Rain-on-snow (ROS) events in mountainous catchments can cause enhanced snowmelt, leading to an increased risk of destructive winter floods. However, due to differences in topography and forest cover, the generation of snowpack outflow volumes and their contribution to streamflow are spatially and temporally variable during ROS events. In order to adequately predict such flood events with hydrological models, an enhanced process understanding of the contribution of rainwater and snowmelt to stream water is needed.
In this study, we monitored and sampled snowpack outflow with fully automated snowmelt lysimeter systems installed at three different elevations in a pre-Alpine catchment in central Switzerland. We measured snowpack outflow volumes during the winters of 2017 and 2018, as well as snowpack outflow isotopic compositions in winter 2017. Snowpack outflow volumes were highly variable in time and space, reflecting differences in snow accumulation and melt. In winter 2017, around 815 mm of snowpack outflow occurred at our reference site (grassland 1220 m a.s.l. – metres above sea level), whereas snowpack outflow was 16 % less at the nearby forest site (1185 m a.s.l.), and 62 % greater at another grassland site located 200 m higher (1420 m a.s.l.). A detailed analysis of 10 ROS events showed that the differences in snowpack outflow volumes could be explained mainly by rainfall volumes and initial snow depths.
The isotope signals of snowpack outflow were more damped than those of
incoming rainwater at all three sites, with the most damped signal at the
highest elevation site because its snowpack was the thickest and the
residence times of liquid water in its snowpack were the longest, thus
enhancing isotopic mixing in the snowpack. The contribution of snowpack
outflow to streamflow, estimated with an isotope-based two-component
end-member mixing model, differed substantially among the three lysimeter
sites (i.e. between
Over the past 50 years, rain-on-snow (ROS) events have become more frequent
in snow-dominated catchments, because rising global mean air temperatures
have led to greater fractions of winter precipitation falling as rain
instead of snow (Barnett et al., 2005; Beniston and Stoffel, 2016; Hartmann et al., 2013; Stewart,
2009). In Switzerland, the mean air temperature is predicted to increase by up
to 1.6
Rain on snow can either be retained in the snowpack or it can enhance snow melt; therefore, the snowpack can either reduce or amplify the volume of water reaching the ground surface compared to snow-free conditions (Kattelmann, 1987; Lee et al., 2010a). In the past, some ROS events that caused enhanced snowmelt have led to severe floods (e.g. Garvelmann et al., 2015; Kroczynski, 2004; MacDonald and Hoffman, 1995; Marks et al., 1998; Sui and Koehler, 2001; Wever et al., 2014). Although catchment models have been applied in the past to predict flood responses during ROS events, these model simulations can be highly uncertain (McCabe et al., 2007; Rössler et al., 2014) because snowpack outflow (snowmelt or a mixture of rainwater and snowmelt) is not generated homogeneously at the catchment scale (Berris and Harr, 1987; Würzer et al., 2016). Peak flows caused by ROS events result from a complex interplay of processes that mainly depend on the initial snowpack properties, rainfall characteristics and energy fluxes (Colbeck, 1977; Garvelmann et al., 2014; Würzer et al., 2016), as well as antecedent catchment wetness.
Snowpack properties such as depth, density and snow water equivalent (SWE) can vary spatially and temporally across the catchment landscape. Additionally, wind drift, landscape topography (i.e. slope, elevation and aspect) and vegetation cover (i.e. forest or grassland) affect the snowpack properties (Marks et al., 1998; Molotch et al., 2011; Stähli et al., 2000). Higher elevations are generally associated with greater snow depths due to higher precipitation rates and lower air temperatures (Beniston et al., 2003; Stewart, 2009). Compared with open grassland, forested landscapes tend to have shallower snow depths due to enhanced canopy interception of snowfall (Berris and Harr, 1987; López-Moreno and Stähli, 2007; Stähli and Gustafsson, 2006). Furthermore, water flow paths within the snowpack can be highly variable, so that calculating or measuring the snowpack outflow can be challenging (Eiriksson et al., 2013; Kattelmann, 2000; Rücker et al., 2019; Unnikrishna et al., 2002; Webb et al., 2018; Yamaguchi et al., 2018).
A detailed understanding of snowpack outflow generation is needed at both
the plot scale and the catchment scale to make runoff predictions during ROS
events more reliable (DeWalle and Rango, 2008;
Marks et al., 1998; Šanda et al., 2014). To track the heterogeneous
contribution of snowpack outflow to streamflow during ROS events,
environmental tracers can be used. Stable water isotopes (
In some studies, the isotopic composition of bulk snow or of individual snow layers has been used as a proxy for snowmelt isotopic composition (Cooper et al., 1993; Dinçer et al., 1970; Huth et al., 2004; Maulé et al., 1994; Sueker et al., 2000). The isotopic composition of bulk snow is known to be variable in time and space, depending on catchment characteristics such as latitude, exposure and elevation gradients (Dietermann and Weiler, 2013), as well as the structure of the forest canopy (Koeniger et al., 2008). Snowfall intercepted by the forest canopy is subject to sublimation, which is the main cause of the isotopic enrichment of winter throughfall (Claassen and Downey, 1995; Koeniger et al., 2008; Stichler, 1987). Although the spatial variations in the isotopic compositions of bulk snow and snowmelt are likely to be similar (Dietermann and Weiler, 2013), estimated meltwater contributions to streamflow can be significantly different when using bulk snow instead of snowmelt as an end-member in IHS (Moore, 1989). Numerous studies have found that IHS in snow-dominated catchments is less uncertain when snowmelt is collected by grab sampling (Obradovic and Sklash, 1986; Penna et al., 2017), with melt pans (Bales and Williams, 1993; Taylor et al., 2001) or with snowmelt lysimeters (Beaulieu et al., 2012; Buttle, 1994; Hooper and Shoemaker, 1986; Laudon et al., 2002; Schmieder et al., 2016; Shanley et al., 1995b; Unnikrishna et al., 2002; Wels et al., 1991). Some snowmelt lysimeter systems facilitate water sampling at regular time intervals (i.e. daily or sub-daily), which is recommended because the isotopic composition of snowmelt can be highly variable over time. This variability is caused by isotopic fractionation in the snowpack during phase changes (i.e. freezing/melting, sublimation/recrystallization/condensation; Judy et al., 1970; Lee et al., 2010a; Schmieder et al., 2016; Sokratov and Golubev, 2009; Stichler et al., 2001; Taylor et al., 2001, 2002a; Unnikrishna et al., 2002) and by ROS events when isotopically distinct rainwater percolates and mixes with the snowpack (Berman et al., 2009; Herrmann et al., 1981; Juras et al., 2016; Shanley et al., 1995a). Furthermore, isotopic exchange and redistribution in the snowpack can cause the snowpack outflow to be isotopically different from incoming rainfall (Judy et al., 1970; Lee et al., 2010a, b; Taylor et al., 2001).
To the best of our knowledge, only two scientific studies have estimated the contribution of snowpack outflow to streamflow during ROS events by IHS (Buttle et al., 1995; Maclean et al., 1995). However, these studies used only one or two ROS events that occurred before and during snowmelt. So far, the effects of forest cover and elevation on the generation of snowpack outflow and snowmelt contribution to streamflow have not been investigated during ROS events.
To fill this research gap, we monitored snowpack outflow and its isotopic
composition in the pre-Alpine Alptal catchment in central Switzerland. Three
snowmelt lysimeter sites were located between 1200 and 1400 m a.s.l.,
altitudes at which precipitation frequently shifts between snowfall and
rainfall (Stewart, 2009). One of the lysimeter
systems was installed under forest canopy, and the other two systems
were installed in open grassland. We measured snowpack outflow volumes every
10 min during the winters of 2017 (1 January–7 May 2017) and 2018
(1 November 2017–6 April 2018), as well as
We hypothesise that snowpack outflow generation during ROS events is
spatially and temporally variable at the catchment scale, depending on
elevation and vegetation cover. Specifically, we will address the following
research questions:
What role do rainfall characteristics and initial snowpack properties play
in the variability of snowpack outflow volumes? What is the relative contribution of snowpack outflow to streamflow during
rain-on-snow events? What is the spatial and temporal variability of snowpack outflow
contributions to streamflow? How does the choice of the event-water end-member (rainwater or snowpack
outflow) affect the results of hydrograph separations?
Field work was conducted in the southern part of the 47 km
The Erlenbach catchment (red outline) in the southern part of the Alptal Valley, showing the distribution of vegetation (grassland and forest), as well as the locations of the three snowmelt lysimeter systems (MG: mid-elevation grassland site; MF: mid-elevation forest site; HG: high-elevation grassland site). A meteorological station is located near the MG site. At the Erlenbach catchment outlet, river discharge and precipitation (snowfall and rainfall) rates are measured, along with stable water isotopes in stream water and precipitation.
In the Erlenbach catchment, winter precipitation is dominated by snowfall,
which accounts for up to one-third of the total annual precipitation of
roughly 2300 mm yr
Two snowmelt lysimeter systems, hereafter referred to as the MG site (mid-elevation grassland) and MF site (mid-elevation forest), were located in the Erlenbach catchment at altitudes of 1216 and 1185 m a.s.l., respectively. The MG and MF sites were installed at field sites maintained by the Swiss Federal Institute for Forest, Snow and Landscape Research (WSL) where power supply was available. A third snowmelt lysimeter system was installed at a high-elevation grassland site at 1405 m a.s.l. (HG site), outside of the Erlenbach catchment at a location with power supply. The measurements from the HG site were assumed to be representative of the 1400 m elevation zone of the Erlenbach catchment; however, we acknowledge that the HG site was located on a flat, slightly north-facing plateau, whereas the Erlenbach catchment is characterised by a sequence of flat plateaus and west-facing slopes. The terrain at the meteorological station and the MG site was relatively flat, whereas the MF site was located on slightly sloped terrain facing west.
The forest at the MF site is dominated by
The snowmelt lysimeter system was designed to measure the natural snowpack
outflow and to collect samples for the analysis of stable water isotopes
(see details in Rücker et al., 2019). The
MG lysimeter system site was installed in March 2016, whereas the HG and
MF lysimeter systems were installed in October and December 2016, respectively.
The design of the HG and MF lysimeter systems has been slightly improved to
facilitate easier access to the technical components (e.g. tipping bucket)
during snow-rich periods. Each of the three snowmelt lysimeter systems (MG,
MF and HG) consisted of three individual funnels (42.15 cm diameter, 1400 cm
Before the snowpack outflow reached the collection vessel, its volume was measured via a tipping bucket mechanism at 5 mL volume increments (each corresponding to 0.036 mm of outflow). The tipping bucket mechanism was installed either directly below each individual lysimeter funnel (MG) or between the end of the silicon rubber tube and the collection vessel (MF and HG). The arrangement of the tipping bucket was changed for the MF and HG systems, so that it could easily be replaced or repaired if necessary. As the tipping bucket mechanisms of the snowmelt lysimeter systems were slightly adapted to the local properties at each field site, the average measurement uncertainties of the snowmelt outflow volume measurements were determined from replicate measurements of known water volumes poured into each funnel. The average measurement uncertainties were 15 %, 7.5 % and 10 % at the HG, MG and MF sites, respectively.
In an earlier study that evaluated the performance of the snowmelt lysimeter
design at the MG site, we found that the three individual funnels registered
highly variable snowpack outflow volumes, thus reflecting temporal and
spatial variability of the snowmelt processes at the plot scale
(Rücker et al., 2019). In the analysis
presented here, we calculated the site-averages of snowpack outflow volumes
collected with the three individual funnels at each lysimeter site. To
express the uncertainty of these measurements at the plot scale, we
calculated the standard errors of these snowpack outflow site-averages
considering both the relative measurement uncertainty of the tipping bucket
and the spatial variability of snowpack outflow generation. Measurements at
10 min resolution were aggregated to daily resolution for the time period
from 06:00 to 05:40 UTC
To prevent freezing in the tubes or in the tipping bucket mechanisms, a
heating cable (Pentair, Raychem, BZV self-regulating heating band, Wisag,
Fällanden, Switzerland) was attached to the silicon tubes of the MG and
HG lysimeter systems in December 2017. In addition, a 12-W heating patch
(110 mm
At all three lysimeter sites, snow depths were measured with stakes located
next to the individual lysimeter funnels. A webcam at each site recorded a
picture of the stakes every hour, and we used one image taken between
09:53 and 11:23 UTC
At the MG and HG sites, snow surveys were carried out at weekly intervals. In addition, snow surveys at monthly intervals were undertaken at the MG and MF sites, meaning that the MG site was surveyed twice in the same week roughly once a month. During each survey, snow depth and bulk snow density were measured along a ca. 30 m snow course with a snow tube (diameter 50 mm, length 1.2 m) to determine the snow water equivalent (SWE) (Stähli et al., 2000; Stähli and Gustafsson, 2006). Additionally, soil conditions were characterised (frozen or not frozen) with a 1 cm diameter aluminium stake. At the MG site, the SWE prior to a ROS event was estimated by the product of the actual snow depth recorded by the snow depth sensor and the bulk density derived from the most recent snow survey. At the MF site, only two surveys were carried out during the 2017 winter season.
ROS events during winter 2017 were identified based on the temperature and
rainfall measurements at the meteorological station and the snow properties
the MG site. For winter 2018, measurements from the meteorological station
and the MF site were used because the MG site was not functional. Criteria
for identifying ROS events are often arbitrary and largely depend on the
research purpose and data availability (Mazurkiewicz
et al., 2008; McCabe et al., 2007; Würzer et al., 2016). In our study,
we identified ROS events based on following criteria: rainfall rates greater
than 0.1 mm h
We measured stable water isotopes (
Composite stream water samples were collected using an automatic water
sampler (6712 full-size portable sampler, Teledyne Isco, Lincoln, NE, USA)
that pumped 100 mL of stream water into a dry 1 L HDPE bottle four times a
day (at 05:40, 11:40, 17:40 and 23:40 UTC
During the snow surveys, bulk snow was collected from the entire snow
profile close to the lysimeter sites with a snow tube (diameter 50 mm,
length 1.2 m) and transferred to a HDPE plastic bag (
All water samples that were collected in the field were stored in sealed
bottles and refrigerated at 4
We used two-component isotope-based hydrograph separation (IHS) to estimate
the relative contribution of snowpack outflow to catchment streamflow. The
calculation of the relative contribution of snowpack outflow to streamflow (
As we only measured snowpack outflow volumes and their isotopic compositions at three locations and not across the entire Erlenbach catchment, we cannot reliably estimate the catchment-wide snowpack outflow contribution during individual ROS events. Instead, we performed IHS for each ROS event and individually for each sampling site using the site-specific measurements of snowpack outflow volume and isotopic composition. Thus, we obtained the relative contributions of snowpack outflow to streamflow for three different scenarios during winter 2017, under the assumption that the catchment-wide average snowpack outflow is represented either by the measurements from the mid-elevation grassland (MG), the mid-elevation forest (MF) or the high-elevation grassland (HG) site. By comparing the IHS results of the three scenarios for each ROS event, we seek to quantify the effects of spatial variability in snowpack outflow generation due to vegetation and elevation. As no snowpack outflow could be measured at the HG and MG sites during winter 2018, a three-scenario comparison was not possible for that period.
We also quantified the relative contributions of rainwater (subscript “R”) and
pre-event water to streamflow as
During winter 2017 (1 January–7 May 2017), most of the Erlenbach
catchment was covered with a seasonal snowpack. At the beginning of January 2017,
when snowfall occurred over several consecutive days during cold
conditions (a snow depth of 22 cm was reached within 6 d, and the mean air
temperature was
Daily precipitation volumes (snow and rainfall) measured at the
meteorological station (MG, right axis,
At the MG site, the snow depth was highest on 17 January 2017 (82.2 cm) and the SWE was greatest on 20 February 2017 (168 mm; Fig. 2b). The seasonal snow cover was established on 3 January, became discontinuous on 17 March 2017 and melted completely by 20 March 2017. Two short-term snowpacks established themselves during additional snowfall events in mid and late April 2017. The snow depth measurements at hourly and daily resolution agreed well for most of the study period except for the last 3 weeks of the seasonal snowpack (1–24 March 2017). For this period, the daily snow depth readings from three measurement stakes indicated lower mean snow depths than the readings of the snow depth sensor (hourly data). These measurement differences can be explained by small-scale spatial heterogeneities of the seasonal snowpack caused by wind drift or enhanced melt around the measurement stakes.
Compared with the MG site, the maximum snow depth at the 200 m higher HG site was reached about 7 weeks later and was 55 cm greater (137 cm; Fig. 2a). The maximum SWE (303.4 mm) occurred about 9 d after the peak in snow depth, and was almost twice the maximum SWE at the MG site. According to Stähli et al. (2000) and Stähli and Gustafsson (2006), snow depths, and thus SWE, are generally larger at higher elevations in the Alp catchment due to lower temperatures and thus a greater tendency for winter precipitation to fall as snow. Due to the greater snow depth at the HG site, the seasonal snowpack lasted around 21 d longer than at the MG site. Similar to the seasonal snowpack, the two short-term snowpacks in mid and late April 2017 reached greater snow depths and SWEs compared with the MG site.
At the forested MF site, the snowpack was generally much thinner compared with the nearby grassland MG site, and thus melted out several times during our study period (Fig. 2c). The maximum snowpack depth at the MF site was around 30 cm lower than at the nearby MG site. Based on monthly surveys, the largest SWE (72 mm) occurred on 25 January 2017, roughly 1 month earlier than the maximum at the MG site. Because of the generally shallower snow depths at the MF site, its seasonal snowpack became discontinuous on 21 February 2017, which was 24 d earlier than at the MG site. Four snowfall events (24 February, 6 March, 18 April and 27 April 2017) built up shallow snowpacks at the MF site that lasted only several days (Fig. 2c). An earlier study in the Alp catchment observed that roughly twice as much snow accumulated at grassland sites than at nearby forested sites (Stähli et al., 2000). Generally, snow accumulation under forest is significantly lower due to interception and canopy effects on radiation (i.e. lower short-wave and higher long-wave radiation (Berris and Harr, 1987; Bründl, 1997; Gustafson et al., 2010; López-Moreno and Stähli, 2007; Molotch et al., 2011; Montesi et al., 2004).
Total volumes of snowpack outflow, cumulated over the entire study period,
were largest at the HG site (
Weekly snow surveys at the MG site showed that the shallow soil was frozen
between 28 December 2016 and 12 March 2017, likely because air temperatures
were mostly below 0
The 2017 study period was characterised by frequent ROS events which altered the snowpack properties at the three snowmelt lysimeter sites. Six ROS events are discussed in detail below to compare the snowmelt processes at the HG, MG and MF sites. Four more ROS events occurred during winter 2018, but only the MF site provided snowpack outflow volume measurements during that winter, so a site-to-site comparison was not possible.
Figure 2 and Table 1 provide
overviews of six ROS events during winter 2017 (events 1–6) and four
events during winter 2018 (only MF: events 7–10). During ROS event 1,
the tipping bucket rain gauge at the meteorological station stopped
working after 13 January 2017 03:40 UTC
Start and end times (UTC
Figure 3a compares the cumulative volumes of rainfall and snowpack outflow of the 10 ROS events during winter 2017 (MG, MF and HG sites) and 2018 (MF site only). The snowpack outflow volumes measured at the three snowmelt lysimeter sites were often associated with large uncertainties (Fig. 3a, b), mainly because snowmelt at the plot scale can be very heterogeneous (Kattelmann, 2000; Rücker et al., 2019; Unnikrishna et al., 2002). As each sampling site consisted of three individual lysimeters, we were able to estimate (at least approximately) this spatial variability of snowpack outflow.
A comparison of 10 rain-on-snow (ROS) events during winter 2017 (HG,
MG and MF site) and 2018 (events 7–10; MF site only) indicates large
spatial and temporal variability of snowpack outflow generation in response to
incoming rainfall.
At the MG site, the snowpack response to ROS events was highly variable (Fig. 3b), i.e. snowpack outflow was less than incoming rainfall (events 1 and 5), similar to incoming rainfall (events 2, 3 and 4) or more than incoming rainfall (event 6). For events 1 and 5 the differences between rainfall volumes and snowpack outflow were 19 and 42 mm, respectively, which were statistically significant, i.e. larger than 2 times their pooled standard errors.
At the HG site, snowpack outflow volumes were similar to those measured at the 200 m lower MG site (within their pooled standard errors), except for events 4 and 6. During event 4, no snowpack outflow was generated at the HG site, which was probably because the local air temperature was lower and the snowpack was deeper (95 cm) and thus retained more rainwater than the snowpack at the MG site (Fig. 2a). Similarly, less snowpack outflow was recorded at the HG site than at the MG site during ROS event 6, probably because the deeper HG snowpack was not yet saturated. For event 2, however, the measurement differences between the three individual lysimeters at the MG site were particularly large, likely due to lateral flow in the snowpack (Eiriksson et al., 2013; Webb et al., 2018).
The MF and MG sites were located close to each other, so differences in snowpack outflow can be mostly attributed to the effects of forest cover. For ROS events 1 and 5, snowpack outflow at the forested MF site was larger compared with the grassland MG site, whereas it was smaller for event 6. For the remaining events 2–4, the differences between the measured snowpack outflow volumes at the MF and MG sites were only 19.3, 24.1 and 5.4 mm, and were not statistically significant. Note that for events 3 and 4 the MF site had snowpacks of only 5 and 8 cm, respectively (although these were still identified as ROS events because snowpacks at the reference site, MG, were 29 and 43 cm, respectively). Larger snowpack outflow volumes (events 1 and 5) at the MF site can be explained by the shallower snow depths below the forest canopy (Fig. 2a). Hence, the shallower snowpack saturated more rapidly during ROS events and additional meltwater was released, so that more snowpack outflow was generated compared with the MG site where the snowpack was deeper (Berg et al., 1991; Berris and Harr, 1987; Wever et al., 2014). During event 6, the MF site was already snow-free so the lysimeter funnels captured only under-canopy throughfall, which was less than the rainfall volumes measured near the MG site due to interception losses (DeWalle and Rango, 2008; Saxena, 1986). For ROS events 7–9, the snowpack outflow volumes of the MF site were larger than the incoming rainfall volumes, indicating enhanced melt. The highest snowpack outflow volumes during winter 2018 were registered during event 9, which followed 1 d after ROS event 8. The snowpack at the beginning of event 9 was thinner (event 8: 23 cm; event 9: 10 cm) and probably more saturated, with a higher snow bulk density compared with event 8.
The detailed analysis of the six ROS events at the three lysimeter sites during winter 2017 shows that incoming rainfall was attenuated differently in the snowpacks (both among sites and among events), illustrating the challenge of adequately estimating snowpack outflow volumes during ROS events at the plot and catchment scale. Previous studies used rainfall characteristics and snowpack properties to predict the effects of ROS events on catchment outflow (DeWalle and Rango, 2008; Kattelmann, 1997; Würzer et al., 2016). Thus, in the following section, we analysed the processes and properties that control the outflow response of the snowpack.
Figure 4a–c compares the snowpack water budgets of
the ROS events to the initial snow properties (i.e. bulk snow density, SWE
and snow depth) and rainfall characteristics during the event period
(maximum cumulative 4 h rainfall, maximum cumulative 8 h rainfall, event
duration and mean air temperature) to better understand their effects on
snowpack outflow generation. The snowpack water budget was calculated as the
volumetric difference between snowpack outflow and incoming rainfall; therefore, positive values of the snowpack water budget indicate enhanced snowmelt,
whereas negative values indicate retention of incoming rainfall in the
snowpack (Table 1). Note that the MF site was
already snow-free prior to ROS event 6; thus, we excluded this data
point from the following analysis. At the MG site, ROS event 6 had the
most positive snowpack water budget (
Correlations between snowpack water budgets (snowpack outflow minus rainfall volume) and initial snow conditions and rainfall characteristics (measured at the MG site) of the rain-on-snow events show the strongest
relationship with initial snow depth. Positive values of the snowpack water
budget indicate enhanced snowmelt, whereas negative values indicate retention
of incoming rainfall in the snowpack. Error bars indicate the uncertainty of
the snowpack outflow, i.e. the combined effects of measurement uncertainty and
spatial variability. The different scatterplots compare the snowpack water
budgets at the MG site during winter 2017 with
Figure 4 shows that most of the relationships between
the snowpack water budgets and initial snow properties or rainfall
characteristics are highly scattered, indicating a large variability in
snowpack outflow generation due to lateral flow and preferential flow
pathways in the snowpack. We estimated a positive correlation between the
snowpack water budget and initial snow depth (Fig. 4c;
No consistent relationships emerged between the snowpack water budget and the
initial SWE (Fig. 4c) or rainfall duration
(Fig. 4e), suggesting that these are poor
predictors for snowpack responses to ROS events. This was further confirmed
by a multiple linear regression analysis (JMP 14 software, 100 SAS Campus
Drive, Cary, NC 27513, USA) that estimated the effects of initial snowpack
properties and rainfall conditions on snowpack outflow volumes for the
15 ROS events measured at the MG and MF sites. The best model fit (
A better understanding of snowmelt processes during ROS events could be
obtained when individual events or event pairs were analysed in greater
detail. Snowmelt at the MG site was most enhanced during event 6, when
66.9 mm of rainfall resulted in
Hourly measurements of precipitation (snowfall and rainfall) and air
temperature
Event characteristics and snowpack water budgets of the 10 rain-on-snow (ROS) events. Snowpack water budgets were calculated for the MG and the MF site by subtracting the snowpack outflow volume from the rainfall volume. Standard errors (SE) of the snowpack water budget were estimated from the measurement uncertainty and the spatial variability of snowpack outflow measurements at the sampling site. Manual snow surveys provided initial snow bulk density and initial snow water equivalent (SWE). Snow surveys were generally performed once or twice a week at the MG site but only monthly at the MF site during the winter of 2017, providing insufficient information at that site (–). Weekly snow surveys at the MF site in winter 2018 did not properly represent the snow properties at the MF site (–) because the snow depth under the forest canopy was much shallower and highly variable over time compared with the MG site; thus, the measured snow bulk densities at the MF site were likely not constant over several consecutive days.
The 90.2 mm ROS event 5 resulted in the most negative snowpack water
budget (
A similar analysis could be carried out for ROS events 1 and 3,
during which 21.6 and 20.0 mm of rain fell, respectively. During event 1,
snowpack outflow volumes at the MG site were
The analysis of the 10 ROS events measured at the MG and MF sites illustrates that the generation of snowpack outflow did not entirely depend on the incoming rainfall volume, but also on the initial snowpack conditions that controlled the retention of rainwater and melt processes. This was further illustrated by the hourly measurements of snowpack outflow that indicated highly variable responses and lag times (e.g. the time between the beginning of the ROS event and the first response of the snowpack outflow; the first response is defined as an increase of snowpack outflow by at least 0.05 mm relative to the previous measurement) across the lysimeter sites (Fig. 5).
For three ROS events (1, 2 and 5), the lag times of the MF site
were the lowest compared with the HG and MG sites, possibly due to the
generally shallower snowpack at the forested site. The snowpacks at the
MG and HG sites were deeper, and for most of the events they likely had a
larger buffer capacity for incoming rainwater than the shallower snowpack at
the MF site. The longest time lags were observed at the MG site during
event 2 (e.g. 26 h) when the ROS event occurred with relatively low
rainfall volumes (21.6 mm per event) on a fresh snowpack with low density
(Tables 1, 2). ROS event 6 occurred when the snowpacks at the HG and MG sites were already
ripe (e.g. more than 3 mm d
Measurements from the MG, MF and HG sites revealed that snowpack outflow generation was highly variable across space and time and as a result, the contribution of snowpack outflow to river streamflow was very heterogeneous across the catchment landscape. For instance, the streamflow response to ROS event 2 was particularly large, probably because of large snowmelt inputs from higher elevations (HG site; Fig. 5b). Daily pulses of snowmelt from the HG site in late March were also reflected in distinct diurnal variations in stream discharge, suggesting the input of snowmelt mainly from high elevations (Fig. 5b). In contrast, during events 1 and 4, no snowmelt was generated at the HG site, so that the observed discharge peak was likely caused by snowmelt from low and mid elevations (MG and MF sites; Fig. 5c, d). However, the synchrony of responses does not allow for any conclusions to be drawn about the water sources of streamflow (McDonnell and Beven, 2014).
Figure 6 compares the isotopic composition of water
samples collected with the three lysimeter systems during the 2017 winter
period. Because the lysimeter funnels were permanently installed in the
field, they collected snowmelt and rain-on-snow during snow-covered periods,
as well as rainfall during snow-free conditions. Thus, we classified the
samples as either rainwater (no snowpack), rain-on-snow or snowmelt to
better quantify the effects of elevation and vegetation cover on the
isotopic signatures of the different water sources. Because of the more
persistent snow cover at the HG site, rainfall only occurred as rain-on-snow
during the study period, so the HG lysimeter system collected predominantly
snowmelt or a mixture of rain and snowmelt. Additionally, the isotopic
composition of bulk snow samples at the HG and MG sites are shown (no
regular bulk snow sampling was carried out at the MF site). We evaluated the
isotopic differences between the water sources at each site and between the
sites for the same source with an unpaired two-sample
Isotopic composition of the samples collected with the three snowmelt
lysimeter systems (rainwater during snow-free conditions, snowpack outflow
during rain-on-snow events and snowmelt) and bulk snow at
We also evaluated the isotopic differences between the sites for the same
source (unpaired two-sample
Rainwater at the MG site and throughfall at the MF site had similar isotopic
compositions (differences in median
At the grassland HG site, the median isotopic composition of bulk snowpack
and snowpack outflow (rain-on-snow and melt) was heavier than at the lower
grassland (MG) site; the median
Due to frequent melt periods and ROS events, values of
Deuterium (
Our daily isotope measurements of rainwater and snowpack outflow across the
catchment landscape allowed us to study the temporal and spatial isotopic
variation of snowpack outflow during rain-on-snow events and snow melt. Our
main observations are as follows:
The isotopic signal of incoming rainwater can be altered as it percolates
through the snowpack, depending on snow metamorphism and isotopic exchange
(Judy et al., 1970). A significant isotopic depletion or enrichment of
snowpack outflow due to such rain-on-snow events has already been reported
in other studies (Herrmann,
1978; Juras et al., 2016; Shanley et al., 1995a; Unnikrishna et al., 2002).
The isotopic exchange in the snowpack is mainly controlled by the residence
time of liquid water (snowmelt and rain-on-snow) in the snowpack, which, in
turn, is determined by the depth and the density of the snowpack (Taylor et al., 2001, 2002b), the
rainfall magnitude (Herrmann et al., 1981) and the flow rate of percolating liquid
water. As a result, deeper snowpacks generally cause slower rainwater
throughflow, which enhances isotopic redistribution in the snowpack and
isotopic exchange between the liquid water and solid ice (Lee et al., 2010a, b;
Taylor et al., 2001). The isotopic contrast between the snowpack outflow of the last day of event 3
and the following day was 13.4 ‰ for
The different results among the three sampling locations reflect the highly
variable isotopic compositions of the snowpack outflow across the catchment.
For example, during event 6, isotope data from the HG site suggested a
significant snowpack outflow contribution to discharge ( This comparison raises an important point of interpretation. Any
two-component hydrograph separation is based on the fundamental assumption
that there are only two end-members (in our case, one of the snowpack
outflows, and “old water” represented by pre-event streamflow). Thus one
cannot interpret the results above as demonstrating that more snowpack
outflow reached the stream from high-elevation sites like HG than from lower-elevation sites like MG. Instead, what these results show is that Although the number of ROS events in our data set is small, our results are
in line with previous studies showing that the differences between
hydrograph separation results obtained for rainwater and snowpack outflow
can potentially be large and should be considered in snow-dominated
catchments (Buttle et al., 1995). Our analysis assumes
that the end-members of the different scenarios (i.e. snowpack outflow at
the MG, MF or HG site) are representative for the whole Erlenbach catchment.
However, the catchment is characterised by a diverse vegetation cover
(22 % partially forested, 53 % forested and 25 % grassland) and
surface topography (altitude 1000–1500 m a.s.l.), so that the “real”
contribution of snowpack outflow to streamflow is likely to lie between the
estimates derived from the three scenarios. The hydrograph separation
estimates for the three lysimeter sites can only provide a probable range of
snowpack outflow contributions to discharge from different landscapes of the
catchment. As shown here, the estimated contributions of snowpack outflow to
streamflow can vary considerably due to differences in landscape
characteristics, rainfall magnitude and snowmelt processes. Future sampling
strategies should take this spatial and temporal variability in snowpack
outflow into account.
Relative contributions of rainwater or snowpack outflow to daily peak
discharge based on two-component isotope hydrograph separation (IHS) using
Relative contribution of snowpack outflow to streamflow at peak flow
based on isotopic hydrograph separation for the six rain-on-snow events from
winter 2017, including the incoming rainwater (blue, not filled) and the
snowpack outflow of the high-elevation site (red, black-shaded), the mid-elevation
grassland site (yellow) and the mid-elevation forest site (green). For some
events, no data (
In many mountain regions, global warming is predicted to lead to more frequent rain-on-snow (ROS) events, which can enhance snowmelt and increase the risk of destructive winter floods. However, the processes leading to such enhanced melt are spatio-temporally heterogeneous, so that model-based predictions of discharge peaks during ROS events can be highly uncertain.
By using three automated snowmelt lysimeter systems, located along an elevation gradient of 1185 to 1420 m a.s.l. in a partly forested pre-Alpine catchment, we were able to capture the spatial and temporal variability of snowpack outflow generated over the winter season (Figs. 2, 5). A comparison of snowpack properties at a grassland and a nearby forested site showed that canopy interception significantly reduced incoming snowfall; thus, the maximum snow depth under forest cover was around 20 cm shallower than that of open grassland. Measurements from two grassland lysimeter sites located at different elevations (1220 and 1420 m a.s.l.) showed that the snowpack was on average 55 cm deeper and snowmelt occurred 21 d later at the higher site.
To better understand how snowpack outflow is generated during ROS events
across the catchment landscape, we studied 10 ROS events in greater detail
(Fig. 3). The ROS events were defined by rainfall
rates greater than 0.1 mm per hour, a total rainfall volume of at least
20 mm within 12 h, air temperatures above 0
We used daily stable water isotope measurements in snowpack outflow, rainwater and stream water to draw inferences about transport and mixing of rainfall within the snowpack during individual ROS events. Depending on the local rainfall characteristics and the snowpack properties, the isotopic responses in snowpack outflow could be either strongly or weakly damped, indicating large spatio-temporal variations of the snowmelt process (Fig. 7). Consequently, isotope-based two-component hydrograph separation (IHS) for estimating snowpack contributions to streamflow often yielded very different results (Fig. 8), depending on which site-specific snowpack outflow isotopic compositions were used. This range of IHS results provides reasonable estimates of relative snowpack outflow contributions to streamflow during individual ROS events, under the assumption that the three lysimeter sites are representative of the snowmelt processes at the catchment scale. Furthermore, our IHS results vary over a wide range, implying that in steep, partly forested catchments like Erlenbach, estimates of snowpack outflow contributions to streamflow derived from bulk snow samples or outflow samples collected at only one location can be highly uncertain. This is in line with the study by Fischer et al. (2017), which showed strong spatial variability in rainwater isotopic composition in the southern Alptal catchment. Using rainwater isotope data in the IHS analysis suggests that the relative contribution of rainwater to streamflow may often be much smaller than the contribution of snowpack outflow, because snowpack outflow is a mixture of both rainwater and snowmelt. Our analysis suggests that snowpack outflow can contribute substantially to streamflow during ROS events and that these contributions depend strongly on the local snowpack properties and rainfall characteristics.
In order to obtain more realistic estimates of snowpack outflow contributions to streamflow during ROS events, snowpack outflow volumes and their isotopic compositions could be interpolated across the study area using a spatially distributed snowmelt model. Recent snowmelt modelling approaches at the catchment scale do not, however, explicitly simulate snowpack outflow during rain-on-snow events (Ala-aho et al., 2017; Lyon et al., 2010; Smith et al., 2016), or use stable water isotopes to track flow pathways (Kormos et al., 2014; Marks et al., 2001; Rössler et al., 2014; Storck et al., 1998). Thus, our spatio-temporally distributed isotope measurements could be beneficial for testing and improving existing snowmelt models (Zappa et al., 2015).
The data are available upon request to James W. Kirchner.
The supplement related to this article is available online at:
SB developed the technology for the field installations and supported the field work. JF provided the precipitation isotope data. AR analysed the data set. AR prepared the paper with contributions from JF and JK.
The authors declare that they have no conflict of interest.
We thank the staff of the Swiss Federal Institute for Forest, Snow and Landscape Research (WSL), especially Massimiliano Zappa for his input on the paper, Karl Steiner for great support in the field, and Alessandro Schlumpf and Bjørn Studer for isotope analysis. We thank Daniel Meyer, Dominic Schori and Stephan Biber for their help in the field and in the laboratory. We are grateful to Martin Brun and the Oberallmeind Schwyz who authorised the installation of a snowmelt lysimeter system on their land. We also thank the reviewers for their valuable suggestions that have much improved the paper, and the editor for the careful handling of the paper.
This paper was edited by Bettina Schaefli and reviewed by Daniele Penna, Roman Juras, and one anonymous referee.
This project was supported by the Swiss National Science Foundation (SNSF) through the Joint Research Projects (SCOPES) Action (grant no. IZ73Z0_152506).