Characterization and mapping of MIS‐2 thermal contraction crack polygons in Western Transdanubia, Hungary

The Pannonian Basin was located in the southernmost, disputed limit of permafrost during the Last Glacial Maximum (LGM). In the western part of the basin, over an area of 1,200 km2, more than 150 sites with polygonal patterned ground were surveyed, and 72 sediment samples from forms identified as relict sand wedges were collected. Ten optically stimulated luminescence ages were obtained from the infills, while morphometric analyses were also carried out on satellite images. Our study revealed that the polygonal networks developed in several phases, from 15.01 ± 1.68 to 23.0 ± 1.7 ka. The polygons have an average diameter of 13–23 m and are mainly present on flat surfaces, intruding into the gravelly, alluvial host of the paleo‐Rába. Statistical analyses highlighted the short transportation period of the sandy infill and multiple sediment provenances. This study adds further data to assess the presence of permafrost or deep seasonal frost and to the interpretation of the LGM in the central European periglacial domain.


| INTRODUCTION
Networks of polygonal cracks in soils can be developed in various environments on Earth. They can be formed due to frost action, desiccation, or even pedogenic processes. [1][2][3] Investigation of these features offers valuable insight into past paleoenvironmental changes as well as recent surface processes. This paper, however, only deals with polygonal networks derived from ground thermal contraction. This process occurs when decreasing temperatures in the frozen ground generate tensile stresses greater than the tensile strength of the ground. Such stresses are favored by both rapid cooling and low temperature, 4,5 and therefore polygonal cracks related to thermal contraction are present in both recent (e.g., Watanabe et al, 6 Wolfe et al 7 ) and relict (Rodríguez-L opez et al, 8 Ewertowksi et al 9 and references therein) periglacial environments. Polygons originating from thermal contraction cracking in frozen ground are one of the most wellrecognized and widespread landforms in permafrost-controlled lowlands. 10 According to the remote sensing investigations of Ewertowski et al, 9 seven types of relict polygonal networks can be distinguished, based on their diameters, regularities, and intersecting angles.
Thermal contraction cracks are of numerous types based on their infilling material (e.g., ice, sand, ground, composite) and their chronological status (active or relict forms). Primary and secondary structures need to be distinguished within the term "sand wedge." When the wedgeshaped cracks are filled exclusively with allochthonous sand during their lifetime, they are called primary structures. 11,12 Former, melted ice wedges that are later filled with sand are called secondary structures. 13 If inactive, the wedges are referred to as relict sand wedges. 14 The distribution of the above-mentioned past periglacial features has been mapped in most of western and central Europe, but the identification and paleoenvironmental significance of such features remain uncertain. 7,14 Based on environmental proxies (i.e., ice wedge pseudomorphs, composite pseudomorphs, sand wedges, cryoturbations, etc.), previous studies have tried to establish the spatial extent/zones of permafrost types within the Pannonian Basin during the Last Glacial Maximum (LGM). 15 Although some early reconstructions proposed that almost the entire basin was affected by permafrost, [16][17][18][19] these estimations did not distinguish between permafrost types. Later, Vandenberghe et al, 15 based on existing archival data, established a map of the Last Permafrost Maximum permafrost extent in the Northern Hemisphere. Caution must be exercised because there are limitations to these reconstructions, such as the topographical influence on climate-controlled permafrost, as stated by Ruszkiczay-Rüdiger and Kern. 20 At that time, however, sand wedges found in France were considered indicative of continuous permafrost, because active sand wedges were only known from Antarctica.
Since then, Wolfe et al 7  Landforms related to periglacial processes were described in the 20th century in the Pannonian Basin. [23][24][25][26][27] Recently, cryogenic features were further investigated by Fábián et al 28 13,22,[30][31][32][33][34][35] In recent decades, only a few studies have investigated the grain size distribution (GSD) of relict sand wedge infillings. [11][12][13][35][36][37][38][39] Primary infillings-according to the papers cited above-are commonly well sorted, allochthonous sandy deposits, which indicates a cold environment with strong eolian activity. In Hungary, grain size analysis of relict sand wedge infillings is lacking. However, other evidence of former intensive wind activity has already been detected. 40,41 This paper describes the characteristics of polygonal patterned ground in the Kemeneshát region, western Transdanubia, discusses the main sedimentological properties of the infilling material, and proposes a time-frame for their formation based on OSL dating.

| STUDY AREA
The analysis was performed on a regional scale, over the Lower and Higher Kemeneshát microregions in the western part of Transdanubia, Hungary ( Figure 1).
Kemeneshát was located ca. 600 km from the maximum southern limit of the Fennoscandian Ice Sheet (FIS) during the LGM (26.5-19 ka 42 ). After retreat of the ice sheet, at its smallest extent (15 ka), this distance increased to 900 km.
The study site is located in an area bounded by three rivers (River Satellite data from the DigitalGlobe Foundation provide a high-quality dataset with a sub-meter horizontal resolution. Due to their high temporal resolution, it is also possible to find images acquired under favorable weather conditions. The current analysis was based on WorldView-1 panchromatic images of 0.5-m horizontal resolution, acquired on January 8 and 21, 2008. Despite the high spatial resolution, there are patches with low polygon visibility. Therefore, satellite images were simultaneously analyzed in Google Earth Pro™ using the "Historical Imagery" function, since images between 2006 and 2017 were also found to be useful to detect polygonal patterns. Detection was also highly influenced by the varying image quality, unfavorable weather conditions or temporary surface cover (snow or tailing material from the mine).
The applicability of satellite images was checked at two test sites (one by mapping cropmarks, and another one with an excavated surface) by using an unmanned aerial vehicle in the summer of 2019, during a field work campaign. Aerial images were taken with a DJI MAVIC 2 PRO drone, with its Hasselblad L1D-20c camera. The flight altitude of the drone was 50-100 m, depending on the investigated area. Surveying was followed by manual processing of the images, which resulted in orthophotos.
Using GIS data, the topographical and geological characteristics of the delineated forms were also studied. 45 Based on SRTM-1 (30-m horizontal resolution) 46 the elevation, slope, and aspect distribution over these polygonal networks were analyzed.

| Field work
In total, 72 samples from the infilling material of the wedges were collected in the pits of the Szemenye gravel pit mine. The quarry is

| Grain size analysis and end-member modeling (EMMA)
The same laboratory performed all sample analyses. After standard drying, the samples were sieved over a 2-mm screen with a Fritsch Analysette Pro 3 shaker. Then, 10 g of each sample was pretreated to remove CaCO 3 and organic matter. The organic matter was removed by hydrogen peroxide (30%, 15-30 ml, depending on the length of the chemical reaction), while carbonates and iron oxide were removed with hydrochloric acid (10%, 15 ml). 47 After sample preparation, grain size was determined with a laser diffractometry Malvern Mastersizer 3000 laser diffractometer (Malver Inc) particle size analyzer.
Raw data were processed in the GRADISTATv8.0 software, 48 where simple descriptive statistical parameters (e.g., mean, sorting, skewness, and kurtosis) were calculated. In the calculations, the Folk and Ward method was used. 49 In addition to this calculation, EMMA was also used for further analysis of the sediments.
EMMA is used to unmix sedimentary dynamic populations and was first developed by Weltje. 50 Since its first use, it has widely been applied in sedimentological analyses. [51][52][53][54][55][56] The calculations include the estimation of necessary end-members (EMs; based on the mean total explained variance), the determination of end-member scores and loadings, and the class-and sample-wise explained variance. For this, the EMMAgeo package for R was used. 57 To obtain a high-quality, reliable model, the possible uncertainties of EMMA 52,58 were reduced in our case as follows: (a) a standard laser diffraction method was used for grain size measurement; (b) the model parameters were optimized (empty bins were removed from the dataset, the number of samples and grain size bins were more

| Dating the infilling material-OSL
By using OSL it is possible to determine the last time sediment grains were exposed to sunlight; that is, the time of sediment formation or the time of burial of the grains. Where sand was readily available, the OSL ages correspond to the formation of cracks within the error limit of OSL dates.
Sampling was done by hammering steel tubes into the sand wedges ( Figure 2). The preparation of samples followed standard laboratory techniques. 59 All procedures were carried out in subdued yellow light provided by low-pressure sodium lamps. The samples were dried to constant weight to determine in situ water content. Coarse grain sand was separated using sieves of 150-and 220-μm mesh size. The carbonate and organic fractions of samples were removed by repeated acid treatment in 10% HCl and 10% H 2 O 2 . Extraction of the quartz fraction was dine using heavy liquid flotation (LST Fastfloat). Finally, 45 min of etching in 40% HF was performed to remove any remaining feldspar residuals and the outer shell of quartz grains. Quartz extracts were spread on 1-cm stainless steel disks using a 4-mm mask size.
The equivalent dose (D e ) of samples was determined using a RISØ DA-20 TL/OSL-type luminescence reader. Irradiation was made using a calibrated 90 Sr/ 90 Y β-source. OSL was detected with an EMI ET9107-type photomultiplier. Stimulation was carried out using blue LEDs (470 ± 30 nm), while detection was made through a Hoya U-340 filter. Throughout the measurements, the single aliquot regeneration (SAR) protocol was applied. 60,61 Preheat and dose recovery tests were performed to identify the best measurement parameters and assess the reproducibility of measurements.
Environmental dose rate (D*) was determined using high-resolution, extended-range gamma-ray spectrometry, applying a Canberra-type HpGe detector and 450-ml Marinelli beakers. Dry dose rates were calculated using the conversion factors of Liritzis et al. 62 Wet dose rates were assessed based on in situ water contents. 63 The rate of cosmic radiation was determined by considering burial depth, and geographic location, following the equation of Prescott and Hutton. 64 Since the wedges were only a few decimeters in width, the radioactive element content of both the sand in the wedges and the gravelly host material were measured. Dose rates were weighed according to the volumetric share of the two materials in a 30-cm-diameter sphere around the sample.  Table S1).
The comparison in Table 1 confirms that the infilling material consists of almost exclusively sand (92.8-96.3%), while the host sediment consists mostly of sandy gravel (49-75% gravel). The clay and silt fraction is also slightly more abundant in the fill (3.5-7%) than in the host sediment (0.8-1.6%).
The EMMA calculations resulted in four different EMs ( Figure S2).
Their weighting scores are at least 16% (EM1), and at most 39% (EM2) of the total modeled volume. Every EM GSD is polymodal, with two EM modes in the medium sand fraction (EM3 at 1.7 φ and EM4 at 1.5 φ), one in the fine sand fraction (EM2 at 2.1 φ), and one in the very fine sand fraction (EM1 at 3.5 φ). Each has additional minor modes, in the sand and silt domains.

| Luminescence characteristics and ages
For samples OSZ 1386 (SZ 2/1 profile) and OSZ 1390 (SZ 2/5 profile), a combined preheat and dose recovery test was performed. In the tests, we investigated the optimal temperature at which the luminescence properties of the sample are the most stable. The parameters considered were: recycling ratio and dose recovery ratio should ideally be close to 1, while recuperation should be below 5%, and each  parameter should have a maximum error of 10%. 61 Those aliquots that did not pass the criteria listed above were rejected and not used for further calculations.
The measurements show that the recycling and dose recovery ratios for most samples had a large uncertainty at each preheat temperature ( Figure S3). For example, for sample OSZ 1390, the recycling ratio was below 1.1 only at 220 C. The results are closest to the ideal usually set at 220 C and the accuracy of results was also best at this temperature ( Figure S4). Thus, preheat temperature was set to this value during the SAR measurements.
SAR measurements generated a low intensity (i.e., the samples were relatively insensitive), which made the evaluation difficult. Thus, more than three-quarters of the aliquots of the measured samples did not meet the SAR criteria and had to be rejected.
Note that some aliquots that produced low signal intensities also showed lower equivalent doses. However, because the results of these samples did not meet the evaluation criteria, they did not affect the calculation of the equivalent dose for the sample.
Samples were evaluated following the recommendations of Arnold et al. 66   Based on our observations, we suggest the use of satellite images for bulk datasets owing to the smaller file size and ease of processing, while drone imagery can be a useful tool for individual study sites for more detailed analysis.

| Characteristics of the infilling
Poor and very poor sorting values suggest that the transportation process of the material did not last long. According to the summary of Murton et al, 14 this seemingly heterogenic infill is uncommon, but not absent for sand wedges. The inconsistency among particle size distributions attests to the complex origin of the infilling material, and the mean sand:silt:clay ratio of 67:32:1 reinforces this uncertainty. As stated previously, the strongly cemented host material is very different from the sandy infilling. In contrast to the observations described by many authors (e.g., Black, 71 Murton and Bateman 72 ) the examined wedges lack vertical lamination. This was probably due to the melting of ice veins or secondary perturbation.
The polymodal distributions of EM loadings, along with the GSDs point to brief wind activity. The grain size distribution suggests that the material was mainly transported by rolling and saltation (coarser fractions), while modified saltation and short-term suspension also played an important role in the process of filling of the frost cracks T A B L E 1 Comparison of fractions of the host sediment and the infilling material of the wedges. Grain size classes were determined according to Blott and Pye. 65 Field ID (lab code) Gravels and boulders >16 mm were removed from the samples before sieving to avoid distorted results. In this case, infilling samples were sieved for comparable results, and therefore the data are in mass percentage.
(cf. figure 1 in Tsoar and Pye 73 ). There is also a notable amount that was probably transported in long-term suspension (silt and clay fractions), although their source has not yet been identified. However, the infill may originate from a mixture of eolian sand and material provided by the collapse of the surrounding host material (in the present case gravel). When eolian sand is unavailable, the collapse of host sediment is the only process involved in the filling of the crack. 74,75 In addition, the fine-grained fraction may derive from washing of the host sediments, so there is no clear evidence of an allochthonous origin of this fraction.

| Absolute ages and regional correlations
The OSL age estimates reliably date the time of wedge development rather than post-depositional wedge modifications. 33  Since OSL ages were determined on aliquots each comprising multiple grains potentially belonging to different phases of wedge filling, the estimates obtained here should be considered indicative of the main periods during which the filling was formed.
Strong wind activity was reportedly present in the Pannonian Basin during the Pleistocene, as testified by mega-yardangs 41 and ventifacts. 40

| The interpretation of paleoclimatic proxies in the Pannonian Basin-limitations and possibilities
There are several proxies which have been used to reconstruct paleoclimatic conditions during the LGM in the Pannonian Basin, such as data retrieved from groundwater recharge temperature, 78 pollen analysis, 79 and malacological data, [80][81][82][83] 93 using the same climate model but considering a representative ground level of À1.5 m, were able to reconstruct discontinuous permafrost in Hungary. Active layer thickness (ALT) was found to have ranged between 2 and 3 m in the region.
Together, this suggests strongly that (a) the area was close to the southern permafrost boundary and (b) the sand wedges developed within the active layer or in a context of deep seasonal frost (depending on local parameters). The sand wedges have an average depth of 2-3 m, which is around the same as the probable ALT. This also suggests strongly that the lack of vertical lamination in the wedges is not due to the melting of ice veins but more probably to secondary perturbation.
After considering European past permafrost-related field evidence, we cannot exclude the possibility of a spatially and temporally fluctuating environment in the Pannonian Basin, thus varying from permafrost to deep seasonal frost. Based on our findings, we can say that among the several, already identified cooling periods (i.e., the possibility of permafrost), the Pannonian Basin may have only been affected by the last one (MIS 2 GS 2.1a-c), since no older OSL ages have yet been found in the region. We must keep in mind that conventional OSL dating of sand wedges does not allow the identification of distinct phases of sand filling, but provides only an estimate of the main period of filling. However, distinguishing the exact paleoenvironmental conditions cannot be made with certainty.

| CONCLUSIONS
Sand-filled thermal contraction cracks are more common relict landforms in the western part of Hungary than elsewhere in the country. Within the study area, the polygons generally appear at 210-230 m a.s.l., mostly in gravel and loess. The infilling material for the wedge-shaped forms is poorly sorted fine sand. The youngest OSL date indicates that eolian activity may have taken place as late as 15.01 ± 1.68 ka BP and such material was able to fill the cracks. As far as the sand filling was synchronous or just slightly delayed compared to crack formation, we assume that climatic conditions during the dated periods were cold enough in winter to generate patterns of soil thermal contraction cracks at the study site. Our research has shown that high-resolution satellite imagery is applicable for the mapping of large polygon networks, while drone photos are more suitable for detailed analysis over a single study site of thermal contraction crack polygons. OSL dating was a useful tool for determining the timing of such geomorphological processes, and based on the compared data (from France), three or four phases of wedge formation are presumed.
Although our data did not ascertain the exact period of frost-related activity in the Pannonian Basin, the geomorphological, sedimentological and absolute dating evidence indicate that cold and dry climate episodes in MIS GS 2 shaped the landscape in the Pannonian Basin.

ACKNOWLEDGEMENTS
DigitalGlobe Foundation is gratefully acknowledged for the satellite images. The authors are also grateful for Professor Barbara Woronko (University of Warsaw) for useful comments that improved the quality of the manuscript. The authors thank reviewers Pascal Bertran and Michel Allard, as well as editor in chief Professor Mauro Guglielmin for their valuable comments and remarks that significantly contributed to the paper.

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