Investigations of polygonal patterned ground in continuous Antarctic permafrost by means of ground penetrating radar and electrical resistivity tomography: Some unexpected correlations

The results of a combined geophysical and geomorphological investigation of thermal‐contraction‐crack polygons near Gondwana station (Germany) in northern Victoria Land (Antarctica) are reported. An area of about 20,000 m2 characterized by random orthogonal polygons was investigated using integrated ground penetrating radar, electrical resistivity tomography, geomorphological surveys, and two trench excavations. The polygons are well developed only at elevations higher than 6–7 m above current sea level on Holocene‐age raised beaches. It is concluded that the polygons are composite in nature because the shallow linear depressions that outline the polygons are underlain by fissures that can contain both sandy gravel and foliated ice (i.e., ice wedges) even in the same polygon network and at distances of just a few meters. Unexpectedly, most of the polygons follow the border of the raised beaches and develop in correspondence with stratigraphic layers dipping toward the sea, imaged by ground penetrating radar (GPR) profiles and interpreted as prograding layers toward the present‐day shoreline.

meters. Unexpectedly, most of the polygons follow the border of the raised beaches and develop in correspondence with stratigraphic layers dipping toward the sea, imaged by ground penetrating radar (GPR) profiles and interpreted as prograding layers toward the present-day shoreline. Foothills. In the latter zone extensive studies of the isotopic composition and thermal regime of ice wedges have been performed. [10][11][12] It is important to highlight that ice wedges are a ground ice type ubiquitous in areas of continuous permafrost, and investigating the response of such features to climate change is helpful to understand future permafrost degradation. 13 Since ice wedges are typically encountered in the upper permafrost, polygonal terrain is particularly susceptible to thermokarst. 14 Polygonal patterned ground and ice wedges develop in a wide range of surficial materials due to thermal contraction cracking of the ground in winter and meltwater infilling of the cracks. However, the physical sensitivity of polygonal terrain varies with spatial patterns in ice-wedge size, thermal regimes, topography, and climate (e.g., Bockheim and Hall 15 ). Recent warming has caused ice-wedge degradation in many sites 16 with further expected effects as a consequence of ongoing increasing warming. Therefore, since polygonal patterned ground responds in a non-linear and nonuniform manner to increasing temperatures, it is essential to fully understand all the factors involved in their origin and evolution through time. Only in this way can such features (and in particular ice wedges) be used as a proxy of climate changes.
With these objectives in mind, 1,17 found that the rate of sublimation is related to substrate and micro-topography, while Hallet et al. 1 modeled the development of sand-wedge polygons in the dry valleys of Antarctica, finding that the age was controlled by long-term rates of trough extension, and ridge growth depends on the equilibrium between sand-wedge widening and relaxation by loose sediments. On the contrary, the geometry and evolution of hexagonal or orthogonal polygonal patterns have been related to different factors, including climate, substrate, and deposit age. 18 It was further pointed out that cracks do not develop all at once, but rather new cracks are influenced by the already-existing ones, in turn resulting in a self-organizing pattern with varying angles of intersections. 19 On the contrary, Mellon et al. 20 found a strict correlation between polygon size and depth of the boundary between almost dry and ice-rich permafrost, analyzing patterns in the Antarctic dry valleys.
New approaches have been used for quantitative analysis of polygon geometry, for instance by Haltigin et al. 21 who used spatial point pattern analysis, showing that polygon networks increase in regularity with time. 22 mapped polygons using high-resolution stereo remote sensing data to extract polygon geometries on wide areas, while several methods based on different approaches, including neural network and deep learning ones, have been recently proposed for automatic or semi-automatic extraction of the polygon shape from satellite and aerial photographs. 13,[23][24][25][26] Polygonal patterned grounds raise peculiar interest also because they were found not only on the Earth surface but also on Mars, and there are several geometrical similarities between the two planets, even though on Mars they sometimes exhibit unusually wide dimensions (diameter > 100 m). 27 However, the presence of polygonal patterns on both planets suggests that thermal contraction cracking is (or was) a generic process acting in both situations. Antarctic polygons provide better terrestrial analogues of Martian polygons than Arctic polygons, due to the cold, hyper-arid environmental conditions of Antarctica; therefore, in recent years several studies focused on polygons as Mars analogues in different zones of the Antarctic continent, including Dry Valleys, 8,28 northern Victoria Land, 9,29 and Antarctic Peninsula, 30 but also in other areas of East Antarctica (e.g., Matsuoka and Hirakawa 31 ).
In our study area statistical evaluations are unpractical and not meaningful due to limited extension in which the polygonal pattern develops. So, we adopted an alternative approach based on integrated morphological and geophysical investigations, the latter mainly focusing on ground penetrating radar (GPR). GPR is the most frequently used geophysical tool for imaging the internal structure of periglacial and permafrost features (e.g., previous studies [32][33][34] ). However, up to now GPR was seldom applied to specifically image materials in areas characterized by polygonal patterns. For instance, Bode et al. 35 combined GPR and remote sensing data to estimate ice-wedge volume.
Three-dimensional (3D) GPR has been successfully used for highaccuracy imaging of subsurface polygon networks 36 and ice complex deposits. Previous studies 37,38 extended 3D GPR into imaging buried networks of sediment-filled relict wedges in the USA, while Watanabe et al. 39 imaged ice and active-layer soil wedges by 3D GPR in the Svalbard, analyzing and discussing factors controlling ice-wedge distribution in a maritime, periglacial environment. 3D GPR was further used for comparative studies between Earth and Mars polygonal patterns. 40 Despite such 3D GPR examples, there are logistical constraints limiting its applicability especially when rough topography and large blocks and boulders are present at the surface.
Electrical resistivity tomography (ERT) is a geophysical technique that can be used to obtain information about variations in subsurface electrical conductivity. Since ice and frozen materials are more resistive than unfrozen and/or water saturated materials, ERT is particularly efficient in glaciological studies and specifically in permafrost systems (e.g., 41 ). As far as patterned ground and ice-wedge investigations are concerned, ERT was seldom used alone 42,43 but because of the different resolutions and sensitivities to electrical conductivity and permittivity of ERT and GPR, they can be complementary to each other and their combined use can produce novel insights. 44 In this paper, we integrated GPR data with ERT focusing on the following objectives: 1) Analyze the relationship between polygonal patterned ground and the geological and sedimentological structures of a permafrost beach environment in northern Victoria Land.
2) Analyze polygon morphology borders where wedges are expected, by characterizing filling materials with non-destructive techniques.

| STUDY AREA
Within the Gerlach inlet in northern Victoria Land, very close to the Korean "Jang Bogo" and German "Gondwana" Antarctic stations, a raised beach studied since the first Italian Antarctic expeditions 45,46 rests directly on bedrock (mainly gneiss of the Wilson Terrane, 47 Figure 1). According to, 48 at least 12 prominent beach ridges can be recognized up to 30 m asl, and the rates of emergence range from <2 mm yr À1 in the last 3 kyrs to >10 mm yr À1 for the oldest beaches (ca 7,500 yrs BP). The analyzed beach is within a continuous permafrost area, although characterized by a quite thin high resistivity permafrost layer overlying low resistivity values. 49 Geomorphologically, the beach is also characterized by the occurrence of abundant "ventifacts" that testify the strong catabatic winds flowing in this area for a large part of the year. 50 Climate is arid and very cold with a mean annual air temperature (MAAT) of À14.0 C at Baia Terra Nova, 51 and an annual net precipitation (precipitation minus evaporation) of less than 100 mm yr À1 . 52  MHz) antennas that gave worse results due to limited penetration depth and higher overall scattering related to the size of stones on and below the surface (see Figure 2). In addition, in the case of lower frequency antennas (which are always bigger in size) the rough surface produces a worse coupling with more diffractions responsible for an overall lower signal-to-noise ratio. The GPR instrument was triggered by an electro-mechanical odometer allowing a constant trace A detailed analysis of the meaning of such attributes is out of the scope of this paper, but a comprehensive analysis related to glaciological application can be found in. 53 ERT encompasses two partially superimposed profiles, collected with a Pasi system in Wenner configuration. Each profile has 64 electrodes, 2 m spaced and a superimposition equal to 16 electrodes showing the smaller diameter. Twelve polygons are located at elevations between T1 and T2, while just one wedge was intercepted by GPR profiles at elevations higher than T2 (11 m asl). In addition, it is interesting to notice that four points are just along the T1 limit, while F I G U R E 2 Photograph of the study area in November 2014 taken during the geophysical data acquisition. Borders of polygonal patterns are emphasized by snow accumulation and are pointed by the yellow arrows. In the foreground is seen a portion of the ERT equipment. View from south to north [Colour figure can be viewed at wileyonlinelibrary.com] six others are along (or very close to) the T2 limit. In Figure 3 it is also apparent that several borders of the polygons follow the limits of both T1 and T2. For completeness in Figure 3 we also map the top of the current berm (Tb), which has an elevation of 1-1.5 m asl.  Figure 4) there is a resistivity lowering in the shallowest portion, sometimes moderate (i.e., about half of the surroundings), and sometimes very high (up to two orders of magnitude). We have to remark that the electrode spacing was equal to 2 m, giving accurate results starting from about 1 m below the topographic surface. Another noticeable behavior is that the resistivity decreases at the beach ridges (T1 and T2) and that in these zones there are more polygon borders (Figures 3 and 4). Close to the actual sea side (i.e., toward SE) in the shallower part there are resistivities below 100 Ωm probably due to salt water-saturated sediments. In the same zone, just below the mean sea level elevation the resistivity increases up to thousands Ωm, values that could be attributed to permafrost conditions below the sea bottom, but further data must be considered to verify such an hypothesis, which is far from the subject of this paper. In any case, such resistivity trend is real and not related to inversion border effects as demonstrated by the comparison between apparent and real resistivity models ( Figure S3 supplementary).

| METHODS
Since the GPR maximum penetration depth does not exceed 4- Attributes are helpful in the interpretation process since they emphasize some data characteristics. For instance, the cosine of the instantaneous phase allows us to follow reflectors by exploiting their lateral phase behavior even when the reflection amplitude (as well as all the amplitude-related attributes) is discontinuous. On the contrary, the sweetness allows us to highlight zones having a similar GPR signature, since it is an attribute that combines together amplitude and spectral characteristics of the signal.
In the study area, we interpreted three main GPR reflectors having different characteristics, geometries, and EM signature highlighted by attribute analysis. In particular, a high-reflective and continuous horizon (H1), sub-parallel to the topographic surface (yellow dashed line in Figures 5 and 6), lies at depths between 30 and 80 cm. Seaward  Horizon H1 can be interpreted as the frost table, as suggested in 55 for an area close to the sea and less than 20 km to the south from F I G U R E 4 Ground penetrating radar (GPR; top) and electrical resistivity tomography (ERT; bottom) profiles collected along the same path (see Figure 3 for location). ERT inversion has RMS % error equal to 4.9 after five iterations. Tab  Gondwana station. It is, as expected, almost parallel to the topographic surface and in some places is laterally continuous, while in others it exhibits an irregular signature. This horizon is interrupted at the polygon sides ( Figures 5, 6, 7, 8) where there is always a strong phase discontinuity emphasized by the cosine of phase attribute. This confirms the pattern of soil temperature found by 56 just some hundred meters away from the study area, which measured ground surface temperatures at the borders of some (albeit smaller) polygons up to 2.8 C warmer than at the centers. However, it can be also explained by a coarser material enhancing the scattering and/or to limitations in GPR imaging of the steeply dipping frost table reflector deepening at polygon borders. In any case, the geophysically estimated frost table depth is in agreement with, 57 which reported an active layer thickness ranging from 2 to 92 cm in unconsolidated deposits (while it exceeds 1.6 m in granitic bedrock).
Horizons H2 are seaward-dipping reflectors mapped along all the central profiles, while they are not present in the two southernmost profiles (Fig. suppl. S2) where the bedrock is almost outcropping (Figures 1, 3). As a consequence of that, they can be interpreted as prograding sedimentary levels related to different beaches, raised during the Holocene. 9,48 They locally deviate from the general SE dipping trend, with deformations and more complex geometries (Figure 7).
Horizon H3 is not continuous in the study area and locally disappears in the central and northern profiles, while it is not present in the southernmost ones (Fig. suppl. S2). It is most probably interpreted as the top of the bedrock below the sedimentary levels, although we cannot exclude that it could be related to different (probably glacial) older sediments.
Besides all the usually considered factors driving and influencing the polygon development such as climate, substrate grain size, deposit age, active layer geometry, permafrost characteristics, and hydrology, we found that most of the polygons borders in our study area develop at beach ridges (i.e., borders of terraces), (Figure 3) and/or where some preexisting stratigraphic layers reach the topographic surface ( Figures 4-7). This is a remarkable result because in a limited area, as the one selected for the study, most of the other factors are constant or very similar.
The analyzed polygons have narrow borders and null to decimetric furrows and therefore are difficult to be studied by any geophysical method. We, therefore, focused on GPR technique, which provides a resolution that is higher than that of the other available geophysical techniques. 58 GPR surveys to study ice wedges are quite common in the literature (e.g., 59,60 ), but all these cases seem to be considerably different both in terms of wedge scale and physical characteristics from our case study. In fact, most of the studies focus on the Arctic where ice wedges typically are several meters wide, with depths exceeding 10 m and well developed on areas of tens or even hundreds square kilometers (e.g., 13 ). Quite surprisingly, in 61 we found wedges with an EM signature similar to the ones analyzed here (see  , the EM signature of the polygon borders is spatially variable, but distinctive, with the local interruption of the frost table, which lies at depths between about 1 and 2 m. However, the geological and geomorphological setting is very different from our study area since within the crater there are mainly carbonates and carbonate breccias covered by thermally derived, poorly sorted angular rocks ranging from a few centimeters to 5 mm in size. 61 Although beyond the specific aims of this paper, we can make some preliminary estimates of the chronology of the polygons, which are present only at elevations above 6-7 m higher than the present sea level. Assuming the maximum uplift rates proposed by Baroni and Orombelli,48 which are equal to 2 mm yr À1 for the last 3,000 years and 10 mm yr À1 before, the minimum age for the polygon formation is ca. 3,300 years BP. We should also consider that at around 15 m asl (i.e., about 4,000 years BP with the same uplift rates) on this beach

| Characterization of the infilling wedges material
Generally, wedges in Antarctica are small: about 1 m wide in the dry valleys 6 and 0.5-1.5 m deep and 0.1-0.5 wide in northern Victoria Land. 9 In continental Antarctica sand wedges and frost fissures (or tessellations according to 9 ) are relatively widespread in areas not perennially covered by ice, being most common in the dry valleys and the later in northern Victoria Land. The presence of ice wedges in vast areas of northern Victoria Land was reported by 66 but here we focus on an ice wedge found by one of the authors (R.R.) and a frost fissure already discussed in 9 filled by an openwork matrix of stones and pebbles, suggesting lateral slippage into the crack (both located in Figure 3).
We remark that while some borders exhibit transparent zones just below their emergence at the surface, others are, in contrast, characterized by strong reflections and diffractions (Figure 8). Most of the borders are in correspondence with stratigraphic layers dipping toward the sea and with the beach ridges T1 and T2 (Figures 5-8).
Since we acquired GPR profiles almost perpendicular to the seashore, they should therefore intersect the stratigraphic structures nearly per- In particular, the frost table, which is usually very continuous on GPR data (yellow dotted lines in Figure 9a and 9c), is interrupted in correspondence to both the ice wedge and frost fissure and occurs roughly at the same depth of the observed trenches and in the range found almost everywhere in northern Victoria land by Raffi and Stenni. 10 The ice does not have a distinctive signature on GPR data due to the limited lateral extension of the wedges and the resolution of GPR signal. The geometry of ice wedges and especially their lateral sub-vertical limits cannot be accurately imaged on GPR data as reported also by, 67 but its presence and location can be indeed detected on GPR data. The lateral limits of the frost fissure are clearer with a distinct phase interruption, possibly because of their lower dip. Below the frost fissure the signal amplitude is generally lower than in the other zones of the profiles at the same depth; this in turn can be due to higher EM attenuation of the materials filling the frost fissure.
In general, the following tentative interpreting criteria to discriminate ice wedges and frost fissures can be proposed:

DATA AVAILABILITY STATEMENT
The data that support the findings of this study are available from the corresponding author upon reasonable request.