Discussion

The results found in this study using measured and simulated pre-flood hydroclimatic factors contribute to improve our understanding of interannual variations in spring flood characteristics. Overall, the ranking of preconditioning factors based on their frequency of appearance as significant predictors in the linear models of Qmax across the twelve basins is as follow: (i) snowmelt intensity (mean and max), (ii) rainfall intensity (mean and max), (iii) snowmelt volume, (iv) rainfall amount, (v) peak SWE and (vi) soil moisture. One note of caution to be mentioned is that the stepwise approach finds the best combination of factors explaining the most variations in Qmax, and as such can remove predictors that are still important on their own, but that are redundant (collinear) in a multivariate context. Still, snowmelt intensity also appears as the most important univariate predictor of Qmax as shown by the correlation analysis (Fig. 5), but peak SWE (SWEmax), which is the second-best univariate predictor of Qmax (Fig. 5) was often excluded from the multivariate models. A thicker snowpack is more likely to survive later into the spring season and be subjected to faster melt rates (e.g. Aygün et al., 2022; Musselman et al., 2017), which could explain the redundant predictive power of these two variables, along with Meltsum, in the multivariate models. Still, SWEmax by itself could only explain 10 to 28% of the variability in Qmax in 8 out of 12 basins as shown by the bivariate correlation analysis (Fig. 5). This shows that the memory of the snowpack in early spring is not sufficient to accurately forecast springtime flood magnitude in southern Quebec.
The pre-flood mean melt intensity (Meltint) emerged as a good predictor of the peakflow magnitude in seven basins (Acadie, Famine, Beaurivage, Etchemin, Matawin, Bras du Nord and York) and was the most skillful predictor for four of these basins (Acadie, Famine, Beaurivage, Etchemin). The maximum melt intensity (Meltintmax) was also a good predictor but only in the Nicolet and Ouelle basins. So, overall, snowmelt intensity appears to be the dominant control on peakflow magnitude for 9 of the 12 basins studied. SWEmax on the other hand was only retained as a significant predictor of Qmax in four basins (Nicolet, Bécancour, Famine and Etchemin).
The sensitivity of peakflow magnitude to antecedent hydroclimatic conditions also varied according to land cover and use. Overall, the interannual variability in peakflow magnitude was primarily controlled by the snowmelt dynamics (initial snowpack SWE, snowmelt amount and intensity) in the more southerly and/or more agricultural basins, while in the northern, snowier and more forested basins rainfall conditions, especially the rainfall intensity, were more important (Fig. 7). This can at first appears counter intuitive, that spring floods in southerly basins with less snowfall and thinner snowpacks are controlled by snowmelt dynamics, while rain events are the main trigger of floods in colder, snowier northern basins. However, snowmelt in forested basins is slower due to shading by the canopy (Ellis et al., 2011; Gelfan et al., 2004), so that Qmax variability is less dependant on SWEmax. Also, the more porous forested soils and reduced soil freezing under thick snow covers have been found to favor snowmelt infiltration, which attenuates flood peaks (Aygün et al., 2022). On the other hand, SWEmax is more variable interannually in agricultural basins (Table 1); melting in open fields is fast, and infiltration is restricted in the often clay-rich and compacted soils (Aygün et al., 2020), which all boost the influence of snowpacks and snowmelt rates on peakflow interannual variability.
Soil moisture was considered to be a key factor in controlling runoff in snow-dominated basin in previous studies (Wever et al., 2017). In this study, no coherent correlation was found between the simulated degree of soil saturation and peakflow variations. Even within the multivariate regression analysis this factor was found to be a significant predictor of peakflow magnitude and timing only for three basins, and a counterintuitive negative effect was found for two of these basins (Bécancour and York). The negative effect of soil moisture on Qmax could be an indirect effect reflecting the depletion of the soil reservoir during cold winters with limited snowmelt and rainfall, which could then be associated with thicker snowpacks and higher flows in the following spring. However, the lumped GR4J model does not consider soil freezing processes, which could be important especially in agricultural basins with thinner snowpacks (Aygün et al., 2020). Therefore, further research is needed in Quebec basins using models that explicitly represents soil freezing and fall moisture ‘soil memory’ in order to better simulate pre-melt soil moisture and its effect on runoff partitioning. Soil freezing is often assumed to play an important role on infiltration, but deep snowpacks can also inhibit soil freezing and cancel its impact on infiltration (Aygün et al., 2021; Aygün et al., 2022).
As for the peakflow timing, a pattern emerged in which an increasing amount of low-intensity rainfall combined with increased snowmelt amounts delayed flood peaks in the more northerly basins, while in the more southerly basins, slower snowmelt and increased rainfall amounts to a lesser extent, led to later flood peaks (Table 7).
Our results are different than those reported by Curry and Zwiers (2018) in the Fraser River basin in western Canada, between the Coast Mountains and the Continental Divide, where the generation of spring runoff was found to be controlled mainly by the maximum accumulated SWE and secondly by the melt rate, with rainfall and soil moisture playing lesser roles. On the other hand, Coles et al. (2016) found that the processes responsible for the generation of runoff in the Canadian prairies hillslopes were, in order of importance, the total snowfall, snow cover amount, fall soil surface water content (0-15 cm) and melt rate. The more humid climate of southern Quebec compared to the Canadian Prairies, and the lower elevation compared to the mountainous basins of western Canada, could explain the fact that interannual variations in accumulated SWE are generally less important than the melt rate and the quantity and intensity of rainfall events during snowmelt. Our results showed that interannual variations in snowmelt volumes were either the prime driver, or as equally important as rainfall, in controlling flood volume variability (Table 5 and Fig. 4). However, our regression analysis showed that snowmelt variables were the most important drivers of peakflow interannual variability in the more agricultural southern basins, even in the southernmost Acadie basin where snowmelt contributes less water than rainfall to flood volumes. Conversely, in the more northerly, snowy and forested basins flood volumes were primarily controlled by snowmelt volumes, whereas rainfall was more important in controlling interannual variations in peakflow.
Initial basin conditions (snow storage, soil moisture) and their forecasting skill are very important for the seasonal prediction of streamflow (Foster et al., 2018; Koster et al., 2010; Li et al., 2009; Mahanama et al., 2012) but these variables are not well measured in most basins. Turcotte et al. (2010) discussed the difficulties encountered by the prediction systems developed for Quebec basins due to errors in the snow observation methods. Therefore, using satellite products of snow cover in conjunction with physically-based models might be a good way forward to improve our understanding of the spring freshet generation mechanisms and the independent role of snow cover, rain on snow events and the soil moisture status in future snow hydrology studies in Quebec. Still, this study showed that knowledge of snow storage in early spring (SWEmax) gives only limited forecasting capability for flood magnitude and timing in southern Quebec, and that synoptic scale weather variability plays an important role in defining rainfall and snowmelt intensity, which contribute largely to runoff processes and ensuing flood characteristics.