Ice‐wedge polygon dynamics in Svalbard: Lessons from a decade of automated multi‐sensor monitoring

Twelve years of continuous monitoring of diverse ground properties reveals the dynamics of three ice wedges and adjacent ground in a low‐centered polygon area in Svalbard. The monitoring documented ground displacements, the timing of crack generation, ground thermal and moisture conditions from the surface to the top permafrost, and snow conditions. The focus is on seasonal ground deformation in and around ice‐wedge troughs, interannual variability of ice‐wedge activity and thermal thresholds for ice‐wedge cracking. Seasonal ice‐wedge activity is mainly associated with frost heave and thaw settlement, as well as thermal expansion and contraction. In mid‐ to late winter, temporary expansion and cracking of troughs by thermal contraction occurs during rapid cooling periods. Following intensive ground microcracking events, troughs show rapid expansion and in some cases major cracking in the frozen active layer. A common threshold for cracking is identified by a combination of ground surface cooling below −20°C and a thermal gradient steeper than −10°C m−1 in the upper meter of ground, indicating that cracking requires both a brittle frozen layer and rapid ground cooling. Our results highlight that in marginal thermal conditions for ice‐wedge activity, the primary control on ice‐wedge cracking is rapid winter cooling enhanced by minimum snow cover.

criteria based on the observed dynamics of both seasonal deformation and thermal contraction cracking. 4,11 In contrast to the significant number of descriptive stratigraphic and morphological studies, the scarcity of quantitative data defining the environmental conditions causing ground motion and cracking in polygonal terrain has prevented direct validation of climatic indicators controlling ice-wedge dynamics. Only a few theoretical and field-based studies have addressed the dynamics of ice-wedge polygons. Based on a visco-elastic theory of thermal contraction, Lachenbruch 12 proposed that the occurrence and geometry of polygonal cracks depended on the coefficient of thermal expansion, thermal conductivity, viscosity and tensile strength of the frozen ground, which are dependent on the ground temperature, cooling rate, soil properties and ice content. Plug and Werner 13

| Key questions
The above review, and that by Christiansen et al., 4 suggest that the following issues need to be solved to understand the dynamic of icewedge cracking and thus to use it as a more detailed climatic indicator: • seasonal movements of polygonal ground (trough-ridge morphology); • timing, magnitude and frequency of ice-wedge cracking; • thresholds for ice-wedge cracking; • controls on interannual variability of cracking.
To address these issues we have conducted a comprehensive field monitoring study of ice-wedge dynamics in a marginal environment for ice-wedge formation (MAAT ca −3°C) using diverse instrumentation and manual techniques. 10,21,22 The monitoring campaign started in 2002 and intensified in successive years. This paper focuses on highresolution ground deformation and thermal contraction cracking of three ice-wedge troughs and associated ridges using 12 years (2005-2017) of data on horizontal and vertical soil movements, cable breaking, acceleration events, air and soil temperatures, soil moisture and snow conditions. The long-term dynamics of polygonal ground, including seasonal deformation, and the spatial variation in cracking activity are under current study, but are beyond the scope of this paper.

| THE STUDIED ICE-WEDGE SITE
The ice-wedge research site is located on the outermost part of a large late Holocene alluvial fan 23,24 27,28 (Figure 2). The surface morphology shifts into small polygons (< 3 m in diameter), earth hummocks and mudboils toward the loess-free area higher on the fan. 22,24 The low-centered polygons are delimited by well-defined troughs 0.2-1 m wide and 0.1-0.4 m deep, centered between symmetrical ridges (eg, 18 ). The large polygons are subdivided in places by secondary or tertiary cracks, with narrower and shallower troughs often lacking ridges. There is a distinct vegetation zonation, with in particular mosses in the wet centers, tall grasses in the troughs and sporadic low grasses on the ridges, in response to the microtopography of the polygons. Drilling and geophysical investigations show that most of the troughs are underlain by ice wedges below the 0.8-1 m deep active layer, and that the width of trough represents the minimum width of the ice wedges (typically 0.5-3 m). 29 The active layer thickness (ca 1 m) is typical of the sediments in the Adventdalen valley. 30,31 The frozen active layer and top permafrost are rich in segregation ice, having gravimetric water content of about 50% (excess ice content about 15%). 25 MAAT was −5.1°C and annual precipitation was 192 mm from 1990 to 2004 at the Longyearbyen Airport meteorological station 10 km northwest of, and at approximately the same elevation as, the monitoring site (data derived from eKlima 32 ). There MAAT rose to −2.6°C during the monitoring period (2005-2017). 32 Large air temperature fluctuations between 0 and − 20°C to −30°C are common in winter, due to the maritime setting with alternating weather systems, with either low-pressures coming from the south or polar high-pressures extending over the polar areas, including Svalbard. The winter condition contrasts with relatively stable positive temperatures (5-8°C) in summer. 30 Shallow snow (< 0.4 m) covers the polygons from October to May, but with varying depth from 0-0.1 m over the ridges to 0.3-0.4 m in the troughs and central polygons. 10 The thin snow cover, mainly due to persistent wind erosion in the valley location, favors efficient ground cooling throughout winter.
Positive temperatures occasionally occur even during midwinter, causing extensive snowmelt and subsequent ground cooling with formation of a thin ice-cover, such as observed from January to

| Instrumentation
Three ice-wedge troughs TR1, TR2 and TR3 were monitored  Cracking activity is monitored using a combination of three independent, complementary subsystems: extensometers (expansion and contraction of the ground), accelerometers (intensity and frequency of cracking; hereafter shock loggers) and breaking cables connected to timing devices (timing of major cracking events). The details of the monitoring set-up at each trough are described in Supporting Information Appendix S1. Data were recorded at 1-hour intervals (otherwise noted in Appendix S1) and offloaded every summer or at shorter intervals. Most of the subsurface sensors and angle-iron frames were installed in boreholes 50 mm in diameter, drilled with a motorized hand-held auger and refilled with the same soil or, when antiheaving was required, with fluvial gravel.

| Thermal parameters
In this paper, the thermal state is reported in terms of the ground surface temperature (T S ) represented by the value at 0.02 m depth, the temperature at the top of the permafrost (T TOP ), the temperature at 1.0 m depth (T 100 ), the thermal gradient in the active layer (G AL ) calculated from the T S and T TOP values, and the cooling rate at the surface (R SC ). When associating R SC with a cracking event, we used an average 3-day rate preceding the event.
The following parameters are derived from data on air and soil temperatures. The annual mean air temperature (AMAT), annual mean ground surface temperature (AMST) and annual mean temperature at the top of the permafrost (AMTTOP) are calculated for hydrological   The ALT in the ridge at TR1 fluctuated between 1.0 and 1.4 m (mean 1.2 m) over the 12-year period. ALT showed a slight increasing trend, but the rate during the second half of the period (ca 0.04 m y −1 ) was twice that during the first half (ca 0.02 m y −1 ) (Figure 6c).
The volumetric water content (VWC) of the active layer had a large consistent seasonal variation ( Figure 7g). VWC remained low and stable (10%-20%) in all winters when the soil was frozen, whereas the value at 0.2 m depth rose to 50%-60% just after thawing and then gradually decreased to about 30% with significant fluctuation until refreezing: note that values below 30% and above 50% are outside the calibration range (Supporting information Appendix S1). The high moisture content in the initial part of the thawing period is consistent with the presence of wet ground and ponded water in many troughs and polygon centers. The daily photographs show that the thawing period is always initiated with 1-1.5 months of lakes filling the lowcentered polygons, occurring in the period from mid-May to mid-July.
During some wet summer periods VWC at 0.2 m depth fluctuated in response to rainfall. When compared with meteorological data at Svalbard Airport, 32 rapid VWC increases coincided with rainfall (usually >10 mm d −1 ), while decreasing VWC occurred during rain-free periods of several weeks, although rain events did not always induce significant wetting. At 0.4 m depth the thaw season VWC values were consistently 40 ± 2% in all summers.

| Ground dynamics at TR1 (first-order ice-wedge crack)
All the automated monitoring results at TR1 had nearly continuous data sets for 12 years (2005-17) (Figure 7). This first-order ice-wedge trough has the longest monitoring record. The vertical extensometer on the ridge recorded annual cycles of seasonal frost heave of 34 ± 8 mm and thaw settlement of 58 ± 2 mm during 5 years  July. This indicates that frost heaving is primarily associated with the upper part of the active layer, whereas heaving due to upward freezing from the permafrost table 33 seems less important.
The horizontal extensometers showed contrasting movements between the trough and ridge (Figure 7b, d). The trough shrank in early winter and expanded slightly in early summer, whereas the ridge did the opposite at the same time but with larger movements. The accumulated seasonal movements resulted in year-by-year expansion of the ridge and contraction of the trough, with the former about 50% larger than the latter (Figure 7b, d). In contrast, the trough extended slightly during the second half period (Figure 7d). The reason is unclear, but the rigid angle irons may have resisted further inward tilting.
Combining the horizontal and vertical movements (Figure 7a, b) suggests that the ridge cracks opened during frost heaving in winter (type a in Figure 8) and partially closed during thaw consolidation in summer (type b in Figure 8), showing a feature of a dilation crack.
The trough did not respond to thaw settlement of the ridge, probably because its delayed thawing prevented the deformation of the rigid frozen substrate. In mid-August, when much of the active layer had thawed, the trough shrunk over a short time, contemporarily with the ridge crack opening by about 20 mm without vertical movement (type c in Figure 8). This opening could be attributed to shrinkage due to desiccation of the topsoil around the ridge crack, although the soil moisture level at 0.2-0.4 m depth was stable during this period ( Figure 8).
The trough-ridge morphology was generally stable throughout the In addition, the broken shallow wire was not replaced in summer 2016 because the water-filled trough prevented excavation and reinstallation, which led to missing data in winter 2016/17. The disconnection

| Ground dynamics at TR2 (third-order icewedge crack)
The near-surface temperature at TR2 showed similar seasonal conditions as at TR1, fluctuating between about 10°C in summer and − 15 to −20°C in winter (Figure 10d). The sensors at 0.9 m or below never rose above 0°C, indicating that the permafrost table was close to 0.9 m.
The horizontal extensometers showed more complex movements than at TR1 (Figure 10a); they involve three types of   (Figure 7). These results imply that this ice wedge is active. Due to the uncertainties, however, the activity of TR2 is excluded from further analysis.

| Ground dynamics at TR3 (first-order ice-wedge crack, most frequently cracking)
TR3 experienced soil thermal regimes similar to TR1 and TR2, but the deeper trough favored snow accumulation, leading to slightly slower cooling and smaller ground temperature variation in winter

| THERMAL CONTRACTION CRACKING EVENTS IN LATE WINTER
The monitoring systems indicated significant cracking activity in at least eight out of the 12 winters. Figure 12 illustrates detailed processes of the late-winter cracking recorded at TR1 and TR3. Significant cracking events were identified by rapid expansion of the trough, the occurrence of wire disconnection across the trough and/or the magnitude and frequency of acceleration events. These cracking events are symbolized as "C" plus year (eg, C05 represents the cracking event in 2005) and, when an event has multiple phases, subdivided  Table 1. For comparison between   the monitored troughs, the temperature near the permafrost table is represented by the value at 1.0 m (T 100 ) instead of T TOP . At TR1 T 100 equals T TOP . At TR3 T 100 is given by (T TOP + T 120 )/2, where T TOP is recorded at 0.8 m and T 120 is the temperature at 1.2 m.
The amount of thermally induced crack opening in winter was estimated from the difference between the extensometer reading just before the rapid expansion and the maximum width in late winter (Table 1). To derive the opening at the ground surface, the data were corrected with reference to tilting of the frame and upheaval of the frame due to frost heave and settlement (Supporting information Appendix S2). Such estimations and corrections were conducted for the winters accompanied by rapid expansion events.

| Event C10
The 2009/10 winter was unusual among the observed winters, encountering extensive melting of snow cover due to rainfall and formation of a water pool in late January, and subsequent intensive cooling that produced extensive ice cover in February. 30

| Event C11
Only the shallower copper wire at TR3 broke on February 16 2011, which was confirmed by excavation in August 2011. The disconnection took place during a cold phase, but cooling was not effective, expansion was small (< 1 mm) and no ground acceleration was registered. Minor, short-lived contraction could have produced the cracking. When the wire was broken T 100 was already low (−11°C) comparable to the cracking condition in the other winters, but T S dropped to −14.6°C and R SC reached only 0.4°C d −1 (Table 1).

| Event C12
Event C12 was contemporarily recorded at bothTR1 and TR3, involving two steps of distinct, rapid expansion (a, b) at TR3 (Figure 12e Table 1). The total expansion at the surface (W S ) including the two steps reached 5.8 mm at TR1 and 6.0 mm at TR3.

| Event C14
The Whether thermal contraction or another mechanism induced this event is unclear.

| Event C17
The

| Thresholds for cracking
The thermal behavior of frozen ground has been described by a viscoelastic model in which the total strain comprises thermal expansion and contraction, elastic deformation and viscous relaxation. 12,34 Assuming that the creep behavior of frozen soil follows Glen's flow law, Lachenbruch 12 (1962) derived the horizontal thermal stress σ H at a given depth that asymptotically approaches a value for a constant cooling rate: where α is the coefficient of linear thermal expansion, R C is the cooling rate and B is a creep parameter (inversely related to viscosity) indicative of deformability of frozen soil. The parameter B is primarily temperature-dependent, increasing with rising temperature toward 0°C. 35 This equation indicates that, for a given soil type and ice content, the tensile stress at a given depth increases with decreasing temperature and increasing cooling rate. The model is in general agreement with observations from northern Alaska, 12 Svalbard 36 and Martian high latitudes, 34 and is supported by our field observations showing that colder (more brittle) frozen ground subject to more rapid cooling favors thermal contraction cracking.
Consequently, based on frozen ground rheology, the threshold for thermal contraction cracking can be expressed in terms of two basic parameters, T TOP (or T 100 ) and the thermal gradient of the frozen active layer (G AL ). We mainly use average values of G AL in the whole active layer instead of the cooling rate, because the latter is variable on both temporal and spatial (depth) scales and thus difficult to define the representative values. The significance of the two parameters is supported by our results as the timing of effective cracking (indicated by rapid expansion and/or wire disconnection) generally corresponds to lowering of both T TOP and G AL ( Figure 12, Table 1). The minimum condition for cracking at TR1 is approximately given by a combination of T 100 < −8°C and G AL < −8°C m −1 . The threshold at TR3, albeit derived from fewer data, appears to be slightly milder (T 100 < −5°C  Table 1).
The threshold applies to fine-grained soils where the buffer layers where the ground is considered to have had minimum snow or vegetation covers. To define more precise and universal criteria for ice-wedge cracking, however, requires long-term monitoring in various permafrost and substrate conditions.

| Interannual variability of ice-wedge dynamics
The 12-year monitoring allows us to assess interannual variability of ice-wedge dynamics and its deviation from the long-term trend. This distinction is crucial for marginal ice-wedge environments like the Svalbard study site, where ice wedges may cease to crack as the climate warms, thereby impacting the microrelief and hydrology. Contrary to common belief, they provide little information about former mean annual air/ground temperature.

| CONCLUSIONS
The long-term, multi-method monitoring of ice-wedge activity in Svalbard leads to the following conclusions.
1. The dominant processes driving near-surface seasonal ground motion at the perimeter of ice-wedge polygons are frost heave, thaw settlement, and thermal expansion and contraction.
2. Rapid cooling events during mid-to late winter often triggered temporary expansion and cracking of the troughs by thermal contraction of the surrounding permafrost, indicated by ground acceleration events, followed by rapid trough expansion and in some cases cracking in the frozen active layer. Thermal contraction cracking occurred in at least eight out of the 12 winters, although the occurrences were spatially variable.