Spatio-temporal features of permafrost thaw projected from long-term high-resolution modeling for a region in the Hudson Bay Lowlands in Canada

Authors

  • Yu Zhang

    Corresponding author
    1. Canada Centre for Remote Sensing, Natural Resources Canada, Ottawa, Ontario, Canada
    • Corresponding author: Yu Zhang, Canada Centre for Remote Sensing, Natural Resources Canada, 588 Booth Street Ottawa, Ontario, K1A 0Y7, Canada. (yu.zhang@ccrs.nrcan.gc.ca)

    Search for more papers by this author

Abstract

[1] Although studies agree that climate warming will cause permafrost thaw, projected permafrost conditions differ widely, and most projections use half degree latitude/longitude or coarser spatial resolution. Using a process-based model, this study projected changes of permafrost from 2010 to 2200 at 30 m by 30 m resolution for a region in the northwest of the Hudson Bay Lowlands in Canada. This long-term spatially detailed modeling revealed some general features of permafrost dynamics with climate warming. Temporally, permafrost degradation at a site can be divided into five stages: gradual-thawing stage, increased-thawing stage, frequent-talik stage, isothermal-permafrost stage, and permafrost-free stage. This study determined the beginning or ends of the stages for each grid cell and mapped the degradation stages in this region. Spatially, permafrost was predicted to become increasingly discontinuous with climate warming. By the end of the 22nd century, only 20% to 65% of the land area in this region will be underlain by permafrost. With the formation of taliks, the maximum summer thaw depth will increase significantly, and near-surface permafrost will disappear in many areas while permafrost at depth can persist for decades. Thus, the spatial distribution of near-surface permafrost and permafrost at depth can be very different. This study also shows that climate scenarios, the depth of permafrost considered, spatial resolution and associated ground conditions used for modeling could cause significant differences in permafrost projections.

1 Introduction

[2] Arctic air temperature increased twice as fast as the global average during the 20th century [ACIA, 2005]. Increase in near-surface ground temperatures, deepening of active-layer thickness, and disappearance of permafrost all have been observed in the last decade [e.g., Wang et al., 2000; Payette et al., 2004; Jin et al., 2006; Osterkamp, 2007; Romanovsky et al., 2010; Smith et al., 2010]. Most climate models project that the warming of northern high latitudes will continue to be greater than the global average during the 21st century [ACIA, 2005], which will induce further permafrost degradation. Permafrost thaw has significant impacts on infrastructure, hydrology, soil organic carbon, ecosystems, and feedbacks to the climate system [ACIA, 2005]. Therefore, understanding and predicting permafrost degradation processes are essential for land-use planning, ecological assessment, and climate modeling.

[3] Many studies have predicted the possible responses of permafrost to climate warming. Woo et al. [1992] projected the position of permafrost boundaries in Canada assuming an increase of ground surface temperature by 4–5°C. Nelson and Anisimov [e.g., Nelson and Anisimov, 1993; Anisimov and Nelson, 1996] estimated changes in active-layer thickness using climate scenarios projected by global climate models (GCMs). These studies are equilibrium projections assuming that the ground thermal regime is in equilibrium with the atmospheric climate. A more realistic approach is based on process-based models, which takes into account the transient responses of permafrost to the changes of various climatic factors [Riseborough et al., 2008]. Such projections can be conducted either using permafrost models driven by assumed or GCM-generated climate change scenarios [e.g., Smith and Riseborough, 1983; Anisimov, 1989; Waelbroeck, 1993; Sazonova et al., 2004; Anisimov and Reneva, 2006; Saito et al., 2007; Marchenko et al., 2008; Zhang et al., 2008c; Wisser et al., 2011] or directly using GCMs' land surface schemes by improving permafrost-related processes [e.g., Stendel and Christensen, 2002; Lawrence and Slater, 2005]. Although all models project that permafrost will degrade with climate warming, the projections vary widely. For example, the projected extent of permafrost in the Northern Hemisphere ranges anywhere from 10% to 80% of the present extent by the end of the 21st century (see Schaefer et al. [2011] for a list of different projections).

[4] Field observations show strong spatial variations in permafrost conditions [e.g., Nelson et al., 1998]; the large areas of the discontinuous permafrost zones reflect the effects of spatial heterogeneity. Most spatial permafrost modeling and projections, however, use very coarse spatial resolutions, with grid sizes ranging from half to several degrees latitude/longitude. At such coarse resolutions, it is difficult to consider the spatial heterogeneity of the ground conditions, especially the features of discontinuous permafrost, which is more sensitive to climate warming than continuous permafrost. More importantly, coarse spatial results are difficult to test via field observations and are difficult to use for land managers and ecologists.

[5] Another related issue is how permafrost at a site will respond to climate warming. Recent observations and modeling studies show that formation of taliks is an important mechanism of permafrost thaw from the top [Jin et al., 2006; Delisle, 2007; Zhang et al., 2008b; 2008c; Schaefer et al., 2011]. With climate warming, the top boundary of permafrost will recede to a greater depth, often developing a year-round unfrozen layer (i.e., talik) between the top seasonally thawing/freezing layer and the permafrost at depth. However, it is not clear how important the permafrost at depth will be to near-surface thawing/freezing processes after taliks developed. Our previous study [Zhang et al., 2008b] predicted that the ground thermal regime will be in strong disequilibrium with the atmospheric climate by the end of the 21st century; therefore, we conjectured that permafrost thaw would continue in the 22nd century even without climate warming. In recent years, most GCMs extended their projections of climate to the 22nd century. Therefore, it would be interesting to see how permafrost will change due to the disequilibrium and climate change after the 21st century.

2 Method and Data

2.1 The Study Area

[6] The study was conducted in Wapusk National Park (WNP) located in the northwest of the Hudson Bay Lowlands in Canada (Figure 1). The park is mainly a peatland plain covering 11,475 km2. It straddles the boundary between continuous and discontinuous permafrost and tree line and contains one of the most concentrated polar bear maternity denning areas in the world [Richardson et al., 2005]. It is a concern that permafrost thaw may induce the collapse of peat banks, which could affect the bears' denning habitat. Permafrost thaw may also affect hydrological and soil biogeochemical processes, causing the release of carbon dioxide and methane to the atmosphere from the carbon-rich soils in this region. This area was selected also because detailed field observations are available, and we have modeled and validated the permafrost conditions in the park at 30 m by 30 m resolution for the 20th century [Zhang et al., 2012].

Figure 1.

Wapusk National Park is in the northwest of the Hudson Bay Lowlands shown in dark gray. The dashed curves indicated by numbers 1 to 4 are the southern boundaries of the continuous, extensive discontinuous, sporadic discontinuous, and isolated patches of permafrost mapped by Heginbottom et al. [1995].

2.2 The Model

[7] The Northern Ecosystem Soil Temperature model (NEST) was used to simulate and map permafrost conditions in WNP. The NEST is a one-dimensional model considering the effects of climate, vegetation, snow, and soil conditions on ground thermal dynamics based on energy and water transfer through the soil-vegetation-atmosphere system (Figure 2) [Zhang et al., 2003]. Ground temperature is calculated by solving the one-dimensional heat conduction equation. The dynamics of snow depth, snow density, and their effects on ground temperature are considered as well. Soil water dynamics are simulated considering water input (rainfall and snowmelt), output (evaporation and transpiration), and distribution among soil layers. Based on observations and analysis for peat soils from Zhang et al. [2008a], the model assumes that the fraction of unfrozen water in a soil layer is linearly related to the soil temperature when the temperature is between −0.5 and 0°C. A linear pattern was used so that the fractions of ice and liquid water can be determined analytically based on energy conservation. Detailed description of the model and its applications and validations can be found in Zhang et al. [2003; 2005; 2006; 2012].

Figure 2.

The structure and processes considered in the NEST model [Zhang et al., 2003].

2.3 The Data

2.3.1 Climate Scenarios and Data Processing

[8] An increasing number of future climate scenarios have been projected by different GCMs. Walsh et al. [2008] assessed 15 GCM-generated climate scenarios by comparing them with reanalysis climate data. From the six top performing GCMs assessed by Walsh et al. [2008], I selected two climate scenarios generated by CCCma (CGCM3.1) and MPI ECHAM5 as they represent the general range of air temperature changes projected by most of the GCMs for this region under the intermediate emission scenarios (A1B). The two scenarios are referred to as CCCma and ECHAM, respectively. The monthly climate scenario data were downloaded from World Data Center for Climate (http://mud.dkrz.de/wdc-for-climate/).

[9] The future air temperature and precipitation were calculated using historical data and the projected differences between the future and the 30 year (1961–1990) averages simulated by the same GCMs. The historical monthly air temperature and precipitation were in 10 km by 10 km resolution interpolated based on climate station observations [McKenney et al., 2006]. From the 2000s (i.e., 2000–2009) to the 2190s, annual mean air temperature was projected to increase by 4.6°C and 7.7°C for the CCCma and ECHAM scenarios, respectively. The projected annual total precipitation was similar between these two scenarios, with a slight increasing trend before the middle of the 22nd century, followed by a slight decrease (Figure 3).

Figure 3.

Air temperature and precipitation observed at Churchill climate station and projected by CCCma and ECHAM.

[10] The model was initialized using the climate at the end of the Little Ice Age (circa 1850) assuming that the ground thermal profiles are in equilibrium with the atmospheric climate during that time. The monthly climate during 1850–1900 was estimated by linearly extrapolating the grid climate data for each month [Zhang et al., 2012]. After initialization, the model ran from 1850 to 2010 using the historical climate data and then ran the two climate scenarios from 2011 to 2200.

[11] The NEST model requires daily climate data as input. The projected monthly air temperature and precipitation were converted to daily data using daily observations from 1939 to 2010 at the Churchill climate station as a template [Zhang et al., 2012]. The period 1939–2010 was selected to avoid the extremely cold and thin-snow years observed in the mid-1930s and to match the leap years after 2011 (except 2100 and 2200) with the leap years in the template. Daily air temperature and precipitation for a grid cell were estimated based on the template with two corrections: first to correct the monthly difference between the grid cell and the Churchill climate station, and then to correct the monthly difference between the projected scenario and a historical year in the template. Daily vapor pressure was estimated based on daily minimum air temperature, and daily solar radiation was estimated based on daily diurnal temperature range, vapor pressure, and the solar radiation above the top of the atmosphere. A detailed description of the calculation method can be found in Zhang et al. [2012].

2.3.2 Other Input Data

[12] The land cover map in WNP was developed using Landsat, dual polarized ALOS/PALSAR images, and a digital elevation model (DEM) at 30 m resolution [Zhang et al., 2012]. Leaf area index was estimated using the simple ratio vegetation index (Band-4/Band-3) from Landsat imagery. Peat thickness was estimated based on DEM and land cover types [Zhang et al., 2012]. Mineral soil conditions were based on Dredge and Nixon [1986; 1992] and bedrock was assumed to be 20 m below the land surface. The spatial modeling was conducted based on the input data domain (i.e., only for pixels with different input data). A detailed description of the input data and calculation procedure has been presented in Zhang et al. [2012].

3 Result and Analysis

3.1 Permafrost Degradation Process in WNP

3.1.1 Permafrost Degradation Stages

[13] Figure 4 shows typical permafrost degradation patterns modeled at two sites under the two climate scenarios. The permafrost degradation process at a site can be divided into five stages: (1) the gradual-thawing stage: the active-layer deepens gradually with small fluctuations; (2) the increased-thawing stage: the active-layer deepens in a faster rate but without formation of taliks; (3) the frequent-talik stage: taliks frequently form but disappear in cold years, and the depth of permafrost table and the active-layer thickness vary significantly due to the formation and disappearance of taliks; (4) the isothermal-permafrost stage: taliks become increasingly thicker and can persist regardless of climate fluctuation; and (5) permafrost-free stage: permafrost disappears completely. Stages 2 to 4 are referred to as severe degradation stages in the following discussions.

Figure 4.

Modeled changes in permafrost conditions at two sites under two climate change scenarios. The stages 1 to 5 are for the gradual-thawing, increased-thawing, frequent-talik, isothermal-permafrost, and permafrost-free stages, respectively. Site 1 is in the northern tip of the park, and site 2 is in the southwest of the park. Both sites are sedge fens. Site 1 has little vegetation, whereas site 2 has a leaf area index of 0.2. Site 1 has a surface peat layer of 0.10 m on the surface, underlain by 2 m of silty sand with 10% of gravels further underlain by sand. Site 2 has a surface peat layer of 0.24 m underlain by sand. The bedrock is assumed to be at 20 m below the surface. The geothermal heat flux at 119 m below the land surface is set as 0.054 W m−2 for all the grid cells in the park. It is assumed that there is no water or ice in the bedrock and no excess ice in the soil [Zhang et al., 2012].

[14] Active layer deepens at a faster rate in the increased-thawing stage because soil water at the bottom of the active layer is not completely frozen in the winter and the latent heat usually accumulates from year to year. With the increase in the fraction of unfrozen water in the soil, some soil layers may develop taliks (i.e., the temperature is above 0°C year-round). The taliks that exist in a year may disappear in the following year due to the fluctuation of climate. Thus, the permafrost table and the active-layer thickness can vary widely from year to year in the frequent-talik stage. With climate warming continuing, the talik becomes progressively thicker so that it cannot freeze back even in cold years. During that time, the permafrost is in isothermal conditions and decays steadily from both the top and the bottom. When talik is developed, the active-layer thickness is determined by the maximum frost depth in winter, which tends to become shallower with climate warming. Near-surface permafrost conditions are sensitive to climate fluctuations in the increased-thawing and frequent-talik stages due to the formation of partially or completely unfrozen soil layers. The climate in the previous year, especially in the previous winter, can influence the maximum summer thaw depth by affecting the fraction of unfrozen water in the soil.

[15] Based on the long-term modeling result, the beginnings or ends of the stages can be determined. The beginning of the frequent-talik stage is the first year when taliks are formed, and the end of this stage is the last year when permafrost table connects with the seasonal thawing/freezing layer (i.e., the last year when talik disappears due to climate fluctuation). The end of the isothermal-permafrost stage is the year when permafrost disappears completely. However, the beginning of the increased-thawing stage is not that obvious. The model results show that the beginning of this stage is indicated by significant, usually continuous for some years, deepening of the active-layer (Figure 4), probably because when a certain amount of unfrozen water is accumulated in the soil, the latent heat fuels the thawing process to maintain continuous active-layer deepening. By testing at different sites, it was found that the beginning of the increased-thawing stage can be determined as the first year when the relative change of the 10 year moving average of the active-layer thickness is higher than 5% (Figure 5).

Figure 5.

Relative changes in 10 year moving average of active-layer thickness (curves with circles) and maximum summer thaw depth (gray curves, the scale is shown on the left) at two sites under two climate scenarios (the same sites as in Figure 4). The vertical dashed lines indicate the beginning of the increased-thawing stage.

3.1.2 Spatial Patterns of the Degradation Stages in WNP

[16] Figure 6 shows the spatial distributions of permafrost degradation stages in five decadal time slices from 2000 to 2200 under the two climate scenarios. With time, permafrost degraded following the sequence of the five stages, although some grid cells skipped the frequent-talik stage, especially under the ECHAM scenario. Spatially, there was a general pattern of degradation from south to north under both climate scenarios. This was mainly due to latitudinal climate gradient and higher and denser shrubs in the south, which favored snow accumulation. Permafrost degradation was more severe or faster under the ECHAM scenarios than under the CCCma scenario because the ECHAM scenario was warmer. However, there were some similarities in permafrost distribution under these two climate scenarios. For example, the spatial pattern of degradation stages in the 2090s under the CCCma scenario was similar to that in the 2050s under the ECHAM scenario.

Figure 6.

Spatial distributions of the permafrost degradation stages in five decadal time slices from 2000 to 2200 under two climate change scenarios in WNP. The top panels are under the CCCma scenario, and the bottom panels are under the ECHAM scenario.

[17] The average durations of the degradation stages can be calculated for the land areas in WNP where permafrost existed in 2000 but disappeared by 2200 (the areas account for 30% and 75% of the land areas in WNP under the scenarios CCCma and ECHAM, respectively). The average durations of the increased-thawing, frequent-talik, and isothermal-permafrost stages were 33, 21, and 21 years, respectively, under the CCCma scenario and were 36, 33, and 26 years, respectively, under the ECHAM scenario. In total, it took 76 and 95 years on average from the increased-thawing stage to permafrost disappearance under these two scenarios, respectively. The time was longer under the scenario ECHAM because a larger area with much deeper permafrost was thawed under this scenario and thawing deeper permafrost requires a longer time. For the same area where permafrost was thawed completely under the CCCma scenario, it took 9 years less under the ECHAM scenario (the average lengths of the increased-thawing, frequent-talik and isothermal-permafrost stages were 30, 21, and 15 years, respectively).

3.1.3 Changes in Areas of Different Degradation Stages in WNP

[18] The model results show that about 95% of the land in WNP contained permafrost at present, and most of the permafrost was in the gradual-thawing stage. With time, the area in the gradual-thawing stage would be reduced, whereas the permafrost-free area would expand gradually under the two scenarios (Figure 7). By the end of the 22nd century, only 8% of the land area was in the gradual-thawing stage under the CCCma scenario, and the entire permafrost area would be in severe degradation stages under the ECHAM scenario. The permafrost-free area expanded to 35% and 80% of the land areas in WNP under the CCCma and ECHAM scenarios, respectively.

Figure 7.

Stack graphs to show changes in land areas in different degradation stages from 2000 to 2200 in WNP under the two climate change scenarios.

[19] The results show that the transitions between degradation stages are related to thresholds in climate since thawing of the permafrost begins only when soil temperature is close to 0°C. For example, large areas in the gradual-thawing stage quickly transformed into the increased-thawing stage in the early 2090s and in the mid 2040s under the CCCma and ECHAM scenarios, respectively. These quick transformations corresponded to the periods when the annual mean air temperature was close to or over −2°C. Similarly, the widespread formation of taliks during 2080–2100 under the ECHAM scenario corresponded to a period when the annual mean air temperature reached 0°C.

[20] The permafrost area in severe degradation stages expanded with time under the CCCma scenario. Under the ECHAM scenario, however, the area expanded quickly in the second half of the 21st century but gradually shrank during the second half of the 22nd century as permafrost disappeared completely in many areas, while all gradual-thawing permafrost had entered into severe degradation stages. Excepting permafrost-free areas, the areas of other stages showed abrupt changes because they were defined based on the thermal conditions near the land surface, whose response to climate is quick and similar among different areas. The permafrost-free areas expanded gradually because thawing deep permafrost is a slow process and various depths of permafrost exist in WNP due to different climate, vegetation, hydrology, and ground conditions.

3.2 Near-Surface Permafrost and Permafrost-in-General in WNP

[21] With climate warming, especially with the formation of taliks, the permafrost table could be several meters below the land surface (Figure 4). Therefore, it is useful to identify and analyze the changes in near-surface permafrost because it directly affects ecological and environmental processes. In this study, near-surface permafrost was defined as areas with permafrost in the top 4 m soil, and permafrost-in-general was the area with permafrost regardless of its depth.

[22] Spatially, permafrost in some southern areas and near the coast disappeared earlier, while northern and inland regions retained permafrost longer (Figure 8). Permafrost persisted longer in inland than near the coast mainly because inland is dominated by bogs with thick peat, while coastal areas are dominated by fens with shallower peat. Faster degradation of permafrost in the south than in the north is related to latitudinal climate gradient and vegetation conditions. With time, permafrost degraded from continuous to increasingly discontinuous for both the near-surface permafrost and permafrost-in-general. The spatial distributions of near-surface permafrost and permafrost-in-general were different since near-surface permafrost disappeared at a much faster rate than the permafrost-in-general (Figure 8). Thawing of deep permafrost was slow because transferring the warming signal from the land surface to the deep ground could take decades to a century. The areal difference between the near-surface permafrost and permafrost-in-general increased during the model period under the CCCma scenario (Figure 9a). Under the ECHAM scenario, however, the difference increased abruptly during 2080–2100 due to the widespread formation of taliks but decreased gradually in the second half of the 22nd century due to disappearance of deep permafrost (Figure 9b). These patterns were similar to that of the areas in the frequent-talik and isothermal-permafrost stages (Figure 7) since disappearance of near-surface permafrost was mainly due to the formation of taliks.

Figure 8.

Spatial distributions of the maximum summer thaw depth in five decadal time slices from 2000 to 2200 under two climate change scenarios in WNP. The top panels are under the CCCma scenario, and the bottom panels are under the ECHAM scenario.

Figure 9.

Stack graphs to show changes in land areas in WNP without permafrost, without near-surface permafrost but with permafrost at depth, and with near-surface permafrost (usually with permafrost at depth) under the two climate change scenarios in WNP.

[23] Although most of the warming in the atmospheric climate occurred in the 21st century (Figure 3), large areas of permafrost disappeared in the 22nd century (Figure 9). Under the CCCma scenario, air temperature increased 3.8°C and 0.8°C during the 21st and 22nd centuries, respectively (based on 10 year moving averages). Near-surface permafrost disappeared in 30% and 36% of the land in WNP during the 21st and 22nd centuries, respectively. Permafrost-in-general disappeared in 13% and 18% of the land in WNP during these two centuries, respectively (Figure 9a). For scenario ECHAM, air temperature increased 6.2°C and 1.6°C during the 21st and 22nd centuries, respectively. Near-surface permafrost disappeared in 63% and 30% of the land in WNP during the 21st and 22nd centuries, respectively. Permafrost-in-general disappeared in 18% and 57% of the land in WNP during the 21st and 22nd centuries, respectively. The disappearance was in a faster rate during the 22nd century, especially after 2150 (Figure 9b).

3.3 Changes in Maximum Summer Thaw Depth and Active-Layer Thickness

[24] The modeled maximum summer thaw depth (i.e., the depth to permafrost table) increased toward the coast mainly due to the reduction in peat thickness and associated changes in ecosystem types (Figure 8). This spatial pattern and the major controls are similar to those of the active-layer thickness modeled for the 20th century [Zhang et al., 2012] (maximum summer thaw depth equals the active-layer thickness when there is no talik). Average maximum summer thaw depth in WNP increased from 0.97 m in the 2000s to 1.64 m and 3.1 m in the 2190s under the two scenarios, respectively (in the calculation of the averages, the maximum summer thaw depth was assumed as 4 m when it was deeper than 4 m). For the areas with permafrost but no taliks all the years during 2000–2200 (23% and 2% of the land area in WNP under the CCCma and ECHAM scenarios, respectively), the average active-layer thickness increased 83% and 394% from 2000 to 2200 under these two scenarios, respectively (Figures 10a and 10b). The increase and fluctuation of the active-layer thickness were larger under ECHAM, especially after 2150, because most of the permafrost was in increased-thawing stage.

Figure 10.

(a and b) Changes in average active-layer thickness for areas with permafrost but without taliks all the years during 2000–2200 in WNP and (c and d) changes in average active-layer thickness and average maximum summer thaw depth for areas with permafrost all years during 2000–2200 and with taliks in some years in WNP. Figures 10a and 10c are under the CCCma scenario, and Figures 10b and 10d are under the ECHAM scenario.

[25] For the areas with permafrost all the years during 2000–2200 and with taliks in some years (40% and 18% of the land areas in WNP under the CCCma and ECHAM scenarios, respectively), the average active-layer thickness deepened gradually in most of the 21st century (Figures 10c and 10d). The deepening was more rapid near the end of the 21st century. The maximum summer thaw depth became deeper than the active-layer thickness after around 2120–2140, corresponding to the formation of taliks in most of the areas. After the formation of taliks, the maximum summer thaw depth deepened much faster, and the average active-layer thickness became somewhat shallower because it was determined by the maximum frost depth in winter, which decreased with time due to the warming in winter.

[26] The model results also show that the depth to permafrost table was usually less than 20 m, even several years before the complete disappearance of permafrost. The base of permafrost ascended gradually but continuously. The reduction of the deep permafrost was mainly due to the rise of permafrost base (thawing from the bottom), although it may be over-estimated because the deep ground was assumed to be composed of bedrock without water or ice. For shallow permafrost, thawing from both the top and the bottom could be similar.

4 Discussion

[27] This high-resolution modeling result shows that permafrost will persist in 65–81% of the land area in WNP by the end of the 21st century. This result is qualitatively in agreement with our previous coarse resolution national modeling result [Zhang et al., 2008c] and the result of Wisser et al. [2011], both of which project that permafrost would still persist in this region at the end of the 21st century. However, some much coarser resolution studies [e.g., Lawrence and Slater, 2005] projected that permafrost would disappear in this region by the end of the 21st century partly because their coarse spatial resolution has difficulties resolving the subgrid ground conditions and their related impacts.

[28] This long-term modeling study shows significant permafrost degradation in the 22nd century although climate warming was not as significant as in the 21st century. This result confirms our previous conjecture that permafrost thaw will continue in the 22nd as ground temperature will be in strong disequilibrium with the atmospheric climate by the end of the 21st century [Zhang et al., 2008b]. Schaefer et al., 2011 also found that permafrost degradation continues in the 22nd century, although the climate scenario has a very small warming rate during the 22nd century.

[29] Although the modeled degradation of permafrost in WNP is expected and the trend is generally consistent with other projections [e.g., Zhang et al., 2008c; Schaefer et al., 2011; Wisser et al., 2011], this long-term, transient, high spatial resolution modeling also revealed some general features of permafrost dynamics with climate warming. Firstly, the degradation of permafrost varied from site to site due to ground conditions and climate, even for a small and relatively flat area like WNP. Permafrost became increasingly discontinuous with climate warming. Such strong spatial difference indicates that modeling and mapping permafrost at a high spatial resolution is important to capture these spatial variations, including the feature of discontinuity, and the results are more useful for land-use planning and ecosystem assessment.

[30] Secondly, these model results show that permafrost degradation at a site can be divided into five stages. As the names and the definitions of the stages implied, they are useful to characterize the processes of permafrost thaw with climate warming. Similar stages of permafrost thaw have been identified by Delisle [2007] and Jin et al. [2006]. Based on observed changes in ground temperature profiles, Jin et al. [2006] divided the process of downward degradation of permafrost into four stages: (1) initial degradation; (2) intensive degradation; (3) layered talik; and (4) no permafrost. Based on numerical modeling, Delisle [2007] indicated that permafrost degradation will pass through three stages: (1) deepening of the active layer but without taliks; (2) formation and deepening of taliks; and (3) permafrost at depth becomes isothermal and decays with time. In the gradual-thawing stage, the relative variation of the active-layer thickness tends to be smaller with the increase in the active-layer thickness. Such a pattern has been observed and can be explained using the Stefan equation [Shur et al., 2005]. Due to the existence of unfrozen water in the increased-thawing stage, the active-layer thickness can be much deeper than the Stefan solution, suggesting that the equation is not suitable for this stage. This study also indicates the importance of taliks in permafrost degradation, which has been identified in other modeling studies [Zhang et al., 2006; 2008b, 2008c; Delisle, 2007; Schaefer et al., 2011] and has been observed in the field as well [Wang et al., 2000; Jin et al., 2006; Romanovsky et al., 2010]. Refreezing of taliks has been observed, and it is suggested that such occurrences may be common [Osterkamp, 2007]. Therefore, it is useful to identify the frequent-talik stage. The depth and duration of the thawing and freezing conditions near the land surface in this stage could fluctuate widely with climate, which would have special implications for ecological and other land surface processes.

[31] Thirdly, this study shows that the persistence of permafrost at depth is important for the thawing/freezing dynamics in the near-surface layers even when taliks developed, especially during the frequent-talik stage. This study also emphasizes the importance of unfrozen water for permafrost degradation. The beginning of the increased-thawing stage is due to water at the bottom of the active layer not completely frozen in the winter. The degrading of the isothermal permafrost is mainly related to the thawing of the small fraction of the ice in the soil.

[32] And finally, this study also provides some insights into the large discrepancies among permafrost projections. In addition to the different models used, this study shows that three other factors may cause different projections: climate scenarios, depth of the ground considered for permafrost modeling and mapping, and spatial resolution and associated ground conditions used for modeling. This study shows that the degradation rates and permafrost extents under the two climate scenarios are very different. The spatial distribution of near-surface permafrost was very different from that of permafrost at depth (Figures 8 and 9). Coarse resolution has difficulties to resolve local ground conditions and usually cannot reflect the discontinuity of permafrost. Therefore, coarse results can be very different from the results of finer resolutions.

[33] This modeling study has several limitations and uncertainties. Firstly, the NEST is a one-dimensional model which did not consider the lateral transfer of heat, especially near water bodies in the park and near Hudson Bay. Secondly, this study did not consider temporal changes in vegetation, land cover types, and soil conditions. It is very likely that vegetation composition, plant heights and density, peat thickness, and even land cover types will change with climate warming and fire disturbances. Other disturbances (e.g., subsidence at peat plateau margins, collapse of plateau surface, formation and migration of thermokarst lakes) could alter hydrological conditions and land cover type as well. All these changes could have significant impacts on the permafrost thawing process [Dyke and Sladen, 2010]. And thirdly, the model was initialized based on the climate in 1850, which was estimated by linear extrapolation of the monthly data from 1901 to 1990. The uncertainty of the initial conditions could affect permafrost thickness and the timing of permafrost disappearance.

5 Conclusions

[34] This study projected changes of permafrost conditions in WNP at 30 m by 30 m spatial resolution during the 21st and 22nd centuries using a process-based model. The model results show that permafrost degradation at a site can be divided into five stages. Permafrost conditions, especially the thawing/freezing conditions near the land surface, show different features in different stages. Therefore, they can be used as a new measure to characterize permafrost degradation processes. This spatially detailed, long-term modeling shows that permafrost degradation in WNP differs from site to site, and permafrost will become increasingly discontinuous with time. By the end of the 22nd century, only 20% to 65% of the land area in WNP will be underlain by permafrost. With the formation of taliks, maximum summer thaw depth will increase significantly, and near-surface permafrost will disappear in many areas, while permafrost at depth can persist for decades. Thus, the distribution of near-surface permafrost and permafrost at depth can be very different in the future.

[35] This long-term, high resolution projection of permafrost also provides some insights into the discrepancies in recently published permafrost projections. In addition to the different models used, climate scenarios, depth of permafrost considered, and spatial resolution and associated ground conditions used for modeling could cause significant differences in permafrost projections.

Acknowledgments

[36] The author would like to thank Drs. Daniel Riseborough, Ian Olthof, and Ms. Wendy Sladen for their critical internal review of the paper. The comments from three anonymous reviewers were very helpful to improve the manuscript. This study is supported by the remote sensing science program and the climate change geoscience program in Natural Resources Canada (ESS contribution number 20110393), a GRIP project (ParkSPACE) funded by Canadian Space Agency, and Canada's IPY funding (CiCAT).