ERTM in Switzerland started in 1999 at Schilthorn, Berner Oberland (Hauck, 2002), as part of the European PACE project (Permafrost and Climate in Europe; Harris et al., 2009). Due to its promising results, the installation was maintained and later extended to a fully automated and continuously operating monitoring site (Hilbich et al., 2011). During the following decade, ERTM was initiated at an increasing number of sites; four of them were subsequently included in the Swiss permafrost monitoring network PERMOS (Hilbich et al., 2008a; Morard, 2011; PERMOS, 2019). Though other ERTM sites exist in the central Alps, the network presented here is composed of a selection of six monitoring sites located in different regions of the Swiss and Italian Alps located in Fig. 1. These six sites (out of the 14 sites where repeated measurements are available; see Table 1) were selected according to (a) length of ERTM time series, (b) existence of long borehole temperature time series for validation and process analysis, and (c) diversity of subsurface properties. Table 1 summarizes the general characteristics of the ERTM network in addition to other ERTM sites introduced below. In this paper, we present datasets until the end of 2017.

The network covers a large variety of site characteristics regarding substrate, ice content, geomorphology, thermal properties, and the observed soil temperature range. The active layer thickness (ALT, i.e. the maximum seasonal thaw layer that is here derived from temperature measurements) given in Fig. 2 varies for all sites between 2.5 and almost 10 m, depending on site characteristics and local climatic conditions (PERMOS, 2019; Pogliotti et al., 2015). During the observation period, a general increase in ground temperatures was observed in all boreholes (PERMOS, 2016; Pogliotti et al., 2015). Figure 2 summarizes the main thermal characteristics of all sites (mean annual air temperatures (MAATs), ground temperature at 10 m depth, and ALT) compared to the average resistivity at around 10 m depth (ρ10 m) at the borehole location. The available ERTM time series of the respective sites shown in Fig. 3 highlight the almost 20-year time series at Schilthorn (SCH) and the ∼ 10-year period at most other sites of the network. Figure 1 and Table 1 show additional existing ERTM time series from permanent electrode arrays, but they are not the focus of the present paper due to (a) the absence of borehole information or (b) interruption of measurements for safety reasons.

ERT is used to image the distribution of the electrical resistivity of the subsurface, based on a four-electrode array, or quadrupole. The electrical voltage (or potential difference) V is measured between a pair of electrodes while a current at intensity I is injected between a second pair of electrodes. Tomographic measurements deploy tens of electrodes for the collection of a few hundreds of quadrupoles, which in combination with adequate inversion algorithms permits a solution for the distribution of the electrical resistivity of the subsurface (e.g. Dahlin, 2001; LaBrecque et al., 1996).

Long-term ERTM requires consistency in all acquisition parameters for comparison purposes. Regarding the configuration of the quadrupoles, surveys were collected in early years using one of three standard configurations: Wenner (W), Wenner–Schlumberger (WS), and dipole–dipole (DD) (e.g. Hauck et al., 2003). Recently, the W configuration has been established as the main configuration used for the monitoring measurements because of a high signal-to-noise ratio. Even though logistical constraints have changed at some sites and new instrumentation became available since the earliest measurements, the same W configuration is still deployed to permit a fair comparison of the data from different periods. However, CER and DRE are exceptions as their monitoring procedure continues to use the initial WS configuration.

We use two different instruments for the acquisition of ERTM data: a SyscalR1 instrument (Iris Instruments, France) with an internal resistivity of 108 Ω, which can connect to 48 electrodes, and a Geotom instrument (Geolog, Germany) with an internal resistivity of 109 Ω, which can connect to 50 electrodes. Permanently installed stainless-steel electrodes are used for current injection and potential measurements. Contact resistances were measured before every collection, and high values related to a poor contact between electrode and ground play, which is typical for highly resistive media like alpine geology and frozen ground. High contact resistances limit the current injections and have a negative effect on the data quality. Thus, different strategies to deal with high contact resistances in permafrost investigations have been addressed in several publications (e.g. Hauck and Kneisel, 2008a; Ingeman-Nielsen et al., 2016; Tomaškovičová et al., 2016). For all our ERTM sites, we tried to maximize the contact surfaces between electrode and ground by either pouring fresh water onto the ground or connecting several electrodes in parallel to reduce contact resistances. Electrode separation varies between 2 and 5 m (see Table 1) accounting for the characteristics of each site and the envisaged depth of investigations, which is between 15 and 40 m for the different sites.

In the context of long-term and automated ERTM at several sites, a consistent processing of the data prior to the inversion is required to evaluate data quality. This has already been addressed for single automated stations in mountain permafrost terrain (Hilbich et al., 2011; Keuschnig et al., 2017; Supper et al., 2014). In a precursor study, a data filtering scheme for several ERTM datasets for the Swiss Alps has been developed (Rosset et al., 2013). Here, we extend this approach by analysing the seasonal and site-specific characteristics of contact resistance and the statistical distribution of measured apparent resistivities. By this, we identify and remove outliers before the inversion of the data to reduce possible artefacts.

We compute the distribution of the specific electrical resistivity for a finite element grid from filtered apparent resistivity datasets. We inverted our data using different inversion schemes and compared the obtained results (similarly to Hellman et al., 2016) and could not detect any significant differences between them that would be relevant for interpretation in a permafrost monitoring context. In addition, previous studies have also revealed that consistent temporal resistivity changes can be resolved using independent inversions of the data (e.g. Caterina et al., 2017; Hilbich et al., 2011; Oldenborger and LeBlanc, 2018; Supper et al., 2014), instead of using a time-lapse algorithm (designed to reduce systematic errors for time series with short time gaps, e.g. hours or days). In this paper, in accordance with previous ERT studies at our field sites, the inversion itself is conducted iteratively using the software RES2DINV (Loke, 2018) in batch mode. We use the robust inversion scheme (L1-norm) with the so-called extended model to minimize the edge effects.

More advanced time-lapse inversion schemes (as proposed e.g. by Karaoulis et al., 2013; Lesparre et al., 2017; Loke et al., 2014) can reduce the occurrence of inversion artefacts, which may originate from the underdetermined parts of the inversion problems of each individual tomogram. Difference plots of individual tomograms without explicit time constraints may therefore contain artefacts, especially in regions with low sensitivity. In the following, small-scale anomalies in the resulting tomograms are therefore excluded from interpretation.

Due to the large number of datasets, as well as the spatio-temporally varying surface and subsurface characteristics at the six sites affecting the rms, the choice of a consistent and comparable stopping criterion for the inversion is not obvious. For simplicity and based on experience, here we use the third iteration of the inversion, which provides a compromise between an efficient reduction in data misfit and avoiding overfitting potential measurement noise (e.g. Hilbich et al., 2011).

After the pre-processing and inversion, the tomograms with sufficient quality are selected for further analyses based on a user-defined site-specific threshold for the inversion error (see filter step 3). Inversion error histograms (see Fig. S3 in the Supplement) illustrate the distribution of the inversion errors for all inverted datasets per site and the site-specific thresholds. As a standardized and non-empirical histogram analysis can indeed not be conducted with such a low number of bins (statistically, the number of measurements is low; see Fig. S3), manually defined thresholds based on expert knowledge were used similarly to in other studies (see Supper et al., 2014).

The resulting set of accepted ERT tomograms amounts to a total of 901 (out of 1013) for the network. One representative tomogram per site for end-of-summer conditions is given in Fig. 6. Mean resistivity values obtained for all sites span over 3 orders of magnitude as shown in Fig. 3, and highlight the considerable variability of alpine permafrost conditions ranging from fine-grained debris and weathered bedrock at SCH (∼ 1 kΩm) to massive ice in a coarse-blocky rock glacier MCO (∼ 1000 kΩm). The four other sites exhibit relatively similar resistivity values around 10 kΩm (yellow to light blue colours in Fig. 6). Note, however, that depending on the prevailing lithology, porosity, and subsurface temperature, similar resistivity values can represent very different subsurface characteristics. For example, values around 20 kΩm are interpreted as the unfrozen coarse-blocky layer at MCO and DRE, as ice-poor bedrock permafrost at CER, and as porous ice-rich permafrost at LAH. Except for MCO and CER, all ERTM profiles show significant lateral variability (in the logarithmic scale; see Fig. 6).

The thaw depth derived from the respective borehole temperatures is used as validation of the inversion results and is shown as horizontal lines crossing each borehole in Fig. 6. At CER, LAH, and MCO the thaw depth is well represented by vertical resistivity contrasts, while the thaw depth at the ice-poor bedrock sites SCH and STO (plateau) does not necessarily cause significant resistivity contrasts. Here, resistivity variations may rather be influenced by variable ground properties (rock type, degree of fracturation) (e.g. SCH-5000 and STO-6000 boreholes), which can prevent clear relationships between absolute resistivity values and thermal state (frozen or unfrozen).

As the borehole at DRE showed positive summer temperatures during most years of the observation period, no thaw depth is indicated in Figs. 2, 6, and 7. At the time of the ERT measurement shown in Fig. 6, the minimum temperature observed in the borehole was +1.6 ∘C (PERMOS, 2016). The resistive anomaly within this talus slope can therefore not (or not entirely) be interpreted as ice-rich permafrost, but rather represents the very porous talus with air-filled voids. The similarity of the resistivity patterns of the talus slopes DRE and LAH is however striking, with ice-rich permafrost confirmed by three boreholes within the resistive anomaly in LAH (and LAV, not shown). Given the potentially complex pattern of the air circulation process (see Wicky and Hauck, 2017) and the undercooled nature of talus slopes (Morard et al., 2010), we cannot fully exclude that parts of the resistive anomaly in DRE may still represent frozen conditions (see also similar geophysical studies on low-elevation scree slopes, Gude et al., 2003; Hauck and Kneisel, 2008b). However, with higher probability the zone above the borehole represents a porous blocky layer with smaller block size or more fine material compared to the maximum resistivity values around the borehole. The DRE tomogram therefore illustrates the potential ambiguity of ERT interpretation in the absence of ground truth or further complementary geophysical data (e.g. seismic data; see Kneisel et al., 2008; Maurer and Hauck, 2007).

Within our network, ERTM series reach up to 19 years in 2017, which is rare if not unique in the permafrost literature (Supper et al., 2014). Although still short in a climate impact context, they allow the analysis of up to almost 2 decades of resistivity changes in the context of permafrost thaw and ground ice melt. Figure 7 presents the percentage change of resistivity logarithm (i.e. log⁡10(ρ2)-log⁡10(ρ1)/log⁡10(ρ1)×100) between the first and last years of ERT monitoring for comparable end-of-summer dates for each site. The end-of-summer period is considered to be between mid-August and mid-September (except at DRE, where mid-October is the comparison period with the highest number of available datasets). All sites show a significant resistivity decrease within the permafrost layer (below the active layer), except at MCO where a clear trend is less obvious. This is consistent with borehole data, which show increasing permafrost temperatures and active layer deepening at all sites for the same specific time spans (PERMOS, 2016).

Amongst all sites, the strongest and spatially most consistent resistivity decrease occurs at SCH and LAH, which are also characterized by high permafrost temperatures (see MAGT10 m in Fig. 2). From Fig. 7, it can be seen that at LAH and SCH the decreasing resistivity zone reaches down to the bottom of the tomogram (to more than 20 and 10 m depth, respectively). At LAH, the central part of the resistive body (i.e. between 7 and 15 m depth and from 80 to 140 m horizontally in Fig. 6) exhibits the smallest temporal resistivity changes, while resistivity decreases above, around, and below the resistive body (see Fig. 7).

At STO, the resistivity decreased between 4 and 10 m depth on the plateau (left-hand side of the tomogram). Both boreholes situated on the plateau indicate a permafrost degradation (see ALT increase in Fig. 7 by 0.9 and 1.7 m respectively), but with a higher rate at the warmer and southern borehole. This is consistent with the thicker negative anomaly at STO-6100 seen in Fig. 7. The strong negative anomaly in the southern slope may also indicate permafrost degradation, but could also be related to increased water saturation at the time of the measurement.

At MCO, the increase of 0.8 m ALT over the 11-year period is the smallest increase rate of the whole network, which is difficult to resolve on the scale of the ERT resolution (5 m electrode spacing). A comparatively small ALT increase is expected considering the high ice content, and it is also consistent with the results from transient model studies (Marmy et al., 2016; Scherler et al., 2013) predicting a much faster permafrost degradation at SCH compared to MCO during the 21st century.

At CER, a significant short-term (during the 4-year observation period) resistivity decrease can be observed along the entire profile at ∼ 3–7 m depth, with the strongest decrease in the area between the two boreholes. However, no permafrost degradation trend can be considered over such a short time period, but measurements show a peak in 2015 due to the very warm summer conditions (see Fig. S4 in the Supplement).

At DRE, a significant resistivity decrease can be observed in the middle of the profile (around 50–70 m) at around 5–15 m depth between 2007 and 2015, which corresponds to the zone interpreted as a blocky layer, which is affected by the air circulation process (see Fig. 6). As the borehole indicates unfrozen conditions for both measurement dates (minimum temperature 0.7 ∘C in 2007 and 1.6 ∘C in 2015), the positive anomaly can probably not be related to thawing processes, but is rather expected to represent differences in temperature and humidity related to temporal variability of the complex ventilation process.

For the sites, which are measured throughout the year (SCH, LAH, and DRE), the seasonal sinusoidal variation in resistivity values demonstrates the general relationship between resistivity and temperature data for the near surface (Fig. 3, higher resistivity values in winter and lower values in summer). Figure 8 shows selected resistivity–temperature scatter plots for three sites (SCH, LAH, and STO), at both shallow depth in the active layer (∼ 1 m depth, Fig. 8a, c, e) and at greater depth within the permafrost (∼ 10 m depth, Fig. 8b, d, f). For the shallow depths, a seasonal signal with regular freezing and thawing is prevailing, which is visible in Fig. 8a and c as a well-defined hysteresis loop at SCH and LAH. At STO, non-existent winter data prevent the observation of a similar signal in Fig. 8e. Above the freezing point, the resistivity ρ increases quasilinearly with decreasing temperature T (see dashed line with a slope of 0.025 K-1 in Fig. 8a, c) following the equation 1ρ=ρ0/(1+α(T-T0)), with ρ0 a reference resistivity at a reference temperature T0 and with α a temperature coefficient for resistivity given as 0.025 K-1 for most electrolytes (Keller and Frischknecht, 1966). The observed hysteresis is caused by the differential behaviour of resistivity during freezing and thawing. During thawing, the resistivity drops abruptly as soon as liquid water becomes available, while the temperature remains at the melting point (zero-curtain effect) during the phase change. At the onset of freezing liquid water remains present in large quantities, enabling electrolytic conduction within the pore space. In addition, ions leaving the frozen phase enhance the pore water conductivity in the liquid phase and decrease the freezing point. Only after sustained freezing does the resistivity start to increase gradually as temperature further decreases.

Within the permafrost (Fig. 8, right column), seasonal cycles have much less influence and an inter-annual signal with a clear thawing trend with decreasing resistivity values is predominant. At 10 m depth at SCH, the temperature seems to reach a maximum constant value of ∼ -0.1 ∘C, which we consider the melting point (see Fig. 8b). The same signal is even more obvious at LAH in Fig. 8d. When temperatures reach the melting point, temperature stays constant, while the ice loss or gain can be followed by decreasing or increasing resistivity data, respectively. Figure 8d shows an example of the zero curtain at LAH, where the resistivity decreases in autumn 2015 and increases again in spring 2016, without temperature variation. At 11.1 m depth, the seasonal changes of temperature are indeed almost reversed, compared to ground surface temperature, because of the time needed by the cold or warm to propagate deeply. Consequently, resistivity still decreases at the onset of winter.

Figure 8e presents a comparison of resistivity–temperature pairs at both boreholes at STO for two depths within the permafrost (5 and 9 m). Temperatures recorded in the southernmost borehole (STO-6100) are generally higher and closer to the melting point. In both boreholes, at 5 m depth, the variability of temperature and resistivity is higher than at 9 m depth, as it is more influenced by surface variations. In borehole STO-6000, the recorded temperatures (∼ -2.5 ∘C) at 9 m depth tend to increase with time with comparably small resistivity decrease, while in the southernmost warmer borehole STO-6100 (∼ -1 ∘C) the temperature increase occurs with a significant resistivity decrease, proving the increase in water-to-ice content ratio. At 5 m depth, both boreholes exhibit a resistivity decrease and temperature increase with a much higher variability in resistivity, i.e. a higher variability in liquid water-to-ice content ratio. Due to the large range of values, a logarithmic scale is used in Fig. 8f, although we find a quasilinear relationship between temperature and resistivity from ∼ 0 ∘C down to our lowest temperature record (-3 ∘C). Within this temperature range, surface conduction is enhanced due to the increase in interface area, whereas in the case of ionic conduction the measured resistivity is influenced by the decreasing amount of liquid water content, leading to decreasing ion mobility and increasing salinity in the liquid phase (Oldenborger and LeBlanc, 2018). In the case of our network sites, the increase in surface conduction is compensated for by a rather large decrease in saturation, which explains the resistivity increase. However, the theoretical exponential temperature dependence of the electric current (e.g. Robertson and MacDonald, 1962; Coperey et al., 2019) is not detectable in our field data, probably because of the narrow temperature range. Similarly, Krautblatter et al. (2010) also observed temperature–resistivity quasilinearity by laboratory experiments on bedrock samples for sub-zero temperatures between -3 and 0 ∘C.

The best example of permafrost degradation detected by ERTM is seen in the almost 20-year time series at SCH. Figure 9 shows the inter-annual evolution of the long-term end-of-summer resistivity changes at SCH, highlighted by the resistivity logarithm change with respect to the median of all considered SCH datasets (shown for all other stations in the Supplement, Figs. S4–S8). The resistivity evolution at SCH clearly shows the effect of the heatwave in 2003 (strong resistivity decrease; see Hilbich et al., 2008b) and the slow return (until 2008) to the initial state (except for the zone between 2 to 6 m depth and 5 to 15 m horizontal distance, which did not recover its original resistivity value). From 2010 onwards, the ground below 5 m depth down to the bottom of the tomogram at 15 m depth exhibits even smaller resistivity values than during the record heatwave of 2003. The temporal evolution of both resistivity and temperature at 7 m depth at the SCH-5198 borehole location (see Fig. S9 in the Supplement) shows this evolution as well, but with continuous temporal details. Since 2009, and coinciding with temperatures reaching the melting point for the second time, the resistivity remained low (< 1400 Ωm at 7 m depth) despite substantial ALT variations (maximum 9.5 m in 2015 and 2017; minimum 4.5 m in 2014; see PERMOS, 2016).

In contrast to the effect of the singular heatwave of 2003 (affecting relatively shallow depths), the sustained resistivity decrease since 2009 affects greater depths and extends over the entire length of the ERTM profile. It represents the cumulative resistivity decrease caused by and in agreement with the change in the ground temperature regime since 2009 (warmer summers and winters). The observed ongoing resistivity decrease therefore confirms the sustained increase in liquid water content within the permafrost layer, which is considered a direct consequence of ground ice degradation.

To facilitate inter-site comparison, we plotted the resistivity values for each site determined from the spatial average of a manually selected representative zone within the permafrost in Fig. 10. Remarkably, all six sites follow a common general trend of long-term resistivity decrease, which is modified by year-to-year variations (e.g. due to meteorological year-to-year variability, irregular dates of the measurement, which may not correspond to the annual maximum thaw depths, or inversion artefacts; see further discussion below). Note the distinct reductions in resistivity that occur simultaneously between ∼ 2009 and 2013 at all sites except CER, where ERTM monitoring only started in 2013.

The reliability of ERT results is a constant concern, as inversion results can be ambiguous, especially where model sensitivity is low. Inverting monitoring datasets without an explicit time constraint in the inversion can produce errors due to noisy data (e.g. Karaoulis et al., 2013; Lesparre et al., 2017; Loke et al., 2014). The objectives of this study include the development and validation of a robust inversion approach that can detect the large-scale climate change signal (permafrost thawing) in the ERT data for very different field sites and conditions, rather than an optimal inversion algorithm for an individual case study. The consistency of the long-term series (e.g. Figs. 3 and 10) proves the high reliability of most ERT datasets presented in this study, even if single datasets may have a relatively high uncertainty (in the case of high electrode contact resistance, high inversion error, or random errors). However, regarding the interpretation of temporal resistivity changes, uncertainties may arise from the effects of temporally varying contact resistances and the number of quadrupoles used for inversion, but these are considered small compared to the amplitude of seasonal and long-term resistivity changes. The reliability of single ERT tomograms can be assessed by either analysing the resolution matrix (Friedel, 2003), the sensitivity uncertainty (e.g. Hermans et al., 2015), or the DOI index (depth of investigation; defined by Oldenburg and Li, 1999), all of which have been applied to datasets obtained on permafrost (e.g. Hauck et al., 2003; Hilbich et al., 2009). Although uncertainties remain, further comparison of resistivity values with borehole data, complementary geophysical data such as refraction seismics (e.g. Maurer and Hauck, 2007; Hilbich, 2010), and hydro-thermal model simulations (e.g. Pellet et al., 2016; Scherler et al., 2010) shows a generally good match regarding the depth of the thaw layer, the layer structure, and its spatio-temporal variability. Moreover, both apparent resistivities and inverted resistivities (Fig. 10a and b) show similar decreasing trends, supporting the assumption that existing inversion artefacts are smaller than the long-term measured variations.

Due to the technical difficulties of maintaining several high-altitude monitoring stations and obtaining high-quality data at the same time, our datasets presented in Fig. 3 have major data gaps and most sites are measured only a few times per year. However, data gaps in the resistivity time series may prohibit the reliable identification of extreme values related to specific thawing–freezing processes, and Hilbich et al. (2008b, 2011) proved the applicability of ERTM in investigating permafrost evolution on different timescales and validated the approach of long-term monitoring with only an annual end-of-summer dataset (at SCH, before the automatization of the site). Within the present network, we confirmed the feasibility of using end-of-summer measurements in a long-term evolution context for sites that are morphologically different than SCH and consider our data sufficient to justify our approach. With respect to resistivity, changes based on the temporal median of resistivity values of the end-of-summer dates (cf. Figs. 9 and 10) permit comparison to sites with only single annual measurements. However, a potential bias might be introduced by variations in the dates used for the end-of-summer measurements alone, or in combination with variations in the timing of the maximum thaw depth (Hilbich et al., 2011). Having continuously operating ERTM systems, as deployed at SCH, should therefore be the goal for future extensions and new installations of ERTM in permafrost areas.

As mentioned above, a further source of uncertainty may arise from seasonally changing electrode contact resistances and the correspondingly changing number of quadrupoles for the inversions of one ERTM profile. Because the filtering process is applied independently to each dataset, the filtered datasets may not consist of the same quadrupoles for each time instance. A time-lapse inversion approach to compute temporal differences would therefore be challenging, especially considering the low number of readings according to the Wenner configuration and the low time continuity of our datasets (gaps in time), which may lead to the occurrence of artefacts. However, for the long-term perspective according to the use of end-of-summer measurements as discussed above, this uncertainty is small as differences in quadrupole numbers are small (below 1 % for SCH, LAH, and CER and between 5 % and 10 % for the other sites for the dates considered in Fig. 7). Moreover, our filtering procedure excludes datasets with more than 20 % filtered quadrupoles (i.e. most winter measurements at STO and LAH; see Fig. 5), which avoids extreme cases of comparing inversion results obtained from a different number of readings. However, for detailed process studies on shorter timescales or at high temporal resolution, such as presented in Hilbich et al. (2011) for SCH, the same number of quadrupoles should be used for inversion to avoid inconsistencies and the occurrence of artefacts in the computed time-lapse differences. In such cases where datasets are derived from continuous monitoring at high temporal resolution, specific time-lapse inversion schemes may likely improve the tomograms significantly (as demonstrated by Doetsch et al., 2015). In our study, we excluded the interpretation of small-scale anomalies to reduce the uncertainty for the individual inversions.

Hysteresis loops in both directions can be observed within our dataset by plotting resistivity against temperature for each borehole location (see Fig. 8a, c). At SCH, visible hysteresis cycles are similar for both boreholes (13 m distant) close to the surface (0–1 m depths, not shown). However, at 1.2 and 1.6 m depths, only borehole SCH-5198 displays a similar (but less pronounced) hysteresis cycle, whereas borehole SCH-5000 exhibits a zero curtain during freezing that results in a hysteresis cycle in the other direction (see Fig. S10 in the Supplement). In sediments, Tomaškovičová (2018) recorded a zero-curtain effect only during the freezing period, similarly to Fig. 8c but in contrast to Fig. 8a, while a thawing zero-curtain effect similar to Fig. 8a is reported by Hilbich et al. (2011), Pellet and Hauck (2017), and Supper et al. (2014). Different local ground effects contribute to the freezing–thawing hysteresis, e.g. pore connectivity, metastable nucleation, capillarity, and pore blocking (Tian et al., 2014), as well as the mineralogy – the carbonate content being an important factor according to Krautblatter et al. (2010) – are assumed to be responsible for these differences.

On larger temporal scales, the comparison of borehole temperatures and resistivities shows that the last two extreme heatwaves in the Swiss Alps, in summers 2003 and 2015 (Scherrer et al., 2016), had a direct effect on ground temperatures and a more sustained effect on resistivities. Heatwaves accelerate the decrease in the ice content in the upper part of the permafrost, and the lost ice is not recovered (air temperatures too high to enable ice aggradation). Therefore, a subsequent warming has a higher effect in the years following a heatwave, as less energy is needed to thaw the decreased amount of ice, leaving more energy for warming of deeper layers (see also Scherler et al., 2013, for a summary of dominant influencing factors for active layer deepening). Figure 9 highlights this effect, which is well documented by the resistivity change below 5 m depth at SCH since 2015.

In their review paper on ERTM in the context of permafrost, Supper et al. (2014) compared the acquisition characteristics mentioned in the very few available ERTM studies at that time and described the SCH installation (Hilbich et al., 2011) as “exceptional”. This was because of the favourable conditions in terms of low contact resistances (due to lithology), permafrost temperatures close to the freezing point, and low resistivity values on the one hand and the unusually low ice content in comparison to other ERTM sites on the other hand. The results of our study confirm this observation in a broader context, with SCH having the lowest resistivities (Fig. 3), the lowest contact resistance values (Fig. 4), and the lowest inversion errors of the entire network. However, the results of the network demonstrate that long-term ERT monitoring is also possible for extremely resistive permafrost landforms such as MCO rock glacier (see Fig. 10a) with much higher contact resistances, but not significantly higher inversion misfits, at least for the summer period. Figure 10c also indicates that analysing several sites simultaneously with a similar methodology helps to increase the representativeness of observed changes in ground ice occurrences. With the ERTM network presented in this study, the SCH data can now be set in relation to ERTM results from other permafrost landforms such as rock glaciers and talus slopes.

The parallel evolution of mean resistivity within the permafrost layer in all parts of Switzerland (and the Italian Alps) confirms the high representativeness of ERTM data for the analysis of ongoing permafrost degradation. Importantly, this spatially averaged resistivity trend represents a much larger footprint of permafrost conditions than would otherwise have been determined according to the point information of borehole temperature profiles (see the standard PERMOS temperature time series plots shown in the Supplement, Fig. S11). Furthermore, ERTM results are directly relatable to ground ice melt without the masking effect found in temperature series due to latent heat release close to the melting point. The observed parallel resistivity evolution at all sites therefore clearly illustrates the ongoing permafrost degradation with a remarkable shift to another resistivity regime after 2009, in agreement with temperature evolution.

In addition to the top-down permafrost degradation observed at all sites, Fig. 7 shows a resistivity decrease at LAH below the ice-rich body. This may be explained by air circulation in the talus slope, particularly below the ice-rich resistive body (Delaloye and Lambiel, 2005; Wicky and Hauck, 2017), which might lead to bottom-up permafrost degradation. Thinning of the permafrost body from below is indeed confirmed by increasing temperatures around 20 m depth (the sensor at 21.5 m depth in borehole LAP-1108 measured constantly negative temperatures until 2012, and recorded positive temperatures since 2015; see PERMOS, 2016). The observed resistivity decrease between 15 and 20 m depth, although close to the limit of the sensitivity of the tomogram, may therefore result from an increase in the ratio between liquid water content and ice content or increasing ion mobility in the liquid pore water at negative temperatures (Oldenborger and LeBlanc, 2018).

The ERTM network is part of a broader permafrost monitoring framework, which includes other variables like air and ground temperature (e.g. Isaksen et al., 2011; PERMOS, 2019), as well as soil moisture (Pellet and Hauck, 2017; Wang et al., 2016), P-wave velocity (Hilbich, 2010; Krautblatter and Draebing, 2014), and permafrost creep velocity (Delaloye et al., 2010; Kellerer-Pirklbauer and Kaufmann, 2012; PERMOS, 2019). Figure 11 shows the simultaneous temporal evolution of electrical conductivity (the inverse of electrical resistivity) together with other monitoring variables from different sites in Switzerland, chosen for their long time series. We present here the conductivity instead of the resistivity for visual consistency with the trend of all other parameters. Interestingly, the increasing trends in rock glacier velocity and the ground temperature both show a break in 2016, while the conductivity trend continues to increase.

Similar trends are also observed at almost all other PERMOS field sites and they represent general tendencies for the entire Swiss Alps (PERMOS, 2019). The observed decrease in ground temperature and creep velocity in 2017 is due to the late onset of snow during the 2016/17 winter, which enabled a more efficient cooling of the ground. This effect is not seen in the resistivity data at SCH because only end-of-summer measurements are considered. Deep ground cooling also occurred during that winter at SCH (shown in Fig. S9 in the Supplement), leading to a seasonal increase in the ice-to-water content ratio. However, this increase was countered by the especially warm spring and summer until the end-of-summer measurement in 2017. The temperature measured at 9 m depth in summer 2017 was indeed the second highest after 2015. The decrease in temperature in 2016/17 was more pronounced at sites with thinner snow cover (and late snow onset) and colder permafrost temperatures like at MCO (see Fig. S11 in the Supplement), and at STO and LAH where it caused small increases in the resistivities in 2017 (Fig. 10).

The fact that all measured variables in Fig. 11 (and similar data for other stations in the Alps) exhibit the same long-term evolution of permafrost, i.e. degradation due to climatic conditions, is very likely due to the current disequilibrium between atmospheric and ground thermal conditions. The resulting permafrost degrading trend is so dominant that it outweighs any impacts of morphological, topographical, climatological, or geological differences between the sites.