NRC Publications Archive Archives des publications du CNRC Future changes in flood hazards across Canada under a changing Climate

: Climate change has induced considerable changes in the dynamics of key hydro-climatic variables across Canada, including ﬂoods. In this study, runoff projections made by 21 General Climate Models (GCMs) under four Representative Concentration Pathways (RCPs) are used to generate 25 km resolution streamﬂow estimates across Canada for historical (1961–2005) and future (2061–2100) time-periods. These estimates are used to calculate future projected changes in ﬂood magnitudes and timings across Canada. Results obtained indicate that ﬂood frequencies in the northernmost regions of Canada, and south-western Ontario can be expected to increase in the future. As an example, the historical 100-year return period events in these regions are expected to become 10–60 year return period events. On the other hand, northern prairies and north-central Ontario can be expected to experience decreases in ﬂooding frequencies in future. The historical 100-year return period ﬂood events in these regions are expected to become 160–200 year return period events in future. Furthermore, prairies, parts of Quebec, Ontario, Nunavut, and Yukon territories can be expected to experience earlier snowmelt-driven ﬂoods in the future. The results from this study will help decision-makers to effectively manage and design municipal and civil infrastructure in Canada under a changing climate.


Introduction
Floods are the most frequently occurring natural hazard in Canada and around the globe [1][2][3][4]. A number of studies have been performed in different parts of the globe to establish methods for effective quantification of floods and their associated risks [5][6][7][8][9][10][11][12][13][14][15][16][17]. Studies have also investigated methods to quantify compound flooding i.e., that are caused by two or more events contributing to flooding example occurrence of extreme rainfall, variations in astronomical tides, storm surge, and wave action, rise in groundwater levels etc., occurring simultaneously or successively [18,19].
Due to continuous greenhouse gas emissions, climate variables and their extremes have exhibited considerable shifts across the globe [20][21][22]. Changes in key hydro-climatic elements and their extremes have been recorded across Canada [21,23] and unprecedented changes are projected for the future [24][25][26]. These changes in climate, coupled with rapid urbanization, have led to increases in the frequencies and magnitudes of flooding events in Canada. A total of 241 flood disasters have been recorded in Canada between 1990 and 2005 [4] [27]. [The trends in hydrological extremes for 248 Reference Hydrometric Basin Network (RHBN) catchments in Canada were examined by [28]. A decreasing trend in annual maximum flows for catchments located in southern Canada, and an increasing trend in catchments located in northern Canada was obtained. In addition, a robust signal of increases in spring snowmelt driven peak flow was obtained in the months of March and April, whereas a decrease in June month peak flow was obtained. These findings highlight that the behavior of extreme floods has changed across Canada as a consequence of climate change. Therefore, as advocated in previous research [29,30], it is important to obtain reliable flood frequency estimates under a non-stationary climate, such that they can be used to design climate resilient civil and municipal infrastructure across Canada.
General Climate Models (GCMs) simulate complex bio-geophysical and chemical processes occurring within the earth system and their interactions [20]. Land surface schemes are the interface within the GCMs that host important energy budget and water balance calculations occurring within a GCM grid-cell. GCM simulations are performed at a coarse spatial resolution of~110-550 km, which hinders the accurate representation of some of the important physical processes, such as convection that shapes the earth's climate [22]. As a result, large uncertainties have been obtained in GCM projections, especially for variables linked to precipitation [22]. For making future flows and flooding projections at catchment(s) scales, typically, coarse resolution climate projections from GCMs are downscaled, and they are used to generate streamflow responses using a hydrologic model. This approach has been adopted in a number of catchment scale studies, including [25,[31][32][33][34], among others.
Another approach commonly adopted by studies making future flow projections at continental or global spatial scales involve the use of coarse scale gridded runoff projections made by GCMs, and downscaling them to obtain higher resolution runoff estimates. Examples of studies adopting this approach include [35] where 45 km resolution streamflow forecasts for northeastern parts of Canada were generated by dynamically downscaling hydro-climatic forecasts from CanESM2 GCM using a CRCM4 Regional Climate Model [36]. A modified version of the WATROUTE hydraulic modelling scheme [37] was used within the CRCM4 model to simulate high resolution flows. Future changes in projected flood hazard across Europe were estimated by [38]. Dynamically downscaled future climatic projections from two regional climate models (RCMs): HIRHAM model of the Danish Meteorological Institute [39] and the Rossby Centre Atmosphere Ocean Model (RCAO) of the Swedish Meteorological and Hydrological Institute [40], were used as inputs into a hydrological model: LISFLOOD [41] to simulate 5 km resolution river discharges across Europe for historical ) and future (2071-2100) timelines. The climatic simulations and projections from 21 GCMs were used as inputs into a global scale hydrologic model, Mac-PDM.09 [42] to simulate current and future flow regimes at 0.5 • spatial resolution and assessed global water scarcity in future [43]. On the other hand, [44] used runoff simulations from GCMs and simulated high-resolution water level dynamics across the Amazon River basin, using a catchment-based macro-scale floodplain model: CaMa-Flood [45]. The same model was used by [46] to obtain 25 km resolution flow projections across the globe, using coarser resolution runoff projections from 11 GCMs in accordance with representative concentration pathways (RCPs): RCP 2.6, RCP 4.5, RCP 6.0, and RCP 8.5 [47].
This study investigates changes in the frequencies and timings of large floods (referred to address 100-year and 250-year return period flooding events hereafter) across Canada under projected future influences of climate change. The analysis presented generates novel information, as only a handful of studies (predominantly global assessments) preceding this study have assessed changes in flood hazards on a Canadian scale. In most of these studies, only changes in flooding frequencies and magnitudes have been assessed. This study extends the assessment to also analyze projected changes in flood timings, which is an important piece of information that is required for effective flood risk management. Finally, this study takes into consideration a larger ensemble of future runoff projections as compared to previous studies, which means that the results generated from this study account for the uncertainty associated with future runoff projections made by GCMs more effectively than the previous studies.

CaMa-Flood Hydrodynamic Model
CaMa-Flood [44][45][46]48,49] is a global scale-distributed hydrodynamic model that routes input runoff generated by a land surface model to oceans or inland seas along a prescribed river network map. Water storage is calculated at every time-step, whereas variables such as: water level, inundated area, river discharge, and flow velocity, are diagnosed from the calculated water storage. River discharge and flow velocity are estimated using a local inertial equation. Floodplain inundation is modelled by taking into consideration the sub-grid scale variabilities in the river channel and floodplain topography. The parameters used in CaMa-Flood model are listed in Table 1. A river channel reservoir has three parameters: channel length (L), channel width (W), and bank height (B). The floodplain reservoir has a parameter for unit catchment area (A c ), and a floodplain elevation profile that describes the floodplain water depth D f as a function of the flooded area, A f . The topography-related parameters i.e., surface altitude (Z), distance to downstream cell (X), and unit catchment area (Ac) are calculated using the Flexible Location of Waterways (FLOW) method [50]. Finally, a Manning's roughness coefficient parameter (n) is used to represent the roughness in the river channel. The CaMa-Flood model has been validated extensively for its ability to simulate runoff in the largest catchments of the globe [44,51]. For instance [45] evaluated the performance of the CaMa-Flood model in simulating flow characteristics in 30 major river basins, including the Amazon, Mississippi, Parana, Niger, Congo, Ob, Ganges, Lena, and Mekong. The model was found to be able to simulate flood inundation characteristics in these basins well. Furthermore, [49] evaluated the importance of adding a new computational scheme to help support the representation of flows through bifurcation channels in CaMa-Flood. The model was found to be able to perform realistic hydrodynamic calculations in the complex, resulting in mega delta with numerous bifurcation channels. Given the high credibility of the model in simulating river flow and flood inundation dynamics, the model has been used to assess the impacts of climate change at regional to global scales [46,[51][52][53][54][55]. This study uses the globally calibrated version of the CaMa-Flood model that was used to generate global scale runoff projections in [46].

Generation of 25 km Resolution Historical and Future Flows across Canada
Coarse resolution historical  and future (2061-2100) runoff simulations obtained from GCMs were used as inputs into the CaMa-Flood model calibrated at 25 km resolution and flow estimates covering the entire Canadian landmass. A spin-up period of two years was considered, and flows generated during this period were ignored during the assessment.

Grid-Wise Estimation of Future Flooding Frequencies of Historical 100-and 250-Year Floods
Generated historical and future flows at each 25 km grid were used to estimate future changes in the frequencies of historical 100-and 250-year floods. Generalized extreme value (GEV) distribution was fitted to the historical annual maximum flow series. The cumulative distribution function of the GEV distribution is expressed in Equation (1): where Q is the random variable, q is a probable value of Q, κ is the shape parameter, ε the location parameter, and α is the scale parameter. Parameters of GEV distribution were estimated using the method of L-moments [56].
Flow quantiles corresponding to 100-and 250-year return period floods were estimated for historical timelines. GEV distribution was then fitted to the annual maximums of the future flow series, and return periods corresponding to historical 100-and 250-year flood magnitudes were estimated.

Aggregation and Uncertainty Assessment of Projected Changes in Flooding Frequencies
Future flood frequency projections from different GCMs (corresponding to a particular emission scenario) were aggregated, and uncertainty magnitudes were quantified for each 25 km grid. Previous studies have found large uncertainties in GCM simulated projections of precipitation-related variables, with even the sign of change being found to be uncertain in many regions of the globe (IPCC 2013). Therefore, in this study, robustness of flood frequency projections was taken into consideration when aggregating flooding frequency estimates. The term 'robust' was used in this paper to refer to projections/projected changes that concurred by more than 50% of the projections analyzed. If equal numbers of projections conveyed increases/decreases in flooding frequencies in the future for a particular grid, then that grid was associated with 'non-robust' flood frequency projections. When aggregating future flood frequencies, projections that concurred with the robust sign of change in flooding frequencies were considered for aggregation and uncertainty assessment. Aggregated flooding frequencies were calculated by finding the median of future return period values, whereas uncertainty was quantified using Equation (2): where U r denotes the calculated uncertainty magnitude, and RP r,0.75 , RP r,0.50 , and RP r,0.25 denote the 75th, 50th, and 25th quantiles of the robust flood frequency projections, respectively.

Estimation of Historical and Future Flood Timing
When assessing changes in flood timing, flow events exceeding the 95th quantile flow value over the entire time-period of concern were considered as flooding events. For each individual CaMa-Flood grid located within Canada, flooding events were identified for both historical and future time-periods that met this criterion. The month of the year corresponding to which the largest frequency of flooding events were simulated was recorded as the time of flooding. The differences in the time of flooding between historical and future time-periods were analyzed to identify the impact of climate change on flood timing across Canada.

Aggregation of Historical and Future Flood Timing
Flood timing values were aggregated at each 25 km grid by only taking into consideration projections from GCMs that were able to accurately simulate the robust flow month. The robust flow month is regarded as the month that is projected with the largest number of flooding events by more than 50% of the projections analyzed. If none of the months have been concurred upon by more than 50% of the projections for a grid, then that grid is marked as having 'non-robust' flood timing results. The aggregated results for historical and future timelines are compared, to assess changes in flood timing between the two timelines.

Study Region
In this study, assessment of future changes in flooding frequencies and timings is performed across the entire Canadian landmass. Canada consists of 10  Different regions of Canada exhibit considerable differences in landscape and climate. Canada encompasses eight climate regions with different geophysical characteristics (Massey and Connors 1985). These are: (i) Pacific Maritime climate that is shaped by the presence of Pacific Ocean and is characterized by mild but extremely wet winters, and cool and dry summers. Regions located along British Columbia's west coast and its border with Yukon Territory are a part of this climate type; (ii) Cordilleran climate is influenced by continental air masses and Pacific air streams. It is found in regions covering eastern British Columbia, Yukon Territory, and small portions of southwestern Alberta. It is characterized by cold and wet winters, and warm and dry summers. The climate experienced within this climate type varies considerably spatially, because of the presence of the Rocky Mountains and insulated valleys. This climate type is found in regions covering eastern British Columbia and the Yukon Territory, as well as small portions of southwestern Alberta; (iii) Atlantic Maritime climate is influenced by western continental air masses, and it is modified by the presence of the Atlantic Ocean. This climate type is characterized by cold and wet winters and hot and wet summers. Regions encompassing New Brunswick, Nova Scotia, Prince Edward Island, and southeastern Newfoundland exhibit this climate type; (iv) Southeastern climate is influenced by the continental air masses that are modified by the presence of the Great Lakes. This climate type is characterized by cold and wet winters, and hot and wet summers. Regions that characterize this climate type include Ontario, Quebec, and parts of Nova Scotia and New Brunswick; (v) Prairies climate type is influenced by the continental air masses, and it is characterized with a wide annual temperature range with very cold winters and very hot summers; the southern regions of Alberta, Saskatchewan, and Manitoba provinces demonstrate this climate type; (vi) Boreal climate is influenced by Arctic and Pacific Ocean air masses. This climate type is characterized with very cold and dry winters, and warm and wet summers. Regions forming a continuous belt from Newfoundland and Labrador passing central Quebec and Ontario, across the Prairies, and west to the Rocky Mountains exhibit this climate type; (vii) Arctic climate region is influenced by the air stream coming from the Arctic ice pack. This region is characterized with a very harsh cold climate, permanent snow-cover, short cool summers, and minimal precipitation. Most of the Nunavut, and northern parts of Northwest Territories and Quebec exhibit this climate type. Lastly; (viii) Taiga climate region is associated with long cold winters for more than six months. This climate has some precipitation in summer and very low precipitation in winter.

Data Used
GCM-simulated daily runoff data for historical    [47] were collected. The list of GCMs considered for analysis in this study is provided in Table 2. Runoff data for above mentioned timelines was collected for a total of 105 (84 future and 21 historical) realizations from the CMIP5 multi-model ensemble. Table 2. GCM-RCP combinations for which at least one set of historical and future realizations were available in the CMIP5 multi-model ensemble. Note that for some cases; more than one realizations were available, and all of them are included for assessment in this study.   [57].

Results and Discussion
This section presents the results obtained from the assessment of projected future changes in flood frequency and timing across Canada.

Projected Changes in Flooding Frequencies
Future flooding frequencies of historical 100-year and 250-year flooding events aggregated using the approach defined in Section 2 (referred as robust GCM median approach hereafter) are presented in Figures 1 and 2 respectively. The regions presented in blue (brown) are projected with future increases (decreases) in flooding frequencies, whereas the regions presented in green are projected with no considerable changes in flooding frequencies. Regions where non-robust (described in Section 2) projections of flood frequencies are obtained are shown in white.   [58] attributed largely to decreases in snowfall and increases in temperatures during the winter months. Finally, small parts of Nova Scotia, Newfoundland, and Labrador, and northernmost regions of Nunavut, and south-west British Columbia are also projected to experience no considerable changes in flooding frequencies in the future. A comparison of changes projected for 100-year and 250-year return period flooding events indicate that the spatial structure of projected changes is similar for flooding events of both magnitudes.
The spatial distribution of the projected changes in flooding frequencies was found to be similar under the RCP 4.5 and RCP 8.5 emission scenarios, whereas different spatial structures of projected changes were obtained under emission scenarios RCP 2.6 and RCP 6.0. For example, under RCP 6.0, the provinces of Yukon Territory, Northwest Territory, and Nunavut were projected with lower flood frequencies than that projected under the RCP 4.5 and RCP 8.5 emission scenarios. It should however be noted that the total number of GCMs from which runoff projections were available under RCP 6.0 (10) was lower than other emission scenarios, i.e., RCP 2.6 (15), RCP 4.5 (19), and RCP 8.5 (19), which can also contribute towards some of these differences. Finally, changes projected under RCP 2.6 were found to be of the smallest magnitudes as compared to other emission scenarios, with large areas projected with negligible changes in the future.
The above results are based on an approach where the robustness of GCMs in predicting the projected sign of change in runoff is taken into consideration when aggregating the projected changes. To investigate the impact of aggregation procedure method on the obtained results, a relatively straightforward method that does not consider the robustness of projections is used to aggregate them. In this method (referred as 'all GCM median approach' hereafter) the median of all projections is taken to obtain the future projected return periods across Canada. The results of future return periods of historical 100-year and 250-year return period flooding events obtained from this approach are shown in Figures 3 and 4, respectively. A comparison of Figures 1 and 2 with Figures 3 and 4 highlight the important similarities and differences in the magnitudes, and the spatial distributions of future projected flood frequencies. It is noted that northern parts of Canada, southwestern Ontario, and northeastern Quebec are projected with an increase in flooding frequencies when the results are aggregated from either approach. Similarly, the northern prairies region and north-central Ontario are projected with decreases in flooding frequencies from either approach. A key difference is obtained in the magnitudes of the projected changes, where higher magnitudes of absolute (positive or negative) changes were obtained from a robust GCM median approach as compared to the GCM median approach. This is likely because projected changes cancel out when averages are taken across all GCMs in the GCM median approach. For the same reason, aggregation using the all GCM median approach was found to result in more areas with no considerable changes as compared to the robust GCM median approach.  Uncertainty magnitudes obtained from future flood frequency projections of historical 100-year floods obtained in the cases of the robust GCM median approach and the GCM median approach are presented in Figures 5 and 6, respectively. To present the spatial heterogeneity of uncertainty effectively, normalized values of uncertainty magnitudes are presented in the figures. The spatial distribution of uncertainty from both approaches was found to be similar; however, the uncertainty magnitudes obtained from all GCM median approaches were found to be higher than that obtained from the robust GCM median approach. Overall, increases in flood magnitudes projected in the northern provinces of Yukon, Northwest Territories, and Nunavut, northern Quebec, and south-west Ontario were found to be among the least uncertain results obtained. The decreases in flood frequency projected for the prairies region, northern Ontario, British Columbia, and Newfoundland and Labrador were found to be among the most uncertain results. These results indicated that there was a larger degree of confidence in the projected increases in flooding frequencies in parts of Canada than the projected decreases. Finally, between the four RCPs, the least uncertainty was found to be associated with the projected changes made under RCP 4.5 as compared to the other RCPs.

Projected Changes in Flood Occurrence Timing
Spatial distributions of flood timing obtained for historical and future timelines are presented in Figures 7 and 8 for two extreme RCPs with regard to greenhouse gas emissions: RCP 2.6 and RCP 8.5, respectively. In the figures, months where wintertime precipitation is likely to contribute to peaks, i.e., November, December, January, February, are shown in the shades of pink, months where snow-melt can be a dominant factor, i.e., March, April, May, are shown in the shades of blue, while months where summertime convection can be a dominant contributor to peaks, i.e., June, July, August, September, October, are shown in shades of green. Grids with non-robust flood timing results are shown in a grey color. Results clearly highlighted projected future increases in the total area effected by snowmelt driven floods (shown in the shades of blue), as well as an earlier onset of snowmelt driven floods in the future. These changes were most evident in the northern and central regions of Canada. Regions in Ontario and Quebec were projected with earlier summertime extreme flows (shifts from April/May to March). Most of the regions from Nunavut and Yukon Territories were projected to have earlier summertime extreme flow changes (from May to April). An earlier onset of snowmelt driven floods in the future was also evident from Figure 9, where grids that are projected with up to two months of early spring melt are shown. These results are in line with the findings from observational studies performed in different locations across Canada, where an earlier onset of snowmelt driven floods has been documented [59][60][61][62], as well as projected under the influences of climate change [34,[63][64][65]. This observation is obtained consistently across all four emission scenarios considered for assessment, although it is noted that the GCMs were more uncertain on the prediction of the peak flow month in the cases of RCP 4.5 and RCP 8.5, than in the cases of RCP 2.6 and RCP 6.0.  robust GCMs i.e., those who concur on the sign of change as projected by more than 50% of the

Conclusions
The impacts of climate change have been detected on the characteristics of streamflow and their extremes in catchments distributed across Canada. Given this non-stationarity in climatic conditions, there is a need to quantify the projected future impacts of climate change on flow extremes to better design civil and municipal infrastructure in Canada. It is also important to account for sources of uncertainties when making future projections, so that policymakers can review them before national flood protection guidelines are put into place.
This study quantifies future changes in the frequencies and timings of flooding events across Canada as a consequence of climate change. An ensemble of 84 future runoff projections made by 21 GCMs are considered for assessment. A state-of-the-art mesoscale hydrodynamic model: CaMa-Flood is used to simulate 25 km resolution historical and projected future flows from coarse resolution GCM runoff estimates. The changes projected by different GCMs are aggregated, and associated uncertainty is quantified using two approaches: (1) where only projections made by 'robust' GCMs (i.e., those who concur on the sign of change as projected by more than 50% of the GCMs) are considered, and (2) where projections made by all GCMs are considered. In general, it was found that the spatial distribution of the projected changes is similar in the results obtained from both approaches, whereas the magnitudes (both positive and negative) are found to be larger in the first approach than in the second approach. In terms of uncertainty, both approaches demonstrate similar spatial structures; however, results from the second approach demonstrate higher uncertainties than the first approach.
The spatial distributions of projected flood frequency changes convey that the northern provinces of Canada: Northwest Territories, Yukon Territory, and Nunavut, and southwestern Ontario, can be expected to have higher flood frequencies in the future, with a return period of 100-year historical floods becoming 10-60 years by the end of the 21st century. On the contrary, the northern prairies and north-central Ontario can be expected to experience lower flood frequencies, with a return period of 100-year historical floods to become 160-200 years in the future. This projected increase (decrease) in flooding frequencies in the above-mentioned areas is also found to be among the least (most) uncertain changes projected for Canada, indicating that there is a high confidence that flood hazard will increase in the above-stated regions of Canada in the future.
An assessment of projected changes in future flood timing indicates earlier snowmelt in almost all regions of Canada. This is expected, given that future temperatures are projected to increase across Canada under the influence of climate change [23]. Signs of increases in snowmelt-driven floods, and earlier snowmelt have been detected in historical flow records [66,67], as well as have been projected for many Canadian rivers [34,[63][64][65]. The results obtained are thus also in line with the findings made in many previous studies performed at catchment scales in Canada.
The flood hazard and risk changes identified in this study can serve as useful guides for decision-makers in Canada to identify flood-hazard areas, and to prioritize appropriate mitigation and response efforts in the face of global climate change. This work can be expanded by overlapping generated flood-hazard maps with exposure elements such as population and water resource management infrastructure, to identify flood risk areas. Efforts in this direction are currently underway.
Author Contributions: A.G. conceptualized the research, performed the formal analysis, and wrote the first draft of the paper. A.G. and S.P.S. provided feedback on the research approach, and reviewed and edited the first draft of the paper. All authors revised the paper and agreed on the final version of the paper.