Water (from streams, springs, and piezometers) was either sampled manually (grab samples) or via automatic samplers placed at the outlet and an upstream location along the stream (HyS1 and HyS2; Fig. 1) for stable isotope analysis (δ2H, δ17O, and δ18O). A three- or four-letter code was adopted for all sampling locations, as shown in the map (piezometers – PZ1, PZ2, and PZ3; springs – GRAS, AUBG, ROCK, BRDG, and ICEC; streams – HyS1 and HyS2; Fig. 1). We used 12 mL amber borosilicate glass vials with polypropylene screw-top caps with PTFE-lined silicone septa for all sample transport and storage. When possible, samples were stored sealed in a refrigerator (∼4 ∘C) until analysis but were, in some cases, stored at ambient temperatures. In all cases, they were kept out of direct light and heat. Automatic sampling was performed with a Teledyne ISCO 6712C compact portable sampler with 24 bottles of 500 mL capacity at HyS1 and a Teledyne ISCO 6712 full-size portable sampler with 24 bottles of 1 L capacity at HyS2 (Lincoln, Nebraska, USA). Automatic samplers were programmed to sample at 6 h intervals over 1 week. The automatic sampler was programmed to fill bottles to half of their capacity, 250 and 500 mL, respectively, to optimize energy usage and to prevent sample loss due to freezing, while still sampling enough water to limit fractionation due to evaporation.

A subsample of water was then taken manually from each bottle, either in the laboratory or in the field. Original installation involved the use of pipettes and tubes inside the autosampler bottles (similar to those described by von Freyberg et al., 2020); however, after some experimentation and due to the Alpine and shaded microclimate of the location, fractionation due to evaporation was deemed minimal and additional components resulted in more contamination and less sampling capacity. When they froze between collection and subsampling, the bottles were closed with a cap and moved to a warmer place until the ice fully melted, and then they could be subsampled.

Samples of rainfall were collected at the Auberge and Chalet meteorological stations (Fig. 1), with a 13 cm diameter plastic funnel connected to an insulated 2.5 L screw-top bag made of 147 µm PET, nylon (NY), and low-density polyethylene (LDPE) plastic (DaklaPack; Perpignan, France), enclosed in a plastic box. The collected water was well mixed, weighed, and subsampled once or twice a week from May to November (outside the snowfall period).

Groundwater was sampled directly from the piezometers (Thornton et al., 2021). Prior to water sampling, the piezometers were emptied using a Geotech peristaltic pump (Geotech Environmental Equipment, Inc.; Denver, Colorado, USA), and the freshly recharged water was sampled with the same pump and stored in 12 mL amber glass vials.

During winter 2017 and winter 2018, snow samples were collected regularly at two locations. Two different sampling methods were used. (i) If distinguishable snow layers were present (visual and textural distinction), then each of them was sampled individually; otherwise, (ii) a single bulk sample of approximately the entire profile was taken.

Snow was sealed in alimentary 700 mL zip bags made of 120 µm biaxially oriented polypropylene (BOPP) and LPDE plastic (DaklaPack; Perpignan, France) after evacuating as much air as possible. The collected snow samples were melted at ambient air temperature (the influence of water vapor from air on the isotopic composition of the sample is discussed in Appendix A). A subsample of well-mixed, melted snow was taken manually in the lab from each bag.

The isotopic composition of the entire snowpack at a given snow pit was obtained, with a weighted average of the values of each sampled layer according to depth as an approximation for the equivalent bulk isotope composition assuming a uniform density.

Stable isotope compositions of water, expressed using the δ notation (δ2H, δ17O, and δ18O), were measured with a Picarro 2140-i wavelength-scanned cavity ring-down spectrometer (Picarro, Inc.; Santa Clara, California, USA), using 2.0 mL glass vials, closed with screw-top caps with Septa silicone rubber and PTFE and filled with 1.8 mL of filtered water. Samples were injected between 6 and 8 times (Penna et al., 2012). The first three injections were discarded to avoid memory effects (Penna et al., 2012). The raw values were then corrected according to a standard curve determined with three internal standards, which are regularly calibrated against the international standards of VSMOW (Vienna Standard Mean Ocean Water) and SLAP (Standard Light Antarctic Precipitation) of the IAEA (International Atomic Energy Agency; Coplen, 1994). Each standard was injected 12 to 15 times, and data from the final six injections were kept. Delta units of isotope compositions (Coplen, 1994) are reported in parts per thousand (per mille), and the strategy used for the analysis is similar to the one described in the work of Schauer et al. (2016). The median analytical errors obtained with this method are 0.4 ‰ for δ2H, 0.01 ‰ for δ17O, and 0.04 ‰ for δ18O.

Based on these measures, we compute the d-excess (deuterium excess; Dansgaard, 1964) and 17O-excess as follows (Barkan and Luz, 2005; Landais et al., 2006): 1d-excess:=δ2H-8⋅δ18O,217O-excess:=106ln⁡δ17O1000+1-λref⋅ln⁡δ18O1000+1, with λref=0.528 (Meijer and Li, 1998; Barkan and Luz, 2005; Landais et al., 2008). From the regression of ln⁡(δ17O/1000+1) against ln⁡(δ18O/1000+1), we obtain a similar slope for our samples (λref=0.528), which confirms the universality of this value. However, memory effects can notably influence the δ17O measurements in cases of larger variations in values within any one sequence measured (Vallet-Coulomb et al., 2021).

In order to gain insights into local evaporative processes, we compute the line-conditioned excess (lc-excess; Landwehr and Coplen, 2006), based on our local meteoric water line (LMWL), as follows: 3lc-excess:=δ2H-a⋅δ18O-b.

We estimated elevation gradients of isotopic ratios, i.e., lapse rates, based on the median of measurements at different locations for precipitation and streamflow. For precipitation, we used the measurements at Auberge station (elevation 1253 m a.s.l.) and Chalet station (elevation 1517 m a.s.l.). For streamflow, we used the measurements of station HyS1 (elevation 1248 m a.s.l.) and HyS2 (elevation 1469 m a.s.l.). For other sampled waters, we did not establish lapse rates due to a varying number of samples and inconsistent sampling dates.

The fluctuations in soil temperature at three different elevations (1240, 1530, and 2640 m a.s.l.) revealed seasonal, daily, and spatial variation (Fig. 3). As winter approached each year, the soil temperature gradually dropped towards but did not reach 0 ∘C. The slightly positive temperature can be explained by ground heat flux. Once snow fell, the insulation it provided was visible as a dampened diel temperature variation that was less correlated with daily air temperature fluctuations than in snow-free periods. Some isolated temperature spikes were observed during winter, probably due to rain-on-snow events (e.g., the spike during winter 2016 is the green line in Fig. 3, which is at the lowest elevation) that were visible in the temperature signal but apparently did not generate any streamflow. Unfortunately, no other observed tracers were available during these periods to confirm this hypothesis.

The negative temperatures measured during the 2016–2017 winter period by two soil temperature probes (at 1530 and 2640 m a.s.l.) can be explained by the cold air temperature and the exceptionally dry winter with a low snow cover. The variation in temperature at the three different elevations in Fig. 3 displays the elevation gradient that alters the start of the snow-free period at the different elevations and the elevation dependency on the timing of the strong warming (of more than 5 ∘C) between March and July. The start of the snow-free period occurred between 4 (2018) and 8 weeks (2017) earlier at 1240 m a.s.l. elevation than at 1530 m a.s.l., whereas between 1530 and 2640 m a.s.l., it was delayed by between 3.5 (2018) and 8 weeks (2016).

Similarly, the soil temperature time series clearly showed the elevation dependency on the arrival of snow. It arrived much earlier (12 weeks) in autumn 2016 at the highest elevation, compared to the two lower elevations. In 2017, the seasonal snow cover onset occurred at a similar time at all elevations, as we see that all diel temperature variation disappears between 22 October and 25 November. A summary of snow-free dates, as extracted from the temperature recordings, is available in the Supplement (Table S3).

Interesting features were observed because soil temperature was examined with respect to the identified streamflow periods. For all three melt periods (M), soil temperature recordings from the highest elevation showed the presence of snow until the start of the recession period (R), which demonstrated the late melt of seasonal snow in some areas of this catchment. The start of the two early melt periods (E) in 2016 and 2017 corresponded to the disappearance of snow at the lowest soil temperature measurement point (1230 m a.s.l.).

Decline in soil temperature started at a similar time across all elevations and occurred close to the start of the streamflow baseflow periods (B), or in other words, the significant decrease in soil temperature only started when the streamflow recession period (R) was already advanced. In the winter of 2016/2017, winter streamflow fluctuations were reflected in soil temperature, whereas the midwinter streamflow rise in January 2018 was not visible in any of the soil temperature recordings. However, this might be due to errors in recording river stage caused, e.g., by accumulated sediment (Michelon, 2022; Chap. 3).

Finally, from the covariation between soil temperature and streamflow, we can deduce that, during M, runoff-generating rainfall events coincide with cold spells, whereas during autumn, runoff-generating rainfall events coincide with warm spells (e.g., October 2016 and 2017).

The average recorded stream temperature at the outlet was 5.0 ∘C, which was slightly higher than 3.1 ∘C, which is the average recorded air temperature at mean elevation 2012 m a.s.l. The fluctuations in water temperature at the catchment outlet (HyS1; Fig. 1) were correlated with the variations in the air temperature (R2=0.87) at the Auberge weather station, and the annual cycle showed no lag between them. The close correspondence is explained by the fact that the instream travel time is long enough for atmospheric heat exchange to exert a strong influence on water temperature (Gallice et al., 2015). The importance of instream atmospheric heat exchange can also explain the high annual and diel temperature amplitudes (Table 1), which corresponded closely to the observed air temperature amplitudes over the year (between 17.5 and 19.5 ∘C at the lower elevations, with a 30 d moving average).

An observation of the patterns in the temperature of springs and groundwater reveals hints of underlying flow-generation processes. Indicators of these processes include the correlation between water and air temperature, diel temperature variations, temperature response to rainfall events, the overall pattern and shape of temperature fluctuations on a seasonal scale, mean values and convergence between different points over the study area, and temperature anomalies and their timing. First, we observe that the correlation between water temperature in springs and in piezometers and air temperature, measured at the Auberge station, varied by location of the groundwater (Table 2); PZ1 correlated the best (R2=0.80) and ICEC the worst (R2=0.56) but still none as well as the surface water (R2=0.87; described above).

In general, diel temperature variations were rare and can probably be explained by poor measurement when water volume was low. The temperature in some springs reacted to precipitation, while that in others did not. The GRAS spring had a permanent but small outflow of only a few liters per minute (personal observation). Since the temperature was recorded directly in the outflowing water, the sensor might have been heated up by atmospheric heat exchange when outflow was low. This most probably explains some strong sub-diel temperature fluctuations in the GRAS (and ROCK) springs (Fig. 3). Despite these diel fluctuations, the GRAS temperature signal did not react to summer rainfall events (visible as peaks on the streamflow), whereas ROCK reacted.

The shape of the curve of temperature fluctuations gives us clues to the flow that fed the springs. For example, that of the BRDG spring differed from the sinusoidal shape of the GRAS and ROCK springs, which match the air temperature variations more closely. The BRDG spring signal showed a constant temperature during winter, with an increase during E and M, when it rose from 4.3 to 5.4 ∘C over 3 weeks at the beginning of M2. The temperature stopped rising around the time when the soil temperature at mid-elevation indicated snow disappearance (blue bar in Fig. 3) and then receded to the winter base temperature. By combining the observation of a strong reaction during melt at low elevations with the return to a base temperature during winter, we can deduce that BRDG spring is fed by snowmelt from low elevations (from the right bank riparian area where it is located) during spring and by groundwater during the rest of the year.

All spring temperatures converged to around 4.3 ∘C at the end of B2 (the only winter period measured in all springs), which corresponds to the almost constant temperature of ICEC spring (annual amplitude of 0.4 ∘C; Table 1). This may indicate that, at this point in the year, all springs are fed by a common or similar groundwater source with little influence of intermediary subsurface or surface flows.

The two piezometers that access the groundwater (PZ1 and PZ3) are in the alluvial floodplain across which the stream meanders into an alluvial plane. During intense rainfall events, PZ1 shows strong positive temperature excursions, which even exceeded the temperature of the main channel in summer; however, its winter anomalies were less extreme. The annual cycle of the PZ1 temperature reached its maximum temperature of 7.9 ∘C with a delay of 74 d (2.5 months) after the air maximum and exceeded the maximum recorded in the springs by 1.5 ∘C. The strong delay of the annual cycle together with the warm temperatures and relatively small-amplitude dampening compared to ROCK and GRAS springs suggests that it is influenced by a large storage volume which induces the delay and is closely connected to heat input from the surface.

PZ3 shows the same annual temperature amplitude as PZ1 but has an even longer delay (21 d with respect to PZ1) and has a negative offset of 1.5 ∘C of its maxima (6.4 ∘C for PZ3) compared to PZ1, possibly related to the higher elevation and more northern aspect of its source area (PZ3 is located 30 m higher, in the more north-facing part of the catchment). PZ3 has, however, an average temperature of 4.8 ∘C closer to the one of the springs.

A distinctive feature of PZ3 is its temperature decrease during M2, which is in phase (but in opposite direction) with the streamflow increase. This suggests a direct, relatively important cold input during snowmelt, resulting from a high hydrologic connectivity of PZ3 to snowmelt water (either directly or via exfiltration from the stream) and a low storage volume during this time of the year.

Dampening depths estimated with the simple temperature model (Appendix B) from groundwater temperature patterns ranged between 1.2 and 5.4 m (Table 2), and the lag in temperature for those same points compared with streamflow ranged from 41 to 133 d. We attribute these values to the delay resulting from heat conduction (depending on the soil's thermal diffusivity D) and advection with water flow. The one exception is BRDG, for which lag estimation fails, perhaps indicating that this spring is not truly groundwater fed. These lag values are furthermore coherent with the dampening, as stronger lags correspond to stronger dampening and are associated with deeper depths. They should, however, be interpreted with care as (i) the presence of an insulating snowpack on the hillslopes prevents heat advection during winter in a similar way that soil would, thereby further contributing to temperature lags and amplitude dampening in the subsurface, and (ii) the model is only based on heat conduction and does not account for advection that could be locally important, particularly during snowmelt inputs.

The BRDG spring highlights these limitations, as the temperature variation over the year (0.9 ∘C) happens over few weeks during the melt periods (M1 and M2). This variation shows a strong reactivity to the snowmelt input, but the resulting estimation of flow path depth (0.2 m) is obviously erroneous. At this time, the maximum air temperature is not reached yet (during R2 and R3), and the expected heat signal transferred from air by conduction later in the year is not visible.

The median δ18O value of all streamflow samples was -12.7 ‰. An isotopic lapse rate of 0.84 ‰ per 100 m for δ2H and 0.13 ‰ per 100 m for δ18O was computed from our precipitation water samples between the Auberge and Chalet stations (with a higher median value at the lower Auberge weather station), which is approximately half the isotopic lapse rate of precipitation observed in Switzerland (e.g., Beria et al., 2018). However, this lapse rate was not seen in stream water isotopes (Fig. 4a–c) measured at HyS1 (1248 m a.s.l.) and HyS2 (1478 m a.s.l.). The distribution of elevations feeding HyS1 (mean elevation 2165 m a.s.l.) was most probably not sufficiently different from the distribution feeding HyS2 (mean elevation 2196 m a.s.l.) to result in a significant shift in the isotopic values at the two streamflow sampling locations, despite the isotopic lapse rate in precipitation. The median isotopic values for the sampled springs showed an elevation gradient (Fig. 4a–c; showing the spring values in order of elevation from Auberge (AUBG) to ICEC), except for the GRAS spring, which showed a significantly higher median isotopic value. This suggests that GRAS spring might receive water only from a small, low-elevation subcatchment and not from the high rock walls located next to it.

The fluctuations in the isotopic composition of the six monitored springs between July 2016 and September 2018 is discussed qualitatively based on the streamflow periods (see Fig. 5). Because the variations in values of δ2H, δ17O, and δ18O correlate well, we will only discuss δ18O. Corresponding figures of the other isotopic values can be found in the Supplement.

Despite some variability, the AUBG spring δ18O values remained relatively constant (δ18O between -12.8 ‰ and -12.2 ‰) during the 2016 streamflow recession R1 and then slowly decreased throughout the 2016/2017 baseflow period B1. Meanwhile, the BRDG spring started with lower isotopic values (-13.3 ‰) but became enriched in heavier isotopes through R1 and B1 to finally have a similar composition during the 2017 early melt period (E1), compared to the AUBG (and ROCK) springs, and also with a subsequent decreasing trend in the heavy isotopes.

The 2017 minimum isotopic values of the AUBG, ROCK, and BRDG springs were reached around the time of 2017 maximum streamflow and then diverged during the 2017 recession period (R2), increasing at a different rate. The δ values of AUBG and ROCK springs increased quickly (+1.0 ‰ in 3 months), while the BRDG spring values only showed a slow increase through the winter period (B2).

The beginning of the 2018 melt period was exceptionally fast, without an early melt period. The springs sampling started 3 weeks after the beginning of the early melt period (E), with a significant part of the snowpack having melted already. During M1, the isotopic composition of the AUBG and BRDG springs over this period showed a constant decrease in the heavy isotopes until the 2018 streamflow maximum. At the inverse of R1, the BRDG composition remained constant during the recession, while the AUBG spring increased quickly in the δ values.

The pattern repeated during the 2018 melt period (M2), with a decrease in δ values, which then diverged at different rates. Again, the BRDG spring δ values increased slowly, while for the AUBG and ROCK springs the increase was faster.

The ICEC spring, located on the western slopes (Fig. 1), followed the same isotopic pattern as the AUBG spring. Although, because of its lower sampling rate, points were missing at the critical moments during the melting periods, which does not allow us to discuss the differences in timing. The ICEC spring showed higher isotopic values compared to BRDG, even though it is located at a higher elevation, which can be explained by the higher maximum elevation of the mountain ridge upstream of BRDG compared to ICEC (see Fig. 1), which most certainly leads to a higher snowfall contribution to BRDG as snowmelt isotopes are more depleted in heavier isotopes than rainfall (Fig. 7).

As discussed earlier, the GRAS spring behaves differently from other springs, with higher δ values than all the others in 2017. EC and temperature measurements indicate that this spring has relatively shallow flow paths, and its δ values also suggested a larger proportion of rainfall-derived water (which has a higher average δ values than snowmelt; Fig. 7).

Because of the close overall correlation between δ18O and δ2H (Fig. S10) and δ18O and δ17O values, we will highlight δ18O in our discussion and in Figs. 4–7. The temporal evolution of the isotopic values in the streamflow showed high δ18O (and δ2H) during winter baseflow, close to the median value of all sampled subsurface waterbodies, and a significant decrease in the isotopic composition during the melt periods. Streamflow is thus largely fed by recent (isotopically light) snowmelt during the melt period; the decrease in the δ18O (or δ2H) is proportional to the amount of snowmelt, with a larger decrease in 2017 compared to 2018. The δ18O (and δ2H) value did not decrease during E.

Rainfall had a median d-excess value of 11.3 ‰ and snowpack of 15.5 ‰. However, as we did not systematically sample fresh snowfall and snowpack separately, it was hard to estimate the extent of snow sublimation in Vallon de Nant.

Surface and subsurface water samples showed median d-excess values close to that of rainfall and considerably lower than the median value for snow. The apparent surface and subsurface water samples bias towards the d-excess value of rain can be explained by secondary evaporation (e.g., from soil); the soil water that remains (and that ultimately recharges groundwater and the streams) thus has a lower d-excess value than either rainfall or meltwater. Thus, the d-excess could be interpreted as the evaporative exposure of water during the time since precipitation, but it is rather difficult to interpret in terms of local-scale process information. Even if the difference in median value between rainfall and snowfall is significant enough for a separation, because of the potential for transformation along the flow path, it will not be as valuable as the δ18O (and δ2H) values directly. For ice melt, d-excess values are too close to those of rainfall for providing further insight into its importance in streamflow.

The LMWL was calculated using linear regression between δ18O and δ2H of 75 rainfall samples with a slope of 7.38 and an intercept of 6.15 (see Fig. S10). The median analytical error was 0.4 ‰ for d-excess and lc-excess, and 8 per meg for 17O-excess. Compared with d-excess, lc-excess for rainfall samples indicated the spread of precipitation isotopes around the LMWL (Fig. 4f). Our LMWL deviated from the global meteoric water line (GMWL). The median lc-excess of snowpack samples was close to rainfall (0 ‰), suggesting no significant snow sublimation in Vallon de Nant. The spring and stream samples showed negative median lc-excess values, indicating that recharged water has undergone some evaporation, which may vary over space and time. Out of all the springs, ICEC spring samples seemed to be less affected by evaporation (as shown by a higher lc-excess value), suggesting that rainfall over the area upstream of this spring directly infiltrated into the ground. This is further supported by the presence of sparse vegetation in this part of the catchment.

We observed that snow has a median value of 17O-excess of 91.3 per meg and is significantly higher than that of rainfall (49.2 per meg). The difference between rainfall, snowpack, and glacier observed for δ2H, δ17O, and δ18O is also visible with 17O-excess, as opposed to d-excess. 17O-excess could potentially be useful to distinguish between rainfall, snowpack, and ice melt, but secondary evaporative processes complicate a direct interpretation.

Given that the local and global reference lines for 17O are very similar (see Sect. 2.3), it is tempting to interpret the spatial differences in 17O-excess values; the median values of all sampled water show a coherent picture, with subsurface and stream water having intermediate values between rainfall and snow samples and thus being a mix thereof. As for d-excess, we can, however, not draw any direct conclusions on mixing ratios, since rainfall and snowfall undergo further evaporative processes during recharge. Furthermore, the temporal dynamic of 17O-excess in springs does not show additional information compared to d-excess. Since the use of δ17O and 17O-excess for tracing Alpine hydrologic processes is still relatively new, our discovery that they offer limited immediate added value is useful for future studies.

During winter, streamflow follows a long recession curve, with almost constant baseflow between January and March. Such a constant winter baseflow can result either from a prolonged emptying of a groundwater store (Cochand et al., 2019), as assumed in many rainfall–runoff models (e.g., Muelchi et al., 2021; Staudinger et al., 2017) or from an interplay of groundwater recession and small amounts of snowmelt occurring at the snow–ground interface in absence of soil freezing (Schaefli et al., 2014). Our streamflow isotope measures suggest that streamflow is indeed resulting from a long recession and thus indicative of sufficient groundwater storage to sustain winter baseflow. However, we observed diverging trends in isotopic ratios in two springs. The lower-elevation spring located near the catchment outlet (AUBG) showed depletion in heavier isotopes during winter, whereas the spring located in the higher part of the catchment (BRDG) showed enrichment in heavier isotopes (Fig. 5). This suggests that there are contrasting processes at play at different elevations in Vallon de Nant during winter, which has also been reported previously in snow-influenced catchments, such as in the Colorado River basin (Carroll et al., 2019). As snowmelt is more depleted in heavier isotopes compared to average annual groundwater recharge (Fig. 4), the lower-elevation spring (AUBG) is likely influenced by winter snowmelt, as previously observed in other Alpine catchments (Rücker et al., 2019). However, the higher-elevation spring (BRDG) is largely influenced by underlying groundwater storage, which is more enriched in the heavy isotopes compared to snowmelt (Fig. 4). This is further supported by increasing trends in EC values at BRDG over the course of winters (Fig. 5d), indicating the prominence of subsurface flow. The lower-elevation AUBG spring shows slightly decreasing EC values over winters, confirming the presence of winter snowmelt. Thus, we conclude that in Vallon de Nant, winter baseflow at lower elevations is the combined result of a long seasonal recession and of winter snowmelt (Fig. 8a). At higher elevations, winter baseflow can be largely explained by the underlying groundwater and hence the longer seasonal recession. It is, however, unclear if snowmelt at lower elevations was due to atmospheric heat exchange or ground heat exchange, which is a question that could be interesting for future research.

The early melt period is rarely discussed in the literature (for a model-based example, see the work of He et al., 2015), despite its importance in terms of water supplied to the catchment and prevalence across Alpine regions, and the streamflow during this period remains challenging to model (see Fig. 9 in Brauchli et al., 2017, or Fig. 3 in Thornton et al., 2022). In the Vallon de Nant, the start of the early melt period coincides with thinning or local disappearance of the snowpack from lower-elevation sites, as can clearly be seen in soil temperature measurements at lower elevations (Fig. 3, and as depicted in our sketch in Fig. 8b). This suggests that the streamflow rise during the early melt might be linked to snowpack melting at lower elevations. During this period, snow cover is still abundant at higher elevations, as can be seen in soil temperature measurements at higher elevations (Fig. 3). It is possible that snowmelt might be happening within the existing snowpack at higher elevations but then be retained within the snowpack (Gerdel, 1945) or be locally stored in the subsurface, thus not leading to a strong streamflow response. An in-depth analysis is beyond the scope of this article but provides an interesting research opportunity for future studies.

Interestingly, streamflow increases swiftly at beginning of the E1 period, whereas EC and isotopic values (δ2H, δ17O, and δ18O) show a lagged response (Fig. 6a–d), suggesting that older water stored within the subsurface (with higher EC and more enriched in heavier isotopes) is the first to be exfiltrated into the stream (Mcdonnell et al., 2010) at the onset of early melt. In other words, the streamflow reaction during this period is not resulting from localized snowpack outflow directly to the stream but from snowmelt transiting through the subsurface. Accordingly, during this period, streamflow rise is most likely limited by both limited snowpack outflow and temporary retention of water in the subsurface.

During the melt season, isotopic compositions of the springs converge to a common value (see M1 period in Fig. 5; around -93.5 ‰ for δ2H, -6.8 ‰ for δ17O, and -13.0 ‰ for δ18O), suggesting that the entire subsurface becomes saturated (Fig. 8c) and flushed with snowmelt. This convergence to similar values during the melt is a priori surprising because (i) we have shown an isotopic lapse rate for the precipitation of 0.84 ‰ per 100 m for δ2H and of 0.128 ‰ per 100 m for δ18O, and (ii) the annual median values of the isotopic values for different springs decrease with elevation (Fig. 4a–c). Accordingly, the convergence of isotopic compositions of the springs during melt to a common value strongly hints towards meltwater coming from a similar elevation range or, rather, towards meltwater being sampled from different elevations in a similar way. This has also been seen in previous studies in mountainous catchments (Feng et al., 2022; Penna et al., 2017).

The higher EC values in the stream at certain instances of the melt period compared to springs (Figs. 5 and 6) are unexpected and might suggest that there is a significant amount of subsurface water reaching the stream that has higher EC values than all sampled springs. This result, however, underlines the importance of subsurface flow paths to stream recharge during the melt periods. The positive temperature anomalies (during summer rainfall events) observed during M2 (ROCK) show the existence of fast surface flow paths but are not enough to explain the high EC values during this period.

The EC values in all springs increase during the recession period, clearly suggesting that they are all fed by subsurface water during this period of the year. The isotopic values of the springs diverge, however, after the melt period, giving insights into the underlying reservoirs that are feeding them, in addition to their relative permeabilities and outflow rates. A smaller increase in δ values indicates a larger subsurface reservoir or slower flow rates or permeabilities (e.g., BRDG), where a larger increase is associated with a smaller reservoir size or higher flow rates or permeabilities (e.g., AUBG). This difference between BRDG and AUBG springs suggests the existence of multiple subsurface reservoirs in Vallon de Nant, which has previously been observed in other high-elevation landscapes (Dwivedi et al., 2019; Mosquera et al., 2016).

The isotopic values in streamflow gradually increase over the seasonal recession period towards those of groundwater, suggesting that groundwater becomes the dominant contributor to streamflow generation during this part of the year (Fig. 6). This can also be seen in increasing EC values of streamflow in the R2 period, suggesting prevalence of deeper flow paths over shallower flow paths. However, during large rainfall events, there are sometimes sudden increases in stream isotopic ratio, suggesting that rainfall can become a significant contributor to streamflow generation during intense rain storms. The role of event-based precipitation on streamflow generation has long been studied (Mcdonnell, 1990; Kirchner, 2003; Kienzler and Naef, 2008), with most of the studies concluding that rainfall helps mobilize older water stored within the catchment and hence the isotopic ratio in stream water often largely resembles subsurface storage (see also the meta-analysis of Barthold and Woods, 2015). Our measurements clearly suggest that this so-called old water paradox does not hold for some rainfall-generated streamflow responses during the recession period (Mcdonnell, 1990; Mcdonnell et al., 2010). What processes might explain the fast contribution of recent rainfall to stream water during the recession in the Vallon de Nant remains to be studied in detail, based also on our results for rainfall patterns and their relation to the stream network (Michelon et al., 2021a). This analysis would, however, require additional analyses of shallow groundwater dynamics and the evolution of hillslope connectivity during rainfall events.

Although temperature is not a conservative tracer, temperature measurements of springs are useful to estimate flow path depth. We provide a quantitative tool in the Appendix B. However, the underlying assumption of our method, which is that heat transfer is driven by conduction, might not always be verified (Kane et al., 2001), and anomalies between measured and modeled temperature (pure sinusoid) could be related to heat transport with subsurface water flows (i.e., to advection phenomena).

At shallow depths (10 cm), soil temperature is strongly influenced by air temperature, and our analysis of soil temperature at different elevations shows that it can be used as a good proxy for the detection of snow cover. Early melt starts when the soil temperature at low elevation (1240 m a.s.l.) rises, showing that snow is melting near the catchment outlet. The soil temperature sensor, albeit not intended for this use, seems to be well positioned to detect the onset of early melt at lower elevations.

For the other soil temperature recordings at higher elevations, there is no direct link to streamflow dynamics. The time elapsed between snow cover disappearance at the soil temperature site at 1530 m a.s.l. and the beginning of the melting period varies significantly but is always positive (8 d in 2016, 3 d in 2017, and 51 d in 2018), underlining that intermediate elevations are actively contributing snowmelt water during the rise in the melt period. For the highest site, there can be some snow left at the beginning of the recession period.

Other studies have also observed the reactivity of soil temperature to snowpack depth. Bender et al. (2020) observed that, in general, ground temperature is more stable under thicker snowpacks and that a high level of temperature variability in ground temperature mainly occurs when the snowpack is absent, but they do notice that, under thin snowpacks, ground temperature can fluctuate dramatically, though without heating, perhaps suggesting the presence of sub-snowpack moisture flow. In fact, they showed that vegetation combined with thin snow delayed the ground warming and discussed in detail the value of soil temperature recordings under snow. A larger number of soil temperature sensors would provide an interesting perspective to identify the relative contributions of the different landscape units, elevations, and terrain aspects more precisely. This could be particularly promising in combination with satellite images for snow cover mapping.

Stable isotope composition of water is particularly promising to track the co-existence of seasonal baseflow and winter melt within springs and shallow groundwater. However, this requires a year-round time series to observe where and when isotopic ratios in groundwater becomes enriched or depleted. This year-round monitoring is particularly important, as many subsurface signals are likely to see a “reset”, or return to baseflow, during the main melt period, as highlighted in our work here.

The range of isotopic composition for each water source informs on the relative snowmelt proportions from their respective subcatchments. Without evidence of a strong isotopic lapse rate in snowfall, the differences measured can be explained by the variation in snowfall amounts with elevation.

The relative proximity of some water sources monitored in this study underlines that the spatial proximity does not necessarily imply similar behaviors (in terms of temperature or isotopic composition), as we see noticeable differences between the sources due to the different characteristics of their subcatchments (i.e., flow path depth, hydraulic conductivity, slope, and aspect).

The lc-excess values might reveal some additional insights in future work, in combination with future analyses of soil water isotopes (to give insights into evaporation effects).

The d-excess and 17O-excess are typically used to investigate the large-scale hydrological cycle and oceanic moisture sources (Nyamgerel et al., 2021). Both d-excess and 17O-excess respond to relative humidity during evaporative processes, but 17O-excess may be less temperature sensitive (Surma et al., 2021; Bershaw et al., 2020) than d-excess and thus indicate compositions of evaporation and climatic conditions even when they would be invisible with d-excess (Risi et al., 2010). Our application is quite different from this typical, until now, application. At this stage, it is not clear what the value of 17O-excess is for hydrological purposes, and the question of whether it conveys local-scale information remains open. While so far not reported for measurements of δ17O, it is known that under certain environmental conditions, the snowpack will experience melting, evaporation, and refreezing, thereby forming a firn rather than snow. This may influence the 17O-excess measured for either fresh snow or the final snowpack (see, for example, Risi et al., 2010). Although much laboratory time was devoted here to the measurement of δ17O and 17O-excess, we gained fewer insights and less added value than we had hoped for from documenting the influence of local-scale snow dynamics, specifically the variation in space and time of accumulation, transport, storage, melt and sublimation, on hydrological processes, except for some unvalidated potential to distinguish glacier melt from snowmelt when combined with temperature measurements. Perhaps our measurements would have been more relevant if fresh snow had been sampled instead of the snowpack or if more relevant reference data were available. However, even if we cannot draw any concrete conclusions at this moment, we hope that the full value of the δ17O data set presented here will be investigated and prove independently valuable in the future.

EC allows qualitative estimation of stream water sources and is very useful to distinguish periods where streamflow is dominated by snowmelt vs. groundwater. During periods of high snowmelt, a drop in stream EC can be explained by dilution from meltwater (Chiaudani et al., 2019), which has a shorter subsurface residence time vs. older water stored within the subsurface and, accordingly, a lower ionic content and EC (Cano-Paoli et al., 2019). However, the difficulty in characterizing the different physical and geochemical properties of soils (influencing EC) does not allow an intercomparison of absolute EC values between the sources. However, variations at a given source may inform on the snowmelt input (low EC) or the flow path dynamic (old water pushed by water input). In catchments like Vallon de Nant that exhibit few elevational gradients in stream isotopic ratios, EC represents an extremely valuable tracer to segregate the relevance of snowmelt in stream water generation.