Time-lapse gravimetry (TLG) is the measurement of acceleration due to gravity, g, at two or more points in time, usually at fixed locations, in order to determine Δg(t). More precisely, TLG measures changes in the vertical component of gravity, which is affected by changes in the distribution of mass: 1Δg(t)=G∭x⋅z^Δρ(x,t)|x|3d3x, where G is the universal gravitational constant, 6.674×10-11Nm2kg-2; z^ is the vertical unit vector; and Δρ is the change in density at location x relative to the measurement point. Δg measurements are generally reported in units of mGal (10-5ms-2) or µGal (10-8ms-2≈10-9g). Individual surveys using a portable relative gravimeter generally include multiple points and, through repeat measurements at one or more reference stations, form one or more closed loops. After corrections to the raw data (Sect. ), the residual effect on g due to mass distribution changes is revealed. Local mass loss due to ice melt or decreased groundwater storage in the vicinity below a measurement location will result in a decrease in gravity between two repeat surveys. As different mass change processes can result in the same Δg values at a given location, interpretation must be informed by other information (Sect. ).

We carried out two microgravimetry surveys on 8 July and 11 September 2024 (Δt=65d) using a Scintrex CG-6 Autograv . A 60 s integration time was used, and, to ensure stability, the device was set to measure until three consecutive measurements within a tolerance of ∼2µGal were recorded. The average of this stable 180 s duration measurement period was used in the below-discussed analysis. To enable time-lapse measurements and to correct for instrument drift, both surveys started and terminated at absolute gravimetric reference point no. 1018 , a first-order point (g=980224831±3µGal) located on a bedrock outcrop at the Maloja Pass. In addition to the MALOJA reference point, nine points were surveyed during both surveys: Silvaplana church, two locations (MURTEL01 and -08) adjacent to the RG, and six locations on the RG (MURTEL02 to -07) (Fig. ). The number of points and their locations were chosen so as to maximise coverage across the RG and to limit the survey duration, ensuring a better drift correction. The total duration of each survey was under 5 h 20 min. Future surveys at Murtèl – or, indeed, at other RGs – could include more measurement locations, requiring multiple survey loops and a longer survey duration. We marked survey points with paint and used a Leica GNSS DGPS

global navigation satellite system differential global positioning system

Several standard corrections were applied to the gravimetric data: tilt, internal temperature, height difference, Earth tides, and drift. The first two, tilt and temperature, are automatically applied by the gravimeter to each individual measurement. Tilt was maintained in the calibrated <10arcsecond range during all measurements; thus, the internal corrections were ≲1µGal. The gravimeter was in a stable internal temperature state during the surveys, having being supplied with continuous power for several days prior to the surveys, thus ensuring internal temperature corrections were within the calibrated range. Corrections for height, Earth tides, and drift were all applied manually. During repeat measurements at a given location, the exact height of the device can vary slightly due, in part, to offsets caused by levelling. Gravity decreases with elevation (free-air gravity gradient): 2∂g∂r=-2GMEr3≃-3.1µGalcm-1, where ME is the mass of the Earth and r is the distance to its centre of mass. While the true value of the local vertical gravity gradient (VGG) varies over Earth's surface, especially in complex terrain, 3.1 µGalcm-1 is adequate for this correction of a few mm.

Earth tides, whose magnitude can reach ∼300µGal, are periodic variations in g primarily due to the relative positions of the Sun and Moon with respect to the Earth. Through tidal harmonic databases, Earth tides can be calculated to a level of precision and accuracy that far exceeds gravimetric measurement accuracy. Modern gravimeters implement these corrections internally; however, to ensure the highest possible accuracy, we chose to manually apply these corrections using the current state-of-the art database, which contains over 28 000 wave groups. Through a Python script, we used the pygtide package to calculate the Earth tides for each individual measurement to sub-nGal precision. These corrections, which ranged from -62.5 to +81.7µGal for the July survey and from -45.9 to +41.1µGal for the September survey, were applied to the measured data. The resulting datasets were used for the calculation of, and correction for, instrumental drift, which occurs in all relative gravimeters. The Scintrex CG-6 has a low instrumental drift of <200µGald-1 and <20µGald-1 residual drift , with independent observations of <1.5µGalh-1 . By comparing the repeat measurements at MALOJA (Fig. a), we calculated and corrected for total drift. Any residual effects of transient atmospheric pressure are also indirectly corrected for during the drift calculation. Finally, the corrected data were referenced to the MALOJA reference point to enable calculation of the temporal difference, Δg, in the two surveys.

The gravimetric investigations of , carried out in Summer 1991, provided information on the internal structure of the Murtèl RG. Raw data, such as the gravity station coordinates and non-terrain corrected gravity values, from the 1991 survey are unfortunately no longer available. Thus, while terrain corrections are not necessary for time-lapse gravimetry studies, since the data in the study are Bouguer anomalies (i.e., terrain corrected), we also perform these corrections to our gravimetric data to enable comparison. Furthermore, while our gravimetric stations MURTEL05, -06, -03, -07, and -08 correspond to “Profile 3” , we did not return to the exact locations of the 1991 stations.

Calculation of Bouguer anomalies (gBA) involves four corrections: 3gBA=gmeasured-[γ0+cFA+cBP+cterrain], where γ0 is “normal” gravity as calculated from the ellipsoid, while cFA, cBP, and cterrain are the free air, Bouguer plate, or terrain corrections, respectively. While methods for all of these corrections (also sometimes referred to as “reductions”) are extensively documented e.g.,, there are several variations in the possible formulas and methods for each of these corrections. The methods reported by , and also detailed in , included an erroneous form of the 1967 international gravity formula in a second-order expansion , a linear (-0.3086mGalm-1)z free-air correction, and a linear Bouguer plate correction of (4.19×10-5mGalm-2kg-1)ρz (where ρ=2670kgm-3). Their reported terrain corrections included two inner prism zones and two outer line-of-mass zones to 167km. Our application of these methods did not result in BAs in their reported range from -157.0 to -155.4mGal. Such issues related to reproducing historic BAs published in the Swiss Gravimetric Atlas have been encountered and documented by others . Swisstopo, the Swiss Federal Office of Topography, has also confirmed these issues related to incomplete documentation of the processing of historic datasets. We thus apply modern standard corrections in order to enable comparisons of BAs relative to others in the same survey. Specifically, we use the full GRS80 Somigliana formula for normal gravity, the Wenzel free-air correction with ETRS89 latitudes, and the Talwani Bouguer plate formula with a density of 2670 kgm-3. Terrain corrections were performed using the Swisstopo GRAVNIV software with the publicly available 0.5 m Swiss digital terrain model swissALTI3D, as discussed in and further detailed in Appendix 3A of .

As discussed in Sect. , the only sources of the 1991 survey data currently available are the graphical data in and (both of which contain the same information). We used a digitisation tool to extract, directly from the figures, the positions of the 1991 gravimetry survey points and the values of the calculated BAs and relative position along “Profile 3” data. Accuracy in the BA data digitisation procedure can be considered ±∼10µGal and ∼2m along the profile. Planar positions can be considered accurate to ±∼3m, although we note some discrepancy between planar positions on the original map and the distances along the original “Profile 3”. In the absence of other information, we present the 1991 survey data as originally published.

On both survey dates, we obtained aerial images using a compact DJI Mini Pro 4 with a 48 megapixel camera. The drone was piloted over the lower part of the RG to ensure coverage across the target zone. Due to the presence of overhead aerial cables and an operating gondola lift, extended zones were not surveyed. In total, 375 images were acquired during the July 2024 survey and 268 images during the September 2024 survey. As non-RTK

real-time kinematic

Digital surface models (DSMs) and orthomosaics were created for both survey dates. We used the photogrammetry software Pix4Dmapper (v4.9.0, 2023, Pix4D S.A., Switzerland). The average ground-sampling distance was 2.04 cm for the July survey and 3.45 cm for the September survey. A high number of median key points per image (>72000 for both surveys) were determined, each with a minimum of four matches across multiple images. We manually georeferenced six GCPs (MURTEL02 to -07) that could be located in multiple images with high precision and accuracy. The georeferenced point clouds were used to produce DSMs and orthomosaics with 5 cm resolution. We used noise filtering and inverse distance weighting, but no surface smoothing, for the DSM processing. Comparison of the final results revealed only very small offsets at the sub-pixel scale.

In order to calculate the effect of snow on measured gravity, snow density needed to be measured. We measured the density using a portable digital balance (precision 0.1 g) and a coring tube (inner diameter 5 cm). The volume and mass of 18 samples from different locations at various depths up to ∼1m were measured. The measured sample sizes ranged from 137–736 cm3 and from 76–410 g. Measured densities ranged from 504–587 kgm-3, a typical range for névé . The average snow density was determined to be 550±27kgm-3.

The gravitational effect of the snow, present in the July survey and absent in the September one, at each of the survey points was calculated through a multistep process: (1) creation of a snow raster mask based on the July orthomosaic, (2) calculation of the height difference between the DSMs, (3) fusion of these data to create a 3-D differential snow volume model, and (4) forward calculation of the vertical gravity effect at the gravity survey points. Python, implementing standard packages (numpy, matplotlib, etc.) and rasterio, was used for these processes. Throughout, manually clipped DSMs and orthomosaics were used, limiting the analysis to snow-covered regions with high-quality DSM data for both surveys.

We first created a spatial mask for the snow-covered pixels using the July 2024 orthomosaic (Sect. ). This georeferenced image was converted to a luminance (L) greyscale format as per the Rec. 709 formula: 4L=0.2125R+0.7154G+0.0721B, where R, G, and B are the red, green, and blue pixel intensities, respectively. The luminance image was blurred using a uniform, centred filter of dimensions 1.2m×1.2m (25pixels×25pixels). This was necessary in order to avoid isolated high-luminosity pixels being identified as snow. A simple 55 % threshold was applied to the processed image, with the exception of the low-albedo snow in the southeast corner of the surveyed zone, where a 38 % threshold was used. This process resulted in satisfactory identification of snow-covered regions, while avoiding misidentification of non-snow-covered regions such as high-luminosity boulders. The height difference between the July and September DSMs was evaluated by a simple difference between the two. The above-discussed snow mask, as well as a 5 cm thickness lower bound, was applied to this raster data. The final result was a 2-D map of snow thickness (Fig. ), which, when combined with either the July or September DSM, provides 3-D information on the change in snow volume distribution.

To calculate the decrease in gravity due to the melting snow, we performed forward gravity modelling using a modified version of Gravi4GW that implements the Nagy prism rather than equivalent point masses. A correction for the height of the proof mass 6.58 cm above the bottom of the gravimeter, relative to our surveyed coordinates (Fig. b) was also applied, although the effect was minor for all points (0.0–0.3 µGal).

Additionally, when comparing the digital surface models from the two surveys, we discovered a new large boulder, ostensibly originating from the unstable zone located adjacent and above (southwest) the RG. We estimated the boulder's mass to be <1×105kg. The nearest gravimetry point, MURTEL04, was located ∼25m away and ∼1m lower than the boulder's centre of mass. The straight-line gravitational pull of this boulder on MURTEL04 was thus ≲1µGal, and the vertical component (i.e., that which we measure in time-lapse gravimetry) was <0.1µGal. Thus, we conclude that this new rock mass can be ignored in our analyses.

In order to correct for the gravity effect of the snow present in the July 2024 survey, the net gravitational effect on the survey points MURTEL01 to -08 was calculated (Sect. ). This vertical Δg effect of the snow ranged from -0.58 (MURTEL04) to -11.39µGal (MURTEL06) (Table ). Only the vertical component of snow-induced gravity change has an effect on measured gravity (Eq.  and Sect. ). While a significant amount of snow was present (Fig. ), the majority of its gravitational effect is non-vertical. For reference, 1 m³ of snow (with a density of 500 kgm-3) at a horizontal distance of 10 m, centred 1 m below a given point, would have a vertical gravity effect of ∼3.6×10-3µGal. We also note that while snow-covered regions extending hundreds of metres from the RG could not be surveyed, these zones are generally at higher elevations than the survey points. The gravitational effect of the snow in those lateral zones would thus be small and positive . The limited spatial coverage of the DSMs would thus have the effect of underestimating net mass loss in the AL and subsurface water. Finally, the statistical uncertainty in the measured snow density was 4.9 %, which translates to an uncertainty ranging from ±0.03 (MURTEL04) to ±0.56µGal (MURTEL06).

Accounting for the snow, net gravity changes were negative at all eight survey points (Table ), indicating a local decrease in subsurface mass. With the exception of MURTEL04, located near the western flank of the RG, all net gravity changes were in the approximate range of -5 to -15µGal. Along the central profile, Δg decreases in magnitude from -15.2µGal near the lower rooting zone (MURTEL05) to -5.2µGal near the front. MURTEL08, located on bedrock ∼20m from the RG front, experienced a decrease in gravity of -10.6µGal. Results from the two lateral survey locations, MURTEL02 near the east flank (-10.3µGal) and MURTEL04 to the west (-29.2µGal), differ significantly, although limited DSM coverage may have a minor influence on the snow-corrected value at MURTEL04. Overall, while there is a spatial trend along the central line, the results point to significant heterogeneity in the subsurface processes. For first-order interpretation, the Bouguer plate approximation (BPA), an infinite flat plane where a decrease of 2.38 cm in water level equivalent equates to a -1µGal change, provides a rough starting point, implying decreases in subsurface water storage from 12–70 cm of water equivalent (see Sect.  for an in-depth discussion and quantitative interpretation).

For uncertainty estimation, we follow the same logic as previous TLG investigations , noting that these two studies used Scintrex CG-5 devices, which have significantly worse temperature and tilt stability and greater drift rates compared to the CG-6 device used in our study . In terms of uncertainty, we assume three significant and uncorrelated sources of uncertainty, in addition to the before-discussed uncertainty in the snow and rockfall correction (Table ): precision, drift, and elevation. Precision groups together uncertainty related to vibrations, tilt corrections, and internal temperature corrections (Sect. ). As determined from standard deviation of the three repeats whose average was used to determine the final values, we observe values ranging from ±0.2 to ±2.4µGal with an average of ±0.9µGal for the July survey and ±0.05 to ±1.1µGal with an average of ±0.6µGal for the September survey. Adding these uncertainties together in quadrature for each survey point, we obtain precision uncertainties in Δg of ±0.7 to ±2.5µGal. We assume an uncertainty associated with the impact of non-linear drift of ±4.4µGal based on survey duration and device specifications. Finally, uncertainty in Δg associated with uncertainty in the height of the device is assumed to be ±1µGal as we manually measured the height of the device to ≲0.3cm precision (Eq. ) in addition to surveying. Calculations of these uncertainties, combined with the snow and rockfall correction uncertainty for individual points, range between ±4.6 to ±5.2µGal and thus we arrive at a total estimated uncertainty in the snow cover-corrected Δg of ±5µGal.

As discussed (Sects.  and ), even though most TLG applications do not require normal gravity, free air, Bouguer plate, and terrain corrections, we needed to calculate and apply these to calculate the Bouguer anomaly for comparison with the digitised 1991 survey data. Furthermore, due to incomplete documentation of the processing steps in and lack of the original DSM files, absolute differences between the published 1991 values of our BA values differ by an offset which cannot be known exactly (Sect. ). While uncertainty of the 1991 survey data is not extensively discussed in or , the techniques and technology of the time clearly limited the achievable accuracy of the data. The LaCoste and Romberg model G gravimeter used in 1991 had a repeatability of 100 µGal and accuracy of <40µGal. The higher density of points in that study (1 every ∼25m in the central line profile) allowed for the calculation of a smoothed trend line which we use for comparisons to the 2024 data on a relative basis (Fig. ).

By displaying the calculated BAs with a fixed offset of 3.9 mGal (Fig. ), we can approximately match our off-RG forefront survey point, MURTEL08 (Fig. ), to the 1991 smoothed line in order to facilitate visual comparisons of the relative values. Our Bouguer anomalies display a larger difference between the forefront region (MURTEL08) and front and central RG zones (MURTEL07, -03, and -06) than in the 1991 survey. In contrast, the difference between the forefront and lower rooting zone at the transition to the talus slope (MURTEL05) is similar to that observed in the 1991 survey. Comparing with the smoothed historic data, the relative difference between the forefront and the front/central zones appears to be approximately 110–190 µGal greater in 2024 than in 1991. Similarly, comparing MURTEL08 to MURTEL04 (Fig. ), the difference between the BAs at these points increases by ∼200µGal between the 1991 and 2024 surveys. Our other survey points, MURTEL01 and -02, do not overlap with any of the 1991 survey points. Their BA values are -152.241µGal (July) and -152.253mGal (September) and -152.550 (July) and -152.564mGal (September), respectively. Comparisons of data from our surveys and the survey must be interpreted with caution due to the unquantified uncertainties in the 1991 survey data and the differences in the locations of our survey points. As such, we stress that the 1991–2024 comparison provides qualitative, relative insights rather than precise, quantitative measurement of Δg.

Comparison of the 1991 survey data and our measurements must be nuanced by the technological and methodological factors discussed in Sect. . Nevertheless, some interpretations and hypotheses can be set forth. While over a single thaw season, or part thereof, melt of ground ice in the AL is the dominant mass redistribution process, over multi-year timescales – in this case 33 years – there are several processes that may significantly affect g.

The starkest change in the spatial trends of the BA (Fig. ) is the relative decrease in gravity in the central and frontal regions (MURTEL06, -03, and -07) as compared to both the forefront measurement point (MURTEL08) and that in the lower rooting zone (MURTEL05). Although MURTEL08 is located ∼25 m from the 1991 survey profile, the others are all located <5m from it. Thus, even if we ignore the forefront point, a clear difference in trend can be observed where the BA is now lower relative to that of the rooting zone than it was in 1991. Mountain permafrost regions are highly dynamic environments characterised by mass-moving processes (rock/debris) and total water (water, snow, ice) storage changes acting on different magnitudes, rates/timescales, and depths below ground, i.e., on the surface, in the AL (supra-permafrost), in the permafrost body (intra-permafrost), and beneath (sub-permafrost). The Murtèl periglacial cirque is typical in this respect . Here, we discuss processes related to both rock mass movements and water/ice storage changes that could be relevant for decadal changes in gravity in alpine periglacial settings (Table ).

Landform and rock mass movements involve gradual or episodic movement of mass. Active rock glaciers undergo creep at rates of a few cmyr-1 to myr-1 , and destabilised rock glaciers can move at rates in the ∼cmd-1 range . Even at slowly deforming RGs, such as Murtèl, the effects of creep on mass distribution, and thus potentially g, will be significant on multi-year timescales. Furthermore, RG creep will eventually render it impossible to return to the same survey points for repeat surveys as the point in space may end up being within the rock mass or high above the land surface. Although Murtèl exhibits comparatively slow creep rates of ≲10cmyr-1, amounting to ∼2-4m of horizontal displacement over 33 years, the displacement alone is very unlikely to account for the 1991–2024 differences. We note that shifting these 2024 BA points by ∼-30m along the x axis in Fig.  would result in an overlap with the 1991 trend line. Rockfall and other landslide-type events also redistribute mass. The interplay of the effects of rock debris from headwalls above rooting zones and those of RG creep is highly complex and would require careful and extended topographic surveying (i.e., with lidar or photogrammetry) at the time of each gravimetric survey. When conceptualising the gravitational effects of these, and, indeed, all mass redistribution processes, it is important to remember the r-2⋅z^ weighting (Eq. ) of mass changes. All things being unchanged, if mass is removed from an elevation above the point and added at an elevation below a point, then gravity will increase. However, the effect of falling rock mass is dependent on the relative distance and vertical orientation relative to a measurement location, both before and after the rockslide episode. Finally, tectonic uplift may be important to consider for multi-decadal TLG surveys, although its effect would be uniform across such a small study area at the Murtèl RG. In the Bernina Alps, tectonic uplift rates are about 1–2 mmyr-1 ; thus, tectonic uplift contributes no more than -0.6µGalyr-1 (Eq. ) or -20µGal to the BA between 1991 and 2024.

Water storage changes, in both liquid and solid form, impart both seasonal (periodic) and decadal changes to g. Surface processes, such as snow accumulation and melt, cause periodic changes and are much more straightforward to measure than subsurface ones, although gravimetry has been demonstrated to be useful in its monitoring on short and long timescales . Annual freeze–thaw cycles in the AL are also primarily periodic in their effects (Sect. ). Decadal progressive deepening (or thinning) of the AL could potentially be measurable with TLG. In periglacial environments in general, permafrost degradation will have measurable effects on g, as will groundwater storage changes (see Sect. ). In RGs, specifically, intra-permafrost melt of excess ice results in a net loss of mass and may result in an increase in bulk density. At a given point in space above the RG, these processes would decrease g, although the effect on BA would be more complex to discern due to the potential for compaction and also due to dynamic effects from compressive thickening or extensive thinning. Overall, our BA results suggest that, compared to 1991, there is now a relatively lower mass density in the central and frontal regions compared to the rooting zone. It is highly likely that both ice/water storage (Fig. ) and rock movement processes have influenced these changes in the BA. Additionally, the periodic drying of frontal springs during melt periods (Sect. ) supports the hypothesis that deep groundwater export may have also occurred. Future high-precision gravimetric measurements that can be compared in absolute terms to our 2024 surveys will enable the evaluation of long-term water/ice storage change processes within the RG.

TLG offers clear opportunities for monitoring the various seasonal and long-term water/ice storage processes that occur in permafrost and periglacial environments. While gravimetry has existed for decades, modern portable devices, such as the Scintrex CG-6 used in our study, offer levels of stability and accuracy that provide new opportunities for monitoring subsurface processes whose effects may only be on the few µGal level. Gravity measurements are non-invasive, integrative, and distributed, and they are possible essentially anywhere that is accessible by foot. Application of TLG to determine absolute gravity changes requires a known reference point, which may not always be feasible. Switzerland has a dense network of maintained absolute gravimetric reference points , and the Murtèl RG has the particularity of being rapidly accessible via cable car for much of the year. This advantageous situation is not always the case for alpine sites. Nevertheless, there are also multi-step strategies that can be employed to establish temporary local references, although they may require large travel distances and long times, depending on the location. These constraints can effectively prevent the carrying out of absolute TLG surveys. In these cases, relative TLG, which involves using a reference point at which the value of g is not known absolutely and may vary with time, may be an option. Relative TLG, wherein a fixed Δg offset between surveys is not known, still provides data on changes in g at survey points relative to one another and has already provided useful knowledge in the growing field of alpine hydrogeology .

The non-discriminatory sensitivity of gravimetry to mass changes can be an advantage in permafrost or periglacial environments, where water co-exists in solid and liquid state, as it yields measurements of the total water storage change including ice and water. Without the need for petrophysical relations, forward modelling can be relatively straightforward. This makes gravimetry an ideal complement to other geophysical measurements that are selective to specific physical properties (e.g., electromagnetic or elastic properties). Conversely, attributing the total water storage change to individual supra-, intra-, and sub-permafrost ice/water storage changes (Table ) is ambiguous and requires independent observations or a priori knowledge of the stratigraphy and hydraulic properties. Gravimetric surveys also require precise positional measurements of survey points to enable accurate temporal comparisons. For example, in the case of subsidence caused by excess ice melt, mass is lost beneath the survey point, which should decrease gravity. Simultaneously, the subsiding point moves closer to Earth's centre of mass, increasing gravity. Only over multi-year timescales can the two competing effects be accounted for, and the loss of the excess ice can only be quantified if elevation and VGG are known to a high precision. As many complex processes occur in periglacial landforms over seasonal, yearly, and longer timescales, TLG surveys at multiple timescales could provide valuable information to help constrain hydrological, structural, and rheological models of these geomorphological structures.

As recently underlined in multiple recent reviews , alpine hydrogeological processes and, in particular, cryosphere–groundwater connectivity are insufficiently understood. In many alpine headwater catchments, RGs are a vital component of annual hydrological and hydrochemical dynamics . Permafrost ice, whether in RGs or other landforms such as frozen talus slopes , is, in essence, immobile groundwater whose presence constitutes a water store and also modifies the subsurface hydraulic properties. Ice build-up and melt are, in turn, controlled as much by the ground thermal regime and the heat fluxes as by water availability from precipitation or snowmelt, be it the seasonal turnover in the AL or permafrost thaw in the underlying rock glacier core. Semi-impervious permafrost bodies can act as aquitards , leading to rapid supra-permafrost runoff of intense precipitation and a “flashy” hydrograph typical of many permafrost-underlain catchments . At Murtèl RG, however, the before-discussed hydrological measurements, stake measurements, and AL energy budgets suggest that up to 20–30 % of the snowpack is retained as AL ice and melts slowly enough to percolate through the semi-impermeable permafrost body. This “leaky” behaviour and the hypothesis that coarse, blocky permafrost landforms contribute to groundwater recharge by seasonally storing and routing winter precipitation to a sub-permafrost aquifer have been investigated at multiple sites . TLG, as discussed in Sect. , is sensitive to all mass changes; therefore, the method would need to be combined with other techniques in order to resolve and differentiate storage changes in permafrost and groundwater.

Due to its extensive documentation and overdeepened bowl-shaped bedrock base, Murtèl would be an ideal site for further studies investigating permafrost–groundwater connectivity. The concave underlying bedrock profile may facilitate the infiltration of basal flow to the fractured bedrock (Fig. ). Characterisation of the fracture network in the granodiorite bedrock would allow for a more comprehensive understanding of the cryosphere–groundwater processes occurring at Murtèl and their implications for the wider catchment. A multi-method approach, combining hydrological and geophysical measurements with numerical hydrological modelling would help us to understand the timings, volumes, and flow paths of water originating from AL ice melt and permafrost degradation.