Geophysical imaging and thermal modeling of subsurface morphology and thaw evolution of discontinuous permafrost

Authors


Abstract

[1] Despite our current understanding of permafrost thaw in subarctic regions in response to rising air temperatures, little is known about the subsurface geometry and distribution of discontinuous permafrost bodies in peat-covered, wetland-dominated terrains and their responses to rising temperature. Using electrical resistivity tomography, ground-penetrating radar profiling, and thermal-conduction modeling, we show how the land cover distributions influence thawing of discontinuous permafrost at a study site in the Northwest Territories, Canada. Permafrost bodies in this region occur under forested peat plateaus and have thicknesses of 5–13 m. Our geophysical data reveal different stages of thaw resulting from disturbances within the active layer: from widening and deepening of differential thaw features under small frost-table depressions to complete thaw of permafrost under an isolated bog. By using two-dimensional geometric constraints derived from our geophysics profiles and meteorological data, we model seasonal and interannual changes to permafrost distribution in response to contemporary climatic conditions and changes in land cover. Modeling results show that in this environment (1) differences in land cover have a strong influence on subsurface thermal gradients such that lateral thaw dominates over vertical thaw and (2) in accordance with field observations, thaw-induced subsidence and flooding at the lateral margins of peat plateaus represents a positive feedback that leads to enhanced warming along the margins of peat plateaus and subsequent lateral heat conduction. Based on our analysis, we suggest that subsurface energy transfer processes (and feedbacks) at scales of 1–100 m have a strong influence on overall permafrost degradation rates at much larger scales.

1 Introduction

[2] Climate warming is predicted to have increasingly severe impacts in high-latitude regions, where the consequent thawing of permafrost could substantially alter hydrological processes. Because permafrost within discontinuous permafrost zones has subsurface temperatures close to 0°C, it is expected that small rises in air temperature will produce rapid degradation of thin permafrost in the southern fringe of these zones, resulting in substantial changes to water storage and runoff pathways [Rouse, 2000]. Recent evidence for increases in stream discharge, particularly during winter months, in rivers draining discontinuous permafrost regions in Eurasia [Smith et al., 2007], Alaska [Brabets and Walvoord, 2009], and Canada [St Jacques and Sauchyn, 2009] suggests that long-term degradation of discontinuous permafrost may already be having an impact on hydrological processes in circumpolar river basins.

[3] Aside from increasing air temperatures, other factors may increase or decrease rates of permafrost degradation. These include thermal properties of local soil, shading due to vegetation cover, amounts of precipitation, thermal insulation provided by snow cover, topographic effects, and energy transfer processes between frozen ground and surrounding surface water and groundwater [Jorgenson et al., 2010; Iijima et al., 2010]. These processes exhibit strong positive and negative feedbacks, which occur at a scale much smaller than typical grid-cell sizes of current large-scale permafrost thaw models (e.g., 50 km) [Delisle, 2007; Zhang et al., 2008a; Avis et al., 2011]. In order to improve current energy transfer models of permafrost degradation in circumpolar regions, it is necessary to understand the subgrid-scale feedback processes.

[4] In low-relief wetland regions like those found in the discontinuous permafrost zone of Canada's Northwest Territories (Figure 1a), permafrost bodies can act as impermeable barriers that control the flow of surface water and groundwater [Hayashi et al., 2004; Quinton et al., 2009]. Wright et al. [2009] demonstrated how groundwater that flows above the relatively impermeable frost table will tend to converge in depressions, increasing soil moisture in these areas. Because moist organic peat has a higher thermal conductivity than dry peat, enhanced vertical energy transfer between the ground surface and the underlying permafrost will create an area of preferential permafrost thaw [Hayashi et al., 2007]. Groundwater may continue to pool in these depressions maintaining enhanced rates of vertical energy transfer and permafrost thaw, resulting in positive feedback between energy and water transfer [Wright et al., 2009; Jorgenson et al., 2010]. These depressions may not completely refreeze during the winter months resulting in semipermanent areas of thaw that may continue to grow [Jorgenson et al., 2010].

Figure 1.

(a) Permafrost distribution in Canada (source: Natural Resources Canada, http://atlas.nrcan.gc.ca). Continuous permafrost includes glacier. Discontinuous permafrost is divided into extensive (50–90%) and sporadic (10–50%) regions. Dotted line defines the Hay River Lowland ecoregion. (b) Location of Scotty Creek research basin (gray filled area) within the lower Liard River valley near Fort Simpson, Northwest Territories. Dashed lines demarcate the major drainage basins. Square defines the location of the geophysics profiles shown in Figure 2. (c) Mean annual air temperature (MAAT) recorded in Fort Simpson.

[5] Despite our current understanding of shallow thaw processes and aside from one-dimensional borehole observations [Smith and Riseborough, 2010], very little is known about how thawing takes place at greater depths. For example, we do not know if shallow groundwater depression storage leads to complete thawing of the underlying permafrost layer; nor do we know how groundwater-induced thaw processes influence the subsurface morphology of the permafrost bodies. Some insight into the impact of future climate warming on subgrid subsurface thaw processes can be gained through small-scale thermal modeling. For example, Smith and Riseborough [2010] used two-dimensional thermal models tied to borehole temperature logs to examine how disturbances along pipeline right-of-way will impact subsurface permafrost conditions under future climate warming scenarios. In order to produce consistent predictions with such models, it is critical to understand the existing form and distribution of permafrost in the subsurface.

[6] Geophysical methods provide effective tools for detecting and charactering permafrost. Prior to the widespread application of two-dimensional tomographic techniques, electrical and electromagnetic soundings were used to investigate the thickness and distribution of subsurface permafrost in high-latitude regions [e.g., Hoekstra, 1978; King and Seppälä, 1987]. In the continuous permafrost zone of Arctic Canada, Todd and Dallimore [1998] used time-domain electromagnetic loop soundings to map the lateral and vertical distribution of permafrost along profiles within the Mackenzie River delta. Multiple geophysical methods were used by Yoshikawa et al. [2006] to investigate the internal composition and structure of pingos in Alaska. In addition, Fortier et al. [2008] have used electrical resistivity logging and electrical resistivity tomography (ERT) methods to map the internal resistivity structure of lithalsas in Nunavik within the discontinuous permafrost zone of Canada. Application of the ERT methods started in alpine permafrost and rock glaciers settings [e.g., Hauck and Vonder Mühll, 2003; Marescot et al., 2003] and has been extended to subarctic [e.g., Lewkowicz et al., 2011] and high-elevation plateaus [e.g., You et al., 2013]. As yet we have little information on the subsurface shape and form of degrading discontinuous permafrost in wetland regions.

[7] For the purpose of imaging the subsurface morphology of thawing discontinuous permafrost, we acquired ERT and ground-penetrating radar (GPR) profiles over different peat plateaus within the Scotty Creek research basin (SCRB), a small low-relief drainage basin that is close to Fort Simpson, Northwest Territories (Figure 1). By using interpretations of the subsurface dimensions of the permafrost and subsurface geology as input to a thermal modeling routine, we investigate processes that are likely to influence future thawing of permafrost. The main objectives of this study are to (1) estimate the shape and dimensions of permafrost bodies beneath the peat plateaus, (2) use interpretations of the geophysical images to examine how feedback processes may control lateral variations in active-layer thicknesses, (3) construct two-dimensional thermal-conduction models using constraints derived from the geophysical data and observed soil-temperature data to investigate whether lateral energy transfer dominates vertical energy transfer in the wetland environment, and (4) use the models to examine what effect changes in land cover will have on rates of permafrost thaw.

2 Study Site

[8] In order to examine the effects of hydrological processes in discontinuous permafrost zones, long-term field studies were initiated within the SCRB, which lies within the Hay River Lowland ecoregion [Quinton et al., 2009] (Figure 1). This region has a dry continental climate with short summers and long cold winters. At the Fort Simpson meteorological station located 50 km north of SCRB (Environment Canada, http://www.climate.weatheroffice.gc.ca), mean annual air temperature (MAAT) gradually increased during 1900–1970 at an average rate of 0.015°C yr−1 (Figure 1c). The rate of increase became greater during 1970–2010 at 0.048°C yr−1 (Figure 1c). Total elevation change within the 152 km2 area of the SCRB is less than 60 m. The basin comprises a mosaic of three main land cover types: peat plateaus, ombrotrophic flat bogs, and channel fens (Figures 2 and 3). The peat plateaus are underlain by ice-rich permafrost and are elevated 0.5 to 2.0 m from the local water table in flat bogs and channel fens due to the ice formation in the peat and frost-susceptible fine-grained material underneath the peat. As a result, the peat plateaus provide stable platforms which support the growth of shrubs and trees. Small, generally circular flat bogs are found within peat plateaus (Figure 2a), whereas larger interconnected flat bogs occupy areas surrounding the plateaus and connect to the channel fens. In the following, bogs refer to flat bogs that are lower in elevation than peat plateaus, unless otherwise indicated. Bogs are characterized by the dominance of Sphagnam spp. and the water with low pH (4.6–5.5) and electrical conductivity (EC) (30–50 μS cm−1). In contrast, channel fens support vascular vegetation such as Carex spp. and have the water with neutral pH and higher EC (> 60 μS cm−1) [Hayashi et al., 2004]. The channel fens form broad (>50 m wide) interconnected channels containing slowly flowing open water and play an important role in conveying water through the drainage basin. A thick layer (3–4 m) of peat covers most of the basin, and hand-auger core samples extracted within the SCRB indicate that this peat layer is underlain by a widespread silty-clay deposit of low hydraulic conductivity (10−10 to 10−9 m s−1) as determined by hydraulic response tests [Freeze and Cherry, 1979, p.340] conducted on drive-point piezometers.

Figure 2.

Elevation maps of the study area showing location of ERT profiles (numbered solid lines) presented in Figures 4-6. (a) Northern part, and (b) southern part of the study area. Dotted lines show locations of additional ERT profiles not shown here. Peat plateaus have slightly higher elevation than surrounding bogs and fens. Circles denoted with “p,” “b,” and “g” indicate locations of weather stations and soil-temperature sensors (see text). Dashed line in Figure 2b delineates depression coinciding with a winter road. Trapezoidal region is shown in Figure 3. Coordinates are referenced to NAD83 UTM Zone 10.

Figure 3.

Photograph of trapezoidal region in Figure 2a showing three different land cover types.

[9] By comparing high-resolution remote-sensing imagery acquired over the SCRB in 2000 and 2008, Quinton et al. [2009] showed that thawing of permafrost is mostly concentrated along the lateral margins of the peat plateaus. Further analysis of optical imagery by Chasmer et al. [2010] indicated that between 1947 and 2008, rates of permafrost area reduction within the SCRB were approximately 0.5% per year. High-resolution digital elevation maps show how the peat plateaus are pockmarked by circular peat bogs (Figure 2). In many places, a crescent-shaped plateau has formed where a portion of the permafrost perimeter surrounding a circular bog has thawed (e.g., c1 and c2 in Figure 2a). Although, prior to this study, permafrost thickness had not been measured within the SCRB, Smith and Riseborough [2010] reported thicknesses of 5–10 m from two boreholes located within 50 km of our site. The site is the location of ongoing hydrological studies and is equipped with a number of hydrometeorological monitoring stations (see Wright et al. [2008, 2009] for the details on instrumentation). In the present study, air temperature, relative humidity, wind speed, and radiation data from the weather station in an open area (circle b in Figure 2a) and precipitation data from another weather station (circle g in Figure 2b) are used to force an energy and mass transfer model (see section 4). Soil-temperature data from peat plateau (circle p in Figure 2a) and bog (circle b in Figure 2a) are used to assess the performance of the energy and mass transfer model.

3 Geophysics Profiles

3.1 Geophysics Acquisition and Processing

[10] Eight ERT and two collocated GPR profiles were acquired within the SCRB during late August 2009 and 2010, at times where the frost-table depths were close to their maxima. Here we present three representative profiles that image subsurface permafrost morphology from different settings within the basin (Figure 2). The ERT tomograms for other profiles are shown in the supporting information accompanying this paper. Since ice exhibits high electrical resistivities relative to liquid water, the ERT method is an ideal geophysical imaging technique for distinguishing subsurface permafrost from unfrozen ground. An Iris Syscal Pro 10-channel system was used to acquire the ERT data. By using an electrode spacing of 3 m and dipole-dipole array geometries with maximum a-spacings of five electrodes (i.e., 15 m) and dipole separations (n-spacings) of up to six times the a-spacings, we were able to image resistivity structure to depths greater than 10 m. Along some profiles, higher-resolution images of shallower regions were produced by reducing the electrode spacing to 1 m. To characterize measurement error distributions, additional reciprocal measurements were recorded for some current-potential electrode pairs [cf. Slater et al., 2000]. From an established linear relationship between measured resistance and measured repeatability, we calculated an error model that was applied to all resistance measurements; this error model was used to compute data mismatch weights in our tomographic inversions.

[11] Cross sections of resistivity were computed for each data set collected along the profiles using the RES2DINV inversion program, an iterative-based smoothness-constrained least squares inversion algorithm [Loke and Dahlin, 2002]. Because we expected subsurface structures to have relatively sharp boundaries, we used the robust (L1-norm) model inversion constraint. For each inverted profile, convergence was reached after no more than five iterations and root-mean-square errors were in the range of 2 to 4%. To evaluate the resolution of our inverted tomograms, we conducted depth of investigation (DOI) tests [e.g., Marescot et al., 2003]. In the final model tomograms used for interpretation, we incorporate the DOI resolution results by plotting only model cells with normalized DOI values greater than 0.2.

[12] Because electromagnetic waves are reflected at boundaries between materials with contrasting electromagnetic properties (i.e., relative permittivity and electrical conductivity), GPR profiling provides a means to image and delineate the shallow surfaces of permafrost bodies and shallow subsurface stratigraphic boundaries. A Malå ProEx GPR system with 100 MHz Rough Terrain Antennas was used to record GPR data along a portion of profiles 1 (Figure 2a) and 7 (data not shown). Processing steps included the following: (1) application of a standard dewow filter, (2) a time-zero correction, (3) trace scaling, (4) band-pass frequency filtering, and (5) a conversion from time to depth. Frost-table depths were measured beneath the peat plateaus using a 1.3 m long graduated steel probe and hand auger, providing ground truth on the topography of the upper surface of permafrost. By collecting core samples using a hand auger within the unfrozen bogs, we gained additional information on the composition and distribution of the shallow sediments.

3.2 Geophysics Results

[13] The ERT tomogram from profile 1 shows a discrete zone with resistivities in excess of 1000 Ωm that we interpret as a permafrost body, which extends laterally under the full width of the 1–2 m elevated peat plateau (Figure 4a; location shown in Figure 2a). Based on our frost-table depth measurements (circles in Figure 4a) and the sharp transition from high to low electrical resistivities at its base, we estimate the thickness of this permafrost body to be 9 ± 3 m. Hand-auger samples measured along profile 1 from within the bog on the eastern side of the plateau indicate that the layer of shallow water-saturated organic peat extends to depths of approximately 3 m and is underlain by silty clay (crosses in Figure 4). The boundary between the layers coincides with a prominent reflection observed on the processed GPR cross section recorded along profile 1 over the bog and a laterally continuous change in resistivities observed on the ERT tomogram (Figure 4). A similar change to lower resistivities occurs at the same depth beneath the fen on the western side of the plateau, indicating that the depth to clay is relatively uniform in this area.

Figure 4.

(a) Profile 1 ERT tomogram with electrode spacing of 3 m recorded over a peninsula of permafrost-cored peat plateau surrounded by channel fen and bog. (b) High-resolution (1 m electrode spacing) ERT tomogram of the rectangular region shown in Figure 4a plotted as a transparent overlay on coincident 100 MHz GPR data. Circles = measured frost-table depths; crosses = peat/clay boundary determined from hand-auger boreholes.

[14] The high-resolution ERT tomogram from the central section of profile 1 shows that the upper surface of the permafrost body is defined by a sharp change from resistivities less than 500 Ωm within the active layer to resistivities in excess of 2000 Ωm in the underlying permafrost (Figure 4b). The location of this boundary correlates with frost-table depths measured at 1 m intervals across the plateau and the depths of a prominent frost-table reflection observed on the corresponding GPR cross section, which terminates abruptly at the eastern edge of the plateau. Three local depressions are observed within the frost-table topography measured by the frost probe (open circles in Figure 4b). In general, resistivities within the active layer of thawed peat are lower above the frost-table depressions (<200 Ωm) than above the frost-table ridges (200–1000 Ωm); the lowest resistivities are observed within the deepest depression (>1.3 m), indicating an anomalously thick region of unfrozen water-saturated peat (D in Figure 4b).

[15] The ERT tomogram from profile 2 reveals resistivity structure under and around the location of an approximately 15 m wide circular bog confined within a peat plateau of slightly higher elevation (Figure 5; location shown in Figure 2a). Two discrete high-resistivity (> 5000 Ωm) bodies that surround the bog are interpreted as permafrost and have thicknesses of between 5 and 8 m. Resistivities of less than 500 Ωm are equivalent to those found within the shallow active layer and indicate that the permafrost beneath the bog has completely thawed. A zone of higher resistivities of approximately 1000 Ωm occurs at depths below 11 m in the central region of profile 2. This could be caused by the presence of ice in the silty clay, but it is unlikely for an isolated ice lens to persist below the thawed layer. Therefore, the higher-resistivity region most likely indicates a lens of higher-resistivity sand within the silty clay, which is a common feature in the glacial deposits in the region [Aylsworth et al., 2000].

Figure 5.

Profile 2 ERT tomogram recorded over an isolated bog that is confined within permafrost (location shown in Figure 2a). Circles = measured frost-table depths; crosses = peat/clay boundary determined from hand-auger boreholes. The area of moderately high resistivity in the central part is interpreted to be a sand lens.

[16] Profile 3 crossed a peat plateau transected by a linear bog that was artificially created by removing trees to create a winter road (Figure 2b). The ERT tomograms for profile 3 show that to the north and south of the road, high resistivities indicative of permafrost (> 5000 Ωm) extend to depths of approximately 10 m (Figure 6). Analysis of aerial photos and historical records indicate that the road was constructed in 1969 [Williams, 2012]. Although the manual frost-table measurements made with a hand auger show that the linear bog is underlain by permafrost (in contrast to the naturally occurring bogs along profiles 1 and 2), the much lower resistivities of less than 2500 Ωm underneath the bog suggest that the permafrost here has a significantly higher fraction of unfrozen water content than at its margins (Figure 6a). The higher-resolution tomogram shown in Figure 6b reveals that this low-resistivity region extends laterally within the permafrost underneath the southern side of the linear bog.

Figure 6.

(a) Profile 3 ERT tomogram with electrode spacing of 3 m recorded over a linear bog coinciding with a winter road (location shown in Figure 2b). Rectangular region is shown in Figure 6b. (b) High-resolution (1 m electrode spacing) ERT tomogram of the rectangular region shown in Figure 6a. Circles = measured frost-table depths.

4 Thermal Modeling

4.1 Model Overview

[17] In order to investigate the processes that control how the discontinuous permafrost bodies respond to seasonal and interannual ground-temperature changes, we run two-dimensional simulations using commercially available finite-element thermal-conduction modeling software Temp/W (http://www.geo-slope.com). Because of its proximity to soil-temperature and meteorological sensors, we elected to base a model on interpretations from geophysics profile 1 (Figures 2a and 4a). Temp/W cannot simulate the complex surface processes such as radiation balance, snow accumulation and melt, and the water table dynamics, all of which affect the energy input to the ground surface [Jorgenson et al., 2010]. Therefore, we use a one-dimensional energy and water transfer model, Northern Ecosystem Soil Temperature (NEST) [Zhang et al., 2003], to simulate the soil temperature in the vadose zone. The output from the NEST model is used to specify the top-boundary condition for the Temp/W simulation. This approach is similar to that of Smith and Riseborough [2010], who used a one-dimensional thermal-conduction model to set up the boundary condition for Temp/W simulation. It should be noted that the NEST model does not account for the advective energy transfer by soil water, even though it simulates the flow and storage of water of soil water and adjusts the soil thermal properties accordingly. This is one limitation of the present modeling approach.

[18] Two sets of simulations are conducted using the NEST-Temp/W model: the first to examine the thawing of a lens-shaped peat plateau surrounded by wetlands (Figure 4a) and the second to examine the effects of linear disturbance (Figure 6a). In the first simulation, two NEST models representing the peat plateau and the bog are used to provide boundary conditions for Temp/W (Figure 7). In the second simulation, one NEST model is used until the tree removal along the road, and a second NEST model is added afterward (Figure 7). Detailed model methodology is described using the first simulation in the following, and the modification of methodology for the second set is described later.

Figure 7.

Schematic diagram showing how the outputs from the one-dimensional NEST model are used to provide the surface boundary condition for the two-dimensional Temp/W model.

4.2 Model Geometry and Boundary Conditions

[19] Noting that the permafrost body below the peat plateau has roughly symmetrical shape (Figure 4a) and that soil-temperature profile data are available for the peat plateau and the bog, we choose to simulate a half of the peat plateau on the bog side to increase the computational efficiency. The width of the peat plateau at the time of the geophysics surveys in 2009 was approximately 25 m, but an aerial photograph from 1970 indicates that at that time the width was roughly 60 m, and the distance between the study plateau and next plateau was approximately 60 m [Quinton et al., 2011]. Mean annual air temperature (MAAT) time series at Fort Simpson (Figure 1c) indicate that temperature was variable during 1900–1970 with a relatively small rate of increase, and a more rapid increase started around 1970. Therefore, we use a 60 m wide model domain with the plateau and bog occupying 30 m each (Figure 8). The relative fraction of peat plateau and bog (0.5:0.5) is consistent with the relative fraction of plateau and wetland for a larger region estimated from the 1970 aerial photograph [Quinton et al., 2011].

Figure 8.

Temp/W model domain showing the surface boundaries, materials, and finite-element mesh.

[20] Field measurements indicate that the peat plateau surface is raised approximately 0.9 m above the bog surface. Field observations from soil pits dug on the peat plateau indicate that the average position of the water table is 0.1–0.2 m above the frost table (i.e., the water table is perched on the ice-saturated impervious peat), and the depth to the frost table varies from 0 in the early spring to 0.5–1.2 m in the late summer, which indicates the active-layer depth. The water table in the bog fluctuates between 0 and 0.2 m below the ground surface. To reflect these near-surface conditions in the top boundary of the Temp/W model, we run the NEST model separately for the peat plateau and the bog using the on-site meteorological data to specify the top-boundary condition. The peat plateau simulation is run with tree canopy, which reduces shortwave radiation input, and with the water table fluctuating between 0 and 0.9 m below the surface. The bog simulation is run with no canopy and with the water table fluctuating between 0 and 0.2 m below the surface. Further details on the NEST model parameterization, boundary condition, and initialization are described in Appendix A. Soil temperatures simulated by the NEST model are compared with measured soil temperatures in the peat plateau and the bog to check the consistency of the NEST model with the field condition.

[21] Our intent is to use Temp/W to simulate the processes below the vadose zone. Therefore, the Temp/W model domain is set up as a rectangle with a flat top surface; the top-boundary condition is provided by the soil-temperature output of the NEST model at 1.05 m below the ground surface for the peat plateau and at 0.15 m below the ground surface for the bog. This configuration allows the complex surface processes to be simulated by NEST and the two-dimensional subsurface processes to be simulated by Temp/W. Since the Temp/W model simulates the saturated part of the subsurface environment, the model has a simplified structure with only two types of porous media: saturated peat extending to a depth of 3 m and saturated silty clay below. The model mesh size varied from 0.5 m near the surface to 2 m near the bottom (Figure 8).

[22] The model domain thickness is taken to be 50 m, and the bottom boundary is treated as a constant flux boundary with a geothermal flux of 0.08 W m−2 estimated for the region [Blackwell and Richards, 2004]. The sides of the model are treated as no-flux (i.e., symmetry) boundaries, noting that the model domain represents a half of the peat plateau.

4.3 Soil Thermal Properties

[23] In our Temp/W modeling we use the thermal properties of saturated peat and silty clay that are listed in Table 1. Volumetric heat capacities for each material are estimated by summing the heat capacities of their individual constituents (minerals, organic matter, water, and ice) [Hillel 1998] multiplied by the volume fractions.

Table 1. Thermal Properties of Saturated Peat and Silty Clay Used in the Simulations
DescriptionPeatSilty Clay
Total water content θtot (m3/m3)0.80.4
Coefficient in equation (1), a (°Cb)0.220.15
Coefficient in equation (1), b, dimensionless−0.15−0.4
Unfrozen thermal conductivity (W m−1 K−1)0.481.07
Frozen thermal conductivity (W m−1 K−1)1.21.75
Unfrozen heat capacity (J m−3 K−1)3.86 × 1062.88 × 106
Frozen heat capacity (J m−3 K−1)2.48 × 1062.19 × 106

[24] Thermal conductivity versus temperature and unfrozen moisture content versus temperature functions for saturated peat and silty clay are shown in Figure 9. Piecewise functions relating unfrozen water content (θu) to temperature (T, °C) and total water content (θtot) are estimated using a simple power law [Lovell, 1957]:

display math(1)
Figure 9.

(a) Unfrozen water content and (b) thermal conductivity as functions of temperature for saturated peat and silty clay.

[25] Empirical constants a and b for each material are shown in Table 1. The empirical constants for saturated peat are taken from Zhang et al. [2008b], who fit observations from soil moisture sensors at our SCRB field site. To estimate the empirical constants for silty clay found at our site, we use unfrozen water content versus temperature data reported for silty clay in Tarnawski and Wagner [1993].

[26] We estimate thermal conductivity versus temperature functions for the three soils using a modified version of the Johansen interpolation method outlined in Balland and Arp [2005]. In this approach the thermal conductivity of the saturated frozen soil (Ksat) is given by

display math(2)

where Vpores and Vwater are volume fractions of the pore space and unfrozen water, respectively. Thermal conductivities for ice (Kice) and water (Kwater) are 2.2 and 0.57 W m−1 K−1, respectively [Hillel, 1998]. For silty clay, Ksolid (the thermal conductivity of the solid fraction of the soil) is 1.62 W m−1 K−1, whereas for the organic matter comprising the peat soils, Ksolid is 0.25 W m−1 K−1 [Hillel, 1998]. Equation (2) gives 0.48 W m−1 K−1 for unfrozen, saturated peat (i.e., Vpores = Vwater), which is consistent with the field-measured values reported by Hayashi et al. [2007], and gives 1.07 W m−1 K−1 for unfrozen, saturated clay, which is consistent with the values reported in the literature [e.g., Jury and Horton, 2004, p.182].

4.4 Model Initialization

[27] The main purpose of modeling is to simulate the thawing of permafrost during 2004–2010, where we have on-site meteorological data, using the Temp/W model with the NEST model providing the top-boundary condition. To initialize the NEST-Temp/W model for 2004–2010 simulation, we first run the NEST model from 1900 to 2004 using the archived meteorological data from Fort Simpson (50 km north of the study site) to estimate the local meteorological data (Figure 7).

[28] In the next step, the Temp/W model is initialized for 1900 by running it in a steady state mode with the top-boundary temperature of −1.8°C for the peat plateau and +1.4°C for the bog (Figure 7). These temperature values are chosen by trial and error so that the transient Temp/W simulation (see below) results in the permafrost thickness in 2004 similar to the observed values of 9 m. Once the Temp/W model is initialized, it is run in a transient mode on a monthly time step from 1900 to 1970 using the monthly average values of soil-temperature output from the NEST model as the top-boundary condition (Figure 7). Within each time step, the Temp/W model iterates the solution typically 10 to 20 times until the convergence is achieved. Fort Simpson mean annual temperature data indicate that noticeable warming started after 1970 (Figure 1c), resulting in the reduction of the peat plateau area [Quinton et al. 2011]. To capture this process in the model, the simulation is stopped every 5 years and if the zone below a node representing the peat plateau becomes permafrost-free, the node is changed to bog, and the next 5 year simulation is run. This mode of simulation is carried out from September 1970 to September 2004. This sets up the model for the main simulation of permafrost thawing during 2004–2010.

4.5 Thermal Modeling Results

[29] Soil temperatures simulated by the NEST model are compared to measured soil-temperature data for the 2004–2010 period (Figures 10a and 10b). Temperatures recorded by the 0.15 m sensor at the peat plateau and the average of temperatures from the 0.1 and 0.2 m sensors at the bog are compared to the NEST-simulated temperatures for the 0.1–0.2 m layer. The bog temperature sensors were installed in August 2008, and the plateau temperature sensors had a data gap during the winter of 2008–2009. Observed and simulated temperatures matched reasonably well with a root-mean-square error (RMSE) of 1.4°C and mean bias error (MBE) of −0.1°C for the plateau, and RMSE of 2.7°C and MBE of −1.1°C for the bog.

Figure 10.

(a) Observed and NEST-simulated daily average soil temperature at a depth of 0.15 m for the peat plateau. (b) The same for the bog. (c) Comparison of observed 2009 and 2010 soil temperatures at depths of 0.1 m within the bog and the peat plateau.

[30] The Temp/W model is run with daily time steps for a 6 year period from 1 September 2004 to 31 August 2010. The simulation is stopped on 31 August every year to adjust the plateau-bog boundary in the manner described above. Figure 11 shows the simulated extent of the permafrost body delineated by the 0°C isotherm. The edge of the permafrost retreats by 3.5 m during the 6 year period or the reduction of permafrost width by 7 m. In contrast, the thickness of the permafrost in the center of the permafrost body (x = 0 m in Figure 11) decreases by 1.4 m, suggesting that the lateral thawing of permafrost is much faster than the vertical thawing.

Figure 11.

Temp/W modeling results showing the extent of the frozen region on 15 September of each year.

[31] To examine the effects of canopy removal along the winter road crossed by ERT profile 3 (Figure 2b), a new Temp/W model is set up with a 60 m domain width similar to Figure 8 but with a different configuration for the top boundary. In this simulation, the entire top boundary is set up as peat plateau. The region between x = 0 and 5 m represents a half width of the winter road, where the canopy is removed in the NEST model to increase the radiation input. The removal in the model occurs in 1970 to approximate the timing of the road construction in 1969.

[32] Two different model scenarios are simulated to evaluate how the water trapped above the frost table may influence subsurface thaw. In the “wet” scenario of the NEST model, the water table is allowed to fluctuate between 0 and 0.2 m below the ground surface, similar to the NEST model simulation for the bog described above, which simulates poorly drained conditions. In the “dry” scenario of the NEST model, the water table is allowed to fluctuate between 0 and 0.9 m to simulate well-drained conditions. These boundary conditions are applied to the winter road section of the model domain, while the same peat plateau boundary condition (described in section 4.2) is used in the rest of the model domain.

[33] The Temp/W simulation with the dry winter road condition shows a deepening of the active layer under the winter road and reduction of permafrost depth starting between 1980 and 2000, but the active layer has not reached the peat-clay boundary by 2010 (Figure 12a). The simulation with the wet road condition also shows the deepening of the active layer starting between 1980 and 2000 (Figure 12b). The active layer has reached below the peat-clay boundary in this simulation, but the permafrost is still present under the winter road (Figure 12b).

Figure 12.

Temp/W modeling results for the hypothetical canopy removal in 1970. Dashed lines show the extent of frozen region on 15 September of each year. (a) Dry scenario. (b) Wet scenario.

5 Discussion

5.1 Observations of Subsurface Morphology

[34] From our analysis of all the ERT tomograms recorded within the SCRB, we find that the discontinuous bodies of permafrost are no thicker than about 13 m (Figures 4-6 and supporting information). In some places, laterally continuous permafrost may be as thin as 5 m (e.g., Figure 5). In general, the lateral margins of the permafrost bodies have steep sides. The high-resolution ERT tomogram for profile 1 reveals that at the margins of permafrost the lateral transition from low-resistivity, low ice content to high-resistivity, high ice content conditions is relatively sharp. Similar observations were made in the subarctic regions of Canada by Fortier et al. [2008] and Lewkowicz et al. [2011] using a larger (2 m or more) electrode spacing and lower resolution, and the high-resolution data in this study clearly demonstrate the sharpness of the boundary. Across the eastern edge of the plateau along profile 1, a prominent frost-table reflection observed on GPR data terminates (at x = 80 m on Figure 4); farther to the east, no reflections from the frost table are evident within 5 m below ground surface, indicating that the thickness of the frozen region tapers out over a lateral distance of just a few decimeters. Steep lateral margins are also observed for the edges of permafrost underlying the small bog along profile 2. These abrupt transitions are in agreement with our manual frost-probe measurements made near the edges of the permafrost bodies, which show that the depths to the upper surface of frozen ground can rapidly increase over just a few decimeters.

[35] Based on observations from our geophysical images and from results of our thermal modeling, we attribute the sharp lateral permafrost boundaries to the efficient lateral transfer of energy to the permafrost from the surrounding unfrozen soils that underlie the bogs and fens that receive a higher amount of vertical energy inputs. Because the peats in the bogs and fens and underlying clays are warmer (during summer and winter) than permafrost under the peat plateau, recent permafrost thaw appears to have been driven by both lateral and vertical energy transfer processes. Consequently, one-dimensional permafrost thaw models that assume only vertical energy transfer cannot capture the multidimensional pattern of energy transfer of the discontinuous permafrost characterized by relatively thin patches of permafrost. For example, the one-dimensional NEST model for the peat plateau simulated the active-layer deepening of 0.43 m between 2004 and 2010 but a very small decrease (0.04 m) in the depth to the bottom of the permafrost during the same period. This is in contrast to the results of two-dimensional simulation (Figure 11), suggesting that one-dimensional permafrost models may systematically underestimate thaw rates in discontinuous permafrost regions.

5.2 The Role of Depression Groundwater Storage on Rates of Thaw

[36] Based on the positive-feedback thaw process between energy and water transfer described by Wright et al. [2009], we interpret different stages of permafrost thaw beneath depressions that have formed within the upper surface of permafrost. Low electrical resistivities observed within a frost-table depression beneath profile 1 indicate a localized region of moist peat (D in Figure 4b). We expect that as this depression and the volume of saturated soil grow, overlying trees and shrubs will drown and a local bog will develop. With the loss of shading provided by the overhead canopy and the continued impoundment of water that flows through the active layer [cf. Quinton et al., 2009; Wright et al., 2009], this area will be subject to enhanced thawing during successive summer seasons until the permafrost layer has been completely penetrated, as is the case beneath the isolated bog along profile 2 (Figure 5). Once complete breakthrough has occurred beneath a bog, a new hydrologic pathway is created such that recharge may take place through the bog to the underlying groundwater aquifer. With continued lateral thawing, the isolated bog will continue to grow and will eventually become connected to surrounding open bogs and fens, allowing more efficient flow of shallow groundwater to the regional drainage network (e.g., as has occurred for features c1 and c2 in Figure 2a).

[37] In contrast to the shallow depression and the isolated bog observed on profiles 1 and 2, respectively, the linear bog along profile 3 is not completely surrounded by permafrost, potentially allowing shallow groundwater to flow through the active layer along the winter road into the channel fen to the west of profile 3 (Figure 2b). Consequently, because groundwater is not impounded, the depression in permafrost beneath the winter road may not be subject to the positive feedback process described by Wright et al. [2009], which may explain why the underlying permafrost has only partially thawed (Figure 6). Alternatively, because the linear disturbance only occurred in 1969, the amount of thawing time may have been insufficient to cause complete breakthrough of permafrost. The results of thermal modeling (Figure 12) suggest that the lateral drainage of groundwater probably contributed to a relatively slow thawing of the permafrost, but these also suggest that the permafrost will be broken through within a matter of a decade or two if the present warm condition is sustained. Running the simulations shown in Figure 12 beyond 2010 by repeating the 2004–2010 boundary condition several times shows that the permafrost is broken through by 2030 for the dry scenario and by 2020 for the wet scenario.

5.3 Influence of Land Cover Type on Shallow Soil Temperatures

[38] Markedly different shallow soil-temperature regimes within the bog and peat plateau are evident from the 2009–2010 observed daily average soil temperature at 0.1 m shown in Figure 10c. Although both data sets show broadly similar temperature trends over the spring and summer months, during winter months, from mid-October to late April, temperatures in the peat plateau were substantially lower than they were at the same depth in the bog. For example, the minimum value of daily average soil temperature at 0.1 m depth in the peat plateau was −7.5°C, whereas the minimum value in the bog was only −1.1°C (Figure 10c). Average air temperature during October–April was −11.3°C at the peat plateau and −11.9°C at the bog, indicating that atmospheric temperatures were similar above the different land cover types over the same time period. Furthermore, despite the different land cover types, the average (and standard deviation) of snow depths measured on a 60 m transect was 0.43 (±0.10) m for the peat plateau and 0.50 (±0.03) m for the bog on 23 March 2010. Similar snow depth measurements on 19 March 2009 indicated 0.97 (±0.09) m for the peat plateau and 0.82 (±0.07) m for the bog. This suggests that there were relatively small differences in winter snowpack thicknesses between the bog and peat plateau, which may have influenced the transfer of thermal energy from the atmosphere to the shallow soils.

[39] The primary factor controlling the temperature difference is the presence or absence of permafrost. Thawing of the active layer in the peat plateau provides the constant temperature (=0°C) at the thawing front throughout the summer, resulting in a much lower temperature at 0.1 m depth in the plateau compared to the bog (Figure 10c). In winter, the soil temperature below the seasonal freezing front in the bog remains positive and provides upward heat flux, thereby maintaining a higher temperature in the frozen layer compared to that in the plateau (Figure 10c). Other possible factors contributing to the temperature difference include the drier condition of near-surface peat in the plateau providing more efficient thermal insulation [Hayashi et al., 2007].

5.4 Influence of Changing Land Cover on Thaw Rates and Limitation of Thermal Model

[40] Because the peat plateaus within the SCRB are only slightly elevated from the local water table, relatively minor subsidence resulting from interannual losses of shallow permafrost can lead to long-term flooding and concomitant changes to the land cover. With the drowning of the peat plateaus, the underlying permafrost loses not only shade provided by the forest canopy but also a layer of insulation resulting from the unsaturated peat cover. This results in much more rapid lateral thawing of permafrost compared to vertical thawing [e.g., Thie, 1974; Payette et al., 2004; Camill, 2005]. Frost-table measurements along the transect located 20 m south of profile 2 showed that the width of the permafrost changed from 38 m in August 2004 [Quinton et al. 2011] to 26 m measured during our survey in August 2010, indicating 12 m reduction in 6 years. The Temp/W simulation gives a width reduction of 7 m (3.5 m on either side) during the same period (Figure 11). Part of the difference between the observed and simulated reduction of permafrost widths may be explained by the complex processes that are not captured in the present model. For example, a 1–2 m wide open water trough has been observed around the retreating edge of the peat plateau providing a local channel for the advection of relatively warm water. The lateral transfer of energy from the flowing water to the frozen peat under the plateau may contribute to the enhanced lateral thawing in addition to conduction.

[41] As the above suggests, the simple thermal simulations using the NEST and Temp/W models have many limitations. The lack of on-site meteorological data prior to 2004 introduces a significant degree of uncertainty in the model boundary conditions simulated by NEST. Also, preliminary sensitivity analyses using Temp/W with different model domain sizes and geometries indicated that the area ratio of peat plateau and wetlands has an influence on the energy balance over a large scale. Both of these can influence the thickness of permafrost, and the model has to be conditioned to match the simulated and observed permafrost thicknesses in 2004 by adjusting the top-boundary temperature in the steady state initialization stage (see section 4.4). In addition, the boundary conditions for peat plateau and bog are specified using the one-dimensional NEST model, which does not take into account the complex lateral processes at the edge of the peat plateau. Despite these limitations, the model is useful for the qualitative understanding of the role of different processes (e.g., vertical and lateral thawing) and for visualizing the transient responses of permafrost to warming (Figure 11) and canopy removal (Figure 12).

6 Conclusions

[42] We use a combination of geophysical imaging and thermal modeling to gain insight into subsurface thaw processes of discontinuous permafrost in a wetland environment. Key findings from our unique geophysical data set are as follows: (1) relatively thin (5–13 m thick) blocks of permafrost under peat plateaus are thawing both vertically and laterally, (2) the permafrost bodies have near-vertical edges as opposed to gently sloping edges, and (3) because of positive feedback, small (0.1 to 0.5 m wide) differential thaw features evolve into recognizable (1 to 5 m wide) depressions and then into isolated (10 to 50 m wide) bogs, which may eventually become connected to the larger-scale drainage network.

[43] By using interpretations from the geophysical images and observed meteorological data, we have constructed a realistic permafrost thaw model that comes reasonably close to simulating the observed rates of recent degradation. Relative to surrounding bogs and fens, peat plateaus have a number of physical characteristics that help slow the thawing of permafrost, including (1) a layer of unsaturated peat found on peat plateaus that has low thermal conductivity and provides thermal insulation to the underlying ground and (2) a forest canopy that shades the ground from solar radiation. In contrast, the shallow saturated soils found beneath surrounding bogs and fens have higher thermal conductivities and are exposed to more solar radiation. A combination of these effects allows seasonal temperature gradients to penetrate to greater depths within the soils under the bogs and fens and promotes thaw at the lateral margins of discontinuous permafrost bodies. As thawing at the lateral edges of the permafrost proceeds, the resulting subsidence allows shallow groundwater to inundate the surface at the margins of the peat plateaus. This process results in long-term saturation of the ground and loss of overlying forest canopy from the peat plateaus, producing a positive feedback that promotes enhanced vertical energy transfer at the edges of permafrost bodies. These subgrid-scale processes likely have a strong influence on overall thaw rates of sporadic and discontinuous permafrost at the scale of regional and global permafrost model grids.

Appendix A

[44] Northern Ecosystem Soil Temperature (NEST) is a one-dimensional water and energy transfer model, which computes the transfer of energy and water through the soil-vegetation-atmosphere continuum, specifically designed to simulate the evolution of permafrost in northern environments [Zhang et al., 2003]. It requires daily input data of air temperature, humidity, precipitation, wind speed, and incoming radiation. These data are all available from the on-site meteorological station during 2004–2010 (winter precipitation for August 2008 onward).

[45] NEST model assigns the soil hydraulic and thermal properties according to the soil type [Zhang et al., 2003]. Based on the geophysical data, the model soil profile is set up with peat for the top 3 m and silty clay for the rest of the model domain. Based on the field data reported by Quinton and Hayashi [2005], the model peat layer consists of highly porous upper layer (0–0.15 m depth) with a porosity of 0.92, dry bulk density of 134 kg m−3, and saturated hydraulic conductivity of 4.2 × 10−3 m s−1 and a denser lower layer (0.15–3 m) with a porosity 0.82, density 248 kg m−3, and conductivity 3.5 × 10−4 m s−1. The underlying silty clay has a porosity of 0.55, density of 1300 kg m−3, and conductivity of 1.7 × 10−6 s−1. The leaf area index of the model is adjusted to match the simulated radiation input to the ground surface (through canopy layer) with the observed radiation.

[46] The NEST model is run for the period of 1900–2010 on a daily time step. Since the on-site meteorological data are unavailable prior to 2004, they have to be estimated using the air temperature and precipitation data from the Fort Simpson meteorological station, located 50 km north of the study site (Environment Canada, http://www.climate.weatheroffice.gc.ca). Mean annual air temperature at Fort Simpson (−2.2°C) was very close to that of the study site (−2.4°C) during 2005–2010. Therefore, the Fort Simpson temperature data are used without adjustment during 1900–2004. Precipitation data at the study site have been corrected for the wind-induced catch deficiency using the procedure of Smith [2007], whereas Environment Canada reports precipitation data without wind correction (C. Smith, 2011, personal communication). Precipitation data at the study site during 2008–2010 were 1.2 times greater on average than the precipitation data at Fort Simpson. Therefore, Fort Simpson data are multiplied by 1.2 to estimate precipitation at the study site. There are numerous data gaps in the Fort Simpson precipitation data prior to 1963. Therefore, daily precipitation time series for each year during 1963–1992 is treated as a stochastic realization of annual precipitation time series and used to generate precipitation time series randomly during 1900–1962. Since no radiation, humidity, and wind speed data are available prior to 2004, the daily average radiation during 2004–2010 is used as the radiation time series. The bottom boundary of the NEST model is set up at a depth of 50 m, and a constant heat flux of 0.08 W m−2 is applied.

[47] The model is initialized by repeating a single-year run using the 1900 data until the permafrost thickness reaches a steady value (spin-up period in Figure 7). After this initialization, the model is run with the 1900–2010 data.

Acknowledgments

[48] We thank Yu Zhang for providing NEST model and useful advice and Bill Quinton for valuable discussion and many years of collaboration. We also thank Chris Hopkinson and Laura Chasmer for digital elevation data; Water Survey of Canada and Liidlii Kue First Nation for logistical support; Allan Bonnetrouge, Tyler Veness, George Sutherland, and Kate Forbes for fieldwork; Matt Wilkinson for assistance with map figures; and Natural Sciences and Engineering Research Council and the International Polar Year for funding. We are grateful for the constructive comments by the reviewers and the Associate Editor; in particular, the comments by Richard Fortier resulted in substantial improvement of thermal modeling results.

Ancillary