FEBRUARY 2017 I S L A M E T A L . 473

Future Climate Change Impacts on Snow and Water Resources of the Fraser River Basin, British Columbia

SIRAJ UL ISLAM AND STEPHEN J. DÉRY Environmental Science and Engineering Program, University of Northern British Columbia, Prince George, British Columbia, Canada

ARELIA T. WERNER Pacific Climate Impacts Consortium, University of Victoria, Victoria, British Columbia, Canada

(Manuscript received 31 December 2015, in final form 9 November 2016)

ABSTRACT

Changes in air temperature and precipitation can modify snowmelt-driven runoff in snowmelt-dominated regimes. This study focuses on climate change impacts on the snow hydrology of the Fraser River basin (FRB) of British Columbia (BC), Canada, using the Variable Infiltration Capacity model (VIC). Statistically downscaled forcing datasets based on 12 models from phase 5 of the Coupled Model Intercomparison Project (CMIP5) are used to drive VIC for two 30-yr time periods, a historical baseline (1980–2009) and future projections (2040–69: 2050s), under representative concentration pathways (RCPs) 4.5 and 8.5. The ensemble-based VIC simulations reveal widespread and regionally coherent spatial changes in snowfall, snow water equivalent (SWE), and snow cover over the FRB by the 2050s. While the mean precipitation is pro- jected to increase slightly, the fraction of precipitation falling as snow is projected to decrease by nearly 50% in the 2050s compared to the baseline. Snow accumulation and snow-covered area are projected to decline substantially across the FRB, particularly in the Rocky Mountains. Onset of springtime snowmelt in the 2050s is projected to be nearly 25 days earlier than historically, yielding more runoff in the winter and spring for the Fraser River at Hope, BC, and earlier recession to low-flow volumes in summer. The ratio of snowmelt contribution to runoff decreases by nearly 20% in the Stuart and Nautley subbasins of the FRB in the 2050s. The decrease in SWE and loss of snow cover is greater from low to midelevations than in high elevations, where temperatures remain sufficiently cold for precipitation to fall as snow.

1. Introduction et al. 2007). The Fraser River basin (FRB) of British Columbia (BC), where complex topography dominates Projected changes in air temperature and precipitation, the landscape, is a prime example of a susceptible basin. It associated with increasing greenhouse gases, will have spans one-fourth of BC, covering roughly 230 000 km2,and wide-ranging impacts on the snowmelt-dominated hydro- thus forms a large basin with significant environmental, logical regimes of western North America. Any change in economic, and cultural importance. The presence of a these driving factors has the potential to modify the dis- seasonal snowpack in the FRB is an essential component tribution of mountainous snowpack that serves as a natural of its water resources and serves as a sensitive indicator of reservoir for cold-season precipitation (Barnettetal. climate change. Snowmelt in the FRB plays an important 2005). Western Canada’s snowmelt-dominated regimes role in regulating water temperature, which supports the are no exception; they too are susceptible to these climate reproductive fitness of salmon (Rand et al. 2006). Change change–driven modifications (Morrison et al. 2002; Ferrari in FRB snowpack levels, extent, and duration will alter the associated snowmelt runoff and water temperatures and may reduce the survival of aquatic species such as salmon, Denotes content that is immediately available upon publica- which in turn will affect local communities and the econ- tion as open access. omy of western Canada. There have been several studies investigating the Corresponding author e-mail: Stephen J. Déry, [email protected] impacts of climate change on snow and its contribution

DOI: 10.1175/JHM-D-16-0012.1 Ó 2017 American Meteorological Society. For information regarding reuse of this content and general copyright information, consult the AMS Copyright Policy (www.ametsoc.org/PUBSReuseLicenses). Unauthenticated | Downloaded 10/02/21 03:48 PM UTC 474 JOURNAL OF HYDROMETEOROLOGY VOLUME 18 to the hydrology of western North America. Mote et al. Gan (2011) projected earlier onset of spring snowmelt (2005) found a declining trend in 1 April snow water and decreased monthly annual peak flow by the end of equivalent (SWE) across western North America from the twenty-first century for the Fraser River at Hope 1925 to 2000 that was driven mainly by rising air tem- using the Interactions between Soil, Biosphere, and perature. Using observational datasets, Park et al. Atmosphere (ISBA) hydrological model driven by (2012) reported declining trends in snow depth across seven global climate models (GCMs). Ferrari et al. North America between 1948 and 2006. Increased (2007) and Shrestha et al. (2012) found that parts of the freezing levels (Abatzoglou 2011) and amplified warm- FRB may already be transitioning to a hybrid, and ing at high elevations (Rangwala and Miller 2012) were perhaps even a rain-dominated, system due to significant related to climate change. Choi et al. (2010) reported a declines of snow accumulations over recent decades, general decline in snow cover duration in BC using re- especially in its lowlands and the Interior Plateau. mote sensing data for 1972–2008. Déry et al. (2012) The seasonal variability and amount of water released concluded that there is increasing interannual variability from the FRB’s snowpack sets the pace of regional ir- in runoff across many streams and rivers in the FRB, rigation, water management, and hydropower genera- which create flow conditions unfavorable to salmon that tion. Assessing how the FRB’s freshwater resources may are related to changing snow cover conditions. Danard change in the future in a way that allows land and water and Murty (1994) reported a decline in SWE across the managers, planners, and government to make informed FRB between 1966 and 1989 concurrent with streamflow decisions requires rigorous scientific attention. This in- decreases in October. Using hydrological model simu- cludes the use of improved future climate projections to lations spanning 1948–2006, Kang et al. (2014) drive hydrological models selected for their represen- revealed a 19% decline in the contribution of snow to tation of key processes in the FRB. The use of large runoff generation for the Fraser River at Hope, BC, GCM datasets in comprehensive hydrologic-model- induced mainly by climate warming. Kang et al. (2016) based projections is becoming more common because showed that the rapidly declining mountain snowpack of the advancement in GCM dynamics as well as the and earlier melt onsets result in a 10-day advance of the availability of their output datasets. It is therefore im- Fraser River’s spring freshet, with subsequent re- portant to use a combination of such tools over the FRB ductions in summer flows over the historical period. to project climate change impacts on its snowpack and Additional studies examined the teleconnections of El hydrology. Particular attention to the FRB’s moun- Niño–Southern Oscillation (ENSO) and Pacific decadal tainous terrain, which exhibits a wide variety of eleva- oscillation (PDO) to western Canada’s hydrology and tion distributions, is also advisable. Given the strong, snow (Moore 1991; Hsieh and Tang 2001; Whitfield et al. often linear, relationship between snowpack and eleva- 2010). Hsieh and Tang (2001) investigated the influence tion (Abatzoglou 2011; Sospedra-Alfonso et al. 2016), of teleconnections on 1 April snowpack accumulations snowpack loss will vary from catchment to catchment in the upper Columbia River basin, with La Niñaanda within different elevations. In estimating future impacts, low Pacific–North American pattern yielding large posi- it is imperative that one simulates and characterizes the tive SWE anomalies. Rodenhuis et al. (2009) reported rate of projected warming-driven changes at different that SWE was lower than average over most of southern elevations within the FRB. BC during warm PDO phases. Shrestha et al. (2016) While the historical hydrology of the FRB has been noted lower flows in the FRB during warm PDO phases addressed in previous studies, the current literature and higher flows in the FRB during cool PDO phases. lacks quantification of future changes in FRB snowpack Climate change is projected to continue into the levels and their variation with different elevations using twenty-first century and its impacts along with it. Over phase 5 of the Coupled Model Intercomparison Project the western United States, Rauscher et al. (2008) found (CMIP5; Taylor et al. 2012) GCM projection forcings. A that by the end of the twenty-first century, projected quantitative evaluation of the CMIP5-projected climate future warming could cause snow-driven runoff to occur change and its impact on the FRB’s snowpack requires a as much as 2 months earlier than present because of an systematic analysis of GCM-driven hydrological model amplified snow-albedo feedback. Elsner et al. (2010) simulations that has not been accomplished thus far. The reported that 1 April SWE is projected to decrease by present study therefore investigates the impact of in- approximately 38%–46% by the 2040s over Washington creasing air temperature, coupled with changing pre- State and the greater Columbia River basin. Vano et al. cipitation phases, on the FRB’s mountainous snowpack (2010) found declines, and eventual disappearance, of accumulation and melting as well as its contribution to the springtime snowmelt peak in the Puget Sound Basin runoff generation. Moreover, the sensitivity of moun- by the 2040s and 2080s, respectively. Kerkhoven and tainous elevations to such climate change is examined in

Unauthenticated | Downloaded 10/02/21 03:48 PM UTC FEBRUARY 2017 I S L A M E T A L . 475 terms of snowpack levels. Thus, the overarching goal of this study is to project changes in FRB snow hydrology with an emphasis on seasonal variations. The simulation tool in this study is the hydrological macroscale Variable Infiltration Capacity model (VIC), which is a surface water and energy balance model designed for large-scale hydrological regions such as the FRB (Cherkauer et al. 2003). To evaluate spatially dis- tributed future responses at 1/48 grid cells, VIC is forced by statistically downscaled outputs of CMIP5 GCMs (Taylor et al. 2012) used in the Intergovernmental Panel on Climate Change (IPCC) Fifth Assessment Report (AR5; IPCC 2014). CMIP5 GCMs run under represen- tative concentration pathways (RCPs) 4.5 and 8.5, two different pathways with stabilized (RCP 4.5) and rising (RCP 8.5) greenhouse gas emissions, over the future time period are considered. Analyses are performed on the daily and annual data for the entire FRB, six of its major subbasins, and three of its hydroclimatic sub- regions, contrasting projected conditions in the future (2040–69) time periods to those from the historical baseline (1980–2009). To estimate future changes re- lated to the snowmelt-dominated regions of the FRB, SWE, snow cover, and runoff variables are analyzed along with the ratio of snowmelt to runoff to estimate the direct contribution of snow to runoff generation for the FRB. These snow hydrology variables are crucial for estimating climate change impacts on the FRB’s snow- melt-driven runoff.

2. Fraser River basin The FRB forms one of the largest and most important basins of western North America, with elevations ranging from sea level to its tallest peak Mount Robson at 3954 m (Benke and Cushing 2005). It spans 240 000 km2 of di- verse landscapes from the dry Interior Plateau to the cool, snowy Rocky Mountains to the northeast and the Coast Mountains to the west (Fig. 1a). Because of its varying terrain, the FRB covers 11 biogeoclimatic zones, in- cluding various vegetation and land-cover types such as alpine terrain, subalpine, montane, and coastal forests, with drier inland forests and grasslands in the Interior Plateau (Bocking 1997; Benke and Cushing 2005). Some of the major tributaries of the Fraser River are the Nechako, Quesnel, Chilcotin, and Thompson Rivers within the Interior Plateau (Kang et al. 2014). Owing to its complex topography, the FRB exhibits FIG. 1. (a) Digital elevation map of the FRB with identification substantial spatial variation of mean annual air tem- of major subbasins: Fraser–Shelley (UF), Stuart (SU), Nautley perature and precipitation. Mean annual air tempera- (NA), Quesnel (QU), Chilko (CH), Thompson–Nicola (TN), and ture ranges from 0.58C in the mountains in the northwest Fraser at Hope (LF). (b) FRB mean elevation (m) per VIC grid cell (see Table 1 for further details). to 7.58C in the lower-elevation area in the south, close to BC’s Okanagan region. The interior portions of the FRB

Unauthenticated | Downloaded 10/02/21 03:48 PM UTC 476 JOURNAL OF HYDROMETEOROLOGY VOLUME 18 have warmer air temperature on average than the high- parameterized. VIC requires a large number of param- elevation Coast and Rocky Mountains. The range in mean eters, including soil, vegetation, elevation, and daily daily air temperature in summer (from 11.08 to 16.58C) is meteorological forcings, at each grid cell (Liang et al. less than the range in winter (from 211.08 to 21.08C). 1994; 1996). To convert runoff and base flow from grid Precipitation varies greatly across the FRB. The cells into discharge from a basin, an external routing 2 mountainous coastal areas receive up to 3000 mm yr 1 model (Lohmann et al. 1996; 1998a,b) needs to be em- of precipitation while the Interior Plateau and regions ployed once the VIC simulation is completed. The to the lee side of the Coast Mountains receive only 400– routing model simulates a channel network with 2 800 mm yr 1 (Kang et al. 2014). The Interior Plateau is a number of nodes, each of which represents in- the driest section in the FRB. More precipitation falls formation from a grid cell. Detailed descriptions of VIC in the Coast and Rocky Mountains, predominantly as are available in Liang et al. (1994, 1996) and Gao et al. snow. Snow normally accumulates throughout the (2009). winter in the FRB, except perhaps at lower elevations VIC (version 4.1.2) with more recent modifications, and coastal areas. Peak flow in the main stem of the such as a parameterization of blowing-snow sublimation Fraser River and its many tributaries usually occurs in and inclusion of lakes and wetlands (Bowling et al. 2004; late spring and early summer and is driven by snowmelt Bowling and Lettenmaier 2010), has been widely ap- primarily from high elevations (Moore and Wondzell plied to assess water resources, land–atmosphere in- 2005). The snowmelt-driven discharge declines rapidly teractions, and hydrological responses over various river following the depletion of the snow storage in late basins around the world (Nijssen et al. 2001a,b; Maurer summer and plays an important role in the regional et al. 2002; Su et al. 2005; Haddeland et al. 2007; Adam water supply. Hydrologic response varies considerably et al. 2009; Gao et al. 2010; Ashfaq et al. 2010; Oubeidillah across the FRB, including snowmelt, hybrid (rainfall et al. 2014; Zhou et al. 2016). As climate change has been and snowmelt), and rainfall-dominated regimes (Wade a critical issue since the 1990s, VIC has also been ex- et al. 2001). tensively used with various forcing datasets provided by Changes to the climate of the FRB over the last few outputs from climate and weather prediction models decades have been documented. Based on the Uni- (Wang et al. 2008; Shukla et al. 2012; 2013). VIC is used versity of Washington gridded observational climate for climate change research because it is a process-based dataset (Shi et al. 2013; Adam and Lettenmaier 2008), model that should respond to changing air temperature the 1948–2006 annual precipitation shows no clear and precipitation in a more physically realistic way than trend over the 1948–2006 record, whereas a warming empirically based or lumped models (Dwarakish and of 1.58C in annual mean temperature occurred over Ganasri 2015). It is also commonly used to simulate the interior sections of the FRB. The northernmost hydrologic response to climate change in snowmelt- subbasins experienced the greatest warming rate, at dominated basins (Christensen and Lettenmaier 2007; 2 2 0.488C decade 1 as compared to 0.288Cdecade 1 for Hidalgo et al. 2009; Cherkauer and Sinha 2010; Schnorbus the remainder of the FRB (Kang et al. 2014). Such et al. 2011). changes have modified the ratio of snowfall to rainfall b. Elevation bands and induced considerable changes in the hydrological regime of the FRB. In mountainous regions with a significant range in elevation, VIC can lead to inaccurate snowpack esti- 3. Hydrological model and data mates if the effects of elevation on snowpack accu- mulation and ablation are not modeled properly. To a. VIC simulate the hydrology of complex terrain, VIC can be VIC is a semidistributed macroscale model (Liang set up by dividing each grid cell into a number of snow et al. 1994, 1996) that resolves energy and water balance elevation bands (Nijssen et al. 2001b). The subgrid by grid cells that are subdivided into elevation bands variability in topography and precipitation is then with equal vegetation fractions per band to represent modeled with a mosaic-type representation by parti- subgrid variability of the land surface. Base flow and tioning elevation bands into a number of topography runoff from a grid cell are determined at the end of each tiles. These tiles are based on high-resolution spatial time step. To consider subgrid variability of the land elevations and fractional area. Each band’s mean ele- surface, VIC uses a statistical approach to cope with vation is used to lapse the gridcell-average air tem- different land and soil types that control soil moisture perature, pressure, and precipitation to a more storage capacity. It captures subgrid variability of accurate local estimate. The snow model is then ap- snow depth using a statistical function that can be plied to each elevation tile separately. The simulated

Unauthenticated | Downloaded 10/02/21 03:48 PM UTC FEBRUARY 2017 I S L A M E T A L . 477 output variables (SWE, snow cover, etc.) for each grid correction was applied using detrended quantile mapping cell are calculated as the area averages of the tiles with delta method extrapolation following Bürger et al. (Gao et al. 2009). (2013). Although BCSD has been applied to daily GCM In this study, 10 elevation bands were used in each grid data and there are other downscaling techniques that have cell (1/48) to better characterize future climate change proven success with downscaling daily data, the monthly impacts on different elevations. Bands were divided into application was maintained, again to align with other different elevation bins based on the minimum and studies and also because the daily evolution of precipitation maximum elevations at finer-spatial-resolution tiles and temperature events was not thought to be necessary to (30-arc-s DEM) over the FRB. The total area and av- investigate the ratio of annual snowmelt to runoff or total erage elevation of the tiles that fall into each bin are contribution of snow to runoff generation in the FRB, the equal to the band area and elevation, respectively. This primary focus of this paper. The target historical dataset allows in-depth analysis of the variation of SWE and (1950–2005) used in the bias correction and spatial snow cover with elevation that is of particular impor- disaggregation was the 10-km resolution observation- tance for the Rocky and Coast Mountains of the FRB, based daily gridded climate data interpolated using the which cover a wide range of elevations. method implemented by the Australian National Uni- versity spline (ANUSPLIN hereafter) (McKenney c. Climate projections data et al. 2011; Hopkinson et al. 2011). Statistically downscaled outputs of 12 CMIP5 GCM projections were used as forcing datasets for the VIC 4. VIC implementation and methodology simulations. The CMIP5 projections (Taylor et al. 2012) were generated using a set of improved GCMs that The model was run at 1/48 spatial resolution using a collectively reflect varying degrees of advancement in daily time step over the entire domain of the FRB in climate science and modeling since CMIP3. The CMIP5 water balance mode along with the built-in snow GCMs were run at higher spatial resolution, with some model. The exception to this is that the snow algorithm models being more comprehensive in terms of their still solves the surface energy balance to determine the physical dynamics (Knutti and Sedlácek 2012). A new fluxes needed to drive accumulation and ablation pro- set of climate forcing scenarios defined as RCPs were cesses. Integrations of VIC over the entire FRB were used to develop CMIP5 climate projections (van Vuuren performed using 34 rows and 42 columns of model grid et al. 2011). These RCPs reflect recent progress in in- cells spanning 488–558Nand1198–1318W. Different tegrated assessment modeling to characterize future tiles characterizing soils and vegetation were propor- greenhouse gas emissions since the scenarios released tionally partitioned within a grid cell. In this model under the Special Report on Emissions Scenarios application, three soil layers were used with depth (SRES; Nakicenovic and Swart 2000). Each of the 12 ranges of 0–0.3, 0.4–0.9, and 1.0–2.0 m, in the top, GCMs, as run under the pathways of stabilized (RCP middle, and bottom layers, respectively. The soil and 4.5) and rising (RCP 8.5) greenhouse gas emissions, vegetation parameters, leaf area index (LAI), and al- were statistically downscaled and used to drive VIC, bedo data were implemented as they were in Kang et al. for a total of 24 GCM–RCP simulations. (2014). Other soil parameters were calibrated based on In the downscaling process, GCMs were selected the agreement between simulated and observed hy- based on the subset with the widest spread in future drographs (Mitchell et al. 2004). projected climate for western North America (Giorgi a. VIC calibration and validation and Francisco 2000), following Cannon (2015). Down- scaled GCMs datasets for the FRB were accessed VIC’s ability to simulate FRB hydrology has been through the Pacific Climate Impacts Consortium (PCIC) reported in recent work by Kang et al. (2014),which data portal (https://pacificclimate.org/data/statistically- provides details of the model implementation, in- downscaled-climate-scenarios). Those downscaled us- cluding the approach to calibration and validation. ing the bias-correction spatial disaggregation (BCSD; For the current study, the VIC configuration (resolu- Wood et al. 2004) were selected to align this work with tion, soil layers and datasets, snowbands, etc.) was others in the region (Schnorbus et al. 2014; Shrestha maintained as in Kang et al. (2014),excepttheforcing et al. 2012; Elsner et al. 2010; Werner et al. 2013). This data were changed to the ANUSPLIN. (Hutchinson implementation of BCSD followed Maurer and Hidalgo et al. 2009; McKenney et al. 2011; Hopkinson et al. (2008) and was modified to incorporate monthly minimum 2011). This is because the CMIP5 GCMs were down- and maximum air temperature instead of monthly mean air scaled using the ANUSPLIN observation dataset, and temperature, as suggested by Bürger et al. (2012). The bias forcing VIC with these downscaled outputs required

Unauthenticated | Downloaded 10/02/21 03:48 PM UTC 478 JOURNAL OF HYDROMETEOROLOGY VOLUME 18 the model to be calibrated using the same observa- TABLE 1. Details of the Fraser River main stem at Hope (i.e., tions.TheANUSPLINdataweredevelopedbyNat- LF) and major subbasins within the FRB. The list includes mean ural Resources Canada (NRCan) and contain gridded elevation, gauged area, percentage gauged area relative to LF, and 8 latitude and longitude of the gauge at the outlet of each subbasin station data of daily maximum air temperature ( C), (Déry et al. 2012). minimum air temperature (8C), and total precipitation (mm) for the Canadian landmass south of 608Nat;10- Gauged Mean basin Gauged area relative Lat Lon km resolution (NRCan 2014). Subbasin elev (m) area (km2) to LF (%) (8N) (8W) Previous studies that applied the ANUSPLIN dataset to modeling watersheds with complex topography in BC UF 1413 32 400 14.9 54.01 122.62 SU 1097 14 200 6.5 54.42 124.27 and Alberta revealed underestimates in runoff, owing NA 1070 6030 2.8 54.08 124.59 to a dry bias in ANUSPLIN precipitation at high alti- QU 1391 11 500 5.3 52.84 122.22 tudes (Eum et al. 2014). As the low precipitation values CH 1756 6940 3.2 52.07 123.54 would make it impossible to calibrate VIC, a pre- TN 1363 54 900 25.3 50.35 121.39 cipitation correction was applied to the ANUSPLIN Total 125 970 58.1 — — LF 1320 217 000 100.0 49.38 121.45 data prior to calibration instead of including a pre- cipitation adjustment in the calibration parameters. This correction was applied to adjust for the dry bias found most notably in the FRB’s mountainous areas. Gridded population method to maximize the NSE coefficient observations (1/168) developed by PCIC following the between observed and simulated runoff. Six training technique of Maurer et al. (2002) and Hamlet and soil dataset parameters were used in the optimization Lettenmaier (2005) were regridded using bilinear in- process, that is, a parameter of the variable infiltration terpolation to match the scale of the current VIC im- curve (b_infilt), the maximum velocity of base flow for plementation (1/48). The precipitation adjustments were each grid cell (Dsmax), the fraction of maximum soil applied at each grid cell, based on the daily climatology moisture where nonlinear base flow occurs (Ws), the bias for the 1979–2006 period of overlap between the depths of the second and third soil layers (D2 and D3), two datasets. Although elevation was not a direct de- and the fraction of the Dsmax parameter at which non- terminate in the bias correction, the difference between linear base flow occurs (Ds). By minimizing the differ- PCIC and ANUSPLIN did have a relationship with el- ence (increasing the NSE coefficient) between the evation due to the increased number of stations and simulated and corresponding observed monthly flow, fi- adjustment of precipitation in the PCIC gridded climate nal values of these six calibrated parameters were esti- dataset with the Parameter-Elevation Regressions mated and were used for all the simulations for their on Independent Slopes Model (PRISM) data [see respective subbasin and for the Fraser River at Hope. Schnorbus et al. (2011) for more details]. To calibrate and validate the simulated flow by VIC, The same adjustment was used to correct the BCSD- daily streamflow data from Environment and Climate downscaled GCM projections prior to the simulations. Change Canada’s Hydrometric Dataset (HYDAT; The calibration of VIC was performed for the years Water Survey of Canada 2014) were used. Déry et al. 1979–90, while the years 1991–2006 were used to vali- (2012) extracted and compiled a comprehensive date the model. These years were chosen to remain streamflow dataset for the FRB spanning 1911–2010. consistent with the CMIP5 baseline period as well as to Once VIC was calibrated and validated, the series of avoid a shift of the PDO phase in year 1976/77 (Marcus VIC simulations were performed for base and future et al. 2011) during the calibration period. The calibra- time periods for the years 1980–2009 (1990s hereafter) tion used the optimization process conducted by min- and 2040–69 (2050s hereafter), respectively. Each imizing the difference between observed and simulated simulation was initiated 5 years prior to the 1990s and monthly streamflow at Hope using the Nash–Sutcliffe 2050s time periods to allow model spinup. VIC was efficiency (NSE) coefficient (Nash and Sutcliffe 1970). driven by the statistically downscaled GCMs output Along with the Fraser River at Hope the calibration data of daily precipitation and maximum and minimum was applied to six subbasins to determine the optimal air temperature. Daily wind speed data were created by parameters by comparing simulated to observed regridding estimates of 10-m wind speed from the Na- streamflow in each basin. The University of Arizona tional Centers for Environmental Prediction–National multiobjective complex evolution (MOCOM-UA) Center for Atmospheric Research (NCEP–NCAR) optimizer was selected for the calibration process reanalyses (Kalnay et al. 1996; Schnorbus et al. 2011) (Yapo et al. 1998; Shi et al. 2008). MOCOM-UA because of their unavailability with the downscaled searches a set of VIC input parameters using the CMIP5 datasets on the PCIC data portal and overall

Unauthenticated | Downloaded 10/02/21 03:48 PM UTC FEBRUARY 2017 I S L A M E T A L . 479 challenges with downscaling wind. The VIC output which is higher in December–March than the other re- variables were found insensitive to different wind gions reflecting a rainfall-driven response (Shrestha forcing datasets during the VIC sensitivity experi- et al. 2012). Figure 1b shows the gridcell division of these ments. Downscaled GCM projections (;10 km), 12 three regions and their elevations. models for both RCP 4.5 and RCP 8.5, were regridded To capture total SWE melt at all locations, including to a common resolution of 1/48 to match the resolution those at high elevations where snow does not disappear of VIC. entirely every year, we use the difference between SWE and SWE . We then quantified the contribu- b. VIC ensembles max min tion of snow to runoff generation across the basin RSR for A realistic model always contains random compo- each water year as (Déry et al. 2005; Kang et al. 2014): nents and uncertainties such as those in boundary forc- N ing or in initial conditions. To alleviate the impact of 5 = å RSR SWEmelt Rt , these random fluctuations the Multi-Model Ensemble t51

(MME) strategy was used throughout the analyses. 2 R 1 N While the simulated climate change varies, to a smaller where is the runoff (mm day ) and is 365 or 366 t 5 or larger extent, between different CMIP5 models, the for a leap year, with 1 indicating 1 October of a given MME can provide better estimates of first-order future water year. The SWEmelt is defined by change as compared to single models (Duan and Phillips 5 2 SWEmelt SWEmax SWEmin , 2010; Miao et al. 2013). Each independent simulation of

VIC, driven by individual GCM output, was used to where SWEmax is max(SWE)t51,.::,N and SWEmin is produce the MME mean. The intermodel spread among min(SWE)t51,.::,N. GCMs was also estimated along with MME and was For each VIC simulation, streamflow was converted to made up of 12 GCMs run under two RCPs (4.5 and 8.5) areal runoff by dividing it by the corresponding subbasin for a 24-member ensemble. area. Daily runoff at the outlet cell was integrated over time to obtain total annual runoff for a selected basin. c. Analysis strategy Future changes were calculated between the 2050s Analyses were performed for the entire FRB and its versus 1990s for both the RCP 4.5 and RCP 8.5 driving six major subbasins, namely, the upper Fraser at Shelley datasets of the VIC outputs for 12 GCMs. Seasonal var- (UF), Stuart (SU), Nautley (NA), Quesnel (QU), iations were assessed by averaging December–February Chilko (CH), and Thompson–Nicola (TN) basins (DJF), March–May (MAM), June–August (JJA), and (Fig. 1a, Table 1). The six subbasins examined in the September–November (SON) months for winter, spring, present study contribute 75% of the annual observed summer, and autumn, respectively. Figure 2 shows an discharge for the Fraser River at Hope (LF), with the overall framework of this study highlighting the GCMs largest contributions from the TN, UF, and QU sub- datasets, VIC configuration, and analysis approach. basins (Déry et al. 2012). However, the focus of this study is on the main stem of the Fraser River at Hope 5. Results and discussion since it is farthest downstream, covering 94% of the a. VIC performance basin’s drainage area, and has a continuous streamflow record over the 1980–2009 historical baseline period. To Table 2 summarizes the performance of the VIC cal- simulate streamflow for these subbasins, a routing net- ibration and validation process for LF and all six sub- work was adapted from Wu et al. (2011). basins in the FRB. The NSE and correlation values The analyses were further expanded for three FRB reveal a good agreement between modeled streamflow hydroclimatic regimes: the Interior Plateau (central part and observations, exhibiting NSE scores greater than of the basin), the Rocky Mountains (Rocky and Co- 0.70 for most subbasins. The calibrated results, including lumbia Mountains in the eastern part of the basin), and low-flow volumes and the timing of runoff peaks, match the Coast Mountains (southwestern part of the basin; those in observations for all subbasins in the FRB Moore 1991). Considering that the future responses for closely, except Fraser–Shelley (Fig. 3). Model perfor- each region could vary considerably, these hydro- mance is best at the outlet, Fraser at Hope, with high climatic regions represent the FRB’s distinct land sur- NSE and correlation values and relatively low bias in face features and hydroclimatic conditions. For both the calibration and validation periods. The larger example, this division helps to evaluate the snow hy- Thompson–Nicola subbasin is also well calibrated con- drology of the Coast Mountains, especially in the lower sidering the high NSE values, whereas the smaller head- Fraser Valley, where there is a unique runoff pattern, water basins, especially the Nautley and Fraser–Shelley,

Unauthenticated | Downloaded 10/02/21 03:48 PM UTC 480 JOURNAL OF HYDROMETEOROLOGY VOLUME 18

FIG. 2. Block diagram of experimental setup and analysis. are the only basins with NSE scores less than 0.70. These perennial snowpack were compared to baseline the- discrepancies are probably due to the problems in the matic mapping to confirm their location (as per observed forcing data, which tend to be more acute at higher el- glaciers) and were masked in the SWE analysis. None- evations where there are fewer meteorological stations. theless, the effects of glaciers may not change the results Additionally, they are more common on the lee side of significantly as VIC is implemented on ;25-km gridcell mountains, where interpolation procedures commonly resolution. Furthermore, glaciers cover only 1.5% of the do not weigh windward and leeward stations accordingly FRB (Shrestha et al. 2012) and provide only a modest (Daly 2006). contribution to streamflow at the Fraser River at Hope, Some of the discrepancies between observations and primarily in late summer (August/early September). simulations may be due to glacier dynamics not being The reasonably high validation results of the VIC represented in the model physics of the version of VIC simulations over the FRB demonstrate that it is a useful applied in this study. However, there are some grid cells simulation tool to assess future climate change impacts. in the FRB domain where a perennial snowpack exists The calibration and validation periods were chosen to that does not melt out, year after year. This is a recog- be the periods before and after the transition from the nized challenge with VIC and is dealt with differently in cool to warm PDO phases, respectively. VIC was also each implementation, ranging from eliminating these tested for sensitivity to variable climate with a series of cells from analysis to introducing a simple, conceptual sensitivity experiments (results not shown). representation of glacier mass balance into VIC to b. CMIP5 precipitation and air temperature model perennial snow with VIC’s built-in snow routines. climatology In this approach, a portion of the model grid cells is identified as glacierized and used to form a glacier mask Being bias corrected and spatially disaggregated, the (Schnorbus et al. 2011). In this study, those cells with downscaled CMIP5 GCMs simulate the baseline mean

TABLE 2. The NSE and correlation coefficients r (all significant at the p level of 0.05) and the relative biases for the VIC calibration (1979–90) and validation (1991–2006) periods for six major subbasins of the FRB and Fraser River main stem at Hope (i.e., LF).

Calibration (1979–90) Validation (1991–2006) Daily Monthly Annual Daily Monthly Annual Subbasins NSE r NSE r Bias (%) NSE r NSE r Bias (%) UF 0.71 0.93 0.82 0.96 228.7 0.62 0.91 0.73 0.95 233.7 SU 0.90 0.95 0.92 0.96 0.9 0.87 0.96 0.89 0.96 212.0 NA 0.64 0.83 0.82 0.91 1.9 0.77 0.91 0.89 0.94 22.9 QU 0.86 0.95 0.92 0.97 212.0 0.75 0.94 0.85 0.96 221.5 CH 0.80 0.91 0.88 0.94 23.6 0.82 0.92 0.89 0.94 1.4 TN 0.89 0.96 0.92 0.97 213.9 0.88 0.95 0.91 0.97 218.0 LF 0.83 0.94 0.94 0.98 29.6 0.85 0.95 0.98 0.93 29.4

Unauthenticated | Downloaded 10/02/21 03:48 PM UTC FEBRUARY 2017 I S L A M E T A L . 481

FIG. 3. Simulated (VIC) and observed (OBS) runoff for the Fraser River and its major subbasins for the calibration period (1979–90).

Unauthenticated | Downloaded 10/02/21 03:48 PM UTC 482 JOURNAL OF HYDROMETEOROLOGY VOLUME 18

FIG. 4. Area-averaged time series of CMIP5 MME mean daily air temperature and precipitation over the (a) Interior Plateau, (b) Rocky Mountains, and (c) Coast Mountains for the 1990s and 2050s (RCP 4.5 and RCP 8.5). Shading corresponds to intermodel uncertainties. climatology of precipitation and air temperature well by MME mean air temperature is projected to increase in design. MME precipitation shows a projected increase in all seasons, with higher seasonal increases projected un- winter (;6%), spring (;12%), and autumn (;12%) over der RCP 8.5 than RCP 4.5 in the 2050s. Mean annual air the northern and southern FRB for the 2050s while temperature increases range from 2.08 to 2.48C for RCP summer precipitation is projected to decrease by 12% 4.5 and 2.48 to 3.08C for RCP 8.5 across grid cells. Ab- over the same period, particularly in the southern FRB solute increases are greater in winter than in summer, under RCP 8.5 (not shown). Precipitation is not projected over the northern FRB, for both pathways. The projected to change by a high percentage during the winter in the air temperature change over the Interior Plateau is es- southern parts of the FRB, but some of the largest annual pecially notable in summer for RCP 8.5 considering the precipitation in the FRB occurs in this region. The spatial relatively high temperature already on record in this re- pattern of precipitation change for RCP 4.5 is more or less gion (not shown). similar to that for RCP 8.5; however, the magnitude of Figure 4 shows the CMIP5 MME mean daily clima- change is not as great (not shown). tology and intermodel spread of daily precipitation and

Unauthenticated | Downloaded 10/02/21 03:48 PM UTC FEBRUARY 2017 I S L A M E T A L . 483 air temperature areally averaged over the FRB’s three While the total precipitation slightly increases in the regions for the base and future periods. The seasonal 2050s, the simulated changes show future increases of up precipitation begins in November and persists until April, to 40% in mean annual rainfall and nearly 50% de- 2 approaching a maximum of 4.0, 6.0, and 10.0 mm day 1 in creases in mean annual snowfall (Fig. 5). The increased winter for the Interior Plateau, Rocky Mountains, and rainfall covers the Rocky and Coast Mountains while Coast Mountains, respectively, for the 1990s. The the decreased snowfall dominates the Interior Plateau maximum intraseasonal variability for CMIP5 MME and Rocky Mountains. These changes in total pre- precipitation occurs in the Coast Mountains, where cipitation and fractional snowfall have competing effects 2 daily rates range from nearly 0.0 mm day 1 in summer on the SWE distribution. 2 to 10.0 mm day 1 in winter. Daily precipitation rates The spatial pattern of projected SWE changes (as a are projected to change less in late spring and summer percentage) shows decreases over the Interior Plateau than in winter and autumn, when rates are projected to and Rocky and Coast Mountains for both winter and increase, particularly in the mountainous regions. The spring, with larger magnitudes in spring (Figs. 6a,b). range of intermodel spread for peak precipitation Most of the Interior Plateau has positive SWEmelt 2 varies between 8.0 and 12.0 mm day 1 during winter for change in winter, while the Rocky and Coast Mountains the Coast Mountains. Precipitation changes in the show nearly zero SWEmelt in the future owing to their Coast Mountains are more variable across GCMs than high elevations with subfreezing conditions (not shown). in other regions of the FRB because of its proximity to The areally averaged seasonal change of SWEmelt the Pacific Ocean, where interactions between steep (Fig. 6c) shows an earlier onset of about 25 (610) days elevation gain and storm-track position are more by the 2050s. This earlier melting of snow increases complex and are represented differently in each GCM. spring runoff. SWEmelt is projected to decrease in spring The high range of uncertainties in projections for and summer, yielding lower total contributions of SWE the Rocky and Coast Mountains is likely due to the to runoff generation. The intermodel range shows a inability of climate models to resolve complex large spread in future SWEmelt, particularly in late topography. spring and early summer. The annual cycle also shows a

Over the 1990s, air temperature remains below decrease in peak SWEmelt magnitude in the 2050s as 0.08C between November and March on average, compared to the 1990s being more obvious in RCP 8.5 rises above 0.08C in early spring, inducing snowmelt than RCP 4.5. The earlier mountain snowmelt and as- across all three FRB regions. In the 2050s, while the sociated runoff by the 2050s will result in a longer dry annual variation of air temperature remains similar season and diminished summer and autumn streamflow to the baseline, a consistent increase of from 2.08 to in the Interior Plateau. Snowpacks are projected to de- 3.08C in air temperature is projected for all three crease because of rising air temperatures accelerating regions. The range of intermodel spread is small in snowmelt and causing more precipitation to fall as rain winter and wide in summer, with a range of 61.58C versus snow and increasing the evapotranspiration (ET). around the warmest MME mean temperature for The projected decrease of seasonal snow cover over the RCP 8.5. In the case of RCP 4.5, the spread is almost entire FRB arises from warming air temperature (not thesameasthatforRCP8.5whilethemeantem- shown). The snowmelt-dominated areas diminish more perature is lower. in winter as compared to other seasons, which reinforces that less precipitation is falling as snow and melting is c. Hydrological response to future climate change occurring more often. Snow cover in the Rocky and In this section, CMIP5-driven VIC simulations are Coast Mountains decreases substantially in spring and analyzed to estimate future changes in SWE, snow summer. These results are consistent with those re- cover, and runoff for the RCP 8.5 pathway. Total runoff ported in Shrestha et al. (2012) using the VIC-driven is calculated using the sum of base flow and runoff. The simulations with CMIP3 model output. spatial patterns for the RCP 4.5 pathway are similar to MME mean seasonal runoff increases in winter and those for RCP 8.5 for the 2050s, but with slightly lower spring and decreases in summer over most of the FRB in magnitudes (not shown). In most of the temporal plots, the 2050s (not shown). The spring increase and summer for example, in Fig. 6c (described in greater detail be- decrease in runoff is more obvious in the Rocky and low), the response under RCP 8.5 is somewhat greater Coast Mountains, which is consistent with declines in the than and distinguishable from the RCP 4.5. It is ex- snowmelt-dominated areas (Figs. 6a,b) and the pro- pected that differences between RCPs will be greater at jected precipitation changes (Fig. 5) in these regions. the end of the twenty-first century than they are in Over the Rocky and Coast Mountains, the projected the 2050s. increase in spring runoff is induced by earlier snowmelt

Unauthenticated | Downloaded 10/02/21 03:48 PM UTC 484 JOURNAL OF HYDROMETEOROLOGY VOLUME 18

FIG. 5. Future change (2050s 2 1990s; %) in the spatial distribution of mean annual (a) rainfall, (b) snowfall, and (c) total precipitation for RCP 8.5 only.

Unauthenticated | Downloaded 10/02/21 03:48 PM UTC FEBRUARY 2017 I S L A M E T A L . 485

FIG. 6. Future change (2050s 2 1990s; %) in the spatial distribution of mean SWE over the FRB simulated by MME VIC for RCP 8.5 as (a) DJF and (b) MAM seasonal means. (c) Areally averaged annual cycle of daily

SWEmelt over the FRB for the 1990s and 2050s for both RCP 4.5 and RCP 8.5. Shading corresponds to intermodel variability.

Unauthenticated | Downloaded 10/02/21 03:48 PM UTC 486 JOURNAL OF HYDROMETEOROLOGY VOLUME 18

FIG. 7. Spatial distribution of mean future change (2050s 2 1990s; %) of (a) runoff, (b) SWEmelt, and (c) RSR. White areas at the outlet of the Fraser do not experience snow and are therefore masked in the percentage differences.

Unauthenticated | Downloaded 10/02/21 03:48 PM UTC FEBRUARY 2017 I S L A M E T A L . 487

TABLE 3. Future change (2050s 2 1990s) in RSR for FRB’s water basins. Differences are calculated using 30-yr MME mean for both RCP 4.5 and RCP 8.5 pathways. The range of uncertainties is also tabulated along with MME.

RSR change (2050s 2 1990s) (%) Subbasin RCP 4.5 RCP 8.5 UF 28.5 6 7.0 212.5 6 8.7 SU 215.0 6 5.6 219.0 6 5.0 NA 217.0 6 9.5 221.6 6 1.0 QU 28.0 6 6.5 213.8 6 7.7 CH 24.0 6 4.0 27.0 6 6.0 TN 29.0 6 4.8 214.0 6 6.6 LF 211.0 6 5.0 216.0 6 5.8 FIG. 8. MME-based VIC-simulated mean daily runoff at Fraser River at Hope. Black, blue, and red curves represent the daily driven by rising air temperature and changing pre- climatology for the base period (1990s), 2050s RCP 4.5, and 2050s cipitation phase. The more pronounced changes are RCP 8.5, respectively. over the Rocky and Coast Mountains, with spatially varying increases of 40%–100% in spring runoff and areas, SWE remains consistent in terms of its accu- nearly 50% decreases in summer runoff, especially over mulation and ablation phases (Déry et al. 2014). the Coast Mountains. Therefore, the underestimation of snowmelt-induced

The spatial distribution of mean RSR change (Fig. 7c) runoff is likely minimal for the FRB. Thus, the ap- shows spatially varying decreases of 15%–30% in the proach used to calculate RSR provides reasonable es- northern portions of the FRB, with maximum decreases timates of snowmelt contributions to runoff. The in the Interior Plateau and the Rocky Mountains. De- projected snowmelt changes for most of the FRB are creases in mean SWEmelt of up to 20% in the Rocky in agreement with projected changes in the Pacific Mountains dominate future declines in RSR (Fig. 7b) Northwest of North America (Chang and Jung 2010; compared to the relative change in future runoff Elsner et al. 2010).

(Fig. 7a). The future decrease (16% for RCP 8.5) in RSR Historically, the low-flow period takes place in win- arises mainly from declines in snow accumulation rather terintheFRB.Streamflowstartstoincreaseinspring, than changes in runoff amounts (Table 3). The RSR for reaches its maximum in early summer, and eventually six subbasins also exhibits considerable decreases in the declines in late summer. While this general seasonal 2050s, except for the high-elevation CH. As the UF, SU, pattern continues in the future, the timing of half the and NA subbasins experience more future warming in annual flow shows seasonal shifts approaching 25 days winter, snowmelt accelerates, contributing more to earlier for RCP 8.5 (Fig. 8) with the intermodel spread runoff in that season. For RCP 8.5, RSR decreases by of nearly 10 days. The flows increase in early spring and nearly 13% (with a range of 69%) in the UF, 19% are followed by an earlier and steeper recession in (65%) in the SU, and 22% (61%) in the NA subbasins summer. The seasonal shifts are mainly induced by the in the 2050s (Table 3). The uncertainties associated with changing phase of projected precipitation in winter and

RSR changes show a wide range of spread depending on spring, along with a transition of snowmelt-dominated the model, probably due to the accumulated intermodel and hybrid regimes to hybrid and rainfall-dominated differences in SWEmelt and runoff. For example, some of regimes in some regions due to increases in air tem- the CMIP5 GCMs are warmer and may cause more perature. The early decline of future SWE storage in- melting than others. duces such seasonal shifting. The shift in winter and

The RSR estimates mentioned above neglect the spring runoff leads to lower summer and autumn runoff snowmelt contribution to other components of the water due to the lower precipitation and snowpack depletion balance, such as ET and deep percolation. The ET is in the previous spring months. Such projected changes projected to increase in spring and summer over the in the FRB runoff are the continuation of observed Rocky and Coast Mountains (not shown). Therefore, climate change that has already been started in the spring snowmelt could be lost to ET, which may cause FRB as seen by Stewart (2009) and Kang et al. (2016). the snowmelt contribution to the water balance to be Daily mean runoff at the NA, QU, CH, TN, UF, and underestimated. However, this could be offset by SU subbasins exhibits almost the same features as the intermittent snowmelt, particularly at lower elevations. LF. A noticeable difference, for all subbasins, is the Furthermore, in most of the FRB’s snowmelt-dominated change in timing of the runoff peak that varies between

Unauthenticated | Downloaded 10/02/21 03:48 PM UTC 488 JOURNAL OF HYDROMETEOROLOGY VOLUME 18

FIG. 9. VIC–CMIP5 MME mean daily runoff for the (a) UF, (b) SU, (c) NA, (d) QU, (e) CH, and (f) TN subbasins. Black, blue, and red curves represent the daily climatology for the base period (1990s), 2050s RCP 4.5, and 2050s RCP 8.5, respectively.

20 and 30 days (median values) for RCP 8.5 within considerable changes in future flows along with in- different subbasins (Fig. 9) with the range of nearly creased water temperature may significantly degrade 10 days. the success rates of upriver salmon migrations into the The projected changes in FRB flows have significant FRB inlands and their reproduction rates (Morrison implications for hydropower generation, such as from et al. 2002; Eliason et al. 2011; Déry et al. 2012). the Nechako Reservoir and Bridge River. The water d. Elevation dependence availability appears to diminish during the period of highest demand (summer) in the 2050s, and the water Elevation is one of the key factors governing cold- resources managers may experience greater year to year season snow processes because of its relationship with air variability and uncertainty in flows because of reduced temperature. This controls the partitioning of pre- snowpacks. The Fraser is one of the world’s most pro- cipitation into snowfall and rainfall and regulates the ductive salmon rivers, with many First Nations relying melting of the snowpack during the ablation season. To on annual upriver migrations for food and as a livelihood investigate how SWE and snow cover may change at dif- (Xie and Hsieh 1989; Beamish et al. 1997; Benke and ferent elevations, analyses are conducted for different el- Cushing 2005; Fraser Basin Council 2006, 2009). The evation ranges using the information from elevation bands

Unauthenticated | Downloaded 10/02/21 03:48 PM UTC FEBRUARY 2017 I S L A M E T A L . 489

The fractional snow cover decreases nonlinearly from high to low elevations. The future variation in relative snow cover with different elevation ranges shows an overall decrease of snow-covered areas within the FRB (Fig. 10b). Snow cover decreases as much as 40% for the lower elevations from sea level to 1200 m, whereas for elevations greater than 1200 m a 20% decrease persists. Seasonal analyses show that the daily SWE climatol- ogy peaks in late winter followed by a sharp decrease in the snowmelt season, with SWE reaching its lowest value in late summer (not shown). With increasing ele- vation, the magnitude of SWE increases considerably for higher elevations. Although all the three regions follow the same SWE variability pattern, the magnitude varies strongly within each region. In the 1990s, the peak SWE value exceeds 1000 mm at higher elevations of the Rocky Mountains, whereas its maximum value sur- passes 800 mm in the wetter Coast Mountains. The daily climatology of snow cover shows a peak value of more than 0.90 (fraction) for higher elevations that remains persistent in spring and decreases to its minimum (0.20) in late summer. SWE changes are evident in the snow accumulation and snowmelt seasons, with a higher rate of SWE de- crease at lower elevations (Figs. 11a,c,e). Compared to the Interior Plateau, the elevation ranges across the Rocky Mountains show larger SWE decreases, with decreases of up to 400 mm in late summer at higher elevations. While the higher elevations show sub- stantial diminishing of SWE magnitude, their relative change compared to the 1990s is less pronounced than those at lower elevations (not shown). In the high el- FIG. 10. MME future change (2050s 2 1990s; %) variation of evations of the Rocky Mountains, maximum SWE mean (a) SWE and (b) snow cover for a range of elevations across occurs nearly 60 days later than it occurs at low ele- the Interior Plateau (green), Rocky Mountains (blue), and Coast vations. A mixed response is seen in the Coast Mountains (red) regions. Curves are for RCP 8.5 only. Mountains over different elevations because of the complex topography of the region interacting with within the Rocky Mountains, Interior Plateau, and Coast influences from the Pacific Ocean. These results are Mountains regions. To facilitate elevation-dependent consistent with those of Schnorbus et al. (2014) for the projected change, the elevation distribution within 10 Peace and Campbell River basins, where SWE was bands is clustered into different elevation ranges. While found to decline substantially at low elevations while VIC mean SWE output matches observations well, its at high elevations, where winter temperatures are output for different elevation bands cannot be compared projected to remain below freezing, SWE may either with observational data because of the paucity of SWE and remain unchanged or increase with increased winter snow cover data at different elevations across the FRB. precipitation. Future change in mean SWE for several elevation The daily mean snow cover shows quite complex dif- ranges shows SWE increases (decreases) with higher ferences at low, mid-, and high elevations, particularly (lower) elevations, which is consistent in all three re- for the Coast Mountains (Figs. 11b,d,f). Considerable gions (Fig. 10a). Future declines in SWE are accentu- fractional differences are found in both snow accumu- ated at low elevations reaching 250% in the 2050s. The lation and melt seasons for all regions. While the frac- impacts of future warming will be more noticeable at tional decrease of snow cover intensifies with increasing low- and midelevation regions, with implications for the elevations, some high elevations, particularly in the mass balance of snow and associated runoff. Rocky and Coast Mountains, show less snow cover

Unauthenticated | Downloaded 10/02/21 03:48 PM UTC 490 JOURNAL OF HYDROMETEOROLOGY VOLUME 18

FIG. 11. MME future change (2050s 2 1990s) of daily mean (left) SWE (mm) and (right) snow cover (as a fraction) for range of elevations across the (a),(b) Interior Plateau; (c),(d) Rocky Mountains; and (e),(f) Coast Mountains for RCP 8.5 only. Annual cycles are filtered using a 5-day running mean.

Unauthenticated | Downloaded 10/02/21 03:48 PM UTC FEBRUARY 2017 I S L A M E T A L . 491 change compared to mid- and low elevations. This may is dominated by snow processes (Kang et al. 2014, 2016; be due to the large snow masses covered by these ele- Stewart et al. 2005; Mote et al. 2005). The results reported vations that remain below freezing throughout the year. in this study reveal considerable changes in snowpack in In addition, the increase of future precipitation at some the 2050s with an earlier occurrence and decreased mag- elevations results in deeper snowpacks and may partially nitude of maximum SWE, which are consistent with ob- mask the effects of warming. In the Rocky Mountains, servation- and simulation-based studies (Boyer et al. 2010; snow cover decreases during the snow accumulation Kerkhoven and Gan 2011; Shrestha et al. 2012; Kang et al. season, with a maximum fractional decrease of more 2014, 2016; Troin et al. 2016) conducted in different regions than 0.30 for most elevations. At high elevations in the across Canada. These results also echo those found in Coast Mountains, snow cover decreases in both the snow studies conducted in other snowmelt-dominated regions in accumulation and melt seasons. Additionally, there is a North America (e.g., Adam et al. 2009; Elsner et al. 2010; considerable shift in the timing (advancement) of the Sproles et al. 2013). snowmelt season in this region. About one-third of the FRB’s snow cover will be lost by 6. Conclusions the 2050s according to the MME (Figs. 10b, 11). Increased This study investigated the possible impacts of twenty- winter and spring surface temperature results in retreating first century climate change on the Fraser River basin snow cover in mountainous terrain, which reduces the (FRB) snow hydrology. VIC driven by 24 statistically surface albedo. This accentuates the snow-albedo feedback downscaled CMIP5 GCM scenarios applied over the and consequently increases the surface temperature, ac- FRB provided estimates of spatial and temporal changes celerating the melting (Déry and Brown 2007; Hernández- in air temperature, precipitation, SWE, snow cover, and Henríquez et al. 2015). With continued warming in the runoff for the 2040–69 (2050s) period relative to the 2050s, declines in snow accumulation and increased melt 1980–2009 (1990s) baseline. The changes in the contri- are projected to no longer be offset by winter precipitation bution of snow to runoff generation (i:e., R ) and increases at some elevations in the FRB. SR variation in change of SWE and snow cover with dif- The projected snowpack and hydrological changes ferent elevation ranges were also analyzed. estimated in this study are influenced by the different The results show that future climate change will con- uncertainties associated with future emission scenarios, siderably affect the FRB hydrology and its snowpack GCM dynamics, natural climate variability, downscaling accumulation and ablation with earlier onset of snow- methodology, and hydrological modeling (Wilby and melt and enhanced winter and spring runoff. The overall Harris 2006). The uncertainties in climate projections findings of this study can be summarized with the fol- (Tebaldi et al. 2005; Bae et al. 2011) and downscaling lowing key points: methods (Fowler et al. 2007; Chen et al. 2013) are the predominant sources of uncertainty in projecting hy- 1) An overall decrease in SWE in winter and spring is drologic changes (Kay et al. 2009; Chen et al. 2011). Our projected over the FRB owing to rising air temper- future work will systematically explore these un- ature that accelerates snowmelt. The SWEmelt in- certainties and those present in the model setup and creases in the Interior Plateau in winter but declines calibration parameters. Generally, air temperature is in spring and summer in the Rocky and Coast the leading controller of summer water availability in Mountains, yielding lower contributions of SWEmelt snowmelt-dominated basins, which makes regional to runoff. snowpacks more vulnerable to temperature-induced 2) SWE and snow cover decrease at a higher rate at effects, rather than precipitation. The IPCC AR5 lower elevations. SWE and snow cover at low GCMs have improved since the AR4, and there con- elevations are more sensitive to increased air tem- tinues to be a very high confidence that these models perature because the climate is warmer. About one- reproduce the observed large-scale time–mean tem- third of the FRB snow cover will be lost by the 2050s. perature pattern (Stocker et al. 2013). The AR5 tem- In the FRB, even the most modest warming will perature projections are considered more certain than convert snowfall to rainfall because large areas hover precipitation and therefore the effect of temperature- around 08C in the present-day climate. Thus, warm- based uncertainties in snowmelt-driven runoff may be ing will accelerate melt at low elevations, contribut- minimal for the FRB. The future estimates presented in ing more SWEmelt to early runoff. this study can therefore prove to be useful information 3) Simulations yield declining RSR for the Fraser River for impact assessments of climate change. at Hope and all six of its major subbasins driven by Rising air temperature has already resulted in sub- projected increases in air temperature. Diminished stantial hydrologic changes in catchments where hydrology SWE along with substantial declines in winter snow

Unauthenticated | Downloaded 10/02/21 03:48 PM UTC 492 JOURNAL OF HYDROMETEOROLOGY VOLUME 18

accumulation and snow-covered areas are projected warmer than CMIP3 projections because of slightly higher for the FRB. Analyses suggest that the snowmelt emissions for the closest comparable forcing scenarios, contribution to streamflow in the FRB will decline such that CMIP5/RCP 4.5 is warmer than CMIP3/B1, for considerably by the 2050s due to warming. example. The results reported here therefore offer valu- 4) The timing and amount of runoff is altered sub- able insight to the FRB’s future state and fate that could stantially by declining mountain snowpacks. Higher have important ramifications for the development of temperatures in the 2050s will result in less winter strategies to address challenges created by climate change precipitation falling as snow and accelerate (by and to design water management systems and plans for 25 days) melting of winter and spring snowpacks. snowmelt-dominated regions of western Canada. Both of these effects lead to a shift in the Fraser River peak runoff toward winter and early spring, when Acknowledgments. This work is supported by the water demand is potentially at its lowest and storage Natural Sciences and Engineering Research Council of facilities are already at capacity. Hence, in the Canada (NSERC)-funded Canadian Sea Ice and Snow future, a larger proportion of annual runoff could be Evolution (CanSISE) Network. The authors are lost to the Pacific Ocean. Although, given tempera- grateful to the colleagues from Pacific Climate Impacts ture and resulting evaporative changes, water demand Consortium (PCIC) for providing the statistically for irrigation will probably also shift to earlier in the downscaled CMIP5 boundary forcing data and their year and could utilize some of this water. assistance. We thank the World Climate Research Programme’s Working Group on Coupled Modelling The key points mentioned above provide critical in- and the climate modeling groups who participated in formation on the range of possible future flows and their CMIP5 for producing and making their model output timing and hence establish possible impacts on the available. We also thank Do Hyuk Kang (NASA Fraser River. Such information can be used in the as- GSFC), Huilin Gao (Texas A&M), and Xiaogang Shi sessment of future water supplies for the region. (Xi’an Jiaotong-Liverpool University) for assisting While this study facilitates the estimation of future with this effort. Thanks to Jeannine St. Jacques (Con- changes in snowpack and water availability of the cordia University) for her input in revising this paper. FRB, some caution in interpreting the results is The authors are thankful to Michael Allchin (UNBC) needed. First, the hydrological model used in this study for plotting Fig. 1a. Thanks to the anonymous re- is integrated on a relatively coarse resolution (1/48), viewers and the handling editor for constructive com- which may not represent small-scale dynamics related ments that greatly improved the paper. to the FRB’s complex topography. Our future work will therefore explore the robustness of the results of REFERENCES thisstudywithanincreaseofresolutionupto1/168. Similarly, to improve physical integration, the daily Abatzoglou, J. T., 2011: Influence of the PNA on declining time step used in VIC integrations may be increased to mountain snowpack in the western United States. Int. J. Climatol., 31, 1135–1142, doi:10.1002/joc.2137. hourly in future efforts. Finally, the accuracy of the Adam, J. C., and D. P. Lettenmaier, 2008: Application of new forcing data at the model grid resolution and modifi- precipitation and reconstructed streamflow products to cation of air temperature and precipitation with pre- streamflow trend attribution in northern Eurasia. J. Climate, set, linear lapse rates across elevation ranges may also 21, 1807–1828, doi:10.1175/2007JCLI1535.1. introduce uncertainties in VIC simulations. Such al- ——, A. F. Hamlet, and D. P. Lettenmaier, 2009: Implications of global climate change for snowmelt hydrology in the twenty- ternative approaches will be investigated in our future first century. Hydrol. Processes, 23, 962–972, doi:10.1002/ studies. hyp.7201. Nevertheless, this work provides useful information Ashfaq, M., L. C. Bowling, K. Cherkauer, J. S. Pal, and N. S. for climate change impact assessments for the FRB, Diffenbaugh, 2010: Influence of climate model biases and especially for water management. Previous studies fo- daily-scale temperature and precipitation events on hydro- logical impacts assessment: A case study of the United States. cusing on the FRB (e.g., Kerkhoven and Gan 2011; J. Geophys. Res., 115, D14116, doi:10.1029/2009JD012965. Ferrari et al. 2007; Shrestha et al. 2012) have estimated Bae, D. H., I. W. Jung, and D. P. Lettenmaier, 2011: Hydrologic the future hydrological changes using CMIP3 or older uncertainties in climate change from IPCC AR4 GCM simu- versions of GCMs. The results from this study are based lations of the Chungju basin, Korea. J. Hydrol., 401, 90–105, on the new and improved CMIP5 GCMs and may le- doi:10.1016/j.jhydrol.2011.02.012. Barnett, T. P., J. C. Adam, and D. P. Lettenmaier, 2005: Potential verage existing literature by providing future estimates impacts of a warming climate on water availability in snow- of FRB snowpack and hydrological changes. For similar dominated regions. Nature, 438, 303–309, doi:10.1038/ locations and climate models, CMIP5 projections are nature04141.

Unauthenticated | Downloaded 10/02/21 03:48 PM UTC FEBRUARY 2017 I S L A M E T A L . 493

Beamish, R. J., M. Henderson, and H. Regier, 1997: Impacts of J. Climate, 7, 1803–1808, doi:10.1175/1520-0442(1994)007,1803: climate change on the fishes of British Columbia. Responding ORCTIS.2.0.CO;2. to Global Climate Change in British Columbia and Yukon, Déry, S. J., and R. D. Brown, 2007: Recent Northern Hemisphere snow Vol. 1, E. Taylor and B. Taylor, Eds., BC Ministry of Envi- cover extent trends and implications for the snow-albedo feed- ronment, Lands and Parks, 12-1–12-16. back. Geophys. Res. Lett., 34, L22504, doi:10.1029/2007GL031474. Benke, A. C., and C. E. Cushing, Eds., 2005: Rivers of North ——, J. Sheffield, and E. F. Wood, 2005: Connectivity between America. Elsevier, 1168 pp. Eurasian snow cover extent and Canadian snow water equiv- Bocking, R. C., 1997: Mighty River: A Portrait of the Fraser. alent and river discharge. J. Geophys. Res., 110, D23106, Douglas and McIntyre, 294 pp. doi:10.1029/2005JD006173. Bowling, L. C., and D. P. Lettenmaier, 2010: Modeling the effects of ——, M. A. Hernández-Henríquez, P. N. Owens, M. W. Parkes, lakes and wetlands on the water balance of Arctic environments. and E. L. Petticrew, 2012: A century of hydrological variability J. Hydrometeor., 11, 276–295, doi:10.1175/2009JHM1084.1. and trends in the Fraser River basin. Environ. Res. Lett., 7, ——, J. W. Pomeroy, and D. P. Lettenmaier, 2004: Parameteri- 024019, doi:10.1088/1748-9326/7/2/024019. zation of blowing-snow sublimation in a macroscale hy- ——, H. K. Knudsvig, M. A. Hernández-Henríquez, and D. S. Coxson, drology model. J. Hydrometeor., 5, 745–762, doi:10.1175/ 2014: Net snowpack accumulation and ablation characteristics in 1525-7541(2004)005,0745:POBSIA.2.0.CO;2. the inland temperate rainforest of the upper Fraser basin, Canada. Boyer, C., D. Chaumont, I. Chartier, and A. G. Roy, 2010: Impact Hydrology, 1, 1–19, doi:10.3390/hydrology1010001. of climate change on the hydrology of St. Lawrence tributar- Duan, Q., and T. J. Phillips, 2010: Bayesian estimation of local ies. J. Hydrol., 384, 65–83, doi:10.1016/j.jhydrol.2010.01.011. signal and noise in multimodel simulations of climate change. Bürger, G., T. Q. Murdock, A. T. Werner, S. R. Sobie, and A. J. J. Geophys. Res., 115, D18123, doi:10.1029/2009JD013654. Cannon, 2012: Downscaling extremes—An intercomparison Dwarakish, G. S., and B. P. Ganasri, 2015: Impact of land use of multiple statistical methods for present climate. J. Climate, change on hydrological systems: A review of current modeling 25, 4366–4388, doi:10.1175/JCLI-D-11-00408.1. approaches. Cogent. Geosci., 1, 1115691, doi:10.1080/ ——, S. R. Sobie, A. J. Cannon, A. T. Werner, and T. Q. Murdock, 23312041.2015.1115691. 2013: Downscaling extremes: An intercomparison of multiple Eliason, E. J., and Coauthors, 2011: Differences in thermal toler- methods for future climate. J. Climate, 26, 3429–3449, ance among sockeye salmon populations. Science, 332, 109– doi:10.1175/JCLI-D-12-00249.1. 112, doi:10.1126/science.1199158. Cannon, A. J., 2015: Selecting GCM scenarios that span the range Elsner, M. M., and Coauthors, 2010: Implications of 21st century of changes in a multimodel ensemble: Application to CMIP5 climate change for the hydrology of Washington State. Cli- climate extremes indices. J. Climate, 28, 1260–1267, matic Change, 102, 225–260, doi:10.1007/s10584-010-9855-0. doi:10.1175/JCLI-D-14-00636.1. Eum, H. I., Y. Dibike, T. Prowse, and B. Bonsal, 2014: Inter- Chang, H. J., and I. W. Jung, 2010: Spatial and temporal changes in comparison of high-resolution gridded climate data sets and runoff caused by climate change in a complex large river their implication on hydrological model simulation over the basin in Oregon. J. Hydrol., 388, 186–207, doi:10.1016/ Athabasca watershed, Canada. Hydrol. Processes, 28, 4250– j.jhydrol.2010.04.040. 4271, doi:10.1002/hyp.10236. Chen, J., F. P. Brissette, A. Poulin, and R. Leconte, 2011: Overall Ferrari, M. R., J. R. Miller, and G. L. Russell, 2007: Modeling uncertainty study of the hydrological impacts of climate changes in summer temperature of the Fraser River dur- change for a Canadian watershed. Water Resour. Res., 47, ing the next century. J. Hydrol., 342, 336–346, doi:10.1016/ W12509, doi:10.1029/2011WR010602. j.jhydrol.2007.06.002. ——, ——, D. Chaumont, and M. Braun, 2013: Performance and Fowler, H. J., S. Blenkinsop, and C. Tebaldi, 2007: Linking climate uncertainty evaluation of empirical downscaling methods in change modelling to impacts studies: Recent advances in quantifying the climate change impacts on hydrology over two downscaling techniques for hydrological modelling. Int. North American river basins. J. Hydrol., 479, 200–214, J. Climatol., 27, 1547–1578, doi:10.1002/joc.1556. doi:10.1016/j.jhydrol.2012.11.062. Fraser Basin Council, 2006: Bridge between nations: A history of Cherkauer, K. A., and T. Sinha, 2010: Hydrologic impacts of pro- First Nations in the Fraser River basin. Fraser Basin Council, jected future climate change in the Lake Michigan region. 24 pp. [Available online at www.fraserbasin.bc.ca/_Library/ J. Great Lakes Res., 36, 33–50, doi:10.1016/j.jglr.2009.11.012. Ab_NonAb_Relations/bridge_between_nations.pdf.] ——, L. C. Bowling, and D. P. Lettenmaier, 2003: Variable In- ——, 2009: Sustainability snapshot 4: The many faces of sustain- filtration Capacity cold land process model updates. Global ability. 2009 State of the Fraser Basin Rep., Fraser Basin Planet. Change, 38, 151–159, doi:10.1016/S0921-8181(03)00025-0. Council, 94 pp. [Available online at www.fraserbasin.bc.ca/ Choi, G., D. A. Robinson, and S. Kang, 2010: Changing Northern _Library/Comm_Indicators/report_ss4_2009.pdf.] Hemisphere snow seasons. J. Climate, 23, 5305–5310, Gao, H., and Coauthors, 2009: Water budget record from Variable doi:10.1175/2010JCLI3644.1. Infiltration Capacity (VIC) model. Algorithm Theoretical Basis Christensen, N. S., and D. P. Lettenmaier, 2007: A multimodal Doc., 56 pp. [Available online at http://www.hydro.washington.edu/ ensemble approach to assessment of climate change impacts SurfaceWaterGroup/Publications/Water_Cycle_MEaSUREs_ on the hydrology and water resources of the Colorado River ATBD_VICmodel_submit.doc.] basin. Hydrol. Earth Syst. Sci., 11, 1417–1434, doi:10.5194/ ——, Q. Tang, C. R. Ferguson, E. F. Wood, and D. P. Lettenmaier, hess-11-1417-2007. 2010: Estimating the water budget of major US river basins via Daly, C., 2006: Guidelines for assessing the suitability of spatial remote sensing. Int. J. Remote Sens., 31, 3955–3978, climate data sets. Int. J. Climatol., 26, 707–721, doi:10.1002/ doi:10.1080/01431161.2010.483488. joc.1322. Giorgi, F., and R. Francisco, 2000: Evaluating uncertainties in the Danard, M., and T. S. Murty, 1994: On recent climate trends in prediction of regional climate change. Geophys. Res. Lett., 27, selected salmon-hatching areas of British Columbia. 1295–1298, doi:10.1029/1999GL011016.

Unauthenticated | Downloaded 10/02/21 03:48 PM UTC 494 JOURNAL OF HYDROMETEOROLOGY VOLUME 18

Haddeland, I., T. Skaugen, and D. P. Lettenmaier, 2007: Hydro- parameterization schemes. Tellus, 48A, 708–721, doi:10.1034/ logic effects of land and water management in North America j.1600-0870.1996.t01-3-00009.x. and Asia: 1700–1992. Hydrol. Earth Syst. Sci., 11, 1035–1045, ——, E. Raschke, B. Nijssen, and D. P. Lettenmaier, 1998a: Re- doi:10.5194/hess-11-1035-2007. gional scale hydrology: I. Formulation of the VIC-2L model Hamlet, A. F., and D. P. Lettenmaier, 2005: Production of tem- coupled to a routing model. Hydrol. Sci. J., 43, 131–141, porally consistent gridded precipitation and temperature doi:10.1080/02626669809492107. fields for the continental United States. J. Hydrometeor., 6, ——, B. Nijssen, and D. P. Lettenmaier, 1998b: Regional scale 330–336, doi:10.1175/JHM420.1. hydrology: II. Application of the VIC-2L model to the Weser Hernández-Henríquez, M. A., S. J. Déry, and C. Derksen, 2015: River, Germany. Hydrol. Sci. J., 43, 143–158, doi:10.1080/ Polar amplification and elevation-dependence in trends of 02626669809492108. Northern Hemisphere snow cover extent, 1971–2014. Environ. Marcus, S. L., O. de Viron, and J. O. Dickey, 2011: Abrupt atmo- Res. Lett., 10, 044010, doi:10.1088/1748-9326/10/4/044010. spheric torque changes and their role in the 1976–1977 climate Hidalgo, H., and Coauthors, 2009: Detection and attribution of stream- regime shift. J. Geophys. Res., 116, D03107, doi:10.1029/ flow timing changes to climate change in the western United States. 2010JD015032. J. Climate, 22, 3838–3855, doi:10.1175/2009JCLI2470.1. Maurer, E. P., and H. G. Hidalgo, 2008: Utility of daily vs. monthly Hopkinson, R. F., D. W. McKenney, E. J. Milewska, M. F. large-scale climate data: An intercomparison of two statistical Hutchinson, P. Papadopol, and L. A. Vincent, 2011: Impact of downscaling methods. Hydrol. Earth Syst. Sci., 12, 551–563, aligning climatological day on gridding daily maximum– doi:10.5194/hess-12-551-2008. minimum temperature and precipitation over Canada. J. Appl. ——, A. W. Wood, J. C. Adam, D. P. Lettenmaier, and B. Nijssen, Meteor. Climatol., 50, 1654–1665, doi:10.1175/2011JAMC2684.1. 2002: A long-term hydrologically based dataset of land Hsieh, W. W., and B. Tang, 2001: Interannual variability of accu- surface fluxes and states for the conterminous United States. mulated snow in the Columbia basin, British Columbia. Water J. Climate, 15, 3237–3251, doi:10.1175/1520-0442(2002)015,3237: Resour. Res., 37, 1753–1759, doi:10.1029/2000WR900410. ALTHBD.2.0.CO;2. Hutchinson, M. F., D. W. McKenney, K. Lawrence, J. H. Pedlar, McKenney, D. W., and Coauthors, 2011: Customized spatial cli- R. F. Hopkinson, E. Milewska, and P. Papadopol, 2009: De- mate models for North America. Bull. Amer. Meteor. Soc., 92, velopment and testing of Canada-wide interpolated spatial 1611–1622, doi:10.1175/2011BAMS3132.1. models of daily minimum–maximum temperature and pre- Miao, C. Y., Q. Y. Duan, Q. H. Sun, and J. D. Li, 2013: Evaluation cipitation for 1961–2003. J. Appl. Meteor. Climatol., 48, 725– and application of Bayesian multi-model estimation in tem- 741, doi:10.1175/2008JAMC1979.1. perature simulations. Prog. Phys. Geogr., 37, 727–744, IPCC, 2014: Climate Change 2014: Synthesis Report. Cambridge doi:10.1177/0309133313494961. University Press, 151 pp. Mitchell, K. E., and Coauthors, 2004: The multi-institution North Kalnay, E., and Coauthors, 1996: The NCEP/NCAR 40-Year Re- American Land Data Assimilation System (NLDAS): Utiliz- analysis Project. Bull. Amer. Meteor. Soc., 77, 437–471, ing multiple GCIP products and partners in a continental doi:10.1175/1520-0477(1996)077,0437:TNYRP.2.0.CO;2. distributed hydrological modeling system. J. Geophys. Res., Kang, D. H., X. Shi, H. Gao, and S. J. Déry, 2014: On the changing 109, D07S90, doi:10.1029/2003JD003823. contribution of snow to the hydrology of the Fraser River Moore, R. D., 1991: Hydrology and water supply in the Fraser basin, Canada. J. Hydrometeor., 15, 1344–1365, doi:10.1175/ River basin. Water in Sustainable Development: Exploring JHM-D-13-0120.1. Our Common Future in the Fraser River Basin, A. H. J. Dorcey ——, H. Gao, X. Shi, S. Islam, and S. J. Déry, 2016: Impacts of a rapidly and J. R. Griggs, Eds, Wastewater Research Centre, Univer- declining mountain snowpack on streamflow timing in Canada’s sity of British Columbia, 21–40. Fraser River basin. Sci. Rep., 6, 19 299, doi:10.1038/srep19299. ——, and S. M. Wondzell, 2005: Physical hydrology and the effects Kay, A. L., H. N. Davies, V. A. Bell, and R. G. Jones, 2009: of forest harvesting in the Pacific Northwest: A review. Comparison of uncertainty sources for climate change im- J. Amer. Water Resour. Assoc., 41, 763–784, doi:10.1111/ pacts: Flood frequency in England. Climatic Change, 92, 41– j.1752-1688.2005.tb04463.x. 63, doi:10.1007/s10584-008-9471-4. Morrison,J.,M.C.Quick,andM.G.G.Foreman,2002:Climate Kerkhoven, E., and T. Gan, 2011: Differences and sensitivities in change in the Fraser River watershed: Flow and tempera- potential hydrologic impact of climate change to regional-scale ture projections. J. Hydrol., 263, 230–244, doi:10.1016/ Athabasca and Fraser River basins of the leeward and wind- S0022-1694(02)00065-3. ward sides of the Canadian Rocky Mountains respectively. Mote, P. W., A. F. Hamlet, M. P. Clark, and D. P. Lettenmaier, Climatic Change, 106,583–607,doi:10.1007/s10584-010-9958-7. 2005: Declining mountain snowpack in western North America. Knutti, R., and J. Sedlácek, 2012: Robustness and uncertainties in Bull. Amer. Meteor. Soc., 86, 39–49, doi:10.1175/BAMS-86-1-39. the new CMIP5 climate model projections. Nat. Climate Nakicenovic, N., and R. Swart, 2000: Emissions Scenarios. Cam- Change, 3, 369–373, doi:10.1038/nclimate1716. bridge University Press, 570 pp. Liang, X., D. P. Lettenmaier, E. F. Wood, and S. J. Burges, 1994: A Nash, J. E., and J. V. Sutcliffe, 1970: River flow forecasting through simple hydrologically based model of land-surface water and conceptual models part I—A discussion of principles. energy fluxes for general-circulation models. J. Geophys. Res., J. Hydrol., 10, 282–290, doi:10.1016/0022-1694(70)90255-6. 99, 14 415–14 428, doi:10.1029/94JD00483. Nijssen, B., G. M. O’Donnell, D. P. Lettenmaier, D. Lohmann, and ——, E. F. Wood, and D. P. Lettenmaier, 1996: Surface soil E. F. Wood, 2001a: Predicting the discharge of global rivers. moisture parameterization of the VIC-2L model: Evaluation J. Climate, 14, 3307–3323, doi:10.1175/1520-0442(2001)014,3307: and modification. Global Planet. Change, 13, 195–206, PTDOGR.2.0.CO;2. doi:10.1016/0921-8181(95)00046-1. ——, R. Schnur, and D. P. Lettenmaier, 2001b: Global retro- Lohmann, D., R. Nolte-Holube, and E. Raschke, 1996: A large spective estimation of soil moisture using the Variable In- scale horizontal routing model to be coupled to land surface filtration Capacity land surface model, 1980–93. J. Climate,

Unauthenticated | Downloaded 10/02/21 03:48 PM UTC FEBRUARY 2017 I S L A M E T A L . 495

14, 1790–1808, doi:10.1175/1520-0442(2001)014,1790: Hydrol. Earth Syst. Sci., 10, 1987–2013, doi:10.5194/ GREOSM.2.0.CO;2. hessd-10-1987-2013. NRCan, 2014: Regional, national and international climate mod- Sospedra-Alfonso, R., L. Mudryk, W. Merryfield, and C. Derksen, eling. Accessed 9 November 2016. [Available online at http:// 2016: Representation of snow in the Canadian Seasonal to In- cfs.nrcan.gc.ca/projects/3?lang5en_CA.] terannual Prediction System: Part I. Initialization. J. Hydrometeor., Oubeidillah, A. A., S. C. Kao, M. Ashfaq, B. S. Naz, and G. Tootle, 17, 1467–1488, doi:10.1175/JHM-D-14-0223.1. 2014: A large-scale, high-resolution hydrological model pa- Sproles, E. A., A. W. Nolin, K. Rittger, and T. H. Painter, 2013: rameter data set for climate change impact assessment for the Climate change impacts on maritime mountain snowpack in conterminous US. Hydrol. Earth Syst. Sci., 18, 67–84, the Oregon Cascades. Hydrol. Earth Syst. Sci., 17, 2581–2597, doi:10.5194/hess-18-67-2014. doi:10.5194/hess-17-2581-2013. Park, H., H. Yabuki, and T. Ohata, 2012: Analysis of satellite and Stewart, I. T., 2009: Changes in snowpack and snowmelt runoff for model datasets for variability and trends in Arctic snow key mountain regions. Hydrol. Processes, 23, 78–94, extent and depth, 1948–2006. Polar Sci., 6,23–37,doi:10.1016/ doi:10.1002/hyp.7128. j.polar.2011.11.002. ——, D. R. Cayan, and M. D. Dettinger, 2005: Changes toward Rand, P. S., and Coauthors, 2006: Effects of river discharge, tem- earlier streamflow timing across western North America. perature, and future climates on energetics and mortality of J. Climate, 18, 1136–1155, doi:10.1175/JCLI3321.1. adult migrating Fraser River sockeye salmon. Trans. Amer. Stocker, T. F., and Coauthors, 2013: Technical summary. Climate Fish. Soc., 135, 655–667, doi:10.1577/T05-023.1. Change 2013: The Physical Science Basis, T. F. Stocker et al., Rangwala, I., and J. R. Miller, 2012: Climate change in mountains: Eds., Cambridge University Press, 33–115, doi:10.1017/ A review of elevation-dependent warming and its possible CBO9781107415324.005. causes. Climatic Change, 114, 527–547, doi:10.1007/ Su, F., J. C. Adam, L. C. Bowling, and D. P. Lettenmaier, 2005: s10584-012-0419-3. Streamflow simulations of the terrestrial Arctic domain. Rauscher, S. A., J. S. Pal, N. S. Diffenbaugh, and M. M. Benedetti, J. Geophys. Res., 110, D08112, doi:10.1029/2004JD005518. 2008: Future changes in snowmelt-driven runoff timing over Taylor, K. E., R. J. Stouffer, and G. A. Meehl, 2012: An overview of the western US. Geophys. Res. Lett., 35, L16703, doi:10.1029/ CMIP5 and the experiment design. Bull. Amer. Meteor. Soc., 2008GL034424. 93, 485–498, doi:10.1175/BAMS-D-11-00094.1. Rodenhuis, D. R., K. E. Bennett, A. T. Werner, T. Q. Murdock, Tebaldi, C., R. L. Smith, D. Nychka, and L. O. Mearns, 2005: and D. Bronaugh, 2009: Hydro-climatology and future climate Quantifying uncertainty in projections of regional climate impacts in British Columbia. Pacific Climate Impacts Con- change: A Bayesian approach to the analysis of multimodel sortium, 132 pp. ensembles. J. Climate, 18, 1524–1540, doi:10.1175/JCLI3363.1. Schnorbus, M. A., K. E. Bennett, A. T. Werner, and A. J. Berland, Troin, M., A. Poulin, B. Baraer, and F. Brissette, 2016: Comparing 2011: Hydrologic impacts of climate change in the Peace, snow models under current and future climates: Uncertainties Campbell and Columbia sub-basins, British Columbia, and implications for hydrological impact studies. J. Hydrol., Canada. Hydrologic Modelling Project Final Rep. Part II, 540, 588–602, doi:10.1016/j.jhydrol.2016.06.055. Pacific Climate Impacts Consortium, 175 pp. [Available Vano, J. A., N. Voisin, L. Cuo, A. F. Hamlet, M. M. Elsner, R. N. online at https://pacificclimate.org/sites/default/files/publications/ Palmer, A. Polebitski, and D. P. Lettenmaier, 2010: Climate Schnorbus.HydroModelling.FinalReport2.Apr2011.pdf.] change impacts on water management in the Puget Sound ——, A. T. Werner, and K. E. Bennett, 2014: Impacts of climate region, Washington, USA. Climatic Change, 102, 261–286, change in three hydrologic regimes in British Columbia, doi:10.1007/s10584-010-9846-1. Canada. Hydrol. Processes, 28, 1170–1189, doi:10.1002/ van Vuuren, D. P., and Coauthors, 2011: The representative con- hyp.9661. centration pathways: An overview. Climatic Change, 109,5– Shi, X., A. W. Wood, and D. P. Lettenmaier, 2008: How essential is 31, doi:10.1007/s10584-011-0148-z. hydrologic model calibration to seasonal streamflow forecasting? Wade, N. L., J. Martin, and P. H. Whitfield, 2001: Hydrologic and J. Hydrometeor., 9, 1350–1363, doi:10.1175/2008JHM1001.1. climatic zonation of Georgia Basin, British Columbia. Can. ——, S. J. Déry, P. Ya. Groisman, and D. P. Lettenmaier, 2013: Water Resour. J., 26, 43–70, doi:10.4296/cwrj2601043. Relationship between recent pan-Arctic snow cover and hy- Wang, A., Y. L. Kaiyuan, and D. P. Lettenmaier, 2008: Integration droclimate trends. J. Climate, 26, 2048–2064, doi:10.1175/ of the Variable Infiltration Capacity model soil hydrology JCLI-D-12-00044.1. scheme into the Community Land Model. J. Geophys. Res., Shrestha, R. R., M. A. Schnorbus, A. T. Werner, and A. J. Berland, 113, D09111, doi:10.1029/2007JD009246. 2012: Modeling spatial and temporal variability of hydrologic Water Survey of Canada, 2014: HYDAT database. Accessed impacts of climate change in the Fraser River basin, British 9 November 2016. [Available online at http://www.ec.gc.ca/ Columbia, Canada. Hydrol. Processes, 26, 1840–1860, rhc-wsc/.] doi:10.1002/hyp.9283. Werner, A. T., M. A. Schnorbus, R. R. Shrestha, and H. D. ——, ——, and D. L. Peters, 2016: Assessment of a hydrologic Eckstrand, 2013: Spatial and temporal change in the hydro- model’s reliability in simulating flow regime alterations in a climatology of the Canadian portion of the Columbia River changing climate. Hydrol. Processes, 30, 2628–2643, basin under multiple emissions scenarios. Atmos.–Ocean, 51, doi:10.1002/hyp.10812. 357–379, doi:10.1080/07055900.2013.821400. Shukla, S., N. Voisin, and D. P. Lettenmaier, 2012: Value of me- Whitfield, P. H., R. D. Moore, S. W. Fleming, and A. Zawadzki, dium range weather forecasts in the improvement of seasonal 2010: Pacific decadal oscillation and the hydroclimatology of hydrologic prediction skill. Hydrol. Earth Syst. Sci., 16, 2825– western Canada—Review and prospects. Can. Water Resour. 2838, doi:10.5194/hess-16-2825-2012. J., 35, 1–28, doi:10.4296/cwrj3501001. ——, J. Sheffield, E. F. Wood, and D. P. Lettenmaier, 2013: On the Wilby, R. L., and I. Harris, 2006: A framework for assessing un- sources of global land surface hydrologic predictability. certainties in climate change impacts: Low-flow scenarios for

Unauthenticated | Downloaded 10/02/21 03:48 PM UTC 496 JOURNAL OF HYDROMETEOROLOGY VOLUME 18

the River Thames, UK. Water Resour. Res., 42, W02419, Xie, L., and W. W. Hsieh, 1989: Predicting the return migration doi:10.1029/2005WR004065. routes of the Fraser River sockeye salmon (Oncorhynchus Wood, A. W., L. R. Leung, V. Sridhar, and D. P. Lettenmaier, 2004: nerka). Can. J. Fish. Aquat. Sci., 46, 1287–1292, doi:10.1139/ Hydrologic implications of dynamical and statistical approaches f89-165. to downscaling climate model outputs. Climate Change, 62, 189– Yapo, P., H. Gupta, and S. Sorooshian, 1998: Multi-objective 216, doi:10.1023/B:CLIM.0000013685.99609.9e. global optimization for hydrologic models. J. Hydrol., 204, 83– Wu, H., J. S. Kimball, N. Mantua, and J. Stanford, 2011: Auto- 97, doi:10.1016/S0022-1694(97)00107-8. mated upscaling of river networks for macroscale hydrological Zhou, T., B. Nijssen, H. Gao, and D. P. Lettenmaier, 2016: The contri- modeling. Water Resour. Res., 47, W03517, doi:10.1029/ bution of reservoirs to global land surface water storage variations. 2009WR008871. J. Hydrometeor., 17, 309–325, doi:10.1175/JHM-D-15-0002.1.

Unauthenticated | Downloaded 10/02/21 03:48 PM UTC