The marine-terminating outlet in Basin 3, Austfonna ice cap, has been accelerating since the mid-1990s. Stepwise multi-annual acceleration associated with seasonal summer speed-up events was observed before the outlet entered the basin-wide surge in autumn 2012. We used multiple numerical models to explore hydrologic activation mechanisms for the surge behaviour. A continuum ice dynamic model was used to invert basal friction coefficient distributions using the control method and observed surface velocity data between April 2012 and July 2014. This has provided input to a discrete element model capable of simulating individual crevasses, with the aim of finding locations where meltwater entered the glacier during the summer and reached the bed. The possible flow paths of surface meltwater reaching the glacier bed as well as those of meltwater produced at the bed were calculated according to the gradient of the hydraulic potential.
The inverted friction coefficients show the “unplugging” of the stagnant ice front and expansion of low-friction regions before the surge reached its peak velocity in January 2013. Crevasse distribution reflects the basal friction pattern to a high degree. The meltwater reaches the bed through the crevasses located above the margins of the subglacial valley and the basal melt that is generated mainly by frictional heating flows either to the fast-flowing units or potentially accumulates in an overdeepened region. Based on these results, the mechanisms facilitated by basal meltwater production, crevasse opening and the routing of meltwater to the bed are discussed for the surge in Basin 3.
The Austfonna ice cap, located on Nordaustlandet in the Svalbard archipelago, is
the largest ice mass in the Eurasian Arctic in terms of area (7800 km
The northern flow unit of the outlet glacier has experienced long-term
acceleration since the mid-1990s (Dowdeswell et al., 1986) along
with stepwise interannual accelerations since 2008. These short-lived
summer speed-up events occurred during the surface melt season
(Dunse et al., 2015). The southern corner of Basin 3 accelerated
to about 290 m a
The 130–140-year-long quiescent phase of Basin 3 is similar to other Svalbard glaciers, but the two-decade-long accelerating phase of the northern flow unit exceeds those of other glaciers such as the 7–11 years of Monacobreen (Strozzi et al., 2002). The stepwise multi-annual acceleration observed since 2008, associated with seasonal summer speed-up events, is also distinguishable from other surging glaciers in Svalbard. Similar melt season speed-up events have been observed in Greenland, and they provide evidence for rapid, large-scale, dynamic responses of the ice sheet to climate warming (Sundal et al., 2011; van de Wal et al., 2008; Zwally et al., 2002). Sundal et al. (2011) pointed out that a simple model of basal lubrication alone could not simulate the fast-flowing manner of the glaciers on Greenland ice sheet, and that an improved understanding of subglacial drainage would be essential in model studies that capture ice dynamic responses to climate warming. This applies also to the surge in Basin 3, which requires a mechanism involving both thermal and hydrologic changes to explain the interannual and seasonal accelerations (e.g. Dunse et al., 2015; Gladstone et al., 2014).
The glacier in Basin 3 (recently named Storisstraumen) is polythermal, with a maximum ice thickness of 567 m, which is sufficient to raise internal ice temperatures to the pressure melting point (pmp) (Dunse et al., 2011). Where the ice is thinner, closer to the margins, the ice is probably frozen to the bed except under fast-flowing outlets. In principle the surge of polythermal glaciers can be explained by a soft-bed mechanism with some constraints for the initiation, such as the unfreezing of the cold bed by the evolution of the thermal regime or by the input of meltwater from englacial channels (Clark, 1976; Lingle and Fatland, 2003; Robin, 1955).
Gladstone et al. (2014) suggested soft-bed sliding mechanisms involving feedbacks in the hydrological system at the ice–till interface. They respond to the penetration of surface melt and explain the summer speed-up events that have been observed since 2008. Surface meltwater can penetrate cold and polythermal glacier ice in High Arctic glaciers and the Greenland ice sheet through moulins and fractures that cut down all the way to the glacier bed (e.g. Benn et al., 2009; Copland et al., 2003; van de Wal et al., 2008; Zwally et al., 2002). Water-filled crevasses can penetrate to the glacier bed regardless of ice thickness or crevasse spacing, as long as the tensile stress acting normal to the crevasse exceeds about 100 kPa (Boon and Sharp, 2003; van der Veen, 1998). Bougamont et al. (2014) investigated the sensitivity of the basal hydrology system in the Russell glacier catchment to the volume of surface melt delivered to the bed and found that increases in surface melt volumes lead to faster summer flow.
Dunse et al. (2015) has suggested a hydrothermodynamic feedback whereby summer meltwater penetrating to the bed is not considered a purely external forcing to the system: meltwater penetrating crevasses to reach the bed enhances basal processes such as lubrication and sediment deformation, resulting in enhanced ice flow and potentially an increase in extensional stress, which may in turn cause increased crevasse formation over a larger area, routing more ice down to the bed.
These earlier studies highlight the importance of time-evolving basal temperature and friction, which are strongly influenced by the evolution of a basal hydrology system. The basal hydrology system can be fed both by in situ melting and by surface meltwater, and has the capacity to not only directly cause sliding but also to alter the thermal regime and hence deformational flow.
Previous crevasse modelling studies simulate the formation of fractures as a continuous process. They treat the development of cracks on a macroscopic scale by either using simplified parameterisation of fracturing effects via variables such as depth of crevasse (Cook et al., 2014; Nick et al., 2010, 2013; Weertman, 1973) or using continuous damage mechanics (CDM), which simulates the continuous process from micro-scale cracks to macro-scale crevasses (Albrecht and Levermann, 2014; Bassis and Ma, 2015; Borstad et al., 2012, 2016; Krug et al., 2014). In this study we take a different approach and apply a discrete element model (Åström et al., 2013, 2014) capable of simulating crevasse formation as a microscopic-scale discrete process in addition to the continuum ice dynamics models. The discrete element model (HiDEM) is used to determine the locations of the crevasses penetrating though the full thickness of the glacier whereby surface water may reach the bed.
In Sect. 2 we present the observational data used to set up the simulation. In Sect. 3 we present the methodology. We used a continuum ice dynamic model to invert the basal friction fields from approximately monthly observations of ice surface velocities between April 2012 and July 2014. These basal friction fields then were input as boundary conditions for basal sliding in our discrete element model, which simulates crevasse distribution in the lower part of Basin 3 at particular points in time. We also converted a satellite image of crevasse distribution to a cartographic map using Radon transform and simplified line integral convolution to validate our modelled crevasse distribution results. In Sect. 4.1 we first investigate the evolution of the basal conditions in Basin 3 during and after the peak of the surge. In Sect. 4.2 we present the modelled crevasse distributions before and after maximum surge velocity and validated the latter with crevasse maps derived from satellite imagery. In Sect. 4.3 we locate the crevasses that reach the bed, calculate basal melt rates and estimate the flow path of the basal water. In Sect. 5 we discuss the mechanisms facilitated by basal meltwater production, surface meltwater and crevasse opening for the surge that occurred in Basin 3. In Sect. 6 we summarise the key findings of the study and present conclusions.
Bedrock elevation (Dunse, 2011) was derived by pointwise subtracting the
measured ice thickness from a 250 m resolution map of surface elevation that
was published by the Norwegian Polar Institute (NPI) in 1998 and InSAR data
of Austfonna acquired in the years 1995–1996 (Unwin and Wingham, 1997). The
ice thickness used for generating bedrock elevation was based on airborne
radio echo sounding (RES; Dowdeswell et al., 1986) supplemented with two RES
data sets from 2008 (Vasilenko et al., 2009). Marine bathymetry data (2 km
horizontal resolution) were from the International Bathymetry Chart of the
Arctic Ocean, version 2.0 (Jakobsson et al., 2008). Bathymetry and inland
bedrock elevation were combined by using an interactive gridding scheme to
eliminate the mismatch along the southern and north-western coastline (Dunse,
2011). We assumed that bedrock elevation did not have any significant changes
over decadal timescales and used it with a set of updated surface elevation
data. The surface elevation was derived from Cryosat altimetry data acquired
during July 2010–December 2012 (McMillan et al., 2014). McMillan et
al. (2014) grouped measurements acquired over a succession of orbit cycles
that are within 2–5 km
We point out several bed topography features (Fig. 1b) that are important to the investigation here. The subglacial hill located at roughly 700 km E and 8850 km N rises to about 250 m above sea level. A corresponding but smaller-magnitude bedrock maximum exists on the opposite side of Basin 3, approx. 15–20 km south-west of the hill. A subglacial valley runs between these bedrock maxima, marked SV in Fig. 1b, and extends several tens of kilometres upstream and downstream. The minimum bedrock height for Basin 3 is within an overdeepening in the lower part of the valley, marked OD on Fig. 1b. The importance of these features is discussed in more detail in Sects. 4 and 5.
Surface and bedrock topography of Basin 3, Austfonna.
We used velocity time series maps (April 2012–July 2014) generated from TerraSAR-X (TSX) satellite synthetic aperture radar (SAR) scenes (Table 1; Schellenberger et al., 2017) as the input surface velocity data for basal friction coefficient inversion. These maps were based on original 2 m resolution TSX scenes provided by the German Aerospace Center (DLR), covering only the lower part of Basin 3 (Fig. 1a). To generate the final velocity maps for the times between two successive TSX images, which were geocoded using a DEM of Austfonna (Moholdt and Kääb, 2012), we needed to use displacement maps. The displacement maps between two consecutive acquisitions were determined using a cross-correlation of the intensity images (Strozzi et al., 2002).
TerraSAR-X acquisitions of Basin 3 and repeat-pass periods.
The coverage of the TSX velocity was smaller than the model domain used by our ice dynamic model (Fig. 1a). Therefore we stitched the TSX data on top of two background velocity fields with larger coverage depending on the acquisition time. The TSX time series maps derived during 19 April 2012–28 December 2012 were stitched with a velocity snapshot from ERS-2 (European Remote Sensing Satellite 2) SAR observations acquired from March to April 2011 (Gladstone et al., 2014; Schäfer et al., 2014), and the TSX time series maps derived after 28 December 2012 were stitched with a velocity snapshot from Landsat 8 imagery acquired in April 2013. We then applied a row-wise recalculation of the velocity value for the grid points on the model mesh that were upstream from the TSX velocity map coverage (Fig. 1a) to create a smoother transition from the TSX velocity map to the background velocity map. The recalculation was carried out by weighting the background velocity data and TSX velocity data according to the distance between the column indices of the targeting grid point and the column indices of the first grid point that had values from the TSX velocity map on the same row.
The velocity recalculated for the upstream area was simply to avoid numerical instability that might appear at the boundary between the TSX and background velocities. So as not to bias the distribution calculation of the crevasses, we confined the discrete element model domain to a smaller region close to the ice front, which was fully covered by TSX velocity map and far away from this transition zone (Fig. 1a).
The continuum ice dynamic model we used is Elmer/Ice, an open-source finite-element model for computational glaciology. In this study, the simulations with Elmer/Ice were carried out by considering a gravity-driven flow of incompressible and non-linearly viscous ice flowing over a rigid bed. Some of the governing equations are presented below. More details can be found in Gagliardini et al. (2013).
The ice flow was computed by solving the unaltered full Stokes equations,
which express the conservation of linear momentum,
The constitutive relation for ice rheology is given by Glen's flow law
(Glen, 1955):
The upper surface,
For all the simulations carried out in this study a linear relation linking
basal shear stress,
We performed inverse modelling of basal friction coefficient distributions
from all the surface velocity observation snapshots using Elmer/Ice based on
the control method (MacAyeal, 1993; Morlighem et al., 2010)
implemented in Elmer/Ice by Gillet-Chaulet et al. (2012). The
inverse modelling determined the spatial distribution of the exponent,
A Tikhonov regularisation term penalising the spatial first derivatives of
As introduced in Sect. 1, ideally, a soft-bed sliding mechanism needs to be
presented in the simulation to be able to capture the surging behaviour.
However, as the main goal of this study is only to find a model approach to
locate the surface meltwater input sources, a linear basal sliding relation
solved with an inverted parameter (
Basal friction coefficient inverted from surface velocity data in
The temperature distribution was calculated according to the general balance
equation of internal energy written as
The Dirichlet boundary condition at the upper surface was prescribed as
An initial guess of the ice temperature,
Spatially varied ice temperature (
All the thermodynamic-coupled inversions were done in chronological order
with a transient simulation after each inversion to evolve the geometry for
the next inversion. A month of geometry evolution was started, with the
All simulations were computed on an unstructured mesh over Basin 3 and generated
with the open-source software GMSH (Geuzaine and Remacle, 2009). The element
size of the mesh increased from
A fixed calving front criterion was adopted in all the simulations in this study due to the lack of ice thickness information corresponding to the observed calving front positions after 2011. The criterion assumed that the calving front was always grounded with a positive height above floatation, which reflected the observation at the terminus in Basin 3. As the near-frontal region was not confined between lateral walls we would not expect a significant impact of different calving front positions on longitudinal stress gradients upstream; i.e. the migration of the calving front would have less of an impact on the basal shear stress distribution in the upstream area than on the uncertainties brought by the observed ice velocity or the lack of ice thickness information at the calving front. The fixed calving front criterion would not distort the basal shear stress calculation at the ice terminus either, as the basal resistance there was already low in 2012. However, as the ice front in the simulation did not migrate, the calving flux might be biased.
HiDEM is a model for fracture formation and dynamics. In HiDEM, an ice body is divided into discrete particles connected by massless beams. The version of HiDEM used here is purely elastic rather than viscoelastic (Åström et al., 2013). The elastic version is sufficient for the purpose of locating fractures governed by glacier geometry and basal friction. If the initial state of a model glacier is out of elastic equilibrium, deformation within the ice will appear as a result of Newtonian dynamics.
The explicit scheme for simulating the Newtonian dynamics and the elastic
modulus can be found in Riikilä et al. (2015). We use
a Young's modulus
All the simulations in this study were carried out with 30 m spatial
resolution (the particles are uniformly shaped and initially uniformly
spaced). We used a time step length of 10
The simulations were set up with input data from marine bathymetry, bedrock
topography,
We created a crevasse map from satellite imagery to validate our modelled crevasse distribution. The map was generated from a Landsat 8 image acquired on 4 August 2013 using the Radon transform technique (Petrou and Kadyrov, 2004; Toft and Sørensen, 1996). We experimented with crevasse maps created from various different satellite sensors (Landsat 7, Landsat 8, ASTER, Sentinel-2), but here we used only the Landsat 8 scene, which combines good spatial coverage with high radiometric quality.
The crevasse maps created from Radon transform.
The Radon transform has been demonstrated to be efficient in detecting along-flow features (Roberts et al., 2013), but can also be used for complex flow patterns, like the one in Basin 3 which has a wide range of crevasse orientations. The advantages of the Radon transform over other detecting methods are that crevasse patterns can be extracted where edge detectors methods (Bhardwaj et al., 2015; Wesche et al., 2013) would fail, and also that it is more robust than frequency-domain methods (Sangwine and Thornton, 1998) in detecting crevasses from incomplete coverage due to clouds, image borders or the calving front.
In this study we followed a similar approach to Roberts et
al. (2013), but used a more robust implementation and a different
post-processing procedure. Firstly, the satellite image was pre-processed
with a Laplacian filter to prioritise the high frequencies, e.g. to sharpen
the edges of surface features. We performed the Radon transform,
Figure 2 shows that the relative errors between the modelled and observed
surface velocity magnitudes during both 18–29 August 2012 and 16–27 August 2013
are the lowest over the fast-flowing region (
Figure 4 shows the friction pattern of the region that is fully covered by
TSX velocity observations between April 2012 and July 2014, spanning the
period of the Basin 3 peak surge velocities in January 2013. To make the
pattern of the
In 2011 the low-friction patches in the central and southern basins were
disconnected from the inland region and lay behind a stagnant terminus.
Before May 2012, the enlarged low-friction area in both northern and southern
glacier termini did not expand across the flat glacier bed in between them,
which might impose some topographic restriction to the expansion of the fast
flow. After the summer melt season (August 2012), the stagnant frontal region
shrank to the glacial terminus, which might have thinned to reach a condition
close to floatation (McMillan et al., 2014). During this period the
low-friction area underneath the southern part expanded further inland and
became connected to the northern low-friction area. In January 2013 the
glacier reached its maximum flow speed
and the low-friction area also expanded across the entire width of the basin
near the calving front with a few particularly deep minima (log
After January 2013 the basal friction pattern in northern basin remained
almost stable. The almost vanishing friction area (log
All the fractures calculated by HiDEM are wider than 0.055 m, and we regard them as crevasses in this study. The fractures marked with black dots (Fig. 5b, in both the upper-left and lower-left corners of the domain) are generated by boundary effects due to the limited domain. Although they might be deep enough to cut through the full depth of the ice we regard them as artificial crevasses. They are irrelevant to the water routing and surge processes we focus on in this paper, thus are excluded from the comparison in this section and the water routing calculation in Sect. 4.3.
The evolution of the basal friction coefficient (
Crevasse distribution from HiDEM on
The crevasse map created by the Radon transform shows a highly crevassed glacial lower region, which comprise sections with crevasses of different orientation (Fig. 3). Transverse crevasses that are almost perpendicular to the flow direction can be found in both northern and southern flow units (Fig. 1b), reflecting large longitudinal tensile stress after dramatic acceleration. However, the detection intensity of the crevasse in the northern flow unit is rather weak. The terminus between the northern and middle flow units has a mixture of crevasses with orientations perpendicular to each other, indicating the expansion and merging of the two flow units. Marginal crevasses can be found above the subglacial valley (Fig. 1b) margins parallel to the local flow direction, reflecting lateral shear stresses and longitudinal compressive stress caused by the presence of the valley margins.
The modelled crevasse distribution reflects the broad features of the basal
friction pattern (Fig. 5b). A high crevasse density is generated in areas
with large tensile stress caused by extending flow on the lower part of Basin
3 as well as at shear margins between low- and high-friction areas. The
orientation of the modelled crevasses in August 2013 above the subglacial valley margins agrees with the representation
map of the observation (Fig. 5c). However, the
orientations of most of the modelled crevasses in the middle-upper area have
a
Although the visual comparison between the two maps shows a general agreement
(Fig. 5c), estimations of statistical quality of the simulated crevasse field
with the observationally derived map are necessary. We calculated the Kappa
coefficient (
To investigate the crevasse distribution after the summer melt season in
2012, we used
By looking at the overall crevasse distributions in August 2012 and August 2013
(Fig. 5a and b), together with their corresponding
We defined “cut-through crevasses” as crevasses that penetrate through two-thirds of the ice depth and assume that they could cut through the full depth of ice if filled with water and potentially route the surface meltwater into the basal hydrology system vertically.
We selected the cut-through crevasses in August 2012 and August 2013 (red dots in Fig. 5a and b). In August 2012 most of the crevasses in the frontal area cut deep enough into the ice and very likely represented future calving locations for the terminus during its advance. Most of the crevasses located between the northern and southern fast-flowing regions were shallow surface crevasses. Many crevasses above the margins of the subglacial valley could reach the bed and potentially route the surface meltwater from upstream to the bed. By August 2013 more cut-through crevasses had been developed in the lower southern and central basins compared with August 2012 as velocity gradients significantly increased after the basin-wide acceleration. There were more cut-through crevasses present above the shear margin but almost no cut-through crevasses above the overdeepened area.
Using the locations of cut-through crevasses above the margins of the
subglacial valley that could potentially route the surface meltwater down to
the bed in August 2012, we calculated the subglacial water flow path
according to the gradient of the hydraulic potential (Fig. 6a). The
hydraulic potential (
The flow paths of different water sources derived from the results
in August 2012.
The flow paths are generated by being integrated through the vector field that
follows the steepest descent in
The surface meltwater entering the bed in the north was predicted to either flow directly to the terminus or stop at the subglacial overdeepened area (Sect. 2.1; Fig. 6a). The surface meltwater entering the bed from the south was routed directly towards the terminus at the southern corner of the glacier, suggesting that surface melt contributed to the dramatic acceleration of the southern part of Basin 3 after the summer melt season in 2012.
In addition to the basal water supplied via the crevasse system, we also
estimated the basal melt rate (Fig. 6b) for the temperate base area of the
glacier. Within Elmer/Ice we computed the energy balance at the bed from
estimated geothermal heat flux, strain heating and basal friction heating
(Gong et al., 2016). Relatively high basal melt rates
(
Previous studies of the surge in Basin 3 (Dunse et al., 2012, 2015; Gladstone et al., 2014) revealed an atypical surge activation phase with multi-decadal acceleration superimposed, for at least 6 years, by short-lived, abrupt seasonal speed-up events that were clearly related to summer melt, which could not be explained solely by the thermal switch mechanism (Murray et al., 2003) typical of polythermal surging glaciers in Svalbard.
We used the discrete element model, HiDEM (Åström et al., 2014), to locate crevasses. In general the modelled crevasse distribution in August 2013 matches the crevasse map derived from satellite observation. However, there is a mismatch of the orientation in the middle-upper area (Fig. 5c). It may be because HiDEM only simulates the ad-hoc formation but not the advection of crevasses; thus no crevasse formation history can be inferred from the model. The inclusion of crevasse advection could be implemented in a two-way coupling of HiDEM with a continuum model accounting for damage transport in future studies. The mismatch of the crevasse density (Fig. 5c) at the northern and southern frontal areas could be caused by a mismatch in the ice front position between reality and the model. Although in reality the ice front advanced for several kilometres after the full surge, it was kept in a fixed position in Elmer/Ice (Sect. 3.1). The shape and steepness of the ice front are likely to affect the behaviour of the discrete element model. However, as they are concentrated at the terminus of the glacier, these crevasses are less likely to affect the basal hydrology system on a larger scale.
We then selected the modelled crevasses in August 2012 that might penetrate the ice far enough to act as routing paths from the glacier surface to the bed. In this study we focused on the cut-through crevasses that formed above the margins of the subglacial valley because the basal flow pattern of the surface melt entering through those crevasses was indicative of potential subglacial water routing and hydrology.
We cannot directly simulate or quantify the effects of the surface meltwater or basal meltwater on the surge development due to the lack of a basal effective pressure-dependent sliding relation. However, based on our results we can still present arguments to emphasise the role of crevasses, summer melt and the basal hydrology system in the seasonal speed-up events.
Firstly, our calculation of the flow paths of both surface melt entering through the crevasses and basal meltwater production suggest the potential for a direct lubricating effect acting beneath the northern and southern fast-flow units. Figure 6 shows that water entering through the crevasses downstream from the subglacial hill (the flow paths in the northern half of Fig. 6a) will flow through the area where the northern fast ice flow unit has developed. The water accessing the bed at the southern part of the basin travels directly towards the terminus at the southern corner of the glaciated system, which has dramatically accelerated during the melt season in 2012.
Secondly, some of the basal water flow paths presented in Fig. 6a and b terminate under a plateau in the hydraulic potential, which occurs in the overdeepened region (see also Fig. 1b). Given the very low gradients of our calculated hydraulic potential in this region and the presence of a local hydraulic potential minimum slightly downstream of the overdeepening, basal water would likely have low flow speeds and possibly even accumulate in the overdeepened bedrock region over time. This may have impacted the surge development in Basin 3. Also, given that the lowest basal resistance during most of 2012 (Fig. 4a) was immediately downstream of the overdeepening area in the northern flow unit, the outflow of accumulated water likely enhanced the surge activation here. If seasonal surface ice accumulates here and drains over a longer period, this may explain prolonged high ice velocities after the melt season has ended.
The temporary speed-up of the southern flow unit in 2008 (Gladstone et al., 2014) could have been triggered by an influx of basal water that was not repeated again until the basin-wide surge was initiated. An outburst of basal water that accumulated in the overdeepened bedrock region could provide one mechanism for this to occur. A ridge in hydraulic potential divides the northern and southern flow units in August 2012 (Fig. 6a). An anomalously high inflow of surface meltwater could have caused this ridge to be flooded if regular drainage channels were of insufficient capacity. We are unable to say how likely this is without a time series of surface melt data including the 2007 and 2008 seasons, but such an event could cause a temporary speed-up in the southern flow unit.
Englacial channels which may cause a redistribution of water within the hydrologic system (Fountain and Walder, 1998) are not directly considered in the current study. We assume that direct transfer of surface run-off via cut-through crevasses exceeds the englacial water transport at Basin 3.
Lastly, we look at the role of basal meltwater in the activation of the southern flow unit. Basal meltwater from further upstream in the northern flow unit can drain toward the southern unit (Fig. 6b; prior to the basin-wide surge, nearly all of the ice drained toward the northern flow unit). If this basal meltwater accumulated upstream due to the lower part of the glacier being below pressure melting point, the accumulated basal meltwater could have caused the speed-up once basal temperatures reached melting point under the southern corner and the hydrologic system extended beneath the southern flow unit. Also, basal meltwater generated locally in the overdeepened area (Fig. 6b) may not have been able to drain completely in one season; thus it could be accumulated locally. However, whether the basal meltwater can eventually burst out from the overdeepening area and contribute to the seasonal speed-up events or refreeze locally also depends on the development of the hydrology system and the thermal regime.
Although we lack either simulated or observed surface melt volumes for summer 2012, we would expect that the surface melt is much larger than basal melt. The run-off output from the HIRHAM5 regional climate model in 1995 (Ruth Mottram, Danish Meteorological Institute, personal communication, 2014; 1995 was not a year with high surface melt) at the location of the cut-through crevasses was at least 10 time larger than the basal melt rate in either 1995 or 2012. The volume of surface melt observed at weather stations located in south-western Basin 3 after the summer of 2004 was also at least 10 times larger (Schuler et al., 2007). Considering the seasonal timings and magnitudes of speed-up events, and the feedback between the surface meltwater input and hydraulic warming at the bed, it is clear that surface melt, when it can penetrate to the bed, causes a much greater impact on sliding and ice dynamics than the basal meltwater.
Then, we also discuss the role of the crevasse formation in the long-term acceleration. These are initiated as a consequence of extensional flow resulting from changes in the basal thermal structure in an early post-quiescent phase and act as the triggering and enhancing factor in the annual hydrothermodynamic feedback proposed by Dunse et al. (2015). While Dunse et al. (2015) are unspecific as to the cause of hydrothermodynamic initiation zone in the long-term glacier acceleration, we propose that the basal meltwater resulting from the gradual thickening of ice (raising basal temperatures) during the quiescent phase could sufficiently enhance flow speeds to initiate cut-through crevassing. The basal temperature distribution inversely calculated from the glacier geometry and velocity (Gong et al., 2016) showed a partially temperate bed in 1995 and expansion of the temperate region from 1995 to 2011, which is consistent with the presence of water at the bed. Given that basal meltwater fluxes are likely to be at least an order of magnitude lower than surface meltwater or run-off fluxes, it probably has a relatively small influence on glacier sliding. We suggest that water at the bed, which is likely to be primarily routed toward the northern rather than the southern flow unit due to topographic constraints (Fig. 6b), caused the speed-up from the quiescent phase during the last part of the 20th century and early 21st century.
This would require two key developments from quiescent to surge phase. Firstly, the initiation of sliding after ice thickening provided sufficient insulation for the bed to reach pressure melting temperature and generate meltwater. This could have occurred during the early 1990s. Then, at some point before August 2012, extensional flow due to sliding became sufficient to cause cut-through crevasses, leading to further acceleration and the surge onset due to the annual hydrothermodynamic feedback. We have demonstrated that cut-through crevasses are likely to be present just prior to the surge in Basin 3, and that surface meltwater can flow along the paths corresponding to the regions of observed fast flow.
It is not clear at which point the hydrothermodynamic feedback cut in, though it is likely to have first occurred in the northern flow unit, due to this unit's earlier acceleration. We suggest that the hydrothermodynamic feedback cut into the southern unit in 2011 or early 2012 due to crevasses penetrating near the southern margin (Fig. 5a), rapidly causing the basin-wide surge.
Direct verification of the long-term evolution of the surge active phase discussed above cannot be provided without quantification of the water reaching the bed and a basal sliding relation engaging the basal effective pressure. However, our approach and results can throw some light on future studies of coupled ice dynamic–thermodynamic–hydrology simulations.
We have forced the discrete element model HiDEM with outputs from the continuum ice dynamic model Elmer/Ice to study the crevasse distribution during the surge in Basin 3, the Austfonna ice cap in the period 2012–2013. Our continuum to discrete multimodel approach provides simulated locations where cut-through crevasses allow the surface meltwater to be routed to the bed. We have demonstrated that automatic crevasse detection through Radon transform may be used to validate simulated crevasse distribution from our continuum–discrete modelling approach. With the future addition of a basal hydrology model, the current study constitutes a step towards a fully coupled ice-dynamic englacial–basal hydrology modelling system in which both input locations of input surface water and basal meltwater generation are represented.
Our results support the hydrothermodynamic feedback to summer melt proposed by Dunse et al. (2015) to explain the seasonal speed-up in Basin 3 and the initiation of the acceleration of the southern flow unit in 2012. The calculated flow paths of the basal water according to hydraulic potential indicate either a direct enhancement to the ice flow through basal lubrication or a lagging mechanism through the outflow of accumulated water in the overdeepened area.
We propose that basal meltwater production caused the speed up from the quiescent phase of Basin 3 during the last part of the 20th century and early 21st century. Then, the hydrothermodynamic feedback initiated during 2011 or early 2012 caused the activation of the southern flow unit and the expansion of the surge across the entire basin. The quantification of the roles and mechanisms involving basal meltwater production, the surface meltwater and crevasse opening for the surge discussed in this study need to be further improved by coupling basal hydrology with the thermal regime evolution and surface mass and energy balance.
This publication also contains Matlab® codes for generating crevasse maps from satellite images (Sect. 3.3) as a supplement.
All data sets used are publicly available. Bedrock elevation
data are available from Dunse et al. (2011). Surface elevation data are
available from McMillan et al. (2014). Surface velocity data from TerraSAR-X
are available from Schellenberger et al. (2017). Velocity data from ERS-2 SAR
are available on request from Tazio Strozzi. Velocity data from Landsat 8 are
available on request from Thomas Schellenberger. HIRHAM5 surface mass balance
data of Austfonna are available on request from Ruth Mottram. Landsat 8
imagery for generating the crevasses map can be downloaded from
The supplement related to this article is available online at:
YG and TZ designed the numerical experiments and carried out the simulation in Elmer/Ice. JÅ carried out the simulations in HiDEM. BA produced the crevasse map with Radon transform. TS processed and produced the TSX velocity time series. YG analysed the model results and designed the figures. YG wrote the manuscript together with RG, JM and TZ. All the authors assisted in data interpretation and commented on/edited the paper.
The authors declare that they have no conflict of interest.
We wish to thank all partners for providing data and constructive discussion during the study, especially Ruth Mottram from the Danish Meteorological Institute for the HIRHAM5 surface mass balance; Tazio Strozzi from GAMMA Remote Sensing and Consulting AG for the ERS-2 SAR surface velocity observation acquired in March to April 2011; Malcolm McMillan from the University of Leeds for the surface elevation derived from Cryosat altimetry data acquired during July 2010–December 2012; Thorben Dunse from the University of Oslo for the bedrock and ice thickness data and Andreas Kääb from the University of Oslo for TSX data funded by the German Aerospace Center DLR (LAN_0211). We wish to thank Fabien Gillet-Chaulet from the Insitut de Géosciences de l'Environnement for making his code available for the inverse modelling. We also acknowledge CSC – IT Center for Science Ltd., Espoo, Finland, for the allocation of computational resources. The work was supported by the Finnish Academy project 286587: Simulating Antarctic marine ice sheet stability and multi-century contributions to sea level rise. Thomas Zwinger was supported by the Nordic Center of Excellence eSTICC (eScience Tools for Investigating Climate Change in Northern High Latitudes) funded by Nordforsk (grant 57001). Bas Altena was funded by the European Research Council under the European Union's Seventh Framework Programme grant agreement no. 320816. Thomas Schellenberger was funded by the Research Council of Norway (RASTAR, 208013), the Norwegian Space Centre as part of European Space Agency's PRODEX programme (C4000106033), and the European Union FP7 ERC project ICEMASS (320816). Edited by: Eric Larour Reviewed by: two anonymous referees