Long‐term monitoring of sporadic permafrost at the eastern margin of the European Alps (Hochreichart, Seckauer Tauern range, Austria)

Abstract Delineating the spatial extent and the altitudinal lower limit of mountain permafrost is difficult due to complex topo‐climatic and variable ground thermal conditions within short distances. Little information exists regarding sporadic permafrost existence, its thermal characteristics and its long‐term changes at the eastern margin of the European Alps. To reduce this gap, permafrost monitoring was initiated in 2004 in the Seckauer Tauern mountains, Austria. Research was carried out in the summit region of Mt Hochreichart (2416 m a.s.l.) and at several nearby cirques and valleys, all with rock glaciers. Geomorphic mapping, numerical permafrost modeling, measurements of the bottom temperature of the winter snow cover, continuous ground temperature monitoring, electrical resistivity tomography and optical snow cover monitoring were applied. Results indicate sporadic permafrost occurrence in the summit region with mean annual ground temperatures slightly below 0°C at the surface and −1.4°C at 2.5 m depth. Permafrost lenses also exist in the transition zone between the rock glacier and the talus slope behind attributed to coarse‐grained, blocky material causing additional ground cooling. Thanks to long‐term data, statistically significant trends of atmospheric and ground warming were observed in 2000–2018. Permafrost at this site will presumably disappear within the next few decades.

An alpine-wide homogenous permafrost modeling approach is the Alpine Permafrost Index Map or APIM. 2 According to the APIM, the extent of permafrost in Austria is between 484 and 2,907 km 2 (0.6-3.5% of the national territory) depending on the permafrost index chosen as a threshold. A reasonable index of ≥0.5 yields a permafrost area of 1,557 km 2 , which is 25% of the modeled permafrost in the entire European Alps. The index describes semi-quantitatively the occurrence of permafrost, from permafrost in nearly all conditions to permafrost only in very favorable conditions. The denomination highlights the problem that despite the fact that APIM is sophisticated in terms of the modeling approach, the environmental heterogeneities of the relevant mountain area cause substantial uncertainty in the model. Earlier estimates of permafrost distribution in Austria 1,9 have suggested that some 1,600-2,000 km 2 of the country (1.9-2.4%) is underlain by permafrost. The field data used for these two national permafrost estimations were also very limited in time and space with vast areas without any permafrost-related data.
Monitoring of permafrost in Austria is carried out at some 23 study areas by different institutions. 10 Most of the sites cluster around two regions in central and western Austria (Figure 1a). Knowledge of mountain permafrost existence and thermal conditions east of those sites and therefore also in the very east of the European Alps, a high-alpine area  Figure 3, modeled permafrost distribution, 2,11 extent of rock glaciers, bedrock geology (simplified from ref. 12 ) and different automatic monitoring sites (for explanation of codes see Table 2). (c) Extent of glaciers during the Last Glacial Maximum in the Seckauer Tauern (based on refs 13,14 ). (d) Geomorphological overview map of the Hochreichart area with modeled permafrost extent [Colour figure can be viewed at wileyonlinelibrary.com] still influenced by permafrost according to models (e.g. 2 ), is very limited. This paper attempts to answer the research question regarding present thermal conditions in a marginally permafrost-affected area at the eastern margin of the European Alps and their relationships to present climate change. To gather such knowledge, permafrost monitoring was initiated in the Hochreichart area, Seckauer Tauern Range, in 2004 and has continued since then. Only preliminary results have been published so far. 11 A comprehensive presentation and discussion of the permafrost-related research at this site for the period 2004-2018 is the content of this paper. No such long-term permafrost-related data series have been presented earlier from anywhere in the Eastern European Alps apart from one study focusing on potential weathering of alpine rock walls. 15 The aim of this paper is to present results on the detection, delin-  (Figures 1d and 2a-d ). As judged from the widespread lichen coverage, the well-developed weathering rinds and the general stable appearance of these sediments, it can be assumed that the present morphodynamical dynamics at these blockfields is insignificant.
Bedrock uncovered by autochthonous block material is not very widespread. Such exceptions are steep cirque headwalls or the flanks of Ushaped valleys (Figures 1d and 2a,e,f). The bedrock geology in the summit areas of the Seckauer Tauern is dominated by granites, orthogneisses and acid gneisses as well as quartzites and conglomerates to the north-east of the main drainage divide 12 (Figure 1b). Further areas of interest in this study were the Brandstätter and Schöneben cirques, the Höll and Dürr valleys, and the summit area of the Kleinreichart. Well-developed rock glaciers characterize all the three cirques and the two valleys of interest. 13,16 The rock glaciers consist predominantly of coarse-grained, blocky, gneissic sediments at the surface. Coarse-grained talus and finer-

| Numerical permafrost modeling
The development of new permafrost models was not within the scope of this paper. Earlier modeling approaches are, however, relevant for the regionalization of the collected and interpreted data and hence are summarized here. Permafrost distribution was quantified using empirical and statistical models. First, the relationship between MAAT and permafrost was analyzed. Second, empirical values of probable lower limits of discontinuous permafrost occurrence of a nearby region in combination with local correction values were used. For details on these two approaches refer to Kellerer-Pirklbauer. 11 Third, results from an alpine-wide homogenous statistical permafrost model based on alpine-wide permafrost observations, MAAT, potential incoming solar radiation and precipitation 2 are considered here.

| Bottom temperature of the winter snow cover
The bottom temperature of the winter snow cover (BTS) is known to be an indicator for the presence or absence of permafrost in alpine terrain. Based on studies in the Western Alps, empirically established thresholds were defined: permafrost unlikely at >−2°C, permafrost possible at −2 to −3°C, and permafrost probable at <−3°C. 18 19 The point data within a 0.04-km 2 area in the Reichart cirque were therefore considered as suitable for interpolation (inverse distance weighting) and subsequent statistical comparison. BTS values were tested regarding normality (Kolmogorov-Smirnov test) and correlated against elevation, snow depth and potential shortwave radiation during the generally snowfree period (JJASON).

| Automatic monitoring
A network for continuous ground and air temperature monitoring by using miniature temperature data loggers in the Hochreichart area was initiated in October 2004 and steadily expanded. Over the years, ground temperature was measured at 12 sites: six at the Hochreichart summit area, five in the Reichart cirque area and one in the Kleinreichart summit area ( Table 2, Figures 1 and 2). Air temperature  was monitored at five sites in close spatial distance to ground temperature monitoring sites. The installed devices are one-or three-channel data loggers (GeoPrecision) equipped with PT1000 temperature sensors measuring and logging hourly the temperature. According to the producer, the PT1000 temperature sensors have an accuracy of +/ −0.05°C, a range of −40 to +100°C and a calibration drift of <0.01°C yr -1 . At the air temperature monitoring sites (n = 5), the loggers are protected from direct insolation by radiation shields (Young).
Sensors at 0 cm depth are shielded from direct solar radiation by thin platy rocks that still allow rather unhampered air circulation within the voids. Raw data were checked and short data errors were corrected or filled using linear interpolation between the neighboring data points. 20 Furthermore, data from a nearby automatic weather station (R-A2)

| Electrical resistivity tomography
Electrical resistivity is a physical parameter, which is related to the chemical composition of a material and its porosity, temperature, and water and ice content. 25 Two-dimensional electrical resistivity tomography (ERT) uses multi-electrode systems and two-dimensional data inversion in order to receive a rather accurate model of the subsurface. 26 (Table 3).
In 2008, an LGM-Lippmann 4-Punkt light hp resistivity-meter using the Wenner-Alfa configuration was used. During all other campaigns a GeoTom-2D system (Geolog2000) with multicore cables has been used applying both the Wenner and the Schlumberger arrays. 26 Saltwater was sometimes used at the electrodes to improve electrical contact. ERT data analyses were carried out in RES2DINV concatenating both array results (in the case of good-quality concatenated Wenner and Schlumberger data with a root mean square [RMS] error < 10) or by using only the Wenner data alone (in the case of poor quality Schlumberger data). The apparent resistivity data were inverted using robust inversion modeling. Bad datum points were removed before the inversion. The number of iterations was stopped when the change in the RMS error between two iterations was small. 29 Granitic rocks are characterized by resistivity values mostly around 5 k Ohm.m. 30 Gneiss is in the range of 0.1-1 kΩ.m. 26 Resistivity values for frozen material can vary over a wide range from several kΩ.m to even millions of Ω.m (see 26

| Modeled spatial permafrost distribution
Permafrost distribution according to the APIM approach 2 reveals low to medium probabilities of permafrost occurrence in the study area.
The lowest area potentially underlain by permafrost according to this approach is about 2,000 m a.s.l. at a north-facing slope in the Schöneben cirque. The lower limit of low-to-medium probable permafrost at north-facing slopes was slightly higher at the Reichart cique always modeled as permafrost free (Figures 1 and 3).             The correlation between ground and air temperature is significant in all cases: strong for annual (r = 0.90) and in particular for July (r = 0.97) and still moderate for January (r = 0.66). The last is related to the influence of seasonal snow cover.

| Subsurface conditions based on geophysics
Results from 13 representative (out of 24) ERT profiles are depicted in    Table 3.
Locations of profiles are shown in Figure 3. assume permafrost conditions at many locations not considered by the spatial permafrost models. 2,11 This highlights the relevance of substrate and landform effects on permafrost distribution. 3 However, a clear definition of a lower limit of permafrost in the study area is not trivial due to the site-specific effects on ground thermal conditions just below or just above perennial freezing.
In addition, large blocks or piles of rocks forming natural cairns may penetrate the general terrain surface and the seasonal snow cover, 33 allowing convective air flow into the coarse-grained sediment layer. 34 Such blocks, often darker in visual appearance due to the weathering Despite this restriction, it is evident that the results from the continuous temperature monitoring reveal suitable thermal conditions for permafrost not only at the summit area but also in the higher elevated parts of the cirques including the rock glaciers. Therefore, the rock glacier in the Reichart cirque is neither relict nor climatically inactive. 37 It appears visually relict with collapse features at the lower part caused by ground subsidence processes and some vegetation cover, but contains lenses of permafrost at the upper end of the landform ( Figure 11).

| BTS data: What are they telling us?
Results from the BTS data measured in four cirques suggest marginal permafrost conditions with a dominance of "no permafrost" sites in all four cirques. In detail, interpretation of the BTS data should be made very carefully. For instance, the multi-annual measurements in the Reichart cirque revealed substantial interannual variations although the general pattern of cooler and warmer areas in a given area remains very similar. Such interannual variations were detected also at other alpine areas (e.g. 39 ). This highlights the problem that using BTS data from only a single year for spatial permafrost modeling is substantially misleading. The spatial interpolation of BTS data for a 0.04-km 2 area for five different years ( Figure 4) illustrates this problem with a moderate difference between the warmest and coldest year of 0.8°C. Permafrost assessments based on only one BTS campaign from one area must therefore be considered as lacking confidence.
BTS conditions in late winter are influenced by atmospheric temperature in early winter or in periods of thin snow coverage where cold air might penetrate through a thin or non-existent snow cover.
Hence, even if a snow cover thickness of >80 cm is measured during a BTS campaign, 18 it necessarily gives little information about the existence or absence of a winter equilibrium temperature at the base of the snow cover. The BTS results presented in Figure 3 15 ). Whereas a warming trend can be seen in July ground and air temperatures, no trend exists for January temperatures since the turn of the millennia.
MAGT at the surface in the study area at the highest and coolest areas is only slightly below 0°C and hence a steady warming, which can be expected in the future, 41 will lead to further ground surface warming and even to positive MAGT values at the coolest sites of the study

FIGURE 11
Rock glacier types as proposed previously. 37,38 The rock glacier in the Hochreichart cirque is considered to be pseudo-relict, and hence a rock glacier which appears to be visually relict but still contains patches of permafrost (modified after 38 24 permafrost will only slowly disappear from the Hochreichart area. A comparison of neighboring ground surface and air temperature conditions at five paired monitoring sites revealed positive surface offsets for the summit areas (i.e. higher ground compared to air temperature as expected; cf. 24 ), but negative surface offsets for rock glacier sites. The latter is attributed to efficient ground cooling of the coarse-grained rock glacier sediments with a strong convective heat transport component. [42][43][44][45][46] This surface offset is of the order of −0.7°C at the rooting zone of the Reichart rock glacier.
At a location where deep circulating air within the rock glacier sediments emerges at the surface (similar observations were made at the Schöneben cirque 21 ), the surface offset is even higher (−3.1°C).
On a seasonal scale, the offset varies between sites with thin block layers from sites with thicker rock glacier sediments. At the rock glacier site, the surface offset is negative in almost all months, meaning lower temperatures at the ground surface compared to the air. This indicates even in winter good thermal coupling of the (upper) rock glacier sediment layer with the atmosphere, a fact related to natural cairns and blocks as well as partly to thin snow coverage as discussed above. 33,34 Normally, the nival offset in winter is larger than the vegetation offset in summer, 47  This study also revealed some methodological drawbacks. ERT data are not straightforward to analyze and interpret, in particular when the substrate is rich in open voids with pore air. However, the pattern of high resistivity values in terms of distinct lenses below a surface (active?) layer often supports the assumption of permafrost.
Furthermore, BTS data must be treated very cautiously in the case of a missing winter equilibrium temperature at the base of the snow cover. The uneven surface of bouldery rock glaciers favors atmosphere-ground air coupling even in mid-winter. Furthermore, very cold periods might influence the ground surface even below thicker snow cover. Finally, multi-annual BTS campaigns in the same cirque revealed substantial interannual differences. This implies that spatial permafrost models based on single BTS campaigns in a given area are highly vulnerable to fail because of some randomness in the data.