The study area consists of fractured rock of the Penninic Bündner schist formation within the Glockner Nappe, comprising calcareous mica schist, prasinite, amphibolite, phyllite, marble and serpentinite . The investigated rockwall is mainly composed of calcareous mica schist, traversed by a scaly serpentinite belt. The rock mass is characterised by a well-developed schistosity dipping in a north-northeast direction, parallel to the slope (∼ 45°), forming an open interface structure that provides pathways for water flow. In addition, optical borehole scanning indicates a high degree of fracturing in the upper meters, with joint openings of up to several centimetres (Fig. ), likely facilitated by stress release and intense frost weathering . The interface set is dominated by two main joint sets : K1, dipping steeply towards the west, and K2, dipping medium-steeply towards the southwest. Less pronounced sets include K3, dipping medium-steeply to flat to south-southeast, and K4, dipping steeply towards northwest. The interaction between the differently oriented joint sets with the schistosity forms cubic to rhomboidal rock blocks, which are predisposed to destabilisation under changing stress conditions. Numerous open fractures, also observed along the ERT profile , act as preferential pathways for water infiltration and are partly filled with fine-grained material that retains water within the discontinuities.

Electrical resistivity tomography is an indirect geophysical method widely used in permafrost research to characterise the subsurface properties, described for example in detail in . At the Kitzsteinhorn north-flank, the investigated ERT profile begins at the north terrace of the cable car station (3016 m a.s.l.) and extends 58 m down-slope (Fig. ). The measurement transect consisted of 30 stainless-steel expansion anchors (each 90 mm in length and 10 mm in diameter), permanently bolted into the rockwall at 2 m intervals. These bolts served as electrodes, offering good electrical contact and long-term durability. Compared to the approach described by , the set-up for automated ERT monitoring was improved through increased mechanical stabilisation of the connections between the take-outs and electrodes, as well as additional protection of the measurement cable from rockfall activity and snow avalanches by encasing it in multiple layers of HDPE and PE tubing.

The automated ERT data set was acquired using the terrameter LS measurement device housed in the summit station to shield it from harsh weather conditions and enable a continuous power supply. The instrument was connected to a laptop, allowing for remote control, monitoring, and immediate interaction in case of errors. As ERT measurements in permafrost bedrock often encounter high contact resistances and, consequently, a low signal-to-noise ratio, all measurements were conducted using the Wenner configuration to ensure maximal signal strength . Additionally, reciprocal measurements were performed to estimate measurement errors and subsequently used for data weighting and determining inversion parameters. Both Wenner and reciprocal measurements were repeated daily during daytime hours (08:00 am–04:00 pm CET) over one year, starting in April 2024.

To increase local ground stability, fifteen boreholes were drilled nearly horizontally (-3°) with a diameter of 150 mm below the cable car station. Each borehole was equipped with 25 m long corrosion-protected rock anchors with a diameter of 57 mm, whereby the upper 18 m section represents the free anchor length, and the lower 7 m section is friction-locked and grouted. The anchors were designed for a load capacity of 1.513 kN and were pretensioned to 600 kN during installation. However, within a few hours, the load decreased to 500 kN due to anchor relaxation and the mechanical response of the rock wall.

In the vicinity of the ERT profile, three anchor heads were instrumented with hydraulic anchor load cells (Glötzl KK 1600 A 75 VW 4, Fig. ) to monitor load at hourly intervals since December 2015. For this study, we analysed data from the two closest anchor load cells (A#9 and A#12 in Fig. ), located 5–15 m from the ERT profile, covering the automated ERT monitoring period from April 2024 to April 2025.

The load monitoring operates on a hydraulic measurement principle, where a vibrating wire measures pressure. This pressure is then converted into anchor loads based on individual laboratory calibration of the vibrating wires by the manufacturer and the well-defined force application surface of the load cell. The change in anchor loads is linked to variations in the mechanical stress regime of the rockwall, since anchor loading represents the relative strain between the activated rock mass and the free anchor length itself.

Rock temperature data are available to a depth of 30 m. The borehole mouth is located between electrode 22 (x=42 m) and 23 (x=44 m). The borehole, which was drilled perpendicular to the surface with a diameter of 90 mm at 2970 m a.s.l. (Fig. ), has been recording rock temperature since December 2015. For the purpose of this study, the borehole temperature data were analysed according to the temporal and spatial resolution of the automated ERT measurements. Specifically, datasets from thermistors at depths of 0.1, 1, 2, 3, 5 and 15 m, covering the period from April 2024 to April 2025, were considered. The former data series (2016–2023) of bedrock temperature over the entire measuring depth of 30 m is shown in . The innovative measuring principle and system is described in detail in . The active layer thickness was defined as the maximum depth of the 0 °C isotherm, derived from a spline interpolation of the borehole temperatures.

A vertical borehole (∅=90 mm, 20.5 m depth) was drilled close to the summit station (Fig. ) in October 2023. A GEOKON 4500S piezometer sensor (accuracy ±0.1 % from 0–7 bar) was installed at a depth of 16.85 m exactly at the stratigraphic boundary between calcareous mica schists and the underlaying scaly band of serpentinite, as determined by optical borehole scanning. To prevent water percolation along the borehole, the lower section of the borehole was filled with impermeable cement. The piezometric sensor was placed inside a bag of filter sand (∅=90 mm, ∼ 30 cm length) to create an isolated, water-permeable collection zone and positioned at the desired tip location. The upper part of the borehole was sealed similarly to the lower section using impermeable bentonite and cement. The measuring principle of the piezometer and data transmission is explained in . Here, we analysed the piezometric pressure data from April 2024 to April 2025, whereby the entire dataset was corrected for temperature and barometric effects.

Meteorological conditions were assessed using two nearby weather stations. Air temperature and snow depth were obtained from the Gletscherplateau station (2920 m a.s.l., ∼ 500 m distance), while precipitation data were taken from the Alpincenter station (2450 m a.s.l., < 2 km distance). In addition, webcam images taken near the Gletschershuttle (2928 m a.s.l.) were used to analyse the subsurface conditions at the rockwall.

The ERT dataset contains up to 135 measured data points per day for each Wenner configuration, provided that electrical coupling was sufficient and no electrodes were excluded before the respective measurement, resulting in a total of 20 148 measured data points over the entire observation period. Due to the size of the dataset, manual filtering procedures and quality controls of each individual observation are impractical, making automated filtering and inversion procedures essential for consistent data processing. With the growing number of automated ERT monitoring sites in recent years, various filtering approaches have been developed, differing in complexity and criteria for identifying and defining data outliers. Some filtering procedures rely on data point correlation over time by incorporating regularisation within the time-lapse inversion scheme e.g.,. However, as indicated by , rapid spatial changes in resistivity occurring within a few hours in the snowmelt period could be filtered out by these time constraints, potentially leading to over-filtering in our dataset.

Therefore, we only applied a technical filter before the datasets were inverted to remove physically implausible and low-reliable data points. Each individual quadrupole was evaluated according to the criteria proposed by , with a slightly higher minimum threshold for apparent resistivity to exclude unrealistic values typical of low-porosity bedrock. Specifically, data points were retained only if they fulfilled the following conditions: (i) Apparent resistivity values: ρa > 0.5 kΩm, (ii) injected current flow: I > 0.02 mA, and (iii) stacking variance within repeated quadrupole measurements (minimum: 2, maximum: 4) CV < 10 %. Quadrupole measurements that did not fulfil one or more of these criteria were excluded from further processing. Compared to other ERT filtering methods e.g.,, we did not implement additional filtering steps due to the specific characteristics of our datasets. During the summer months, resistivity values decreased over more than one order of magnitude . Applying a filter based on a fixed standard deviation factor across the dataset would have been overly strict for the summer measurements, excluding plausible data points, or insufficiently sensitive in winter, failing to identify anomalously high resistivity values. Applying a moving median filter along each depth level would exclude data points showing rapid spatial changes in resistivity values. However, these variations, attributed to water infiltration or freezing corridors along fractures , should be preserved in the dataset. Our conservative filtering approach, which is based solely on technical limitations, may not detect all poor-quality data points, particularly during freezing periods with poor galvanic contact. To address this, only datasets retaining more than 80 % of data points (n = 108) after technical filtering were inverted.

Before performing the inversion routines, we assessed the quality of the raw data set and accurately quantified the measurement error (noise) on the basis of the filtered data sets. The crossplot of normal (Rn) and reciprocal (Rr) measurements suggests a negligible occurrence of systematic and random errors (Fig. a). In previous studies, outliers were defined, for example, over a relative reciprocal error exceeding 10 % or 25 % . Due to technical constraints, mainly related to high-resistivity measurements at very low currents, reciprocal measurements could not fully cover the entire normal data set. Nevertheless, applying these thresholds to the available reciprocal pairs shows that only 100 data points (0.5 % of the total data) would be classified as outliers using the 25 % threshold, and 266 data points (1.4 %) using the 10 % threshold, confirming the overall high quality of the data set (Fig. b).

Precise error quantification is crucial to avoid misinterpretations of the inversion models . Overestimating the error can result in low-resolution, overly smoothed models, whereas underestimating may lead to overfitting and to the introduction of artefacts. In inversion processes, error models typically assume a normal distribution (Gaussian noise), where the reciprocal error (εrn) is described as a linear function of the mean resistance (Rrn) from normal and reciprocal readings : 1|εrn|=arn+brnRrn. To determine the error model parameters arn and brn, data points with a relative reciprocal error (ΔRrn/Rrn) exceeding 25 % or a resistance above the 99th percentile (Rrn > 19.193 Ω) were excluded. Following the approach of , the resistance range was logarithmically divided into equal-sized bins (n = 20). For each bin, the mean resistance and standard deviations were calculated and subsequently fitted to the linear model of Eq. () (Fig. c). The fitting yields arn = 0.9871 Ω (minimum absolute resistance error) and brn = 2.44 % (relative increase in error). The error can change over time-lapse observations e.g., which is pronounced by increased values during freezing periods (beginning and towards the end of the observation period, see Fig. d). However, previous ERT studies in permafrost environments have reported elevated data uncertainties e.g.. To account for this and address the sensitivity to high-resistivity conditions, we applied a fixed relative magnitude error of brn = 8 % within the inversion framework, which exceeds the empirically calculated data errors across the entire time series (Fig. d). We preferred a constant error level rather than scaling brn per seasonal phase, to avoid applying different degrees of smoothing to the individual datasets, which could introduce or suppress resistivity contrasts between the seasonal phases.

The ERT data sets were inverted using the Python-based software ResIPy (version 3.6.2; ), which interfaces with the finite-element forward and inverse modelling code R2. An unstructured triangular mesh was employed, whereby the mesh element size increases with distance from the electrodes and was generated using the meshing code Gmsh . The inversion follows a regularised, smoothness-constrained (Occam-type) least-squares scheme, minimising an objective function that combines the data misfit with a model roughness term, both evaluated in the L2 norm. The data were log-transformed and weighted according to the reciprocal error model (Eq. ). A time-lapse algorithm was applied, implementing regularisation relative to the previous data sets following the difference inversion approach of .

The ERT dataset was subdivided according to the defined seasonal phases I-V, with the first measurement of each phase used as background models in the inversion (see Sect. ). To account for the pronounced decrease in resistivity observed during the snowmelt period (phase II), this time period was further divided into two sub-periods to enable the development of appropriate reference models. The selected baseline dates and corresponding final RMS misfit were as follows: phase I – 17 April 2024 (RMS 1.01), phase IIa – 22 May 2024 (RMS 1.02) and 20 June 2024 (RMS 1.17), phase IIb – 31 July 2024 (RMS 1.01), and phase IIIa – 14 September 2024 (RMS 1.00).

The mean anchor load was calculated for each defined seasonal phase, and one representative ERT data set with minimal outlier filtering was inverted (Fig. a, b). To better relate the ERT models to the prevailing environmental conditions, the inversion dates were aligned with the corresponding webcam images of the rockwall shown in Fig. b.

During stable freezing conditions (phase I), anchor loads reached a mean value of 562 kN, while the mean value of the resistivities of the corresponding model was 170 kΩm. The ERT model showed predominantly high resistivity values (> 20 kΩm), with isolated low-resistivity zones. In phase II (snow melt & subsurface warming), mean anchor load decreased to 518 kN and mean resistivity dropped to 6 kΩm. The tomogram showed extensive low-resistivity zones (< 10 kΩm) within the upper layer (∼ 2–3 m depth) from the central to lower section of the profile (x = 24–58 m). During late summer (phase III), mean anchor loads (493 kN) and resistivity (2.5 kΩm) dropped to their seasonal minimum. Almost the entire ERT model exhibited values below 10 kΩm, except for a localised zone near the surface (∼ 2.5 m depth) in the upper part of the profile (x = 0–18 m). This anomaly might be attributed to limited water infiltration, potentially due to shielding by the summit station and the presence of relatively intact rock with fewer discontinuities compared to the lower section of the profile. With the onset of superficial cooling (phase IV), resistivity in this near-surface zone and in a rock outcrop adjacent to borehole B2 (x = 40–46 m) further decreased. Despite this, mean resistivity remained low at 3.4 kΩm, while anchor loads moderately increased to an average of 533 kN. During the deep freezing period (phase V), mean anchor load peaked at 574 kN.

To assess whether the stress dynamics observed from April 2024 to April 2025 reflect typical seasonal patterns, long-term anchor load trends are shown in Fig. . Across all observation years, anchor loads exhibit a distinct seasonal pattern: high stress levels during phase I, followed by a pronounced decrease throughout phases II and III, and minimum loads typically occurring during late summer (phase IV). Subsequently, loads increase again during phase V. This seasonal variation was evident for both monitored anchors #9 and #12. The overall long-term decline in maximum and minimum anchor loads is assumed to be attributed to progressive deepening of the active layer over the years, where ice melt at the base of the active layer acts as the primary driver .

To examine the interrelation between electrical resistivity, anchor loads, and rock temperature across the defined seasonal phases, these parameters were analysed at high temporal resolution (daily scale), as shown in Fig. . For comparison of relative changes, we focus on apparent resistivity from deep-reaching signals (DOI 5–9; electrode distances 10–18 m), as the rock mass mechanically affected by the anchors reached an effective perpendicular depth of 13 m.

During phase I (stable freezing), high resistivity values (> 19 kΩm, indicative for frozen conditions; ) and high anchor loads correspond well with borehole temperature data, collectively indicating a rock mass predominantly controlled by frozen conditions. All monitored parameters (resistivity, anchor load, and freezing front depth) remained largely stable throughout this time period.

A sharp decline in resistivity to below 10 kΩm occurred during snow melt and subsurface warming (phase II, Fig. a, b). The continuous apparent resistivity gradient (DOI 3–9) suggests that individual precipitation events had negligible effects, contrasting with automated ERT observations at the Schilthorn in the Swiss Alps by . The onset of decreasing anchor loads and resistivity coincided with the initial development of the active layer (Fig. d).

During summer (phase III), the ongoing subsurface warming maintained low resistivity values, while anchor loads continued to decline in response to progressive active layer thickening (Fig. ). While resistivity primarily reflected the presence of water-filled discontinuities and multilayered flow near the surface, melting of ice in deep-reaching fractures proceeded more slowly due to latent heat consumption, resulting in delayed anchor load responses. At the point of maximum active layer thickness (transition between phases III–IV), cleft ice melt is likely most advanced, coinciding with minimum (stable) anchor loads.

With persistently negative air temperature and the onset of a snow cover (phase IV), superficial ground cooling initiated a slight increase in resistivity observed in the upper layers (DOI 1–2, Fig. b). Freezing of water in deep-reaching fractures likely commenced in the subsequent phase V (deep freezing), as both resistivity and anchor loads increased with the downward propagation of the -0.5 °C isotherm (freezing front).

Figure  shows scatter plots of electrical resistivity versus borehole temperature for selected depths ranging from 0.5 to 4.5 m. Borehole temperature data were interpolated at discrete depth levels (0.5, 1.5, 2.5, 3.5, and 4.5 m) using adjacent thermistor measurements. Corresponding median resistivity values were extracted from depth-specific zones of interest, defined as rectangular areas with a vertical extent of 1 m and a horizontal width equivalent to one electrode spacing, bounded laterally by the adjacent electrodes of the borehole (E22 and E23, Fig. ). Between 0.5 and 3.5 m depth, a seasonal pattern of freezing and thawing was evident, whereas at 4.5 m depth, the temperature remained below 0 °C throughout the entire observation period despite notable variations in resistivity over more than an order of magnitude. During the thawing season (phase II and III), pressurised water flow with hydraulic heads of up to 1.2 bar was observed (Fig. a). The observed decrease in resistivity at temperatures down to approximately -2 °C at depths of 2.5–4.5 m may result from the freezing point depression of water and the opening of hydraulic fractures , which allows fluid water flow to persist below the nominal freezing point. This suggests that substantial amounts of liquid water may be present even under sub-zero conditions, significantly lowering electrical resistivity despite nearly stable temperatures. Such pressurised, water-saturated discontinuities explain the pronounced discrepancies between field and laboratory temperature–resistivity relations during snowmelt and thaw periods (phases II and III, Fig. ). Laboratory experiments by , conducted on seven saturated core samples with a pore water conductivity of 0.058 ± 0.002 S m⁻¹, comparable to local snowmelt water (0.014 S m⁻¹), captured the temperature–resistivity relation under both frozen and unfrozen conditions but did not replicate the influence of transient water flow in discontinuities observed in the field ERT measurements.

At shallower depths (0.5–1.5 m), a hysteresis loop in the temperature-resistivity relation was observed (Fig. ), likely due to increased contact resistance during the initial surface cooling process and differing electrical behaviours between the freezing and thawing phases . During the onset of freezing (phase IV), resistivity values remained low at first, as liquid water was still present in the pore spaces. The transition from liquid to frozen pore water increases the ion concentration in the remaining liquid phase, lowering the freezing point and enhancing electrolytic conduction within the pore space. Once the pore spaces were completely frozen, as observed in phase I (stable freezing), resistivity values increased gradually as temperatures decreased. Despite this hysteresis in phase IV (deep freezing), a generally good match between field and laboratory data was found in the upper layers (0.5–1.5 m), as shown in Fig. .

The discrepancy in phase IV is likely due to electrode placement on a disintegrated rock outcrop (Fig. ), where near-surface rock is subject to more rapid drying and freezing. This interpretation is supported by a comparison of apparent resistivity values involving electrodes E22/E23 near the borehole and equivalent electrode spacing at the profile centre (Fig. a), which shows substantial deviations during superficial cooling (phase IV). Thus, near-surface resistivity appears to be more strongly influenced by rapid changes in conditions (e.g. saturation, drying, or freezing) associated with local rock surface heterogeneity, while rock temperature responses are delayed and moderated due to thermal damping.

emphasised that the resistivity gradient between the upper (active) and lower layer is a key indicator for identifying the presence and temporal evolution of permafrost. In our study, the comparison of resistivity values at 7 m depth (borehole temperature confirmed a permafrost body, Fig. a) and 1.5 m depth (seasonally unfrozen) further supports the hypothesis of pronounced water flow in discontinuities even in greater depths (Fig. b). When the active layer remained shallower than 1.5 m, resistivity at 7 m depth consistently exceeded 19 kΩm, indicating frozen conditions according to the laboratory calibrations of . However, as the active layer deepened beyond 1.5 m during the thawing season, resistivity values in the deeper permafrost zone dropped by more than two orders of magnitude and remained at low levels even when superficial cooling was initiated.

The combined approach of automated ERT, anchor load, and piezometric monitoring revealed seasonal phases of high hydrostatic and cryostatic pressures, which contribute to the conditioning of planes of weakness and subsequent rock instabilities. High hydrostatic pressures increase the effective stress on the rock mass, potentially promoting fracture displacement or intensified rockfall activity after rainfall events . Additionally, percolating water also transports heat more efficiently from the surface to deeper layers than conductive heat transfer . This non-conductive heat transport can lead to rapid changes in subsurface temperatures, as indicated by thermal anomalies in borehole temperature , potentially forming thaw corridors, accelerating permafrost warming and ice erosion in fractures. When the infiltrating water reaches frozen contact surface, initial ice aggregation can occur, releasing latent heat and thereby rapidly warming the surrounding cold bedrock .

During snow melting and subsurface warming (phase II), increased pore water content, water-filled discontinuities and multilayered water flow along schistosity reduced the median resistivity values rapidly by more than one order of magnitude from 140 to 3.2 kΩm (Fig. ) and across all investigated depths (Fig. a). The pattern of enhanced subsurface electrical conductivity during snowmelt is also reported in other automated monitoring studies in similar settings . As borehole temperature data indicate that subsurface conditions exceeding 4.5 m depth remained frozen throughout snow melting and subsurface warming (phase II), snow melt water is likely perched above ice-filled fissures (Fig. a), which exhibit lower hydraulic permeability than unfrozen ones . As a consequence, the pressurised water flow can mechanically widen fractures , which increases hydraulic conductivity and thus enhances the flux per unit drop in hydraulic head (Q/Δh). Based on Darcy's law, laminar flow through a planar and equally spaced fracture, while accounting for surface roughness (f) and a constant for the flow field geometry (C), is proportional to the cube of the fracture aperture (2b), and described by 2QΔh=C(2b)3f. The onset of the decline in resistivity values coincided with a reduction in anchor load (Fig. ), the latter being indicative of stress redistribution within the rock mass. This stress response was likely triggered by the initial development and subsequent deepening of the active layer (Fig. ) and the associated melting of segregated pore and fracture ice .

During this time window of snow melting and subsurface warming (phase II), shear strength may decrease to the point where it is exceeded by shear forces, potentially leading to slope failures. Recent examples include the collapse of a rock pillar on the Matterhorn , a rockslide at the Bliggspitze , and a rock slope failure at the Piz Scerscen and at the Fluchthorn .

The initiation of autumnal superficial cooling (phase IV) is reflected by increasing electrical resistivity values in the shallow subsurface (DOI 1 and 2; Figs. , ). Although rock temperatures dropped below 0 °C from mid-October onward (Fig. a), resistivity values in deeper layers and anchor loads remained stable at low levels (Fig. ), suggesting that the transition from water- to ice-filled discontinuities at depth and refreezing of the water at the base of the active layer had not yet initiated (Fig. b). This can be explained by a depression of the freezing point below 0 °C, caused by rock properties, salinity, or pore pressure . Therefore, a rise in hydrostatic pressure is prevented by the water storage capacity of the snow cover , and cryostatic pressure remains low due to the absence of ice-crystal formation within cracks and pores, while the cooling increases shear resistance of the rock mass .

With the ongoing cooling of the subsurface in winter (phase V), the phase transition from water to ice in discontinuities is initiated, releasing latent heat, while cryostatic pressure develops through volumetric expansion and ice segregation. The density change and resulting volume increase (∼ 9 %) during the transition from liquid pore water to ice can theoretically generate ice-induced pressures of up to 207 MPa , sufficient to fracture rock. However, the development of high ice-induced pressures, which requires fully water-saturated conditions and rapid, spatially uniform freezing , is unlikely at the investigated flank, where freezing proceeds only from one side due to the rockwall geometry and often occurs in unconfined spaces.

In contrast, crystallisation pressure might develop during this time window, as water migration through the porous rock is promoted by the freezing of the water within cracks and pores . The slow freezing rates and sustained subzero rock temperatures observed in winter (Fig. a) fulfil the conditions needed for ice segregation . Cooling at both the rock surface and the base of the active layer (Fig. a) indicates bidirectional freezing, which promotes the flow of water toward the permafrost table and the rock surface, favouring the formation of segregated ice.

Laboratory tests on metamorphic rocks, including mica schist samples from the Austrian Alps, indicate that the upper thermal limit for ice segregation is close to the pore freezing point , provided that temperature-gradient-driven suction can develop and sufficient liquid water is available. Theoretical models and field measurements suggest a lower limit of the ice segregation window between -15 and -6 °C, which depends strongly on site-specific conditions and rock type . As the thermal regime of the Kitzsteinhorn north flank lies within this range during the deep freezing in winter (Fig. a), ice segregation is likely an effective process at the investigated site. Water supply for ice segregation is probably governed by discontinuities, as the effective porosity of the local calcareous mica schist is low (0.3 %–0.4 %; ). Optical scanning of borehole B1 revealed clefts with apertures up to 71 mm (Fig. ), which could act as preferential pathways for rapid water movement.

According to , the stress on the rock mass caused by internal ice pressure pi can be described by a stress intensity factor KI, depending on crack length c and assuming a narrow crack width w with 3KI=4cπpi,w≪c whereby ice lenses must grow spatially independently and uniformly, governed only by internal ice pressure as well as the solid, elastic–brittle properties of the rock mass. At the investigated rockwall, increasing anchor loads (Fig. ) provided evidence for rising stresses on the rock mass induced by ice segregation (Fig. b). This process was triggered by rockwall cooling, indicated by the downward shift of the -0.5 °C isotherm and supported by the increase in resistivity values observed with larger electrode spacings (DOI 5–9, Fig. ). While the shear strength of rock, ice, and rock–ice contacts increases with cooling, shear forces induced by cryostatic pressure might rise even faster than the shear resistances, potentially preparing rock instabilities .

The bulk electrical resistivity of water-saturated rock reflects several conduction mechanisms: electrolytic conduction through the pore fluid, electronic conduction through the rock matrix, and surface conduction along mineral grain surfaces e.g.. Beyond these intrinsic rock and pore-water properties, the bulk resistivity of a fractured rockwall is further governed by the distribution and orientation of open cracks and fractures .

In our case, pressurised water flow through discontinuities plays a decisive role by reducing the overall electrical resistance, which is further supported by the hydraulic opening of fractures. This mechanism processes lowered resistivity during snow melt infiltration, and in combination with the contrasting resistivity responses during freezing and thawing, produced characteristic hysteresis cycles (Fig. ); a phenomenon also reported from the Schilthorn by . In contrast to the homogeneous lithological conditions and less jointed rock mass investigated by inside a tunnel, our dataset indicates a stronger influence of heterogeneous rock conditions such as disintegrated blocks, which result in differential behaviour of freezing and drying of the rock subsurface. Despite these challenges of transferring laboratory-scale results to field observations, our dataset shows generally good agreement between the bilinear temperature–resistivity relation derived in the laboratory experiments and the field data, particularly in the upper layer (Fig. ).

The maintenance of an automated ERT monitoring system in steep, permafrost-affected rockwalls requires substantial effort due to harsh weather conditions and natural hazards such as rockfalls, avalanches, and lightning strikes. These factors make the system time-consuming to maintain and require ongoing repairs to ensure high-quality data acquisition. To the author's knowledge, similar automated ERT applications in steep permafrost rockwalls have only been reported from two case studies: the same profile at the Kitzsteinhorn and at the Aiguille du Midi , both of which reported significant data gaps. In contrast, damages to cables in the present study were effectively mitigated through protective tubing (e.g. HDPE or PE), short-distance fixation to the rock surface, and careful alignment of the profile along a stable rock rib. In addition, manually disconnecting the cables from the measuring instruments during lightning storms prevented severe damage to the equipment.

Beyond interpreting ERT results, a key aim of this study is to assess the benefits of automated ERT monitoring in light of the associated technical challenges, operational demands, and maintenance efforts. While manual ERT surveys generally yield lower uncertainties due to individual electrode checks, targeted improvement of contact resistance, manual outlier filtering, and optimised inversion settings, the temporal consistency of resistivity values in the time series (Fig. ) and the overall low deviation between normal and reciprocal measurements (Fig. ) offer promising reliability. Consequently, continuous ERT measurements with short time intervals should prioritise the detection of transient phenomena such as infiltration of snowmelt water in spring (phase II), active layer deepening (phase III), and the onset of freezing in autumn (phase IV), all of which are key for understanding coupled thermo-hydro-mechanical interactions rather than focusing on analysing individual datasets. While most long-term assessments of alpine permafrost rely solely on annual end-of-summer measurements without direct temperature borehole validation e.g., our ERT data exhibit stable median electrical apparent resistivity values during late summer (Fig. ), confirming the general feasibility of this approach. Nevertheless, identifying the optimal time window to capture extreme thawing-freezing events or maximum active layer thickness remains a challenge , although such periods are crucial for accurately resolving long-term permafrost evolution.

As mentioned above, automated ERT data are more susceptible to uncertainties arising from high contact resistance, random data errors and model-data misfits. While removed data points affected by seasonal changes in water and ice content to minimise systematic errors, our study explicitly aims to capture such changes driven by seasonal environmental forcing (Fig. a). Therefore, instead of filtering these temporal variations, we only applied a technical filter for identifying data outliers and used the time-lapse inversion approach of , which effectively mitigated systematic errors and coherent data noise while enabling the detection of small temporal resistivity changes . Standard reliability assessment methods for manual ERT surveys, such as resolution e.g and sensitivity matrices e.g. or the DOI index , are typically not applied for large datasets of automated ERT studies e.g.,. However, to reduce uncertainties associated with inversion artefacts, incorporating complementary observations and previously acquired geophysical data can be beneficial. Borehole temperature records showed a generally good match between active layer thickness and resistivity data during seasons without pronounced pressurised water flow (Fig. ), indicated by a consistent temperature–resistivity relation observed in field and laboratory experiments (Fig. ). Previous ERT surveys with single-inversion routines in 2013 and 2023 revealed a similar seasonal pattern of drastic resistivity decrease in summer. The similarity between measured and inverted resistivity trends (Fig. a) further supports the minor influence of inversion artefacts relative to the significant temporal variations in the time-series. Other automated ERT studies in alpine regions have reported seasonal variability in electrode contact resistances e.g., which can help identify and filter unreliable data. While technical constraints in our system prevented electrode contact resistance checks before each measurement, earlier ERT surveys at the investigated rockwall indicated acceptable, low contact resistances (< 190 kΩ) in summer .

While soil moisture information can support the quantification of ground ice when combined with ERT measurements in coarse-debris layers e.g.,, new approaches are required to characterise hydrological conditions at greater depths (> 10 m), particularly within fractured permafrost bedrock. Piezometric pressure measurements indicated that during late spring and summer snowmelt, water-saturated conditions are not limited to near-surface layers; instead, liquid water is capable of percolating into deep-reaching fractures, resulting in water level build-up of up to 12 m (Fig. a). These hydrological dynamics are essential to interpret the substantial decrease in electrical resistivity observed across the entire depth range of ERT measurements (∼ 10 m, Figs. b, ), while borehole temperatures still indicated permafrost conditions below a depth of 4.5 m (Fig. a). Water-saturated conditions with additional hydraulic opening of fractures might trap electrical current, resulting in attenuated current penetration into deeper subsurface layers.