With an ice-covered area of about 14 km2, the Otemma glacier (45∘56′03′′ N, 7∘24′42′′ E) in the western Swiss Alps is amongst the 15 largest glaciers of Switzerland . The glacier is characterized by a relatively flat tongue, which has retreated by about 2.3 km since the Little Ice Age (LIA) and 50 m yr-1 since 2015 . A recent study suggested an almost complete glacier retreat by 2060 .

A Tyrolean-type water intake has been constructed for hydropower production about 2.5 km downstream of the current glacier terminus and is used in the present study as the outlet of what we call the Otemma basin (Fig. b). It has an area of 30.4 km2, a mean elevation of 3005 m a.s.l. (2350 to 3780 m) and a glacier coverage of 45 % in 2019 (adapted from ).

The geology of the underlying bedrock is composed of gneiss and orthogneiss from the Late Paleozoic Era with some granodiorite inclusions . The main geomorphological forms comprise bedrock, with some vegetation cover above the LIA limit (46 %), steep slopes (30 % post-LIA lateral moraines and 10 % talus slopes), gently sloping debris fans and morainic deposits (13 %) and a flat glaciofluvial outwash plain (0.9 %) (Fig. b). One main subglacial channel at the glacier snout provides water to a large, highly turbid and turbulent stream, which quickly reaches a flat outwash plain composed of sandy–gravelly sediments; this leads to a braided river network, which eventually converges in a more confined channel about 1 km downstream and extends to the hydropower intake. A few tributaries from small hanging glaciers or valleys also contribute to river discharge during the snow-free season.

In July 2019, we installed an automatic weather station (Fig. a) at the glacier snout at an elevation of 2450 m a.s.l., which recorded with a 5 min resolution air temperature, humidity, atmospheric pressure (Decagon VP-4) and liquid precipitation (Davis tipping rain gauge). After July 2020, total incoming shortwave radiation was also recorded by the device (Apogee SP-110). For the present analysis, winter solid precipitation data were provided by SwissMetNet, the Swiss automatic monitoring network, using information from the Otemma station (2.7 km from the glacier snout) or the Arolla station (10 km from the glacier snout). Data with a detailed description are available on Zenodo .

Hourly river stage was recorded from 2006 to 2018 at the water intake corresponding to the catchment outlet (GS3, Fig. a) by the local hydropower company (Force Motrice de Mauvoisin, FMM); corresponding discharge was estimated using a theoretical stage–discharge relationship provided by FMM. We post-processed the data by in-filling data gaps related to regular sediment flushing events (of duration <1 h) with linear interpolation. Winter discharge was also recorded, although a data gap usually occurred from October to December.

In August 2019, we installed three river gauging stations, one in the vicinity of the glacier snout (GS1), one at the end of the outwash plain (GS2) and one at the catchment outlet (GS3) (Fig. a). River stage, water electrical conductivity (EC) and water temperature were recorded continuously at 10 min intervals using an automatic logger (Keller DCX-22AA-CTD). Periodic EC and discharge measurements were also made in many tributaries and water sources, with a main focus on three representative tributaries along the outwash plain. Finally, we installed seven groundwater wells consisting of fully screened plastic tubes at an averaged depth of 1.5 to 2 m in the outwash plain, which covered four transects (A to D) perpendicular to the river in the direction of the base of the hillslope. Water table elevation was recorded in each well at a 10 min interval using SparkFun MS5803-14BA pressure sensors. Sensor bias was verified and corrected by bimonthly manual groundwater stage measurements. A more detailed description of the data is available on Zenodo .

Electrical resistivity tomography (ERT) was used to map the sediment structure in the outwash plain. We performed a total of 21 lines from 2019 to 2021 using a Syscal Pro Switch 48 from Iris Instruments (Fig. a). The electrode array consisted of 48 electrodes with a spacing between 1.5 and 4 m, and dipole–dipole (DD) and Wenner–Schlumberger (WS) schemes were systematically used for better data interpretation. We processed the data using the open-source pyGIMLi python library . All data inversions were calculated using a robust scheme (L1 norm) with different regularization strength (lambda from 1 to 100) to assess overfitting and underfitting. The depth of the outwash plain sediments was estimated by performing multiple transects in different parts of the outwash plain and by identifying the transition from water-saturated sediments with a resistivity value between 500 and 2000Ω m and the bedrock layer with a resistivity of 4000 to 7000Ω m, similar to other studies e.g.,. More detailed information on the data, results, codes and maps is available on Zenodo .

While some literature exists to characterize most geomorphological landforms in glaciated catchments, data on post-LIA outwash plains in Alpine environments are scarce. We therefore used two different methods to estimate the saturated hydraulic conductivity (Ks) of the outwash plain.

The first method applied the pressure wave diffusion method documented in the work of . Given a certain hydraulic diffusivity (D), this method was used to relate the aquifer head variations (h) at a distance x from the stream to the diel stream stage cycles (hx=0) generated by ice melt. It furthermore made use of a simplified 1D Boussinesq equation, where advective fluxes were neglected (Eq. ). This procedure is only valid for relatively flat aquifers with a thick unconfined saturated layer and where evapotranspiration losses can be neglected , which makes this approach well-suited for high-elevation outwash plains. By comparing the phase shift (time lag) and the amplitude dampening between the river stage and the groundwater signals, the aquifer hydraulic diffusivity (D) was estimated and related to Ks using the aquifer thickness (B) and assuming that the specific yield (Sy) was similar to the aquifer porosity (Eq. ).

1δhδt=Dδ2hδx22D=KsBSy

For this analysis, we used the two upstream and downstream well transects (B and D; see Fig. a) for two periods: during high flow in mid-August 2019 and during a lower-flow period in mid-September 2019. An additional groundwater well “B3” on transect B was also used for this analysis. The 1D partial differential equation was solved using a central-differencing scheme in space and a Crank–Nicolson method in time, imposing the measured river stage variations as a boundary condition. Prior to solving the equation, both river stage and groundwater heads were detrended by substracting the linear trend of each dataset as suggested by . We then calibrated the model parameter D using a Monte Carlo approach where we minimized the root mean square error and maximized the Spearman rank correlation between observed and modeled groundwater heads. Hydraulic conductivity was finally calculated based on the aquifer thickness estimated by ERT, and porosity was estimated by measuring saturated water content (Decagon 5TM) at five locations in the upper sediment layer.

A second independent estimation of the hydraulic conductivity was obtained with salt tracing, using ERT time lapse with a measurement cycle of about 30 min. We injected 3 kg of salt dissolved in 15 L of water in a 1 m-deep pit in the center of the outwash plain and recorded the timing of the passage of the salt plume at a downstream transect (distance 9.38 m) using ERT, similarly to the work of . We only installed one ERT line perpendicular to the groundwater flow consisting of 48 electrodes with a 1 m spacing. Hydraulic conductivity can be calculated by solving Darcy's law for the mean pore velocity as follows: 3vp=Ksθsdhdx, where dhdx is the aquifer gradient, θs is the aquifer porosity and vp is the mean pore velocity corresponding to the travel distance divided by the travel time of the center of gravity of the salt plume.

We analyzed the storage–discharge relationship at the catchment scale by using a classical recession analysis during periods when both water inputs (snow, rain) as well as outputs (evapotranspiration) can be neglected, i.e., during periods when discharge is only related to aquifer storage . Following , we describe the recession behavior of the aquifer storage with a nonlinear storage (S)–discharge (Q) function, 4S=eQc, whose derivative, using dSdt=-Q, is given by 5-dQdt=1ceQ(2-c). This is usually summarized as -dQdt=aQb, where a=1/ce is the recession coefficient and b=2-c is the slope coefficient . The release behavior of the catchment-scale storage was characterized by identifying zones where the slope of the relationship between the rate of change (-dQdt) and discharge (Q) is constant in the logarithmic scale, which allowed calculation of the slope coefficient b.

We performed the recession analysis for the 12-year period of discharge data provided by FMM at the catchment outlet (GS3). The recession periods were automatically selected by identifying periods where flow is constantly decreasing for at least 10 d and is extended until the first increase in flow. The discharge recession data were smoothed (moving average with a span of 50 % of a given recession period) to remove small step-like decreases or small drops due to sensor failures, so that only the averaged trends are analyzed. Finally, we plotted the relationship between (-dQdt) and discharge (Q), and we average the recession points from all years in bins with an equal number of points (we selected 100), as suggested in the work of , to which we apply a linear regression (nonlinear least squares method, MATLAB R2019a). This procedure allowed estimation of the slope coefficient b. Once b is identified, we fitted a power-law function on the raw discharge data (without any smoothing) for each winter recession, using the analytical solution of Eq. () in order to estimate the recession coefficient e. Finally, this allowed us to relate the maximum baseflow discharge Q0 to the catchment-scale baseflow storage S0 using Eq. ().

Similarly to the catchment-scale recession analysis, the same relationship between storage and discharge can be applied to specific landforms, which allows estimation of the rate of water storage and release in different parts of a glaciated catchment. For instance, the form of the water table in an aquifer can be linked to the shape and physical properties of the landform . Using some simplifications, the Boussinesq equation provides a physically based means of estimating the temporal variation of the aquifer table along a one-directional aquifer and thus allows estimation of discharge based on the groundwater gradient and physical properties of the aquifer . For flat aquifers with homogeneous conductivity, a slope b of 1.5 (c=0.5) is common for the late recession . Here, an analytical solution of the Boussinesq equation was proposed, leading to the discharge solution shown in Eqs. () and (). 6S=eQ0.57Qt=Q0(1+αt)-28α=Q00.5e≈KshmϕL2 A physical description of α was proposed (Eq. ) based on the aquifer conductivity (Ks) and porosity (ϕ), the aquifer length (L) and the aquifer thickness at distance L (hm) .

In the case of a significantly sloping aquifer (>10∘), a value b=1 is usually proposed for the late drainage . In this case, if the aquifer thickness was small enough, the aquifer flux would be mostly advective and conducted by the bedrock slope , so that discharge recession becomes linear (Eqs.  and ). Due to the nonlinearity of the Boussinesq equation, the parameter α could only be approximated using numerical linearization approaches . In this study we used one of the simplest proposed descriptions for α (Eq. ), similar to the previous one, where only hm/L (the aquifer slope) is replaced by sin⁡(θ) and θ is the bedrock slope . 9S=eQ10Qt=Q0e-αt11α=1e≈Kssin⁡(θ)ϕL In both equations (Eqs.  and ), the rate of aquifer decline can be related to a recession constant (1/α), corresponding to the characteristic response time of the aquifer. Based on this approach, we reviewed the range of estimated hydraulic conductivity values reported in recent studies for typical landforms in glaciated catchments. Combined with realistic aquifer properties (slope, porosity, aquifer length) for each type of landform, we applied the proposed relationships for flat (Eq. ) or sloping aquifers (Eq. ) and finally assessed the recession timescales (1/α) at which different storage compartments provide water for baseflow.

Landform classification was performed by combining a visible band orthoimage from 2020 with a 10 cm resolution and a 2 m resolution digital elevation model (DEM) . We calculated the slope from the DEM and classified it in categories as suggested in the work of : <8∘ for outwash plains; 8–22∘ for mildly sloping glacial deposits and debris cones; 22–42∘ for lateral moraines below the LIA limit and talus slopes above the LIA limit; >42∘ for bedrock. We then downscaled the orthoimage to 2 m and combined the RGB bands with an additional band corresponding to the slope classes. We manually identified small zones corresponding to the main landform features and performed a supervised classification using a random tree classifier (ArcGIS Pro v2.3). We finally calculated the median class for a moving window of 10 cells by 10 cells (20×20 m) to smooth out noise in the results. A specific class for grass was used, since many grass patches were identified above the LIA line on shallow soils on top of bedrock. Lateral moraines below the LIA line were distinguished from coarser debris talus slopes with similar slopes in zones where glaciers were absent during the LIA. The glacier extents from 1850 (LIA limit) and 2016 are provided by the Swiss Glacier Inventory 2016 . The results are presented in Fig. b.

Based on the previously discussed recession theory (Sect. ), we propose a simple methodology to estimate the seasonal storage and discharge contribution of each individual superficial landform storage compartment in the Otemma catchment. In order to estimate the maximum water storage, we used the total area (Ai) of each classified landform (Sect. ) and an estimation of their sediment thickness, similarly to other studies . Sediments are however never fully water-saturated, so that it remained difficult to estimate the maximum aquifer thickness for each landform. To overcome this limitation, we defined a simple hydrological model where we simulated a realistic daily water input (Qin) in the form of rain (Prain) and snowmelt (Psnow) and estimated storage (S) and outflow discharge (Qout) based on the nonlinear storage–discharge relationship (Eq. ). We defined c based on the landform slope and estimated e following Eq. () or () using realistic hydrological characteristics of each landform: hydraulic conductivity was based on our measurements (Sect. ) or from a review of the literature, while the aquifer slope and length were estimated for each landform based on our landform classification by manually measuring the averaged landform length (Fig. b).

Following this approach, we defined Eqs. () to () in order to simulate the seasonal storage and discharge over a whole year. 12δStδt=Qin,t-Qout,t,13Qin,t=Psnow,t+Prain,tAi+Qglacier,t/Acatchment,14Qout,t=Ste1/c, where δSδt is the change in storage in mm d-1, Qin,t is the daily water input at time t and Qout,t is the generated daily output discharge based on the nonlinear storage–discharge equation. Psnow and Prain are the daily snowmelt and daily liquid precipitation in mm d-1, Ai is the area of each landform, Qglacier is the daily river discharge from the glacier in L d-1 and Acatchment is the total catchment area in square meters. Finally, e is the recession parameter estimated based on α (Eq.  or ) and c the slope coefficient (1 for slopping aquifers >10∘ and 0.5 for flatter aquifers). In these equations, the landform storage (St) was scaled by dividing the volume by the entire catchment area, which allowed ready comparison of the storage associated with each landform.

The snowmelt input was modeled with a snow accumulation routine (rain transitions to snow from an air temperature between 1 and 2 ∘C) and a degree-day model for daily snowmelt estimation following , with a degree-day melt factor of 6.0 mm ∘C-1 d-1 when air temperature is higher than 1 ∘C. The catchment was separated into 50 m elevation bands with a calibrated temperature lapse rate of 0.5 ∘C 100 m-1 and a precipitation lapse rate of +10 % 100 m-1. Winter precipitations from SwissMetNet were adapted using a correction factor for each year. The melt parameters, precipitation correction factor and lapse rates were estimated by minimizing the error between modeled and observed snow water equivalent (SWE) based on 92 snow depth measurements and two snow pits for density measurements made near the maximum snow accumulation on 28 May 2021. It was further calibrated by matching the snowline limit during the snowmelt season as suggested in , based on daily 3 m resolution Planet images . Snowmelt and rain inputs were considered to recharge entirely the whole aquifer (no surface flow), and there was no routing or water exchange between the different landforms, so that our estimates represent the maximum potential storage linked to a realistic maximum recharge.

In the case of the outwash plain, an additional glacier melt input (Qglacier) was provided, since this is the only landform directly recharged by the river network in Otemma. Only a small fraction of the total river discharge was allowed to recharge the outwash plain aquifer. An infiltration rate of 100 L s-1 (2 % of mean summer discharge) from May to October was used, estimated from dilution gauging along the stream and preliminary modeling results. This amount was also found to realistically approximate the rate of recharge observed using the groundwater wells. Finally, the maximum storage (sediment thickness) of the reservoirs cannot be exceeded in any landform.

Based on the three sources of water (rainwater, snowmelt, glacial stream), a small routine was also added to calculate the source water partitioning in each landform. At each time step, the reservoir was assumed to be fully mixed and a water amount for each water source was removed, proportional to the estimated partitioning at the previous time step and so that the total water removed equates the calculated discharge (Qout,t). The amount of water recharge from each source is then added, and a new partitioning is calculated. This allowed tracking of the seasonal contribution of different water sources in each landform.

From the groundwater head observations in the outwash plain, we computed the daily averaged lateral (perpendicular to the stream) and longitudinal (parallel to the stream) aquifer gradients (Fig. ). During the summer, the lateral upstream gradient (well D1–D2) is mostly comprised between 0 % and 0.5 %. The EC at well D2 is similar to tributary 2, which suggests a hillslope recharge from tributary 2 or a constant deeper bedrock exfiltration which maintains a mild lateral gradient towards the stream. The lateral downstream gradient (wells B1–B2) shows a stronger slope of about 1 % in the direction of the stream, which gradually decreases to values close to 0 % by September. This gradient seems closely related to the snowmelt-fed tributary 3. Indeed, well B2 shows a low EC in the early melt season, similar to tributary 3, which only increases in mid-August, when this tributary runs dry (Fig. ).

The longitudinal gradient seems to maintain a larger slope of about 1 % to 2 % during the summer. Interestingly, the daily discharge in 2020 shows a similar weekly dynamic to the upstream gradient, although the gradient tends to react with a delay of 1 to 2 d. This suggests a strong influence of the stream discharge magnitude on the upstream gradient, which starts decreasing only in early September, i.e., at the moment when discharge peaks decrease.

River stages at the well transects could not be measured continuously due to the high discharge and unstable sediments; a few isolated measurements show that, in the upper part of the floodplain (B, C, and D transects), the river stage is always 10 to 40 cm higher than the groundwater level in the wells closest to the river, indicating a lateral gradient from the stream to the well and thus a losing stream reach. Higher discharge therefore leads to a higher river stage, which increases the hydraulic gradient through the riverbed and therefore promotes higher stream infiltration.

Based on the hydraulic gradients, it appears that groundwater flows in the same direction as the terrain's main slope are recharged in its upstream part by the stream and re-emerge at the end of the outwash plain. This re-emergence results from the underlying bedrock with much lower hydraulic conductivity, which forces water to exfiltrate in the river as the sediment thickness decreases towards the end of the plain. This groundwater upwelling is also supported by the EC in well A1 (Fig. ), which shows the highest EC in the floodplain, although it is located at 5 m from the river, indicating long flowpaths and no direct contact with the river at this location.

In order to disentangle the relative contribution of different superficial landforms, we suggest comparing the recession constant (1/α), which provides a way to compare how fast each aquifer compartment releases water and what their significance is for maintaining flow during dry periods. We reviewed studies focusing on specific landforms in glaciated catchments where hydraulic conductivity (Ks) was estimated in Table .

We then estimated the storage and response time of each unit in the Otemma catchment using the landform-based model (Sect. ) based on Ks values from Table , including maximum and minimal Ks values to account for uncertainty. We also defined aquifer properties realistic for our catchment (Table ). For lateral moraines , we selected Ks to be smaller than for flatter deposits , which probably reflects the lesser degree of compaction at the valley bottom. We separated talus slopes from lateral moraine as talus slope material is coarser and lay above the LIA line. For the outwash plain, we used our own estimate of the hydraulic conductivity, and for mildly sloping glacial deposits, comprised between a slope of 8 to 22∘, we used a mean slope of 10∘ as the majority of those deposits were rather flat.

Supported by a simple degree-day model for snow accumulation and melt, we estimated the catchment-scale average rainfall and snowmelt during the year 2020. Rainfall amounts to a total of 204 mm and snowmelt to 1732 mm of water equivalent (see Fig. a). Figure b shows the resulting estimated maximum storage for each landform.

The resulting maximum baseflow storage in the flat glacial deposits is 19 mm (with an uncertainty margin from 13.5 to 32.5 mm) or a maximum aquifer thickness of 1.1 m (0.8 to 1.8 m) during peak snowmelt. The storage in the outwash plain gradually increases due to constant recharge from the river and rapidly reaches its maximum storage of 11.3 mm (or an aquifer thickness of 10 m). The lateral moraines show a very flashy storage response linked to their short recession constant. Their storage reaches 23 mm (15 to 52 mm) during snowmelt, corresponding to an aquifer thickness of 0.55 m (0.35 to 1.25 m). Due to their very low retention capacity, talus slopes only transmit water, and their storage is low, with only 1.8 mm (1 to 4.5 mm) and a maximum aquifer thickness of 0.11 m (0.06 to 0.27 m). After peak snowmelt, storage decreases quickly in the lateral moraines and somewhat more slowly in the flatter glacial deposits, while maximum storage is maintained in the outwash plain due to the stream recharge. During fall, lower discharge leads to a storage decrease in the outwash plain too, so that by early December the total remaining storage becomes very limited, with only 8.8 mm (5 to 20 mm) remaining from the outwash plain and flat moraine deposits.

Discharge recession was analyzed from 2006 to 2017 at the catchment outlet by calculating the averaged relationship between recession rates (dQ/dt) and river discharge (Q) (Fig. ). A change in slope occurs for discharge higher than 0.33 mm d-1, probably due to the transition between discharge dominated by ice melt to discharge fed by groundwater. Due to this slope change, we assume that the recession starts when baseflow discharge is smaller than 0.33 mm d-1, and higher values are excluded for the linear regression shown in Fig. .

The estimated regression has a slope of b=1.56, leading to a quadratic relationship between storage and discharge (Eq. ). Due to the low values computed in Fig. , a change in the smoothing process of the raw discharge data may have an impact on the recession. We have tested different processing parameters and assessed the impact on the linear regression; overall, the slope varies between 1.45 and 1.65.

Using the same recession periods, the recession trends of each individual year are assessed (Fig. ) using a quadratic relationship (Eq. ) and fitting the maximum baseflow discharge (Q0) and the recession coefficient (e). The corresponding calculated recession constant (1/α) seems to decrease in recent years, but the trend is unclear due to the overall short time period, while the temporal evolution of Q0 and S0 does not show any trend, suggesting no clear increase in groundwater storage over the 12-year period (Fig. ).

Overall, we obtain a similar estimation of the baseflow storage in the Otemma catchment during each winter, with a mean maximum baseflow discharge of 0.34 mm d-1, a mean maximum storage of 42.5 mm and a recession constant (1/α) comprised between 90 and 155 d. Finally, at the end of the recession periods in late winter, discharge has decreased by a factor of 3, which indicates that the baseflow storage does not completely empty and still retains on average 58 % of the maximum baseflow storage of early December.

Those results are in contradiction with the landform-based model (Fig. ), where a maximum baseflow storage during early December was estimated to only 8.5 mm. Accordingly, the landform-based analysis seems to miss a relatively important storage compartment.

Our analysis has shown that the landform- as well as catchment-scale hydrological responses critically depend on (i) the sediment structure defining Ks and (ii) the landform characteristics in terms of slope and aquifer flowpath length. These key properties can then be combined to estimate an averaged response time (1/α) of each landform, although the storage–release behavior may be more complicated when considering more complex aquifer geometries , heterogeneous landforms with varying physical properties for Ks and ϕ, preferential flowpaths or non-stationary processes . In this study, we focused on characterizing the “slow” groundwater compartment, which is relevant for baseflow only, but an initial part of the water release may also occur in a faster superficial layer, as suggested in other studies e.g.,. Our approach, while simple, relies on physical properties of the aquifer. The calculated values for 1/α were similar to studies which estimated this parameter based on direct observations of discharge recession. This supports the validity of our approach for analyzing the storage–release behavior and the relative importance of different landform units in a glaciated catchment.

With this analysis, we have shown that only flat aquifers release water at timescales longer than weeks. In addition to Ks, the bedrock slope plays an important role, as it changes the relationship between storage and discharge, illustrated in our landform-based model by the slope coefficient c. Indeed, steeper slopes promote stronger advective fluxes and modify the recession equation (Eqs.  and ), so that a sloping aquifer (c=1) would lose 50 % of its storage 1.4 times faster and 99 % 4.5 times faster than a flat aquifer (c=0.5).

The seasonal landform-based analysis of superficial storage proposes an example of the groundwater dynamics in a glaciated catchment. The estimated storage amounts are likely not accurate due to a strong simplification of the recharge processes and the absence of superficial overland flow; it nevertheless illustrates (i) the strong relationship between recharge and storage, (ii) the importance of the timing of the water input and (iii) the relative speed at which different reservoirs may empty. Accordingly, we can establish a sound perceptual model (see Sect. ).

Prior to introducing this model, we first discuss and summarize hereafter what new insights we gain from our case study on the hydrological functioning of the main classes of geomorphological landforms.

While our approach identifies the relative size and seasonal hydrological response of proglacial landforms, we use a simplistic recharge model. In reality, hydrological connectivity from the water sources and between landforms will ultimately drive the amount of actual recharge. Due to the coarse and barren nature of the sediments in such environments and the limited presence of soils, it can be expected that any water input will infiltrate into the sediments . It has also been shown that groundwater flow is driven by the bedrock topography underneath the landform, where a strong change in hydraulic conductivity drives the water downslope . We can therefore assume that recharge occurs directly at the location of the water input, percolates until the bedrock and is then directed downslope. In the case of snowmelt, this recharge will gradually move upslope with the snow line during summer, a zone where talus slopes and bedrock are frequent. Water will rapidly be directed downslope at the bedrock interface and directed in zones of bedrock depression, concentrating the flow and thus providing little recharge to other downhill-sloping deposits. Water may also reach a flatter zone in hanging valleys, where flatter morainic material may be present in rock overdeepenings, which likely act as an immobile storage, where groundwater only overflows above the bedrock, similarly to a fill-and-spill mechanism . The concentrated groundwater flow eventually reaches either the main stream or a flat glacial deposit (moraine or outwash plain) and acts as point recharge, so that only areas located below a zone of bedrock convergence will receive recharge. Similarly, glacier melt recharge will mostly occur along the reach of the glacial stream at the valley bottom and will maintain high groundwater storage in outwash plains or flat moraines exclusively.

We summarize here the gained insights into a perceptual model (Fig. ) of the hydrological functioning of the Otemma catchment, augmented with an additional “missing” storage which we tentatively allocate to bedrock (Sect. ). In this representation, the partitioning between the different sources of water recharging each landform are taken from the results of our landform-based model of the Otemma catchment (Fig. ). We also provide a comparison of the discharge amounts provided by each landform proportional to the results of Fig. .

The perceptual model illustrates well how steep lateral moraines may provide large water amounts during peak snowmelt or strong rain events in mid-summer but drain very rapidly in fall. Talus slopes were not included in the perceptual model, as they play a marginal role in the Otemma catchment and have even faster drainage than steep moraines. In contrast to steep slopes, the baseflow provided by the bedrock aquifer appears more stable, although its storage decreases by half during winter. In a perspective of future early snowmelt, the model shows that most landforms may become dry much more quickly, with the exception of (i) the outwash plain, which receives water from the glacial stream, and (ii) the bedrock, which drains slowly, highlighting the future increasing importance of such aquifers for providing wetness and maintaining favorable ecological conditions.

In this representation we also neglected the impact of permafrost melt, although it is likely present at high elevation and in north-sloping moraines and may provide some future additional meltwater in glaciated catchments, as shown in the work of . Rock glaciers were also not included as their presence is marginal currently in Otemma, but their role in storing and releasing water may become increasingly important since they have a capacity to store water on timescales of months, as shown in Table  and as discussed in more deglaciated catchments in Austria .