P. (Eds.), 2008. Climate Change and gradual shifting from snow to rain regime. It is projected that by Water. Technical Paper of the Intergovernmental Panel on Climate Change. IPCC Secretariat, Geneva, 210 pp. the end of the century the difference between mean monthly dis- Bergeron, N.E., Roy, A.G., Chaumont, D., Mailhot, Y., Guay, E., 1998. Winter charges will be small during winter and spring. Increase in temper- geomorphological processes in the Sainte-Anne River (Quebec) and their ature during the winter/spring period will drive most of the change impact on the migratory behaviour of Atlantic tomcod (Microgadus tomcod). Regulated Rivers: Research and Management 14, 95–105. that are projected in river discharge. Warming of air temperature Bisson, J.L., Roberge, F., 1983. Prévision des apports naturels: Expérience d’Hydro- in February to April will induce an important decrease in the pro- Québec. Compte-rendu de l’Atelier sur la prévision du débit, Toronto, November portion of precipitation falling as snow, with a simultaneous in- 1983. Boé, J., Terray, L., Martin, E., Habets, F., 2009. Projected changes in components of crease in winter runoff and a reduction in the volume of water the hydrological cycle in French river basins during the 21st century. Water stored into the snowpack compared to the reference period. This Resources Research 45, W08426. doi:10.1029/2008WR007437. will lengthen the snow melting period and reduce spring runoff. Boyer, C, Verhaar, P.M., Roy, A.G., Biron, P.M., Morin, J., 2009. Impacts of environmental changes on the hydrology and sedimentary processes at the With these changes, the WS center-volume date is expected to confluence of St. Lawrence tributaries: potential effects on fluvial ecosystems. be in advance by 22–34 days depending of the watershed. The lat- Hydrobiologia (Ecosystem studies of the St. Lawrence River). doi: 10.1007/ itude of the river governs the timing of occurrence of the maxi- s10750-009-9927-1. Burakowski, E.A., Wake, C.P., Braswell, B., Brown, D.P., 2008. Trends in wintertime mum change (sooner for tributaries located south) and the climate in the northeastern United States: 1965–2005. Journal Geophysical duration of the period affected by marked changes in the temporal Research 113, D20114. doi: 20110.21029/22008JD009870. distribution of discharge (longer time scale for rivers located at Cayan, D.R., 1996. Interannual climate variability and snow pack in the western higher latitudes). United States. Journal of Climate 9, 928–948. Charron, I., Lalonde, O., Roy, A.G., Boyer, C., Turgeon, S., 2008. Changes in riparian Higher winter discharges may have a significant effect on river habitats along major tributaries of the St. Lawrence River, Québec, Canada. River geomorphological processes as sediment transport events may oc- regulations and application 24, 617–631. cur more frequently under ice-cover conditions than under current Chartier, I. (2006). Influence du modèle hydrologique sur la modélisation des apports dans une perspective de changements climatiques : une application sur conditions. Conversely, lower spring discharges may promote sed- le bassin versant Baskatong. In: Proceedings, Canadian Dam Association Annual imentation into the tributary and at their confluence with the St. Conference, Quebec City. Lawrence River. The combined effects of modifications in river Chaumont, D., Chartier, I., 2005. Développement de scénarios hydrologiques à des fins de modélisation de la dynamique sédimentaire des tributaires du Saint- hydrology and geomorphological processes will likely impact Laurent dans un contexte de changements climatiques. 46 pp. riparian ecosystems. Chiew, F.H.S., Teng, J., Vaze, J., Post, D.A., Perraud, J.M., Kirono, D.G.C., Viney, N.R., The use of multiple climatic models in this study has allowed us 2009. Estimating climate change impact on runoff across southeast Australia: method, results, and implications of the modeling method. Water Resources to examine plausible scenarios of changes into the hydrological re- Research 45, W10414. doi:10.1029/2008WR007338. gime of the St. Lawrence tributaries. Convergence between the Cleveland, W.S., Devlin, S.J., 1988. Locally weighted regression: an approach to models and the consistency of the results with other studies in- regression analysis by local fitting. Journal of the American Statistical Association 83, 596–610. creases the reliability of the outcome of this study. The strong link Cooley, K.R., 1990. Effects of Co2-induced climatic changes on snowpack and between winter-spring center-volume date and winter-spring streamflow. Hydrological Sciences Journal 35, 511–522. mean temperature and ratio of snow/precipitation (winter-spring) Court, A., 1962. Measures of streamflow timing. Journal of Geophysical Research 67, for the St. Lawrence tributaries suggests that the quality of the 4335–4339. Croley II, T.E., 2003. Great Lakes Climate Change Hydrologic Impact Assessment: IJC hydrological projections for northern rivers is highly sensitive to Lake Ontario-Saint Lawrence River Regulation Study. NOAA Technical temperature which drives most of the changes in the winter-spring Memorandum GLERL-126. NOAA, Great Lakes Environmental Research ratio of snow/precipitation. Laboratory, Ann Arbor, MI. 77 pp. Crossley, J.F., Polcher, J., Cox, P., Gedney, N., Planton, S., 2000. Uncertainties linked to The perturbation method used here has given interesting re- land-surface processes in climate change simulations. Climate Dynamics 16, sults for an early phase assessment of the impact of climate 949–961. changes on the hydrological regime of the St. Lawrence tributaries Crowder, A.A., Smol, J.P., Dalrymple, R., Gilbert, R., Mathers, A., Price, J., 1996. Rates of natural and anthropogenic change in shoreline habitats in the Kingston Basin, and in the identification of regions that are expected to be more Lake Ontario. Canadian Journal Fisheries Aquatic Science 53 (Suppl. 1), 121– sensitive. A further step needs to be taken in order to reduce and 135. 82 C. Boyer et al. / Journal of Hydrology 384 (2010) 65–83

Denis, B., Laprise, R., Caya, D., 2003. Sensitivity of a regional climate model to the Hodgkins, G.A., Dudley, R.W., Huntington, T.G., 2003. Changes in the timing of high spatial resolution and temporal updating frequency of the lateral boundary river flows in New England over the 20th Century. Journal of Hydrology 278, conditions. Climate Dynamics 20, 107–126. 244–252. DesGranges, J.-L., Jobin, B., 2003. Knowing, mapping and understanding Saint- Houghton, J. T., Ding, Y., Griggs, D.J., Noguer, M., Van der Linden, P.J., Dai, X., Maskell, Lawrence biodiversity, with special emphasis on bird assemblages. K., Johnson, C.A., 2001. Climate Change 2001: The Scientific Basis. Contribution Environmental Monitoring and Assessment 88, 177–192. du Groupe de travail au Troisième rapport d’évaluation du Groupe d’experts Diaz-Nieto, J., Wilby, R.L., 2005. A comparison of statistical downscaling and climate intergouvernemental sur l’évolution du climat Cambridge University Press, change factor methods: Impacts on low flows in the River Thames, United Cambridge, Royaume-Uni and New York, NY, ÉU, 881 pp. Kingdom. Climatic Change 69, 245–268. Hughes, F.M.R., 1997. Floodplain biogeomorphology. Progress in Physical Duan, Q., Sorooshian, S., Gupta, V.K., 1992. Effective and efficient global Geography 21, 501–529. optimization for conceptual rainfall-runoff models. Water Resources Research Huntington, T.G., Hodgkins, G.A., Keim, B.D., Dudley, R.W., 2004. Changes in the 28, 1015–1031. proportion of precipitation occurring as snow in New England. Journal of East, P., Magnan, P., 1989. Étude de la vitesse de nage du Poulamon Atlantique Climate 17, 2626–2636. (Microgadus tomcod) en relation avec les travaux de la traversée du fleuve Hupp, C.R., 1988. Plant ecological aspects of flood geomorphology and paleoflood Saint-Laurent par la ligne Radisson-Nicolet-Des-Cantons. Université du Québec history. In: Baker, V.R., Kochel, R.C., Patton, P.C. (Eds.), Flood Geomorphology. à Trois-Rivières, Hydro-Québec Environnement. 18 pp. Wiley, New York, pp. 335–356. Feser, F., 2006. Enhanced detectability of added value in limited area model results Hurrell, J.W., 1996. Influence of variations in extratropical wintertime separated into different spatial scales. Monthly Weather Review 134, 2180– teleconnections on northern hemisphere temperature. Geophysical Research 2190. Letters 23, 665–668. Fortin, V., 2000. Le modèle météo-apport HSAMI: historique, théorie et application. Hurrell, J.W., Deser, C., 2009. North Atlantic climate variability: the role of the North Rapport de recherche, Institut de recherche d’Hydro-Québec, Varennes. Atlantic Oscillation. Journal of Marine Systems 78, 28–41. Fortin, R., Léveillé, M., Guénette, S., Laramée, P., 1992. Contrôle hydrodynamique de Hurrell, J.W., Visbeck, M., Busalacchi, A., Clarke, R.A., Delworth, T.L., Dickson, R.R., l’avalaison des oeufs et des larves de poulamon atlantique (Microgadus tomcod) Johns, W.E., Koltermann, K.P., Kushnir, Y., Marshall, D., Mauritzen, C., Mccartney, sous le couvert de glace de la rivière Sainte-Anne, Québec. Aquatic Living M.S., Piola, A., Reason, C., Reverdin, G., Schott, F., Sutton, R., Wainer, I., Wright, Resources 5, 127–136. D., 2006. Atlantic climate variability and predictability: a CLIVAR perspective. Fortin, J.P., Moussa, R., Bocquillon, C., Villeneuve, J.P., 1995. HYDROTEL, un modèle Journal of Climate 19, 5100–5121. hydrologique distribué pouvant bénéficier des données fournies par la Jain, S., Lall, U., 2000. Magnitude and timing of annual maximum floods: trends and télédétection et les systèmes d’information géographique. Revue des sciences large-scale climatic associations for the Blacksmith Fork River, Utah. Water de l’eau 8, 97–124. Resources Research 36, 3641–3651. Fortin, L.-G., Turcotte, R., Pugin, S., Cyr, J.-F., Picard, F., 2007. Impact des Jain, S., Lall, U., 2001. Floods in a changing climate: does the past represent the changements climatiques sur les plans de gestion des lacs Saint-François et future? Water Resources Research 37, 3193–3205. Aylmer au sud du Québec. Canadian Journal of civil Engineering 34, 934–945. Johnson, F., Sharma, A., 2009. Measurement of GCM skill in predicting variables Fowler, H.J., Blenkinsop, S., Tebaldi, C., 2007. Linking climate change modelling to relevant for hydroclimatological assessments. Journal of Climate. doi:10.1175/ impact studies: recent advances in downscaling techniques for hydrological 2009JCLI2681.1. modelling. International Journal of Climatology 27, 1547–1578. Kay, A.L., Jones, R.G., Reynard, N.S., 2006. RCM rainfall for UK flood frequency Giorgi, F., Christensen, J., Hulme, M., von Storch, H., Whetton, P., Jones, R., Mearns, L., estimation II. Climate change results. Journal of Hydrology 318, 163–172. Fu, C., Arritt, R., Bates, B., Benestad, R., Boer, G., Buishand, A., Castro, M., Chen, D., Kendall, M.G., 1938. A new measure of rank correlation. Biometrika 30, 81–93. Cramer, W., Crane, R., Crossly, J., Dehn, M., Dethloff, K., Dippner, J., Emori, S., Kendall, M., 1975. Multivariate Analysis. Macmillan Co. III, New York, 210 p. Francisco, R., Fyfe, J., Gerstengarbe, F., Gutowski, W., Gyalistras, D., Hanssen- Laprise, R., 2008. Regional climate modeling. Journal of Computational Physics 227, Bauer, I., Hantel, M., Hassell, D., Heimann, D., Jack, C., Jacobeit, J., Kato, H., Katz, 3641–3666. R., Kauker, F., Knutson, T., Lal, M., Landsea, C., Laprise, R., Leung, L., Lynch, A., Ludwig, R., May, I., Turcotte, R., Vescovi, L., Braun, M., Cyr, J.-F., Fortin, L.-G., May, W., McGregor, J., Miller, N., Murphy, J., Ribalaygua, J., Rinke, A., Chaumont, D., Biner, S., Chartier, I., Caya, D., Mauser, W., 2009. The role of Rummukainen, M., Semazzi, F., Walsh, K., Werner, P., Widmann, M., Wilby, R., hydrological model complexity and uncertainty in climate change impact Wild, M., Xue, Y., 2001. Regional climate information-evaluation and assessment. Advances in Geosciences 21, 63–71. projections. In: Houghton, J.T. et al. (Eds.), Climate change 2001: The Malanson, G.P., 1993. Riparian Landscapes. Cambridge University Press, Cambridge. Scientific Basis. Contribution of Working Group to the Third Assessment 296 pp. Report of The Intergovernmental Panel on Climate Change. Cambridge Maurer, E.P., Hidalgo, H.G., 2008. Utility of daily vs. Monthly large-scale climate University Press, Cambridge, United Kingdom and New York, USA. 881 pp. data: an intercomparison of two statistical downscaling methods. Hydrology Gleick, P.H., Adams, D.B., 2000. Water: The Potential Consequences of Climate and Earth System Sciences 12, 551–563. Variability and Change for the Water Resources of the United States. Report of McKenney, R., Jacobson, R.B., Wertheimer, R.C., 1995. Woody vegetation and the Water Sector Assessment Team of the National Assessment of the Potential channel morphogenesis in low-gradient gravel-bed streams in the Ozark Consequences of Climate Variability and Change, Pacific Institute for Studies in Plateaus, Missouri and Arkansas. Geomorphology 13, 175–198. Development, Environment and Security, Oakland, CA. ISBN #1-893790-04-05. Meehl, G.A., Covey, C., Delworth, T., Latif, M., McAvaney, B., Mitchell, J.F.B., Stouffer, Gordon, C., Cooper, C., Senior, C.A., Banks, H.T., Gregory, J.M., Johns, T.C., Mitchell, R.J., Taylor, K.E., 2007. The WCRP CMIP3 multimodel dataset – a new era in J.F.B., Wood, R.A., 2000. The simulation of SST, sea ice extents and ocean heat climate change research. Bulletin of the American Meteorological Society 88, transports in a version of the Hadley Centre coupled model without flux 1383–1394. adjustments. Climate Dynamics 16, 147–168. Merritt, W.S., Alila, Y., Barton, M., Taylor, B., Cohen, S., Neilsen, D., 2006. Hydrologic Graham, L.P., Hagemann, S., Jaun, S., Beniston, M., 2007. On interpreting response to scenarios change in sub watersheds of the Okanagan basin, British hydrological change from regional climate models. Climatic Change 81, 97–122. Columbia. Journal of Hydrology 326, 79–108. Hartley, S., Keables, M.J., 1998. Synoptic associations of winter climate and snowfall Minville, M., Brissette, F., Leconte, R., 2008. Uncertainty of the impact of climate variability in New England, USA, 1950–1992. International Journal of change on the hydrology of a Nordic watershed. Journal of Hydrology 358, 70– Climatology 18, 281–298. 83. Hay, L.E., Wilby, R.L., Leavesly, H.H., 2000. Comparison of delta change and Mortsch, L.D., Quinn, F.H., 1996. Climate change scenarios for Great Lakes Basin downscaled GCM scenarios for three mountainous basins in the United States. ecosystems studies. Limnology and oceanography 41, 903–911. Journal of the American Water Resources Association 36 (2), 387–397. Naiman, R.J., Décamps, H., Pollock, M., 1993. The role of riparian corridors in Hayhoe, K., Wake, C.P., Huntington, T.G., Luo, L., Schwartz, M.D., Sheffield, J., Wood, maintaining regional biodiversity. Ecological Applications 3, 209–212. E., Anderson, B., Bradbury, J., DeGaetano, A., Troy, T.J., Wolfe, D., 2007. Past and Naiman, R.J., Décamps, H., 1997. The ecology of interfaces – riparian zones. Annual future changes in climate and hydrological indicators in the US Northeast. Review of Ecology and Systematics 28, 621–658. Climate Dynamics 28, 381–407. Nash, J.E., Sutcliffe, J.V., 1970. River flow forecasting through conceptual models. Haylock, M.R., Cawley, G.C., Harpham, C., Wilby, R.L., Goodess, C.M., 2006. Journal of Hydrology 10, 282–290. Downscaling heavy precipitation over the United Kingdom: a comparison of Pope, V., Gallani, M.L., Rowntree, P.R., Stratton, R.A., 2000. The impact of new dynamical and statistical methods and their future scenarios. International physical parameterizations in the Hadley Centre climate model: HadAM3. Journal of Climatology 26, 1397–1415. Climate Dynamics 16, 123–146. Hewitson, B., 2003. Developing perturbations for climate change impact Prudhomme, C., Reynard, N., Crooks, S., 2002. Downscaling of global climate models assessments. Eos, Transactions, American Geophysical Union 84, 337–348. for flood frequency analysis: Where are we now? Hydrological Processes 16, Heywood, I., Cornelius, S., Carver, S., 2006. An Introduction to Geographical 1137–1150. Information Systems, third ed. Prentice Hall, New Jersey. 464 pp. Prudhomme, C., Jakob, D., Svensson, C., 2003. Uncertainty and climate change Hirst, A.C., Gordon, H.B., O’Farrell, S.P., 1996. Global warming in a coupled climate impact on the flood regime of small UK catchments. Journal of Hydrology 277, model including oceanic eddy-induced advection. Geophysical Research Letters 1–23. 23, 3361–3364. Quilbé, R., Rousseau, A.N., Moquet, J.-S., Trinh, N.B., Dibike, Y., Gachon, P., Chaumont, Hirst, A.C., O’Farrell, S.P., Gordon, H.B., 1999. Comparison of a coupled ocean- D., 2008. Assessing the Effect of climate change on river flow using general atmosphere model with and without oceanic eddy-induced advection. Part 1: circulation models and hydrological modelling–Application to the Chaudière Ocean spin-up and control integrations. Journal of Climate 13, 139–163. River, Québec, Canada. Canadian Water Resources Journal 33, 73–94. Hodgkins, G.A., Dudley, R.W., 2006. Changes in the timing of winter–spring Roeckner, E., Arpe, K., Bengtsson, L., Brinkop, S., Dümenil, L., Esch, M., Kirk, E., streamflows in eastern North America, 1913–2002. Geophysical Research Lunkeit, F., Ponater, M., Rockel, B., Suasen, R., Schlese, U., Schubert, S., Letters 33, L06402. doi:10.1029/2005GL025593. Windelband, M., 1992. Simulation of the Present-day Climate with the C. Boyer et al. / Journal of Hydrology 384 (2010) 65–83 83

ECHAM4 Model: Impact of Model Physics and Resolution. Max-Planck Institute Verseghy, D.L., 1996. Local climates simulated by two generations of Canadian GCM for Meteorology, Report No. 93, Hamburg, Germany, 171 pp. land surface schemes. Atmosphere–Ocean 34, 435–456. Seguí, P.Q., Ribes, A., Martin, E., Habets, F., Boé, J., 2009. Comparison of three Visbeck, M., Hurrell, J.W., Polvani, L., Cullen, H.M., 2001. The North Atlantic downscaling methods in simulating the impact of climate change on the oscillation: past present and future. Proceeding of the National Academy hydrology of Mediterranean basins. Journal of Hydrology. doi:10.1016/ Science 98, 12 876–12 877. j.jhydrol.2009.09.050. Whitfield, P.H., Cannon, A.J., 2000. Recent variations in climate and hydrology in Sen, P.K., 1968. Estimates of the regression coefficient based on Kendall’s Tau. Canada. Canadian Water Resources Journal 25, 19–65. Journal American Statistics Association 63, 1379–1389. Whitfield, P.H., Wang, J.Y., Cannon, A.J., 2003. Modelling future streamflow Stewart, I.T., Cayan, D.R., Dettinger, M.D., 2004. Changes in snowmelt runoff timing extremes—floods and low flows in Georgia Basin, British Columbia. Canadian in Western North America under a ‘business as usual’ climate change scenario. Water Resources Journal 28, 633–656. Climatic Change 62, 217–232. Wilby, R.L., Wigley, T.M.L., Conway, D., Jones, P.D., Hewitson, B.C., Main, J., Wilks, St-Hilaire, A., Lachance, M., Bobée, B., Ouarda, T.B.M.J., Gignac, C., Rioux, P.-J., D.S., 1998. Statistical downscaling of general circulation model output: A Gaudet, J., Thiémonge, N., Pion, C., 2003. Assessment of the impact of comparison of methods. Water Resources Research 34, 2995–3008. meteorological network density on the estimation of basin precipitation and Wilby, R.L., Charles, S.P., Zorita, E., Timbal, B., Whetton, P., Means, L.O., 2004. runoff: a case study. Hydrological Processes 17, 3561–3580. Guidelines for Use of Climate Scenarios Developed from Statistical Thompson, D., Wallace, J., 2001. Regional climate impacts of the Northern Downscaling, IPCC Task Group on Scenarios for Climate Impact Assessment Hemisphere annular mode. Science 293, 85–89. (TGCIA). Toner, M., Keddy, P., 1997. River hydrology and riparian wetlands: A predictive Zhang, X., Harvey, K.D., Hogg, W.D., Yuzyk, T.R., 2001. Trends in Canadian model for ecological assembly. Ecological Applications 7, 236–246. streamflow. Water Resources Research 37, 987–998.