As discussed earlier, from the weather perspective this study focuses on quantifying the roles of orography, local and remote moisture sources, the occurrence of rain-on-snow and soil moisture phase in the Alberta flood event through the analysis of CRCM5 experiments. An ensemble of nine CRCM5 reference simulations (CRCM5_Ref) only differing in their initial conditions serves to evaluate the performance of the model and to help assess the significance of the various factors that may have contributed to the flood event. The CRCM5_Ref integrations are initialized six hours apart between 0000 UTC 12 June and 0000 UTC 14 June 2013, continuing until 0000 UTC 22 June 2013. The simulations are performed in LAM mode with 0.11° resolution (Fig. 1b), and are driven by the ERA-Interim reanalysis (Dee et al. 2011) at the lateral boundaries. Spectral nudging is applied to the temperature field and the horizontal wind components, with a half-response wavelength of 410 km and a relaxation time of 24 h. Nudging strength is set to zero at the surface and increases with height, reaching full strength above the lowest 15 % of the atmospheric column. For all simulations, unless specified otherwise, the atmosphere is initialized from ERA-Interim fields and the soil is initialized using fields from a continuous ERA-Interim driven run of CRCM5 over the same domain. As accurate surface conditions are critical for the simulation of streamflows, initial snow depth, snow density and snow temperature in the BRB (Fig. 1a) are taken from the Snow Data Assimilation System (SNODAS; Barrett 2003), and the soil is considered frozen for grid cells covered by snow. Initial conditions for WATROUTE are obtained from a spin-up hydrologic simulation, which uses CRCM5-simulated runoff and starts in January 2013.
The performance of CRCM5_Ref in representing precipitation, the main synoptic-dynamic characteristics of the event and streamflow in the BRB is first assessed. To this end, the simulated precipitation is compared to Environment Canada’s six-hourly Canadian Precipitation Analysis (CaPA; Mahfouf et al. 2007), available at 15 km spatial resolution, and to the daily ANUSPLIN gridded dataset (McKenney et al. 2011), available at 10 km spatial resolution. Hourly rainfall accumulations from 30 rain gauges of Alberta’s AgroClimatic Information Service, located in and around the western BRB are also used (Fig. 1a). The six-hourly ERA-Interim reanalysis at 0.75° spatial resolution is used to evaluate simulated synoptic and dynamic features, such as geopotential heights, upper and lower level circulations and vertically integrated moisture fluxes. Finally, the hydrographs generated by CRCM5 and WATROUTE are compared to those from two flow monitoring stations on the Bow River, one located at Banff and the other at Calgary (Fig. 1a). All experiments performed, from weather and climate perspectives, are discussed below and summarized in Table 1.
Impact of orography
Milrad et al. (2015) concluded that orographically forced ascent played a role during the event. Here, the impact of orography on the location and intensity of precipitation is quantified by performing an experiment (CRCM5_Oro) with reduced orography over the southern Canadian Rockies, i.e. by reducing terrain heights above 1200 m by 75 %, following an approach similar to Flesch and Reuter (2012). The orography used for CRCM5_Ref and the reduced orography used for CRCM5_Oro are shown in Fig. 2a, b. Differences in precipitation are quantified and linked to differences in the moisture flux convergence, which originate from the expected differences in atmospheric circulation.
Impact of antecedent atmospheric moisture
Moisture already present in the atmosphere on 14 June, when high values of vertically integrated water vapour (IWV exceeding 25 mm) extended from the southern US into Saskatchewan and Montana, might have contributed to the precipitation event over southern Alberta. To quantify its impact, a simulation (CRCM5_dryair) with reduced initial atmospheric moisture content between 700 and 925 hPa, i.e. specific humidity set to 0.001 kg kg−1 and condensed water set to zero, over Alberta, the Great Plains and the east Canadian Prairies (Fig. 2c) is compared to CRCM5_Ref. The temporal evolution of the region with decreased atmospheric moisture is followed for the 14–21 June period, and its impact on moisture fluxes and precipitation over southern Alberta is assessed.
Impact of the state of the land surface on precipitation
Milrad et al. (2015) suggested that ET from the land surface in regions to the east and southeast of Alberta acted as a moisture source. To study the impact of soil moisture in the Great Plains, the east Canadian Prairies and Alberta (Fig. 2c) on ET, atmospheric moisture and precipitation, several simulations are performed and compared to CRCM5_Ref. For each of these regions, two simulations are performed: one with initial soil moisture set to zero, in order to suppress ET, and the other with initial soil moisture set to saturation, in order to maximize ET. These simulations are referred to as CRCM5_SMdryGP and CRCM5_SMwetGP for the Great Plains, CRCM5_SMdryCP and CRCM5_SMwetCP for the east Canadian Prairies, and CRCM5_SMdryAB and CRCM5_SMwetAB for Alberta.
Impact of the state of the land surface on streamflow
The state of the land surface in an individual drainage basin plays a critical role in determining both the amplitude and the timing of streamflows in that basin. For example, rain-on-snow events can lead to flash flooding, especially when soils at and downstream of the snow covered regions are saturated and/or frozen. To assess the role of the land surface in the BRB on the magnitude and timing of peak flow, two sensitivity experiments are performed and compared with CRCM5_Ref. In the first experiment (CRCM5_nosnow), snow is initialized to zero for the BRB. In the second experiment (CRCM5_unfrozen), in addition to no initial snow, the state of initial soil moisture is changed from frozen to liquid in the BRB.
The Alberta flood from a climate perspective
From the climate perspective, the frequency of occurrence of a precipitation event similar in magnitude to that of the Alberta flood event, defined as the cumulative precipitation over the 19–21 June period, is determined using extreme value analysis, for observed and modelled data for the 1981–2010 period. As the CaPA dataset is not available for this entire 30-year period, the ANUSPLIN dataset is used. The CRCM5 simulation considered (CRCM5_ERA) is driven by ERA-Interim and has a horizontal resolution of 0.11°. Extreme value analysis is performed on yearly May–June maximum 3-day precipitation amounts for the 1981–2010 period, for both ANUSPLIN and CRCM5_ERA. The May–June transition months are considered since the chances of extreme flooding resulting from both meteorological and hydrological factors combined (e.g. heavy precipitation on snow or frozen and/or saturated soil) is the highest at this time of year. For the analysis, the Gumbel distribution is fitted by the method of L-moments to the 3-day precipitation extremes separately for each grid-cell in the province of Alberta and adjoining regions. The fit of the distribution is tested with the standard Kolmogorov–Smirnov goodness-of-fit (KS) test at the 5 % significance level. The estimated return time of the 3-day precipitation event, defined as discussed above, is obtained from the fitted distribution and the resulting spatial pattern studied for ANUSPLIN and CRCM5_ERA.
The temporal evolution of the estimated return time of an event similar in magnitude to that of the Alberta flood event during the 21st century is assessed using two CRCM5 transient climate change simulations corresponding to Representative Concentration Pathways RCP4.5 and RCP8.5 (van Vuuren et al. 2011). These simulations (CRCM5_CanESM2_4.5 and CRCM5_CanESM2_8.5) are available at 0.44° resolution and are driven by the second generation Canadian Earth System Model (CanESM2) at the lateral boundaries. In this analysis, the Gumbel distribution is fitted to 30-year moving windows of extreme May–June 3-day precipitation amounts shifted in 5-year increments for grid cells located within the western BRB. The likelihood of future occurrences of a precipitation event similar in magnitude to that of the Alberta flood event in the western BRB is studied and discussed.
The influence of anthropogenic GHG emissions on the probability of occurrence of an event similar in magnitude to the Alberta flood event is explored. This is accomplished by comparing large ensembles of 1-year CRCM5 simulations, in GVAR configuration (Fig. 1c), for present-day and pre-industrial cases. As in Kay et al. (2011) and Christidis et al. (2013), the present-day and pre-industrial ensembles differ in their GHG concentrations, sea surface temperatures (SST), and sea-ice concentrations (SIC). In this study, the ensemble for the present-day case (CRCM5_Ind) uses GHG concentrations corresponding to the year 2013, while SST and SIC evolution is taken from ERA-Interim for 2013. Three ensembles for the pre-industrial case (CRCM5_preInd1, CRCM5_preInd2 and CRCM5_preInd3) are considered, where GHG concentrations correspond to their values for 1850, while SST and SIC correspond to pre-industrial conditions.
Pre-industrial SST is obtained by subtracting the SST change attributable to anthropogenic GHG emissions from the 2013 ERA-Interim SST. These attributable SST changes are calculated on a monthly basis from the all-forcings and natural-forcings-only runs of three coupled Atmosphere–Ocean Global Climate Models from the fifth phase of the Coupled Model Intercomparison Project (CMIP5): CanESM2 for CRCM5_preInd1, GFDL-ESM2M (Geophysical Fluid Dynamics Laboratory) for CRCM5_preInd2 and GISS-E2-H (Goddard Institute for Space Studies) for CRCM5_preInd3. Pre-industrial SIC is then estimated using regression models relating SIC to SST (Pall et al. 2011), derived from ERA-Interim.
Each ensemble consists of 1000 members, initialized 1 h apart between 0800 UTC 20 November and 2300 UTC 31 December 2012. Initial conditions are taken from respective spin-ups, which start in early 2012, one for the present-day scenario and one for each pre-industrial scenario. The analysis focuses on comparing the present-day and pre-industrial distributions of May–June maximum 3-day and 1-day precipitation and surface runoff values over different flood-affected regions and average ET over the Great Plains.