The water level gauge is located in the lower forefield after the confluence of all four rock-glacier springs, and it captured the catchment-integrated surface outflow (Fig. , Table ). The discharge Qw is expressed as a power-law relation of water level hw (stage) with empirically fitted coefficients c1 and c2, 1Qw=c1hw-h0c2, where h0 is the stage at zero discharge (standing water in logger pool). The coefficients for the stage–discharge (hw–Qw) relation (Eq. ) were constrained by dilution gaugings with sodium chloride as a chemical tracer or the volumetric method (bucket method) if discharge was too low for dilution gaugings. The water level is obtained from the total pressure Ptot measured by the submerged logger (pressure compensation) via the hypsometric equation that corrects the barometric pressure measured on the rock glacier to the elevation of the gauging station, 2hw=Ptot-Patm′ρwg,Patm′=Patm⋅exp⁡ΔzgRT‾v. Δz = 42 m is the elevation difference, T‾v [K] the layer-averaged virtual temperature (approximated by the actual temperature Ta), Patm′ the elevation-corrected air pressure, ρw = 103 kg m⁻³ the water density, g the gravitational acceleration [9.81 m s⁻²], and R the specific gas constant [287 J kg⁻¹ K⁻¹].

The electrical conductivity (EC) of the water was monitored at the two main springs at the rock-glacier front and at the total outflow in the bedrock step downstream of the confluence (Fig. ). We report water EC referenced to 25 °C, ϰ [µScm-1], calculated from the measured conductivity ϰϑ [µScm-1] at water temperature ϑw [°C] via 3ϰ=ϰϑ1+αϑw-25°C, with a linear temperature compensation factor α = 0.019 °C⁻¹ . Additionally, we measured EC manually using a WTW LF 320 with a TetraCon 325 probe as a reference for the two quasi-continuously measuring conductivity logger models (Table ) and when the water loggers were found dry (water level too low).

We took grab samples of the spring snowpack before onset of snowmelt (coring with a plastic tube), the rock-glacier outflow, cumulative rainwater, and supra-permafrost water in a rock-glacier furrow where the ground ice is accessible (at the stake measurement spot; Sect. , Fig. ). The water samples were stored in polypropylene bottles with little headspace and tightly sealed with Parafilm in order to minimise evaporation effects.

Water stable isotope composition (δ2H, δ18O) was analysed by cavity ring-down spectroscopy at the Institute of Geology of the University of Bern using a Picarro L2120-i analyser attached to a V1102-i vaporiser.

We report the water stable isotope composition as a δ ratio [‰] of the sample to the Vienna Standard Mean Ocean Water (VSMOW), where δ is the ratio of 18O/16O and 2H/1H. Analytical precision is ±1.0 ‰ for δ2H and ±0.1 ‰ for δ18O. The altitudinal gradient in δ18O does not exceed -0.2 ‰ for every 100 m (annual average) . The isotopic differences over the catchment elevation range (2600–3100 m a.s.l.) are within 1 ‰.

The PERMA-XT pointwise snow depth measurements hS (sonic ranger data, Table ) located on a windswept rock-glacier ridge are converted to snow water equivalent (SWE) [kg m⁻² = mm w.e.] with the semi-empirical parsimonious Δsnow model . Additionally, to harness the larger footprint of the SEB for a spatially averaged snow depth estimate on the rugged rock-glacier surface, the deviation of the SEB, dev_SEB, during the snowmelt months is back-calculated to SWE, 4SWE≈devSEBΔt/Lm, where dev_SEB := Q∗-QH-QLE-QG . The SEB deviation is the remainder of the turbulent fluxes (QH+QLE) and ground heat flux (QG) subtracted from the surface net radiation (Q∗). Δt refers here to the duration of the snowmelt period as estimated from temperature measurements in the snowpack (0 °C) and using time-lapse imagery from an on-site camera (Table ).

Liquid precipitation data are taken from the on-site rain gauge, assuming that precipitation is liquid based on a threshold air temperature of Twb = 2 °C. The rainfall heat flux QPr was estimated via 5QPr=CwrTP-0°C, where Cw = ρwcw [4.18 MJ m⁻³ K⁻¹] is the water volumetric heat capacity, and r [m³ m⁻² s⁻¹] is the rainfall rate intercepted at the surface as measured by our on-site rain gauge or from MeteoSwiss data (Sect. ). Precipitation temperature TP was approximated using the wet-bulb temperature Twb, calculated from air temperature and relative humidity .

The evaporative water flux (including sublimation if Ts < 0 °C) is derived from the latent turbulent flux QLE of the SEB that is estimated with the bulk aerodynamic method . The flux–gradient relation is expressed as 6QLE=ρaLvqa-qsrq. QLE is driven by the specific humidity difference between the atmospheric air qa and the snow (if snow covered) or debris surface (if snow free) qs; qs under snow-free conditions is taken from humidity measurements in the near-surface AL (qa at 0.7 m depth; Table ). Otherwise, the surface is considered saturated at the radiometrically determined surface temperature. The bulk aerodynamic resistance for vapour transport in the near-surface atmosphere rq [s m⁻¹] decreases with the strength of turbulence: a thermally unstable atmosphere or strong winds enhance turbulent transport. rq is estimated using the Monin–Obukhov similarity theory (MOST) and the parameterisations detailed in and .

During the thaw season, the ground heat flux Qg [W m⁻²] downwards into the coarse blocky AL is spent on warming the debris ΔHalθ (sensible heat storage changes), melting ground ice in the AL ΔHali (latent heat storage change or ground-ice storage change), and conducting into the permafrost body beneath QPF (“permafrost heat flux”) () (Fig. ), 7∂∂t∫0ζ(t)CvTal(z,t)-0°Cdz︸ΔHalθ+Lmρi∂∂t∫0ζ(t)fi(z)dz︸ΔHali,deval,Qm=Qg-QPF︸Qnet[Wm-2], where ζ is the depth of the ground-ice table (AL thickness) [m], Cv the volumetric heat capacity of the debris [J m⁻³ K⁻¹], and Tal the AL temperature [°C]. fi [–], Lm [3.35×105 J kg⁻¹], and ρi [kg m⁻³] are the volumetric ice content, latent heat of melting, and ice density, respectively. The AL energy budget (Eq. ), derived from the conservation of energy principle, is estimated based on data from the instrumented main cavity (Fig. ) as outlined below. We compare two independent estimates of the ground-ice melt ΔHali: one from the deviation of the AL energy budget denoted by dev_al (Sect. ) and one from the Stefan model based on direct stake measurements of ζ(t) in a nearby rock-glacier furrow denoted by Qm (Fig. , Sect. ). Details on the measurement set-up and data processing are in .

We estimate the thaw-season ground heat flux from two measurements: from the AL net long-wave radiation QCGR3rad and the heat flux plate QHFP (Table ). These two measurements are correlated and represent the downwards heat flux Qg in the absence of buoyancy-driven convection, i.e. in conditions of stably stratified AL air column which prevails during the thaw season. These Qg measurements are at 1.5–2.0 m depth in the AL, not at the surface. Details about data pre-processing are in .

The sensible heat ΔHalθ stored/released by temperature changes of the blocks in the coarse blocky AL beneath the Qg measurement depth is estimated by 8ΔHalθ≈Cvh∂〈Tal-0°C〉∂t, where Cv = 1-ϕalρrcr is the volumetric heat capacity [0.6 × 2690 kg m⁻³ × 780 J kg⁻¹ K⁻¹] (porosity ϕal = 0.4), h the distance from the Qg measurement level to the AL base [2 m], and 〈Tal〉 the spatially averaged AL temperatures [°C].

The deviation dev_al for closure of the AL energy budget (Eq. ), after assessment of the uncertainties, corresponds to the latent heat storage changes, i.e. heat spent on melting ground ice, 9deval:=Qg-QPF-ΔHalθ.

The heat flux across the permafrost table QPF is estimated with the gradient method from PERMOS borehole temperature data via Fourier's heat conduction equation: 10QPF≈-kPFΔTPFΔz, where the borehole temperatures are measured at 4 and 5 m depth in the permafrost body beneath the AL. We take a thermal conductivity kPF value of 2.5 W m⁻¹ K⁻¹ .

The ground ice is accessible at a few spots, all located in furrows where the AL is thinner (1–2 m; Fig. ). In one spot, a plastic tube was drilled ca. 120 cm into the ice in August 2009 but subsequently abandoned (Christin Hilbich, personal communication, 2022). We made serendipitous use of it as an “ablation stake” (Fig. b1), manually measuring the depth of the ground-ice table ζ(t) [m] at each field visit in the summers of 2022 and 2023 . Assuming that changes in ice content fi only occur at the phase change boundary ζ(t) coinciding with the 0 °C isotherm (negligible melting point depression in the coarse material), the ice melt heat flux Qm can be expressed as 11Qm=fiLmρidζdt.

Given the sparse and pointwise stake measurements, we use a Monte Carlo simulation and a Stefan model (Eq. ) to assess a plausible range of ground-ice melt for a range of input parameter values as expected on the landform scale (probabilistic uncertainty estimate, Sect. ). The Stefan model has been widely used to simulate the freezing and thawing fronts in permafrost (e.g. ), including the Cold Regions Hydrological Model . We parameterise the cumulative heat flux from ground-ice melt on Murtèl Hali(t) = ∫0tQmdt′ = fiLmρiζ(t) as a function of the depth of the ground-ice table ζ(t) using a modified Stefan equation by that is appropriate for a two-layered stratigraphy (Fig. ) , 12∫0tQm(t′)dt′:=ΣtQm=fiLmρih12+2keffI(t)-I1Lmf2ρi, where the surface thawing index I(t) is defined as 13I(t):=∫0tλ52Tsdt′≈86400∑iλ‾52T‾s, and the thaw index of the ice-poor AL overburden I1 is defined as 14I1:=h12Lmf1ρi2keff. The factor λ5 ≤ 1 corrects for the sensible heat storage in the thawed layer and is a polynomial of the Stefan number Ste, λ5 = 1-0.16Ste+0.038Ste2 . The depth-averaged dimensionless Stefan number Ste is proportional to the ratio of sensible heat to latent heat absorbed during thawing, 15Ste:=CvT‾sLm〈f〉ρi, with the bulk volumetric heat capacity Cv [J m⁻³ °C⁻¹] of the (unfrozen, ice-free) AL (identical for both layers), the average surface temperature T‾s for the time t elapsed since onset of the thaw season, and the latent heat consumed by the melting of the ground ice Lm〈f〉ρi (different in each layer, and depth-averaged quantities are denoted by 〈⋅〉; details in ).

The empirical stage–discharge (hw–Q) relation (Eq. ) based on eight gaugings (Fig. , Table ) yields the fitted coefficients h0 = 780 mm (stage of the standing water), c1 = 7429, and c2 = 5.15 ± 0.753 (1σ uncertainties; hw in m, Qw in m³ s⁻¹). Because of the wide channel (plane-bed type stream morphology) in the slightly dipping forefield, the water level covered by gaugings varies by merely 9 cm that covers a discharge range from 3 L min⁻¹ (detection threshold) to 27.7 L s⁻¹ (Fig. ). The channel remained stable during the study period of August 2020–September 2022. Discharge estimates in the wet summer of 2021 rely on extrapolated stage–discharge relation data (often exceeding 27.7 L s⁻¹), while discharge in the dry summer of 2022 is mostly interpolated and better constrained (except the early snowmelt period and peak discharge of event water). Importantly, no stage measurements could be made beneath a snow cover; hence, snowmelt discharge before complete melt-out of the forefield is not gauged. We consider our stage–discharge relation and discharge estimate sufficient for our purpose of season-cumulative water balances and emphasise that our focus is on the low-discharge summer periods where the contribution from ground-ice melt is potentially largest. In the context of the hydrological significance of Murtèl rock glacier, the (reliably measured) zero-discharge estimates will be important.

Discharge in the small catchment is variable and shows a seasonal trend that decreases as snowmelt progresses (Figs. , ), superimposed by regular diurnal fluctuations related to radiative forcing/snowmelt. Total measured discharge was 160×103 m³ in summer 2021 (snowmelt and thaw season) and 97×103 m³ in summer 2022. After completion of the snowmelt, outflow is flashy where dry phases without measurable baseflow (⪅ 3 L min⁻¹) are interrupted by precipitation-fed discharge spikes (event water). The qualitative discharge pattern is similar in both summers. Field observations and additional EC measurements at the bedrock step downstream of the confluence (mixing calculations) suggest that the discharge of SW exceeds that of SE if SW is active. Water temperature of the outflow and the deep seep are stable and always at 0–1 °C.

The rainfall–streamflow relation from the threshold analysis (Fig. ) is calculated for the late-summer discharge after completion of the snowmelt in the entire catchment. For our purpose, only one observation is relevant: rainfall less than 9 mm d⁻¹ in most cases (8/10) does not generate measurable outflow at the gauging station (and this finding is independent of the quality of the stage–discharge relation). This agrees with field observations: total spring discharge below the detection threshold of ∼ 3 L min⁻¹ seeps into the ground along its way from the rock-glacier springs to the gauging station located ∼ 50 m below the rock-glacier front (“not measurable baseflow” refers to maximal discharge of ∼ 3 L min⁻¹). We observed rapid recession and drying out of the stilling pool at the gauging station after the discharge estimate in the morning of 24 August 2021 (Fig. c, Table ). We thus constrained the detection limit of ∼ 3 L min⁻¹ using the bucket method (discharge too small for dilution gaugings).

EC of the outflow seasonally increases from ∼ 50 to ∼ 150 µScm-1 (Figs. , ). Both main springs SE and SW show a qualitatively similar behaviour in both summers, despite different weather conditions: they transition from a snowmelt regime (daily oscillations) to a rain regime (rapid EC drop after onset of event discharge that stabilises). The lower-lying main spring SE (2624 m a.s.l.) has a lower EC than SW (2626 m a.s.l.) and appears more strongly buffered in terms of smaller daily oscillations and smaller seasonal shift. Still, the different weather conditions might show up in the late-season EC: the SE–SW EC difference is largest in autumn 2022 after the dry hydrological year of 2021–2022 (SWE and summer rainfalls below average; Fig. ), where the EC of SW reaches > 250 µScm-1. EC is in general anti-correlated with discharge from seasonal down to hourly timescales, suggesting dilution behaviour or longer water residence times: high EC at low discharge which is best seen during snowmelt, although the hourly EC evolution during late-summer rainfalls can be complex with a high-EC peak at the onset of event-water discharge (a timescale beyond the scope of this study). Point measurements of the seep S∗ show a strong enrichment up to 250–350 µScm-1 of the autumn seep water. The supra-permafrost water co-evolves with the outflow. Notable are the persistent EC difference between two nearby sampling spots in the same rock-glacier furrow: one next to the stake measurement spot and the other next to thermistor TK4/5 (Fig. ). Spatially varying EC speaks for channelised supra-permafrost flow as interpreted from tracer tests by .

We collected a total of 145 samples from different rock-glacier springs and seeps (referred to as outflow), supra-permafrost water, snowpack, and rainwater: 2 samples collected in 2020, 57 samples collected in 2021, and 86 samples collected in 2022 (Table ). All but a few rainfall samples are aligned on our local meteoric waterline (LMWL) given by δ2H=8.19δ18O+15.35 ‰ (R2 = 0.9919) (Fig. ). The δ18O–δ2H relationship of the outflow samples is δ2H=8.06δ18O+13.78 ‰ (R2 = 0.9969), with its slope of 8.1 similar to the local (LMWL, 8.2) and global meteoric waterline (GMWL, 8.0), suggests that the source waters of the outflow have undergone little if any evaporation. This finding is consistent with a sparsely vegetated coarse blocky landform with rapid infiltration , and it is consistent with the measured specific humidity gradients in the Murtèl AL : moisture for evaporation is (generally) drawn from a rain-fed reservoir in the shallow AL and not from the supra-permafrost water in the deep AL. Moisture transport in the AL is generally downwards, leading to condensation; upwards transport and evaporation from the deep AL occurs only episodically during droughts.

δ18O of the outflow and supra-permafrost water showed an enrichment during the thaw season from -16 ‰, also measured in the spring snowpack, to -10 ‰, levelling off in late summer to autumn, and repeatedly interrupted by short excursions towards isotopically heavier values of -10 ‰ that co-occur with the isotopically enriched rainfall (typically -10 ‰ to -6 ‰). δ18O was overall higher in summer 2022 and reached the plateau phase sooner (by July), likely reflecting a proportionally smaller amount of snowmelt in the catchment after the snow-poor winter of 2021–2022 (consistent with the EC pattern discussed above). We do not observe the seasonal late-summer isotopic depletion reported by . The two main springs SE and SW showed, overall, the same pattern, although SW showed seasonally somewhat more extreme values, i.e. isotopically lighter than SE during snowmelt and heavier in late summer. Discrepancies were smallest at high discharge during major rainfall periods. δ18O of the supra-permafrost water was always close to the outflow δ18O. Analogous to its EC, the δ18O value of the supra-permafrost water in the eastern stretch of the furrow is closer to the eastern main spring SE, while the one in the western stretch of the furrow is often closer to spring SW (if active). δ18O values of the rainwater varied considerably between -13 ‰ and -6 ‰ but were generally heavier than all other sampled waters. The denser 2022 data set shows the well-known seasonal pattern with maximum enrichment in July–August. The pattern agreed with 1994–2022 measurements from the nearby Global Network for Isotopes in Precipitation (GNIP) station in Pontresina (1724 m a.s.l.) with δ2H=8.095δ18O+9.53 ‰ (n = 321, R2 = 0.9943) (accessible via https://nucleus.iaea.org/wiser/, last access: 4 December 2023; ). δ18O values of the snowpack were from -15 ‰ to -19 ‰. Although snow δ18O is sensitive to the sampling timing , its δ18O does not exceed outflow δ18O and is meaningful as a qualitative end-member. The deuterium excess, in cases used as in indicator of multiple freeze–thaw cycles , shows no clear seasonal trend (Fig. ).

The ground ice is rarely accessible in coarse blocky landforms. Here, we present one of the few (to the best of our knowledge) direct observations of the seasonal evolution of AL ice in rock glaciers. The ground-ice table as observed in a rock-glacier furrow in the thaw seasons of 2022–2024 underwent seasonal build-up and melt of ∼ 70 cm within the coarse blocky AL, showing the seasonal build-up and melt of AL ice (Figs. , ). The seasonal release of water from melting ground ice in the coarse blocky AL is 200–300 L m⁻² over a thaw period of ∼ 100 d, corresponding to 250 mm water equivalent (w.e.) or a melt rate of 1–4 mm w.e. d⁻¹ (1–4 kg m⁻² d⁻¹). A thin (≤ 6 cm) mud layer covered the ground ice in the late summer of 2022. The mud cover loses its insulating effect under the continuous belowground thermal radiation emitted by the blocks .

Our observations indicate two different mechanisms of ground-ice build-up: (1) refreezing of snowmelt onto cold blocks (analogous to the formation of basal ice at the ground–snow interface) and (2) blowing in of snow with subsequent metamorphosis to ice. The “warming spikes”, the AL energy budget, and the ice coatings/icicles in sheltered cavities indicate the former, while the observation of ripe snow on top of the fresh ground ice in exposed cavities indicate the latter mechanism.

The Stefan model (Eq. ) describes the observed lowering of the ground-ice table (“Stefan AP53” in Fig. ) and relates it to a modelled ground-ice melt Qm. The effective thermal conductivity keff = 3 W m⁻¹ K⁻¹ is derived from the heat flux measurements , the thickness of the ice-poor AL overburden h1 = 3 m, and the ice content f1 = 0.01 is calibrated with the 2022 stake measurements. This 2022 parameter set also describes the 2023 ground-ice melt (Fig. c). The ground-ice melt derived from the stake measurements Qm is shown in Fig.  (“Ground ice melt AP53”) along with the estimate of the AL energy budget (dev_al, Sect. ). Systematic stake measurements were not performed in summer 2021. The modelled 2021 ground-ice melt qualitatively agrees with observations insofar as a nearby thermistor (TK4/5) remained embedded in ice in summer 2021 and was only released in July 2022 (Fig. b2, Tal in furrow shown in Figs. , ).

Given the sparse and pointwise observations on Murtèl and the few other published observations of seasonal ground-ice melt in coarse blocky landforms, we estimate the uncertainty with a probabilistic Monte Carlo approach. We make an educated guess of the value range of the input parameters. The Stefan melt parameterisation is most sensitive to (Eq. ), namely the effective thermal conductivity keff (2.0–3.5 W m⁻¹ K⁻¹, Fig. c; ), the AL overburden thickness h1 (1.5–4.0 m, Fig. d), and ice content f2 (0.2–0.8, Fig. e), assuming that these parameters are independent of each other and of the ground surface temperature Ts. Fifty thousand model runs with the 2021 and 2022 Ts forcing to represent two contrasting thaw seasons yield the outcome distribution of maximum thaw depth ζmax⁡ and the total amount of ground-ice melt (Fig. b, a). The expected amount of meltwater released in summer 2021 is 100–200 mm w.e. (150–350 mm w.e. in summer 2022), with a large range of overlapping values of 150–250 mm w.e.

We obtained consistent estimates of seasonal ground-ice melt from the AL energy budget (dev_al, Fig. ) and stake measurements (Qm, Figs. , ), although we estimated the AL energy budget dev_al beneath a broad ridge and made stake measurements Qm in a narrow furrow. End-of-thaw-season values (cumulative ice melt) are 160 mm w.e. in the thaw season of 2021 and 260 mm w.e. in 2022 (Table ). Although our pointwise observations cannot exclude some differential build-up/melt in furrows and ridges (a micro-topographic variability mentioned by , and ), the agreement suggests that our estimates of end-of-thaw-season ice storage changes are fairly representative over the landform and within an uncertainty of ±50 % (Fig. ). Smaller discrepancies in the sub-seasonal evolution of Qm and dev_al (Fig. ) likely arise from differences in the micro-topographic setting, duration of snow cover, debris texture, and AL thickness at the two measurement points. Our estimates of the ice melt agree with prior studies on Murtèl by and .

The stake measurements show no local AL thickening for the years 2021–2023 (Fig. ). The ground ice that melted during the thaw season was regenerated by refreezing and build-up in the following winter and spring. In the exceptionally cool and wet summer of 2021 with late snow melt-out (beginning of thaw season in July), even net build-up occurred at least locally in the rock-glacier furrow (frozen thermistor TK4/5; Fig. a). The surviving snow patches (Fig. ) tentatively support the net positive mass balance observed in 2021 but cannot provide conclusive evidence on the landform scale.

A net increase in belowground ice content during snowmelt, i.e. the conversion of snowmelt to ground ice, as observed on Murtèl is arguably most efficient on well-drained coarse blocky permafrost landforms. Such terrain, abundant in periglacial high-mountain areas, features a distinct seasonal chain of coupled heat and water–ice transformations co-controlled by the snow cover (illustrated in Fig. a: 1–3). In autumn–early winter, the permeable AL contains little water to freeze, enabling a rapid and pervasive ground cooling to large depths before the onset of an insulating snow cover (Fig. a: 1). In late winter and during spring snowmelt, whenever a warm and melting snowpack releases water into the subfreezing AL, the large cold content (sensible heat) is partly transformed into the build-up of new ground ice (latent heat) (Fig. a: 2; warming spikes in Fig. ). The timing of ground-ice build-up, from snowmelt in spring rather than by freezing soil water in autumn–early winter , is distinct from fine-grained material with a larger water retention capacity and is observed in other coarse blocky permafrost landforms (e.g. ). From a hydraulic perspective, the large and permeable pores provide ample storage space for the infiltrating snowmelt to refreeze and for in-blown snow to be trapped at depth without blocking the water infiltration/percolation pathways by the newly formed ground ice or basal ice at the ground–snow interface (but see also ). During the thaw season, the melting AL ice largely (∼ 80 %) absorbs the ground heat flux (Fig. a: 3; ). Numerical simulations by of the coupled heat and water/ice balance apply to Murtèl. They showed that alone this interplay between the subsurface water/ice balance and ground freezing/thawing leads to negative ground thermal anomalies (up to 2.2 °C in their modelling scenarios), even without air convection which is an additional efficient cooling mechanism common for such terrain . The regeneration of ground ice in the AL partly explains the climate robustness of coarse blocky landforms . If the lost ground ice is not regenerated, the permafrost landform is preconditioned towards irreversible degradation . Finally, permafrost conditions might be required in seasonally snow-covered terrain to keep winter ground temperatures well below the freezing point until the spring snowmelt, depending on the timing and insulation of the snow cover. This is shown by the “bottom temperature of snow cover” (BTS temperatures) , a winter-equilibrium ground surface temperature attained beneath a closed/insulating snow cover that is near or above the freezing point in permafrost-free (seasonally frozen) Alpine terrain. To sum up, ground-ice build-up by snowmelt refreezing relies on strong (preferentially convective) ground cooling and water supply into the subfreezing AL, in turn sensitive to debris texture, thermo-hydraulic properties, and weather/snow conditions.

In our warming climate, net build-up as possibly occurred in 2021 is the exception. Murtèl rock glacier is slowly degrading (AL thickening; ) and must be on a multi-year average releasing meltwater from the old permafrost core. Neither our stake measurements nor our energy balance measurements that span only 2 years can reliably separate melt from young AL ice from melt of old permafrost ice (or a possible transient layer in between; ). Inter-annual to decadal surface subsidence estimates from kinematic surveys and vertical borehole deformation revealed slow subsidence attributed to ground-ice melt in the over-saturated permafrost body. report 2–5 cm yr⁻¹ for the period 1987–1996. The most recent 2022–2023 geodetic measurements (Isabelle Gärtner-Roer, personal communication, 2024) yield 1 cm yr⁻¹ in the mid–frontal parts of the rock glacier (no significant net change in the surface elevation: the ice melt compensates compressive thickening; subsidence approximately equal to ice loss in massive ice). The amount of meltwater released from the permafrost core is an order of magnitude smaller than the seasonal ice turnover in the AL and negligible compared to precipitation, surface outflow, and even evaporation, agreeing with previous studies .