Results
Interannual and spatial variability of peak streamflow and
its
timing
The variability of the spring specific streamflow magnitude
(Qmax) observed in the twelve basins is shown in Fig.
3a, while volumetric streamflow values are given in Table 4. The highest
specific Qmax values are observed in the more
agricultural basins (Acadie, Nicolet and Beaurivage), while the forested
basins (Batiscan, Matawin, York and Godbout) have lower specific flows,
except Bras-du-Nord (#9) and Ouelle (#8) (Fig. 3a).
Qmax is also more variable in more agricultural compared
to forested basins, again with the exception of the Bras-du-Nord and
Ouelle basins. The seasonality of the spring peak flow is shown in Table
4. For the two northernmost basins (Godbout and York), 90% of the
spring flow peaks occurred the latest, in May. For the two southernmost
basins (Nicolet and Acadie), melting occurred earlier with 40% of
peakflow events occurring in March and 40% in April. For the remaining
basins, 65% of the peakflow events occurred during April. The peakflow
timing shows pronounced interannual variability as well as spatial
variability between basins (Fig. 3b). The general increasing trend from
south to north in the peakflow timing also appears clearly. Also, for
the three completely forested basins located on the Canadian Shield,
i.e., Matawin (#7), Bras du Nord (#9) and Batiscan (#10), and melting
occurs later compared to basins at the same latitude with less forested
areas, such as Beaurivage (#5: 60% forest cover) and Etchemin (#6:
74% forest cover). Therefore, the spatial distribution of the
Qmax timing appears to respond primarily to latitude and
secondarily to land cover.
Contribution of snowmelt and rain to flood
volumes
The contribution of pre-flood water volumes (snowmelt and rainfall)
available for runoff to the total flood volume estimated from the
multivariate regression method is illustrated in Fig. 4 for the twelve
basins. The rainfall contribution is high for the southernmost Acadie
basin (#1) compared to the other basins. While the median rainfall
contribution in Acadie (0.25) is only slightly higher than that of the
other basins, the interannual variability is large, with the third
quartile of the distribution reaching near 0.75, and in some extreme
years, rainfall was the sole contributor. For the other basins, the
median rain contribution is around 0.20, but can be as high as 0.60,
which shows that the rain contribution to the spring flood volume in all
basins can be important.
Annual variations in water volume (snowmelt + rainfall) available for
runoff explain between 67% and 93% of the interannual variability in
flood volumes (Table 5). Rainfall and snowmelt volume variability had a
comparable effect on flood streamflow volume for five basins (Nicolet,
Acadie, Batiscan, Matawin and Bras du Nord), whereas for the other seven
basins, the interannual variability in flood volume is more controlled
by snowmelt volume than rainfall (Table 5), without a clear relation
with latitude or land cover.
Correlation between antecedent factors and spring flow peak
and
timing
Correlations between Qmax and the six antecedent factors
for the 12 basins are displayed on a correlogram (Fig. 5). Snowmelt
intensity (Meltint) is positively correlated with peak
flow in all basins (r = 0.23 to 0.64); the correlation is significant
(p < 0.05) in most basins, except for Ouelle, Batiscan
and Bécancour. Peak SWE (SWEmax) is also positively
correlated with Qmax in all basins; correlations are
significant in all basins but Ouelle, Bras du Nord, Beaurivage and
Bécancour. Thus, years with higher snow accumulation and faster melting
rate (intensity) generally tended to result in higher peak flow. The
pre-flood accumulated snowmelt (Meltsum) is positively
correlated with Qmax in all basins (r = 0.07 to 0.50)
but only significant in five of them. Pre-flood accumulated rainfall and
its mean intensity do not show any significant univariate correlation
with Qmax, except for the two basins Bras du Nord and
Godbout, where Qmax is positively correlated with
rainfall intensity (Rainint) and in Famine where
Qmax is positively correlated with
Rainsum. Soil moisture (Smean) is
significantly and positively correlated with Qmax in
only four basins (Godbout, Batiscan, Famine and Matawin) and negatively
correlated in Bécancour.
The correlation coefficient for Meltint is stronger than
for SWEmax in six basins, while SWEmaxis a better predictor in only two basins, Matawin and Batiscan.
Overall,
the correlation analysis shows that the pre-flood melt rate
(Meltint) is the best overall univariate predictor of
the spring peakflow, followed by the maximum SWE
(SWEmax) and accumulated snowmelt
(Meltsum), which is logical given the strongly nival
character of the hydrological regime of rivers in Quebec. On the other
hand, the correlation coefficients are overall only moderate, suggesting
that a combination of several factors would be required to better
explain the variability in spring flow magnitude.
For the peakflow timing, the pre-flood accumulated rainfall
(Rainsum) significantly controls QmaxTin all basins, except for the three basins Famine, York, and Godbout
(Fig. 6).
This means that spring flood peaks occur later during years with high
rainfall volumes during the pre-flood period. A larger amount of
accumulated snowmelt (Meltsum) and a slower melt rate
(Meltint) also seem to favor a later occurrence of flood
peaks, however with varying levels of statistical significance
(Fig. 6). The correlation with soil moisture is not spatially coherent,
being significantly anti-correlated with flow timing in two basins
(Famine, and Beaurivage) and positively correlated in Ouelle.
Multivariate prediction of spring flow peak and
timing
The stepwise multivariate regression models explain 40 to 74% of the
variation in Qmax in all basins except in Beaurivage and
Batiscan, where the models only explain 20-28% of the variation (Table
6). The snowmelt intensity (Meltint) was the predictor
most often retained in the regression models (7/12 basins). It has a
positive effect on Qmax, i.e., more rapid melting led to
higher peak flow, in all basins but York, where the effect of
Meltint was negative. Meltin was notably
the sole significant predictor of Qmax in two basins,
Acadie and Beaurivage, where it explained 60% of the variability in
Qmax for the former but only 20% for the later. Either
SWEmax or Meltsum was retained in 7
basins, but never together, as these two variables are partly collinear
(r = 0.45-0.69), i.e., thicker snowpacks led to larger pre-flood
snowmelt volumes, with a positive effect on Qmax in both
instances, i.e., leading to higher flood peaks.
Among the rainfall-related variables, the rainfall intensity
(Rainint) was the most frequent predictor (5/12 basins)
and the one with the largest overall effect among all predictors. The
accumulated rainfall amount (Rainsum) contributed to
Qmax variability in four basins, with an overall lesser
effect than Rainint. Pre-flood soil moisture was
significant in only three basins, with a small and positive effect in
Matawin and a counter-intuitive, negative effect on
Qmax, (Bécancour and York), i.e., drier pre-flood soils
leading to higher peak flows in these two basins.
In order to try improving the prediction performance in basins where the
initial predictor set led to a low R2 (e.g., Nicolet,
Beaurivage, Ouelle, Batiscan), the maximum rainfall
(Rainintmax) and snowmelt (Meltintmax)
intensity during the pre-flood period were added as potential
predictors, to see if these or any of the other basins were sensitive to
the most extreme pre-flood rainfall and snowmelt events.
Meltintmax significantly and positively contributed to
explain Qmax in two basins only, Nicolet and Ouelle,
while Rainintmax also had a large positive effect in
Ouelle only. The Beaurivage and Batiscan models were not improved,
remaining with low R2 values of 0.20 and 0.28,
respectively.
Interannual variations in peakflow timing (QmaxT) were
comparatively well explained by a different combination of factors
between basins, with adjusted R2 varying between 0.27
and 0.65. (Table 7). Accumulated rainfall and its mean intensity explain
most of the variation in all the basins, however with opposites effects.
In southern basins, larger rainfall volumes led to earlier flood peaks,
while the opposite occurred in more northern basins. Inversely, more
intense rainfall events tended to delay flood peaks in southern basins
while the opposite occurred in northern basins. Accumulated snowmelt
(Meltsum) had a positive effect in five basins across
the latitudinal gradient, i.e., more snowmelt delayed flood peaks,
except in Ouelle where the effect was opposite. The snowmelt intensity
Meltint had a noticeable negative influence on flood
timing in southern basins (5/12 basins), i.e., slower melt rates led to
delayed flood peaks.