Subarctic peatlands underlain by permafrost contain significant amounts of
organic carbon. Our ability to quantify the evolution of such permafrost
landscapes in numerical models is critical for providing robust predictions of
the environmental and climatic changes to come. Yet, the accuracy of
large-scale predictions has so far been hampered by small-scale physical processes
that create a high spatial variability of thermal surface conditions,
affecting the ground thermal regime and thus permafrost degradation
patterns. In this regard, a better understanding of the small-scale
interplay between microtopography and lateral fluxes of heat, water and snow
can be achieved by field monitoring and process-based numerical modeling.
Here, we quantify the topographic changes of the Šuoššjávri
peat plateau (northern Norway) over a three-year period using drone-based
repeat high-resolution photogrammetry. Our results show thermokarst
degradation is concentrated on the edges of the plateau, representing 77 %
of observed subsidence, while most of the inner plateau surface exhibits no
detectable subsidence. Based on detailed investigation of eight zones of the
plateau edge, we show that this edge degradation corresponds to an annual
volume change of
Using the CryoGrid3 land surface model, we show that these degradation patterns can be reproduced in a modeling framework that implements lateral redistribution of snow, subsurface water and heat, as well as ground subsidence due to melting of excess ice. By performing a sensitivity test for snow depths on the plateau under steady-state climate forcing, we obtain a threshold behavior for the start of edge degradation. Small snow depth variations (from 0 to 30 cm) result in highly different degradation behavior, from stability to fast degradation. For plateau snow depths in the range of field measurements, the simulated annual volume changes are broadly in agreement with the results of the drone survey. As snow depths are clearly correlated with ground surface temperatures, our results indicate that the approach can potentially be used to simulate climate-driven dynamics of edge degradation observed at our study site and other peat plateaus worldwide. Thus, the model approach represents a first step towards simulating climate-driven landscape development through thermokarst in permafrost peatlands.
Observations show that permafrost is warming at a global scale (Biskaborn et al., 2019). Its thawing has major consequences on arctic and boreal ecosystems and landscapes (Beck et al., 2015; Farquharson et al., 2019; Liljedahl et al., 2016) and potentially represents an important climate feedback through the decomposition of thawed organic matter (Koven et al., 2015; Schuur et al., 2009, 2015). Carbon emissions from permafrost regions towards the atmosphere are already observed (Natali et al., 2019); field measurements show that these emissions are influenced by the timing of the active layer deepening (Morgalev et al., 2017) and by the state of degradation of the permafrost terrains (Langer et al., 2015; Nwaishi et al., 2020; Serikova et al., 2018). In particular, abrupt thawing of ice-rich permafrost is expected to become a significant factor for carbon emissions, potentially offsetting the negative feedback by increased ecosystem productivity that is expected for gradual thaw (McGuire et al., 2018; Turetsky et al., 2020).
As such, our ability to quantify and represent the physical evolution of permafrost landscapes is critical to provide robust predictions of the environmental and climatic changes to come (Aas et al., 2019; Andresen et al., 2020; Teufel and Sushama, 2019). While permafrost affects about 14 million square kilometers in the Northern Hemisphere (Obu et al., 2019), the ground thermal response to climatic signal and morphological changes of permafrost are governed by processes occurring within a spatial scale of a few meters (Gisnås et al., 2014; Jones et al., 2016; Martin et al., 2019; Way et al., 2018). Indeed, lowland permafrost landscapes (such as peat plateaus and polygonal tundra) are characterized by low amplitude (0–3 m vertically) and high frequency (10–100 m horizontally) spatial variations of their topography, often referred to as microtopography (French, 2018). This microtopography drives the lateral redistribution of snow, liquid water and heat, which can dramatically modify the ground thermal regime and water content (Martin et al., 2019). In many cases, the microtopography results from the presence of excess ice in the ground, i.e., the volume of ice that exceeds the total pore volume that the ground would have under unfrozen conditions (NSIDC glossary), so permafrost thawing results in surface subsidence (“thermokarst”; Göckede et al., 2017, 2019; Nitzbon et al., 2019, 2020). In models, the representation of this feedback between small-scale fluxes and dynamical topography is still in its infancy and large-scale permafrost modeling studies usually lack these processes (Park et al., 2015). Robust predictions of the physical state of permafrost landscapes thus require further field observations and model development to improve our understanding of these phenomena.
Peat plateaus are permafrost landforms covering extensive regions throughout
the boreal and arctic realms that store nearly 200 Pg of carbon (Lindgren et al., 2018). They are mainly
located in the sporadic permafrost zone (Seppälä,
1972; Sollid and Sørbel, 1998) and are typically associated with a
climatic envelope characterized by a mean annual ground temperature around
0
The ongoing degradation of Fennoscandian peat plateaus is a potential
analogue for the future of much larger peat plateau areas found in Russia,
Canada and Alaska. It provides the opportunity to conduct field measurements
and test process-based model approaches to further understand the local
drivers of permafrost peatland dynamics. Both field measurements and
numerical modeling experiments have contributed to our understanding of how
microtopography drives the lateral fluxes of heat, water and snow and
impacts the ground thermal regime (Martin
et al., 2019; Sannel, 2020; Sannel et al., 2016; Sjöberg et al., 2016).
Transport of snow and water from the plateau to the surrounding mire are
critical factors leading to lower ground temperatures in the peat plateaus
(mean annual temperature at 1 m depth is 2 to 3
In this study, we quantify volume changes of a peat plateau in northern Norway using repeat digital elevation models compiled from drone aerial imagery. Using photogrammetry, the surface topography of the plateau was reconstructed and compared for the years 2015 and 2018. Subsequently, we adapt the laterally coupled, tiled version of the CryoGrid3 model (Langer et al., 2016; Nitzbon et al., 2019; Westermann et al., 2016) to reproduce observed patterns of microtopography change, including an analysis of model sensitivity towards snow depth. The work presented here builds on field observations and simulations for peat plateaus in northern Norway (Martin et al., 2019) and other arctic landscapes (Aas et al., 2019; Nitzbon et al., 2019, 2020, 2021).
The Šuoššjávri peat plateau (69.38
The Šuoššjávri site consists of a laterally incised peat plateaus and smaller peat mounds surrounded by wet mires and ponds. These peat bodies extend over meters to several tens of meters with irregular geometries and rise 1 to 3 m above the surrounding wet mire (Fig. 2). At many locations, the peat plateau edges show signs of advanced degradation and lateral erosion.
We used drone-based
Measurements of snow depths for winter 2016, as presented in Martin et al. (2019), were extended to the winters of 2017 and 2018 for selected points on the plateau top. Snow depths were measured at the end of March with an avalanche probe at the same points in all years using a dGPS system to define the locations within 5–10 cm accuracy. In March, the elevated peat plateaus are commonly covered with 10 to 40 cm of snow, with most of the values between 10 and 30 cm (Fig. 3). We used these observations to design numerical simulations with four idealized snow scenarios in which maximum snow depth on the plateau is limited to 0, 5–10, 10–20 and 20–30 cm (Sect. 3.3.2).
March snow depth distribution on top of the Šuoššjávri peat plateau for 2016, 2017 and 2018. Data for 2016 are from Martin et al. (2019). Measurement locations are displayed in Fig. 2.
Our drone-based photogrammetric approach (Sect. 3.1) enabled us to derive
repeat DEMs of the peat plateau for September 2015 and 2018. We computed the
elevation difference (elevation from 2018 minus elevation from 2015) to
quantify the spatial pattern of elevation changes throughout the plateau.
From this elevation difference map, we selected eight degrading zones
(referred to as
Permafrost degradation in peatlands creates different subsidence patterns
depending on the size of the ice-rich features. Small structures like palsas
tend to sink uniformly from the edge to the top, while peat plateaus show
stability of their top part and pronounced lateral retreat. To distinguish
between these two types of thermokarst patterns, we introduce a so-called
For the simulation results (Sect. 4.2), a 10 m long window was used to capture the topography from the base of the plateau to its flat top. For these 10 m profiles, the five required points were determined and the HvsV index was computed over three-year-long time periods.
Schematic representation of the HvsV index used to quantify the observed and simulated thermokarst patterns. The index ranges from 1, when the plateau undergoes pure lateral edge retreat (subsidence restricted to the areas at contact with the mire), to 0, when it experiences uniform subsidence.
We simulate the ground thermal regime and related topographic evolution of the Šuoššjávri peat plateau using the CryoGrid3 model (Westermann et al., 2016). CryoGrid3 is a land surface model designed for permafrost modeling, which consists of a physically based description of 1D heat transfer in the soil column, including freeze–thaw processes of soil water or ice. The model features a simple snowpack module, which includes heat conduction, dynamic buildup, melt, sublimation, water infiltration and refreezing. At the upper boundary, the model uses the surface energy balance module to calculate the ground surface temperature. The turbulent fluxes of sensible and latent heat are calculated using a Monin–Obukhov approach (Monin and Obukhov, 1954). Computation of dynamic soil moisture is accomplished with a bucket scheme (Martin et al., 2019, Nitzbon et al., 2019) in which each grid cell can hold water up to its field capacity, while excess water is moved to the next grid cell until a water table on top of the permafrost (or bedrock layer) is reached. Evapotranspiration is adjusted by soil moisture availability and the water loss is distributed vertically, so it decreases exponentially with depth.
CryoGrid3 can represent ground subsidence resulting from the melt of the excess ice in the ground (see Westermann et al., 2016). The subsidence calculation is based on soil stratigraphy, in particular volumetric ice content and natural porosity, i.e., the porosity of the soil matrix in unfrozen conditions. When excess ice melts in a grid cell, the grid cell size shrinks accordingly. This excess ice scheme was first implemented by Westermann et al. (2016) and later used in Nitzbon et al. (2019, 2020, 2021) to represent the transient evolution of polygonal tundra landscapes for different future climate scenarios. In the present study, this scheme is adapted to simulate microtopography changes and thermokarst patterns of the Šuoššjávri peat plateau.
Following Nitzbon et al. (2019, 2020), CryoGrid3 includes a parallel
framework to simultaneously compute several 1D tiles that can exchange
water, snow and heat at defined time steps. This approach, denoted as
The snow depth is a major control for the ground thermal regime (Gisnås et al., 2014; Martin et al., 2019; Sannel, 2020; Sannel et al., 2016). Strong wind redistribution of snow from the plateau to the lower-lying mire leads to a shallow snow cover on the plateaus (Sect. 3.1). In the laterally coupled tiling approach of CryoGrid3, wind drift of snow is not computed in a physically based way. Instead, fresh snow is redistributed at regular time intervals between all tiles based on the relative surface elevations of the snow covered tiles. Tiles gain or loose snow proportional to the difference between their surface elevation and the average surface elevation of all tiles in a mass-conserving scheme. Thereby, snow is redistributed between all the tiles without taking their relative location into account. To represent immobile snow trapped by vegetation and/or rough surfaces, snow is only considered movable if its depth exceeds the “immobile snow height”, which can be adjusted as a model parameter. In the setup used for this study, the elevation difference between the plateau and the mire leads to complete redistribution of snow that exceeds the immobile snow height from the plateau to the mire. The immobile snow height can therefore be used to adjust the overall snow depth on the plateau in our modeling experiments.
The present-day topography of the peat plateau is clear evidence that the long-term evolution is not only governed by lateral thermokarst at the plateau edges but also by thermokarst processes in the interior of the plateau, with pond formation and drainage as well as drainage gully development and deepening. As a fully 3D simulation of these phenomena is beyond the capability of our model approach, we focus on the simplified situation of a laterally homogenous peat plateau edge, for which all fluxes in the third spatial dimension are assumed zero (translational symmetry). For this purpose, we couple 40 tiles in a linear configuration, with subsurface water and heat transfer only allowed between neighboring cells (Fig. 5). The wet mire is divided into three tiles with widths of 50, 2 and 0.5 m at a surface elevation of 300 m a.s.l. They are composed of a 3 m thick layer of unfrozen saturated peat above a 7 m thick silty mineral layer that also extends below the plateau. The outermost mire tile is linked to a hydrologic reservoir (Nitzbon et al., 2019) at 300 m a.s.l. to ensure a stable water level with permanently water-saturated conditions. The peat plateau tiles are 0.3 m wide so the initial width of the plateau amounts to 11.1 m. They contain the same total amount of peat above the mineral base layer as the mire tiles but include additional excess ice, which increases their surface elevation. In line with observations (Table 1), the initial excess ice content is adjusted so that the flat top of the plateau is located 2 m above the wet mire at 302 m a.s.l. The excess ice is initially distributed homogeneously between the mineral base layer and the bottom of the active layer (assumed 0.7 m deep). The selected peat and excess ice stratigraphy implicitly ensures that the plateau surface reaches the surface elevation of the mire when the excess ice has fully melted. Our setup leads to an initial excess ice content of 47 % (volume of excess ice/volume of unfrozen soil) in the plateau, which is in the range of commonly reported field values (Bockheim and Hinkel, 2012; Kokelj and Burn, 2003; Lacelle et al., 2013; Morse et al., 2009; Subedi et al., 2020).
Field observations at the eight edge transect areas presented in Fig. 2. Elevation changes and surface measurements are derived from the digital elevation models. The normalized annual volume changes are obtained by dividing the annual volume changes by the length of the retreating edge in the zone. The HvsV shape index (Fig. 4) was calculated according to Eq. (1). See Sect. 3.2 for more details.
While this is clearly an idealized setup, it is still possible to compare
the magnitudes of modeled volumetric plateau degradation with field
observations for sufficiently straight sections of the plateau edge (Sect. 3.2, Fig. 2). As field observations of snow depth show a considerable spread
of snow depths on the plateau (that cannot be reproduced by modeling), we
investigate model sensitivity towards snow depths on the plateau by
adjusting the immobile snow height using four different values within a
realistic range. In each configuration, the same immobile snow height was
applied to all tiles. During the simulations, the snow depth on the plateau
varied within ranges of 5–10 cm due to snow fall, snow drift and snow melt.
Therefore, we named the scenarios based on their snow depth range, i.e.,
Setup used to simulate peat plateau degradation. We coupled 40 CryoGrid3 tiles to reproduce the contact between the mire and the peat plateau. The surface elevation was linearly interpolated between 300 and 302 m a.s.l. over a lateral distance of 2.4 m to represent a typical geometry of peat plateau edge. AL stands for active layer. We linearly interpolated them between 0.9 (leftmost tile) and 0.7 m (rightmost tile). The model implements lateral fluxes of snow, subsurface water and heat between the tiles as well as ground subsidence due to excess ice melt. The bottom part of the setup has been truncated because it consist of silt over 7 m for all tiles. This setup is an idealized setup derived from our field observations. It does not aim to represent one particular natural setup of the edge transect areas detailed in this study.
Left: subsidence over the Šuoššjávri peat plateau. The black lines indicate the profiles used to derive the edge retreat metric described in Sect. 3.2 (HvsV index). Note that the color scale is truncated for subsidence values higher than 40 cm and lower than 0 cm (which corresponds to an increase in surface elevation, generally due to vegetation change between the two years). Right: edge transect areas of the plateau used to compare observed and simulated lateral thermokarst patterns (Sect. 3.2).
As described in Martin et al. (2019), field measurements from the Iškoras peat plateau (40 km east of the site of the present study) were used to establish the soil stratigraphy. The peat layer in the mire has total volumetric contents of 5 % mineral and 15 % organic material and a porosity of 80 %. It is underlain by a saturated mineral silt layer with 50 % porosity above a mineral bedrock layer (3 % porosity, as in Westermann et al., 2013). Over the Šuoššjávri plateau, the soil stratigraphy features a significant spatial variability. The stratigraphy assumed in the model matches the western parts of the plateau, where strong subsidence is observed and from where annual volume changes for model comparison are obtained. In the eastern and southern parts (Fig. 2), the organic soil thickness is only shallow with morainic deposits close to the surface. This part is comparably stable and not targeted by our modeling.
Snow and soil parameters are based on the field measurements and the
sensitivity tests from Martin et al. (2019). The simulations use a snow
density of 230 kg m
As presented in Martin et al. (2019), we use model forcings for the
hydrological year 2015–2016 that have been compiled by dynamical
downscaling of the ERA-Interim reanalysis (Dee et al., 2011) with the
Weather Research and Forecasting model (WRF v.3.8.1; Skamarock and Klemp, 2008). The WRF model was
run in two nested domains with 15 and 3 km grid spacings from August 2015
to July 2016. To generate the model forcing for CryoGrid3, we used 3-hourly
output from the nearest grid point in the 3 km domain. The other model
parameters for WRF were selected as in Aas et al. (2016), with the exception of
slightly higher vertical resolution (45 model layers compared to 40) and
excluding the climatic mass balance glacier module. The one-year forcing
data are looped to generate a 100-year time series with steady-state climate
forcing. To also achieve a realistic initial temperature profile in deeper
layers, a 100-year spin-up is performed for all simulations using the /textit0 cm
snow scenario for which the peat plateau is stable (Sect. 4.2). Note that
the other snow scenarios cannot be used for model spin-up, as the plateau
edge starts to retreat instantly so a true steady state cannot be
reached. As shown in Fig. 1, the hydrological year 2015–2016 was
relatively warm. It is 0.9
The topographic changes of the Šuoššjávri peat plateau
between September 2015 and September 2018 are presented in Fig. 6. From the
DEM difference, we found that 19 % of the plateau displays 5 cm or more
subsidence (i.e., the sensitivity threshold of the measurements), which
consequently implies that 81 % of the plateau is stable during the
observation period. The mean subsidence value (considering values larger
than 5 cm) is
We extracted the outermost 2 m of the plateau edge which corresponds to one third of the total plateau surface but represents 77 % of the total subsidence (including the rims of the depressions within the plateau). The distribution of subsidence values for the whole plateau and the 2 m edge zone are presented in Fig. 7.
Due to the spatial variability in the peat plateau stratigraphy described in
Sect. 3.3.3, the west side of the plateau features higher subsidence values
than the east side. On the east side, ground subsidence is lower due to the
limited thickness of the peat layer, with mineral soils at a depth of less
than 1 m below the surface. A description of the eight edge transect
areas and their subsidence between 2015 and 2018 is presented in Table 1.
For these eight zones, the normalized annual volume change is
Subsidence distribution for the Šuoššjávri peat plateau. The edge is taken as the outermost 2 m of the plateau. Only subsidence values greater than 0.05 m are considered in this graph to guarantee a meaningful subsidence signal. Data are derived from the topography difference between 2015 and 2018. The percentages indicate which proportion of the total area (whole plateau or plateau edge) is affected by a subsidence higher than or equal to (in absolute value) a given subsidence value (e.g., 36 % of the edge exhibits a subsidence higher than or equal to 5 cm).
Results from the model simulations are presented in Fig. 8. The temporal
evolution of the peat plateau microtopography shows an edge retreat, while a
large part of the plateau is stable, as is observed in the DEM difference
(Sect. 4.1). The temporal evolution and patterns of simulated edge retreat
show a pronounced dependence on the snow depth on the peat plateau. The
From the evolution of the idealized topography in the three simulations, we
can identify three different types of simulated lateral thermokarst. For
both the
Surface elevation profiles of the peat plateaus as simulated with CryoGrid3 for different snow depths on top of the plateau in time increments of 10 years. Three phases are identified (see text): initial slope adjustment (slope modifications along time), constant edge degradation (slope conserved) and plateau collapse (subsidence over the full plateau). Note that the /textit0 cm snow simulation (not shown) did not produce any changes to the initial topography.
The top panel of Fig. 9 presents the normalized annual volume change. For
the
A comparison of the Šuoššjávri peat plateau lateral thermokarst patterns between field data and simulations is presented in the two bottom panels of Fig. 9. Field values represent average and standard deviation of the field measurements of the measured variables (Sect. 3.3 and Table 1). For each simulation, we average the volume loss and shape index over the ISA, CED and PC phases(Sect. 4.2).
Overall, field-based and simulated volume changes are in a similar range.
The mean field value of
For the HvsV shape index, the initial slope adjustment phases for both the
Based on dGPS measured ground control points, the vertical accuracy of the drone-based DEMs is estimated at 2.6 cm (Sect. 3.1), but shadows, changing cloudiness or strong reflectance contrasts near water bodies can create artifacts in the acquisitions, which locally might cause larger deviations. When comparing elevation differences between two DEMs, vegetation growth, the presence or absence of leaves and water level variations can add noise to the results. To account for these possible flaws when computing elevation differences, we only considered variations higher than 5 cm, which is double that of the mean difference between the elevation of the ground control points (measured with a dGPS) and their counterpart on the DEMs (2.6 cm). This value finds good consistency with values from the literature (Forlani et al., 2018; Jaud et al., 2016). In comparison, our results show that actively degrading zones of the plateau are associated with subsidence values higher than 20 cm that can reach 1 m and more. These values are significantly higher than the 2.6 cm average discrepancy between the DEMs and dGPS measured ground control points, so the DEM accuracy does not affect the volume changes strongly (Table 1). Yet, the evaluation of elevation accuracy derived from this technique will benefit from additional studies confirming these results. Additionally, as described in Sect. 3.2, we acquired the volume changes for the plateau based on an estimation of the elevation of the inflection point of its edge, from which we derived its contour in the 2015 and 2018 DEMs. In case of high vegetation and uneven or gentle slopes, this method to delineate the peat plateau contours can introduce additional uncertainty. However, we carefully checked that this was not the case for the sections analyzed in Table 1.
At the western edge of the Šuoššjávri plateau, subsidence is highly variable, ranging from 0 to more than 1 m within 3 years. This pattern highlights the highly complex and irregular behavior of ice-rich permafrost landscapes (Nitzbon et al., 2019; Osterkamp et al., 2009). When an initial perturbation, for example intense rainfall or above-average snow accumulation, triggers subsidence (Seppälä, 1988, 2011), both the snow redistribution and the subsurface drainage towards the mire are affected, which creates warmer surface conditions and, in return, results in more subsidence. Considering the complex geometry of the Šuoššjávri plateau edges, meter-scale variability of the snow and hydrological conditions likely contribute to observed variability of ground subsidence. Furthermore, heat transfer between the wet mire and the plateau is likely influenced by the geometry of the plateau-mire interface. As an example, zones 1, 2, 4 and 6 belong to convex features of the plateau edges and show particularly high subsidence rates. Finally, the distribution of the excess ice in the ground plays an important role for the timing and magnitude of subsidence. Heterogeneous excess ice distribution throughout the plateau may be an important driver of the observed spatial variability of the edge degradation.
Our results confirm that edge degradation is a major degradation pathway of
peat plateaus with 77 % of the total subsidence occurring within the
outermost 2 m of the Šuoššjávri plateau. This result
shows consistency with Jones et al. (2016) who reported that
85 % of the degradation of forested permafrost plateaus was due to
lateral degradation along the margins. Between 2015 and 2018, we find that
the Šuoššjávri plateau lost 3.2 % of its surface area.
Applying Eq. (A1) (Appendix A5), the aerial change corresponds to an average
annual rate of surface loss of 1.1 % yr
Our modeling framework relies on an idealized geometry and steady-state climate forcing, so the full variety of the observed thermokarst patterns cannot be reproduced. However, the comparison between model results and observations clearly shows that the numerical model framework can capture the correct order of magnitude of the degradation processes, while also reproducing key patterns in the observed ground temperature regime (Appendix A1, Fig. A1).
Among the different degradation phases (initial slope adjustment, constant edge degradation and plateau collapse), the CED phase is most relevant for the comparison to field observations, as it is characterized by steady edge retreat in response to the steady-state climate forcing while the bulk of the peat plateau remains stable. On the other hand, the ISA phase is essentially an adjustment to the change in snow depth conditions from the no snow scenario used for initialization to the scenarios with non-zero snow depth, which are characterized by edge retreat. The PC phase corresponds to the sustained collapse of a plateau with ground subsidence in all parts, which is not observed for the Šuoššjávri peat plateau but regularly occurs for smaller circular palsas in the vicinity. As palsas are often small rounded peat bodies, the assumption of translational symmetry inherent in our model setup (Fig. 5) is not valid. For these features, simulations should be performed for cylindrical symmetry, which better describes the geometry of small palsas (as done in simulations by Aas et al., 2019). This suggests that our simulations are indeed most realistic during the CED phase, whereas changes of the overall geometry of the peat plateau must be taken into account to model the final stages of degradation.
Our simulations confirm the crucial role of snow on the ground thermal
regime and peat plateau degradation. They show that a stability threshold is
crossed between zero (stability) and 10 cm snow depth (lateral thermokarst).
Even though the absolute value of this threshold cannot be generalized due
to our simplistic snow model and the interplay of climatic parameters, it is
broadly consistent with field experiments involving human-made snow clearance in
permafrost-free mire areas in northern Scandinavia, which resulted in the
formation of new palsas (Seppälä, 1982, 1995).
However, it is possible that our simulations slightly overestimate the
sensitivity of edge retreat to snow depth variations, with the true
stability threshold at higher snow depths. While measured March snow depths
in 2015–2018 regularly exceeded 20–30 cm (Fig. 3), our simulations show
higher than measured volume changes for the
Our idealized model approach assumes snow depths to be constant on the entire peat plateau, which does not capture the significant spatial and interannual variability of snow depths on the plateau observed in measurements (Fig. 3). In particular, the complex geometry of snow drift patterns along the plateau edges, with snow drifts forming on lee sides, is not captured by the simple snow redistribution model implemented in CryoGrid3. Field observations show that snow drifts along the plateau edges feature considerably higher snow depths than the surrounding wet mire, thus introducing additional winter warming in the zone of maximum change. Additionally, persistent wind patterns can strongly influence the distribution of snowdrifts. In CryoGrid3, on the other hand, snow removed from the plateau is evenly distributed over the entire mire, not taking edge effects into account.
Furthermore, our model assumes a fixed snow density and thus snow thermal
properties, while snow densities in reality vary with, e.g., snow depth and
time. A density increase from 200 to 300 kg m
In this study, we demonstrate that the snow depth on the plateau exerts a strong control on subsidence patterns. Our experiment shows that snow depth alone can drive important surface temperature changes and permafrost disappearance. This result illustrates that permafrost disappearance is not only a function of temperature (Chadburn et al., 2017) and that plateau systems can react sensitively to different climatic parameters affecting surface temperature. As such, snow precipitation and wind speed variations (which both affect snow pack buildup) should also be regarded as important drivers of the evolution of peat plateaus. In this regard, future precipitation patterns are expected to show an increase of rainfall, partially at the expense of snowfall because of atmospheric warming (Bintanja and Andry, 2017). Yet at the regional level, these changes are highly uncertain (O'Gorman, 2014) but must be taken into account when projecting the future evolution of peat plateaus in the subarctic.
The presented model approach (including excess ice and small-scale
representation of lateral fluxes) is clear evidence of the importance of a
small-scale thaw feedback mechanism on permafrost degradation. The feedback
between the dynamic microtopography and the lateral fluxes of water, heat
and snow shows how a limited increase in snow cover (e.g., from the
Our simple approach is not able to capture the complex patterns of different subsidence rates that are observed around the edges of the plateau (Fig. 6). In addition to small-scale variations of ground stratigraphy, excess ice content and plateau heights, our observations suggest that the irregular plateau outline with both concave and convex shapes affects the lateral fluxes of heat, water and snow, which in turn exert a control on the edge dynamics (Sect. 5.1). While computationally demanding, our multi-tile approach could be embedded in an ensemble framework to represent a range of edge geometries and other critical parameters, yielding a range of different degradation scenarios and therefore capture the high spatial variability of subsidence at the plateau scale. Further sensitivity tests with steady-state climate forcing should focus on the role of air temperature (colder or warmer), total precipitation and excess ice content on peat plateau stability and lateral thermokarst patterns.
In our experiment, the modeling scheme shows a sensitivity of the plateau retreat to different surface temperatures resulting from the different prescribed snow depths. Because other climatic parameters than snow depth can affect surface temperature, this indicates that our scheme may also be able to simulate the plateau response to a temperature increase, paving the way for climate change simulations. Transient simulations should ideally be initialized with a model spin-up for a period during which the peat plateau is stable; otherwise, lateral thermokarst will already occur during the model spin-up phase. For Scandinavia, the ideal period would be the Little Ice Age, when most of the present-day peat plateaus were formed (Kjellman et al., 2018). Future studies should therefore investigate if the simple multi-tile setup can capture changes in peat plateau stability in the transition from the Little Ice Age to the warmer conditions of the 20th century during which peat plateaus in Finnmark likely entered their current state of accelerating degradation (e.g., Borge et al., 2017). Further benchmark simulations with the multi-tile model could focus on peat plateau areas in colder climates that are still stable today. A significant challenge, in particular for model simulations on long timescales (e.g., extending to the Little Ice Age), is to obtain accurate enough model forcing, as biases in the model forcing could shift or even mask climatic thresholds for peat plateau stability.
Most land surface models (LSMs) that simulate the future response of permafrost to climate change rely on simplified 1D implementations of permafrost thaw dynamics, which ignore subsidence and only reflect gradual top-down thawing of the frozen ground (Andresen et al., 2020; Burke et al., 2020). Excess ice melt and the resulting microtopography changes exert a major control on the evolution of hydrologic conditions, which in turn strongly influence the timing of permafrost degradation, as demonstrated for polygonal tundra (Nitzbon et al., 2019, 2020). Aas et al. (2019) presented a similar approach for peat plateaus in northern Norway. It is based on two tiles (one for the wet mire, one for the plateau) and reproduces both climate-induced stability and degradation. However, in this approach, the plateau subsides as a whole when a climate-related threshold is exceeded and excess ice begins to melt. This contrasts with our field observations, which show ongoing edge retreat on decadal timescales, while the plateau interior is largely stable. Our approach uses a larger number of tiles to explicitly represent the temperature and soil moisture gradients across the plateau edges, which causes excess ice melt to only occur in a narrow zone at the plateau edge, in agreement with observations. Over longer timescales, on the other hand, this process leads to the reshaping and finally the complete collapse of the entire peat plateau. In Earth system model frameworks, implementing a multi-tile approach for the land surface scheme is challenging due to its complexity and computational demands. Yet, parameterized approaches could eventually be developed based on sensitivity tests with future generations of higher-complexity multi-tile frameworks (Sect. 5.3.2). In particular, future studies should investigate to what extent the two-tile approach demonstrated by Aas et al. (2019) can emulate the results of a multi-tile model, especially when not only applied to single sites, but to the entire sub-Arctic where peat plateaus occur today. However, our multi-tile setup clearly produces different thaw dynamics at the scale of individual sites, which might affect the modeled carbon balance. To investigate this issue further, a multi-tile model coupled to a carbon cycling scheme would be required.
We present field measurements and numerical modeling of lateral thermokarst
patterns of the Šuoššjávri peat plateau in northern Norway.
We use high-resolution digital elevation models derived from drone-based
photogrammetry to quantify changes of surface elevations of the plateau
between September 2015 and 2018. The study shows that the edges of the peat
plateau are hot spots for thermokarst, where 77 % of the total measured
volume change occurred, while most of the total plateau surface does not show
detectable changes in surface elevation. Lateral thermokarst is therefore
the main pathway for the degradation of the peat plateau. We show that this
retreat corresponds to a normalized annual volumetric loss of
Using the CryoGrid3 land surface model we show that these thermokarst patterns can be reproduced numerically in a framework that implements lateral redistribution of snow, subsurface water and heat, as well as excess ice-melt-triggered subsidence. Overall, the modeled annual volume changes are on the same order of magnitude as the measurements. Based on a steady-state climate forcing, our simulations demonstrate the importance of the shallow snow cover on the plateau due to wind drift of snow to the lower-lying mire areas. The modeled peat plateau is fully stable when all snow on its top is removed towards the mire (0 cm snow depth on the plateau), whereas its edges retreat at increasing rates with increasing snow depths. For the model forcing applied in our simulations, a maximum of 5–10 cm of snow on the plateau only triggers an edge retreat of 4–5 m within 100 years. A snow cover of 10–20 cm depth fully degrades the plateau (assumed 11 m wide) in 100 years, while this time is reduced to 40 years for an even higher snow cover of 20–30 cm.
These results highlight the fast dynamics and high spatial variability of permafrost landscape evolution in response to climate change. They also show that the related changes in microtopography and the thermal and hydrological regime can be represented in numerical models, thus showing a way forward towards substantial improvements in simulating permafrost landscape evolution and its impact on greenhouse gas emissions.
Figure A1 compares the mean annual ground surface temperatures (MAGSTs) and the active layer thickness (ALT), as they are simulated in this study and in Martin et al. (2019) with field measurements from the same study. Note that Martin et al. (2019) used a 1D model, which did not include the possibility to simulate thermokarst processes. Overall, our simulations show good agreement with field measurements. However, they feature a smaller variability than the observations because the variability of the simulations is diagnosed for one idealized peat plateau profile, whereas the variability in the observations is derived from individual points distributed over the plateau, which each feature different overall conditions (e.g., snow cover buildup, drainage regime). As discussed in Sect. 5.3.2 ensembles of simulations exploring different geometries and parameter sets would be required to match the variability of the observations.
While MAGSTs from the present study are in good agreement with the
measurements on the peat plateau, the model underestimates temperatures in
the mire slightly (0.5
Mean annual ground surface temperature (MAGST) and active layer thickness as they are simulated in this study and in Martin et al. (2019) compared to the field measurements from Martin et al. (2019) for the same region. Values indicated with the letter
Geomorphological map of the surroundings of the study site. The white rectangle indicates the study site. Source: Geological Survey of Norway (NGU).
Comparison between the weather station and forcing data.
Comparison between the weather station and forcing data.
To compare with previous studies that only report aerial (and not volume)
changes (Sect. 5.1), we define an average annual rate of surface loss of a
plateau (Eq. A1). The percentage of annual loss of plateau area is
The model code and settings used for the simulations are
permanently deposited at
The data presented in this article are permanently
deposited in the Norwegian research data archive NIRD at
The supplement related to this article is available online at:
LM and SW designed the study and conducted the numerical simulations. LM led the manuscript preparations. SW, ML, JN and LM contributed to the model development. TE, SW, LM and JS acquired field data. TE processed the DEMs. KA provided forcing data. LM and SF analyzed field data. All authors contributed to result interpretation and to manuscript preparation.
The authors declare that they have no conflict of interest.
Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
This work was funded by PERMANOR (Norwegian Research
Council, KLIMAFORSK program, NFR project 255331), Nunataryuk (EU grant
agreement no. 773421), ESA Permafrost_CCI
(
This research has been supported by the Norges Forskningsråd (grant no. 255331) and the Horizon 2020 (grant no. Nunataryuk (773421)).
This paper was edited by Christian Hauck and reviewed by two anonymous referees.