Warming permafrost and active layer variability at Cime Bianche , Western European Alps

The objective of this paper is to provide a first synthesis on the state and recent evolution of permafrost at the monitoring site of Cime Bianche (3100 m a.s.l.) on the Italian side of the Western Alps. The analysis is based on 7 years of ground temperature observations in two boreholes and seven surface points. The analysis aims to quantify the spatial and temporal variability of ground surface temperature in relation to snow cover, the small-scale spatial variability of the active layer thickness and current temperature trends in deep permafrost. Results show that the heterogeneity of snow cover thickness, both in space and time, is the main factor controlling ground surface temperatures and leads to a mean range of spatial variability (2.5± 0.1 C) which far exceeds the mean range of observed inter-annual variability (1.6± 0.1 C). The active layer thickness measured in two boreholes at a distance of 30 m shows a mean difference of 2.0± 0.1 m with the active layer of one borehole consistently deeper. As revealed by temperature analysis and geophysical soundings, such a difference is mainly driven by the ice/water content in the sub-surface and not by the snow cover regimes. The analysis of deep temperature time series reveals that permafrost is warming. The detected trends are statistically significant starting from a depth below 8 m with warming rates between 0.1 and 0.01 C yr.

For monitoring the huge spatial variability of mountain permafrost, a number of monitoring sites has been established through the Alps during the last years (e.g., Cremonese et al., 2011).At present the collection of temperatures in boreholes provides the best direct evidence of permafrost state and evolution.Nevertheless, the combination of geophysical methods and thermal monitoring is particularly suitable for long-term monitoring of mountain permafrost because it provides crucial information on ground ice/water content and structure (e.g., Hilbich et al., 2008;Haeberli et al., 2010;PERMOS, 2013).The site of Cime Bianche has been designed with the main objective of monitoring the spatial variability of mountain permafrost.Moreover Cime Bianche site is a permanent observatory in the southern side of the European Alps, a region where permafrost observa-P.Pogliotti et al.: Warming permafrost and active layer variability tions are more sparse and younger compared to the northern side (e.g., Cremonese et al., 2011), and where significant climatological differences occur (e.g., Frei and Schär, 1998;Evans and Cox, 2005).
At Cime Bianche, the spatial variability of ground surface temperature (GST) is measured because it has crucial implications on the initialization, calibration and validation of numerical models (e.g., Guglielmin et al., 2003;Noetzli and Gruber, 2009;Hipp et al., 2014), and it is often used as an indicator of permafrost occurrence.One of the main challenges in the study of GST variability is the quantification of the thermal effect of snow cover given the influence that it can have on thermal regime trough different processes (Zhang, 2005;Luetschg et al., 2008;Guglielmin et al., 2014b).On gentle slopes, snow cover mostly causes a net increase of mean annual ground temperature due to the insulating effect during winter, but timing of onset and meltout, duration, thickness and interaction with ground surface characteristics strongly control the local magnitude of this effect (Hoelzle et al., 2003;Brenning et al., 2005;Gruber and Hoelzle, 2008;Pogliotti, 2010;Gubler et al., 2011;Rödder and Kneisel, 2012).Although a number of studies focused on snow-GST interaction exists (e.g., Zhang, 2005, for a review), little is known on its spatial and temporal variability especially over complex alpine terrains.
Beside the GST, the active layer thickness is also measured at Cime Bianche.The World Meteorological Organization recognizes permafrost and active layer as one of the essential climate variables selected for quantifying the impacts of climate change (e.g., Harris and Haeberli, 2001).In the Alps, the active layer is of particular interest because it directly affects slope processes (e.g., Fischer et al., 2012) and infrastructures stability (e.g., Bommer et al., 2010;Springman and Arenson, 2008).The active layer dynamics are controlled by a number of variables such as air temperature, solar radiation, topography, ground surface characteristics, ground ice/water content and the timing, distribution and physical characteristics of the snow cover (Zhang, 2005;Luetschg et al., 2008;Scherler et al., 2010;Wollschläger et al., 2010;Zenklusen Mutter and Phillips, 2012).As a consequence, the active layer thickness has an high spatial and temporal variability (Anisimov et al., 2002;Wright et al., 2009) which in the Alps may occur at very small scale.
Compared to the active layer which responds more to short-term variations like seasonal snow and air temperature conditions, the deep (10 to 200 m) thermal regime of permafrost reacts to long-term changes in climate (Beltrami, 2002).The deep permafrost temperature regime is a sensitive indicator of the long-term climate variability and changes of the surface energy balance (Romanovsky et al., 2002).The trend analysis of deep temperature time series allows the detection of signals of past and ongoing changes of permafrost (e.g., Isaksen et al., 2001).
The overall objective of this paper is to provide a first synthesis on the state and recent evolution of permafrost at Cime Bianche.In particular we present (i) the spatial and temporal variability of GST and its relation with snow cover (ii) the small-scale (30 m) ALT differences and (iii) the warming trend of deep permafrost temperature.

Site description
The Cime Bianche monitoring site is located in the Western Alps at the head of the Valtournenche valley (Valle d'Aosta, Italy, 45 • 55 N-7 • 41 E) on the Italian side of the Matterhorn, at 3100 m a.s.l.(Fig. 1).The site is located on a small plateau slightly westward, with degrading characterized by terracettes, convexities and depressions that result in a high spatial variability of snow cover thickness during winter.
The bedrock lithology is homogeneous, mainly consisting of garnetiferous micaschists and calcschists belonging to the upper part of the Zermatt-Saas ophiolite complex (Dal Piaz, 1992).The bedrock surface is highly weathered and fractured, locally resulting in a cover of coarse-debris deposits with a thickness ranging from few centimeters to a couple of meters.The presence of small landforms like gelifluction lobes (between 0.6 and almost 5 m in length) and sorted polygons of fine material (with diameters ranging between 0.6 and 3.4 m) suggests the presence of permafrost.
The climate of the area is slightly continental.The longterm mean annual precipitation is reported to be about 1000 mm yr −1 for the period 1931-1996 (Mercalli and Cat Berro, 2003) while the in situ records show a mean of 1200 mm yr −1 for the period 2010-2013.The mean annual air temperature (MAAT) is about −3.2 • C (mean 1951-2011).Mean monthly air temperatures are positive from June to September, while February and July are, respectively, the coldest and the warmest months.The site is very windy and mainly influenced by NE-NW air masses.The wind action strongly contributes to the high spatial variability of snow cover thickness.
Permafrost research in the area started in the late 1990s (Guglielmin and Vannuzzo, 1995) with repeated campaigns of measurements of the bottom temperature of snow (BTS) and glaciological observations showing that the monitoring site was probably ice covered during the climax of the Little Ice Age.In 2003, as a preliminary investigations for site selection, the potential permafrost occurrence in the area was assessed using results from BTS, vertical electrical soundings and ERT (electrical resistivity tomography) and the application of numerical models like Permakart (Keller, 1992) and Permaclim (Guglielmin et al., 2003).

Instrumentation
The site instrumentation started in 2005 and has been progressively upgraded during the following years.The current setting is nearly unchanged since August 2008 and consists of two boreholes, a spatial grid of ground surface temperatures measures and an automatic weather station (AWS).

Boreholes
A deep (DP) and a shallow (SH) borehole, reaching a depth of 41 and 6 m, respectively, located at a distance of about 30 m (Fig. 1), have been drilled in 2004 with core-destruction method.Both boreholes are 127 mm in diameter with a 60 mm sealed PVC pipe for sensor housing.The boreholes are equipped with thermistor chains based on resistors type YSI 44031 (resolution 0.01 • C, absolute accuracy ±0.1 • C).The entire setup (thermistor chains attached to the data logger) have been calibrated by the manufacturer before the installation.Sensors depths in meters from the surface are 0.02, 0.3,0.6,1,1.6,2,2.3,2.6,3,3.3,3.6,4,4.6,5.9 for SH and 0.02,0.3,0.6,1,1.6,2,2.6,3,3.6,4,6,8,10,12,14,15,16,17,18,20,25,30,35,40,41for DP.In each borehole, the shallower sensors (0.02 and 0.3 m) are cabled on two independent chains and are used to measure the ground surface temperature outside the PVC tube in order to avoid the thermal disturbance of the casing.Temperatures are sampled every 10 min and recorded by a Campbell Scientific CR800 data logger.The system is equipped with a GPRS module for daily remote data transmission.

Ground surface temperature grid (GSTgrid)
A small grid (40 m × 10 m) is used for monitoring the spatial variability of GST.The grid consists of five nodes: four at the corners and one in the center (Fig. 1).Each node is equipped with two platinum resistors, PT1000 (resolution 0.01 • C, accuracy ±0.1 • C), buried in the ground at depths of 0.02 and 0.3 m (according to Guglielmin, 2006).Ground temperatures are recorded hourly by a Geoprecision D-Log12 data logger.
For the analysis, GST measured at the two boreholes is also included; thus data from seven nodes are used for the analysis.Ground surface at each node is mainly characterized by coarse debris with a fine matrix of coarse sand and fine gravel.At each node, the sensors are placed in the matrix thus local ground conditions are nearly homogeneous between all nodes.In contrast, snow cover depth and duration sharply differ across the grid nodes.For this reason, based on field observations and temperature time series analysis (Schmid et al., 2012) the data set is divided in snowfree and snow-covered nodes.The first group includes three nodes characterized by shallow or intermittent winter snow cover, while the latter group includes four nodes that clearly show a long lasting deep snow cover damping temperature oscillations during winter (Fig. 1).

Automatic weather station
An AWS has been installed just above the borehole SH since 2006.Air temperature and relative humidity, atmospheric pressure, wind speed and direction, incoming and outgoing short-and long-wave solar radiation and snow depth are recorded every 10 min by a Campbell Scientific CR3000 data logger.The system is equipped with a GPRS module for the daily remote data transmission.In September 2011 a second snow depth sensor was installed in the surroundings of the DP borehole.Finally, solid and liquid precipitation has been measured since January 2009 by an OTT Pluvio 2 system.

Data analysis
This section reports a short description of the methods used for the calculations of synthesis parameters considered in this study.
MAGT is the mean annual ground temperature at a specific depth (m) (e.g., MAGT 10 ).
MAGST is the mean annual ground surface temperature.ALT is the active layer thickness defined as the maximum depth (m) reached by the 0 • C isotherm at the end of the warm season.It is calculated considering the maximum daily temperature at each sensor depth and interpolating between the deepest sensor with positive value and the sensor beneath.The maximum of the resulting vector and the corresponding day are named ALT and ALTday, respectively.This procedure is applied on the warmest period of the year, here fixed from 1 August to 30 November.The uncertainty of ALT estimation is evaluated considering the amplitude of thermistors noise (inferred from calibration of the manufacturer) and the interpolation distance between the sensors.Considering these factors, the uncertainty of ALT estimation is ±0.15 m in borehole SH and ±0.2 m in borehole DP.
TTOP is the MAGT at the top of the permafrost table (Smith and Riseborough, 1996).It is calculated by interpolation of the MAGT at depth of the ALT that is considering the first sensors above and below the ALT.
THO is the thermal offset within the active layer and is computed as TTOP-MAGST (Burn and Smith, 1988).Zero annual amplitude oscillation (ZAA) is the depth beneath which there is almost no annual fluctuation (AF) in ground temperature, nominally smaller than 0.1 • C (van Everdingen, 2005).The AF is calculated at each sensor depth as the difference between annual maximum and annual minimum of the mean daily temperatures.The ZAA is calculated by interpolation between the deepest sensor with AF greater than 0.1 • C and the sensor beneath.When necessary, a moving average, with a window of 360 days, is applied on deep nodes data (below 8 m) before daily aggregation to remove electrical noise (±0.01 • C).
All the parameters listed above, with the exception of ALT, are computed considering the hydrological year (beginning 1 October) as a reference period.All the analyses are performed with the free statistical software R (R Core Team, 2014).When appropriate, the variability of the results is expressed in terms of standard error (se = sd/ √ n, where se is standard error, sd is standard deviation and n is the sample size).

Snow cover duration and snow-free days
In order to investigate the effect of snow cover duration and air temperature on MAGST, the method of Schmid et al. ( 2012) is applied on snow-covered nodes using the sensors at 0.02 m.This method allows us to infer the date of snow onset (OD) and the date of snow melt (MD) from the amplitude of ground temperature oscillation.Subsequently, starting from OD and MD, it is possible to calculate (i) the duration of snow cover (SD, Fig. 2) as the number of days between OD and MD and (ii) the number of snow-free days as the sum of remaining days of autumn and summer.The latter period is used as reference for calculating the mean annual air temperature of snow-free days (MAATsf, Fig. 2).Step 2: seasonal detrending (thick black line).
Step 3: trend fitting (dashed red line).Vertical dashed lines represents 1 October, materializing the limits of the hydrological years.

Trend analysis
In order to look for linear trends that might reflect warming, two non-parametric methods are applied to borehole temperatures: Mann-Kendall test (MK) (Mann, 1945;Kendall, 1948) and Sen's slope estimator (SS) (Sen, 1968).These methods are commonly used to assess trends and related significance levels in hydro-meteorological time series such as water quality, stream flow, temperature and precipitation (e.g., Gocic and Trajkovic, 2013;Kousari et al., 2013).The reason for using non-parametric statistical tests is that they are more suitable for non-normally distributed data and are not sensitive to outliers or abrupt changes.
The procedure chosen includes (i) the pre-whitening of the data to remove the lag-1 autocorrelation components as recommended by von Storch and Navarra (1999) (see also Hamed, 2009 andBence, 1995), (ii) the fitting of the trend's slope with SS and (iii) the testing of trend significance level (p value) with MK.Such a procedure is implemented in the R package zyp (Bronaugh et al., 2013).
Given the short climatological time span of the borehole observations, a seasonal detrending is recommended, as suggested by Helsel and Hirsch (1992), for better discerning the long-term linear trend over time.Thus, a seasonal decomposition based on loess smoother (Cleveland, 1979;Cleveland et al., 1990) is applied to the monthly aggregated time series of each borehole before applying SS and MK (Fig. 3).Such a seasonal detrending method is implemented by the R function stl (R Core Team, 2014).

Geophysics
At the end of summer 2013, two geophysical surveys have been realized with the objective to assess the composition of The Cryosphere, 9, 647-661, 2015 the subsurface.A first, explorative geoelectric (ERT) profile was performed on 16 August 2013 and repeated on 9 October 2013 in combination with one refraction seismic tomography (RST) along the same line (see Fig. 1).Combining refraction seismic and ERT measurements enables us to unambiguously identify the subsurface materials in the ground.Due to very different specific resistivities, ERT is best suited to differentiate between ice and water, whereas the distinction between air and ice can more easily be accomplished by RST because of large contrasts between their respective p wave velocities.

Electrical Resistivity Tomography
A 94 m long electrode array composed of 48 electrodes with 2 m spacing was installed along a straight line less than 2 m away from the two boreholes (Fig. 1).Current was injected using varying electrode pairs, and the resulting potential differences were automatically measured by a Syscal (Iris Instruments) for each quadrupole possible with the Wenner-Schlumberger configuration (529 measurements, 23 depth levels).The electrode locations were marked with spray paint and a number of electrodes were left on site to facilitate further measurements.
The measured apparent resistivity data sets were then inverted using the RES2DINV software (Geotomosoft, 2014) with the following setup.A robust inversion constraint was applied to avoid unrealistic smoothing of the calculated specific resistivities.Additionally, the depth of the model layers was increased by a factor 1.5 and an extended model was used to match the model grid of the corresponding seismic inversion.Note that for geometric reasons, the two lower corners of the resulting tomograms have very low sensitivity to the obtained data and should not be over-interpreted.Finally, a time-lapse inversion scheme was applied to the two ERT data sets, yielding the percentage of resistivity change from the first measurements to the second.Here, an unconstraint inversion was chosen, meaning that the ERT measurements were inverted independently.

Refraction seismic tomography
The measurements were conducted using a Geode seismometer (Geometrics) and 24 geophones placed with 4 m spacing.A seismic signal was generated in-between every second geophone pair by repeatedly hitting a steel plate with a sledge hammer.To improve the signal-to-noise ratio, the signal was stacked at least 15 times at each location.Two additional offset shot points were measured (3 m before the first geophone and 6 m beyond the last one) in order to maximize the spatial resolution and match the ERT profile length and depth of investigation.
The first arrivals of the seismic p wave were manually picked for each seismogram using the software REFLEXW (Sandmeier, 2014).A simultaneous iterative reconstruction

Ground surface temperatures
Figure 4 shows MAGST at 0.02 and 0.3 m on the seven GST nodes.Some years (e.g., 2009, 2011, 2013) show a MAGST spatial variability, evaluated as the range of MAGST measured in all nodes and greater than 3 • C, that clearly exceeds the inter-annual variability.In general, considering all 7 years, we observed that mean spatial variability (2.5 ± 0.1 • C) is greater than mean inter-annual variability (1.6 ± 0.1 • C).The results are similar at both depth.The difference between MAGST measured at 0.02 and 0.3 m is, on average, 0.4 ± 0.1 • C, with deeper sensors usually warmer than the shallower ones.On average, the thermal offset due to snow cover is about 1.5 ± 0.2 • C with snow-covered nodes being warmer than snow-free nodes.These observations confirm that the warming and cooling effects of, respectively, a thick and thin snow cover (Zhang, 2005;Pogliotti, 2010) can coexist over short distances (< 50 m) and lead to high spatial variability of the GST.The duration of snow (SD) on snow-covered nodes is on average 270 ± 6 days with a mean range of spatial variability of 28 days and a mean range of inter-annual variability of 48 days.To disentangle the influence of snow and air temperature on surface temperature in snow-covered nodes, we tested the relationship between MAGST and MAAT and between MAATsf and SD.We found no significant correlation between MAGST and MAAT. Figure 5 shows the scatterplot comparing SD (A), MAATsf (B) and MAGST: MAGST is significantly correlated to both SD (p < 0.05) and MAATsf (p < 0.001), with the latter explaining the higher portion of variance (R 2 = 0.39).Being computed on snow-free days, MAATsf is mainly controlled by air temperature but partially also by the duration of snow cover, therefore integrating the relative contribution of both components (snow duration and air temperature) on MAGST.

Active layer
Table 1 summarizes the active layer parameters observed in the two boreholes.Since August 2008 data are available at both SH and DP boreholes, results of ALT can be compared over 6 years while MAGST, TTOP and THO over 5 years (shaded rows in Table 1).Missing values (column % NA) in both boreholes are lower than 4 % in all years.
ALT is the parameter showing the greater difference between the two boreholes with a mean of 2.7 ± 0.3 m in SH and 4.7 ± 0.2 m in DP.The mean inter-annual difference of ALT between the two boreholes is 2.0 ± 0.1 m, while the mean absolute inter-annual variability of ALT at borehole level is 1.0 ± 0.1 m.In both boreholes the maximum ALT has been recorded in 2012 and the minimum in 2010.ALT (date) is normally anticipated in DP (except 2013) with differences ranging from a few days (e.g., 2009) to more than 3 weeks (e.g., 2012).The MAGST is on average slightly lower in SH, which normally shows a thinner winter snow cover compared to DP (Fig. 6).The TTOP values are very similar, around −0.9 • C. The THO is negative in both boreholes (except 2013) with a mean value of about −0.5 • C in DP and −0.3 • C in SH.
The values of Table 1 show that all the active layer parameters are very similar between the two boreholes with the only exception of ALT, which in DP is nearly double than in SH.To better understand the causes of this difference, the daily mean temperatures at selected depths within the active layer of both boreholes and the corresponding snow cover thickness are compared in Fig. 6.Although a consistently thinner snow depth is recorded on SH compared to DP (mean difference ∼ 41 ± 14 cm during the winter sea-  sons 2012 and 2013), the duration of the insulating snow cover is similar (250 ± 16 days for SH vs. 254 ± 17 days for DP) and effectively does not determine a large difference in MAGST (Table 1).Consequently, the snow cover regimes of the two boreholes can be considered equivalent.
For these reasons we hypothesize that ALT difference may be related to a greater ice/water content in SH compared to DP.This is revealed by the geophysical survey (see Sect. 3.4 and Fig. 9) and can be inferred by temperatures at greater depth.At 1.6 m (red lines, Fig. 6) a pronounced zero-curtain effect can be observed in SH (dashed lines) twice per year, (i) from snow melt to mid-summer and (ii) from the snow onset to mid-winter, while a similar behavior is missing in DP.The occurrence of the zero-curtain reflects a large consumption of energy, both for ice melting during summer and water freezing during winter, resulting in lower temperatures of SH.Deeper down, the summer heat wave in SH is further delayed if compared to DP: at 3 m in SH (dashed blue lines) the zero-curtain effect is almost continuous from late summer to early winter (e.g., in 2010 and 2011), and it is not possible to see a breaking point between melting and freezing processes.Such a behavior is totally missing in DP.It is also interesting to observe that freezing zero-curtain ends nearly contemporary at 1.6 and 3 m and is followed by a rapid temperature drop.
In conclusion, the ALT at Cime Bianche shows a pronounced spatial variability probably caused by the variability of ice/water content in the sub-surface and associated energy consumption resulting from freezing and melting processes.

Permafrost temperature and warming trend
Due to the small depth reached by the borehole SH, the analysis of permafrost temperature is limited to the borehole DP.Looking at temperature profiles with depth (Fig. 7), the permafrost layer at Cime Bianche has a thickness greater than 40 m and a mean temperature of about −1.2 • C. The ZAA varies across years and during the observation period ranged from a minimum of 14.2 m in 2011 to a maximum of 16.2 m in 2013 (Table 2).During the observed years, both minimum (solid lines) and maximum (dashed lines) temperature profiles (deeper than 6 m) tend to progressively shift toward warmer temperatures (Fig. 7).The only exceptions are represented by the 2011 maximum and the 2009 minimum, with the latter only above 10 m of depth.
The observed temperature shift is also quantitatively supported by the trend analysis.The analysis was conducted at all depths, but only deeper temperatures (below 8 m) show significant trends (Kendall's p value < 0.01).Figure 8 reveals that a pronounced warming rate ranging from 0.1 • C yr −1 at 8 m to 0.007 • C yr −1 at 41 m can be observed.The upper boundaries of the confidence intervals are systematically unbalanced toward higher values and the lower boundaries are always above zero.This means that, at all depths, the statistical distribution of all possible fitted trends www.the-cryosphere.net/9/647/2015/The Cryosphere, 9, 647-661, 2015 is positively skewed.Based on this analysis, it is concluded that permafrost at Cime Bianche is warming because significant positive warming rates are reported below 8 m.

Geophysics
Figure 9 shows the final distribution of specific resistivity for the two ERT measurements, the percentage of change in the model resistivity between the two time steps and the p wave velocity distribution over the same subsection.Additionally, the surface characteristics and a detailed analysis of the geophysical properties at the two borehole locations (SH and DP, Fig. 10) are included in the analysis.
The overall characteristics of both ERT profiles are very similar (Fig. 9a and b) and can be divided into three main zones: a low resistive layer directly below the surface, varying between 2.5 m thickness at the top of the slope and 7 m thickness at the bottom; two high resistive areas with values exceeding 20 000 m, located below the superficial layer (from the start of the subsection to the superficial borehole: 0-34 m and 40-52 m); and a less-high resistive area on the lower part of the profile below 5 m depth.
Comparing the two ERT data sets (cf.also the time-lapse image in Fig. 9c), one can observe a clear increase of the uppermost low resistive layer between August and October which is coherent with a thickening of the active layer observed in the borehole temperature during this period.Another main difference between the two measurements is the apparition of two low resistive zones at 34 and 60 m, visible down to 10 and 15 m depth, respectively.These areas can also be seen in the ERT tomogram from August but much  less developed and limited to a few meters.In addition, the very high resistive area located in the upper part of the profile is much smaller and displaced by about 5 m towards the lower part of the profile in the second measurement.These changes are clearly visible in blue (increase) and red (decrease) colors in Fig. 9c.As stated before, the two data sets were inverted independently within the time-lapse scheme.A constrained inversion (results not shown here) would yield very similar overall distribution of resistivity changes; the only difference is a much smaller range of values.The large area of resistivity increase, located just above the superficial borehole location and reaching down to the bottom of the profile, corresponds to the displacement of the high resistive area observed in the ERT tomograms.
The RST tomogram exhibits much less lateral variations than the ERT results (see Fig. 9d), pointing to the influence of liquid water in the ERT results.One can clearly see a relatively slow layer with velocities between 300 and 1500 m s −1 (red and dark red colors) just below the surface, with varying thickness between 3 and 5 m.This layer is thickest in the vicinity of SH and thinnest at DP (64 m).Below this first layer the velocities increase steadily until reaching the maximum (around 6400 m s −1 ).The rate of velocity increase is strongest around 40 m and there is a clear distinction between the upper part of the profile (until 45 m) and the lower one.At depth the high velocity zone is present in the upper part and not in the lower part of the profile.Conversely, the velocities at the surface are much higher in the lower part (especially around DP) than in the upper part.
Both geophysical profiles show clear differences in the subsurface properties as well as surface composition at the borehole locations (Fig. 9e).The upper part of the profile (until 50 m) is more or less homogeneously covered by medium size blocks and has the deepest layer of coarse-debris deposits, whereas the granulometry in the lower part is much more variable at the surface and the debris layer is thinner.The boreholes are located in very different conditions: DP is located in-between two zones composed of big blocks (from pluridecimetric to metric), whereas SH is located at the junction between medium size blocks (from pluricentimetric to decimetric), mixed and non-mixed with soil.To relate in detail the results yielded by the geophysics and the measured temperature, the vertical distribution of specific resistivity, seismic velocity and ground temperature at SH and DP are shown in Fig. 10.

Ground surface temperatures
In this study both the inter-annual and the spatial variability of MAGST within a restricted area has been analyzed and compared: the results show that at Cime Bianche, the mean range of spatial variability (2.5 ± 0.1 • C) far exceeds the mean range of observed inter-annual variability (1.6 ± 0.1 • C).Given the comparatively homogeneous characteristics of the ground surface at the sensors locations, such a variability is essentially caused by the heterogeneity of the snow cover thickness both in space (effect of wind redistribution and micro-morphology) and time (effect of variable weather conditions and precipitations).In particular, the combination of snow cover duration and air temperature during the snow-free period is the main factor controlling MAGST values.This is true not only for snow-free nodes but also for nodes experiencing long-lasting (270 days) yet highly variable (28 days) snow cover.
The thermal effect of snow cover on ground surface temperature has been extensively analyzed (e.g., Goodrich, 1982;Keller and Gubler, 1993;Zhang, 2005;Luetschg et al., 2008;Langer et al., 2013).In recent years, with the advances of minilogger technology, the number of field experiments aimed at the characterization of the spatial variability of GST has grown.Recently Gubler et al. (2011)  The inter-annual variability of MAGST caused by snow is also well known and documented by a number of studies (Romanovsky et al., 2003;Hoelzle et al., 2003;Karunaratne and Burn, 2004;Brenning et al., 2005;Etzelmüller, 2007;Ødegård and Isaksen, 2008;Schneider et al., 2012)  The obtained results are very similar at both measurement depths.Given such a small difference and the agreement of temperature fluctuations between 2 and 30 cm, it is arguable that to describe the spatial variability of GST and run longterm GST observations, measurements at two or more depths are not needed.

Active layer
In this study, both ALT and temperature fluctuations within the active layer of two adjacent boreholes have been compared.Such experimental design provides direct evidence of the small-scale spatial variability of the ALT and allows to evaluate the effect of ice/water content on sub-surface temperature.
From 2009 to 2013 the ALT at Cime Bianche varied within 2.0 and 5.5 m with a mean inter-annual variability of 1.0 ± 0.1 m.These ranges and the observed inter-annual variability of ALT are comparable to those recorded in other alpine sites (Anisimov et al., 2002;Christiansen, 2004;Schneider et al., 2012;Smith et al., 2010;PERMOS, 2013) In the Swiss Alps, the thickness of the active layer typically varies between 0.5 and 8 m depth (Gruber and Haeberli, 2009;PERMOS, 2009PERMOS, , 2013)).
ALT in the borehole SH is systematically lower than in DP (mean difference 2.0 ± 0.1 m) even though all the active layer parameters (MAGST, TTOP, THO see Table 1) are very similar between the two boreholes.On one hand, such a similarity suggests that snow cover regimes above the two boreholes are nearly equivalent; thus snow probably plays a major role only on the inter-annual variability of ALT.On the other hand, the pronounced spatial variability of ALT is probably caused by the variability of ice/water content in the sub-surface and associated variation of energy consumption resulting from freezing and melting processes.Langer et al. (2013) confirms this hypothesis by observing, in a tundra lowland landscape, that ALT is related mainly to ground properties (ice content), whereas snow physical properties have greatest influence on the ground surface temperatures.Probably the different ice/water content between SH and DP is caused by snowmelt and meltwater infiltration along preferential discontinuities (a borehole acts a discontinuity itself).Hilbich et al. (2008) observed at Schilthorn (Swiss Alps) a similar situation between two boreholes 15 m apart, ascribing the lower ALT of one borehole to the higher moisture contents (and related freezing) caused by preferential water flow paths from the surrounding slopes.Schneider et al. (2012) analyzed the thermal regime of four adjacent boreholes drilled on differing material (coarse debris, fine debris and bedrock) at Murtèl-Corvatsch (Swiss Alps) and recognized meltwater and ice content as mainly responsible for the observed ALT spatial variability.
The different amount of available water in the active layer of the two boreholes is also reflected by the occurrence of the zero-curtain in the borehole SH and its absence in the borehole DP.In the upper part of the active layer, a pronounced zero-curtain can be observed two times per year: (i) from snow melt to mid-summer (spring zero-curtain) and (ii) from the snow onset to mid-winter (autumn zero-curtain).Recently, Zenklusen Mutter and Phillips (2012) deeply analyzed similar behaviors on a sample of 10 boreholes in Switzerland, observing that, on average, the duration of the spring zero-curtain is usually shorter than the autumn one and is strongly dependent on snow depth at the end of the winter.At Cime Bianche, such a distinction between spring and autumn zero-curtain is not always possible in the deeper part of the active layer.As also observed by Rist and Phillips (2005), it may happen that, below a certain depth, the ground temperature does not become positive because the energy from the summer heat wave is not sufficient to melt all ice before the onset of the subsequent winter season.This continuous zero-curtain is more probable when a higher amount of meltwater is available (Scherler et al., 2010;Kane et al., 2001) and can occur at a different depth from year to year, strongly influencing the resulting ALT.

Permafrost temperature and warming trend
In order to look for trends that might reflect warming, two non-parametric methods were applied to borehole temperature time series.The detected linear trends are statistically significant (Kendall's p value < 0.01) only at depth below 8 m.Probably, in the first meters, the seasonal and interannual variability of temperatures is so strong that significant trends are not detectable, although a seasonal detrending being applied to remove such high-frequency oscillations (see also Sect.2.5).The detected trends span the range 0.1-0.01• C yr −1 , suggesting that at Cime Bianche permafrost is warming.
As also discussed by Zenklusen Mutter et al. (2010), the detection of trends on time series covering a short time span requires caution and adoption of specific criteria.Moreover, the estimation of uncertainties and significance levels is also fundamental for facilitating the comparisons of trends between differing sites and for reproducing trend detection methods on others data sets.
Recently in the Alps, Zenklusen Mutter et al. ( 2010) detected trends on daily temperature time series of two boreholes in the Muot da Barba Peider ridge (Eastern Swiss Alps).For the deep frozen bedrock between 8 and 17.5 m, a general warming trend was found with significant (p value < 0.05) values ranging from 0.042 to 0.025 • C yr −1 .At Cime Bianche a similar range of warming rate was found between 16 and 20 m.The substantial difference between the two sites is that the Swiss boreholes are drilled at the top of a NW-oriented ridge with a mean slope of 38 • and thus have a strong 3-D thermal effect induced by topography (Noetzli et al., 2007).In the mountains of Scandinavia, Isaksen et al. (2007) reported warming trends between 20 and 60 m of depth ranging from about 0.05 to 0.005 • C yr −1 over three sites, while Isaksen et al. (2011) found an increase in mean ground temperature between 6 and 9 m of depth at two sites, with rates ranging from about 0.015 to 0.095 • C yr −1 .Recently at Tarfala mountain station (Sweden), Jonsell et al. (2013) found trends over 11 years (2001)(2002)(2003)(2004)(2005)(2006)(2007)(2008)(2009)(2010)(2011) ranging from 0.047 to 0.002 • C yr −1 between 20 and 100 m of depth.
The absolute values of warming rates are difficult to compare because of different site characteristics, geographical regions and methods used for trend detection.Nevertheless, some similitudes exist between our and the above-mentioned case studies: (i) trends are difficult to detect at shallower depth because of the higher seasonal variability of temperatures; (ii) warming trends are mainly significant below 8-10 m of depth; (iii) warming trends exponentially decrease with depth; (iv) there is no evidence of negative (cooling) trends at any depth in recent literature.

Geophysics
Given the relatively high resistivity and p wave velocities along the profiles, the presence of permafrost observed in the borehole data is confirmed by the geophysics over the whole profile length (Fig. 9).Moreover, a clear discrepancy between the upper part of the profile where SH is located and the lower one with borehole DP can be seen in both the ERT and the RST data.
At DP, the comparatively high p wave velocities indicate the presence of weathered bedrock close to the surface, whereas at SH a layer of coarse-debris deposits in the uppermost 5 m is confirmed by very low p wave velocities.

P. Pogliotti et al.: Warming permafrost and active layer variability
Conversely, p wave velocities at depth are higher for SH (around ∼ 6000 m s −1 ) than for DP (around ∼ 5000 m s −1 , see also Fig. 10).This difference, also seen in the resistivity data (around 17 000 m at SH and 13 000 m at DP), would indicate that a larger ice content is present in the upslope part of the profile than in the lower part.This is in good agreement with the spatial variation of ALT highlighted in Sect.3.2 and the zero-curtain phase observed only at SH (see Fig. 6).
The low-resistivity and low-velocity layer near the surface, the thickness of which increases visibly between August and October in the ERT data, is considered to be the active layer.Figure 10 compares the vertical distribution of specific electrical resistivity, p wave velocity and temperature for both boreholes and dates.At first glance, there seems to be a mismatch between resistivity and temperature regarding ALT for SH.However, borehole temperatures at SH in August show constant values at the freezing point between 1 and 3 m depth (between 2 and 4 m in October), the deeper level being the depth of the sharply increasing resistivity values.As resistivity is sensitive to the liquid water content, its values will not increase significantly before most of this liquid water has been frozen, coinciding with a temperature increase to values below the freezing point (e.g., Hauck, 2002).Due to the higher water/ice content in SH, this phenomenon (∼ vertical zero-curtain) is only seen in SH and not in DP.
The two low resistive areas (34-40 and 53-60 m), already visible in August and more pronounced on the second ERT profile in October, are interpreted as the preferential water flow path.Since the melt water cannot infiltrate through the two ice-rich (high resistive) bodies close by (at 20-33 and 40-52 m horizontal distances), it is forced to follow a preferential path in-between.The lower infiltration area (53-60 m) is constrained in the upper part by the ice-rich zone and in the lower part by the presence of bedrock near the surface.
Finally, the displacement of the high resistive area observed near SH (blue zone at depth on the time-lapse tomogram) is most likely an inversion artefact (overcompensation) due to the appearance of the low resistive area in the second ERT profile (cf.Hilbich et al., 2009).

Conclusions
This paper presents a first synthesis on the thermal state and recent evolution of permafrost in the monitoring site of Cime Bianche, one of the few permanent observatories on the southern side of the European Alps.The analysis focused on (i) the spatial and temporal variability of MAGST in relation to snow cover, (ii) the small-scale (30 m) spatial variability of ALT and (iii) the warming rate of deep permafrost temperatures.The results of analysis show the following: 1. Spatial variability of MAGST is greater than its interannual variability and is controlled by a combination of air temperature during the snow-free period and snow duration.
2. The ALT at Cime Bianche has a pronounced spatial variability caused mainly by a different ice/water content due to very different surface and subsurface conditions in terms of weathering and fracturing of bedrock.
3. Permafrost at Cime Bianche is warming at significant rates below 8 m of depth.
The Supplement related to this article is available online at doi:10.5194/tc-9-647-2015-supplement.

Figure 1 .
Figure 1.Overview of the Cime Bianche monitoring site.

Figure 2 .
Figure 2. Example of detection of snow cover duration from GST time series with the method of Schmid et al. (2012).OD is the onset date of snow; MD is the melting date of snow.The periods used for the calculation of MAGST, SD and MAATsf are represented by the scheme on top.

Figure 4 .
Figure 4. Mean annual ground surface temperatures at depths of 0.02 m (red) and 0.3 m (blue).Star symbols indicate snow-covered nodes, while bullets indicate snow-free nodes.The horizontal lines indicate the mean MAGST for each year and each depth.Black rectangles are used to highlight the min-max envelope to facilitate the inter-annual comparison.

Figure 5 .
Figure 5. Scatterplots of SD (a) and MAATsf (b) against MAGST.The solid line represents the linear fit while the dotted lines are the confidence intervals.The metrics of the fitting are also reported.

Figure 6 .
Figure 6.Fluctuations of snow cover thickness (Hs) and ground temperatures (daily mean) at selected depths in the active layers of Cime Bianche from 1 October 2010 to 30 September 2013 determined from borehole temperature data.Lines type: dashed is for SH, solid is for DP.Colors: red is for shallower temperatures (1.6 m), blu is for deeper temperature (3 m), grey is for snow.

Figure 7 .
Figure 7. Minimum (solid lines) and maximum (dashed lines) temperature profiles in the borehole DP below 6 m of depth for the period 2009-2013.

HFigure 8 .
Figure 8. Warming rate calculated over the period 2009-2013 below 8 m of depth in the borehole DP as a function of depth.Black dots represent linear trends as • C yr −1 .The uncertainty of trend values is represented by the dashed bars, which indicate the lower and upper boundaries of the 95 % confidence interval of the fitting model (see Sect. 2.5 for details).

Figure 9 .
Figure 9. Tomograms of the specific resistivities for both ERT measurements: (a) 16 August 2013, (b) 9 October 2013, (c) percentage change in model resistivity between the two dates and (d) seismic velocities.The location of SH and DP is figured with vertical black lines of respective length.A rough description of the surface aspect along the profile is also shown (e).

Figure 10 .
Figure 10.Vertical distribution of specific resistivity and P wave velocity at the borehole locations, extracted from the tomograms shown in Fig. 9, as well as borehole temperatures for the dates of the ERT and RST measurements.The horizontal lines represent the active layer thickness at the respective time periods.
but has rarely been explicitly analyzed and quantified.An exception in the Alps is represented by Hoelzle et al. (2003), who reported an inter-annual variability of ±2.7 • C measured during two seasons on eight mini-loggers with different surface characteristics in the Murtèl-Corvatsch area.Our results thus report a more robust quantification of the mean inter-annual GST variability (1.6 ± 0.12 • C), based on a longer time series (7 years).

Table 1 .
Synthesis parameters of active layers recorded in the two boreholes of Cime Bianche.The average values (Avg.) are calculated over the period 2009-2013 where data from both boreholes are available.

Table 2 .
Interpolated depth of zero annual amplitude oscillation (ZAA) and corresponding mean temperatures in the borehole DP.