Introduction

The hydrology of cold regions is characterized by long winters dominated by snowfall and rapid spring melting, which is the main cause of the high spring streamflow (Pomeroy et al., 2016). In the province of Quebec, Canada, the amount of accumulated snow is very important, with a mean annual maximum of 200 to 300 mm of snow water equivalent (SWE) (Brown, 2010). As a result, the streamflow regime is nival to nivo-pluvial and strongly influenced by the snowmelt contribution, which occurs between April and June depending on the basin geographic location and the year. In the southern part of the province, snow begins to accumulate in November and melting occurs between March and May. Peak flow typically occurs in the spring following snowmelt while a second flood peak typically occurs in summer in response to convective rainstorms, or in the fall caused by the advection of moist air masses with above-freezing temperatures. In northern Quebec, snow accumulation begins earlier in October and melting occurs later, in June and July, with a single streamflow peak mainly driven by snowmelt (Assani et al., 2010a; Buttle et al., 2016; Saint-Laurent et al., 2009).
Knowledge of the SWE stored in the winter snowpack and of ablation dynamics in the spring is key for accurate streamflow predictions and operational management of reservoirs in Quebec. As such, a reliable seasonal forecast of spring freshet based on winter and early spring conditions is essential for reservoir operators to optimize two conflicting objectives, namely flood protection and hydropower production (Turcotte et al., 2010), as well as for governmental agencies to prepare flood mitigation and disaster relief measures. Nevertheless, the relation between snowpack conditions and the inter-annual variations in the magnitude and timing of the spring peak flow is not straightforward, due to the complexity of spring runoff generation mechanisms (Merz and Blöschl, 2003; Tarasova et al., 2019). In fact, the same annual snow accumulation can induce more or less severe floods because of the multiplicity of antecedent hydrological conditions that can control runoff in addition to snowpacks, such as meteorological conditions during the melt period, the occurrence of rain-on-snow events, soil moisture and soil freezing. Therefore, a good understanding of the flood generation mechanisms and of the relative contribution of the key driving factors involved is essential to explain the interannual variability of the spring peak flow characteristics and guide future forecasting efforts.
The variability in flood characteristics in North America has been linked with large-scale climatic indices, and several previous studies have studied how these indices influence extreme floods (Assani et al., 2010a; Assani et al., 2010b). Mazouz et al. (2012) studied the relationship between the interannual variations of high spring flow characteristics in southern Quebec (magnitude, duration, period of occurrence, frequency, and variability) and several global climatic indices using canonical correlation analysis. A significant correlation between the Atlantic Multi-Decadal Oscillation (AMO) and North Atlantic Oscillation (NAO) indices and four flood characteristics (duration, period of occurrence, frequency, and variability) was found, while no relationship was found between these indices and the flow magnitude. This correlation was explained by the low temperature during the negative phases of the AMO and the positive phases of the NAO, which causes a later date of occurrence, a higher frequency, a longer duration and lower variability of heavy spring floods (Mazouz et al., 2012).
Heavy rainfall events during spring can also contribute significantly to runoff while also accelerating snowmelt, causing more devastating floods (Fang and Pomeroy, 2016; Pomeroy et al., 2016; Sui and Koehler, 2001) depending on the antecedent conditions of the snowpack (Garvelmann et al., 2015). The relative contribution of melting and rainfall to runoff and floods becomes more complicated during rain-on-snow events and affects the results of forecasting studies. Rain-on-snow events in Canada have been addressed by several studies (Dyer, 2008; Mccabe et al., 2007; Pomeroy et al., 2016; Wayand et al., 2015). In Quebec, many devastating spring floods have been caused by a combination of heavy rainfall during melting and a deep accumulated snowpack, such as for the Richelieu river floods in 2011 (Saad et al., 2015). Teufel et al. (2018) studied the devastating spring floods that occurred in Montreal during May 2017, showing that heavy rainfall events during April and May combined with snowmelt were the culprit of these extreme events. Likewise, antecedent moisture conditions in catchments plays a key role in runoff generation during melt; the degree of soil saturation below the snowpack determines the infiltration and runoff of snowmelt water in snow-covered basins (Koster et al., 2010; Mahanama et al., 2012). These two studies quantified the contributions of snow accumulated on January 1st and soil moisture to the skill of seasonal forecasts of spring snowmelt in 23 basins of the eastern United States. They demonstrated that despite the important role of snow, the contribution of soil moisture to the skill of streamflow forecast was significant. Several studies showed also the importance of ’soil memory’, i.e. soil moisture conditions before soil freezing (Curry and Zwiers, 2018; Mahanama et al., 2012; Webb et al., 2018; Wever et al., 2017) so that understanding the relationship between floods, soil moisture and snow cover in these basins is necessary to understand the spring streamflow generation.
The main challenges in studying how antecedent hydrological variables control spring floods are the choice of predictors, the possible interaction between them, the period over which these factors are calculated and the unavailability of observations for some variable such as soil moisture and SWE (Coles et al., 2016; Curry and Zwiers, 2018; Fang and Pomeroy, 2016; Nied et al., 2013; Nied et al., 2014). In western Canada, Curry and Zwiers (2018) investigated the influence of hydroclimatic conditions on the variability of annual maximum daily flow magnitude using multivariate linear regression models in a snow-dominated basin. Potential predictors were ranked according to their degree of control on the maximum basin peak flow. The maximum annual snowpack (SWEmax) ranked first, followed by the snowpack melting rate calculated between the date of SWEmax and that of peak flow, the Pacific Decadal Oscillation (PDO) and El Niño-Southern Oscillation (ENSO), and finally the rate of air temperature warming between April 1stand the date of peak flow. Some of the variables used were measured while others were simulated by a hydrological model, such as the snowmelt rate and soil moisture. Coles et al. (2016) studied snowmelt runoff generation in the Canadian prairies hillslopes using a decision tree learning approach to rank the processes responsible for the generation of runoff. The impact of antecedent hydroclimate conditions on flow peaks were, in order of importance, total snowfall, snow cover, fall soil surface water content, snowpack melt rate, melt season length, and fall soil profile water content. A salient result was the importance of the degree of soil saturation during the fall before the frost period, or ‘soil memory’, in controlling runoff. Fang and Pomeroy (2016) studied the sensitivity of the June 2013 flood in Calgary to pre-flood conditions as simulated by the physically-based Cold Region Hydrological Model (CRHM). They studied streamflow generation processes by varying the amount of precipitation, the land cover and the soil storage capacity during the pre-flood period. They showed that runoff increases rapidly in response to prior accumulation of snow and soil moisture and that antecedent soil moisture in the basin is a better indicator of flood magnitude than the antecedent snowpack in this basin. Using multiple linear regression, Maurer and Lettenmaier (2003) found that soil moisture dominates runoff predictability in the Mississipi River basin for lead times of 1.5 months, except in summer in the western part of the basin where snow dominates. In western Canada, Dibike et al. (2021) found that basin average maximum SWE, April 1stSWE and spring precipitation were the most important predictors of both annual maximum and mean springtime flow, with the proportion of explained variance averaging 51.7%, 44.0% and 33.5%, respectively.
Outside North America, Zhang et al. (2014) used path analysis to identify influential climatic factors on spring floods in an alpine catchment in Xinjiang, China. They found that winter snowfall and mean thawing degree days in spring had the most direct influence on flood peaks, while accumulated freezing degree days in winter had an indirect influence on floods.
In Quebec basins, the hydroclimatic drivers, or ‘predictors’ of interannual variations in the magnitude and timing of spring flow peaks are not well identified and have not been studied except in relation with global climatic indices (Assani et al., 2010a; Assani et al., 2010b; Mazouz et al., 2012). Hence, the main objective of this study is to identify and better understand the factors that control the interannual variability of spring freshet characteristics in the tributary catchments of the St. Lawrence River in view of improving seasonal flood forecasts. The limited availability of snow depth, SWE and soil moisture observations has always been an obstacle when analyzing historical hydrological datasets. In this study, we use outputs of simplified conceptual models to simulate snow accumulation and melt as well as soil moisture storage in the basins. We seek to answer the following questions: (i) is the inter-annual variability of the spring freshet magnitude mainly dependent on the antecedent snowpack, with higher flow peak occurring in years with deep snowpack? (ii) Does the quantity and intensity of rainfall during the pre-flood period have a strong influence on the characteristics of the spring freshet? (iii) How do the preconditioning factors vary between basins, according to their latitude, physiographic region and predominant land cover?