Measurements took place in the Kammstollen, a private tunnel beneath the E-W oriented ridge of Zugspitze (2962 m a.s.l., Fig. a–b), located at the German-Austrian border. This mountain is part of the Northern Calcareous Alps, and its summit is composed of Upper Triassic carbonate rocks, mainly massive limestones of the Wetterstein Formation . The thick-bedded, locally dolomitized rocks commonly exhibit fracturing and karstification due to dissolution, which are frequent and well documented on the nearby Plateau . Minor dolomite and occasional marl intercalations reflect depositional variability, while brecciated zones up to 1 m thickness are present in the summit area . A prominent fault zone extends from the investigation area up to the summit , but overall, the steep summit cliffs consist of comparatively pure, competent carbonate rocks. The study area is accessible year-round thanks to the infrastructure and logistical support of the research station, Scheefernerhaus (UFS), which is directly connected to the tunnel.

Long-term meteorological records on the summit by the German Meteorological Service (DWD) are available since 1900 (Fig. c). Air temperature increased drastically in the last 30 years, with mean temperatures in the last decade (2013–2022) reaching -3.3 °C, which is 1.5 °C warmer than the reference period 1961–1990. Additionally, 2011, 2020, 2022, and 2024 were the warmest years on record. Mean annual precipitation is more than 2500 mm of which 80 % is snowfall, but the steep north walls remain snow-free for most of the winter. The hydrological behavior of clefts at this location has been thoroughly measured for many years and interpreted by .

The earliest records of permafrost on the Zugspitze date to the construction of its infrastructure . In recent years, 3D modeling has confirmed the possible presence of permafrost on the shaded and steep north side of the SW-NE ridge, which includes the site of this study, beginning at approximately 2350 m a.s.l., but suggests a rather warm permafrost . Based on this modeling, one N-S borehole was drilled in 2007 by the Bavarian Environmental Agency (LfU) about 30 m under the summit to document permafrost presence and evolution under the summit cable car, at a distance of 700 m from our field site . These measurements support the modeling, showing average core temperatures between -1.2 and -0.7 °C, with a warming trend of 0.4 K in the last ten years. Modeling by forecasts the disappearance of this lens by 2070. conducted the first ERT monitoring in the tunnel in 2007, setting the basis for this study. The tunnel comprises a main tunnel (MT) and a side tunnel (ST). Ice accumulated across the entire ground of the side tunnel and in parts of the main tunnel, reaching a depth of up to 50 cm in some locations. Recently conducted intense geophysical measurements and mechanical modeling to monitor an ice-controlled instability on the ridge 400 m from the tunnel. This location showed strongly degrading permafrost due to the thermal influence of the southern slope on a thin ridge. Other geophysical measurements in the area utilized passive seismic methods to document permafrost degradation and relative gravimetry to detect mass changes associated with hydrology . Finally, documented the dramatic recession of the three glaciers at the Zugspitze, which have been shrinking since the 1980s and could completely disappear within the next decade.

The setup of our study is based on the feasibility study conducted in 2007 and 2008 by and remained constant to obtain comparable measurements. Monthly monitoring of electrical resistivity (ρ) commenced in 2014 and continues to this day. Measures were undertaken along 6 transects, with electrode distances of 4.6 and 1.5 m, using Wenner and Schlumberger arrays, for a total of 1310 quadrupoles. The monitoring settings are presented in the Supplement (Fig. S1), and a detailed description of the procedures is provided in . Resistivity data were primarily collected using an ABEM Terrameter LS, with alternating use of an ABEM SAS 1000 and SAS 300C before 2018. Contact resistance was improved by spraying the electrodes with saltwater shortly before the measurements, after our tests showed it did not affect the results. Electrodes that still had too high contact resistivity, mostly close to the active layer because of ice on or around the electrodes, were eventually excluded from the measurement. Problematic electrodes with rust or poor coupling have been consistently replaced. The final number of measured quadrupoles reached on average 90 % of the total, but was constantly above this average in the last 5 years (Fig. d). Data were collected using a stacking of 2, increased to 4 in case of a measurement error above 1 %. Between 2017 and 2018, data availability and quality were limited due to repeated cable failures; therefore, 14 measurements from this period have been excluded from this analysis.

Hourly monitoring of rock temperature (T) follows with updates and extensions as listed hereafter and in Table . Old loggers of Type UTL-1 were replaced with new Geoprecision loggers, offering higher accuracy and reliability. New loggers were installed along the main tunnel at a depth of 40 cm for rock and air temperature (MT and AT). One logger has been installed outside to measure rock surface temperature (RST) at 15 cm depth. The ST, AT, and RST loggers (Geoprecision) were calibrated at 0 °C by the manufacturer using a high-precision water triple-point cell, reaching a temperature stability of ±0.005 °C. This process ensures high accuracy and reproducibility in calibrating the measurement system. The iButton loggers (MT) were calibrated in an ice bath on multiple occasions throughout the study to account for drift effects. The analysis presented here focuses on the period after 2018, where complete data have been recorded. Data availability for temperature loggers is presented in Fig. d.

To analyze temperature dynamics, we linearly interpolated the four temperature measurements along the side tunnel (transect A–A') from 2019 to 2023, treating them as a borehole (Figs.  and ).

Successively, T-ρ field calibration (Figs. , , and S6) was conducted by combining these temperature values with single apparent resistivity values from the closest quadrupole to each temperature logger (within a 0.3 m radius). Logger ST-5, located at a depth of 5 m below the surface, was excluded from the calibration because resistivity measurements at this location were only available during summer, resulting in an insufficient dataset for robust analysis. To validate the results, we analyzed the probability density function of the apparent resistivity before inversion (Fig. ).

After field work, each database was imported into MATLAB. Winter measurements from December to June produced, on average, 4 negative resistivity measurements, which were excluded, while summer measurements showed none. Data were filtered using the coefficient of variability (CV, or VAR (%) according to ABEM) of the resistance, i.e., the standard deviation divided by the mean, as computed by the measuring instrument, to remove systematic errors. The threshold for the CV was set to 1 %, resulting in an average exclusion of 6 data points, mostly concentrated in the winter months.

For the ERT inversion, we utilized an updated version of the software CRTomo . This inversion algorithm fits a model f(m)i to the data di based on an adequate data error description ϵi, instead of just minimizing the data misfit as commonly done. Data fit is achieved when the data misfit ϵRMS (root mean square) value is close to one, which indicates that the model reproduces the data within the expected error margins on average, as shown in Eq. () 1ϵRMS=1N∑iNdi-f(m)iϵi2.

Following , we performed independent inversions without applying a time-lapse approach. Because time-lapse inversion can significantly influence results , we aimed to minimize temporal constraints, such as those introduced by prior or initial inversions, that may bias the results. We argue that this approach enhances the robustness of our results, as consistent trends emerge without temporal regularization.

Further on, the error model for the inversion was updated with new measurements. Image quality is highly dependent on the error estimates in the inversion: an overestimation smooths the image, reducing resolution, while an underestimation leads to artifacts . The random error (e) can be quantified using the difference between normal (RN) and reciprocal (RR) resistivity measurements, although this is only a measure of precision and not accuracy and does not represent the total experimental error. The error can be modeled as a linear function of the mean resistance R (Eq. ), depending on an absolute (a) and a relative (b) term 2|RN-RR|=e=a+bR. To solve this, expresses R as a function of e, removes obvious outliers (e>10%R), and obtains a and b from the envelope of all remaining measurements. instead rejected only values of e>100%R and then divided the measurements into logarithmically equally sized bins. The standard deviations of the reciprocal errors in each bin were least-square fitted with the linear error model (Eq. ). adapted the approach of to fit the discrepancies between readings acquired at different times, proposing data error estimates for time-lapse measurements.

After updating the used error estimates, we computed a new inversion for all measurements, including those from 2007 published in , and analyzed temporal changes by classifying different areas of the tomogram, such as the active layer and permafrost core. Some authors manually define an area of interest , while others choose an automated algorithm . Here, the inversion results were clustered using the MATLAB k-means algorithm. The function analyzed all tomograms from 2014 to 2023 and computed, for each grid cell, the squared Euclidean distance to the cluster centroids (the cluster means). Cells were then categorized into the nearest centroid. The number of clusters k is defined by the user, and assignment is mutually exclusive. This approach converges to a local minimum, though lower minima may exist. In practice, the global minimum is approximated by running multiple replicates with random starting points. We used five cluster classes, based on the elbow method , which evaluates the sum of squared distances between samples and their nearest cluster center as a function of the number of clusters. The optimal number is identified at the elbow of the resulting curve.

The temperature distribution in the tunnels indicates permafrost presence in transect A–A' (Fig. ). Due to repeated measurement gaps before 2019, it is not possible to determine longer temperature trends. Over the last five years, the increase in active layer thickness (ALT) from 8.5 to 10 m, combined with the overall rise in annual maximum temperatures, indicates permafrost degradation. The 10 m logger has now reached 0 °C and maintained this level for more than 2 months in summer 2023. It is likely to overcome this threshold in the next few years. No permafrost is recorded in transect B–B' (Fig. S2, Supplement), while seasonally frozen rock is recorded between electrodes E30 and E39. All loggers in the main tunnel recorded their maximum temperature in 2023, with a more pronounced increase in the area to the right of the side tunnel, specifically electrodes E32 to E41. Outside the tunnel, the mean annual rock surface temperature of the steep bedrock facing north is -1.7 °C (RST-Logger, Fig. c).

Ice on the ground of the main tunnel decreased between 2019 and 2021, especially in the area of the permafrost lens between electrodes E31 and E38 (Fig. S5). This decrease aligns with the temperature increases recorded by the loggers in the main tunnel. As documented by , large quantities of water from snowmelt are recorded in June and July in the northern part of the main tunnel (electrodes E1 to E20, x=200 to 275 m). In the side tunnel, water accumulation on the ground is also common, but only at the end of summer and between the ST-5 and the north face. This water results from melting surface ice on the tunnel walls and from ice-filled fractures within the tunnel.

The correlation between rock temperature inside the tunnel and air temperature outside is clear (Fig. d), with correlation coefficients between 0.77 for the ST-5 logger at 5 m depth and 0.4 for the ST-20 logger at 20 m depth. As expected, with increasing depth, the time lag grows, reaching up to 75 d at 10–15 m. This represents the time required for the thermal signal to propagate to the permafrost core within the bedrock. The correlation between air temperatures inside and outside the tunnel is strong at electrode 30 (x=143 m), near one of the tunnel exits, but becomes absent for deeper loggers (Fig. e), indicating the influence of additional processes at greater depths (e.g., solar radiation or snow cover from the south side slope). A comparison of ST-Loggers with the LfU-borehole , which is installed on the summit, ca. 150–200 m higher than our site, shows that ST-5 and ST-10 are quite similar to the LfU-borehole measurements in summer. At the same time, winter temperatures are much lower in the LfU-borehole (Figs. S3 and S4).

The newly collected thermal information, together with the ERT measurements, allows us to validate the temperature-resistivity (T-ρ) laboratory calibration by , which was often used to interpret ERT field measurements so far e.g.,. Figure  shows that, despite differences in resistivity ranges, there is a high correlation between temperature and resistivity.

Resistivity varies from 30 to 120 kΩm for ST-10, from 15 to 25 kΩm for ST-15, and from 40 to 130 kΩm for ST-20. These differences lead to a strong spreading of data when plotting all the values together (Fig. a). Still, when analyzing each temperature logger separately, the correlation appears clear (Fig. b and Fig. S7b–e). This suggests that the relation between resistivity and temperature can not be assumed to be unique for this transect. Logger ST-5 records only a few values below 0 °C: during the cold months, a thick ice cover in the tunnel limits electrode contact, whereas in the warm months, positive temperatures are measured. Therefore, ST-5 will be excluded from our analysis. The other loggers show a good linear fit: the best is at ST-10 (R2=0.84), whereas at ST-20 the fit is poor (R2=0.41). Resistivities decrease over time, possibly due to reduced ice content; therefore, an annual analysis could yield a better fit. Generally, most recorded data are close to the freezing point, which should be taken into consideration. Supercooling effects and similar phenomena from the calibration are not detectable in the field calibration due to the absence of freeze-thaw cycles in the permafrost core.

The freezing point at 29 kΩm agrees with values from the laboratory, but in the field, resistivity increases at much stronger rates than laboratory experiments can reproduce. Field calibration results for ST-10 and ST-20 can be resumed in the “average” Eq. (). 3ρ=(-106±49)T+29 ST-15 shows much lower resistivity values than the other locations and behaves similarly to the laboratory calibration. This could depend on the degree of fracturing at this location.

To explain the big discrepancies in the field calibration, we analyze the rock that composes the tunnel. While the majority is made of compact rock (material A in Fig. ), short parts of the tunnel are highly fractured due to the fault zone (material B in Fig. ). As suggested by , we compare the probability density function of the apparent resistivity of all the measurements in the permafrost area (side tunnel). The ERT transect on the right (Fig. d) presents only one peak, while on the left (Fig. c) two peaks are visible. Supposing that each peak relates to a different rock type, this indicates the presence of two materials in the left area, as confirmed by field mapping (Fig. a). The probability density curves for material A on the right side of the tunnel clearly show variations with time towards lower values. At the same time, smaller changes are recorded on the left for material A and no changes for material B. A combination of both materials would likely produce results similar to material B, as current follows the easiest path of lower resistivity; however, this should be further investigated.

During previous research at this site, analyzed 3000 reciprocal measurements according to and obtained values of a=48 Ω and b=8%. We recorded reciprocal measurements for four months in summer and autumn to validate these results and analyzed them using three methods, as described in , , and . All methods suggest that the previous absolute error (a) was strongly conservative. The newly computed value of a approaches zero (a=0.22 Ω), as Fig.  shows. The value of b by appears to be correctly estimated: we obtained b=7.5% according to , or lower values (b=4%) according to and . For the following analysis, we will use the error estimation from , as lower values of b can lead to overfitting in the tomograms.

In Fig. b, we present the inversion differences between the error model chosen for this study and the model previously used in , applied to the inversion of the ERT monitoring data from 2022. Using the updated error model, resistivities in the permafrost core (x=65 to 115 m) are higher in both winter and summer. The maximum differences occur from May to July, especially on the right side of the side tunnel (an increase of >10 %), while the smallest differences occur between January and April. An exception is represented by the area between y=0 and -10 m right of the side tunnel (Fig. c), which is very close to the rock surface. Here, resistivities are generally lower, with peaks during the warm months: this better represents the active-layer dynamics measured with temperature loggers. Considerable differences are also evident in the area corresponding to the fault zone crossing the permafrost body. The inversion with the updated error model reports lower resistivities along the fault axis; however, these values do not extend to the main tunnel, as field observations would suggest. The area between x=150 and 200 m shows higher resistivity, but only in the winter months. The tunnel in this area is covered by concrete (indicating a more fractured bedrock), and it is seasonally covered by ice on the ground and walls. The tunnel surface (y<-25 m) shows smaller differences: frozen patches are found at different x-positions, but the most evident is at x=135 m, which corresponds to one tunnel exit where snow normally accumulates in winter. Lower resistivities are produced by the inversion at x>225 m from January to July, with new values reaching -10 %.

Precise knowledge of the T-ρ correlation is crucial for a correct interpretation of the thermal state of the permafrost core derived from resistivities. Many laboratory calibrations are nowadays available for different sites and materials examples in, but most studies do not validate them quantitatively in the field e.g.. Information at permafrost core depths is often missing, as few studies have compared field and laboratory measurements. previously conducted field calibration but did not compare it with laboratory values. reported lower resistivity values in field measurements compared with laboratory calibrations, attributing this discrepancy to water pressure effects. To date, no field calibration procedures have been proposed.

demonstrates that resistivity behaves differently above and below the freezing point, a finding that we also observed at the ST-5m logger (see Fig. S7b). He could also show that the ρ-T gradient below the freezing point (29.8±10.6 % °C⁻¹) is, on average, ten times higher than that above the freezing point (2.9±0.3 % °C⁻¹). Our new field calibration yields similar values at the freezing point, but higher resistivity gradients at low temperatures. Similar results are obtained when comparing field calibration by with laboratory measurements by , or for laboratory measurements consistent for different types of rocks .

Assuming that laboratory measurements are typically performed on intact rock samples, it can be inferred that such results provide a reasonable approximation of a saturated rock matrix. Laboratory calibrations are typically performed on fully saturated samples, whereas field conditions often involve partial saturation, which can result in higher resistivity. In this case, the conditions we observe at the Zugspitze are always nearly fully saturated (>90 %), except for the outer 8–10 cm of the rock surface . In the specific setup where we measure from inside the gallery, meters to decameters from the rock wall, surface variations do not play an important role (compared to measurements with electrodes on the surface). The higher resistivity values observed in the field are likely attributable to the presence of ice within frozen discontinuities. However, the extraction of rock samples inevitably removes them from their in situ stress conditions, which may induce minor fracturing or the opening of pre-existing fractures, thereby potentially reducing the measured resistivity of the specimen.

Additionally, recorded supercooling effects just after the equilibrium freezing point (-0.5 °C) during initial freezing in laboratory tests. Although these effects are not directly apparent in our dataset, an increase in the standard deviation of resistivity measurements is observed near the freezing point in the ST-5m logger data (see Fig. S7b), which may be linked to similar processes. It is important to note that such effects were only observed in the laboratory during the initial freezing phase and for very short periods, conditions that cannot be replicated with our setup.

Both field and laboratory calibrations at our site agree on the freezing point around 29 kΩm; however, the interpretation of laboratory measurements above this point should be taken with caution. In fact, it is impossible to accurately reproduce the steadily variable alternation of bedrock, fractures, ice/water, and voids found in nature in a sample; however, this composition strongly influences ERT results. This is evident in the high variability of the calibration results presented here, despite our temperature loggers being just 5 m apart. For this reason, caution should be exercised when extrapolating the T-ρ relation to other locations, even within the same site. We suggest that field calibrations are totally valid only for the measured spot. Further limitations are a reduced temperature range in nature and the lack of freeze-thaw cycles when measuring permafrost. Recent studies confirm the presence of hysteresis effects in field measurements and demonstrate their importance in the energy balance of permafrost .

Our measurements recorded resistivities greater than 200 kΩm, values that exceed the laboratory calibration range. has explained these values in terms of ice presence, but not numerically proven. Our results indicate that laboratory calibration should be validated in the field and, if necessary, refined based on field data.

The chosen procedure, the filtering criteria for outliers, and the bin placement can affect the quantification of the error to varying degrees . Different arrays generate distinct error types, corresponding to different error models. Further complexity arises when multiple measurements are combined into a single dataset, as in this case. In addition, the reciprocal error may vary over time during long-term monitoring due to changing external conditions, such as seasonal fluctuations in water state or long-term variations in underground water content. Given these factors, the best inversion would theoretically require one error model per array and per epoch, which would be time-consuming during data acquisition. As a consequence, the use of reciprocal errors is very rare in field applications . When reciprocal measurements are conducted, most studies use a uniform error level for all configurations and time steps to reduce the degrees of freedom , thereby ensuring consistency among results at different times , while other authors use a time-varying error model . With the new computation of the error model suggested here, i.e., the drastic reduction of the absolute term a, we possibly introduce some artifacts in the inversion results. Still, the significant advantage is a substantial increase in resolution, enabling us to investigate permafrost degradation in much greater detail (see features in Fig. ).

As shown in Figure S9a, the updated error model has little impact on the RMS error, which remains within the same range (0.99–1), as expected, since this is the value minimized by the inversion algorithm. However, the number of iterations increases from 3–4 to 6, which was also foreseeable, since smaller errors require a more precise fit of the model to the data to achieve the same RMS values and therefore more iterations. Inspection of the RMS formulation (Eq. ) shows that, if ϵRMS remains approximately constant, smaller values of the error ϵi result in a reduced data misfit (i.e., di-f(m)i). In our case, with the newly computed values of a (Fig. ), we observe smaller errors for R<104 Ω, leading to stronger data weighting. This, in turn, allows for closer model adjustment, potentially improving the fit, though it also increases the risk of overfitting and introduces model artifacts, while reducing smoothing and enabling the resolution of finer-scale features.

New temperature calibrations and error models enable us to accurately interpret the inverted resistivities, particularly in the permafrost core. Results confirm the presence of permafrost near the melting point, and its degradation is readily quantifiable. Still, complex inter-annual behavior due to latent heat effects during thawing, seasonal water infiltration, and variable snow cover makes the processes highly non-linear and therefore difficult to interpret. The degree of fracturing also seems to contribute to this non-linearity. A survey of fractured zones was already presented in , but the effects of the fault zone were neglected. Here, thanks to the new field calibration, differences are distinctly highlighted. Clustering, as suggested by and , also provided powerful insights in this direction. Thanks to simple algorithms, large amounts of data are used to detect similar patterns, yielding an objective, data-driven analysis. This allows for a more robust interpretation of results than the “user-defined areas of interest”, which are mostly rectangular, as previously used in the literature .

Despite this complexity at high temporal and spatial resolution, when looking at slope scale over one decade, it is evident that ERT values strongly correlate with rock and air temperatures, as shown in Fig. b. Mean resistivity of the tomograms decreases from 37 kΩm in 2018 to 28 kΩm in 2023 (-25%). The decrease is stronger from 2018 to 2021, whereas in the last 3 years it is less pronounced. Meanwhile, air temperatures have steadily increased since 2012, rising from -4.2 to -3.2 °C.

This supports the theory that external thermal forcing and conductivity are the primary processes promoting permafrost degradation here. Contradicting trends, such as the increase in frozen areas between 2007 and 2014, as shown in Fig. , can also be explained by thermal forcing (e.g., rising air temperatures). In fact, the hydrological year 2006/2007 recorded extreme temperatures, 1 °C warmer than the 30-year average. On the contrary, from 2007 to 2014, temperatures were mostly below the 30-year average (Fig. b, inlet). Therefore, 2007 can be considered an anomaly, and the measurements from that year are not representative of the 2000–2010 decade.

Analyzing the resistivity ratio between active layer and permafrost ρPL/ρAL as suggested by , we compare clustering class 1 with class 2 and class 1 with class 3 and focus on the maximum values of each year (Fig. a). Although they varied, both ratios increased over the last 10 years. The ratio class 1/class 2 closely follows the trends of mean annual air temperature (MAAT): it decreased in 2018 and in 2021 after mild years, while it increased in all other years. In 9 out of the last 10 years, MAAT was above the long-term average of -3.8 °C, with records reaching -2.6 °C in 2019/2020 and 2022/2023. Excluding the possibility that the permafrost core has increased its resistivity (ρPL), this trend can be explained only by a decrease in resistivity in the active layer (ρAL), to a stronger degree than the decrease in the core, as confirmed by Fig. b. Compared with Fig. , it is remarkable that, although the number of frozen cells has not decreased since 2021, the ratio ρPL/ρAL increased in 2022 and 2023. This confirms that non-linear degradation processes are occurring in the active layer and that sudden degradation may occur in the next few years. On the contrary, it seems that class 1/class 3 have stabilized over the last four years, possibly indicating that degradation has already reached a stable level in these areas.

The resistivity ratio between the active layer and the permafrost in our data falls within the same range as reported by . However, while they observed a long-term decrease in this ratio, our results show a contradictory overall increase. This suggests that, at our site, the degradation of the active layer has not yet significantly impacted the permafrost core. Notably, our unconventional measurement setup–conducting measurements from the inside out rather than from the surface may influence this interpretation. In typical surface-based ERT surveys, accuracy decreases with depth due to the limited number of data points, resulting in higher confidence near the surface and lower confidence at greater depths. The configuration used here yields improved accuracy along the main tunnel (deep bedrock) and along the side tunnel (permafrost core), but provides lower resolution for the rest of the outside bedrock slope.

After more than a decade of monitoring, we share key lessons learned from challenges like equipment malfunctions and data inconsistencies to improve future monitoring efforts and contribute to the scientific community. Long-term monitoring requires strict consistency in measurement parameters and procedures to ensure data comparability over time. However, when changes in software, hardware, or operational protocols are necessary or beneficial, it is advisable to implement an overlap period during which both the old and new setups are operated simultaneously. This overlap should be sufficiently long to capture variability across different environmental conditions, including both warm and cold periods.

Instruments deployed in humid environments are particularly susceptible to corrosion, which can compromise data quality and equipment longevity. Regular inspection and maintenance are therefore essential and require the continuous replacement of faulty or degraded electrodes. All components of the monitoring system, including cables, measurement units, and data loggers, should be routinely tested to promptly detect and address failures. Immediate, on-site verification of collected data – especially at remote locations – is strongly recommended to identify potential issues early. Quality control procedures – such as threshold checks, outlier detection, and trend analysis – in the data workflow can help identify inconsistencies early. Establishing clear criteria for data validation and flagging enhances the overall reliability of the dataset. In this context, optimizing and automating data analysis procedures can significantly enhance efficiency and reliability.

Comprehensive documentation, regular calibration (with redundancy where possible), and robust data management – including automated logging, standardized formats, and secure backups – are essential to ensure the accuracy, reproducibility, and long-term usability of monitoring data. We also encourage the publication of data in open-access repositories and active participation in collaborative initiatives such as the IPA Action Group IDGSP.

At this site, permafrost degradation is mainly driven by thermal forcing, as shown in Fig. b. Future heat waves are projected to increase in magnitude and frequency . As a result, this trend is expected to persist in the coming decades, particularly for the active layer (cluster classes 2 and 3), although the permafrost core (class 1) will ultimately be affected as well. The fault crossing the tomogram, clearly visible thanks to the newly developed error model, may further accelerate degradation due to its high permeability. In fact, when unfrozen, it could allow large volumes of water to infiltrate, as demonstrated in , thereby introducing advective heat transport processes that compound the thermal degradation. Consequently, atmospheric forcing is expected to play a significant role in the dynamics of permafrost degradation; however, its influence remains difficult to quantify. Increasing the measurement frequency from April to July could provide valuable insights into the delayed thermal response of deep rock layers.

On steep slopes, the increase in rock temperatures is also linked to irreversible ice losses . Although this is not expected to influence the hydrological cycle in alpine environments strongly, it has strong stability consequences for slope stability . This might also be true for the analyzed area, where the fractured zones might be a source of new instabilities, given the almost vertical dip of this fracture area.

The current setup was not fully automated and, consequently, it does not enable real-time monitoring. The potential of automated ERT has been assessed in . The calibration and error estimates in this paper contribute to a more robust implementation of ERT in monitoring and early warning systems for critical infrastructure, e.g., mountain huts and cable car stations, where less-calibrated approaches are currently being tested. Fixed and automated ERT monitoring systems have been implemented on steep permafrost slopes (e.g., studies by ). In principle, such systems allow continuous resistivity measurements and could provide near-real-time insights into permafrost conditions. However, to our knowledge, these approaches have not yet reached a level of maturity that enables reliable operational use for early warning purposes. Achieving this would require developing robust automated workflows for both data acquisition and processing and establishing clearly defined early-warning criteria. Potential criteria could include thresholds for active-layer thickness or for resistivity values approaching the thawing point. It is important to remember that ERT-derived information alone is insufficient for early warning applications. It must be complemented by detailed geological characterization, including surface fracture mapping and, ideally, subsurface information obtained from boreholes or tunnels to identify critical discontinuities and preferential ways for water flow. Furthermore, reliable quantitative interpretation of resistivity data requires field calibration over at least one complete freeze-thaw cycle and across multiple depths, as demonstrated in this study.