Permafrost in monitored unstable rock slopes in Norway – New insights from temperature and surface velocity measurements, geophysical surveying and ground temperature modelling

The warming and subsequent degradation of mountain permafrost within alpine areas is an important process influencing the stability of steep slopes and rock faces. The unstable and monitored slopes of Mannen (Møre and Romsdal county, southern Norway) and Gámanjunni-3 (Troms and Finnmark county, 35 northern Norway) were classified as high-risk sites by the Norwegian Geological Survey (NGU). Failure initiation has been suggested to be linked to permafrost degradation, but the detailed permafrost distribution at the sites is unknown. Rockwall (RW) temperature loggers at both sites have measured the thermal regime since 2015, showing mean rock surface temperatures between 2.5 °C and -1.6 °C depending on site and topographic aspect. Between 2016 and 2019 we conducted 2D and 3D electrical 40 resistivity tomography (ERT) surveys on the plateau and directly within the rock wall back scarp of the unstable slopes at both sites. In combination with geophysical laboratory analysis of rock wall samples from both sites, the ERT soundings indicate wide-spread permafrost areas, especially at Gámanjunni-3. Finally, we conducted 2D thermal modelling to evaluate the potential thermal regime, along with an analysis of available displacement rate measurements based on Global Navigation Satellite System 45 (GNSS) and ground and satellite-based interferometric synthetic aperture radar (InSAR) methods. Surface air and ground temperatures have been warmed significantly since c. 1900 with 1 °C and 1.5 °C, and highest temperature are measured and modelled since 2000 at both study sites. We observed a seasonality of displacement, with increasing velocities during late winter and early spring and highest velocities in June, probably related to water pressure variations during snow melt. The displacement rates of 50 Gámanjunni-3 rockslide co-vary with sub-surface resistivity and modelled ground temperature. Increased displacements rates seems to be associated with sub-zero ground temperatures and higher ground resistivity. This might be related to the presence of ground ice in fractures and pores close to the melting point, facilitating increased deformation. The study demonstrates and discusses the possible influence of permafrost, at least locally, on the dynamics of large rock slope instabilities. 55 * Corresponding author: E-mail address: bernd.etzelmuller@geo.uio.no (Bernd Etzelmüller):

northern Norway) were classified as high-risk sites by the Norwegian Geological Survey (NGU). Failure initiation has been suggested to be linked to permafrost degradation, but the detailed permafrost distribution at the sites is unknown. Rockwall (RW) temperature loggers at both sites have measured the thermal regime since 2015, showing mean rock surface temperatures between +2.5 °C and -1.6 °C depending on site and topographic aspect. Between 2016 and 2019 we conducted 2D and 3D electrical 40 resistivity tomography (ERT) surveys on the plateau and directly within the rock wall back scarp of the unstable slopes at both sites. In combination with geophysical laboratory analysis of rock wall samples from both sites, the ERT soundings indicate wide-spread permafrost areas, especially at Gámanjunni-3. Finally, we conducted 2D thermal modelling to evaluate the potential thermal regime, along with an analysis of available displacement rate measurements based on GNSS and ground and satellite-based 45 INSAR methods. Surface air and ground temperatures have been warmed significantly since c. 1900 with +1 °C and 1.5 °C, and highest temperature are measured and modelled since 2000 at both study sites. We observed a seasonality of displacement, with increasing velocities during late winter and early spring and highest velocities in June, probably related to water pressure variations during snow melt. The displacement rates of Gámanjunni rockslide co-vary with sub-surface resistivity and modelled ground 50 temperature. Increased displacements rates seems to be associated with sub-zero ground temperatures and higher ground resistivity. This might be related to the presence of ground ice in fractures and pores close to the melting point, facilitating increased deformation. The study demonstrates and discusses the possible influence of permafrost, at least locally, on the dynamics of large rock slope instabilities. 55

Introduction
Permafrost or permanently frozen ground is a globally widespread phenomenon, and covers c. 15 % of the northern hemisphere land surface (Obu et al., 2019). Permafrost is purely thermally defined, with ground temperatures below 0 °C over at least two consecutive years (van Everdingen, 1998). In southern 65 Norway, permafrost is widespread above c. 1500 m a.s.l., in northern Norway above c. 800 m a.s.l. (e.g. Gisnås et al., 2016a). In steep rock walls, permafrost is located several hundred meters lower in (Magnin et al., 2019), and many rock faces in Norway are within or close to the mountain permafrost limit. Furthermore, steep rock walls efficiently cool the ground and its surroundings because of low or lacking snow cover (Myhra et al., 2017), and they maintain strong thermal gradients in transition areas compared 70 to more snow-covered regions, forming environments of more intense frost weathering (Myhra et al., 2019). While permafrost degradation in the lowlands of the Arctic is mainly associated with ground ice melt (Hjort et al., 2018) and/or release of greenhouse gasses (Schuur et al., 2015;Davidson and Janssens, 2006;McGuire et al., 2010), slope instability is the major concern in mountain areas (Haeberli et al., 2010;Gruber and Haeberli, 2007). An increase in rockfall and rockslide activity has been documented 75 following atmospheric warming (Gruber and Haeberli, 2007;Ravanel et al., 2010;Fischer et al., 2012;Frauenfelder et al., 2018;Ravanel et al., 2017). Furthermore, the increase in sub-zero rock temperatures reduces shear strength in steep slopes by affecting the strength of intact rock, ice and rockice interfaces (Krautblatter et al., 2013;Mamot et al., 2019). The specific sensitivity of metamorphic rocks similar to those investigated in this study for temperature-dependent weakening of the rock-ice interfaces 80 has recently been demonstrated (Mamot et al., 2020) and is complemented by rock fatigue in zones with transitional rock freezing (Jia et al., 2015;Mamot et al., 2018). Recently, large rockslide detachments in Karrat Fjord, West Greenland, were associated to permafrost degradation (Svennevig et al., 2020).
Large rockslides are the most destructive processes in terms of single event landslide disasters (Evans and DeGraff, 2002) and caused massive destruction and loss of life in historic time, hitting water bodies 85 and causing displacement waves or filling valley bottoms (Hermanns et al., 2014;Hermanns et al., 2013;Svennevig et al., 2020). The Norwegian Geological Survey has systematically mapped relevant areas over most of the Norwegian high-relief land area for such potentially destructive unstable slopes, and classified them according to their risk {Hermanns, 2013 #3893; Majala, 2016 #4282}. Seven unstable rock slopes have been identified as high-risk objects based on their risk to cause loss of life, and are 90 therefore permanently monitored. At least two of them are situated within the permafrost realm or close to the lower limit of mountain permafrost in Norway: Gámanjunni-3 in Kåfjord/Troms, northern Norway (69.5°N, 20.6°E) and Mannen in Romsdalen, southern Norway, (62.5°N, 7.8°E). Both sites were deglaciated prior to the Younger Dryas (c.c. 12 ka BP), and showed initial displacement long after deglaciation, with calculated ages from c. 7-8 ka at Mannen and c. 6.6-4.3 ka at Gámanjunni-3 (overview 95 in Hilger et al. 2021). Paleo slip rates variation during the Holocene and slip initialisation have been discussed in relation to Holocene permafrost dynamics at these sites (Böhme et al., 2019;Hilger et al., 2021), and demonstrated that present movement rates are much higher than the estimated averages rates during the Holocene. While Vick et al (2020) mostly relates these instabilities to structurally controlled rock slope deformation, we hypothesize that these higher rates might be influenced in addition by a change 100 in the ground thermal regime, and thus permafrost dynamics since the onset of atmospheric warming after the Little Ice Age (LIA).
This study evaluates the permafrost conditions and recent thermal development in these two unstable slopes of Norway. We present updated movement and rock wall temperature data, electrical resistivity tomography (ERT) combined with seismic refraction surveys, along with numerical modelling of recent 105 thermal behaviour of the unstable rock slopes.

Setting
The study focusses on two main sites monitored by the Norwegian Water and Energy Directorate (NVE) since 2009 (Mannen) and 2016 (Gámanjunni-3), respectively ( Figure 1) (Blikra et al., 2016). Both sites The mean annual air temperature (MAAT) was -3.2°C during 2016-2020 on the top plateau at 1200 m a.s.l. The MAAT during the 1961-1990 normal period was -4.3 °C (Lussana et al., 2018;Saloranta, 2012), thus considerably cooler than in the recent years. During the 4 years of meteorological data, mean annual precipitation was 655 mm. The ground is usually snow covered from November until June, with an approximate thickness of 1 m. The site lies in the discontinuous mountain permafrost region of northern 145 Norway, and permafrost has been modelled or even measured within the slip face of the unstable rock slope (Magnin et al., 2019;Farbrot et al., 2013;Gisnås et al., 2016a;Obu et al., 2019).

Mannen (Møre and Romsdal County)
The Mannen rockslide is situated in the Romsdalen valley on a north-facing slope, between 900 m and c.
150 1300 m a.s.l. (Figure 1). The glacially over-steepened Romsdalen valley cuts through mountains comprised by gneisses that formed during the Caledonian orogeny (Saintot et al., 2012). The instability consists of an intensely folded high-grade metamorphic unit with alternating sillimanite and kyanite layers, amphibolites and pegmatites (Saintot et al., 2012). There is an exposed slip surface or back scarp, building up a 20 m high rock wall, ending in a mostly snow-filled deep rock fracture ( Figure 2b). The 155 valley floor of Romsdalen is covered by 15 large postglacial rock slope failures, and below Mannen, 6-9 rockslide deposits are mapped, of which 3 occurred in the first millennia after the deglaciation, and 3-6 slides are inferred climatically triggered during a climatic phase with increased precipitation following the Holocene Thermal Maximum (HTM) (Hilger et al., 2018). In September 2019, a smaller rockslide "Veslemannen" on the western flank of Mannen occured (Figure 2b), after episodic acceleration over 160 several years, leading to numerous evacuations of the local population living below Mannen (Kristensen et al., 2021). Mannen was previously proposed as a translational failure (Henderson and Saintot, 2007) and wedge failure (Dahle et al., 2010). Two scenarios define the Mannen instability (Dahle et al., 2008), where the largest has no detected movement. Scenario B is approximately 10 Mm 3 of rock with 7 displacements of 5 mm a -1 to north dipping 20°. Scenario C has displacements of 2.5 cm a -1 dipping 60°, 165 which is steeper than the topographic surface, and possibly sinking into a graben structure. Since the installation in 2010, the meteorological station on top of the Mannen plateau measured an average annual precipitation of 1250 mm and mean air temperature of -0.5 °C. During the last normal period 1961-1990, the MAAT was -1.3 °C at the Mannen plateau. Atmospheric warming is therefore 180 evident at this site during the last decades. The plateau is usually covered by a 2 to 3.5 m thick snow cover during the period November-June. The site lies at the lower limit of mountain permafrost, where permafrost can be expected in shaded patches or deeper fractures (Magnin et al., 2019;Gisnås et al., 2016a;Gisnås et al., 2014;Westermann et al., 2013). Recent modelling for the small rockslide "Veslemannen" indicated at least deep seasonal frost and a thermal influence on the dynamics of the 185 rockslide (Kristensen et al., 2021).

Methods and data processing
This study uses various data series from different measurements related to climate, rockslide movement, 190 thermal regime and subsurface conditions. All locations of the devices and profiles used in this study are given in Figure 1.

Movement of the rock slopes
For Gámanjunni-3, the real time monitoring was initiated in 2015, and includes Global Navigation 195 Satellite System (GNSS) antennas, crack meters, extensometer, a laser to measure distances to a reflector plate, a meteorological station, a ground based interferometric radar system (GB InSAR, sensitive to displacement towards the valley floor along the radar line-of-sight (LOS)) and web-cameras. In addition, three satellite corner reflectors were installed in 2012, and their displacement measured along satellite LOS (ascending: down and towards east, descending: down and towards west) are currently being 200 operationally monitored using Sentinel-1 satellite interferometric synthetic aperture radar (InSAR instrumentations, a meteorological station, web-cameras and two GB-InSAR systems. For this study, we use some selected GNSS monitoring stations, the laser monitoring, the GB-INSAR and the satellite corner reflectors to evaluate changes in movements of the unstable slopes. The location of the systems used in this study are given in Figure 1B. The distance laser sensor used at both sites is a Dimetix DLS-B 15, which measures with an accuracy of 215 1.5 mm and ±1 mm in good weather conditions. The laser device registers 10 measurements per second, which are averaged for every 10 minutes. The distance measured is in Line Of Sight (LOS) which may deviate from the true displacement vector direction, and daily averages are shown and used. The Trimble NetR9 GNSS Reference Receiver with a Trimble Zephyr 2 antenna is measuring position continuously, and processes an average position every 12 hours. The standard deviation calculated for a period between 220 1.8.2016 and 1.9.2020 is ±0.86 mm and ±0.69 mm in north and east direction, respectively, and ±1.89 mm vertically. However, one of the permanent points located outside of the unstable section of the slope showed movement because of water intrusion under the foundation, hence the time series with reliable data is short.
The ground-based radar from LiSALab is a GB InSAR system (© Ellegi Srl) (Bardi et al., 2016;Crosta et 225 al., 2017), placed on a concrete foundation in the respective valleys. The data were processed in the LiSALab software using an atmospheric correction region that covers a selected and assumed stable part of the slope. The reported data is derived primarily from 24 h averages and 3-day average measurements, which have been used to create interferograms, where the selected time-averaging period depends on the particular displacement rate in LOS at the site. The data is georeferenced to a 10x10 m DEM. At 230 Gámanjunni-3, time series for 8 points on the rockslide part, and 10 points on the rock glacier are extracted and further analysed ( Figure 1b).
All measurements included uncertainties and white noise in the data. To reduce these effects, we calculated daily averages of movement rates, and filtered the data with a moving average filter, with variable window sizes. This procedure allowed for identifications of long-term displacement trends and 235 possible seasonal variations of movement.

Measured and reconstructed rock wall temperatures
During 2015, five temperature data loggers (Geoprecision, M-Log5W-ROCK) were installed in the back scarps of Gámanjunni-3 (3 loggers) and Mannen (2 loggers), recording rock surface temperatures (RST) 240 with a 2-h interval, and with an accuracy of ±0.05°C (Magnin et al., 2019) (Figure 1b). The installation procedure followed the approach described by Gruber et al. (2004). To avoid rapidly fluctuating surface temperatures sensors were placed at a depth of c. 10 cm below the surface. We also tied to place the loggers above ledges to minimise snow influence (Magnin et al., 2019).
Automatic data loggers (Hobo and iButton) were placed on the Gámanjunni-3 rockslide and rock glacier 245 in 2013 and 2014, measuring ground surface temperatures (GSTs), i.e. temperature below the snow cover, and temperatures in air-filled voids between crushed blocks below the surface (Eriksen, 2018). The GST loggers are distributed in three clusters over the rockslide, with data points on the moving block, the rock glacier and the toe area of the rockslide, and maintained by NVE today (Figure 1). At the Mannen site, data loggers were placed on the plateau to measure GST ( Figure  Both sites are equipped with automatic weather stations, measuring surface air temperatures (SAT), precipitation (P) and snow depth (SD). To reconstruct temperature development since the end of the LIA at the study sites, both in terms of SAT and RST, two strategies are followed: First, we used gridded climate data (daily SAT and P) available for all Norway on a ground resolution of 1 km since 1957. The 260 data-set, in the following called "seNorge", is established by interpolation between meteorological stations (Lussana et al., 2018), and is daily operationally updated. Secondly, for the period before 1957 until the start of the meteorological observation period, which is during the end of the 19 th century in Norway, data from the weather stations on site and the rock wall loggers were combined with long-term series from nearby stations using simple or multiple linear regressions. For Gámanjunni-3, a good 265 correlation with the Tromsø weather station was obtained (Figure 1a), where the data series started in 1867. For the Mannen site, we used both the stations in Dombås and Fokstua in central southern Norway (Figure 1a), where SAT measurements reach back to 1864 and 1923, respectively. We considered R 2scores of above 0.7 sufficient to use these data-sets to derive upper boundary conditions for the numerical modelling.

Laboratory analysis of rock properties
In order to relate the resistivity results of the geophysical surveys to the freezing transition of specific rock types and approximate frozen rock temperatures, six representative rock samples from the study sites 275 were tested in the freezing laboratory at the Technical University of Munich. The two samples from Gámanjunni-3 are fine-grained greenish gneiss with indicated schistosity (density (ρ) = 3.1 g/cm³, porosity ca. 0.7 %). Some layers contain a significant high proportion of feldspar. The one sample from Nordnesfjellet (10 km NW of Gámanjunni) is a dark grey fine gneiss to quartz-rich mica schist (ρ = 2.8 g/cm³, porosity ca. 0.6 %). Minor slightly weathered but closed clefts in different orientations and the 280 anisotropy due to foliated minerals accounts for certain deviations in the measured laboratory arrays. These correspond to variations in the field where small-scale changes of meta-sediment rock types appear. From Mannen, the three greenish to dark grey gneiss samples are medium-grained with dark and light bands of biotite, quartz and feldspar. The sample Mannen03 is coarser with a higher proportion of cmbig feldspar minerals and therefore pronounced white bands. 285 The method of the resistivity calibration follows Krautblatter et al. (2010). The samples had a cuboid shape of c. 20x20x30 cm and a mass of 20 to 45 kg. All blocks were submerged in undisturbed tap water (473 µS cm -1 conductivity) in atmospheric pressure for at least 72 hours to approach close to natural fluid saturation and chemical equilibrium with the pore-surrounding rock material. Each sample was equipped with three lines (L = 21 cm) of four M6 stainless steel screws in a Wenner-type array to calculate 290 resistivity assuming an undisturbed half-space measurement geometry as the half space with the median depth of investigation controlled by electrode spacing is significantly smaller than the sample dimensions.
To overcome the challenge of loss of electrical contact upon freezing, the electrodes were fitted tightly ca. 10 mm deep into the rock. Contact grease was applied to the electrodes in order to further improve galvanic contact. Two Greisinger GMH 3750 thermometers were put in each specimen (5 and 20 mm 295 depth) to record both the near-rock surface temperature and the temperature at mean depth of investigation (Barker, 1989) every 30 seconds. We used an ABEM Terrameter LS, operating in monitoring mode, to obtain resistivity measurements every 15 minutes while the rock specimen where going through a freezethaw cycle between 10 °C to -5 °C in a 1 m³ cooling box equipped with a specially designed Fryka TK1041-LK-s ventilated cooling system controlled by a temperature probe close to the sample. The 300 cooling rate was controlled manually to not exceed a temperature gradient of more than 1 K between the temperatures at the rock surface and at mean depth of investigation. We used a low minimum current of 0.1 mA and high maximum voltage of 600 V to allow measurements even at high resistances supported by the high internal resistance of the ABEM Terrameter. Variance between repeated measurements (stacks) in the critical temperature interval of -2 to +2 °C was well below 1 %. 305

Field ERT and refraction seismic tomography
We used non-invasive geophysical surveys along profile lines in order to map permafrost at the individual sites and provide information on possible ice-rich or ice-poor zones in the ground. We used Electrical Resistivity Tomography (ERT) at all sites, and in addition refraction seismic tomography at Gámanjunni-310 3.
The electrical resistivity distribution of the subsurface is evaluated by injecting a current, and measuring the resulting electrical potential differences along the profile. The investigation depth depends mainly on the distances between the current electrodes employed along the profile and the profile length, with larger distance giving greater penetration depth. The obtained apparent resistivity measurements have to be 315 inverted using suitable inversion algorithms yielding the specific electrical resistivity distribution along the 2D profiles. High electrical resistivity is normally associated either with frozen conditions/ground ice occurrences or dry blocky layers. Low electrical resistivity points to (high) liquid water contents and unfrozen conditions (Hauck, 2002). At Gámanjunni-3, ERT is combined with seismic tomography. Seismic shots along the profiles produces P-waves, which velocity distribution and resulting travel times 320 are used and applied in a subsequent data inversion. The combination of ERT and seismic refraction are used to quantify to what extent the subsurface pores are filled with ground ice, water or air (Hauck et al., 2004;Mollaret et al., 2020), by applying the «4-phase model» (4PM) (Hauck et al., 2011). The 4PM is based on both ERT and refraction seismic tomographic surveys. The combination of both methods is able to distinguish between ice (high resistivity and medium P-wave velocities), water (low resistivity and P-325 wave velocities) and air (high resistivity, low P-wave velocities). For modelling details, we further refer to Hauck et al. (2011) and Mewes et al. (2017). The ERT profiles at Gámanjunni-3 and Mannen were either located on the plateaus, along the valley slopes or in the rock walls. In the rock walls we used steel screws drilled into the bedrock as electrodes, while outside the rock walls, normally steel rods were used. All measurements were carried out during late summer. At the Gámanjunni-3 site, four major datasets were obtained between 2012 and 2019, while 335 at the Mannen site, two major datasets were collected in 2012 and 2018, respectively. Location and details of the profiles are given in Figure 1b and Table 1.
ERT data acquisition was conducted with ABEM Terrameters (SAS1000 or LS) using Wenner or Wenner-Schlumberger protocols, with the Wenner protocol providing the best signal-to-noise ratio in difficult rock wall terrains (Dahlin and Zhou, 2004). All ERT profiles were inverted using common 340 inversion parameters within the software Res2Dinv (Loke and Barker, 1995). The colour coding followed the values obtained through the temperature-depending resistivity analysis performed in the freezing lab of TUM (cf. section 4.1), 3 to 5 inversion iterations showed sufficient convergence without overfitting. For profiles G-NVE1 and G-NVE2 at Gámanjunni-3 ( Figure 1b) the 4PM was applied to investigate relative contents of water, ice and air in the ground along the survey line (Hauck and Hilbich, 2018). 345 3.5. Bottom temperature of snow (BTS) survey BTS measurement is a simple and rapid method to estimate the possibility of permafrost conditions in field. The principle is that during the late snow season, under a snow cover of at least 80 cm or more, the temperature under the snow cover is decoupled from the temperature in the atmosphere, hence is governed 350 by heat flow from the ground (Haeberli, 1973). BTS values of below -3°C indicates a high probability of permafrost, while BTS>-2°C indicates no permafrost. This method has been widely used and validated in mountain areas, especially since the 1980s, and has been used also in Norwegian mountains for local permafrost mapping (e.g. Isaksen et al., 2002;Brenning et al., 2005;Lewkowicz et al., 2012). At both Gámanjunni-3 and Mannen BTS surveys were carried out on the 9. and 1. March 2017, respectively 355 ( Figure 1b). The survey was done using a long stick with a thermistor mounted at the bottom. At each site, is penetrated through the snow, and the BTS temperature is registered using a standard multi-meter. At least to measurements are carried out at each site to address small-scale variability.

3.6.
Ground temperature modelling 360 The transient heat flow model CryoGrid 2D (Myhra et al., 2017) solves the two-dimensional heat diffusion equation. The thermal properties (e.g. volumetric heat capacity and thermal conductivity) depends on temperature and material type. We used the MATLAB-based finite element method MILAMIN package (Dabrowski et al., 2008). CryoGrid 2D models conductive processes, thus nonconductive heat flow processes such as convective water or air flow are neglected. The model domain is 365 constructed as a 2D slice through a slope up to a chosen depth. An unstructured triangular mesh is generated for various subsurface thermal regions, i.e. regions with a distinct combination of water (liquid and ice), mineral, organic and air volumetric contents. The maximum allowed triangle area, which is a measure of the spatial resolution, increases typically with depth, and is assigned to every thermal region. We used bedrock thermal conductivity of 2.5 Wm -1 K -1 for both Mannen and Gámanjunni-3 slopes. Along 370 the right and left boundaries we prescribe zero-flux boundary conditions. The lower boundary (at 6000 m depth) is defined by a geothermal heat flux of 50 mWm -2 (Slagstad et al., 2009).
The upper boundary conditions are GST time series for the surface nodes. To calculate GST at each node, we first used SAT extracted from seNorge for elevations between valley bottom and top plateau at both sites, where SAT is linearly interpolated between the surface nodes, following a lapse rate of 0.64 °C 375 (100 m) -1 for Mannen and 0.48 °C (100 m) -1 for Gámanjunni (Magnin et al., 2019). We subsequently estimate GST using N-factors between 0 and 1 that link SAT and GST, hence account for the surface offset (Riseborough et al., 2008). The Nf-factor describes the winter surface offset due to snow coverage, where values close to 1 indicate no to little snow coverage, while values closer to 0 indicate a thick snow cover. Nt factors relate surface offsets during summer, which depends on factors such as vegetation cover, 380 direct solar radiation/shading, albedo and soil moisture. We applied empirical results derived from snow studies in Norway, relating Nf factors to annual mean snow height (Gisnas et al., 2013;Gisnås et al., 2016a). In our modelling we assume Nf=1 for steep rock walls (slope > 60°) that are frequently snowfree, whereas values of 0.25-0.5 can be used on the mountain plateaus (slope < 30°), where snow cover is thicker depending on precipitation and wind redistribution (Gisnås et al., 2016a). For the plateau, Nf 385 was set to 0.5 on Gámanjunni and 0.3 at Mannen due to higher snow cover at the latter site. The intermediate values of Nf are used at slope gradients between 30° and 60°.
The 2D geometry of the model domains has been extracted from a gridded 1-m-digital elevation models (www.hoydedata.no) along an approximately west-east (Gámanjunni) or south-west to north-east transect (Mannen) ( Figure A1). There are normally larger temperature gradients close to the surface than in deeper 390 layers. Hence, we constructed nodes with a distance of 0.05 m at the upper boundary. The subsurface thermal regions are constructed according to the geological profile for Gámanjunni ( Figure 2) and our mapping of surficial sediments along the profile using orthophotos for Mannen ( Figure A1), We then applied a stratigraphy, i.e. volumetric contents of the ground constituents, for the various surficial sediment classes at both surface and depth as presented in Westermann et al. (2013) (Figure A1). 395 For both sites, the model was initialized at deglaciation. When we assume warm-based conditions at the bottom of the ice-sheet (0 °C), we used the methods to reconstruct deglaciation curves and climate data described by Hilger et al (2021). We ran the model yearly until 1 st September 1873, then at weekly time steps, and in the period 1.1.2000-31.12.2018 at daily time steps.

Laboratory analysis
The laboratory analysis relates rock temperatures to measured electrical resistivity. The rock samples 420 from Gámanjunni-3 showed a similar pattern, with a sharp resistivity increase between 20 and 40 kΩm at the equilibrium freezing temperature of c. -0.5°. For areas with electrical resistivity above this range we expect negative temperatures, below we expect unfrozen conditions. The rock samples from Mannen revealed a sharp increase of resistivity below the equilibrium freezing temperature of c. -0.3°C, depending on bedrock type and freeze or thaw setting. The sharp increase in resistivity was between 15 kΩm and 425 close to 50 kΩm, a range we defined as the transition zone between freezing and thawing conditions.

Gámanjunni-3
Temperature monitoring and reconstruction -The three rock wall loggers at the unstable slope are all located at c. 1200 m a.s.l., and oriented towards south (RW-S), north (RW-N) and north-west (RW-NW), 430 respectively ( Figure 1). The RW-S site showed average rock surface temperatures of +0.05 °C between 2015 and 2019, while the two other loggers had clearly sub-zero temperatures of -1.6 °C and -1.3 °C for RW-N and RW-NW, respectively ( Figure 4a). During the same period mean SAT on the plateau was -3.1 °C, showing that rock wall temperature was at least +1.5 °C higher than air temperature. For the southoriented rock wall, close to +3 °C warmer temperatures than SAT on the plateau were recorded. The 435 measured RW temperatures represent a period of high temperatures in comparison to the reconstructed RW temperatures since 1870, as shown in Figure 4c. At Gámanjunni, the north-exposed logger showed sub-zero annual RW temperatures during the whole reconstruction period, while for RW-S positive annual averages were mostly estimated since 2000, along with some years during the 1930s. The reconstructed long-term series clearly demonstrates the warming since the LIA, which increased with 440 between 1 °C and 1.5 °C for both sites.
The GST loggers placed in the rockslide area ( Figure 1b) showed mostly average annual temperatures below 0 °C, with some exceptions. Average annual temperatures on the toe of the rockslide at c. 750 m a.s.l. revealed values between -1°C and -1.5°C, which normally indicate high permafrost probability. On the rock glacier, annual GST values are warmer and between -1°C and 0°C. Close to the moving block at 445 c. 1050 m a.s.l. a mean annual GST of -2°C is measured (Eriksen, 2018).
The few BTS measurements available (Figure 1b) all showed values below -3°C. These observations all together place the site into the discontinuous mountain permafrost zone (Magnin et al., 2019;Farbrot et al., 2013;Gisnås et al., 2016a).   Figure 5b). Time series derived from GB-InSAR in all the 18 points revealed velocity variations between 160 and 580 mm a -1 along GB-radar LOS for the 10 points on the rock glacier, and between 12 and 290 mm a -1 for the 8 points in the rockslide area ( Fig. 5c-d). The data showed a clear seasonality, which is attributed to the insecurity of the data gathered when the ground is snow covered. Monthly velocities during the snow-free months revealed in 490 most cases (1) higher displacement rates early in the melting season than later, and highest summer velocities during 2017 for both the rockslide and the rock glacier part of the instability (Figure 5d). According to the RW and SAT observations, 2017 was a particular warm year, and RW-S show abovezero mean annual RW temperatures (Figure 5c). The distribution of velocity over the rockslide area shows relatively similar velocities over most of the body including the sliding block, with 40-80 mm a -1 , 495 decreasing rapidly towards the slide front at 600 m a.s.l. (Eriksen et al., 2017) (Figure 6a-b). Highest velocities are obtained on the rock-glacier like landform forming the southern part of the instability, with surface velocities of > 100 mm a -1 (Figure 6b). GB-InSAR revealed similar displacement patterns as the TerraSAR-X data (Figure 6a).
Numerical modelling -The temperature field revealed by the 2D temperature modelling clearly showed 500 permafrost conditions in the slopes of Gámanjunni-3, down to 600-700 m a.s.l. at the end of the 19 th century , which includes most of the moving unstable part of the slope (Figure 7a). Since the end of the LIA, permafrost has warmed and degraded at its lower boundaries (Figure 7b), today probably only around half of the moving part of Gámanjunni-3 is influenced by permafrost, while the lower parts are modelled to be permafrost-free today (Figure 7c). On the plateau, maximum snow cover is around 1 m 505 thick, warming ground temperatures, while the steep rock walls are snow free. Snow cover and water content are sensible parameters for the modelling, modulating the permafrost temperature and geometry, as shown in the sensitivity tests provided in the appendix ( Figure A2). Maximum permafrost thickness of c. 300 m is modelled, which is in agreement with similar settings where we measure deep permafrost temperatures, such as in Tarfala in northern Sweden or Juvvasshøe in southern Norway (Isaksen et al., 510 2001). The lower permafrost boundary is modelled to be relatively stable during the 150 year period, demonstrating that relative thick permafrost can be expected in places. Geophysical surveys -The two long profiles down the slope from the moving block at ca. 1050 m a.s.l.
(G-NVE1, Figure 8a) and along the slope at c. 750 m a.s.l., crossing a rock-glacier like feature (G-NVE2, Figure 8b), show consistent patterns (see also Figure B1b). General resistivity values of the unfrozen and intact bedrock at depth seems to be around 1-10 kΩm. Surface resistivity values in the down-slope profile show a maximum between 700-900 m a.s.l. (40-100 kΩm) and further decrease to < 2 kΩm towards lower 545 elevation (Figure 8a). This altitudinal transition roughly coincides with the numerical temperature modelling (cf. Figure 7c). The profile G-NVE2 is oriented from south to north (Figure 8b). Both profiles show maximum resistive surface layers of up to 50 m in thickness. In combination with the GST values by Eriksen (2018) and NVE this resistive near-surface layer could indicate permafrost patches. The overall resistivity values within the rock glacier are lower (10-20 kΩm), and the more resistive surface 550 layer is somewhat shallower (c. 25 m) compared to the rockslide part in the centre of the profile.
The 3D profiles (G-TUM-S1-S4 and G-TUM-E1-E4) on the plateau clearly demonstrate the cooling influence of the NW-oriented rock wall, which becomes less pronounced with distance from the rock wall (Figure 9a-c, Figure B1b). In addition, low-resistivity areas (< 20 kΩm) are visible, probably indicating thawed conditions associated to water-filled fractures and cracks in prolongation of the exposed sliding 555 surfaces (Figure 9 a-b). These discontinuities oriented parallel and perpendicular to the profiles may account for differences in overlapping tomograms. The east-oriented profiles (TUM-E) are clearly influenced by the SW-oriented rock wall, with generally lower resistivity compared to TUM-S close to the NW-oriented rock wall. (> 60 kΩm) areas, which are highly influenced by the rockwalls (Figure 10, Figure B1c).

560
The G-EDY1 ERT-profile crosses the south-exposed rockwall, and shows lower resistivity values close to the rock wall surface (< 20 kΩm), decreasing towards the north-side of Gámanjunni-3 ( Figure 10a). The transition between the rockwall and the moving block below is covered by blocky scree material and show high resistivity (>100 kΩm) (Figure 10a). The G-EDY2 ERT-profile transverses the NW-oriented rock wall, with higher resistivity at the surface, clearly indicating the temperature differences between the 565 two rock faces (Figure 10b). Also here, high resistivity patches are found under the moving block and the cooled rock wall, while lower resistivity values are found under the snow-covered plateau (< 15 kΩm) (Figure 10b). The moving block is an area of high resistivity, with either possible permafrost influence or the presence of air-filled fractures. These higher values are also visible at a cross profile over the structure (G-TUM-Block) (Figure 10c).  Figure 10: ERT surveys over the back scarp at Gámanjunni-3, (A) G-EDY1 which is oriented over the exposed part of the slip surface. While the rock wall shows low resistivity, there is a clear transition towards the scree and the moving block with considerable higher resistivity. (B) G-EDY2 which is placed over the north-western exposed part of the rock wall. There are significant higher resistivity in both rock wall and plateau, illustrating that this side is more influenced by the cooler rock wall. For both profiles, the location of the rock wall loggers are indicated as 595 circles. (C): G-TUM-block oriented NW to SE over the moving block below the slip face. The resistivity is relatively high in relation to the plateau and below the block.
The combination of ERT and refraction seismic tomography within the 4PM revealed clear patterns in relation to possible permafrost and ice saturation (Figure 11, (Hauck and Hilbich, 2018)). At the G-NVE2 ERT profile, the considerable ice content values of up to 50-80% saturation suggests permafrost 600 conditions. We also see heterogeneities in vertical and horizontal directions along the profile line ( Figure  11b). The overall water contents are mostly low, except a possible fracture zone at depth (horizontal distance 180-230 m) (Figure 11a), showing greater ERT heterogeneities and low seismic velocities ( Figure C1b). Such a pattern is normally associated with high subsurface air contents. This kind of low P-wave velocity (i.e. high air contents) at greater depths is not common, and more prominently discussed 605 in the reports concerning the original seismic and ERT results of the area (GeoExpert, 2016). Unfrozen surface layers with no ice along G-NVE2 for the uppermost 5-10 m are further suggested by the analysis, while the northern part of the profile show a higher ice saturation within the upper 30 m (Figure 11b). Finally, overall dry conditions are suggested by modelled high air contents near the surface. The overall ice-content is probably low even if the model indicates high ice-saturation. This is related to the low-610 porosity bedrock (porosity c. 0.7%, (Leinauer, 2017)), which is in accordance with resistivity values of only ~10 kΩm, which are more atypical for high ice contents ( Figure C1b)). We prescribed a laterally homogeneous porosity model in the 4PM, which probably led to an overestimation of ice saturations due to low-porosity bedrock in the right-hand side of the profile.
Also along the slope of the instability (G-NVE1) the results suggest permafrost conditions. This is 615 especially evident for the upper part of the profile (Figure 11a). We also could identify an unfrozen surface layer, even if this feature is less visible due to geometry reasons, and a fracture zone characterised by high air contents (150-280 m horizontal distance, Figure 11a). We also note that the values for the geophysical profiles at the crossing points of the two profiles are quite correspondent. There, the transition between predominantly high ice saturations and high water saturations (which could be interpreted as a transition 620 zone between frozen and unfrozen conditions) is in both cases at around 40-50 m depth. However, also here we prescribed a gradient model for porosity which was homogenous along the profile, so such transitions in the data could also be due to change in material properties. However, we are relatively confident of the reliability of our analysis, as the 4PM resulted in similar values at the cross-over area of both profiles. 625 The results indicate permafrost conditions in both profiles, with an at least 30-50 m thickness. In addition, strong heterogeneities especially regarding de-compaction and fracture zones have been found, indicating significant air contents at larger depths, which is seldom found in thermally stable mountain permafrost bodies.

640
Temperature monitoring and reconstruction -The two rock wall loggers are oriented towards north (RW-N) and east (RW-E), respectively (Figure 1c). Both loggers showed positive annual average temperatures during 2015 and 2019, with +1.2 °C and +2.6 °C, respectively (Figure 4b). During the same period, the mean air temperature on the plateau was 0 °C, showing that rock wall temperature was at least +1 °C higher than air temperature. The reconstructed RW temperature series since 1970 revealed above-zero 645 temperatures in the rock wall, with an increasing trend (Figure 4c). The north-exposed rock wall certainly featured sub-zero temperatures in some cold years, such as in 2011.
GST loggers distributed along the rock scarp ( Figure 1b) showed mean GST between +0.9 and +1.6 °C, showing the warming influence of the thick snow cover. The TinyTag loggers in «Veslemannen» ( Figure  1c) recorded BTS temperatures in fractures between -1.3 and -1.8 °C in late April -early May 2015. In 650 late April 2016 the BTS recorded was between 0 and -2.3 °C. The mean temperatures recorded by these loggers are not representative, as they all lack complete annual data, but have to be around 0 °C in annual average. These data are described in more detail in Kristensen et al. (2021). Most of the BTS measurements were conducted close to the edge of the back scarp. While the BTS-values were mostly below -2 °C behind the north-exposed scarp, BTS values above -2 °C dominate behind the east-oriented 655 edge ( Figure 1c). The data confirm that permafrost patches likely occur along the plateau edge (Magnin et al., 2019). and 2016 seems to be associated with a lower snow cover during the winters of these years (Figure 5f). In terms of seasonal variations the cumulative movement plots indicate a step-wise pattern, with higher 665 velocities during spring/summer and lower velocities during fall/winter (Figure 5f). This is different to what was observed at «Veslemannen», where velocity accelerations started during the snow melt period, but was much higher and more variable in the fall period, and after heavy precipitation events (Kristensen et al., 2021).
The distribution of velocities over the moving slope body was derived from GB-InSAR, and shows 670 highest velocities in the upper part just below the back scarp and the plateau, with >20 mm a -1 ( Figure  6c). This high-velocity area defines scenario C for Mannen (Fig. 1c).
Simulated ground temperatures -The temperature field revealed by the 2D temperature modelling indicates possible permafrost conditions in the steep part of the slope during the onset of modelling at the end of the LIA, with permafrost thicknesses of between 50 and 100 m depending on model initialisation 675 procedure ( Figure 7d) and snow cover parameters ( Figure A2b). During the 150 years of the model run, steady warming reduced and degraded the modelled permafrost. However, isolated patches might still be possible in the steepest part with less snow, depending on model parametrisation in terms of snow coverage and water content in the model domain ( Figure A2b). Today, deep seasonal frost is modelled in the steep parts which is coincident with the rock wall measurements in the back scarp (Figure7f, Figure   680 4b). The plateau is heavily snow covered, and frost penetration is only possible laterally from the snowfree steep slopes.
Geophysical surveys -The ERT profiles at Mannen show generally higher resistivity than at Gámanjunni-3, probably related to different background resistivity of the bedrock and less surficial sediment cover. The 1 km profile (M-NGU1, (Dalsegg and Rønning, 2012)) covers both the plateau and the steep unstable 685 slope, and showed comparatively low resistivity (10 -40 kΩm) at depth (probably indicating the resistivity of the unfrozen intact bedrock), and higher resistivity at and below the scarp close to the surface down to c. 1150 m a.s.l. (50 ->100 kΩm) (Figure 12a). These high resistivity areas reveal crushed airfilled and well-drained bedrock and may contain permafrost patches (Dalsegg and Rønning, 2012). The ERT profile along the crest (M-TUM1-scarp) shows decreasing resistivity from NW to SE (Figure 12b). 690 High resistivity (> 100 kΩm, possibly indicating frozen conditions at depth) are observed close to the rock wall, while low resistivity (< 30 kΩm) dominates in the southeast, where the profile departs from the crest, and in the upper c. 20 m of the profile. The highest values (> 300 kΩm) are observed around a deep crack delimiting one of the moving blocks at Mannen, which defines a fractured zone with high porosity and unsaturated conditions (Figure 12b). 695 The rock wall profiles (M-EDY1-4) show mainly resistivity < 40 kΩm, also at depth, on the plateau, and higher resistivity (> 50 kΩm) over the back wall and over the fracture between the back scarp and the moving block (Figure 12c-f). Again, the highest values are measured below the back scarp over large fractures, which contain much air and are possibly partly snow and possibly ice filled. An exception is the M-EDY-2 profile (Figure 12d), where high resistivity is also obtained on the plateau. This profile has 700 2-m spacing, giving a higher resolution close to the surface, and therefore the coarse and high-porosity block cover on the plateau might result in higher resistivity values.  (Figure 1). The circles show the mean annual rock wall temperatures in the two loggers on site. The exposed back fracture below the slip surface is indicated by dashed lines, while the back scarp is indicated by a red line.

Hypothesis and uncertainties
Both Gámanjunni-3 and Mannen are considered as high-risk unstable slopes {Hermanns, 2013 #3893; Majala, 2016 #4282} and are continuously monitored. The movement was initiated several millennia after 720 deglaciation, thus climatic changes have been discussed as a factor influencing the dynamics of the instability (Hilger et al., 2021). Cosmogenic nuclide (CN) dating shows that the current displacements of the slopes is larger than the Holocene average (Böhme et al., 2019;Hilger et al., 2021), indicating atmospheric warming as a likely influencing factor. For both sites, we therefore hypothesise that permafrost warming and/or degradation might be a substantial explanation for the temporal displacement 725 pattern.
Our study combines a variety of methods, ranging from point observations (e.g. rock wall temperature measurements) via local surveys (e.g. geophysical measurements) to larger-scale modelling along an entire slope setting (e.g. geothermal modelling). Each method has uncertainties, and includes different 730 pre-conditions, representing many pieces of a puzzle to form a consistent picture. For example, the heat flow modelling represents the lower extent of possible permafrost at Gámanjunni-3 as indicated in the ERT surveys, but does not represent local thawed areas and variations indicated by the ERT.
The methods used are independent of each other, and may contradict in places, but result in an overall 735 explainable pattern. The temperature modelling does not account for fractures and other structures in a rockslide area (e.g. where water can penetrate); Cryogrid-2D is a two-dimensional model purely based on heat conduction, which results in a smoothed and simplified version of reality {Myhra, 2017 #3967; Myhra, 2019 #4134}. The ERT profiles were measured in rough terrain, both the resistivity values and data noise are very sensible to cracks and fractures, strong topographic variations, or local water 740 penetration. This leads to a high variability of resistivity and may produce inversion artefacts, such as in the transition between plateau and rock wall at Gámanjunni (Figure 10), which is also represented by the partly high root mean square errors (RMS) of the inverted tomograms, which vary between 6 and 20%. Therefore, comparing ERT and a more large-scaled temperature model is not meaningful on a local site level. Our ERT and 4PM results for Gámanjunni clearly show potentially frozen areas within the rock 745 slide area, disappearing down-slope. The potentially frozen areas are at a depth of c. 30-50 m, which is close to the results of the temperature modelling at these places. However, locally thawed areas in the rockslides and possible taliks as indicated in the ERT are certainly realistic, but cannot be covered by the simplified heat flow model. 750

Permafrost conditions and recent ground thermal development
At Mannen, we do not measure sub-surface RW or GST temperatures below 0°C on an annual average at present, except for shaded locations in fractures (Kristensen et al., 2021). Over the last 140 years, MAAT 755 has increased, and since the cooling in 1970s the temperature rise was around +1.5°C (Figure 4c). Rock wall temperatures oriented towards north must have had sub-zero surface temperatures during several periods of the last 150 years, indicating permafrost development in the past in shaded topographic settings (Figure 4c). This confirms the modelling by Magnin et al. (2019), and the results from Kristensen et al (2021) for "Veslemannen" which indicate sporadic permafrost zones at Mannen in certain locations such 760 as fractures, snow-free patches and in shaded locations. It is also well documented that cracks and fractures in rock walls locally significantly decrease ground temperatures (Magnin et al., 2015a;Hasler et al., 2011). This is also supported by the ERT surveys showing highest resistivity values close to the rock wall and large fractures (high porosity), which may be partly filled by ice ( Figure 12). However, there are no observations of ice in the fractures as in the Jettan rockslide in northern Norway, where permafrost is 765 observed and probably influences seasonal kinematic variations (Blikra and Christiansen, 2014). The thermal modelling indicates that permafrost patches could develop at and below the upper scarp, and that the unstable area is modelled as being cooler than the plateau (Figure 7), a pattern also visible in the long ERT profile (Figure 12). The mountain plateau of Mannen can hardly develop permafrost because of a very thick and long-lasting snow cover.

770
For Gámanjunni-3, MAAT have risen over the last 140 years, and the rise was around +1.8°C since 1880. Estimated rock-wall temperatures in all orientations have been mostly negative between 1880 and 2020. Since c. 2000, however, the south-oriented rock wall showed mean annual temperatures close to or above 0°C (Figure 4c). Permafrost warming and possible degradation might have accelerated since c. 2000, which could influence the geotechnical properties of the site. The ERT measurements suggest permafrost 775 at Gámanjunni-3, but resistivity differences between topographic aspect and laterally over the plateau indicate changes in ice content and ground temperature including the potential occurrence of taliks {Krautblatter, 2010 #2104; Gruber, 2007Gruber, 2009 #4229}. Those can form during general atmospheric warming, extreme warm years or along water-filled fractures (Luethi et al., 2017). These processes result in high resistivity variations (Hilbich et al., 2008;Krautblatter and Hauck, 2007;Mollaret 780 et al., 2019). This interplay, together with air and water advection in fractures produces a complicated thermal pattern, which is not reproduced by our heat flow modelling. The pattern is further highly modulated by snow cover, which in Scandinavian high-mountain settings is highly variable due to wind re-distribution (Gisnås et al., 2014;Gisnås et al., 2016b). This redistribution of snow is the major source for high spatial variability of surface temperatures (Haberkorn et al., 2015), which can vary with several 785 °C (Gisnås et al., 2014;Marmy et al., 2016;Magnin et al., 2015a;Magnin et al., 2017;Magnin et al., 2015b;Hasler et al., 2011;Haberkorn et al., 2017). However, ice-free north-oriented rock walls show a cooling influence on the surrounding subsurface.
In summary, for both sites, we can expect at least local permafrost conditions, clearly more widespread 790 at Gámanjunni-3 than at Mannen, and a warming with accelerated pace during the last two decades, following similar observations all over Europe (Etzelmüller et al., 2020).

795
Spatial pattern of movement -The spatial distribution of surface displacement is slightly different at the two sites. At Mannen, relatively high displacement rates of c. 20 mm a -1 are measured in the upper part of the unstable slope, while low velocities of < 5 mm a -1 dominate the other parts (Figure 6c). At Gámanjunni-3 displacement rates of > 50 mm a -1 are registered over most of the mapped rockslide area, with some higher values in the upper part. Maximum velocity values of >150 mm a -1 are observed in the 800 rock glacier in the southern part of the area (Figure 6a-b).
Displacement rates, ground temperatures and ERT results were related along the ERT lines G-NVE1-2 and M-NGU1 (Figure 13). At both sites GT is clearly associated with measured resistivity, confirming the lab analysis and our interpretation of possible permafrost at these sites (Figure 13a-b). For Gámanjunni-3 we observe a positive relationship between electrical resistivity and displacement rates 805 (more displacement when higher resistivity) and associated lower displacement with higher ground temperatures along the longitudinal profile over the rockslide mass (Figure 13c-d).
At Mannen similar observations were made, but are not that clear (Figure 13f). This seems contradictory as permafrost is seen as a stabilising factor for slopes (Gruber and Haeberli, 2007;Krautblatter et al., 2013). An explanation for this behaviour can be found in e.g. Davies et al. (2001) who found Factor-of-810 safety (FS) values below 1 for ice-filled fractures close to the melting point, and FS-values at 1 or above when the ice has melted or is very cold. The stability of both ice in fractures and rock-ice interfaces strongly declines with warming temperatures below 0°C (Mamot et al., 2018). Both at Gámanjunni-3 and Mannen possible ice occurrences are close to the melting point and thus deformable.
It is here important to note, that other factors than the presence or absence of permafrost may govern the 815 observed spatial differences in displacement rates. Changes in the geo-mechanical behaviour may, for example, cause similar variations of displacement rates as large unstable slopes are typically controlled by pre-existing weak structures and their reactivation by gravitational processes. This applies also when analysing the cross profile (G-NVE2) at Gámanjunni-3, which covered both the rock glacier and the rockslide. Here, the geophysical surveys indicate that an unfrozen near-surface layer 840 (ice contents ~ 0 in the uppermost 5-10 m) overlies frozen areas. Further on, lower resistivity values in 20-30 m depth are measured in the rock glaciers than in the moving part of the rockslide, even if velocities in the rock glacier landform is much higher (Figure 13e). These observations may relate to different processes of movement in the two parts of the instability. While in the rockslide the movement is influenced by possible ice deformation due to inferred higher ice content with depth in this part or block 845 movement below a frozen layer, the rock glacier movement seems governed by movement related to water/ice mixtures close to the melting point, where shear strength is greatly reduced and ice deformation increases {Arenson, 2002 #2018; Davies, 2001 #1417; Cicoira, 2019 #4261}.
The rock glacier has markedly higher velocities . Such velocities are common for 850 rock glaciers in alpine environments (Kääb et al., 2007), and often facilitated by a block motion within a deforming massive ice body {Arenson, 2002Haeberli, 2006Haeberli, 1998Cicoira, 2019 #4261; #4262}. The ERT measurements show a local resistivity peak under the rock glacier (at c. 120 m distance, Figure 8b), and GST loggers indicate permafrost presence in the landform (Figures 1b and 8b). A rock glacier in the neighbouring valley from Gámanjunni-3 (Adjet rock 855 glacier) had velocity averages increasing from ~4.9 to ~9.8 m a -1 (2009-2016) and maximum velocities from ~12 to ~69 m a -1 . There, permafrost warming, topographic controls, and increased water access to deeper permafrost layers and internal shear zones have been used to explain the kinematic behaviour at Adjet rock glacier  are observed in these cracks, governing movement rates (Blikra and Christiansen, 2014). The ERT measurements indicate very high resistivity values in this zone (> 100 kΩm), indicating either air or snow/ice fill. However, there are no direct observations of ice. Hörnligrat ridge (Matterhorn, Swiss Alps). They showed that reversible displacements dominate in winter while irreversible enhanced fracture displacements are mainly observed in summer, likely indicating thawing-related processes (e.g. meltwater percolation into fractures). However, this behaviour can strongly differ from one fracture to another as seen at the Aiguille du Midi 3842 a.s.l, France (Guillet et al., unpublished). A similar pattern is observed for «Veslemannen» where meltwater infiltration and 880 thawing of seasonal frost along with precipitation episodes are discussed (Kristensen et al., 2021). The instability on the Zugspitze crest (Germany/Austria) shows movements of c. 20 mm a -1 and highest displacement rates during summer, with a reduction of up to 85% during the remaining seasons (Mamot et al., 2020). Gischig et al. (2011) found high winter and low summer velocities at the Randa rock slope instability (Switzerland), and no correlation with rain fall. They could reproduce this pattern by thermo-885 mechanical modelling, where surface temperature governed the variation. In the Jettan site near Gamanjunni-3 Blikra et al. (2014) documented ice in fractures, and highest velocities during summer, probably caused by melting of ice-patches in fractures.
The possible higher early spring and summer velocities and lower displacement rates during fall and winter might be related to high water input in the fractures due to snow melt causing hydraulic/hydrostatic 890 pressures and contributing to the melting of ice/snow in fractures formed during the winter. During summer and fall, the fractures might be free of ice/snow at the end of the melting season and water infiltration might have less impact. The lower velocities at Mannen during years with lower snow cover (Figure 5f) also supports this interpretation. However, only thermo-mechanical modelling, like applied by Gishig et al (2011) or Mamot et al. (2020), may increase the understanding of how this signal can 895 influence rock mass deformation.
In both study sites, long-term Holocene displacement variations seem to be related to climate signals (Hilger et al., 2021). These observations agree also with other studies, e.g. Philips et al (2017) report on 6000 years old ice derived from tension crack at a rock pillar. The rock pillar collapsed in 2014, and had a volume of around 150000 m 3 . Permafrost aggradation and degradation decrease the stability of intact 900 rocks, both by weakening rock bonds (rock fatigue) and by critical and subcritical fracture propagation at sites with strongly varying cryostatic and hydrostatic conditions (Draebing and Krautblatter, 2019;Voigtländer et al., 2018). There certainly has been a long-term warming of our study sites since the LIA, and an accelerated warming since c. 2000. This warming trend has been documented all over Europe (Etzelmüller et al., 2020), and is responsible for permafrost degradation in Norway (Borge et al., 2017), 905 possibly influencing both rock glacier velocities and landslide triggering Frauenfelder et al., 2018).
In summary, both sites show corresponding seasonality with increased early summer velocities. Combined with the knowledge of at least discontinuous permafrost to patchy permafrost at the sites, snow and ice melt processes with associated water drainage in cracks are realistic explanations for a possible 910 seasonality. There is evidence that the recently measured higher displacement rates in relation to Holocene values (Hilger et al., 2021) may be related to a warmer atmosphere, and can accelerate into the future. The triggering of «Veslemannen» described in detail by Kristensen et al. (2021) might be a first sign.
For Gámanjunni-3, a rapid acceleration of the rock-glacier like landform forming the southern part of the rockslide, is possible, as described for various cases in the recent past in northern Norway (Eriksen et al.,915 2018), in the European Alps (Delaloye et al., 2008) and in Central Asia (Kääb et al., 2020). It could possibly lead to the triggering of secondary rock falls or debris flow, as described elsewhere (Lugon and Stoffel, 2010;Kummert et al., 2018), or movement can stop when the deforming ice is melted out.

920
The following conclusions are drawn based on this study:  Temperature measurements, numerical modelling and geophysical soundings demonstrates the existence of permafrost at both study sites. At Gámanjunni-3 permafrost seems to extend down to 700 m asl. today, while at Mannen sporadic pockets of permafrost are possible.

925
 Surface air and ground temperatures have warmed significantly since c. 1900 with +1 °C and 1.5 °C, and highest temperature are measured and modelled since 2000 at both study sites.  Displacement rates of Gámanjunni rockslide co-vary significantly with sub-surface resistivity and modelled ground temperatures. Increasing displacements rates are seems to be associated with subzero ground temperatures and higher ground resistivity. This might be related to the presence of 930 ground ice in fractures and pores close to the melting point, facilitating increased deformation.  A seasonality of displacement has been observed, with increased velocities during late winter and early summer at both sites. This pattern may be linked to the timing of snow melt and water infiltration, leading to high water pressure. At Mannen, inter-annual variations may be related to snow cover thickness.

935
 The rock glacier associated to the Gámanjunni-3 rockslide show two to three times higher velocities (> 100 m a -1 ) and lower electrical resistivity than the rockslide part. The movement mechanism are clearly different for both systems, and a mixture of water and ice contributing to the rock glacier movement is suggested.  The permafrost in the study sites has certainly warmed and probably degraded since the LIA, with an 940 accelerated pace since c. 2000. This atmospheric and associated permafrost warming might be a factor for the high deformation rates in relation to the Holocene  A possible permafrost degradation and probable thawing at Gámanjunni-3 may result in the destabilization of the upper part of the plateau south of Gámanjunni-3, considerably increasing the susceptible volume for worst-case collapse scenario.

APPENDICES APPENDIX 1 -Thermal modelling
The thermal modelling requires a set of parameters and boundary conditions. For our modelling we 950 defined zones with crisp boundaries, defining surface sediment cover, bedrock or fractured bedrock. For each of the zones a set of material properties were defined, following the system in earlier publications (e.g. Westermann et al., 2013). Most cover sediments are quite coarse-grained, with no organic material ( Figure A1). For Gámanjunni-3 we used a well-defined geological model to delineate the rockslide Böhme et al., 2019), while for Mannen the instability is much less defined. In the 955 latter surficial material is thin, and bedrock at the surface in the slope, and coarse blocks at the plateau, dominates. For the sensitivity analysis we varied forcing air temperature, snow cover (by changing the nF factor) and the water content, the latter only for Gamanjunni because of bedrock dominance with assumed low water content for the Mannen site ( Figure A2). Permafrost distribution and geometry varies with these parameters, indicating that reality is probably somewhere in between. Especially for the 960 Mannen site located in the sporadic permafrost zone, the parameter variations show the influence of less snow or cooler SAT on the possible permafrost presence at the site ( Figure A2b). profile. MTA = maximum triangle area, which is a measure describing the spatial resolution of the triangles in the employed finite element method solver. The table shows the stratigraphy chosen for the different regions, along with the depth parametrisation and volumetric contents of water/ice, mineral, organic and air components, following Westermann et al. (2013). Figure A2. Sensitivity plots for the modelled ground thermal regime of (A) Gámanjunni-3 and (B) Mannen. The "Main" run is the presented run in Figure 7, the subplots show modelled ground temperature in response to changes in nF factor ("nF+0.1" means that nF is increased by 0.1; "nT-0.1" means that nF is decreased by 0.1), forcing SAT ("T+1 °C" means that SAT is 975 increased by 1°C; "T-1 °C" means that SAT is decreased by 1 °C) and water content in the subsurface ("50% water" means that water content is halved and the remaining fraction is added to the mineral fraction; "200% water" means that water content is doubled by reducing the mineral fraction). GT and permafrost geometry changes in response to these variations. It is noteworthy at Mannen that only small changes in snow or forcing temperatures would produce considerably more permafrost in the unstable area. Due to limited assumed sediment cover for the Mannen site, we did no sensitivity plot for the water content 980 in bedrock, which is low.

APPENDIX B -3D visualisation of ERT profiles
We projected all ERT profiles in a 3D topographic model in MATLAB (© Mathworks) ( Figures B1 and  B2). The inversion was performed with the same parameters, and the colours follow the transition between 985 possible thawed to possible frozen derived from the laboratory analysis ( Figure 3). The profiles in the rock walls and steep slopes do not always follow straight lines because of security issues in the field. In the plots we indicated the moving block at Gámanjunni-3, and the plateau crests at both sites. It is clear that the inversion procedures may have produced artefacts at strong topographic transitions. However, the clear patterns described in the main text associated to fractures and snow-free rock walls are visible. While Figure 11 in the main text shows and discusses the results of the 4PM for two profiles at 1055 Gámanjunni.3, the Figure C1 shows the original inverted ERT and RST tomograms of the two profiles. NVE-2. Note that the profiles are subsets of G-NVE-1 and -2, and thus shorter than shown in Figure 8 and Figure  B1. Note also that the colours for the electrical resistivity do no correspond to the colour scale derived from the laboratory analysis (Figure 3), but are the original results first presented in Hauck and Hilbich (2018). The possible fractures zones mentioned in the manuscript are indicated as a box, while the possible lower permafrost limit is drawn as a line in (A). 1065 reflectors were initially funded by the Norwegian Space Agency, and the processing of this data is now carried out as part of the InSAR Norway ground motion service, hosted by NGU (http://insar.ngu.no). Jan Steinar Rønning from NGU provided data raw data for the ERT profile M-NGU at Mannen.

Contributions
BE took the initiative for the study, and coordinated the synthesis of the various data sets. He wrote the first drafts of the manuscript, and designed most of the figures. JC carried out the numerical modelling of the ground thermal regime at both sites, provided background information and modelling results and developed visualisation tools for the ERT profiles. SW developed much of the principles of the numerical 1090 modelling code, and supervised the analysis. FM coordinated and participated in the field investigation of the ground temperature and ERT rock wall surveys at both sites. PAD and EM participated at the ERT surveys at Mannen and PAD and LR at Gámanjunni-3. AA, LK and IS contributed with on-site knowledge, interpretation and displacement data from laser, GNSS and GB-InSAR, on behalf of NVE. BJ, JL and MK contributed with the laboratory analysis of the bedrock samples at the Technical 1095 University of Munich (Germany), and ERT profiles conducted in field at both sites. RH and MB from NGU contributed with a structural-geological model of the sites. CHa and CHi from the University of Fribourg, Switzerland, applied the 4-phase-model and submitted a report from the two long profiles at Gámanjunni. They also commented and improved the inversion and interpretation of all the other ERT profiles. HE provided and interpreted the TerraSAR-X data from Gámanjunni, along with ground 1100 temperature logger data. TRL coordinated installation and development of the satellite corner reflector processing chain and supervised the TerraSAR-X InSAR processing. All authors contributed actively to the final versions of the manuscript.

Competing interests:
The authors declare that they have no conflict of interest. 1105 7.