Bathymetry Beneath Ice Shelves of Western Dronning Maud Land, East Antarctica, and Implications on Ice Shelf Stability

Antarctica's ice shelves play a key role in stabilizing the ice streams that feed them. Since basal melting largely depends on ice‐ocean interactions, it is vital to attain consistent bathymetry models to estimate water and heat exchange beneath ice shelves. We have constructed bathymetry models beneath the ice shelves of western Dronning Maud Land by inverting airborne gravity data and incorporating seismic, multibeam, and radar depth references. Our models reveal deep glacial troughs beneath the ice shelves and terminal moraines close to the continental shelf breaks, which currently limit the entry of Warm Deep Water from the Southern Ocean. The ice shelves buttress a catchment that comprises an ice volume equivalent to nearly 1 m of eustatic sea level rise, partly susceptible to ocean forcing. Changes in water temperature and thermocline depth may accelerate marine‐based ice sheet drainage and constitute an underestimated contribution to future global sea level rise.


Introduction
Floating ice shelves in Antarctica play a key role in buttressing its grounded continental ice sheets. Freezing and/or melting at ice shelf bases influences the overall mass balance of the ice sheets that feed them. An increase in mass loss of an ice shelf is followed by increased drainage of its ice sheet (Dupont & Alley, 2005;Gudmundsson et al., 2019). The most important process by which Antarctica's ice shelves lose mass is basal melting by interactions with warm seawater (Depoorter et al., 2013;Oerter et al., 1992;Rignot et al., 2013), whose circulation beneath the shelves largely depends on subglacial bathymetry Jenkins et al., 2010;Tinto et al., 2019). According to Goldberg et al. (2019), accurate bathymetry is the leading requirement for correctly estimating basal melt rates in circulation models. Consistent and accurate bathymetric models therefore benefit estimations of ice shelf and ice sheet stability. However, most compilations of subglacial topography incorporate low-resolution interpolated sub-ice shelf bathymetries (Arndt et al., 2013;Fretwell et al., 2013;Morlighem et al., 2019).
The cavities beneath the ice shelves of western Dronning Maud Land (wDML) have been only sparsely and incompletely sounded by seismic reflection data on the Fimbul and Ekström ice shelves (Nøst, 2004;Smith et al., 2020). Bathymetry beneath the neighboring Atka, Jelbart, and Vigrid ice shelves is simply unknown. Covering 63,000 km 2 (Figure 1), they represent 10% of East Antarctica's total ice shelf by area and experience ©2020. The Authors. This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited.

10.1029/2019GL086724
Key Points: • We present subglacial topography models beneath Ekström, Atka, Jelbart, Fimbul, and Vigrid ice shelves • Water cavities beneath ice shelves of wDML are secluded due to moraine formations at LGM and subsequent shallow water entry points • Ice shelves are currently protected by sills but are highly sensitive to future warming ocean temperatures and changing thermocline depth Supporting Information: • Supporting Information S1 mass loss predominately from iceberg calving and basal melting in almost equal parts (Rignot et al., 2013). The ice shelves' catchment area is mainly drained through the Jutulstraumen Glacier ( Figure 1; Mouginot et al., 2017b;Rignot et al., 2013). The area currently contains 341,000 Gt of ice above sea level, a sea level equivalent (SLE) of about 0.95 m (Supporting Information S1). Around 18% (17 cm SLE) of this is grounded below current sea level, making it susceptible to ocean forcing. Inland, where prograde slopes impair ocean intrusion to the remaining catchment, mass loss is expected to be dominated by atmospheric forcing (DeConto & Pollard, 2016).
The ice shelves of wDML are in contact with seawater at temperatures close to the surface freezing point. This classifies them as cold-cavity ice shelves (Rignot et al., 2013). Increasing water temperatures in cavities are generally related to warming ocean temperatures (Gille, 2008;Schmidtko et al., 2014), redirected currents, or increased upwelling of Warm Deep Water (WDW; Carmack & Foster, 1975;Hattermann, 2018;Schröder & Fahrbach, 1999). WDW is carried by the Weddel Gyre traversing a narrow continental shelf at Figure 1. Overview of survey region in wDML with (a) magnetic anomaly data (Golynsky et al., 2018;Mieth & Jokat, 2014) showing the proposed boundary of the Archaean Grunehogna Craton (Unit 01), Jurassic mafic intrusions (Unit 02), and seaward dipping Jurassic basalt flows (Unit 03) (Jokat et al., 2003(Jokat et al., , 2004Riedel et al., 2013). (b) Map of data used for this study with gravity (green) and ice thickness radar data (yellow), bathymetric references from prevalent seismic reflection data (red), and shipborne swath bathymetry data (blue). (c) Extent of survey region. Grounded ice, ice shelves, and ocean are illustrated in gray, white, and blue, respectively. wDML (Schröder & Fahrbach, 1999), permitting strong interaction between the coastal current and water cavities and implying high sensitivity to rising ocean temperatures (Nicolaus & Grosfeld, 2004). Increasing temperature could lead to a rapid increase in basal melt rates and negatively influence ice shelf mass balance (Hellmer et al., 2012). Langley, von Deschwanden, et al. (2014) observed a network of inverted channels at the base of the Fimbul Ice Shelf using ground-based radar data, hinting at small-scale active melting. In addition, Zhou et al. (2014) have shown that heated Antarctic Surface Water (ASW) finds its way beneath the Fimbul Ice Shelf, whereas the entry of WDW is currently minimal and strongly dependent on bathymetric features .
We address the lack of consistent and reliable bathymetric estimates beneath the ice shelves of wDML by inversion of airborne gravity data for bathymetry. Our models reveal deep troughs beneath the ice shelves and allow us to identify sites of possible present or future inflow of WDW into the cavities.

Geologic Framework
A clear understanding of underlying geology is vital for modeling bathymetry from gravity because of the effects of variable rock densities on gravitational accelerations (Brisbourne et al., 2014). The geology of wDML is largely interpreted by extrapolation from sparse outcrops along magnetic anomalies (Jacobs et al., 1998;Jacobs et al., 2017, Figure 1a). This reveals strong geologic affinities to South Africa, its conjugate in Gondwana (Jokat et al., 2003). Along the continental shelf, a belt of strongly positive magnetic anomalies ( Figure 1a, Unit 03) is related to thick packages of intercalated basalts and sediments that dip toward the ocean, as interpreted in seismic reflection data (Kristoffersen et al., 2014). Beneath the Ekström Ice Shelf, these basalt packages form the so-called Explora Wedge (Hinz & Krause, 1982;Jokat et al., 2004;Kristoffersen & Haugland, 1986), which we describe in greater detail in section 3.1 using newly acquired high-resolution magnetic data.

Materials and Methods
Bathymetry models for the Ekström and Atka ice shelves are based on aerogeophysical data collected during the austral summer of 2015/2016 within the Geodynamic evolution of East Antarctica (GEA) project conducted by the Alfred Wegener Institute (AWI) and Federal Institute for Geosciences and Natural Resources (BGR). The Fimbul and Vigrid bathymetry models are based on data collected between 2001 and 2004 as part of AWI's VISA project (Riedel et al., 2012). Bathymetric models for Jelbart are based on a compilation of GEA and VISA data. Both projects collected coincident gravity, magnetic, and radio echo sounding (RES) data ( Figure 1b). Seismic data are incorporated for Ekström (Smith et al., 2020) and Fimbul ice shelves (Nøst, 2004).
Gravity data of differing quality were delivered by a GT-2A gravimeter during the GEA campaign and a ZLS Ultrasys Lacoste & Romberg Air/Sea gravimeter during VISA campaigns ( Figure S2.1; Riedel et al., 2012). Confidence in the bathymetric models also depends on the availability of reliable ancillary information on subsurface geological density variations. Our estimates of combined uncertainties in the model bathymetries are ±175 m (Ekström and Atka), ±225 m (Jelbart), and ±210 m (Fimbul and Vigrid), consistent with previous empirically derived root mean square misfits between inversion-derived and seismic depths for other Antarctic ice shelves (Brisbourne et al., 2014;Greenbaum et al., 2015).
Gravity data are inverted for bathymetry using the module "GM-SYS 3D" within Geosoft Oasis montaj, which implements a 3D forward-modeling approach based on Parker (1972). Reference and control points from seismic data over the ice shelves, swath bathymetry along the calving fronts, and grounded ice thickness from RES data are used where available to design preprocessing filters to suppress nonbathymetric signals. These specific filters are applied to enhance the resemblance between gravity and sparsely known existing bathymetric data (Figures 2 and 3) as a step toward minimizing residuals prior to inversions. Free-air anomaly (FAA) data over the Ekström and Atka ice shelves ( Figure 2a) were preprocessed using a 70-km high-pass filter, oriented at 45°using a cosine function to suppress features trending parallel to the main crustal and geological structures of the continental shelf, as revealed in the magnetic anomaly data ( Figure 2d). FAA data from the Jelbart Ice Shelf were isotropically band-pass filtered over 6-100 km (Figures 3a and 3b), and those across the Fimbul and Vigrid ice shelves were band-pass filtered over 6-150 km (Figures 3c and 3d) to suppress aircraft noise and long wavelength signals related to crustal thickness variations and sedimentary basins at the extended continental margin (Figure 3; Jokat et al., 2003Jokat et al., , 2004. RES ice thicknesses from the GEA campaign are derived after correction for firn (Blindow, 1994) and subtraction from the Reference Elevation Model of Antarctica (REMA; Howat et al., 2019) to determine the base ice position. Ice thicknesses from VISA campaigns are incorporated from Riedel et al. (2012).
Along the coastline and calving fronts of wDML, swath bathymetry data acquired during various surveys conducted by RV Polarstern are implemented to evaluate the progression of possible troughs Airborne magnetic data across wDML were acquired during the GEA campaign using a Scintrex Cs-3 cesium magnetometer, including at 2-km line spacing above Ekström (Figure 2d). We used these data to interpret shallow subsurface geological variations at wavelengths passed by prefiltering that must be accounted for in bathymetric modeling (Figure 1). For further details on materials and methods, see Supporting Information S2. The unfiltered FAA data for Jelbart (a) and Fimbul/Vigrid (c) with crossing flight lines (5-to 10-km spacing) are band-pass filtered in panels (b) and (d) to achieve a starting resemblance to bathymetries (e) constrained from swath bathymetry, seismic reflection point data (Nøst, 2004; 5 × 5-km squares), and ice thickness radar data in grounded areas. Positive, long-wavelength anomalies close to the continental shelf in the north in panels (a) and (c), possibly representing crustal thinning, were removed by described filtering techniques in panels (b) and (d). Legend in panel (d) is valid for panels (a) throughout (d).

Geologic Constraints From Magnetic Data Interpretation
Changes in magnetic field strength are far less sensitive to bathymetric than to geological variations, meaning well-founded magnetic interpretations can be employed to reduce uncertainty in the gravity models. SW-NE-trending positive magnetic anomalies oriented perpendicular to the ice shelf flow are identified in the aeromagnetic data set across the Ekström Ice Shelf (Figure 2d). A linear high reaching 250 nT in the eastern part of Ekström bifurcates westwards into 300-nT-high branches. These anomalies are confidently relatable to the shallow subcrop of the Explora Wedge (Kristoffersen et al., 2014). Ferraccioli et al. (2005) and Riedel et al. (2013) interpret a triangular magnetic high across the Jelbart Ice Shelf and coincident circular gravity and magnetic anomaly highs over the southern part of the Fimbul Ice Shelf close to the outlet of the Jutulstraumen (Figure 1a, Unit 02) as Jurassic mafic intrusions. These intrusions are reflected in the density maps estimated in the 3D modeling process (Supporting Information S2) showing a distinct correlation to magnetic anomalies. Figure 4 shows models of bathymetry beneath the ice shelves, together with water column thicknesses estimated by subtraction of the ice draft calculated from RES data. Our first-order observation in wDML is that the ice shelves overlie a seabed (mostly <1 km deep) that deepens between shallow sills close to the present-day calving fronts and the landward grounding line. In greater detail, troughs cross the shelves, and the sills are interrupted by a series of distinct narrow gateways (Figure 4a). Depth maxima, especially beneath the Ekström and Fimbul ice shelves, correlate with thick water columns and are separated by small bathymetry highs. This may result in several possible water circulation regimes (Figure 4).

Bathymetric Models Beneath Ice Shelves of wDML
Modeled bathymetry beneath the Ekström Ice Shelf is characterized by a trough parallel to the direction of main ice flow (Figures 1d and 4a). It has two distinct maxima with water depths of about 1,130 m in the south and 1,030 m in the center separated by a bathymetric high of 860 m. The water column (Figure 4b) is thickest over the trough, with maxima slightly exceeding 600 m in the south and center, interrupted by the bathymetric high. Swath bathymetry along the calving front shows a 390-m-deep gateway to the cavity (EK-1, Figure 4a) within a sill following the course of the calving front.
The neighboring Atka Ice Shelf has a similar shape, with a NNW-trending trough in its center exhibiting two depth maxima, at 780 m modeled close to the grounding line and 620 m, observed in bathymetry data, north of the calving front. Gateways in the sill reach an observed depth of 500 m beyond the calving front and a modeled depth of 450 m beneath the ice shelf (A-1, Figure 4).
The Jelbart Ice Shelf is underlain by two bathymetric troughs, separated by a broad bathymetric high over which the water column thins to 50 m. The western trough reaches maximum depths of 1,200 and 1,050 m in the south separated by a minor bathymetric high. The trough terminates against a rampart-like bathymetric high (J-BH, Figure 4a). The eastern trough coincides with the present-day surface ice flow maximum (Figure 1d). Its maximum depth is about 1,000 m close to the grounding line and 900 m in its center. It progresses north toward the continental shelf edge where it crosses a bathymetric sill via a 430m-deep gateway, observed with swath bathymetry, into the Weddell Sea (J-1, Figure 4a).
The seabed beneath the Fimbul Ice Shelf is characterized by a central trough in line with the Jutulstraumen glacier outlet (Figure 1d). Three depth maxima of 1,300, 1,250, and 1,150 m, from south to north, are set between 1,000-m-deep bathymetric highs. The main gateway, F-1 (Figure 4a

Discussion
Smith et al. (2020) report large differences between seismically derived depths beneath the Ekström Ice Shelf and those derived by interpolation in Bedmap2 (Fretwell et al., 2013), which has been the baseline data set for the majority of modeling studies in the area. Our models show similar differences, reaching up to several hundreds of meters, for all ice shelves of wDML. Based on these differences, Bedmap2 underestimates the volume of wDML water cavities by 63% compared to our bathymetric models. Excluding the area of Nøst's (2004) seismic measurements, which were incorporated in Bedmap2, Bedmap2 underestimates cavity volume by 444% (table in Figure 4b), because it interpolates between the sparse reference points.
Seabed variations beneath the ice shelves of wDML are all of similar shapes, including deep troughs separated by small bathymetric highs, suggesting shared or related origins and former glacier extents. Smith et al. (2020) interpret periodic retreat of the grounding line position based on the presence of multiple bathymetric highs interspersed between deep points along the trough beneath Ekström. Our bathymetric model displays similar features and shows the bathymetric highs to continue laterally across the seafloor, supporting this interpretation. We also observe similar segmentation of troughs crossing the neighboring ice shelves.
These troughs vanish in shallow water close to the calving fronts. The repetition of this pattern throughout wDML suggests a link to former glacial extents, most likely via the formation of end moraines by ice sheets that advanced to the continental shelf edges during the Last Glacial Maximum (LGM, at 23-19 kyr BP; Grobe & Mackensen, 1992;Hillenbrand et al., 2014). Subglacial sediments deposited at shelf edges might be displaced downslope or be remobilized by currents to form contourites, as observed in seismic data close to the ice shelves of Dronning Maud Land by Huang and Jokat (2016).
The modern-day calving fronts of these ice shelves overlie shallow sills cut by gateways with depth maxima at 550 to 600 m, which limit the possible entry of WDW (Schröder & Fahrbach, 1999) into the cavities due to similar average thermocline depths of about 600 m in water depths of 1,000 m (Hattermann, 2018). Hattermann et al. (2014) have shown that only small amounts of modified WDW creeps into the Fimbul cavity. The sills thus currently protect the ice shelf base from basal melting from considerable WDW ingress. Seasonal variability in basal melting beneath Fimbul  is probably linked to the inflow of solar-heated ASW (Zhou et al., 2014). A similar setting for neighboring ice shelves leads to the assumption that present-day basal melting processes are mainly driven by ASW throughout wDML. Our new results show long continuous cavities at depths greater than 1,000 m leading landwards from the sills at average depths of 400 m. If enhanced surface ocean warming or a shallower thermocline in the future leads more significant quantities of warm water to breach the sills, these deep troughs will guide fast access to the grounding line with little to stop basal melting from eroding wDML's ice shelves. Since our bathymetric models are limited by their depth uncertainties and a horizontal resolution of about 5 km, it is necessary to further investigate these decisive bathymetric features close to the calving fronts in the future.
Buttressing by grounding at pinning points plays an important role in stabilizing ice shelves by braking ice flow (Dupont & Alley, 2005). We suggest that a broad bathymetric high with a water column of 50 m in the center of the Jelbart Ice Shelf (J-BH, Figure 4a) acted as a pinning point in the past. This semicircular bathymetric high suggests its formation as a moraine, caused by former glacial extents at the western trough and/or less erosion in the center of the ice shelf due to main ice streams being situated west and east. From this we infer a depositional character for J-BH, whose depth may be underestimated owing to unaccounted-for lower density. Loss of buttressing might explain a high ice mass loss of 41 Gt for the drainage area of the Jelbart Ice Shelf from 1979 until 2017, compared to stable mass balances in neighboring drainage areas . A similar paleo-pinning point (V-BH, Figure 4a) is interpretable from the short (40 m) water column beneath the northern Vigrid Ice Shelf. Buttressing loss from these points might have led to enhanced thinning and accelerated drainage (Dupont & Alley, 2005); ice flow velocities over the Vigrid Ice Shelf are currently faster than the neighboring eastern Fimbul Ice Shelf ( Figure 1d). However, ice velocities have not increased in the time span from 1994 to 2012 (Gudmundsson et al., 2019).

Summary
We have constructed new bathymetric models for seafloor beneath the Ekström, Atka, Jelbart, Fimbul, and Vigrid ice shelves using gravity inversions constrained by seismic, multibeam, and radar data and geological variability interpreted from magnetic anomalies. Our bathymetric models show that current topographic compilations used, for example, in ice sheet/ocean modeling heavily underestimate the vulnerable ice shelf cavities of wDML by up to 798% for the Ekström Ice Shelf and by 63% in total.
The newly mapped seabed underlies about 10% of East Antarctica's ice shelf inventory by area (Rignot et al., 2013). It is of direct significance for understanding and modeling the current and future stability of the wDML ice sheet, which contains water equivalent to about 0.95 m of eustatic sea level rise and is susceptible to ocean and atmospheric forcing.
The models show deep seawater cavities under all of the ice shelves, providing increasing evidence that this is a trend in DML. These cavities are partially secluded from the open ocean by sills perched near or at the continental shelf edge. The sills currently protect the ice shelves from basal melting by WDW ingress. At present, the ice shelves probably experience basal melting mainly by warm surface waters. However, shallowing thermocline depths with an increased entry of WDW into the cavities pose a significant risk to these ice shelves in the future.

Data Availability Statement
Bathymetric models and raw gravity data can be accessed at pangaea.de (10.1594/PANGAEA.913742).