Spatial distribution of soil organic carbon in northwest Greenland and underestimates of high Arctic carbon stores



[1] The amount of soil organic carbon (SOC) in the high Arctic is generally poorly constrained. Because of periglacial processes such as frost churning and sequestration in frozen soils, a substantial amount of SOC is typically not inventoried. This study provides a detailed study of SOC content by depth in 55 soil pits in a high Arctic ecosystem of northwest Greenland. Sampling sites spanned ecosystems from mires to polar deserts, from sea level to the margins of the Greenland Ice Sheet, and across various periglacial features. The amount of SOC in the various ecosystems was mapped using a correlation of SOC with high-resolution ASTER satellite imagery and the normalized difference vegetation index (NDVI) classes from the Circumpolar Arctic Vegetation Map. On the basis of this correlation, the total carbon was extrapolated to greater areas of the high Arctic. Our study found the amount of SOC in the high Arctic has typically been grossly underestimated, remarkably by the greatest amount in the most barren environments of the polar desert. We estimate that the high Arctic contains about 12 Pg SOC, a factor of over 5 times greater the most cited values previously reported. Since our estimate was only assessed in seasonally frozen ground, additional carbon frozen in the permafrost is likely present and potentially available in the event of permafrost thawing due to warming of the Arctic.

1. Introduction

[2] One direct feedback to rising greenhouse gases in an anthropogenically warmed environment is the release or storage of organic carbon in vegetation and soils [Smith and Shugart, 1993; Oechel and Vourlitis, 1994; Christensen et al., 1999; Kirschbaum, 2000; Joos et al., 2001; Melillo et al., 2002; Luo, 2007]. To assess this potential release of carbon and evaluate its effect on climate change, accurate estimates of the quantity and distribution of soil organic carbon (SOC) are required. Current estimates assert that global soil ecosystems contain 2–3 times as much carbon (1395–2400 Pg) [Bohn, 1982; Post et al., 1982; Schlesinger, 1984; Batjes, 1996] as is in the atmosphere (750 Pg) [Intergovernmental Panel on Climate Change (IPCC), 2001], elucidating the importance of this pool and its potential to both affect and be affected by future global climate change [Jenkinson et al., 1991; Smith and Shugart, 1993; Oechel and Vourlitis, 1994; Christensen et al., 1999; Cox et al., 2000; Kirschbaum, 2000; Joos et al., 2001; Melillo et al., 2002; Jones et al., 2005; Knorr et al., 2005].

[3] While it has been recognized that the Arctic (particularly the low Arctic) contains substantial stores of SOC [Post et al., 1982; Miller et al., 1983; Gorham, 1991; Bliss and Matveyeva, 1992; Kimble et al., 1993; Gilmanov and Oechel, 1995; Michaelson et al., 1996; Tarnocai, 2000; Ping et al., 2008], few comprehensive, quantitative studies of SOC have been conducted in the high Arctic (a region defined as north of the 4°–6°C July mean average temperature [Bliss, 1979]). One widely cited reference of high Arctic SOC content sampled only carbon in the upper 25 cm of soil [Bliss and Matveyeva, 1992]. As soil profiles in the high Arctic generally extend beyond 25 cm depth, and cryoturbation transports carbon to depth, this estimate does not quantify fully the amount of high Arctic SOC. Horwath et al. [2008] reported that 62% of SOC associated with pervasive periglacial nonsorted stripes in fine-textured soils of the high Arctic was located below 25 cm.

[4] Permafrost conditions drive many physical processes that shape the landscape and mediate soil carbon dynamics in the Arctic. Water migration toward freezing fronts results in ice lens formation in soils and resultant frost heave. Repeated heave cycles lead to soil mixing (cryoturbation), often expressed as broken or irregular horizons and convoluted deposits of buried organic carbon [Tedrow, 1962, 1965; Vandenberghe, 1988; Bockheim and Tarnocai, 1998]. Soil carbon also accumulates along the permafrost contact or is frozen within the permafrost as it aggrades [Tedrow, 1965; Kimble et al., 1993; Bockheim and Tarnocai, 1998]. SOC in the low Arctic has been underestimated substantially by not accounting for SOC at depth [Kimble et al., 1993; Michaelson et al., 1996; Bockheim et al., 1999], hence it is probable that SOC in the high Arctic has also been underestimated since most estimates are based on the top 25 cm of soil.

[5] Because of its short growing season and near-surface permafrost, Arctic ecosystems are expected to elicit profound responses to global climate change (summaries given by IPCC [2001]). As air temperatures continue to rise and permafrost thaws, previously inactive stores of Arctic SOC may enter the active soil carbon cycle and be released as CO2 or CH4 [Gorham, 1991; Smith and Shugart, 1993; Christensen et al., 1999; Tarnocai, 1999; Hobbie et al., 2000]. The high Arctic region encompasses approximately 2 million km2 of ice-free land [Bliss and Matveyeva, 1992] and therefore represents a large region of terrestrial carbon storage with the potential to release or store more carbon. With the Arctic region already exhibiting warming [e.g., Lachenbruch and Marshall, 1986; Oechel et al., 1995; Osterkamp and Romanovsky, 1996; Serreze et al., 2000; Comiso and Parkinson, 2004; Steffen et al., 2004; Overpeck et al., 2005; Ekstrom et al., 2006; Jorgenson et al., 2006], it is increasingly relevant that accurate quantifications of high Arctic SOC are completed.

[6] The objectives of this research were to (1) provide a detailed quantitative estimate of active layer SOC in northwest Greenland and determine if previous estimates of SOC in the high Arctic have been underestimated, (2) examine the spatial distribution of SOC in northwest Greenland by investigating correlations of SOC with surface variables (e.g., slope, aspect, NDVI, aboveground biomass), and (3) further determine the utility for using such correlations to estimate SOC in the circumpolar high Arctic.

2. Methods

2.1. Site Description

[7] Field research was conducted out of Thule Air Base on the northwest coast of Greenland (76°N, 68°W). The air base is located on an 800 km2 sparsely vegetated, ice-free peninsula that is bounded to the east by an outlet glacier of the Greenland Ice Sheet (Store Landgletscher), and by Baffin Bay to the south and west (Figure S1). The region near Thule Air Base is known by its Greenlandic name, Pitugfik, but henceforth in this paper, the study area will be simply referred to as the Thule region or the Thule peninsula. Place names will be introduced by Greenlandic name, followed by the English name in parentheses.

[8] Portions of the region have been covered by three to four ice sheet advances during the Quaternary [Funder and Houmark-Nielsen, 1990; Kelly et al., 1999], which deposited a relatively thin veneer of glacial drift across the majority of the peninsula. The timing and extent of these glaciations is somewhat uncertain, with periods of ice-free coverage ranging from 9,000 to 30,000 years across the peninsula [Davies et al., 1963; Funder and Houmark-Nielsen, 1990; Kelly et al., 1999]. The vegetation is generally prostrate and is dominated by Salix, Dryas, Cassiope, Saxifraga, and Carex species. Using the Circumpolar Arctic Vegetation Map (CAVM) classification system, the vegetation communities of the study area are classified as (P1) prostrate dwarf shrub, herb tundra; (P2) prostrate/hemiprostrate dwarf shrub tundra; (W1) sedge/grass, moss wetland; (B1) cryptogam, herb barren; (B3c) noncarbonate, mountain complex; (G1) rush/grass, forb, cryptogam tundra; and small regions of (B4c) carbonate mountain complex [CAVM Team, 2003] (Table 1). Vegetation cover ranges from 0% to 100%, consisting of vascular plants, lichens, and cryptogamic crusts within nonvegetated bare soil and rock (Table 1). Limited wetland (W1) communities occur in small areas of low-lying valleys in the north, extreme south (Manîserqat (Green Valley)), and in isolated pockets across the peninsula (Figure S1). The B3c community is confined to the crystalline rock region of Pingorssuit (P Mountain) in the south central region of the Thule peninsula (Figure S1). The sparsely vegetated B1 and P1 communities are generally confined to regions of higher elevation or in close proximity to the Store Landgletscher, while the remaining plant communities (P2 and G1) are widely distributed across the peninsula and occur normally at lower elevations.

Table 1. Summarized SOC Data and Site Characteristics for 55 Soil Pits in Northwest Greenland
PlotFAO Soil ClassificationVegetation CommunityaCAVM ClassNDVI ClassNDVI ValuePercent Vegetation Cover bPercent Vascular CoverDominant SpeciesAboveground Biomass (g/100 cm2)Deepest Depth (cm)Elevation (m)Percent SlopeAspectTotal Organic Carbon (kg/m2)
  • a

    Bliss and Matveyeva [1992].

  • b

    Includes nonvascular species; PSD, polar semidesert; PD, polar desert; P1, prostrate dwarf shrub, herb tundra; P2, prostrate/hemiprostrate dwarf shrub tundra; W1, sedge/grass, moss wetland; B1, cryptogam, herb barren; B3c, noncarbonate mountain complex; and G1, rush/grass, forb, cryptogam tundra.

06-2004Endoskeleti-Turbic CryosolPSDP120.123523Dryas1.9821600–12027.92
07-2004Turbic CryosolPSDP120.082715moss20.04726751807.88
08-2004Turbic CryosolPSDP1n/an/a5832crypton/a713431–206.11
11-2004Gleyi-Histic CryosolMireW140.3210074sedge5.8341070–122515.16
12-2004Skeleti-Leptic CryosolPDB110.0111n/a0.0803810n/a4.83
14-2004Turbic CryosolPSDP120.093723Dryas2.38623661357.38
15-2004Turbic CryosolPSDP240.289153Cassiope4.3602365–6011.19
19-2004Endoleptic CryosolPSDP120.082315Dryas17.1542590–124715.15
22-2004Turbic CryosolPSDP120.094122Dryasn/a972671–21351.49
30-2004Oxyaquic CryosolMireW140.329541moss6.2502210–11808.00
33-2004Gleyi-Turbic CryosolPSDP120.08n/an/an/an/a70151–204.84
39-2004Turbic CryosolPSDP1n/an/a2622Salixn/a80343329214.15
46-2004Lepti-Turbic CryosolPSDP230.186960Cassiope24.624213171809.44
49-2004Turbic CryosolPSDP240.299554Cassiope3.070168107.52
50-2004Turbic CryosolMireW150.399753moss10.0441830–1022.87
52-2004Turbic CryosolPSDP130.176851Dryas8.0541522–42256.65
59-2004Chromi-Calcic Cryosol (Skeletic)PDB120.0411n/a0.08015261800.46
61-2004Endoleptic CryosolPSDP1n/an/an/an/an/an/a40213162702.54
64-2004Lepti-Turbic CryosolPDB3c10.004612crypto8.3595338901.33
71-2004Gleyi-Turbic CryosolPDB120.055111crypto1.3804721–22252.95
74-2004Parayermi-Turbic CryosolPDB110.0054Salix0.06032823153.09
75-2004Turbic CryosolPSDP130.166739crypton/a522620–103.07

[9] The generalized bedrock geology of the Thule region consists of Proterozoic-age carbonate-rich sedimentary rocks in the northern portion, Archaean-age crystalline metamorphic basement rocks the southern region, and interspersed diabase sills and dikes [Davies et al., 1963; Escher and Pulvertaft, 1995]. Soils meet the criterion for Cryosols and Histosols [Food and Agriculture Organization (FAO), 2006] and are generally developed in glacial drift of mixed lithology or in bedrock outcrops as listed above.

[10] On the basis of weather data from Thule Air Base (1978–2003), the mean annual air temperature of the study area is −11.9°C. The average maximum temperature over this record is 8.0°C, and the average minimum is −37°C (Thule Air Base, unpublished data, 2004). Growing season (June, July, and August) temperatures from 1978 to 2002 average 3.8°C. Mean annual precipitation is approximately 108 mm, with the growing season averaging 14.3 mm (Thule Air Base, unpublished data, 2004). Soil thaw usually begins in the first week of June, with freeze back starting in early to mid-September.

2.2. Soil Pit Locations

[11] Seventy-five soil pits were excavated and sampled during midsummer (July to August) in 2003, 2004, and 2005. Data from 55 of these pits are presented in this paper, as some pits were outside the bounds of the satellite imagery or were not described in sufficient detail for this study. Soil pit widths were typically 1 m wide with depths dictated by the base of the annual thaw layer, or in some cases the depth of high water tables or extremely gravelly soil conditions. The soil pits depths ranged from 20 cm in organic-rich soils (peat over permafrost) to 96 cm in mineral soils. These active layer depths are similar to those observed in soil studies of the Canadian high Arctic by Muc et al. [1994]. Soil horizons were field mapped to scale, and digital photos were taken. Soil samples (∼500 g) were collected from diagnostic soil horizons (i.e., O, A, C, etc.), and soils were classified according to the World Reference Base for Soil Resources classification system [FAO, 1998, 2006] (Table 1). Additional parameters recorded for each site include slope, aspect, elevation, aboveground biomass, vascular vegetation cover, total vegetation cover (including lichens and cryptogamic crusts), degree of surface disturbance (i.e., patterned ground formation), and the dominant geological setting. Vegetation coverage was obtained by making point descriptions every 25 cm along a 25 m transect in four cardinal directions from the center of the pit. Aboveground biomass was obtained from a 10 × 10 cm sod sample collected at each site. Living biomass and litter were plucked from the sod, dried for 24 h at 60°C, and then weighed.

[12] The selection of soil pit locations was based primarily on normalized difference vegetation index (NDVI) classes derived from an Advanced Spaceborne Thermal Emission and Reflection (ASTER) satellite image (15 × 15 m pixels in the visible near infrared) (refer to Horwath [2007] for details). Eight NDVI classes, representing an index of relative vegetation greenness, were selected on the basis of classes used by the Circumpolar Arctic Vegetation Map (Figure 1). Low NDVI values represent sparse vegetation cover, while high values indicate bright green vegetation and/or densely vegetated areas. Translating these NDVI classes to CAVM vegetation classes, B1 and B3c correspond to NDVI classes 1 and 2; P1, P2, and G1 correspond to NDVI classes 2–4; and W1 corresponds to NDVI classes 3–8. As the surface of the study area is extremely heterogeneous, NDVI provided an objective criterion for selection of sampling sites. Secondary land surface characteristics such as bedrock geology, surficial geology (Quaternary sediments), and elevation were also considered in site selection in order to broadly distribute sites across the region and to capture the variability in geology and topography. Using GIS maps of the above characteristics, sampling sites were chosen by randomly selecting locations in the center of contiguous patches with similar NDVI values (to avoid possible edge effects) prior to beginning fieldwork. In the field, a GPS and laptop computer were used to navigate to the preselected locations. This was done to avoid a sample bias in the field. The number of pits in each NDVI class is loosely weighted by the areal coverage of each class across the study area to account for the amount of subsurface carbon variability.

Figure 1.

Color-classified NDVI map from a 26 July 2004 ASTER image. The range of NDVI values in each of the eight classes is identical to those used in the CAVM map [CAVM Team, 2003]. Small NDVI values represent sparse vegetation cover, while high values represent dense or greener vegetation.

2.3. Remote Sensing

[13] A ratio of the red (R) and near-infrared bands (NIR) (0.63–0.69 μm and 0.76–0.86 μm, respectively) (equation (1)) from a level 2 (reflectance) 25 July 2004 ASTER image was used to produce eight classes of NDVI (Figure 1). Of the cloud-free, summer images available, this date was selected because of its proximity to peak biomass. Approximately 15% of the study area was covered by clouds and cloud shadows. These pixels were eliminated prior to the areal estimation of each NDVI class. To estimate areal coverage, the number of pixels in each class was counted in Photoshop (Adobe Systems Incorporated, San Jose, California, United States, version 6.0) on a density-sliced color image generated using ENVI software (Research Systems, Incorporated, Boulder, Colorado, United States, version 3.5). The number of pixels was multiplied by the 15 m ASTER pixel size to estimate the areal coverage of each NDVI class (Table 2).

equation image

Numerous “external” variables (e.g., shade, rock, bare soil, litter, etc.) contribute to the spectral signals used to calculate NDVI [Hope et al., 1993]. The combination of these variables produces a “mixed” pixel which integrates the various signals. Each variable contributes a unique spectral signature that affects the NDVI value above the soil pit (1 m2) and, in turn, is affected by the satellite averaging spectral reflectance signals over a 15 × 15 m area (one pixel size). While this may become an important issue later, no adjusted indexes (such as soil adjusted vegetation index (SAVI)) were used to remove external factors from the spectral signals. One goal of this research was to find correlations between SOC storage and surface parameters such as remotely sensed NDVI, so that this technique might be extended to other areas of the high Arctic to better estimate SOC storage. To that end, NDVI, one of the most commonly used indexes of surface vegetation assessment, was used for comparison to other Arctic carbon studies for potential extrapolation to larger areas.

Table 2. Summary of SOC and pH Data Based on Eight NDVI Classes in the Thule Regiona
 NDVI ClassTotal
  • a

    These data show how SOC varies across NDVI classes, how SOC is unevenly distributed across the study area, and the total estimated SOC storage to the top of the permafrost table for the Thule peninsula.

  • b

    Additional four pits not included in these calculations, as they were outside of the ASTER image boundaries.

  • c

    Remaining 19.8% (158.5 km2) found in snow, ice, lakes, and cloud cover.

SOC average (kg/m2)
SOC maximum5.319.515.826.415.8
SOC minimum1.
Standard deviation1.
Average soil pH5.986.165.104.744.924.794.794.35
Number of pits5151610121151b
Number of pixels408,9841,714,500537,689174,71610,237738398124n/a
Percent cover11.548.315.
Area of class (km2)
SOC per class (1012g)0.302.381.080.500.050.0020.00150.00174.3

2.4. Soil Particle Size Analysis

[14] Soil samples were oven-dried at 60°C for 24 h and sieved with a 2 mm mesh sieve on an electric shaker (Model 150, Derrick Manufacturing, Buffalo, New York, United States). To determine the percent gravel content of each sample, the <2 mm sample fraction was weighed and subtracted from the total dry weight of the original sample. The particle size distribution of the <2 mm soil fraction was determined using a Beckman-Coulter LS 13–320 laser diffraction particle size analyzer. Procedures for sample preparation and analysis are as follows. A 5–10 g sub sample was split using a small sample splitter (Soiltest Incorporated, model CL-242A), and organic matter was removed using 10–20 mL 30% hydrogen peroxide (samples were slowly heated to 80°C to ensure complete organic digestion). The organic-free sample was centrifuged for 10 min, the clear supernatant siphoned off, and the remaining sample dried at 60°C. Samples were further split to the appropriate sample sizes, dependent on relative grain size, for the LS 13-320 analyzer (0.1–0.5 g for fine-textured to 1–2 g for coarse-textured soils). Samples were placed in 20 mL vials with 5 mL of 10% sodium hexametaphosphate and 15 mL of deionized water and sonicated for 5 min. The entire contents of the vials were then rinsed into the LS 13-320 for analysis.

2.5. Bulk Density

[15] Bulk density is a key component in the calculation of total SOC; however, it is often difficult to measure in the gravelly soils of the high Arctic [Lal and Kimble, 2001]. Traditional techniques such as excavation, standard core, and clod methods [Blake and Hartge, 1986] are impractical in remote areas with extremely gravelly soils. In relatively gravel-free areas, samples (n = 36) of each horizon were collected by pressing 4 oz. soil moisture tins into the pit face and removing carefully; bulk density is then calculated from the tin volume and dry soil weight (Table S1).

[16] For most soil samples the original volume was not precisely known, so bulk density was determined for each horizon by combining estimated bulk density for the <2 mm and >2 mm fractions. The bulk density of the <2 mm fraction was calculated by the Soil Water Characteristics Hydraulic Properties Calculator software program (version 6.02.52) to estimate bulk density based on the percentages of sand, silt, clay, and organic matter (K. E. Saxton and W. Rawls, Soil Water Characteristics: Hydraulic Properties Calculator, 2005, available at A correction term in the model accounts for natural state of soil compaction (i.e., less dense for O horizons and denser for deep C horizons) (Figure S1). This “fine” fraction bulk density was then weighted by the percentage of the <2 mm and >2 mm (“coarse fraction”) (by weight) (equation (2)). A bulk density of 2.3 g/cm3 was used for the >2 mm fraction based on average granite gravel bulk densities used by Flint and Childs [1984] and Corti et al. [2002] in studies of gravelly soils. (Even if values of coarse fraction bulk density were increased by 10% (i.e., 2.53 g/cm3), the organic content of each horizon would change by only 2%–3% in moderately gravelly soils (i.e., 30%–40%) and 3%–4% in more severely gravelly soils (i.e., 70%–80%). The impact of this increase on the total pit SOC estimate is a function of the fraction of the pit depth that the horizon occupies, with most of the very gravelly horizons occupying a small portion of the pit.) The bulk density value that includes the gravel component is referred to as the “whole soil bulk density” (Table S1).

equation image

2.6. Carbon Analysis and Calculation

[17] Organic carbon was analyzed using a Perkin Elmer 2400 CHN (carbon, hydrogen, nitrogen) elemental analyzer at the University of Washington. A 2 g subsample of the <2 mm fraction was dried at 60°C for 24 h and handground with a mortar and pestle. If carbonates were present, samples were pretreated with a 1:1 HCl solution to remove the carbonate prior to analysis. Approximately 30–50 mg of sample was placed in a tin capsule and ignited in the elemental analyzer. A standard or duplicate check was run every 10 samples.

[18] Calculations of the distribution and area of cryoturbated soil horizons requires determining the fractional area of each horizon in the soil profile [Kimble et al., 1993]. This was accomplished by tracing outlines of the soil horizons from scanned field maps of the soil horizons and converting these traces to a geoTIFF file using ArcGIS software (Environmental Systems Research Institute, Redlands, California, United States). The area of each horizon was normalized to the full profile that was sampled. The total mass of organic carbon per horizon was determined by multiplying the carbon content of the <2 mm fraction (kg/kg) by the fraction of <2 mm content, the bulk density of the whole soil (kg/m3), the fractional area of each horizon, and the total pit depth (m). These horizon values were then summed to obtain the total carbon content normalized to 1 m2 surface area of the pit (equation (3)). This equation corrects for the >2 mm gravel fraction and assumes no carbon in the gravel itself. The log of the total pit SOC estimate was used for most all statistical analyses in this study (plotted on a log linear plot) because of a nonnormal distribution of data and in order to weight all data more equally.

equation image

3. Results and Discussion

3.1. Particle Size Analysis

[19] Soils of the Thule region range from fine-textured clay and clay loam soils to sand and sandy loam textures with the majority of the soils of the region (∼68%) classifying as sandy loam and loamy sand (Table S1). Most fine-textured soils are found in carbonate-rich areas, with coarser-textured soils found in more recently deglaciated soils and overlying crystalline bedrock. Examining soil particle size in the context of carbon content is important because (1) soils dominated by silt-sized particles are more susceptible to frost heave [Penner, 1968] and thus redistribution of carbon and (2) finer-textured soils have the ability to retain almost twice as much carbon as coarse textured soils [Bouwman, 1990; Bird et al., 2002]. Although some researchers have found correlations between clay content and percentage of soil organic carbon in global soils [Paul, 1984; Jobbágy and Jackson, 2000], SOC shows little to no relationship to soil texture in the Thule region.

3.2. Bulk Density

[20] Bulk densities of the <2 mm soil fraction range from 0.50 to 1.80 g/cm3. Peat bulk density (when available) calculated by soil moisture tins ranges from 0.15 to 0.24 g/cm3. These are similar to peat densities found by Gorham [1991]. Whole soil bulk density values range from 0.16 g/cm3 (gravel-free, organic-rich) to 1.95 g/cm3, with a few as high as 2.2 g/cm3. Densities generally increase with depth because of compaction and lower carbon content. These bulk density values are similar to those found in previous studies of Arctic soils (Table S2).

3.3. Organic Carbon

[21] Organic carbon content in individual soil horizons ranges from 0.10% in C horizons to 44.8% in peat soils (Table S1). The total carbon content for horizons summed on a pit basis ranges from 0.5 kg C/m2 to 26.4 kg C/m2 (Table 1). The average carbon content for all soil pits is 9.0 kg C/m2 with a standard deviation of 6.8 kg C/m2. This average is higher than average estimates of Miller et al. [1983] (7.2 kg/m2 (weighted average) for polar desert, polar semidesert, and wet sedge), but is slightly lower than averages observed in Arctic Canada by Tarnocai [2000] (10.8 kg/m2) and in northeast Greenland by Elberling et al. [2004] (11.0 kg/m2).

[22] Recent circumpolar data for Turbels (a general classification for the common cryoturbated soils in the Thule region) [Tarnocai et al., 2009] (0–100 cm mean soil organic carbon content of 32.2 kg/m2) is much higher than estimates presented here. More specifically, comparable data for the Thule peninsula given by Tarnocai et al. [2009] (Table 3) show an average SOC content of 25.5 kg/m2 and an SOC mass for the Thule peninsula of 18.4 Tg (compared to 9 kg/m2 and 4.3 Tg presented here). Although data from Tarnocai et al. [2009] are based on the upper 100 cm, and data presented here are generally less than 100 cm in depth, the magnitude of the difference cannot be fully explained by pit depth alone. Data from our regionally detailed study suggest that the broad-range extrapolations of Tarnocai et al. [2009] may be overestimating SOC in the high Arctic by not making distinctions between high and low Arctic Turbel data.

Table 3. Estimates of Circumpolar Arctic SOC Storage by Various Authorsa
StudyDescriptionArea (km2)C Density (kg/m2)Amount
  • a

    Note how each study varies in its terminology, classification system, and areal coverage.

  • b

    Weighted average based on published densities and areas.

  • c

    Soil depth 0–100 cm.

Post et al. [1982]Tundra (including dry, moist, wet, rain)8.8 × 10621.8191.8 Pg
Boreal forest, boreal desert, and tundra21.9 × 10617.8b394.1 Pg
Miller et al. [1983]Total Arctic (excluding biomass) (polar desert, semidesert, wet sedge, tussock, low shrub, tall shrub)5.71 × 1069.6b55 Pg
 Total Arctic (excluding biomass) (polar desert, semidesert, and wet sedge only)3.3 × 1067.2b24.2 Pg
 Polar desert0.80 × 1060.090.02 Pg
 Polar semidesert1.50 × 1067.210.08 Pg
Bliss and Matveyeva [1992]Total Arctic (low + high Arctic) (to 25 cm depth)5.60 × 1065.4b26.33 Pg
 Low Arctic (tall shrub, low shrub, tussock, mire, semidesert)3.616 × 1067.6b23.98 Pg
 High Arctic (polar desert, semidesert, mire)1.98 × 1061.2b2.35 Pg
Gilmanov and Oechel [1995]North American tundra4.12 × 10622.291.3 Pg
Tarnocai [2000]Canadian Arctic (high Arctic, mid-Arctic, and low Arctic)1,509,00028.543 Pg
 Canadian high Arctic315,00019.66.19 Pg
 Canadian mid-Arctic399,00034.313.7 Pg
 Canadian low Arctic795,00029.123.17 Pg
Tarnocai et al. [2003]Northern hemisphere Cryosols (0–100 cm)7.77 × 10636.6268 Pg
 Northern hemisphere Cryosols (0–30 cm)7.77 × 10615.3119 Pg
Ping et al. [2008]CAVM Lowland active layer220 × 10329.26.4 Pg
 CAVM Upland active layer2067 × 10325.753.2 Pg
This studyHigh Arctic (polar desert, semidesert, mire) (extrapolated data)1.98 × 1066.1b12.06 Pg
Thule specific data    
Tarnocai et al. [2009]USDA Histel, Orthel, and Turbel soil classifications722.825.5c18.4 Tgc
This studyCAVM NDVI classes and FAO soil classifications640.79.04.3 Tg

[23] SOC ranges from 3.2 kg/m2 in NDVI class 1 to 25.9 kg/m2 in NDVI class 8 (Table 2). The categorization of SOC by NDVI reveals a slightly nonlinear positive relationship between NDVI class and the log of SOC content (Figure 2b). The variability of carbon (as shown by standard deviation) appears highest in NDVI classes 2–4, with lesser variability observed in classes 1 and 6 (Table 2). Using an assessment of surface disturbance as an indicator of cryoturbation, it is evident that classes 2–4 are more active, and thus higher variability in SOC content is expected due to cryoturbation redistributing SOC throughout the profile (Figure S2). The degree of surface disturbance was visually assessed on-site using an index (1–5) to describe how disturbed the surface appeared at each of the 55 soil pit sites. An index of 1 indicates little to no evidence of patterned ground, and an index of 5 indicates the most well-developed patterned ground features. Although only single samples were collected for classes 5, 7, and 8, it is suggested that the variability of these high NDVI classes is much smaller than that observed in lower NDVI classes. This is attributed to less active cryoturbation resulting from the insulating effects of peat accumulations and fibrous materials that are less susceptible to cryoturbation (Figure S2).

Figure 2.

The relationship of NDVI and SOC in northwest Greenland. Shown are (a) the mean and one standard deviation of log SOC per NDVI class and (b) individual July NDVI values plotted against the log of SOC. Tabular data below the graphs show the statistical results of linear regression models of July NDVI values and the log of SOC.

[24] As previously shown, the variability of SOC in each NDVI class was often large. In order to estimate total SOC of the region, the mean SOC value was used for each NDVI class. Mean carbon content per NDVI class was then extrapolated across the areal coverage of each class to estimate total SOC storage for the Thule peninsula (Table 2). The total SOC on the Thule peninsula ranges from 0.002 Tg (1012 g) for NDVI class 8 (smallest area) to 2.4 Tg for NDVI class 2 (largest area). Summing the extrapolated SOC content of the eight classes results in approximately 4.3 Tg of SOC in the active (seasonally thawing) layer of the Thule region.

[25] The NDVI map of the Thule region (Figure 1) reveals that SOC is unequally distributed across the landscape. The majority of the region is covered by NDVI classes 1 and 2 (60%), with classes 5–8 covering only 0.03% of the region (Table 2). Organic-rich peat soils of Manîserqat (Green Valley) and adjacent valleys (Figure S1) at the extreme southern end of the peninsula are dark brown to black, nutrient-rich, peats that result from the fertilization effects of large colonies of Dovkies or Little Auks (Alle alle) that nest in the talus slopes. This highly localized, lush, green vegetation is typical of high Arctic regions that are fertilized by bird colonies, thus highlighting the nutrient limited nature of this region [Eurola and Hakala, 1977; Odasz, 1994, 1996]. The valley is dominated by Alopecurus alpinus and Luzula confusa and is often grazed by small herds of musk oxen. Soils of this valley have the highest organic carbon content and the highest NDVI values for the region (class 6, 7, and 8); however, they cover an extremely small area (Table 2). Similar ecosystems can be found in Svalbard and Franz Josef [Eurola and Hakala, 1977; Odasz, 1994, 1996], but these are also highly localized systems of the high Arctic. Although low NDVI classes have a lower mean SOC content, the large areal coverage generates a significant storage of SOC. This is important to note because sparsely vegetated surfaces are common across much of the circumpolar high Arctic and cover approximately 847,000 km2 [Bliss and Matveyeva, 1992].

3.4. SOC Correlation With Surface Parameters

[26] Nine surface parameters were analyzed for possible correlations to subsurface SOC: NDVI, total vegetation cover, vascular vegetation cover, aboveground biomass, slope, aspect, pit depth, elevation, and surface disturbance. Total and vascular vegetation cover differ in that total vegetation cover includes mosses, lichens, and cryptogamic crusts, while vascular cover is limited only to vascular plants. These nine surface parameters were chosen as they are believed to influence or reflect carbon accumulation in the soils. Soil texture was previously shown to not have a strong relationship with SOC. Geologic substrate was also not included because of insufficient sample sizes and lack of relationship with SOC.

[27] Linear regression analyses show that vascular vegetation cover and NDVI had the strongest correlations to SOC storage in the Thule region (R2 = 0.519 and R2 = 0.385, respectively) (Figure 2b and Table S3). Total vegetation cover and elevation each explain approximately a third of the variance observed, while deepest depth, slope, aboveground biomass, surface disturbance, and aspect parameters explain very little (Table S3).

[28] To analyze whether the addition of nonvegetation parameters could improve the prediction of SOC (as indicated by linear regressions of vascular plant cover), the residuals of the log SOC–vascular vegetation regression (Table S3) were plotted against the five nonvegetation parameters (slope, aspect, elevation, depth, and surface disturbance). None of the nonvegetation variables provided much additional explanation in the variation in log SOC, but the highest explanatory variable, aspect (R2 = 0.038), was included in a multiple regression of log SOC with vascular vegetation, and slight improvements were observed (R2 = 0.556) (Table S4). It is likely that large improvements did not occur with the addition of the aspect parameter because it (as well as elevation) is biophysically associated with vegetation and as thus is cross-correlated. All nine variables were not regressed together, as the other vegetation parameters (e.g., biomass) are also cross-correlated.

[29] While vascular plant cover has the strongest correlation to SOC in northwest Greenland, it is a variable that cannot be directly sensed from a satellite. Although NDVI is less strongly correlated with SOC than vascular plant cover, it can be directly sensed from satellite imagery. As ASTER-derived NDVI calculated at peak biomass (July) explains 71% of the variations in vascular vegetation cover (Figure 3), NDVI is viewed as an appropriate proxy for vascular vegetation. Therefore, as vascular vegetation cover best predicts SOC, and NDVI is a good proxy for vascular vegetation, it suggested that NDVI represents the best remotely sensed variable for SOC prediction in northwest Greenland. In light of this, the residuals of the log SOC–NDVI regression (Table S3) were plotted against the five nonvegetation parameters (slope, aspect, pit depth, elevation, or surface disturbance) to explore additional improvement in SOC prediction. As with the residuals of the vascular vegetation regression, none of the additional variables provided much additional explanation in the variation in SOC. Aspect was again the highest additional explanatory variable (R2 = 0.077), and it was included in a multiple regression of log SOC with NDVI. As with vascular vegetation, slight improvements were observed (R2 = 0.436) (Table S4). It appears that both vegetation variables (NDVI or vascular vegetation) explain the bulk SOC variability and that the addition of aspect (also influential in vegetation growth) provides additional model improvement.

Figure 3.

Scatterplot showing the relationship between ASTER-derived measurements of NDVI with ground measurements of percent vascular vegetation cover. Results here indicate that NDVI is a good proxy for vascular vegetation cover.

[30] Discovering a statistically significant relationship between NDVI and belowground SOC was somewhat unanticipated. Previous Arctic remote sensing research has shown strong NDVI relationships to aboveground biomass [Shippert et al., 1995; Walker et al., 2003], leaf area index (LAI) [Reidel et al., 2005], time since deglaciation [Walker et al., 2003], and temperature [Walker et al., 2003], but no strong relationships to SOC have been reported. Even using a ratio of a “mixed” pixel (which incorporates external (nonvegetation) parameters) to calculate NDVI, approximately 38% of the subsurface SOC variation is explained. Although NDVI does not form a perfect correlation with belowground SOC, it does provide a step toward the use of remotely sensing to predict SOC storage.

[31] The use of NDVI as a parameter for the estimation of high Arctic SOC storage is important for two reasons. The first is that NDVI is used in other Arctic studies as a parameter in biophysical properties (e.g., LAI, CO2 flux, biomass estimates, etc.) [e.g., Shippert et al., 1995; Walker et al., 2003; Reidel et al., 2005; Steltzer and Welker, 2006]. The second reason is that NDVI is easily extracted from remotely sensed imagery (under cloud-free conditions), thus providing a possible means by which satellite imaging could help predict SOC over larger regions. As the CAVM NDVI classes have been used in this study, the potential exists for extrapolating SOC-NDVI prediction models from the Thule region to larger areas of the Arctic. In order to achieve this, it is suggested that the SOC-NDVI relationships be tested in other regions of the high Arctic.

3.5. Comparison to Previous High Arctic Carbon Studies

[32] As mentioned above, previous research has shown that estimates of low Arctic SOC have been substantially underestimated by not accounting for SOC at depth. Existing studies provide limited carbon data from soils in the high Arctic [Tedrow and Douglas, 1964; Tedrow, 1966; Ugolini, 1966; Tedrow, 1970; Walker and Peters, 1977; Bliss and Svoboda, 1984; Bliss et al., 1984; Mann et al., 1986; Kimble et al., 1993; Muc et al., 1994; Christensen et al., 1995; Levesque, 1996; Broll et al., 1999; Lev and King, 1999; Tarnocai, 2000; Stutter and Billett, 2003; Bekku et al., 2004; Elberling et al., 2004] (Table S5), but most only provide carbon content (%) for select soil horizons, and total SOC (kg/m2) usually is not determined. These studies use various sampling methods, soils depths, and biological boundaries, and only one [Kimble et al., 1993] estimates carbon to the top of the permafrost table for each pit, making it difficult to draw conclusions about previous estimates of high Arctic carbon storage. Circumpolar estimates of SOC (from soil cores, remote sensing, etc.) also use nonstandardized methods and terminology. While the high and low Arctic are traditional biogeographic boundaries, few researchers publish circumpolar carbon data defined by these regions. This creates additional difficulties when comparing regional carbon data using nonspecific terms such as “tundra,” which cross the low/high Arctic boundary (Table 3).

[33] Equal comparisons were able to be made, however, with one of the most cited estimates of high Arctic SOC. Bliss and Matveyeva [1992] published SOC values compiled from soil pits dug across the high Arctic to a depth of 20–25 cm. This study focused primarily on the ecological and physiological aspects of the high Arctic, but also includes accompanying SOC data. To compare current SOC results from the Thule region with those of Bliss and Matveyeva, soil pits were reclassified into the same vegetation categories used by Bliss and Matveyeva [1992].

[34] Bliss and Matveyeva [1992] categorize the high Arctic into three general vegetation types: polar desert, polar semidesert, and mire. Polar deserts are sparsely vegetated landscapes composed of 0%–10% vascular plant cover and are dominated by Draba, Saxifraga, and Dryas plant species [Bliss and Matveyeva, 1992; N. Grulke, personal communication, 2006]. Polar semideserts are moderately vegetated landscapes with 5%–25% vascular vegetation, >50% total vegetation cover (including mosses and lichens), and are dominated by Dryas, Salix, Saxifraga, Draba, Minuartia, and Papaver plant species [Bliss et al., 1973; Bliss and Matveyeva, 1992; N. Grulke, personal communication, 2006]. Mires are distinct communities in that they often have standing water during summer months and are dominated by Carex, Eriophrum, Alopecurus, Dupontia, and Juncus plant species [Bliss and Matveyeva, 1992]. Soils in this community often contain peat of varying thickness (0.20–5+ m) and have a very shallow active layer (15–25 cm). These class descriptions are general guidelines for site classification and do not constitute a complete and precise continuum of plant communities (i.e., there is a degree of overlap among all the classes) (Figure S3).

[35] Reclassification of Thule soil pit locations into the three categories used by Bliss and Matveyeva [1992] was done using vegetation transect data collected at the majority of sites in the Thule study area (Table S5 and Figure S3). To reclassify locations, the percentage of plant cover was first examined. For locations where percent plant cover overlapped classes, or where little cover data were provided (e.g., 25%–50%), the plant species composition was used to classify sites. Where transect data are not available, visual estimates of cover and field observations were used to classify sites.

[36] Upon reclassification of the Thule region soil pits (Table 1), results show that carbon in the polar semidesert (n = 31) ranges from 1.5 to 19.5 kg/m2, with an average of 7.9 kg/m2. Polar desert sites (n = 9) have a much narrower range of 0.5 to 5.3 kg/m2, with an average of 2.2 kg/m2. Carbon in mire sites (n = 15) range from 2.7 to 26.4 kg/m2, with an average of 15.4 kg/m2 (Table 1). The large range in this category stems from the inclusion of peat-rich soils of Green Valley with less organic-rich mires from the rest of the region.

[37] As a first-order estimate of high Arctic SOC comparisons to the results of Bliss and Matveyeva [1992], the mean SOC value from each of the three categories was multiplied by the high Arctic area of each vegetation type, as defined by Bliss and Matveyeva [1992] (Table 4). By using estimates of SOC to the top of the permafrost table in the Thule region and extrapolating to the circumpolar high Arctic region, this study estimates that the polar semidesert regions of the high Arctic contain 7.9 Pg of SOC, polar deserts contain 2.2 Pg, and mires contain 2.0 Pg. This estimate is, again, based on a well-distributed sampling of 55 soil pits dug only to the base of the active layer, or as limited by standing water or impenetrable gravels. Just as the Thule region is quite variable in SOC distribution, it is thought that the larger region of the high Arctic is equally as varied. As previously noted, SOC estimates would be higher if carbon in the permafrost was included.

Table 4. Comparison of Circumpolar High Arctic SOC Data From Bliss and Matveyeva [1992] to SOC Estimates of the Thule Region Extrapolated to the Circumpolar High Arctic Regiona
Vegetation TypeBliss and Matveyeva [1992]This Study
Area (106 km2)O.M. (kg/m2)SOC (Pg)Organic Carbon (kg/m2)SOC (Pg)Average Organic Carbon Above 25 cm (kg/m2)Average SOC Above 25 cm (Pg)
  • a

    Results from this study compare (1) the total SOC estimates for the circumpolar high Arctic by multiplying the high Arctic area by the estimated carbon content of each vegetation type (“SOC” columns), (2) the average carbon content in the upper 25 cm of soil (for equal methodology comparisons to Bliss and Matveyeva [1992]), and (3) the average total SOC storage for the upper 25 cm (high Arctic area multiplied by average carbon content in upper 25 cm). O.M., organic matter.

  • b

    Used area values from Bliss and Matveyeva [1992].

Polar semidesert1.0052.190.9787.837.87b5.255.28b
Polar desert0.8470.0460.0172.552.16b0.930.79b

[38] The results of the Bliss and Matveyeva [1992] comparison suggest that previous values of SOC have been dramatically underestimated, particularly in the less vegetated areas. Over 127 times more SOC is found in polar desert regions, approximately 8 times more carbon in polar semidesert regions, and approximately 1.5 times more in mires (Table 4). That the estimates of SOC in mire communities are only slightly underestimated by Bliss and Matveyeva [1992] is logical as permafrost tables are shallow (because of the insulating effects of thick vegetation) and relatively stable with respect to cryoturbation; thus the majority of SOC would already be accounted for in the upper 25 cm. Unlike the low Arctic where thick peat deposits occur [Gorham, 1991], the high Arctic tends to have limited peat deposits except in areas near bird nesting sites [Eurola and Hakala, 1977)].

[39] When the sum of the SOC in the three communities is compared, results from this study (12.1 Pg) are overall more than 5 times greater than the sum put forth by Bliss and Matveyeva [1992] (2.4 Pg) (Table 4). It is noteworthy that although the greatest underestimate of SOC is in the polar desert class, the largest pool of SOC is found in the polar semidesert class (Table 4). Although some of this difference may be attributed to the greater depths to which pits were dug in the Thule region, it is apparent that SOC brought to depth by cryoturbation is unaccounted for in previous estimates by Bliss and Matveyeva [1992]. In order to make more equal comparisons of SOC data, Table 4 also shows average SOC data above 25 cm for each of the three communities in the Thule region. Here it is illustrated that, on average, polar deserts, polar semideserts, and mires still contain 46.5, 5.4, and 1.2 times greater amount of SOC, respectively. Proposed reasons for differences between these estimates include (1) a more accurate accounting of cryoturbated horizon dimensions in the Thule region and (2) more accurate estimates of bulk density which account for gravel content.

[40] Although not exclusively comparable, results infer that other Arctic carbon studies may also be underestimating circumpolar SOC storage (Table 3). Miller et al. [1983] estimate that SOC storage in polar semidesert regions is somewhat similar to data presented here; however, polar desert regions of the high and low Arctic have 108 times less SOC that this study predicts for approximately the same area. Extrapolated high Arctic SOC values presented here (12.1 Pg) are 50% of the low Arctic SOC (24.0 Pg) estimated by Bliss and Matveyeva [1992] (Tables 3 and 5), indicating that their estimates of low Arctic SOC may be underestimated as well. It is apparent that larger stores of SOC for the high and low Arctic are present that could be affected by climate warming.

Table 5. Distribution of SOC Above and Below 25 cm in the Thule Region Based on Two Methods of Surface Characterizationa
Surface CharacterAverage Percent Above 25 cmAverage Percent Below 25 cmStandard DeviationMaximum Percent Above 25 cmMinimum Percent Above 25 cmMaximum Percent Below 25 cmMinimum Percent Below 25 cmAverage Pit Depth (cm)
  • a

    (1) Bliss and Matveyeva [1992] vegetation categories of polar desert, polar semidesert, and mire and (2) NDVI classes modeled after those used by the CAVM map [CAVM Team, 2003].

  • b

    Both methods of surface characterization highlight the fact that sparsely vegetated regions of northwest Greenland have more SOC stored below 25 cm than above.

Polar desert4357209020801068
Polar semidesert6832171004159075
NDVI class 14555289020801062
NDVI class 25546127335652675
NDVI class 37228211004159074
NDVI class 4742624962377441
NDVI class 5544644
NDVI class 6100020
NDVI class 7100036
NDVI class 8851530

[41] The percentage of SOC in the upper 25 cm was also calculated using methods previously explained. In the polar semidesert sites, the majority (68%) of the SOC was found in the upper 25 cm (Table 5), while polar desert sites had an average of 57.0% below 25 cm. Although pit depth might be a suspect parameter in this estimation, polar desert pits average shallower depths (68 cm) than polar semidesert pits (75 cm). Soils of the mire sites have shallow depths to permafrost, are relatively stable with respect to cryoturbation, and therefore contain an average of 84% of the SOC above 25 cm.

[42] A linear regression of NDVI and percent SOC below 25 cm shows a statistically significant relationship with a negative slope; NDVI explains nearly one third of the variance (R2 = 0.306) (Figure 4b). Relatively larger percentages of SOC below 25 cm occur in lower NDVI classes, than in most of the higher NDVI classes. Nearly half of the SOC in classes 1 and 2 (55 and 46%, respectively) is stored below 25 cm in comparison to the higher NDVI classes which store significantly less than 50% below 25 cm. One exception to this relationship is class 5, in which only one pit was excavated, and therefore the comparison is not accurate. As pit depth is an important variable in the estimation of SOC, it is important to compare pits of equal depth. Thus model improvement is observed when the deepest pit depth variable is incorporated into the regression model (R2 = 0.449).

Figure 4.

Relationships between NDVI and the amount of SOC found below 25 cm in northwest Greenland. Shown are the (a) mean and one standard deviation of the percentage of SOC found below 25 cm for eight NDVI classes and (b) individual July NDVI values plotted against corresponding percentages of SOC below 25 cm. Tabular data below the graphs show the statistical results of linear regression models of July NDVI values and percent SOC below 25 cm (Figure 4b) and July NDVI coupled with pit depth and SOC below 25 cm (graph not shown).

[43] In the broader context of carbon cycling and climate change, the estimated storage of high Arctic SOC presented here (12.1 Pg) represents 1.6% (or 5.6 ppmv) of the CO2 content of the atmosphere (750 Pg) [IPCC, 2001]. Although this high Arctic pool seems relatively small, a change of 5.6 ppmv (or 12 Pg) carbon due to climatic warming would be equivalent to 3.7 times the annual increase of CO2 from anthropogenic sources (1.5 ppmv) [IPCC, 2001]. It is unclear at what rate this carbon might be released [Davidson and Janssens, 2006], or whether warming in the high Arctic would lead to an increase in carbon storage due to accelerated plant growth as suggested by recent studies of “greening” in Alaska [Sturm et al., 2001; Tape et al., 2006]. Improved estimates of high Arctic SOC are valuable for estimating potential Arctic terrestrial feedbacks to climate change.

4. Conclusions

[44] This study examined 55 soil pits to assess the soil organic carbon content above the permafrost table in the Thule region, northwest Greenland soils. SOC ranged from 0.5 kg C/m2 to 26.4 kg C/m2 across eight classes of NDVI as defined by the CAVM project. It was estimated that 4.3 Tg of carbon as SOC is stored to the top of the permafrost table in the 800 km2 study region.

[45] Compared to previous estimates of high Arctic SOC using the classification system of Bliss and Matveyeva [1992], this study suggests that 127 times more carbon is stored in polar desert regions, 8 times more in polar semidesert regions, and 1.5 times more in mires. This is due, primarily, to not accounting for organic carbon mixed to depth by cryoturbation. This study therefore finds that 57% of the total SOC is stored below 25 cm in sparsely vegetated polar desert landscapes, while 31% and 16% is found below 25 cm in more densely vegetated polar semideserts and mires, respectively. The 12 Pg of SOC estimated for the circumpolar high Arctic represents only 1.6% of the total carbon in the atmosphere; however, additional carbon is likely stored in the permafrost, and thus this estimate is likely to be somewhat low.

[46] Although lower NDVI classes (less vegetated) do not store greater amounts of total SOC than higher classes, more SOC is stored at depths deeper than normally sampled (25 cm) in past studies. Soil studies not accounting for carbon at depth are likely underestimating SOC, particularly for sparsely vegetated regions (NDVI class 1 and 2). This finding is particularly important considering that nearly 60% of the Thule region landscape, and much of the high Arctic, is NDVI class 1 and 2 (Table 2). Soils in higher NDVI classes contain more total SOC than soils in lower NDVI classes; however, the areal extent of these classes is less and decreases even further at more northern latitudes. Future studies should confirm the importance of sparsely vegetated areas on total SOC storage and measure SOC that is frozen in the permafrost.


[47] Funding for this research was provided by the National Science Foundation (OPP-0221606); University of Washington Department of Earth and Space Sciences; Geological Society of America (7691-04); Association of American Geographers (AAG) Dissertation Research Award; AAG–Geomorphology Specialty Group Dissertation Research Award; and Arctic Institute of North America Grant-in-Aid Award. Logistics and field support provided by VECO Polar Resources and the U.S. Air Force. We thank J. Welker for informative discussions and project leadership; D. Sabol for satellite imagery assistance; B. Hagedorn, K. Burnham, D. Garnas, H. Kokorowski, M. O'Neal, and A. Vigna for field assistance; and P. Sampson and R. Schmick for statistical assistance.