Study area and data
This study was carried out on twelve tributary catchments of the St.
Lawrence River located in the province of Quebec, Canada with a natural
hydrological regime (Fig. 1). Spring floods in the populated areas of
the St. Lawrence valley cause frequent damages. For example, the recent
major flooding events in the spring of 2017 and 2019 forced the
evacuation of several thousands of people from flooded neighbourhoods in
the Province and resulted in significant psychosocial and material
damages (Benoit et al., 2022; Lin et al., 2019; Teufel et al., 2019) The
area of the catchments varies between 367 and 4504 km2(Table 1). They are spatially distributed between the north and south
shore of the St. Lawrence River and within four
homogeneous
hydrological regions used by the Quebec Expertise Center for Water
Expertise (CEHQ) in charge of streamflow monitoring and forecasting. The
Northwest St. Lawrence region (Batiscan, Bras du Nord and Matawin
basins) is characterized by a continental climate; the Saint-Laurent
Southwest region (Nicolet, Acadie) is
characterized
by a
maritime
climate; the Saint-Laurent Southeast region (York, Beaurivage,
Bécancour, Famine, Etchemin, Ouelle) is characterized by a
mix of maritime and continental
climate and the Saint-Laurent Northeast region (Godbout) by a maritime
climate (Assani et al., 2010a; Assani et al., 2010b; Mazouz et al.,
2012). The studied basins also encompass three different
physiographic
regions (i) the St Lawrence Lowlands characterized by flat Proterozoic
and Paleozoic sedimentary rocks covered by glaciomarine deposits and
mostly clayed soils (Acadie); (ii) the Canadian Shield on the north
shore of the St. Lawrence River, with its rugged Precambrian gneissic
rocks and sandy soils concentrated in incised glacial valleys (Batiscan,
Matawin, Bras du Nord and Godbout); (iii) the Appalachians on the south
shore of the St. Lawrence, composed of tilted and folded Paleozoic
sedimentary rocks, rolling hill topography, and loamy soils (Ouelle,
York, Etchemin, Bécancour, Famine, and Nicolet) (Table 1). In terms of
land cover, five basins (Batiscan, Godbout, Matawin,
Ouelle, York and Bras du Nord) are
nearly completely (>92%) forested basins, with the
remaining area covered by agriculture and lakes. The southernmost Acadie
basin is dominated by agriculture (72%) with only 25% forest cover,
while the remaining basins (Nicolet,
Etchemin, Bécancour, Beaurivage and Famine) have mixed covers
(agriculture 12-39%, forest: 61-87%) (Table 1).
Daily historical streamflow observations at the outlet of the 12 basins
were obtained from the Quebec Center for Water Expertise (CEHQ). The
length of the observed flow data varies between basins, from 17 to 55
years (Table 1).
Having a good estimation of pre-flood snowpack conditions is one of the
challenges to understand the contribution of snowmelt to peak flow
variability. The difficulty of measuring snow depth and snow water
equivalent (SWE) typically results in limited data being available both
over time and space. While remote sensing can be used to estimate SWE in
low vegetation areas, problems remain in forested areas (Bergeron et
al., 2014; Brown, 2010; Larue et al., 2017; Sena et al., 2016). In
Quebec, a network of snow survey sites has been installed in forested
areas to measure snow depth and SWE every two weeks in the winter and
spring seasons, but the spatial distribution and density of these
stations is low (Nemri and Kinnard, 2019). Consequently, using outputs
of snow and hydrological models seems the only solution to derive long
and continuous SWE and soil moisture records. In a previous study
conducted in the same basins by Nemri and Kinnard (2019), the GR4J
hydrological model (Perrin et al., 2003) coupled to the Cemaneige snow
model (Valéry et al., 2014b) was calibrated and validated in order to
properly simulate basin-wide SWE, soil moisture and daily streamflow.
The calibration methods and validation results are well described in
Nemri and Kinnard (2019). They found that a multi-objective calibration
strategy gave the best simulation of both streamflow and SWE, and the
simulation results using this method were used in the present study. The
GR4J-Cemaneige model was forced by daily precipitation and temperature
date extracted from daily grids produced by the Atmospheric Environment
Information Service (SIMAT) in collaboration with the CEHQ (Bergeron
2015). Historical SWE measurements at 12 measuring points of the Quebec
snow survey network located in or very close to the selected basins were
used in the calibration along with the observed streamflow (see Fig. 1
for locations). The Cemaneige snow module simulates the accumulation and
snowmelt in five altitude bands. The precipitation phase (rain, snow) is
determined using the mean
temperature
of each altitude band. The snow/rain fraction is calculated according to
the function used in the Hydrotel model (Fortin et al., 2001) based on
the minimum (Tmin) and maximum (Tmax)
daily temperature at each altitude band: when Tmax ≤
0 °C, all precipitation fall as snow, while if Tmin ≥
0 °C all precipitation fall as rain, else the snowfall fraction is
estimated as 1-Tmin/(Tmax-
Tmin). These functions are well described by Valéry
(2010) and Valéry et al. (2014b).
In addition, soil moisture measurements were not available for the study
so that the soil moisture simulated by the GR4J model was used. In GR4J,
the hydrological processes in the basin are simplified into two
interconnected reservoirs. The soil reservoir has a maximum capacity
(mm), which is a free parameter to be calibrated, and determines the
amount of water in the basin according to the degree of soil saturation,
which itself is a function of the ratio between the quantity of stored
water and the maximum storage capacity. A summary of the calibrated
model parameters and Nash-Sutcliffe performance criteria is given in
Table 2.