Rapid Basal Channel Growth Beneath Greenland's Longest Floating Ice Shelf
Abstract
Nioghalvfjerdsfjorden Glacier (N79) is one of the two main outlets for Greenland's largest ice stream, the Northeast Greenland Ice Stream, and is the more stable of the two, with no calving front retreat expected in the near future. Using a novel surface elevation reconstruction approach combining digital elevation models and laser altimetry, previously undetected local phenomena are identified complicating this assessment. N79 is found to have a complex network of basal channels that were largely stable between 1978 and 2012. Since then, an along-flow central basal channel has been growing rapidly, likely due to increased runoff and ocean temperatures. This incision threatens to decouple the glacier's northwestern and southeastern halves.
Key Points
-
We created a novel ice surface elevation reconstruction with annual change rates by fusing altimetry and digital elevation models
-
A rapidly growing basal channel is identified near the grounding line of Nioghalvfjerdsfjorden glacier, with implications for stability
-
We believe this channel growth to be the result of warming ocean water and increased runoff leading to more intense meltwater plume activity
Plain Language Summary
Nioghalvfjerdsfjorden Glacier (N79) has one of the longest floating ice tongues in Greenland and is one of two outlets for the island's longest ice stream. While many of Greenlands's outlet glaciers have been retreating due to climate change, it was believed that N79 would remain relatively stable. By combining multiple data sources, we have created an improved reconstruction of the glacier, revealing previously overlooked features which may threaten that stability. Most notably, a large, rapidly growing along-flow channel was identified at the bottom of the ice shelf near the grounding line, which threatens to cut completely through the glacier. We attribute this behavior to the ice bottom topography and pre-existing patterns of stress in the ice interacting with warming ocean water and increasing meltwater discharge, focusing the melting of the ice tongue in specific locations.
1 Introduction
The Northeast Greenland Ice Stream (NEGIS) is the largest ice stream of the Greenland Ice Sheet, draining approximately 12% of its surface area and containing a sea-level rise equivalent of 1.1 m (Mouginot et al., 2015). Its discharge is routed through two major outlet glaciers, ZachariæIsstrøm (ZI) and Nioghalvfjerdsfjorden Glacier (N79; Figure 1), the former of which has demonstrated a pattern of accelerating mass loss (Mouginot, Rignot, Bjørk, et al., 2019). N79 features a ∼70 km long floating ice tongue which provides a substantial buttressing effect to this branch of the ice stream (Mayer et al., 2018). At the end of its containing fjord, the ice tongue is anchored by a series of islands (An et al., 2021; Morlighem et al., 2017).
(a) Map of the study area showing calving-front and grounding-line locations in 1978 and 2020, and the locations of transects and Surface Elevation Reconstruction and Change Detection times series from this and later figures. The red box indicates the area of panels (d)–(f), and the black box panel (b), Figures 2–4 and S5–7. (b) Interpretation of observable surface features with reference designations, and, for flow-perpendicular channels, year of formation. (c) Ice-penetrating radar profile from 03 April 2017, showing the central region of steep thickness gradient (C2) and the (forked) southern cavity (S1, S2, see also in panel e). (d) 1978 shaded relief ice bottom reconstruction, showing the buoyancy-inferred bottom depth for floating ice, and bedrock elevation elsewhere. Also shows the location of a bedrock canyon (white arrow). (e, f) Similar reconstructions for 2017 and 2020. Satellite images and aerial photographs shown in the figures are listed in Table S2 in Supporting Information S1 and are chosen for temporal proximity.
Possibly due to this configuration, N79 experienced only minor changes in calving front and grounding line position throughout our observational period of 1978–2020 (Figure 1a, S1). Moreover, modeling studies suggest that its grounding line and calving front are unlikely to change significantly over the next century (Choi et al., 2017). However, cosmogenic exposure and radiocarbon dating of surrounding rocks indicate that both N79 and ZI have retreated inland from their current extents during the Holocene (Larsen et al., 2018), indicating N79 may be more sensitive to climate change than previously thought. Indeed, velocity measurements show that, while not as dramatically as ZI, N79 has accelerated in the vicinity of its grounding line (∼10%) (Mouginot et al., 2015). Moreover, the average ocean temperature is increasing, and the depth to the upper interface of warm Atlantic Intermediate Water (AIW) is decreasing in the open sea beyond the calving front (Schaffer et al., 2020). Evidence of AIW has even been found in a marginal surface lake (Blåsø, Figure 1a) near the N79 grounding line (Bentley et al., 2022). This coincides with other glaciers in northern Greenland seeing an onset of widespread calving events over the last decade (Ochwat et al., 2022).
Laser altimetry time series indicate a dynamic thinning of less than 0.5 m annually from 1999 to 2009 on N79 (Csatho et al., 2014), with longer time series revealing an increased rate of thinning after 2012 near the center of the glacier close to the grounding line (Narkevic et al., 2020). However, this accelerated thinning is only observed in a few locations because of the sparse spatial coverage of airborne altimetry between 2009 and 2018 (Figure S2 in Supporting Information S1). This obscured whether accelerated thinning was a localized phenomenon or a more widespread trend that largely evaded the altimetry flight lines. This has significant implications, as reconstructions based on altimetry (e.g., Khan et al., 2022) make broad conclusions about N79 and other glaciers based on these sparse data. Even localized but growing channels, say, at the base of an ice shelf can indicate vulnerability to mechanical break-up (Alley et al., 2023; Rignot & Steffen, 2008).
Our understanding can be improved by including digital elevation models (DEMs), which have a much denser spatial distribution of elevation data, albeit at the cost of poorer precision than altimetry. Using altimetry as control data for correcting any systematic error present in DEMs from multiple years can produce an ice surface elevation reconstruction with high spatiotemporal resolution and accuracy. Here we present a novel elevation reconstruction of the N79 grounding region using such a technique.
2 Methods
Repeat coverage of WorldView (WV) stereo satellite imagery since 2011 enables the determination of ice surface elevation changes with high spatial resolution and accuracy (Porter et al., 2022; Shean et al., 2019). We used ArcticDEM strips, generated from WV images using the Surface Extraction with TIN-based Search-space Minimization (SETSM) approach (Noh & Howat, 2015), to reconstruct elevation change in the N79 region from 2012 to 2020. These DEMs, calculated using satellite ephemeris information without ground control, have vertical errors on the order of 4 m (Porter et al., 2022), which is unsuitable for precise change detection and investigating ice dynamic processes of outlet glaciers. We developed a correction algorithm, based on the approach of Schenk et al. (2014) to reduce this error.
Altimetry time series, serving as control, were generated from Airborne Topographic Mapper (ATM) (1993–2019), ICESat (2003–2009), and ICESat-2 (2018-present) data using the Surface Elevation Reconstruction and Change Detection (SERAC) method (Schenk & Csatho, 2012). A spline-based approximation algorithm (Shekhar et al., 2021) infers the elevation at the date of the DEM acquisition for each time series, and a polynomial correction surface is fitted to the resulting residuals for a given DEM. Once added to the DEM, the error is reduced, and separate DEMs can be mosaicked together with minimal edge discontinuity and a final uncertainty on the order of ∼1 m (Text S1 in Supporting Information S1). The pipeline also accounts for tidal flexure and the inverse barometric effect on floating ice (Text S2 in Supporting Information S1). In this manner, ice surface DEMs, covering the N79 grounding line region, are created for 2012, 2014–2017, and 2020, with nominal dates in the spring to early summer and 30 m horizontal resolution (Table S1 in Supporting Information S1). The Greenland Ice Mapping Project (Howat et al., 2014) surface DEM is used outside the spatial extent of the corrected DEMs. A DEM generated from 1978 stereo aerial photographs (Korsgaard et al., 2016) is used for long-term comparison. Landsat and Sentinel imagery is used for qualitative assessment of surface features (Table S2 in Supporting Information S1). Demarcated surface features are digitized from satellite images in conjunction with final DEMs.
The 2012–2020 gridded surface elevation reconstructions form the basis of several other data sets. Eulerian (static reference frame) annual elevation change is calculated as the direct difference between surface heights for consecutive years. Furthermore, a hydrostatic assumption was used to infer the depth to the ice shelf base for each year, and a bathymetry model from An et al. (2021) was used to estimate grounding line location (Text S3 in Supporting Information S1). The accuracy of the derived ice bottom elevations is assessed by comparing them with airborne ice-penetrating radar (IPR) returns (CReSIS, 2020) and estimated to have an error of 9.4 m (Text S3 in Supporting Information S1). Subglacial drainage patterns are inferred for each year from the surface and bed DEMs using the MatLab Topo Toolbox (Schwanghart & Nikolaus, 2010), and tested for robustness using a Monte Carlo analysis as described in Narkevic (2021). Using this reconstructed subglacial routing to demarcate a subglacial drainage basin for N79, annual aggregate runoff is estimated using values from the Regional Atmospheric Climate Model v2.3p2 (Noël et al., 2018), assuming all runoff reaches the bed immediately. Basal melt is neglected, as runoff is sufficient to demonstrate an increased melt flux.
Velocities for the period of interest, derived from a combination of radar and optical images using feature tracking and interferometry, are from Mouginot, Rignot, Scheuchl, et al. (2019). These are summer-to-spring annual averages from 2012 to 2017. From these, the surface strain rate components are derived per van der Veen (2013) and used as a proxy for surface stresses, and have an estimated uncertainty of ∼0.01 yr−1 based on the velocity uncertainty. Eulerian change rates on floating ice are complicated by the advection of large features, so velocity are used to reconstruct Lagrangian (moving reference frame) elevation change in the manner of Shean et al. (2019)., that is, taking the difference between elevation at an initial pixel, and the pixel to which that ice parcel would have advected by the subsequent DEM date (Text S4 in Supporting Information S1).
Finally, to investigate the propagation of dynamic thinning to the grounded ice, SERAC time series derived from altimetry were partitioned into components due to surface processes as estimated by the IMAU-FDM v1.2G firn densification model (Brils et al., 2022) and ice dynamics, as per (Csatho et al., 2014; Shekhar et al., 2021).
3 Results and Interpretation
3.1 Morphology of the N79 Ice Shelf
The floating ice shelf of N79 exhibits a lateral dichotomy. The northwestern ice shelf (NWIS) gradually becomes thinner with distance from the grounding line and has a relatively uniform pattern of crevasses, while the southeastern ice shelf (SEIS) is characterized by larger, sparser flow-perpendicular channels (P1, P2, etc. in Figure 1b) separated by ∼5–10 km with surface bulges between. The bulges create an across-flow step-wise thickness discontinuity of ∼100 m at the center of the floating tongue (Reeh et al., 2000; Figure 1c). This dichotomy remains visible within ∼40 km of the grounding line, beyond which the ice shelf appears more uniform. There are also three major along-flow channels or cavities, one at each margin, and one in the center. The central channel is less continuous than the others, consisting of segments (e.g., C1, C2) extending downstream into the northwest end of the flow-perpendicular channels. Around 2000, a second band of channels with a more oblique orientation appeared closer to the margin (O1–O3) on top of the SEIS marginal cavity, causing a fork in the cavity (S1, S2). Overall, from 1978 to 2020, the ice shelf became thinner, with more intense and complex channelization, with channels reaching the grounding line by 2020 (Figures 1d–1f, S5).
The flow-perpendicular channels in SEIS are not necessarily analogous to the crevasses in NWIS. Near the grounding line, one can observe “ripples” in the ice shelf that first appear angled upstream toward the center (Figure 1b), possibly representing nascent flow-perpendicular channels. These are reminiscent of the basal channel pattern predicted for laterally heterogenous ice tongues under no-slip conditions by Sergienko (2013). The flow-perpendicular channels generally rotate until perpendicular to flow, and ultimately become associated with new segments of the central channel, forming a hook shape, and developing a complex morphology with internal ridges (Figure 2b). Yet the flow-perpendicular features predicted by Sergienko differ from our observations: they are generally uniform in size and frequency and lack the complex morphology observed in the perpendicular channels on the surface (Figure 2b). If the ripples are a natural consequence of melt patterns for a heterogenous ice shelf, their growth into larger channels implies the influence of other factors. If in hydrostatic equilibrium, these ridges would correspond to subglacial keels, otherwise they may be uncompensated compressional features. There are no radar flights spanning the flow-perpendicular channels to indicate which is the case. Over time (i.e., with distance from the grounding line), the flow-perpendicular channels tend to become narrower along-flow and more subdued in vertical relief. Considering all this, we may conjecture that channelization is initiated by lateral topographic heterogeneity leading to heterogenous melt distribution; then other factors periodically exaggerate and then compress some of the perpendicular features.
(a) Shaded relief 2017 ice surface elevation with grounding line and transect locations. (b) Surface elevation profiles across flow-perpendicular channels, starting at the upstream side, illustrating complex morphology. Dotted lines show elevations from the 30 m resolution DEM mosaic and solid lines after smoothing with a 600-m Gaussian kernel to emphasize surface topography reflecting basal channels. (c) Along and (d) Across-flow profiles of ice surface and bottom elevation showing central channel growth (c, d), grounding line retreat (c), and thinning in the shear zones (d). Shear zone extent is inferred from shear strain rates (Figure 4D) and 2017 grounding line flexure zone is from (ESA, 2017).
Around 2012, two flow-perpendicular channels (P1, P2) emerged ∼10 km downstream of the grounding line in close proximity, with P1 not fully rotating into flow-perpendicular position in subsequent years (Figure 1b). The connected channel C1 has since grown in both depth and length, thinning the ice in its location at a prodigious rate, reaching nearly 100 myr−1 between 2017 and 2020 at the intersection of the transects in Figures 2c and 2d. This thinning is nearly twice the 50 myr−1 2011–2015 estimated steady-state melt rate near the grounding line (Wilson et al., 2017).
3.2 Grounding Zone Changes
According to SERAC time series and DEM elevation change, the effects of rapid thinning were detectable upstream of the grounding line by ∼2015 (Figures 3a and 4c), indicating that thinning of C1 had propagated from the ice shelf to grounded ice (Figure 4c, S6). By 2020 the hydrostatically-inferred grounding line had experienced significant local retreat upstream of C1 and the SEIS marginal cavity (Figures 2a and 2c). This effect was sufficient to shift the subglacial drainage patterns in the area. The potential drainage pathways from the reconstruction indicate three major outlets into the fjord: two corresponding to the marginal cavities and one that, before 2016, entered the fjord ∼1 km southeast of C1. By 2016 the channelized thinning shifted this pathway directly into C1 (Figure 3a, S5). This would mean the subglacial drainage network and basal ice shelf channels had become connected, directly linking C1 to a bedrock canyon (Figure 1d) that likely delivers much of the subglacial meltwater that reaches the grounding line (Figure S8b in Supporting Information S1). The ensuing inferred grounding line retreat then proceeded along the central and southern subglacial drainage paths. These results, however, come with the caveat that bed elevation in this area is uncertain and cannot be strongly claimed without additional evidence such as IPR.
(a) Surface elevation change rate from 2016 to 17 showing the Eulerian difference on grounded ice and the Lagrangian difference on floating ice, with inferred major subglacial drainage pathways. Insets show the mean annual elevation change and drainage pathways for other years. Arrows indicate where the drainage outlet shifts. There is no 2017–20 Lagrangian change due to the lack of velocity. (b) Along-flow radar returns near the center line from 29 April 2014. (c) Radar returns from the same flight path on 3 April 2017. The yellow arrow shows the area of grounded ice change, and the red arrow shows the side echo of C1.
While there are no radar flights over the basal channel system, inferences supporting its rapid thinning and corresponding grounding line retreat can be made from 2014 to 2017 along-flow IPR repeat transects, which are slightly southeast of and parallel to C1 (Figures 3b and 3c). In the grounding zone (7,500–13,000 m along-track) one can see the ice bottom horizon by 2017 has become both more reflective and slightly higher, indicating the interface is wetter, and the ice bottom at this location may have been lifted up, that is, become ungrounded between 2014 and 2017. The ice bottom hydrostatically derived from surface elevation underestimates the bottom of the floating ice (Figure 3c), suggesting the ice is not in hydrostatic equilibrium. It is likely that the channel's narrowness and bridging effects of adjacent ice prevent the surface from fully compensating for basal mass loss (Drews et al., 2017). The 2017 radar profile also depicts the rapidly thinning channel C1 as a new “ghost” horizon ∼200 m above the ice bottom picks, which is likely a side echo from the bottom of the basal channel (Figures 3b and 3c). One can also see the expression of the basal channel at the point where the flight crosses the hook-shaped connection between C1 and P1, and it is thinner than predicted by the surface DEM-based reconstruction. Finally, the flattening of the ice shelf surface “bump” along the basal channel between 2014 and 2020 (Figures 2c and 2d, around CCh) also indicates the ice bottom may have risen, potentially to the point of ungrounding.
The increasing dynamic thinning of the grounded ice is illustrated by the SERAC elevation time series reconstructions shown in Figure 4c. About 2.5 km upstream of the 2015–2017 grounding line and C1, there is a sudden increase in the rate of surface thinning beginning ∼2015 (CCh) and a smaller increase as far as ∼7 km upstream (CUp) by the following year. It appears this onset of thinning may be unique to the central channel, as there is a more subdued pattern upstream of the grounding line along the northern subglacial drainage route (NCh; there is insufficient data to construct a time series along the southern drainage route), nor is there any detectable change in thinning rate at a typical “background” point (BG).
(a) Mean surface velocity 2016–17. (b) Inferred along-flow strain rate component. (c) Time series of elevation change relative to 30 August 2020 for select locations, with total change in blue and the dynamic component in red. Purple indicates the background thinning rate. (d) Inferred shear strain rate component.
3.3 Dynamics of N79
The NW-SE lateral dichotomy may be due to rheological heterogeneity in the ice shelf. N79 contains ice from NEGIS, which originates in a region of elevated geothermal heat (Rogozhina et al., 2016), and from a tributary that merges from the west very near the outlet upstream of the NEIS, which is likely colder and less plastic. Thus, stress may be more prone to build up in SEIS, being accommodated in short episodes of high deformation. The inferred surface strain rates (Figures 4b and 4d, S7) seem to confirm this. Entering the confines of the fjord imposes along-flow compressive strain on the ice shelf. In NWIS, the compression is fairly uniform, but in SEIS it is specifically concentrated along the flow-perpendicular channels, perhaps explaining their complex morphology and narrowing over time. There is also a concentration of shear strain on C1 by 2016–17 which was not present previously (Figure 4d, S7), suggesting the NWIS and SEIS halves of the ice tongue may be decoupling as C1 becomes more incised. What effects decoupling could have on the ice shelf's stability are not immediately obvious, although the pattern suggests NWIS is exerting resistance on SEIS, and SEIS is attempting to break away as the union of the halves weakens.
Yet the same basic channel pattern has existed since at least 1978 and does not appear to have caused thinning of this magnitude in the middle of the ice shelf until recently. The likely causes of this severe localized thinning are several-fold. It is probable that infiltration of AIW has increased (Bentley et al., 2022), leading to more intense meltwater plume activity at the ice-ocean interface. Runoff rates have also continued to rise over the past few decades (Figure S9 in Supporting Information S1); the most recent flow-perpendicular channel formed in 2012, which was a year of particularly intense melting for Greenland (Nghiem et al., 2012), and significant calving events across the north (Ochwat et al., 2022). The potential shift in subglacial drainage patterns would also eventually direct this runoff directly into C1. Moreover, the non-perpendicular angle of P1 may have allowed it to better serve as an extension of the central channel than prior perpendicular channels, creating a lengthy conduit for housing a stable subglacial melt plume.
4 Conclusion
Despite its complicated system of subglacial channels, we find that N79 was relatively stable for many years (at least from 1978 to 2012, Figure S10 in Supporting Information S1). A flow-perpendicular channel/central channel complex periodically appeared and grew modestly for 5–10 years, but that growth would significantly diminish when a new flow-perpendicular channel formed and the cycle repeated. One might liken this to the configuration of Jakobshavn Glacier prior to the disintegration of its floating ice tongue in 1998 (Thomas et al., 2003). Like N79, Jakobshavn is sourced from two tributaries and had a large basal channel near the grounding line along the seam between these two branches. This channel began to grow, likely as a result of increased warm water at the bottom of the fjord (Motyka et al., 2011). However, disintegration did not occur until the channel drew close to the calving front, and for the central channel of N79 this is decades away given its length and velocity. There is also a resemblance to recent events at Petermann Glacier, which has a similarly long ice shelf, where grounding line retreat has been facilitated by warming ocean temperature (Washam et al., 2019) and fractures causing sections of the ice to become decoupled from one another (Millan et al., 2022).
While the impact of these developments on the ice shelf may not be felt for years, there are still insights to be gained. Firstly, the importance of continued high-density data collection must be stressed. Our combined data approach reveals dynamically relevant processes on relatively small spatiotemporal scales, and we believe similar reconstructions for other locations could be equally revelatory. Such observations cannot be made without a high spatiotemporal density of altimetry, DEMs, and surface velocities. Ideally, there would also be a greater density of radar observations, as there are presently no more effective methods of determining the true shape of the ice shelf bottom, and our conclusions could be strengthened by additional feature-informed radar flight lines. As demonstrated above, reconstructing basal elevation based on the surface can easily underestimate channelized thinning, since narrows features may not be in hydrostatic equilibrium.
Perhaps more significantly, the results also hint at the weaknesses of our current fundamental ability to model ice sheets. The channels of N79 and their varied behavior are too small-scale and temporally variable to be easily incorporated in a model, both in terms of including the relevant physics, and the necessary computational power for resolving such features. Yet the effects are dramatic, and there is no quick and obvious way to replicate them. When one smooths out the inputs, generalizing from localized data, the results can be misleading. Consider Mayer et al. (2018), which reconstructs the mass loss of N79 over multiple decades largely based on observations of a single feature near the NWIS margin. Our results suggest that no single location is likely to be representative of the entire ice shelf on large time scales. A more limited study can be instructive, but with the increasing availability of data, more comprehensive studies are becoming more feasible.
It is our hope that other researchers and funding agencies will continue to strive for greater density and accuracy of data, and increased model complexity. The tools developed for this research are being prepared for public release, and we envision significant potential for improving our understanding of processes with high spatiotemporal variability, and generating data sets to validate numerical representations of these processes.
Acknowledgments
The authors wish to thank Laurie Padman and Susan L. Howard for assistance with tidal corrections, Cornelis van der Veen for helpful discussion, and Ivan Parmuzin for assistance with data management and visualization. Noël B., Brils, M., van den Broeke, M., and Kuipers Munneke, P. are acknowledged for providing the RACMO 2.3p2 runoff model, and the IMAU-FDM firn-densification model. This research was made possible with the following funding: NASA Sea-Level Change Team (Grant 80NSSC21K0322), ICESat-2 Science Team (Grant 80NSSC21K0915), and NSF Polar and Cyberinfrastructure programs (Grant OAC-2004826).
Open Research
Data Availability Statement
The software used to generate the elevation reconstructions is the Mosaic Utility and Large Data set Integration for SERAC (MOULINS) (Narkevic, 2021), which is still in development for public release. It includes spline-based curve fitting based on (Shekhar et al., 2021), and tidal correction based on software available at (https://github.com/tsutterley/pyTMD). The altimetry data used come from the Airborne Topographic Mapper (ATM; https://nsidc.org/data/ilatm2/versions/2), ICESat (https://nsidc.org/data/glah12/versions/34), and ICESat-2 (https://nsidc.org/data/atl06/versions/4). Uncorrected DEMs from 2012 to 2020, generated from WorldView imagery by the ArcticDEM project and are available at https://data.pgc.umn.edu/elev/dem/setsm/ArcticDEM/strips/s2s041/2m). The 1978 DEM is from https://www.ncei.noaa.gov/access/metadata/landing-page/bin/iso?id=gov.noaa.nodc:0145405. Surface elevations outside the reconstructed region are from BedMachine v4 (Morlighem et al., 2017), available at https://nsidc.org/data/idbmg4/versions/4, and the bed elevation is from (An et al., 2021), available https://datadryad.org/stash/dataset/doi:10.7280/D19987. Subglacial drainage reconstructions are made with Topo Toolbox (Schwanghart & Nikolaus, 2010), available at https://topotoolbox.wordpress.com/. The velocities used can be found at https://datadryad.org/stash/dataset/doi:10.7280/D11H3X. N79 calving fronts are from (Goliber et al., 2022) and available at from https://doi.org/10.5281/zenodo.6557981. All Landsat imagery is courtesy of USGS and obtained from https://earthexplorer.usgs.gov/. All new data sets generated by this study (surface elevation mosaics, corresponding ice bottom elevation, Lagrangian elevation change, and select partitioned time series), are accessible through Zenodo https://doi.org/10.5281/zenodo.7518206.
