Climatic, geomorphologic and hydrologic perturbations as drivers for mid‐ to late Holocene development of ice‐wedge polygons in the western Canadian Arctic

Abstract Ice‐wedge polygons are widespread periglacial features and influence landscape hydrology and carbon storage. The influence of climate and topography on polygon development is not entirely clear, however, giving high uncertainties to projections of permafrost development. We studied the mid‐ to late Holocene development of modern ice‐wedge polygon sites to explore drivers of change and reasons for long‐term stability. We analyzed organic carbon, total nitrogen, stable carbon isotopes, grain size composition and plant macrofossils in six cores from three polygons. We found that all sites developed from aquatic to wetland conditions. In the mid‐Holocene, shallow lakes and partly submerged ice‐wedge polygons existed at the studied sites. An erosional hiatus of ca 5000 years followed, and ice‐wedge polygons re‐initiated within the last millennium. Ice‐wedge melt and surface drying during the last century were linked to climatic warming. The influence of climate on ice‐wedge polygon development was outweighed by geomorphology during most of the late Holocene. Recent warming, however, caused ice‐wedge degradation at all sites. Our study showed that where waterlogged ground was maintained, low‐centered polygons persisted for millennia. Ice‐wedge melt and increased drainage through geomorphic disturbance, however, triggered conversion into high‐centered polygons and may lead to self‐enhancing degradation under continued warming.

degradation was, for example, recorded during the early Holocene thermal maximum in the western Canadian Arctic. [5][6][7][8][9] Geomorphological processes triggered by lake drainage or sea level rise may, however, affect topography and surface hydrology on a local to subregional level. This may cause polygon growth or degradation independently of the regional climate trend. [10][11][12] The respective influence of climate and geomorphology on the evolution of different types of ice-wedge polygons is not well understood because of large temporal and spatial discrepancies between climatic forcing and geomorphological response processes. In this study we therefore investigated past landscape dynamics on millennial time-scales to discriminate climate-driven and geomorphology-driven changes on ice-wedge polygon development.
We addressed the spatial heterogeneity 13,14 within individual icewedge polygons, by applying a multi-proxy approach, studying six peat cores from three different ice-wedge polygons, each with one core from the polygon center and one core from the polygon rim or margin.
We addressed the following specific research aims:

Reconstruction of ice-wedge polygon development on the Yukon
Coastal Plain during the mid-to late Holocene.

| BACKGROUND
Ice-wedge polygons are most widespread in regions underlain by continuous permafrost. 15,16 They develop in areas with a very low relief energy, where drainage is impeded and the ground stays permanently waterlogged. 17 They are characterized by wedge-shaped ice in the ground, which builds up over decades to millennia through repeated thermal contraction cracking during winter and meltwater infiltration into the cracks in spring. 18 We are using the term ice-wedge polygon in the sense of polygonal peatlands, i.e., peat-forming areas underlain by a network of ice wedges that show a surface expression in the form of raised rims and/or low-lying troughs.
One way to classify different morphological types of ice-wedge polygons is to distinguish low-centered polygons from high-centered polygons. 15,19 Low-centered polygons are characterized by raised rims on either side of polygonally adjoining ice wedges enclosing a central depression. Surface flow is impeded, yet not completely prevented, where this type prevails. High-centered polygons are thought to develop from low-centered polygons due to (i) improved drainage causing (melt) water flow and thermal erosion along ice wedges, (ii) self-organization through lateral material displacement as the underlying ice wedges grow wider and rim material is pushed toward the centers of polygons 15 and/or (iii) increased air temperatures promoting ice-wedge thaw and wetter polygon troughs. 4 Relief inversion and an altered landscape hydrological regime result from the conversion. 4 The raised center consecutively dries up and may be eroded, 20,21 while thermal erosion along ice wedge pathways may enhance transport of material into adjacent landscapes.
Ice-wedge formation may be related to large-scale climate trends.
Thermal contraction cracking requires severe ground frost in winter, 19,[22][23][24] which may be provided by a combination of low ambient temperatures and a thin snow cover. Cracking has been shown to be more frequent in peat than in mineral soil. 22,23 Ice-wedge polygon development also requires sufficient moisture supply. Ice wedges are fed primarily by hoar formation within cracks in winter and by water from snowmelt and rain in spring. These drivers of ice-wedge polygon development may, in turn, be influenced by the vegetation cover.
In particular, growth height and functional group composition determine effectiveness of ground insulation [25][26][27][28] and snow retention potential. 29 Alterations in any of these factors (winter temperatures, snow cover, moisture supply, vegetation composition) may cause changes in cracking frequency or degradation of ground ice, and ultimately trigger changes in ice-wedge polygon morphology.
Ice-wedge polygons also experience drastic geomorphological changes, most recently induced by permafrost thaw. Increased thaw has been observed to produce thicker active layers and degrading ice wedges, 4,12,30,31 while stabilization of deeply degraded ice wedges has been reported to be a result of thermal insulation through the accumulation of organic debris. 30 Increasing wetness due to increased thaw of ice-rich permafrost is thought to be reversed in the long run, as increased evapotranspiration during warmer, longer summers is predicted to reduce moisture in the active layer as well as surface water in ponds and lakes. [32][33][34] Such ambiguous effects acting on various temporal and spatial scales all relate to the interplay between climatic and geomorphological drivers.
Studies of long-term ice-wedge polygon development have shown that ice-wedge polygons may exist in a relatively stable state over millennia. 12,[35][36][37] They are, however, vulnerable toward changes in air temperatures, precipitation and geomorphological disturbance. Recent studies have underlined that ice-wedge polygons may degrade over the course of years to decades as a response to such changes. 20,30

| STUDY AREA
The study area is situated on the terrestrial part of the Canadian Beaufort Sea shelf. It is characterized by a subarctic, maritime climate, a flat to slightly undulating topography, and ice-rich unconsolidated sediments shaped by periglacial processes in the western part and by Pleistocene glaciations superimposed by periglacial processes in the eastern part. 7 TheYukon Coastal Plain stretches across 240 km of coastline from the Mackenzie Delta in the east to the Alaskan border in the west and is bor- About half of the scarce precipitation falls as snow, resulting in a thin snow cover (mean 25 cm), which is locally variable due to strong wind redistribution and prevails for 250 days per year on average. The topography of the plain is characterized by a flat coastal zone and rolling hills toward the Mountain range. This study focused on the flat coastal reaches, which were shaped by (i) Late Pleistocene advances of the Laurentide Ice Sheet, which reached its furthest extent about 16.2 ka BP 9,39 and (ii) paraglacial and periglacial processes thereafter. The unglaciated landscape west of about 139.6°W was subject to periglacial conditions throughout the Quaternary, and is characterized by flat, low-lying wetlands and ice wedge growth. The moraine landscape in the eastern part has higher coastal cliffs composed of thick glacigenic deposits. This leads to large elevation differences between the tops of moraines and the base level of stream erosion and results in relatively deeply incised valleys and generally larger elevation differences than in the unglaciated part. Typical periglacial features on the Yukon Coastal Plain include thermokarst lakes, many of them at least partly drained, ice-wedge polygons, pingos and retrogressive thaw slumps.
Peatland development is favored by continuous permafrost with a shallow active layer (mostly less than 50 cm) and an abundance of low-lying ground.
A permafrost depth of 142 m has been documented near Roland Bay. 40 The tundra vegetation is dominated by mosses, sedges and dwarf shrubs, 41,42 with sedges (Carex sp) dominating sites with impeded drainage, and tussock cottongrass (Eriophorum vaginatum) dominating better drained, elevated surfaces. 31 Dwarf shrubs associated with wetlands include various Ericales, Salix spp., Betula glandulosa and Rubus chamaemorus, while in river valleys sheltered conditions promote taller growth of the shrubby taxa Salix spp., Alnus crispa and Betula glandulosa. 43 We investigated the mid-to late Holocene development of three ice-wedge polygons situated in the western and central coastal reaches of the Yukon (Figures 1, 2). Polygon morphology and vascular plant taxa composition have been summarized in Wolter et al. 31 Komakuk Polygon (Figure 2a) was formed outside the recon-  Polygon, an additional permafrost core (PG2161) was drilled directly subjacent to the active-layer core we retrieved from the polygon center, as the active layer itself was rather shallow (14 cm beneath the ridge and 22 cm beneath the center). The total core length for Ptarmigan Polygon center was 88 cm, including both active layer core and permafrost core. Due to logistical considerations, the permafrost core was photographed, described, and subsampled in 4-5 cm increments in the field before it thawed.

| Laboratory analyses
The six active-layer cores were photographed and described in the laboratory, before being subsampled in 1 cm increments. In three cores (Komakuk Polygon ridge, Roland Polygon centre, Ptarmigan Polygon ridge), the lowermost samples could not be reasonably divided further, so the lowermost 1.5 or 2 cm was taken as one sample. In total, 24 radiocarbon dates were obtained from terrestrial plant macrofossils (Table 1) picked from selected samples. The plant fragments were pretreated with standard acid-alkali-acid (AAA) extraction using 1% HCl (1 h at 60°C plus 10 h at room temperature) and 1% NaOH (4 h at 60°C), which was removed by washing with MilliQ water. 44 For very small or very fragile samples the extraction time  46 We calibrated the radiocarbon dates using CALIB 7.1 (calibration dataset Grain size analyses were carried out on carbonate-free and organic-

| Data and statistical analyses
The zonation presented for the cores was delineated using the In a second PCA we assessed all 165 core samples based on sediment data (TOC, TOC/TN, δ 13 C). Sediment data were rangetransformed to ensure the data are on the same scale. PCAs were conducted using the function "princomp" in R. Hellinger and range transformations were executed using the function "tran" in the package analogue 53 in R.
Zone KP c 2 (0-13 cm depth) uniformly showed very high TOC contents, increasing TOC/TN toward the top of the core and decreasing δ 13 C ( Figure 3a). The grain size composition was classified as silty sand. The amount of plant material rose to 100% in this zone ( Table 2).

| Ptarmigan Polygon
The permafrost core from the center of Ptarmigan Polygon had a  In zone PP c 2 (0-23 cm depth), organic matter was characterized by high TOC contents, while TOC/TN ratios were similar to those found in zone PP c 1, and δ 13 C was slightly lower than in PP c 1 ( Figure 4a). The sediment texture was silty sand. In this zone, samples consisted nearly entirely of Cyperaceae peat, yet identifiable plant macrofossils were nearly absent, consisting of one fragment of a dwarf shrub twig and one Carex sp. seed ( Table 3).  (Table 3). In the lower part of the core, they consisted of Cyperaceae with a low amount of Bryophytes and one small leaf

| Roland Polygon
The active layer core from the elevated center of Roland Polygon (YC12-RP-Mc) showed a hiatus but no age inversions ( Table 1)   In RP r 2 (0-11 cm depth), TOC contents were high and TOC/TN ratios increased strongly upcore, while δ 13 C was stable (Figure 5b).
There was no information on grain size composition for RP r 2, as the peat contained very little inorganic material. Plant macrofossils were dominated by abundant remains of the mesic terrestrial taxa Betula glandulosa, Ledum decumbens and Eriophorum vaginatum (Table 4).

| DISCUSSION
The results of sediment and plant macrofossil analyses on the six short cores suggest that all sites experienced change ( Figure 6). In the mid-   The peat samples are subdivided into submerged, wet and mesic polygon environments. The wet polygon samples show a wider range (larger dissimilarities) for TOC contents, but especially in δ 13 C (associated with PC2) than those of submerged polygons, but they overlap largely.
Overall, organic-rich lake sediment from shallow lakes is clearly separated from peat of submerged, wet and mesic ice-wedge polygon microsites.

| Center core
The center core exhibited two zones that we interpreted as lake sediments overlain by peat. In zone KP c 1 (ca 1600-1450 cal years BP), fine-grained sediments with TOC values around 10 wt.% indicated the presence of a lake environment rather than peat from an icewedge polygon. While lake sediment in deeper parts of lakes in the region exhibits slightly lower TOC, lower C/N and higher δ 13 C, 55,58 the sediment we found resembled a transitional phase between lake drainage and ice-wedge polygon initiation identified in a study from Herschel Island 12 as well as a phase of low lake water level reconstructed in a study from a lake near Roland Bay. 38 The plant macrofossil assemblage showed a mixture of mesic terrestrial, wet terrestrial, emergent and submerged aquatic taxa, indicating a highly productive shallow-water environment in close proximity to drier terrestrial reaches. Modern satellite imagery showed the outline of a drained lake basin (Figure 2a). The studied polygon was situated in the marginal part of that former lake, which still existed during KP c 1.
In KP c 2 (at least 400 cal years BP to modern) peat established.
Strong parallels in composition, plant macrofossil assemblage and thickness of this peat layer and the dated peat in KP r 1 from the same ice-wedge polygon suggested a similar age range for KP c 2 ( Figure 3).

| Rim core
The rim core featured a lower sediment facies typical of a low-centered polygon superseded by a hiatus that we interpret as an erosion surface, and recent peat accumulation in the upper part of the core.
KP r 1 was dated to the two millennia around 5000 cal years BP, with an age inversion in the lowest sample ( Figure 3, Table 1), suggesting a mid-Holocene age range for KP r 1. Fluctuating high TOC contents indicated either decomposing peat or varying input of inorganic material.
Good preservation of plant macrofossils and narrow ranges in C/N and δ 13 C showed that organic material composition was stable, while organic matter contents varied, suggesting that peat decomposition played a minor role. The pattern was probably caused by varying input of finegrained sediment originating from sporadic disturbances. Plant macrofossils comprised mesic and wet terrestrials, and emergent and submerged aquatics. This assemblage indicated a highly structured wetland as found in low-centered polygons with sufficiently deeply submerged centers to allow the growth of submerged Potamogeton (cf Hannon and Gaillard 56 ).
KP r 1 was followed by a hiatus of~5000 years, which coincided with a facies break. The polygon center core had a basal age of 1600 cal years BP, which placed the facies break in that core in a time slice lost from the rim core, indicating that lake sedimentation could have been active there at least after 1600 BP. We interpreted the upper surface of KP r 1 and KP c 1 as an erosion surface.
The peat in KP r 2 developed within the past 300 years, as indicated by the results of radiocarbon dating (Table 1). Radiocarbon dates from this timeframe are generally ambiguous (deVries effect, 66 Suess effect, 67 Table 3), suggesting that the site was not located within the productive littoral zone of a lake, but in a slightly deeper, more central part, in which few terrestrial plant remains would be expected. In PP c 2 (post-bomb dates), sedge peat established, as evident from stable high TOC contents, consistently low C/N ratios ( Figure 4) and Cyperaceae remains. These stable modern conditions in the center of the low-centered polygon showed no indication of drier or wetter conditions or disturbances.

| Rim core
The polygon rim core consisted of one peaty sediment horizon. The core showed peat accumulation since 1100 cal years BP (Table 1, Figure 4). During that time, polygon rim conditions remained relatively stable, as indicated by stable TOC contents and grain size composition.
A rise in C/N ratios was accompanied by an increase in dwarf shrub macrofossils toward the top of the core (Figure 4). This indicated drier conditions on the polygon rim in the recent past. Improved aeration in drier peat facilitates microbial activity and peat decomposition, and the gradual increase in δ 13 C values along the core could have been caused by increasing microbial utilization of carbon, which discriminates against the lighter 12 C and thus leads to 13   A sharp increase in C/N and a drop in δ 13 C indicated that carbon increasingly derived from terrestrial plant sources. 73 TOC stayed very high and exceptionally stable, and thus we infer that the carbon signature did not present a decomposition signal, but an alteration in carbon source, toward more mesic plant taxa, particularly to an increase in the deciduous dwarf shrub Betula glandulosa.

| Margin core
The core showed peat of different genesis: the lower zone indicated a shallow submerged environment superseded by peat typical for lowcentered polygons and a hiatus we interpreted as an erosion surface, until in the upper zone peat formation was re-initiated. The margin core from Roland Polygon was located only 4 m from the center core, and basal dates (~7000 cal years BP) matched the center core. RP r 1A was, however, not made up of lake sediment but of peat from wet terrestrial plants, as indicated by very high TOC, relatively low C/N and high δ 13 C.
The plant macrofossil record contained no aquatic plants. Instead, mesic terrestrials, large amounts of Carex seeds and some Daphnia resting eggs (ephippiae) were found ( Figure 5, Table 4). The genus Carex contains semiaquatic species such as Carex aquatilis, which often dominates aquatic communities in tundra ponds associated with ice-wedge polygonal terrain (eg, Bliss 74 ). Daphnia is found in partly submerged areas around lakes or in ponds (eg, Gliwicz 75 ). We suggest that during the mid-Holocene an ice-wedge polygon with a seasonally or permanently submerged pond existed in the shallow reaches of a lake as seen around modern lakes in the region (Figure 2a, c).
During the period corresponding to RP r 1B drier conditions established, indicated by decreasing δ 13 C, rising C/N, decreasing amounts of Carex seeds, absence of aquatics and increasing dominance of mesic terrestrials ( Figure 5, Table 4). The vegetation mosaic reflected typical moisture gradients found in ice-wedge polygons in the region (eg, Wolter et al 31 ). Radiocarbon ages in RP r 1B ranged from ca 5000 cal years BP to dates within the last 300 years. The zone was capped by a distinct facies break, at which a hiatus of nearly 5000 years occurred within 3 cm of sediment (Table 1). This may have been caused by lateral displacement or decomposition of peat. We suggest that erosive action, rather than decomposition alone, caused the removal of material, as no signs of intensive decomposition were found in adjacent layers. A similar erosion surface was found in Komakuk Polygon, where it was most prominent in the polygon margin as well.
RP r 2 comprised modern peat that formed within the last 300 years.
Very high and uniform TOC contents indicated stable peat accumulation.
The shift toward drier conditions that we saw in the polygon center core was repeated here, with C/N decreasing strongly and Carex disappearing.
This supported evidence for conversion from a low-centered polygon to a high-centered polygon, probably as recently as 100 years ago.
Roland Polygon was located at the margins of a lake during the mid-Holocene and at least seasonally submerged. At some point after lake drainage, erosive removal of material created a~5000-year hiatus. The central part stabilized and has been accumulating peat in a low-centered polygon since 600 cal years BP, and the margin followed during the last 300 years. The modern high-centered polygon probably emerged during the last century.

| Climate vs. geomorphic disturbances as drivers of change in ice-wedge polygons
The prerequisites for ice-wedge polygon development (waterlogged ground, permafrost, extreme ground-penetrating cold during winter) are determined by climate and geomorphology. Ice-wedge polygon initiation and conversion of low-centered into high-centered polygons is therefore strongly related to the dynamics of and the interplay between the two.
Investigations into radiocarbon dates have revealed broad climateinduced simultaneous patterns of peatland initiation. 76,77 Strong seasonality and high summer temperatures have been suggested as drivers of intensive peatland formation during the Holocene thermal maximum in Alaska. 77 Our study of mid-to late Holocene ice-wedge polygon development found spatially heterogeneous peat formation in polygons around 7000 cal years BP (after the regional Holocene thermal maximum), under conditions much wetter than today ( Figure 6). We found no climate-induced peat initiation in the following millennia, when regional climatic patterns were largely stable. In the last millennium, however, evidence for lake drainage and peat accumulation in Komakuk Polygon and Roland Polygon during the regional Little Ice Age (ca AD 1600-1850; see [78][79][80] ) suggests a climatic link. Polygon was the most stable over time, yet the polygon rim changed from Cyperaceae-dominated to dwarf-shrub-dominated, indicating drying (Table 3). Roland Polygon showed complete development from a low-centered to a high-centered polygon. All three polygons showed signs of recent ice-wedge degradation. 31 The conversion of one polygon type to another may result from internal self-organization through two main processes: lateral movement of material adjacent to ice wedges may widen ice-wedge troughs and displace material toward the polygon center, where a mound establishes. 19 Vegetation growth in polygon centers exceeding the upward growth of the surrounding ice wedges may also result in a well-drained mound of peat surrounded by water-filled trenches. 21,83 Both processes act on time-scales of centuries to millennia, contrasting with the rapid conversions we found.  89 The question of whether disturbance triggered later drainage of the polygon centers and finally led to relief inversion cannot be answered at this stage, but will be worth investigating.
The changes we observed in peatland initiation and change from low-centered to high-centered were mostly caused by geomorphological change (sea-level rise, tapping and draining of adjacent lakes, changes in drainage pathways across the landscape). In permafrostaffected landscapes, climatic change may trigger widespread geomorphological change, especially where unconsolidated ice-rich sediments dominate. Such climate-induced geomorphological change may have locally variable impacts, but its frequency is likely to increase under climatic change. Regionally synchronized ice-wedge polygon development requires a higher amplitude and seasonality of temperature and precipitation change than evident for the mid-to late Holocene. Our findings indicate that modern warming, however, may have triggered regional-scale conversion from low-centered polygons to highcentered polygons. This process may rapidly initiate irreversible selfenhancing erosion of ice-wedge polygons.
Roland Polygon experienced stability for at least 2000 years during the mid-Holocene (ca 7000-5000 cal years BP, Figure 5, Table 4 Our results indicate that a non-negative water balance was the main factor promoting stability during low-amplitude climatic fluctuations.
Stable low-centered polygons prevail when continued moisture supply outweighs drainage and evapotranspiration. In contrast, decreasing moisture supply from the surrounding landscape or increasing drainage caused by geomorphological processes such as coastal erosion, thermal erosion or thermokarst trigger conversion into high-centered polygons.

| CONCLUSIONS
We reconstructed mid-Holocene and Holocene landscape features in coastal lowland tundra on the Yukon Coastal Plain. We traced the development of shallow lakes to low-centered ice-wedge polygons and subsequently to high-centered polygons. During the mid-Holocene, the studied sites contained shallow lakes or submerged polygon centers and paleobotanic data generally reflect wetter conditions than today. An erosional hiatus of ca 5000 years indicates disturbance and erosion in high-and intermediate-centered polygons which formed later on. In recent decades, ice-wedge polygons on the Yukon Coastal Plain experienced degradation and drying through warming-induced geomorphological change. In our study, the main driver of (i) icewedge polygon initiation was lake drainage. The main driver triggering (ii) conversion of low-centered polygons to high-centered polygons was improved drainage through ice-wedge degradation and changes in the local topographic gradient. By contrast, we found that stable conditions prevailed for millennia during the late Holocene in icewedge polygons under low-amplitude climatic change as long as a non-negative water balance was maintained in the polygon field.
Hence, geomorphic disturbance was the main driver of locally variable ice-wedge polygon dynamics in periods of low climatic forcing.
Extreme climatic change triggered simultaneous developments, such as widespread peat initiation during the Holocene Thermal Maximum.
Modeling of the development of polygon fields through time must thus focus on temperature constraints as well as landscape water balance and flow paths.