20-year permafrost evolution documented through petrophysical joint inversion, thermal and soil moisture data

This study investigates the ground characteristics of the high altitude (3410 m a.s.l.) permafrost site Stockhorn in the Swiss Alps using a combination of surface and subsurface temperature, soil moisture, electrical resistivity and P-wave velocity time series data including a novel approach to explicitly quantify changes in ground ice content. This study was motivated by the clear signal of permafrost degradation visible in the full dataset at this long-term monitoring site within the PERMOS (Permafrost Monitoring Switzerland) network. Firstly, we assess the spatio-temporal evolution of the ground ice and water content by a combined analysis of all available in situ thermal (borehole and ground surface temperature), hydrological (soil moisture) and geophysical (geoelectric and seismic refraction) data over two decades (2002–2022) regarding the driving factors for the spatially different warming. Secondly, we explicitly quantify the volumetric water and ice content and their changes in the subsurface from 2015 to 2022 using a time-consistent petrophysical joint inversion scheme within the open-source library pyGIMLi. The petrophysical joint inversion scheme has been improved by constraining the rock content to be constant in time for six subsequent inversions to obtain consistent changes in ice and water content over the monitoring period based on jointly inverted resistivity and traveltime data. All different data show a warming trend of the permafrost. The ice content modeled from the petrophysical joint inversion has decreased by about 15 vol.% between 2015 and 2022. Changes in ice content are first observed in the lower, south-facing part of the profile. As a result, resistivity and P-wave velocity have been decreasing significantly. Permafrost temperatures measured in the boreholes have increased between 0.5 °C and 1 °C in 20 years. Our study shows the high value of joint and quantitative analysis of datasets comprising complementary subsurface variables for long-term permafrost monitoring.


Introduction
Permafrost, defined as ground at or below 0 • C for at least two years (Permafrost Subcommittee NRC Canada 1988), is an invisible thermal phenomenon.Its evolution is therefore harder to observe than other cryospheric processes.Mountain permafrost occurs in a variety of landforms (e.g.bedrock, rock glaciers, and talus slopes).For more than 30 years, studies have been addressing the effects of atmospheric warming on mountain permafrost (e.g.Haeberli 1992) by developing operational networks to detect and monitor permafrost degradation (Harris et al 2001, Etzelmüller et al 2020).In recent decades, higher air temperatures have significantly increased active layer thickness (ALT) and permafrost temperatures (Noetzli et al 2022, PERMOS 2023b) and decreased ice content (Hilbich et al 2008a, Mollaret et al 2019).
Mountain permafrost degradation, accelerated by heatwave events, affects slope stability (Ravanel et al 2017, Hock et al 2019) leading to subsurface structural damage (Kääb and Haeberli 2001, Farquharson et al 2019, Guglielmin et al 2021).In the last 20 years, the European Alps have been affected by five hot summers (2003,2015,2018,2019,2022) characterized by heatwaves (NCCS 2022).In the future, hot events may increase in intensity and frequency (Caretta et al 2022).Additionally, the evolution of permafrost is influenced by near-surface properties (i.e.soil moisture, surface cover, microtopography, etc.) (Clayton et al 2021), snowpack distribution and duration.The timing and duration of the snow cover significantly influence the permafrost temperatures down to large depths through insulation effect and control of the radiation balance at the surface (e.g.Marmy et al 2013, PERMOS 2023b).
In Switzerland, several permafrost monitoring sites have been operationally investigated for more than a decade as part of the Swiss Permafrost Monitoring Network (PERMOS) using borehole, air and ground surface temperature (GST) measurements, but also non-invasive geophysical methods, such as electrical resistivity tomography (ERT, Hilbich et al 2008b).A regular geoelectric monitoring allows to capture 2D subsurface information, and thus to spatially expand the permafrost evolution derived from point-scale borehole information.ERT measurements are hereby often combined with refraction seismic tomography (RST) measurements to quantify volumetric ice and water content (Hauck et al 2011).At some sites, additional sensors measuring soil moisture and the full surface energy balance are located in the vicinity of the boreholes and geophysical survey lines (Pellet and Hauck 2017, Hoelzle et al 2022).These different datasets can be used to assess permafrost evolution and quantify the changes in subsurface properties by forcing, calibrating or validating various kinds of models, such as the thermal models CryoGrid (Westermann et al 2023) and COUP (Jansson 2012, Marmy et al 2016, Pellet et al 2016) or the petrophysical joint inversion (PJI) framework (Wagner et al 2019, Mollaret et al 2020).
A clear warming signal in the past two decades visible in all available datasets from Stockhorn (3410 m a.s.l.), a high-altitude site in the Swiss Alps, prompts this study to investigate permafrost degradation.This work aims to assess the spatio-temporal evolution of the ground ice and water contents by (1) combining all available in-situ thermal, soil moisture and noninvasive geophysical data over the last 20 years  to identify and analyze the driving factors leading to spatially different warming and (2) explicitly quantifying the volumetric water and ice contents and their changes in the subsurface using a timeconsistent PJI scheme.

Field site description
The PERMOS monitoring site Stockhorn is an eastwest oriented rock plateau with a gentle slope of 8 • located at an altitude of 3410 m a.s.l. in the western Swiss Alps above the village of Zermatt ( 45• 59 ′ 12 ′′ N, 7 • 49 ′ 27 ′′ E, figure 1(a)).The northfacing rock wall is covered by (strongly retreating) glaciers (figure 1(b)), whereas the south slope is unglaciated.Consequently, 3-dimensional temperature gradients influence the thermal regime of the permafrost below the plateau (Gruber et al 2004, Hilbich 2009).The frozen bedrock (figure supplementary material 1.1(a)) is covered by an up to 2 m-thick layer of coarse debris (weathered bedrock) (Marmy et al 2016, Mollaret et al 2019).In addition, the site is characterized by a complex geology (suppl.material 1.2) with two distinct units: gneiss from the Paleozoic crystalline Monte Rosa nappe and a mica schist (albite-muscovite schist) layer (Ebert 2001, Gruber et al 2004, Mollaret et al 2019).
Permafrost investigations in the Stockhorn region began in 1984 (King 1990), with operational monitoring activities starting in 1998 within the European PACE project (Harris et al 2001, figure suppl. material 1.1(d)) and continuing today within PERMOS (PERMOS 2023b).At Stockhorn, data are available from two boreholes, a meteorological station, RST monitoring (RSTM) and ERT monitoring (ERTM) (figure suppl.material 1.1(d)), GST loggers and soil moisture sensors (figure 1(a)).The historical background and the details of the measurement setup are described in suppl.material 1. Monitoring measurements are conducted once a year at the end of the summer (figure suppl.material 1.1(d)) using a Geotom resistivity meter (Geolog).Further details on ERT data acquisition and filtering can be found in Mollaret et al (2019) and PERMOS (2023b).Seismic measurements are performed annually on the same day as the ERT measurements along the central part of the ERTM line, using a Geode seismometer (Geometrics).A sledgehammer is used as source, and shot points are generally between every other geophone, with 10-20 individual stacks.First breaks are picked manually using the Reflexw software (Sandmeier 2020).

Geophysical inversion
This study focuses on the datasets from 2015 to 2022 with comparable measurement dates (first week of September).ERT and RST data were individually inverted using the open-source library pyGIMLi (Rücker et al 2017) to derive the electrical resistivity and P-wave velocity distribution (section 4.4).

Ice content estimation
We applied the petrophysical joint inversion (PJI) scheme introduced by Wagner et al (2019) to estimate the ice and water content distributions based on the measured apparent resistivities and seismic traveltimes.We follow the approach of Mollaret et al (2020) using Archie's law as the electrical mixing rule and a Wyllie-type averaging of P-wave slownesses (Timur 1968, Hauck et al 2011).The inversion and PJI parameters used in this study are given in Suppl.Material 3. To obtain consistent results over time regarding the non-changing porosity distribution, we first applied the PJI individually for each year and then used the (temporally) averaged porosity distribution for the final PJI runs shown in section 4.4.

Borehole temperature
Borehole temperatures in STO6000 (north) and STO6100 (south, details about instrumentation are given in Gruber et al 2004), drilled 30 m apart, show an overall warming (figure 2) with a similar interannual variability (suppl.material 2), but differences regarding absolute temperatures and warming rates due to their respective locations.Mean temperatures in 2022 at around 10 m and 18 m depth in both boreholes have been −1.4 • C and −1.9 • C (at 9.3 m and 18.3 m in STO6000) and −0.5 • C and −0.8 • C (at 9 m and 17 m in STO6100) (PERMOS 2023b) (figure 2(a)).Borehole STO6000 is therefore about 1 • C colder than STO6100, and decadal warming rates are generally higher at STO6000 (at 13.3 m +0.  per decade from 0.8 to 28.3 m in STO6000, while the same warming rate is found between 0.2 and 4 m in STO6100 (figure 2(a), suppl.material 2.4).In contrast, the temperature change between 2002 and 2012 at greater depths than 8.3 m in STO6000 and 5 m in STO6100 is with ±0.3 • C much smaller (see also figure 1

in Hoelzle et al 2022).
Moreover, ALT is generally larger at STO6100 compared to STO6000, while both boreholes show a pronounced ALT increase of −2.3 m and −2.8 m, respectively, from 2002 to 2022 (figure 2(b)).Since 2013, ALT has increased at a significantly higher rate (−0.2 m yr −1 ) than between 2002 and 2012 (−0.025 and −0.021 m yr −1 for STO6000 and STO6100, respectively).This is in accordance with other ALT measurements from boreholes in Switzerland (PERMOS 2023a).From 2001 to 2012, the difference in ALT between the two boreholes is relatively small, varying between 0.1 and 0.5 m.In contrast, the difference increases up to 1 m from 2013 onwards.

Ground surface temperature (GST)
Figure 3(a) shows GST monitoring data across the Stockhorn plateau since 2011 measured by UTL3 temperature loggers placed a few cm below the surface (see Staub and Delaloye 2017).Logger S001, located close to the north face, records the lowest temperature values and the largest temperature amplitude.In contrast, loggers S003 and S004, located around the edge of the small cliff to the south, show higher temperatures and a pronounced zero-curtain effect from 2012 to 2016.
The number of positive temperature days (figure 3(e)) and thawing degree days (TDD, figure 3(d)) have increased over the past decade for all GST loggers except S005 (see table 2.1 in suppl.material 2 for regression analysis and statistical details).

Soil moisture
Soil moisture is measured along a vertical profile located between the STO6100 borehole and the ERT profile (figures 1(a) and 4, SM1-SM6).Two types of sensors are installed at depths of 10 cm (1 SMT100), 30 cm (2 SMT100, 1 PICO64), and 50 cm (1 SMT100, 1 PICO64, see Pellet and Hauck 2017).In addition, to capture the spatial variability of the heterogeneous terrain (Pellet and Hauck 2017), 3 SMT100 sensors are spatially distributed at 30 cm depth around the ERT profile (figures 1(a) and 4(b), SM11-13).Sensors SM2 and SM6 are not shown due to poor data quality.All sensors show a similar pattern throughout the year, characterized by a constant volumetric water content (VWC) during the winter months (frozen state), an abrupt increase during the snowmelt season, a large variability throughout the summer, and a decrease at the end of the year when freezing begins.The observed seasonal VWC evolution corresponds to the GST stages, frozen (winter minimum), zero-curtain (large moisture increase/decrease), and unfrozen (summer maximum and high variability).
A comparison of the three different depths shows that the value of the winter VWC minimum increases with depth.Overall, the observed VWC values increased by 0.3-1.2vol% at all depths over the last 10 years.

Non-invasive geophysical data 4.4.1. ERT results
The ERT tomograms in figures 5(a)-(f) show a highly heterogenous resistivity distribution in the subsurface.Two distinct conductive zones (boxes R1 and R3) and one resistive zone (box R2) are observed.R1 (1000-5000 Ωm) is located between the two boreholes (figure 5(a)), where the surface material consists of fine debris (figure suppl.material 1.1(b)).This conductive zone was less pronounced before 2007 (not shown, see PERMOS 2019), but has become larger and more conductive since 2012 (figures 5(a)-(f)).The second conductive zone R3 is located in the lower part of the profile, where a blocky surface covers more fine-grained soil.From 2007 to 2012, R3 showed high resistivities around 100 000 Ωm, which decreased since 2014 but to a lesser extent than the conductive zone R1.When comparing resistivities from 2015 to 2022 (figure 5(g)), there is a general decrease in resistivity in the first uppermost 20 m.At the same time, resistivity increases in certain areas, mainly near the surface (figure 5(g), in blue).

RST results
In contrast to the resistivity models, P-wave velocity distribution mainly shows a horizontal layering.The surface layer (figure 5(h), box V1) corresponds to the low velocity layer in the top 3 m of the whole profile.Thickness of layer V1 increases with time, especially in the lower part of the profile (figures 5(h)-(m)).A high velocity zone occurs at greater depth with maximum values near borehole STO6100 (figure 5(h), box V3).From 2015 to 2020 this anomaly becomes smaller, and the velocities decrease.In 2021 and 2022, this high P-wave velocity zone of 2015 disappeared.The overall velocity also generally decreased in the first 10 m of the ground from 2015 to 2022 (figure 5(n)).

PJI results
PJI-based estimates of volumetric ice content at Stockhorn (figures 6(a)-(e)) generally range from about zero near the surface up to 18% at depth.As the minimum estimated porosity at depth is 16% (see suppl.material 3.1), this corresponds to ∼100% ice saturation in the deepest parts of the profile.The ice content shows large inter-annual variations (figures 6(a)-(e)) with significant ice loss over the observation period in the first 10-15 m (figure 6(f)).
The PJI-estimated maximum water content remains fairly constant throughout the period (between 31% and 35%), except in 2021 (28%).However, within the first 10-20 m the estimated water content generally increases with time (figures 6(g)-(k)).The spatial distribution of the estimated water content remains stable over the monitoring period, but water content maxima increase in some areas (figure 6(l), see arrows).

3D permafrost characteristics
Thermal, hydrological and geophysical observations show that the state and evolution of permafrost on the Stockhorn plateau differ spatially.The upper-/northern end of the plateau shows colder temperatures and higher resistivities, while the central part is warmer and characterized by a less resistive zone.Over the period 2002-2022, observations show an overall increase of near surface ground temperature (figure 3) and permafrost temperatures (figure 2) and a general decrease in resistivity (figure 5(g)) typically associated with ground ice loss (Hilbich et al 2011, Oldenborger and LeBlanc 2018, Mollaret et al 2019, Scandroglio et al 2021).These observations are consistent with the measured air temperature increase at Stockhorn (+0.5 • C for the period 2003-2018, see Hoelzle et al 2022) as well as the disappearance of the ice cover protecting the northern rock face of the plateau from the atmosphere in recent years.The loss of insulation from the ice cover led to changes in the 3D thermal regime.
Observations show that surface and permafrost temperatures in the northern borehole are colder and warming faster than in the southern one (figure 2).The lower temperatures are consistent with the proximity to the north face and the warming rates are higher since less liquid water (colder temperatures and less infiltration due to the slope) is present and therefore less latent heat is released.Changes in the 3D thermal regime of the plateau are also evidenced in the geophysical data and may explain the resistivity decrease observed in zone R1 (figures 5(a)-(f)).
We attribute the conductive anomaly R1 (figures 5(a)-(f)) mainly to geological differences (see suppl.material 1.2).In addition, we hypothesize that an impermeable frozen layer prevents water from infiltrating to larger depths giving rise to the varying resistivity in this zone.Due to the slightly inclined slope in this part, the water is drained towards borehole STO6100, enhancing the conductive zone and causing its variations.STO6100 also exhibits higher temperatures than STO6000, which may be due to the lateral influence of the southern cliff combined with the additional heat input from the higher water content, both leading to higher thermal conductivities.Sensor SM12, located at the southern edge of the plateau (figure 1(a)), has the highest VWC values (figure 4(b)) of all soil moisture sensors during the summer season.This is consistent with the gentle slope of the plateau, draining the water by gravity towards the south and accumulating in small depressions.
Compared to the resistivity tomograms, the Pwave velocity distribution does not distinguish these two main rock units.However, the layer V1 (around 800 m s −1 ) extends to depths that are in good accordance with the observed 1.4 m ALT increasing from 2015 to 2022.We therefore assume that the line separating zone V1 from zone V2 in figures 5(h)-(m) represents the unfrozen/frozen limit of the subsurface.The resistivity anomalies R1-R3 (figures 5(a)-(f)) are not visible in the seismic results.The maximum difference in P-wave velocity is about −1500 m s −1 , which indicates a change over time in pore fractions from ice to liquid water and/or air.

Accelerated thawing
All soil moisture sensors show an increase in VWC since 2014.This is consistent with the decrease in near-surface resistivity observed in recent years (figure 5(g)), a trend that has been observed since 2008 (Mollaret et al 2019).During the summer field campaign of 2022, we observed a newly formed thermokarst depression (suppl.material 1.1(b) and (c) at the location of the resistivity anomaly R1 (figure 5).Until 2021, the main low resistivity anomaly at larger depths was disconnected from the smaller surface anomaly.We expect that some ice was still present around 5 m depth until summer 2021, which disappeared completely in summer 2022, thus creating the thermokarst depression.In 2022, the ALT reached more than 6 m depth for the first time since measurements began.In contrast to ice content, the thermal regime (i.e.ALT) can recover from year to year.The summers of 2021 (cold) and 2022 (hot) illustrate this process.A cold summer like 2021 may reduce ALT temporarily, but the following heatwave in 2022 again had a strong impact, reaching a new overall record in ALT.
The effect of hot summers on the ALT is shown in figure 2(b) (arrows).After a strong increase of ALT due to the heat wave in 2003 (+1.4 m compared to the previous year in STO6000), ALT decreased again in the following years.In 2006, ALT returned to its pre-heatwave state (note, that data for 2005 are not available).The thermal conditions at Stockhorn thus recovered after one hot summer (Hilbich et al 2008a).
After the next heat wave in 2015 (Russo et al 2015, Lhotka and Kysely 2022), ALT has not decreased again due to consecutive heat waves in 2018, 2019, and 2022.This is consistent with a change in the onset and duration of snow cover (figure 3(b)), with a trend to earlier spring snowmelt leading to an increase in the number of positive temperature days.On the other hand, freezing degree days (FDD, figure 3(c), see table 2.1 in suppl.material 2), the onset of snow cover and autumn air temperatures show no significant trend and vary considerably from year to year.

Petrophysical joint inversion
Using the PJI, we were able to estimate the ice, water, air, and rock contents, which are in good agreement with the borehole temperatures.Compared to previous studies (e.g.Wagner et al 2019, Mollaret et al 2020), the PJI scheme was applied to multiple datasets over a 8 years period and improved by constraining the rock content to be constant in time for the six subsequent inversions.This enabled us to consistently estimate ice and water content changes over the monitoring period based on jointly inverted resistivity and traveltime data.
It has been seen that the water content estimates of the PJI are mainly dominated by the resistivity data, while the ice content estimates are mainly dominated by the P-wave velocity data (figure 6(f)).This can be explained by the PJI equations, where water content is directly related to resistivity (and porosity) by Archie's law (Archie 1942), in contrast to ice content, which is only related to velocity by the Wyllietype equation (Hauck et al 2011, Mollaret et al 2020).Using electrical relationships that include the resistivity of ice and/or rock (such as the geometric mean model, Mollaret et al 2020) would be an alternative with a more equal dependence of water/ice content on both input datasets.
The small positive anomaly (i.e. increase in ice content, marked by an arrow in figure 6(f)) near the steep topographic step towards the southern slope is probably an artefact, as processes related to thawing and freezing within the cracks observed on this cliff cannot be resolved with our geometric setup.Similarly, the unrealistic ice content values >0 within the active layer (figures 6(a)-(e), near the surface) are due to the insufficient spatial resolution, especially of the RST measurements (4 m geophone spacing).

Ice-rock ambiguity in the PJI
In September, we do not expect ice at the surface (i.e.ALT around 5 m).However, the PJI results show some ice content near the surface, especially in the earlier years of the monitoring period (figures 6(a)-(e)).Wagner et al (2019) and Mollaret et al (2020) have already highlighted the ambiguity between rock and ice content due to their similar P-wave velocities and resistivities.Resistivity values for ice and granite range from 104 to 106 Ωm (see appendix 3, table 3.6; Hauck and Kneisel 2008).Velocity values for permafrost range from 2500 to 4250 m s −1 , for magmatic rocks from 2500 to 5000 m s −1 , and metamorphic rocks from 3000 to 5750 m s −1 (see appendix 3, table 3.6; Hauck and Kneisel 2008).Additionally, Mollaret et al (2020) have emphasized that the PJI often fails to produce zero ice content, where positive temperatures are measured (although it correctly predicts minimum ice content values at the surface), which is at least partly due to the resolution capacity of the data.The resolution of the measured traveltime data is coarse (4 m spacing), which also prevents the detection of very low velocity zones in the unfrozen shallow layers.Moreover, data quality is affected by the complex topography (e.g.cliffs).Despite the presence of estimated ground ice content at the surface, the PJI results remain consistent over time.They show a clear decrease in ice content from 2015 to 2022, which is in good agreement with the increase in subsurface temperature and hence the thickening of the active layer.

Conclusion
In our study, we investigated the ground characteristics and permafrost evolution of a high altitude (3410 m a.s.l.) permafrost site in the Swiss Alps using a novel combination of surface and subsurface temperature, soil moisture, electrical resistivity and Pwave velocity time series data including an innovative approach to explicitly quantify changes in ground ice content and water content.Although different complex processes related to substrate and microtopography may complicate data interpretation, all the different data collected over the last 20 years show a warming trend of the permafrost with a corresponding decrease in ground ice content.Key findings of the study include: Further studies should include the application of different petrophysical equations in the PJI scheme for Stockhorn to better understand the processes between ice melt and liquid water transport and to improve the quantification of absolute ice content values.
Constraining the P-wave velocity of the two main rock units (gneiss and mica schist) could better differentiate rock from ice content.Our results may further serve as input and validation data for permafrost models such as the hydrothermal model CryoGrid Community Model (Westermann et al 2023), which can then be further used to analyze how quickly the permafrost will degrade in future.

3. 1 .
Non-invasive geophysical data acquisition ERT data have been collected along a dedicated monitoring profile since 2005 (Hilbich et al 2008a).

Figure 2 .
Figure 2. (a) Vertical ground temperature profile of the hydrological years 2002, 2012 and 2022 in boreholes STO6000 and STO6100.There is an increase in temperature at all depths from 2002 to 2022 in both boreholes.Borehole STO6000 warms faster at depths, but borehole STO6100 is closer to the melting point.(b) Active layer thickness (ALT) for both boreholes, with black arrows indicating the occurrence of summer heat waves and black vertical lines representing the uncertainty of the ALT.The ALT trend for borehole STO6000 is −0.025 m yr −1 from 2001 to 2012 and −0.200 m yr −1 from 2013 to 2022 (blue dashed line).The ALT trend for borehole STO6100 is −0.021 m yr −1 from 2001 to 2012 and −0.201 m yr −1 from 2013 to 2022 (red dashed line; Data: PERMOS 2023a).

Figure 3 .
Figure 3. (a) Daily ground surface temperature (GST, colored lines) and air temperature (grey line).(b) Daily snow height.Note, the data gaps from spring 2018 onwards (Data: PERMOS 2023a).(c) Freezing degree days (FDD)and (d) thawing degree days (TDD) for the hydrological year.The TDD of logger S001 increases significantly.(e) Number of days with positive temperature for the hydrological year.The number of days with positive temperature increases significantly for the sensors S001 and S002.Note, that loggers with a star indicate that the linear regression mentioned in the text is statistically significant at p < 5%.

Figure 4 .
Figure 4. Daily time series of Volumetric Water Content (VWC) (solid lines) and VWC averages over the hydrological year (dots) at different depths (10 cm, 30 cm, 50 cm).Sensors SM1, SM3, SM4, and SM5 are installed vertically in the same pit, while sensors SM11, SM12, and SM13 are distributed spatially (see figure1(a)).Note, that sensors with a star indicate that the linear regression is statistically significant at p < 5%.Sensors with the symbol ( * ) have a p-value between 5% and 10%.All sensors show a significant increasing trend (except sensor SM13).

Figure 5 .
Figure 5. (a)-(f) Independently inverted electrical resistivities obtained from ERT and (h)-(m) seismic P-wave velocities obtained from RST. (g) and (n) show the resistivity and P-wave velocity differences between 2015 and 2022, respectively.Resistivity difference is calculated as follows: (log(res_2022) − log(res_2015))/log(res_2015) * 100%, and velocity difference: (vel_2022−vel_2015)/vel_2015 * 100%.The maximum thaw depth at the survey dates is represented by the horizontal line at the borehole locations (vertical lines).Zones R1 and R3 are conductive zones.Zone R2 represents a resistive zone.There is a general decrease over time in resistivity in the uppermost 20 m.Zone V1 corresponds to the low velocity layer at the surface, roughly representing the thickness of the active layer, increasing by about 1 m from 2015 to 2022, as indicated by the dashed lines.Note, that the RST panel survey in 2018 (panel j) was conducted with a spacing of 2 m instead of 4 m and does therefore not cover the same area as the other RST surveys.Zone V3 represents a high velocity zone which disappears in 2021 and 2022.The overall velocity also generally decreased in the first 10 m of the ground.

Figure 6 .
Figure 6.Ice content (a)-(e) and water content (g)-(k) are derived from jointly inverted apparent resistivities and seismic traveltimes using the same rock content distribution within the PJI.The thaw depth at the measurement date is represented by the horizontal line at the borehole locations (vertical lines).The differences in ice (f) and water (l) content are calculated as f_2022-f_2015.The estimated ice content shows a significant ice loss from 2015 to 2022.The estimated water content generally increases with time within the first 10-20 m.See table 3.5 in suppl.material 3 for X 2 and RMS values.See suppl.material 3.2 for the air content and suppl.material 3.3 for the inversion of the data set from 04.09.2018 which was measured with a different measurement geometry (cf figure 5(j)).

•
The sum of positive temperature values and the number of days with positive temperature derived from GST loggers have increased between 2011 and 2021.•The permafrost temperatures measured in the boreholes have increased by 0.5 • C-1 • C in 20 years.The coldest permafrost conditions are observed in the northern borehole, which also shows a stronger warming compared to the southern borehole.• The ALT increased by 2.3 m (STO6000) and 2.8 m (STO6100) between 2002 and 2022.• Resistivity and P-wave velocity decreased significantly.The ERTs and RSTs are consistent with the spatially variable thermal state and evolution of the permafrost observed using standard monitoring techniques (i.e.GST, borehole temperature, ALT).• Thermokarst features have developed at the surface of the plateau in 2022.• The ice content modeled from a petrophysical joint inversion scheme (PJI) has decreased by about 15 vol.% between 2015 and 2022.Changes in ice content are first observed in the lower, south-facing part of the profile.