Conclusions
This study used observed streamflow records from 12 tributary catchments
of the St. Lawrence River in southern Quebec, Canada, combined with
observed and simulated antecedent hydroclimate conditions during the
pre-flood period, to investigate their control on spring flood magnitude
and timing. The following main conclusions can be drawn from the study:
- Interannual variations in the volume of meltwater available for runoff
were usually more important, or else equally important, than rainfall
volumes to explain the interannual variability of spring flood
volumes.
- In contrast, the late winter snowpack (peak SWE) was not by itself a
strong predictor of spring flood magnitude and timing; as such, the
‘snowpack memory’ offered only a limited potential for seasonal flood
prediction.
- The ‘soil memory’ effect, represented here by the simulated
soil moisture content, was poorly related to flood characteristics;
however, the effect of soil freezing was not considered and should be
studied further.
- The snowmelt rate during the pre-flood period was the most ubiquitous
and skillful predictor of spring flood magnitude.
- SWE and snowmelt dynamics dominated the interannual variability of
flood magnitude in the more southerly and agricultural basins, due to
more variable snowpack conditions from year to year, faster snowmelt
and restricted infiltration. In the northern, snowier forested basins,
rainfall variability was instead more important in driving interannual
variations in flood magnitude, which is attributed to the documented
slower melt rates under forest canopies and the buffering effect of
the more porous and less frozen forested soils.
- Seasonal to sub-seasonal (S2S) spring flood prediction in the
humid-continental climate of southern Quebec would require an accurate
knowledge of pre-melt snowpack SWE, but also S2S predictions of
rainfall and temperature, a proxy for snowmelt rates, which is a more
challenging requirement (White et al., 2017). Using correlations with
large scale climate indices (Mazouz et al., 2012) might represent an
option, while advances in the application of Machine Learning (ML)
hydroclimate predictions techniques (Başağaoğlu et al., 2022; Mosavi
et al., 2018) might help to unravel more complex relationships between
potential hydroclimate predictors and flood conditions and help
forecasting seasonal flood characteristics.