Subsurface Porewater Flow Accelerates Talik Development Under the Alaska Highway, Yukon: A Prelude to Road Collapse and Final Permafrost Thaw?

Abstract

The presence of taliks (perennially unfrozen zones in permafrost areas) adversely affects the thermal stability of infrastructure in cold regions, including roads. The role of heat advection on talik development and feedback on permafrost degradation has not been quantified methodically in this context. We incorporate a surface energy balance model into a coupled groundwater flow and energy transport numerical model (SUTRA-ice). The model, calibrated with long-term observations (1997–2018 on the Alaska Highway), is used to investigate and quantify the role of heat advection on talik initiation and development under a road embankment. Over the 25-year simulation period, the new model is driven by reconstructed meteorological data and has a good agreement with near surface soil temperatures. The model successfully reproduces the increasing depth to the permafrost table (mean absolute error <0.2 m), and talik development. The results demonstrate that heat advection provides an additional energy source that expedites the rate of permafrost thaw and roughly doubles the rate of permafrost table deepening, compared to purely conductive thawing. Talik initially formed and grew over time under the combined effect of water flow, snow insulation, road construction and climate warming. Talik formation creates a new thermal state under the road embankment, resulting in acceleration of underlying permafrost degradation, due to the positive feedback of heat accumulation created by trapped unfrozen water. In a changing climate, mobile water flow will play a more important role in permafrost thaw and talik development under road embankments, and is likely to significantly increase maintenance costs and reduce the long-term stability of the infrastructure.

Key Points

• A fully coupled groundwater flow and energy transport model integrated with a surface energy budget model is developed and validated with field observations

• Talik formation is triggered by the combined effect of snow cover, mobile porewater flow, road construction, and atmospheric warming

• Connectivity of two isolated taliks located up-gradient and down-gradient of the road embankment led to accelerated permafrost thaw

Plain Language Summary

Formation of taliks (perennially unfrozen zones in permafrost areas) affects the thermal, hydrological, and biogeochemical processes at and below the land surface. The presence of taliks beneath or adjacent to constructed structures has been reported in many permafrost settings, but the processes controlling the initiation, formation, and extent of a talik under a road embankment remains largely unknown. In this study, a fully coupled groundwater flow and energy transport model integrated with a surface energy budget model is developed and validated with field observations from a Canadian sub-Arctic discontinuous permafrost zone. Our results show that positive feedback mechanisms related to mobile porewater flow through the taliks greatly increased permafrost degradation. Talik formation creates a new hydrothermal state under the road embankment and poses a serious threat to overlying infrastructure. Additionally, dispatching the snow from the road pavement onto the embankment slopes may initiate an irreversible permafrost thaw feedback through the creation of taliks. With a warming climate, subsurface water flow will have a more important role in permafrost thaw and talik development. The results from our research highlight the importance of adequate drainage systems below road embankments to preserve permafrost.

1 Introduction

Globally, permafrost is thawing due to climate change (Biskaborn et al., 2019; Huang et al., 2017), particularly in the Arctic (McCrystall et al., 2021; Pithan & Mauritsen, 2014). Amplified Arctic warming has increased permafrost temperature by ∼2°–5°C since the 1980s (Cai et al., 2021; Cohen et al., 2014). The thawing of permafrost substantially reduces the bearing capacity of the soil (Doré et al., 2016; Schneider von Deimling et al., 2021), leading to potentially catastrophic situations for structures and roads (Hjort et al., 2018) as well as increased maintenance costs and reduced lifespans (Melvin et al., 2016; Reimchen et al., 2009). Under the RCP8.5 climate scenarios, projected Arctic warming will significantly increase infrastructure costs to nearly $70 trillion USD in the Arctic by the end of the 21st century (Yumashev et al., 2019), and $5.5 billion USD in Alaska from 2015 to 2019 (Melvin et al., 2016). However, the effect of mobile water flow on subsurface thermal processes has largely been overlooked in assessing the vulnerability of northern transportation infrastructure to permafrost thaw (McKenzie et al., 2021).

In the Arctic, roads and highways are a local cause of permafrost thaw due to their alteration of the subsurface thermal regime. Disturbance from roads is a result of the road's construction, and the alteration of the land surface energy budget (SEB) from the presence of the road and road maintenance/snow clearing operations (Chen et al., 2020). The net result of a road is an increased downward heat flux leading to permafrost warming at a higher rate than in natural conditions (Schneider von Deimling et al., 2021; Wu et al., 2007). When more heat accumulates in the subsurface and the heat loss in winter is insufficient to freeze back the active layer to the permafrost table, a supra-permafrost talik or perennially unfrozen zone is created and it will enlarge with time (Connon et al., 2018; Nelson et al., 2001).

In addition to the increasing air temperature (AT) due to climate change, many local factors affect talik initiation and development, including thick snow deposition (O'Neill & Burn, 2016; Jafarov et al., 2018), fire (Zipper et al., 2018), surface water dynamics (Gao et al., 2019; Roux et al., 2017), surface water infiltration (i.e., snowmelt and rainfall infiltration) (Chen et al., 2020; Westermann et al., 2011), and subsurface water flow (Gao et al., 2019; Lamontagne-Hallé et al., 2018; Rowland et al., 2011; Yoshikawa & Hinzman, 2003). These surface and subsurface thermal processes increase the amount of net annual heat flux in the ground and the underlying permafrost, causing permafrost warming, additional permafrost thaw, and even triggering talik formation.

For northern infrastructure, construction practices often change or damage vegetation and replace the natural surficial cover by engineered materials such as gravel, sand, and/or asphalt (Doré et al., 2016). Due to the high thermal conductivity and low albedo of construction materials, greater amounts of solar radiation and energy are transferred downward into the underlying subsurface, which may result in talik development in the permafrost supporting the infrastructure (Chen et al., 2021; Wu et al., 2015). The presence of taliks beneath or adjacent to constructed structures has been reported in many permafrost settings, including along the Qinghai-Tibet Plateau (QTP) Highway (Mu et al., 2020; Wu et al., 2015; Zhang et al., 2019), the Alaska Highway (Fortier et al., 2017; Sliger, 2016; Stephani, 2013) and the Dempster Highway in northwestern Canada (O'Neill & Burn, 2016).

Field observations and numerical simulations have demonstrated that the formation of taliks affects the thermal, hydrological, and biogeochemical processes at and below the land surface (Carey & Woo, 2002; Guan et al., 2010; Liljedahl et al., 2016; Walvoord et al., 2019). The formation of a talik enlarges subsurface water pathways (Lamontagne-Hallé et al., 2018) and changes the heat exchange processes between the land surface and the subsurface (Rowland et al., 2011; Roy-Leveillee & Burn, 2017), resulting in accelerated permafrost thaw which can be irreversible (Devoie et al., 2019).

The processes controlling the initiation, formation, and extent of a talik under a road embankment remains largely unquantified due to lack of long-term field observations. The feedback between talik formation and the thermal state of road embankments is not well studied. In particular, the role of heat advection on the magnitude and rate of talik development in the permafrost under a road embankment is uncertain. For example, in permeable materials heat advection driven by subsurface water flow is now recognized to affect permafrost degradation (e.g., Dagenais et al., 2020; Evans & Ge, 2017; Farquharson et al., 2022; Fortier et al., 2007; McKenzie et al., 2021), yet many modeling studies neglect this process and only consider heat conduction to predict and evaluate the thermal stability of road embankments (e.g., Fortier et al., 2011; O'Neill & Burn, 2016; Chen et al., 2022). Moreover, the complex thermal interactions between the atmosphere and a road embankment or the land surface are often simplified by using an n-factor approach (e.g., Karunaratne & Burn, 2004; Lunardini, 1978; Walvoord et al., 2019), surface offset (e.g., Chen et al., 2022; Lai et al., 2003), regression approach (e.g., Park et al., 2018; Zhang et al., 2019), or a surface heat transfer layer (e.g., Dagenais et al., 2020; McKenzie et al., 2007). These approaches simply interrelate the air and ground temperature without considering individual energy transfer processes at the surface, such as the change of snow regime and surface cover types. Further, the feedback of subsurface thermal processes on the ground energy fluxes is often neglected. Although one-dimensional heat conduction models driven with meteorological data were developed to partly couple the shallow surface energy component (Dumais & Doré, 2016; Hermansson, 2004), impacts of lateral and vertical subsurface heat transfer on the thermal regime of road embankment are not investigated and quantified. To our knowledge, no studies have evaluated the thermal regime of a road embankment in a changing climate through a cryohydrogeologic model fully coupled with a SEB routine.

Our study presents a novel approach to couple the individual energy processes at the top surface of a road embankment with hydrothermal processes in the subsurface layer. The general question is that: “what is the impact of heat advection induced by subsurface water flow on permafrost thaw and talik development under a road embankment?” The objectives of this study are (a) to integrate a SEB model into a groundwater flow and heat transport numerical model; (b) to simulate and reproduce 25 years (1993–2018) of thermal observations beneath a road embankment at Beaver Creek, Yukon, Canada; (c) to investigate the role of heat conduction and heat advection via subsurface porewater flow in accelerating permafrost thaw and talik development; and (d) to estimate the role of talik growth on the active layer dynamics. We hypothesize that (a) the combined effect of snow insulation, water flow, road construction, and climate warming contributes to rapid permafrost degradation and talik formation; (b) heat advection associated with water flow provides additional energy to expedite permafrost thaw; and (c) formation of a talik accelerates permafrost thaw and creates an irreversible thermal state under infrastructure unless mitigation measures are implemented.

2 Study Site

In 1993, an experimental road segment was constructed to monitor the thermal state of permafrost under a paved road embankment at Beaver Creek, southwestern Yukon, Canada (Figure 1a) (Doré et al., 2006; Reimchen et al., 2009). The Beaver Creek region is in the Klondike Plateau of the Western Boreal Cordillera. The study site is characterized by ice-rich degrading permafrost (de Grandpré et al., 2012; Sliger, 2016; Stephani et al., 2014). Mean annual air temperature (MAAT, 1981–2010) at Beaver Creek is −4.8°C, with a mean January temperature of −25.2°C and a mean July temperature of 14.1°C (Figure S1 in Supporting Information S1). The MAAT increased by 2.0°C from 1981 to 2010 (Environment and Climate Change Canada, 2018). Mean annual precipitation (1981–2010) is 417 mm, of which 117 mm (28%) falls as snow. Snowfall usually starts in September and snowmelt starts in April. The average thawing and freezing indices (i.e., yearly cumulative ATs above and below freezing) are 1537.7 thawing degree-days and 3321.8 freezing degree-days, respectively (1981–2010) (Environment and Climate Change Canada, 2018).

Figure 1

Location of the Beaver Creek-Road Experimental Section (BC-RES), south-western Yukon (62° 20′N, 140° 50′W). (a) Map of Yukon Territory showing the location of the BC-RES with a yellow star and roads with red lines (modified from Natural Resources Canada, 2014); (b) false-color satellite image of the experimental road test site near Beaver Creek (WorldView II image, taken 4 August 2010 at 21h30), preferential subsurface water flow is identified by the lighter tone vegetation aligned perpendicular to the slope; (c) photo of soil pit in the undisturbed natural ground (NG) made in early October 2017. The pit location is near the NG thermistor string; (d) photo of a thick (∼60 cm) snow cover deposited on the embankment slope, taken in March 2013; and (e) photo of asphalt-paved embankment center, taken in early October 2017 (Photographs by the authors).

The Beaver Creek - Road Experimental Section (BC-RES) is located at km 1865 of the Alaska Highway (Figure 1b). The permafrost temperature at 8 m deep below the undisturbed natural ground (NG) is about −2.2°C (Fortier et al., 2021; Stephani, 2013). The local vegetation is part of the taiga biome, with stunted spruce, shrubs, tussocks, and mosses characterizing the area (Sliger, 2016). Surface peat deposit ranges from 0.2 to 0.4 m in thickness overlying a sandy silt layer (Sliger, 2016). The active layer thickness is approximately 1 m (Fortier et al., 2021; Stephani, 2013). Subsurface water flow occurs preferentially in water tracks as observed on Figure 1b (lighter tone vegetation aligned perpendicular to the slope) (de Grandpré et al., 2012; Sliger, 2016).

The previous road embankment (Figure 1b) at the BC-RES was decommissioned and relocated in 1993 (Doré et al., 2006). Thermal monitoring started in April 1997 (yellow points, Figure 1b). The studied road embankment is a reference test section, which operates without any mitigation techniques, and has been experiencing severe damage due to the permafrost thaw (Figure S2 in Supporting Information S1). The maximum monitored settlement was 10 cm/month in summer (de Grandpré et al., 2012). The reference section was used to represent the standard construction practices for the Alaska Highway (Doré et al., 2006). During the reconstruction in 2008, there is no excavation for the reference section. In contrast to the reference section, the previous embankment materials were partly or fully removed for other test sections to implement the mitigation techniques (Doré et al., 2006; Reimchen et al., 2009), such as installation of air duct, replacing the embankment material with crushed-rock layer, etc. The removal of berms paralleling the embankment (Text S1 and Figure S2 in Supporting Information S1) revealed that the embankment material sunk into the NG due to permafrost degradation and isolated water-saturated taliks developed under the embankment berm (Reimchen et al., 2009; Stephani, 2013). There was significant water flow in the embankment materials after the excavation (Movie S1).

3 Methods

3.1 Field Observations and Measurements

To investigate talik initiation and formation beneath the unmitigated road embankment, we analyzed the 22-year (1997–2018 on Alaska Highway) record of ground temperatures at the BC-RES. Three land cover types were monitored, modeled, and compared: the undisturbed NG (seasonal snow-covered vegetated surface), embankment slope (seasonal snow-covered gravel/sand surface), and embankment center (snow-free asphalt-paved surface) (Figure 1b). Detailed information about the field survey and measurements are presented in Text S1 in Supporting Information S1.

Soil temperatures were continuously measured at multiple depths with a thermistor string starting in 1997, with a data gap from 1 January to 30 September 2008, due to road reconstruction (Fortier & Chen, 2022a, 2022b; Fortier et al., 2021). The thermistor strings were installed in boreholes located at the centerline of the road, midway along the embankment slope and in the NG (Text S1, Data Set S2, and S3 in Supporting Information S1). The thermal regime of the road embankment was estimated using kriging interpolation with linear semi-variogram model (Bargaoui & Chebbi, 2009) of the thermistor string data for each site, while the depth of the permafrost table was identified by locating the 0°C isotherm in mid-October. Mid-April is used to evaluate the talik development as this is generally when the coldest conditions of the underlying permafrost occur. The latent heat effect is identified by the slow rate of 0°C isotherm change over extended periods during the thawing soils (Outcalt et al., 1990). Snow insulation was reflected by the smaller amplitude of winter ground surface temperature (GST) variation, compared to winter AT variation (Chen et al., 2020). The thickness of the talik(s) was determined by the distance between the seasonal freezing depth and the permafrost table in mid-April when the soil is the coldest. Furthermore, the buffer effect of vegetation and organic layer was identified by the shorter thawing period and smaller amplitude of summer ground surface temperature variation, in comparison to the paved and bare surfaces (Chen et al., 2020; Sharif et al., 2019). For the results presented below, we specify the soil elevation relative to the vertical position of the NG surface (located below the embankment). The coordinates at the intersection between the embankment centerline and the horizontal surface of natural ground are (0, 0). A negative value represents the soil below the NG surface, while a positive value indicates material located in the embankment, which is above the NG surface.

The meteorological data set at the BC-RES is available from 2008 to 2018 (Fortier & Chen, 2020). The meteorological station was installed about 30 m from the road embankment at the BC-RES and provides hourly AT, rain precipitation, wind speed, and incoming solar radiation data (Data Set S1). The measurements of AT and wind speed started in October 2008, while rainfall and solar radiation started in October 2012. The missing meteorological data for the 1993–2008 period at the BC-RES was interpolated by empirical and linear fitting equations with available measurements from the BC-airport (8 km away from the BC-RES) and the BC-YGT (4 km from the BC-RES) (Environment and Climate Change Canada, 2018). Additionally, due to the lack of direct measurements at the three weather stations over the 1993–2013 period, the missing solar radiation data were estimated empirically (Text S2 and Table S1 in Supporting Information S1). Extraterrestrial radiation for each day of the year can be estimated from the solar constant, the solar declination, the latitude, and the time of the year by the empirical equation provided by Allen et al. (1998). A sensitivity analysis was used to quantify the uncertainties of the interpolated meteorological data on GSTs estimates. Specific information about interpolation methods is in Text S2 in Supporting Information S1.

3.2 Numerical Model and Modeling Approach

To simulate the interactions between surface energy processes, the road embankment subbase, and the subgrade, we fully integrated a SEB model (Chen et al., 2021) into the SUTRA groundwater flow and heat transfer model modified to include freeze-thaw processes (SUTRA-ice, McKenzie et al., 2007; Voss & Provost, 2010). The road embankment subbase and subgrade are a layer of fill material and the native material under an embankment, respectively. The modeling approach is illustrated in Figure S8 in Supporting Information S1.

The reconstructed meteorological data (Figures S3, S4, S5, and S6 in Supporting Information S1) from 1993 to 2018 and the simulated soil temperatures at depth (outputs of SUTRA-ice) drive the SEB model, while the outputs (either specified heat fluxes or surface temperatures) of the SEB model are passed to the surface grid nodes of SUTRA-ice model as the surface thermal boundary conditions (TBC) (Figure S8 in Supporting Information S1). The inputs of subsurface soil temperature in the SEB model are updated with new simulated soil temperatures provided by the SUTRA-ice model. Model assumptions are listed in Text S3 in Supporting Information S1.

The performance of the SUTRA-ice model integrated with the SEB model was evaluated by comparing the measured and simulated soil temperatures at depth. The validation period of the integrated model from 2014 to 2018 was selected due to the availability of meteorological measurements. To quantitatively evaluate the performance of the SUTRA-ice code combined with the SEB model to reproduce the thermal regime of a road embankment, we used the following statistical approaches: (a) The root mean square error (RMSE) was used to evaluate the uncertainty of the simulations; (b) The mean absolute error (MAE) was used to estimate the absolute difference between simulated and measured values; (c) The mean bias error (MBE) is the average error representing the systematic error of a prediction model to under or over prediction; (d) The standard deviation (SD) was used to quantify the extent of deviation from a set of data values; (e) The coefficient of determination (R²) was used as a measure of precision in predictions in comparison to the long term observations.

3.2.1 Surface Energy Balance (SEB) Model

The SEB model reproduces surface temperatures and surface energy components with varying snow depths and snowpack properties (Chen et al., 2021). The SEB model is driven by the net radiation (Q_net), sensible heat flux (Q_s), heat flux of vapor exchange from the surface (Q_v), conductive heat flux through the snow and ground (Q_c), rainfall heat (Q_p), and stored energy (Q₀). The SEB is expressed as:

(1)

The SEB model has been successfully developed, calibrated and validated with 10-year field observations at the same study site (Chen et al., 2021). A more detailed description of the SEB model equations and a sensitivity analysis showing the effects of snow depth, duration and timing can be found in Chen et al. (2021). The summary of the SEB model employed in this study was provided in Text S6 in Supporting Information S1. For the present study, the energy balance in the SEB model is fully coupled with the SUTRA-ice model and all parameters used in the SEB model are listed in Table S3 in Supporting Information S1.

3.2.2 Saturated-Unsaturated Groundwater Flow and Heat Transfer Model (SUTRA-Ice)

The SUTRA-ice model (McKenzie et al., 2007; Voss & Provost, 2010) simulates transient groundwater flow and heat transport in variably saturated conditions with freezing and thawing dynamics. It has been benchmarked against other coupled cryohydrogeologic models within the InterFrost consortium (Grenier et al., 2018; Rühaak et al., 2015) and tested against the analytical solutions (McKenzie et al., 2007). The SUTRA-ice model has successfully simulated the effect of climate change on thermal and hydrological processes (e.g., Evans et al., 2018; Kurylyk et al., 2014; Kurylyk et al., 2016; McKenzie & Voss, 2013) and the feedback of permafrost degradation on the temporal variations of groundwater discharge (e.g., Ge et al., 2011; Lamontagne-Hallé et al., 2018; Walvoord et al., 2019). The governing equations of SUTRA-ice are presented in Text S4 in Supporting Information S1. All parameters used in the SUTRA-ice model are listed in Table S4. Sensitivity analysis of mean annual soil temperature at depth to permeability factors is shown in Figure S7 in Supporting Information S1 and the related discussions are in Text S5 in Supporting Information S1.

3.2.3 Model Domain and Boundary Conditions

The computational domain consists of four sub-domains, including an asphalt layer, an embankment fill, a peat layer, and a sandy silt layer, as descripted in Figure 2. The geometry, based on the Beaver Creek site, corresponds to an embankment with an upper surface width of 12.0 m, a height of 5.0 m, and sides with a 2:1 slope. The top surface is a 10.0 m wide asphalt-paved driving surface with a 1.0 m wide bare gravel-sand embankment shoulder on either side. The porous, low-moisture embankment fill lies directly on the NG (Figure 2). The thickness of the peat layer is 0.4 m. The model domain is 237 m long, and the thickness of the NG ranges between 65.9 m on the upgradient edge of the model to 54.0 m on the downgradient edge. The vertical mesh size of the computational domain coarsens with depth, from 0.1 m in the peat layer to 3.0 m at the model bottom. The horizontal mesh size spacing increases from 0.2 m at the embankment subgrade to 2.0 m on the left and right sides.

Figure 2

Two-dimensional model domain with temperature and hydraulic boundary conditions. Note that the figure is not-to-scale.

Model simulations were run in three phases. Phases 1 and 2 do not include the road, and are used to generate the permafrost distribution in 1993 before road construction started. The first phase generated the frozen ground with a specified temperature of −6°C and a constant pressure of 0 Pa at the top surface of the NG, shown in Figure 2. The model was run for 1,000 years to reach steady state. In the second phase, the model was repeatedly driven by meteorological data from the SEB model from 1 May 1992 to 30 April 1993 for 100 years to reach a quasi-dynamic equilibrium. The second phase led to an active layer thickness of 1.0 m and the ground temperature of −2.5°C at a depth of 8 m, like observed values below the NG (Fortier et al., 2021). The outputs of Phase 2 were used as the initial temperature and pressure conditions for Phase 3. The initial temperature of the embankment subgrade, added only for Phase 3, was set at the average summer AT (10.8°C) in 1993, the time when road construction started. During Phase 3, the model was driven by meteorological data from the 1993–2018 period. The time step size was 1 year for Phase 1, 1 day for Phase 2, and 8 hr for Phase 3.

For the TBCs, the flux of energy varies with the change of meteorologic and land surface conditions. To represent the complex heat exchange with the atmosphere at the top model surface, the specified time-dependent TBC in Phase 2 and 3 was simulated with the SEB model with inputs of independent variables of meteorological data. This novel TBC, shown in Figure 2, considers the effect of subsurface processes and inter-annual climate. The TBC also allowed the GST to be simulated from surface energy components with varying snow properties and depths (Chen et al., 2021). More details about the other TBC approaches are discussed in Chen et al. (2021). An adiabatic boundary condition was applied to both lateral sides of the model. A heat flux representing the local geothermal gradient (0.08 W·m⁻²; Grasby et al., 2012) is assigned across the bottom of the model domain. The top embankment surface is asphalt-paved with an albedo of 0.13 in the snow-free period and 0.3 in the snow-covered period (Chen et al., 2021; Dumais & Doré, 2016). The top surface of the embankment shoulder and embankment slope are bare and covered by gravel and sand with an albedo of 0.3 (Schneider von Deimling et al., 2021).

For the hydraulic boundary conditions at the model surface, the flux of fluid mass into the model was dependent on rainfall and snowmelt water. The temperature of inflowing water was always equal to the GST. Water may only exit the model at the NG surface or on the embankment sides, through the coupling of the time-dependent specified recharge boundary condition with a drain boundary condition. When the pressure of the surface node was higher than 0 Pa (i.e., the water table intersects the top surface), the drain boundary condition simulated discharge by removing water (Figure 2). The loss of liquid water (discharge) varied linearly with pressure (Provost & Voss, 2019). No flow boundaries were present at the vertical sides, representing the theoretical limits of a hydrogeological watershed, and at the bottom, representing the low-permeability bedrock. The left and right boundary conditions were both set considerably far from the road embankment (100 m lateral distance each), therefore limiting their potential effects on the modeling outcomes for the region of interest for this study. The road pavement surface is also a no-flow boundary, as we consider asphalt to be impermeable.

3.2.4 Modeling Strategy

To investigate the role of advective heat transfer on the permafrost degradation below the road, two additional simulations with (scenario 1) and without (scenario 2) groundwater flow are compared. To deactivate the groundwater flow for scenario 2, we set permeability to a very low value (10⁻¹⁹ m²). The specified GST of scenario 2 was obtained from the SEB model with the same input of meteorological data of scenario 1. The thermal regimes in mid-October and mid-April of 2018 were selected to compare the difference of permafrost table depth and talik formation area between scenarios 1 and 2. The difference of simulated soil temperatures (scenario 1 vs. scenario 2) is presented to investigate the role of the subsurface water flow on permafrost degradation.

Furthermore, the amount of sensible energy stored in the unfrozen zone and the mean soil temperature were used to quantify the feedback of talik development on permafrost degradation. Sensible heat storage in unfrozen water is expressed by Dincer et al. (1997):

(2)

where Q is the amount of heat stored in a single model element (W), T_i is the initial temperature (°C), T_f is the final temperature (°C), m is the mass of heat storage medium (kg), and C_p is the specific heat of liquid water (J·kg⁻¹ °C⁻¹). In this study, the initial temperature (T_i) was set as the initial temperature at a certain depth at the beginning of the simulation, the mass (m) is equal to the volume (Δ x × Δy × ε; assuming a model thickness of 1 m) of the pore soil in a single element multiplied by its saturation (S_L) and its density. The selected elements having 0.2 m in width (Δx) and 0.25 m in height (Δy) were located at 2.5-m, 4.0-m, and 5.0-m depth, respectively.

3.2.5 Model Performance and Sensitivity Analysis

Due to the availability of meteorological data and soil temperatures, we selected the period from 1 May 2014 to 30 April 2018 for simulation validation and compared the measured and simulated soil temperatures at multiple depths below three different surface cover types: road pavement, embankment slope, and NG. The limitations of the modeling approach are discussed in Text S5 in Supporting Information S1.

Sensitivity analysis was used to quantify the uncertainty of the interpolated meteorological data to the simulated GSTs. The sensitivity analysis method changed one parameter at a time, while holding all others constant. For the AT, the uncertainty was quantified by comparing (a) simulations using raw measurement data as a reference case, and (b) simulations using the interpolated AT (Figure S3 in Supporting Information S1). Similarly, the uncertainty related to reconstructed solar radiation data was quantified by comparison of measured values and the interpolated data from the empirical equation as shown in Figure S6 in Supporting Information S1. Additionally, the uncertainties of wind speed (Figure S5 in Supporting Information S1) and rainfall inputs were quantified by comparison of measurements and average values of the measured data over the 2009–2018 period. The uncertainty of snow depth variations was quantified by decreasing or increasing the maximum yearly snow depth by 0.12 m, determined from the deviation of snow depth along the test sections according to the snow survey in 2010 (Stephani, 2013). Furthermore, the mean absolute error (MAE) and mean bias error (MBE) of GST deviations were used to evaluate the uncertainty.

4 Results

4.1 Reconstruction of Meteorological Data (1993–2018)

During the 1993–2018 period, AT experienced an increase, rising at an average rate of 0.07°C·yr⁻¹ (Figures 3a and 3b). Over this period, the MAAT was −4.1°C. The thawing and freezing indices represented an average of 1449.0 thawing degree-days and 2951.2 freezing degree-days, respectively. Mean annual solar radiation (MASR) was 113.4 W·m⁻² (SD = ±3.2 W·m⁻², Figure 3d). The total annual solar radiation was 46.9% of which occurred during summer, while winter radiation represented only 3.8% of the total annual solar radiation. Total annual rainfall exhibited a significant inter-annual variability, with an average of 302.2 mm (SD = 1.2 mm), of which 75% falls in summer (Figures 3e and 3f).

Figure 3

Meteorological data at the BC-RES over the simulated period from 1993 to 2018; (a) daily air temperature (AT) and (b) mean annual AT (MAAT) (°C); (c) daily solar radiation and (d) mean annual solar radiation (MASR) (W·m⁻²); (e) daily rainfall and (f) total annual rainfall (TAR) (mm). The data from 2013 to 2018 were measured at the BC-RES meteorological station (see Figure 1b).

4.2 Thermal Observations (1997–2018)

For the NG (Figure 4a), the permafrost table as identified by the 0°C isotherm, remained stable at approximately 1.0 m depth (SD < 0.05 m). The permafrost temperature below the NG remained around −2.2°C at a depth of 8.0 m. Conversely, for the embankment slope (Figure 4b), the permafrost table deepened from 1.3 to 3.8 m, with an average increase of approximately 0.11 m·yr⁻¹. The increasing rate was not constant, with a relatively larger rate (0.17 m·yr⁻¹) from 2007 to 2018. In 2009, a 1.5-m thick talik initially formed and grew gradually over the observed period. The soil temperature in the talik reached a maximum of 7°C in summer. Further evidence of permafrost warming, and degradation was shown by the depth of −0.2°C isotherm which lowered from 1.6 m in 2002 to 5.4 m in 2018. Similarly, the depth of the permafrost table below the embankment center lowered by 1.2 m (Figure 4c), with an increase rate of 0.05 m·yr⁻¹. The permafrost warmed gradually by about 1.0°C at the 1.5 m depth. A 0.2 m thick talik initiated in 2012 and enlarged to 0.8 m by 2018.

Figure 4

Measured soil temperature variations below the ground surface from 1997 to 2018 in (a) the natural ground, (b) the embankment slope, and (c) the embankment center. The blue dash line in (b) and (c) represents the depth of the interface between the natural ground and the embankment fill. The white column shows the missing data from 1 January to 30 September 2008.

Figure 5 illustrates the variability of soil temperature beneath the embankment slope at different depths. Soil temperature at the 1.5-m depth increased at a rate of 0.05°C·yr⁻¹. The permafrost slightly warmed by 0.5°C in the first decade of monitoring, reaching 0°C in 2007, and completely thawed in 2008. Afterward, the soil temperature had a strong seasonal variation and its annual amplitude quickly increased to approximately 2°C by 2013. During the 2008 winter, the soil did not completely freeze back, and its temperature stayed around 0°C. Similarly, permafrost temperature at the 3.0-m depth experienced a gradual increase, rising at a rate of approximately 0.04°C·yr⁻¹ and exhibiting a low intra-annual variability.

Figure 5

Measured soil temperature variations at the 1.5 m depth (blue) and 3.0 m depth (red) below the embankment slope from 1997 to 2018.

4.3 Model Performance and Sensitivity Analysis Results

A comparison between simulated results and field measurements for three different surface cover types showed good agreement between measured and simulated soil temperatures (R² > 0.93) (Figure 6). The deviation between simulated and observed soil temperatures was largest at the top surface and decreased with depth (Table S2 in Supporting Information S1). By comparison, between simulated and observed GSTs, the mean annual errors (MAE) were 1.0°C for the NG, 1.8°C for the embankment slope, and 2.4°C for the embankment center. Also, the differences between measured and simulated mean annual GSTs were less than 0.4°C for the three surface cover types.

Figure 6

Simulated and measured soil temperatures below the different surface cover types; (a) 0.3-m depth below the embankment center pavement; (b) 0.1-m depth below the embankment slope surface; and (c) 0.3-m depth below the natural ground from 2014 to 2019. Statistic values (R², RMSE, MAE) are shown in the graphs. Note that the mean annual ground temperature (MAGT) is the average near-surface soil temperature at the depth of 0.3, 0.1, and 0.3 m, for the natural ground, embankment slope, and embankment center, respectively, over the simulation period.

Based on the sensitivity analysis results, the model was found to be most sensitive to snow depth. The mean bias error (MBE) ranged from 0.44°C to −0.39°C when the maximum snow depth varied ±12% (Table 1). The next most sensitive parameters are solar radiation and AT. Their MAEs and MBEs were less than 0.4°C. Finally, the model is the least sensitive to changes in wind speed and rainfall.

Table 1. Sensitivity Analysis of Ground Surface Temperatures (GSTs) to Uncertainty of Reconstructed Meteorological Data Inputs

Input parameter	Absolute variation	Sensitivity ranking	+MAE (°C)	+MBE (°C)	−MAE (°C)	−MBE (°C)
Maximum Snow depth	±0.12 m	1	0.39	−0.39	0.44	0.44
Solar radiation	±23.64 W·m^−2	2	0.38	−0.21	0.38	0.21
AT	±1.28°C	3	0.38	−0.10	0.28	0.00
Wind speed	±0.54 m·s^−1	4	0.05	−0.02	0.06	−0.02
Rainfall	±3.5 mm	5	0.00	0.00	0.00	0.00

4.4 Simulated Permafrost Thaw and Talik Development Beneath Road Embankment

4.4.1 Simulated Ground Surface Temperature (1993–2018)

The long-term GST data, shown in Figure 7a, indicate that the simulated mean annual GSTs below the three surface cover types were much warmer than the MAAT. The surface offsets were approximately 3.6°C for the NG, 9.1°C for the embankment slope, and 5.3°C for the embankment center (Figure 7b). Additionally, significant ground warming occurred over the 2006–2018 period at a rate of 0.19°, 0.14°, and 0.24°C·yr⁻¹ for the NG, embankment slope, and center, respectively. These warming rates were about twice greater than those from 1993 to 2005. Additionally, the simulated thawing period under the NG was approximately 20 days shorter than the duration of positive AT, while the corresponding values for the embankment slope and center surfaces were about one week longer (Figure 7c).

Figure 7

(a) Measured mean annual air temperature (AT) and simulated mean annual ground surface temperature (GST) for natural ground (NG), embankment center, and embankment slope for 1993–2018; statistical ranges of (b) mean annual simulated GST and (c) thawing days over the 25-year period, with mean values and standard deviations were represented with thick lines and error bars, respectively; and mean values of (d) AT and (e) GST for each day of year over the 25-year period.

Figure 7e illustrates daily specified GSTs for the SUTRA-ice simulation. In winter, the GST of the embankment center was relatively close to the AT, while the surfaces of the embankment slope and NG were much warmer than the atmosphere, by about 16.1°C and 10.9°C, respectively. Conversely, in summer, the GST of the NG was nearly the same as the AT, while the surfaces of the embankment center and slope were much warmer than the atmosphere, about 13.0°C and 6.6°C higher, respectively.

4.4.2 Permafrost Degradation and Talik Development

Figure 8a shows that the soil temperatures in the embankment subbase and subgrade exhibited an asymmetric distribution. The maximum temperature difference between the up-gradient and down-gradient sides of the embankment was about 5°C for the soil temperature, while the permafrost table was approximately 1 m deeper on the up-gradient side in 2018. After the road construction, the permafrost table at the embankment center moved upwards by about 1.0 m above the NG and was more than 3.0 m shallower at the embankment slope in 1998. As the climate warmed, permafrost degraded quickly and the depth to the permafrost table increased by 4.5 m between 1993 and 2018. The mean soil temperature of the whole embankment subbase increased by an average of 3.5°C, from 3.0°C in 1993 to 6.5°C in 2018.

Figure 8

Simulated (a) temperature distribution in mid-October, (b) temperature distribution in mid-April, and (c) liquid water saturation in mid-April for the years 1998, 2003, 2008, 2013, and 2018. This figure focuses on the road embankment and does not show the entire model domain.

A fast-growing talik initiated below the up-gradient embankment slope in 2003, enlarged laterally and vertically over time, and eventually expanded to the down-gradient side in 2008 (Figure 8b). By 2018, the talik has grown more than three times larger than that of 2003. Furthermore, with the increase of permafrost thaw, more unfrozen water accumulated beneath the embankment (Figure 8c). Since 2003, a saturated unfrozen zone had been located at the embankment slope and enlarged with time.

Figure 9a presents the measured increasing depth to the permafrost table from 1997 (1.3 m) to 2018 (3.8 m), below the down-gradient embankment slope. Similarly, the simulated depth of the permafrost table below the same slope experienced a similar increasing trend at a rate of 0.14 m·yr⁻¹ from 1993 to 2018, where the measured value was 0.12 m·yr⁻¹. The MAE between the measured and simulated depth of the permafrost table was about 0.16 m. In contrast, permafrost thawed faster beneath the up-gradient embankment slope, with an average rate of 0.19 m·yr⁻¹. Over the same period, the average of simulated talik thickness was about 2.1 m below the down-gradient side, which is similar to the observed value (2.0 m). In comparison, the simulated talik thickness underneath the up-gradient side was deeper by 1.4 m on average. By 2018, the talik thickness was 5.3 m, approximately twice larger than that at the down-gradient side (3.2 m).

Figure 9

Simulated and measured (a) permafrost table depth versus time from 1993 to 2018; and (b) statistical analysis of the talik thickness evolution. The depth of the permafrost table was relative to the natural ground, mean values and standard deviations were represented with thick lines and error bars, respectively. The box-plot graphs display the distribution of the data with minimum, first quartile, median, third quartile, and maximum.

Figure 10a demonstrates that permafrost at 2.5 m depth experienced a steep increase in temperature, rising by about 1.0°C in the first 3 years, then gradually warmed and completely thawed in 2005. The soil temperature remained positive over the rest of the simulation period. Conversely, the liquid water saturation had an exponentially increasing trend, reaching the maximum (1.0) in 1997, and remained fully saturated after 2006. There was a similar trend for permafrost at 4.0-m (Figure 10b) and 5.0-m (Figure 10c) depth. Permafrost completely thawed at 4.3-m, and 6.0-m depth in 2010 and 2016, respectively.

Figure 10

Simulated soil temperature below the up-gradient embankment slope and water saturation variations at (a) 2.5 m, (b) 4.0 m, and (c) 5.0 m depth from surface from 1993 to 2018; and mean annual temperature and stored sensible energy in liquid water at a single node located at (d) 2.5 m, (e) 4.0 m, and (f) 5.0 m depth relative to the interface between the natural ground and the embankment fill. Note: the size of the rectangular nodes used for the calculation of stored sensible energy is shown in Panel (d).

Over the 1993–2018 period, mean annual soil temperature had a relatively linear (R² > 0.94) increase, at a rate of 0.19°C·yr⁻¹ at 2.5-m depth (Figure 10d), 0.13°C·yr⁻¹ at 4.0-m depth (Figure 10e), and 0.08°C·yr⁻¹ at 5.0-m depth (Figure 10f). In response to soil warming, the amount of sensible heat stored in the liquid water, calculated for a single rectangular node (0.2 m width × 0.25 m height) located at different depths, exhibited an abrupt and nonlinear increase, rising by 2 W during the thawing period (Figure 10d–10f). Once the permafrost thawed, the amount of stored sensible energy in the liquid water had a relatively linear rise. The corresponding rising rates were 0.16 W·yr⁻¹ at 2.5-m depth, 0.17 W·yr⁻¹ at 4.0-m depth, and 0.20 W·yr⁻¹ at 5.0-m depth.

4.4.3 Comparison Between Heat Advection and Heat Conduction

The permafrost underlying the road embankment for scenario 1 (heat conduction + heat advection) was much warmer than that of scenario 2 (heat conduction only) (Figure 11). The warming rate in scenario 1 was 0.18°C·yr⁻¹ at 1 m depth, about three times than that in scenario 2 (0.06°C·yr⁻¹). The mean annual soil temperature at 1 m depth rose from −1.7°C to 2.9°C in scenario 1, while it remained frozen in scenario 2. The subsurface water flow roughly doubled the permafrost thawing rate (from 0.1 to 0.19 m·yr⁻¹) which caused the permafrost to thaw up to 4.5 m deeper by 2018 (Figure 11a).

Figure 11

Temperature isotherms of scenario 1 (with water flow illustrated by red solid lines) and scenario 2 (without water flow showed with black dashed lines) in (a) mid-October and (b) mid-April of 2018; and temperature difference between the two scenarios (scenario 1 - scenario 2) in (c) mid-October and (d) mid-April of 2018. This figure focuses on the road embankment and does not show the entire model domain.

Similarly, the cross-sectional area of the talik in scenario 1 was twice as large as scenario 2. For scenario 2, two isolated taliks were located at the up-gradient and down-gradient sides of embankment slope. However, these two taliks were replaced by a larger connected talik in scenario 1. Additionally, the consideration of subsurface water flow led to higher temperatures, up to approximately 7°C in October (Figures 11c) and 2°C in April (Figure 11d), respectively, although a few shallow sections of the road embankment are slightly colder in scenario 1.

5 Discussions

5.1 Performance of SUTRA-Ice Model Integrated With SEB Model

We incorporated the SEB model into the SUTRA-ice model to couple the individual surface energy processes with subsurface heat transport and enhance the physical representation of surface conditions in cold regions groundwater models (Figure S8 in Supporting Information S1). The coupled model successfully reproduced the observed seasonal trends of soil temperatures and talik formation for different surface cover types (i.e., undisturbed vegetated NG, seasonally snow-covered embankment slope, and snow-free asphalt pavement, Figure 6). However, compared to the measured soil temperatures, the simulated soil temperatures exhibited a quicker and earlier decreasing trend in fall, and a slightly quicker rise in spring (Figure 6). The main reason is due to the lack of snow depth measurements which leads to poorly represented inter-annual variations of snowfall and snowmelt timing. An earlier snow cover in fall and a shorter snow-covered period in spring linearly reduce the amount of heat released to the atmosphere and warm the ground (Chen et al., 2021; Ishikawa, 2003). The infiltrated snowmelt water causes a step temperature rise by more than 5°C at the ground surface (Chen et al., 2020). The development of cracks due to differential embankment subsidence is not considered in our model. These macropores facilitate heat advection by rainfall water infiltration.

Our model effectively simulated the thermal insulation effect of the snowpack, identified by smaller amplitudes of temperature variations of the embankment slope (Figures 6b and 6c). Similarly, the surface offset caused by the surface albedo was clearly illustrated by the difference (6.4°C) of the summer GSTs between the embankment center and slope (Figure 7).

The discrepancies of simulated GSTs are first due to the uncertainty of snow depth input data, followed by solar radiation and AT, and, to a lesser extent, by wind speed and rainfall (Table 1). Our model is able to estimate daily surface temperature with reasonable accuracy (MAE < 0.4°C, Table 1) with an AT-based empirical equation (Text S2 in Supporting Information S1). Additionally, our model showed that snow depth variations and subsurface temperature feedback greatly improved the accuracy of GST predictions, in comparison to other approaches that neglect these processes (e.g., Dumais & Doré, 2016; Hermansson, 2004; Lunardini, 1978).

The larger variations between the measured and simulated permafrost table, timing of talik initiation, and talik thickness on the down-gradient side of the road (Figure 9) are related to uncertainties of the geometry caused by differential embankment subsidence and recharging of embankment material on side slopes, water flow direction, and temperature of the recharge water. As the coarse embankment material (sand and gravel with large interstitial voids) subsides and eventually intercepts the water table, the ground permeability increases and more groundwater flow (positive feedback) occurs. Furthermore, a lack of available spatial and temporal variation of the snow cover distribution and snow properties also contributes to the discrepancies between measured and simulated results (Cao et al., 2020; Chen et al., 2021). Because the embankment acts as a snow fence which changes the wind field and facilitates snow accumulation on the leeward slope (Doré et al., 2016; Fortier et al., 2011). Additionally, the difference in the solar radiation at the north-facing and south-facing slopes causes the discrepancies. The south-facing slope receives more heat, which will cause a warmer ground surface and lead to the earlier disappearance of the seasonal snow cover in spring. These processes can trigger the asymmetric thermal distribution in the subsurface, variable thawing rates of underlying permafrost, and eventually cause the uneven settlement. The coupled cryohydrogeological model does not simulate deformation processes and/or mechanical characteristics of embankment subsidence caused by the permafrost thaw. Mechanical considerations are needed for future advancement of cryohydrogeological model due to changes in the embankment geometry and soil properties related to the permafrost thaw.

5.2 Permafrost Degradation and Talik Development

Over a period of 25 years, the permafrost degraded at a rate of 0.11 m·yr⁻¹ vertically, and a talik formed under the embankment slopes (Figures 4 and 8) due to a combined effect of snow accumulation on the embankment slope, subsurface water flow, increased surface albedo on the road pavement, and climate warming. We use the modeled NG subsurface conditions as a reference to discuss the main factors affecting these processes.

5.2.1 Impact of Vegetation and Organic Layer

Even with the increased AT of almost 2.0°C during the 1993–2018 period, the permafrost temperature under the NG remained relatively stable (Figure 8). The reason for this stability relates to the thermal insulation effect of the vegetation and soil organic layers (Luo et al., 2018; Sharif et al., 2019), which reduces heat propagation downward and buffers the impact of AT warming (Figures 4 and 7). In summer, heat advection induced by subsurface water flow does not contribute significantly to permafrost degradation below the NG, which is like the simulated results in peatlands (e.g., Kurylyk et al., 2016; McKenzie et al., 2007).

5.2.2 Impact of Subsurface Water Flow

Heat advection driven by groundwater flow (scenario 1) transports additional heat that triples the permafrost warming rate (from 0.06° to 0.18°C·yr⁻¹) and roughly doubles the thawing rate (from 0.1 to 0.19 m·yr⁻¹) (Figures 10 and 11). Subsurface water flow, which is an effective mechanism to transfer heat to greater depths (Chen et al., 2020; Sjöberg et al., 2021), enhances the heat exchange between the underlying permafrost and the overlying embankment. As a result, heat carried by mobile water causes an additional permafrost thaw of 2.1 m on average beneath the embankment (Figure 11a) during the study period. In general, the model scenario considering water flow (scenario 1) better reproduced the thawing rates (0.2 m·yr⁻¹) observed at different test sections (350 m away) at the same study site (Chen et al., 2020; Dumais & Doré, 2016) and are consistent with observations (Fortier et al., 2007; Sjöberg et al., 2016) and other simulated results of the conceptual hillslopes (Dagenais et al., 2020; Ge et al., 2011; Jafarov et al., 2018; McKenzie & Voss, 2013).

Thicker active layer beneath the embankment slope enables a large amount of the subsurface water to be trapped (Figure 8c). Due to the latent heat effect, the presence of trapped free and unfrozen water (Figure 8c) is a positive feedback mechanism to retard freeze back of the active layer, increases the net energy gain to the underlying permafrost body (Figure 10d–10f), and contributes to general higher permafrost temperature (Figure 8b). Over time, the presence of water flow triggers talik initiation and development and connects the isolated taliks from the up-gradient and down-gradient sides (Figure 11). The connection of taliks develops new groundwater flow paths, which enlarge with time along with further permafrost degradation (Figure 8).

Moreover, when the active layer is hydraulically active, heat advection increases warming rates at depth (Figures 8 and 10). High inflow temperature, high permeability, and large hydraulic and thermal gradients enhance heat advection and, as a result, the thawing rate of permafrost (Devoie et al., 2019). The strongest heat advection occurs during the summer months when the temperature difference between the flowing water and the permafrost is maximal (Kurylyk et al., 2016). The presence of taliks in winter allows for perennial flow which slows down the freezing of the active layer.

The asymmetric temperature distribution in the embankment subbase and subgrade (Figure 11) is mainly due to the thermal impact of water flow. Mobile water first reaches and accumulates below the up-gradient embankment slope, due to a thicker active layer resulting from snow insulation over several years (Figures 7 and 10). Meanwhile, the uplifted permafrost table (1 m above the NG surface, Figure 10a), caused by the absence of snow cover on the road pavement at the embankment centerline, acted as a frozen, impermeable, wall that impeded water flow. Over time, the permafrost degraded, and a thicker active layer hydraulically allowed water flow beneath the embankment (Figures 8a and 11). Thereafter, the talik propagated to the down-gradient side of the embankment (Figure 8b).

5.2.3 Impact of Snow Insulation

In winter, insulation from the snowpack reduces the ground heat loss due to its low thermal conductivity. As a result, the ground covered with snowpack is more than 10°C warmer in winter, in comparison to the snow-free surface of the embankment centerline (Figure 7e). When the heat loss in winter is not enough to freeze back the active layer to the permafrost table a talik initiates and grows beneath the road embankment. Similarly, thermal insulation created by thick snowpack has been recognized elsewhere as an important factor for talik formation through observations of sub-Arctic roads (Fortier et al., 2011; O’Neill & Burn, 2017) and with modeling approaches (e.g., Devoie et al., 2019; Jafarov et al., 2018). The flux of heat released across the embankment slopes exponentially decreases with the increase of snow depth and follows a linear relationship with the timing of the first snowfall and the duration of the snow-covered period (Chen et al., 2021; Dumais & Doré, 2016).

5.2.4 Impact of Climate Warming

The simulated GST showed a rapid response to the increasing AT due to climate change (Figure 7). The increasing rate (about 0.09°C·yr⁻¹) of surface temperature is mainly related to the AT warming rate (0.07°C·yr⁻¹), as the total solar radiation remains constant over the 25-year period (SD = ±86.9 W·m⁻²) (Figure 3). This positive trend is like other results reproduced by surface energy models (Kurylyk et al., 2016) and the long-term GST observations in the Qinghai-Tibet Plateau (Wu et al., 2007). Due to the absence of snow cover, the GST at the embankment center exhibited stronger seasonal and inter-annual variability than the GSTs of the embankment slope and the NG (Figure 7).

Increasing AT also contributed to permafrost warming and thawing. Increased AT leads to a rise of the GST by about 2°C (Figure 7a) and a simultaneous increase of the embankment temperature by an average of 3.6°C (from 2.3°C to 5.9°C) in mid-October, both of which result in the rise of sensible heat storage and high thermal gradient between the ground and overlying embankment. As a result, more heat propagates downwards into the underlying permafrost, which leads to permafrost thaw and even creates the formation of taliks. Similar behavior is reported by other modeling approaches (Zhang et al., 2008). Moreover, the increase of AT lengthens the thawing period of the surface (Figure 7b). Longer thawing periods enhance water flow below the embankment and increase the energy absorbed by underlying permafrost (Lamontagne-Hallé et al., 2018). Furthermore, an earlier thawing period will enable more snowmelt water infiltration to reach the frozen front (instead of run-off) and this will warm the ground by latent heat release (Hinkel et al., 2001).

5.2.5 Impact of Road Construction

The simulation results demonstrate that the response of permafrost temperature to road construction is relatively fast, as demonstrated by the fast-warming rate (0.7°C·yr⁻¹) in the first 3 years (Figure 10) a period with stable AT. After road construction, the low-albedo bare surface of the embankment and asphalt pavement enhances the atmospheric heat absorption, resulting in the increase of summer GSTs by about 6.0°C (Figure 7). In summer, the warm road embankment, with an average temperature of 12.4°C (Figure 11), serves as a heat source that is partially transported by the flowing porewater. This process provides an additional energy source to expedite permafrost thaw via heat advection (Figure 10). Alternatively, a relatively slow increase of the talik thickness is observed due to pore-ice latent heat (Figure 4). Over time, the talik thickness is expected to double, once the soil in the zone of −0.2°C isotherm thaws, and triple once the soil in the zone of −0.4°C isotherm thaws (Figure 4b). The thaw-consolidation triggered by talik development may cause pavement damage and embankment shoulder cracking, which is consistent with observations from drill surveys along the Alaska Highway (Stephani, 2013).

6 Conclusions

This study investigates and quantifies the effects of heat advection caused by water flow, snow insulation, and atmospheric warming on permafrost degradation and talik development under a road embankment in a permafrost region. To our knowledge, this is the first study to couple a surface energy balance (SEB) model with a cryohydrogeologic model to quantify the role of heat advection on the thermal stability of a road embankment, supported by long-term field observations. This novel application of surface thermal boundary conditions (TBCs) enables us to consider the inter-annual variations of meteorological data and subsurface processes. Thermal processes beneath three surface cover types are reproduced successfully: undisturbed vegetated NG, snow-covered embankment slope, and snow-free asphalt pavement.

In summer, the warm road embankment, with an average internal temperature of 12.4°C, serves as a heat source and loses heat to the flowing groundwater. Porewater flow induces heat advection which increases heat transfers, tripling the warming rate and expediting thaw, by roughly doubling the permafrost thawing rate. The mobile water produces an asymmetric temperature distribution in the road embankment, which induces the uneven permafrost thaw and adversely affects the thermal and presumed mechanical stability of the infrastructure.

At the study site, a talik formed and grew over time under the combined effect of water flow, snow cover, road construction, and climate warming. The talik is a new thermal state for the road embankment which accelerates the underlying permafrost thaw. The trapped unfrozen water acts as a positive feedback mechanism to retard freeze-back of the active layer, increases the net energy gain to the underlying permafrost body, and contributes to increased permafrost temperatures. Moreover, the results presented herein demonstrate that snow removal from roads should be adapted for permafrost regions. Dispatching the snow from the road pavement onto the embankment slopes may initiate an irreversible permafrost thaw feedback through the creation of taliks.

With a warming climate, subsurface water flow will have a more important role in permafrost thaw and talik development. The results from our research highlight the importance of adequate drainage systems below road embankments to preserve permafrost. For the sustainable development of cold regions, it is necessary to limit subsurface flow in and under a road embankment, and future research focusing on developing effective methods to improve drainage in these structures would be beneficial.

Acknowledgments

The research is supported by the National Science and Engineering Research Council of Canada (Discovery Grant to Daniel Fortier) and Transport Canada. Appreciation is extended to China Scholarship Council (CSC) for a PhD Research Scholarship (Lin Chen, No. 201504910680); the Centre d’études Nordiques (CEN) and Fonds de Recherche du Québec-Nature et technologies (FRQNT) for internship grant (Lin Chen) at U.S. Geological Survey. The authors thank Yukon Highways' Public Works for providing thermal data and access to the test site. The authors also thank Dr. Michelle A. Walvoord (U.S. Geological Survey) and Ms. Natalie Latysh (U.S. Geological Survey) for providing insightful comments. Any use of trade, firm, or product names is for descriptive purposes only and does not imply endorsement by the U.S. Government.

Open Research

Data Availability Statement

The data on which this article is based are available in Fortier and Chen (2020), Fortier et al. (2021), Fortier and Chen (2022b), and Fortier and Chen (2022a). The meteorological data sets of BC-airport and BC-YGT are available on Environment and Climate Change Canada (https://climate.weather.gc.ca/historical_data/search_historic_data_e.html). The SUTRA code and updated version are available on U.S. Geological Survey (https://water.usgs.gov/water-resources/software/sutra/).