The study area is the south-western part of Antarctic Drainage System 4 (as defined by Zwally et al., 2012) (Fig. 1a). Here, the EAIS discharges into the Weddell Sea via the Coats Land ice margin: 600 km of marine-terminating glaciers, glacier tongues and ice shelves along the Luitpold and Caird coasts between the Filchner and Riiser-Larsen ice shelves from 74 to 77∘ S. From north to south the main glaciers include the Stancomb-Wills Glacier, Dawson Lambton Ice Stream, Hayes Glacier, “Lichte Glacier”, Mann Ice Stream, Weldon Ice Stream, Mosley Ice Stream and the Lerchenfeld and Schweitzer glaciers. These discharge into the Brunt Basin, Dawson-Lambton, Caird, Lichte, Möller, “Hoffmann” and Albert troughs, and an unnamed trough in Vahsel Bay, respectively (quotation marks denote unofficial names). The Stancomb-Wills Glacier and Dawson Lambton Ice Stream form glacier tongues extending 250 and 20 km from their grounding lines, respectively. Ice shelves include the BIS, which is formed from a series of unnamed glaciers extending from the grounding line and is pinned to the bed at McDonald Ice Rumples (75∘28′ S, 26∘18′ W). Together with the SWGT, it forms a continuous floating ice mass with a number of large cracks and chasms and a melange of icebergs and sea ice where it forms at the grounding line (Anderson et al., 2014). At the southern limit of the study area the Lerchenfeld and Schweitzer glaciers feed a small (30 km long, 10–20 km wide) ice shelf maintained through lateral buttressing in Vahsel Bay (Fig. 1b).

All surveyed glacial troughs were characterised by reverse-bed slopes ranging from moderate (e.g. Möller Trough, which deepens by 100 m along a 34 km profile; Fig. 2c, e) to steep (e.g. Albert Trough, which deepens by 450 m along a 26 km profile; Fig. 3c, d).

Over-deepened basins are present in all troughs with the deepest basins being at the eastern limits of the surveys, closest to the modern ice front. In some cases, large, deep basins were observed extending under the modern ice front, for example in the Caird and Albert troughs (e.g. Fig. 3b, c). In other cases, such as the Hoffman, Möller, Lichte and SWGT troughs, there are a series of two or more deep basins becoming progressively more poorly defined with increasing distance from the modern ice front (Figs. 2, 5). Some of the deep basins have 20 to 60 m high moraines on their seaward side, for example in Caird and Albert troughs (Fig. 3b, c, e). Other deep basins are separated by bedrock sills, for example in Dawson-Lambton Trough, where the deep basins have complex morphologies that appear to be bedrock controlled (Fig. 3a).

The volume of sediment infilling in the troughs and deep basins is variable with troughs downstream of the lower-velocity glaciers (< 250 m yr-1; Fig. 1b) typically characterised by ice-moulded bedrock and low volumes of sediment (e.g. Hoffmann, Möller and Lichte troughs; Fig. 2a–c). This was confirmed by gravity and box cores, which only recovered 18 cm of diamicton overlain by 40 cm of diatomaceous ooze in Möller Trough (BC608, Fig. 2c) and 47 cm of sediments in Lichte Trough (GC613, Fig. 2d), or did not recover any sediments (box core BC606 and gravity cores GC609 and GC610; Fig. 2c). In contrast, troughs downstream of higher-velocity glaciers (> 250 m yr-1), including the Albert and Dawson-Lambton troughs (Fig. 3a–c) and the Brunt Basin (Figs. 4, 5), contain high volumes of sediments. In Albert Trough a sediment drape is present in the 900 m deep inner basin. A box core from this basin (BC566; Fig. 3c) consists of a gravelly mud unit overlain by a diatomaceous ooze with abundant dropstones. A similar sequence was found in a 4.08 m long gravity core (GC569; see Fig. 3c for location, Fig. 6 for core log) which included a glacial deposit of stiff, greenish-grey gravelly mud and clasts below 3.88 m overlain by laminated diatomaceous muds and oozes with occasional sand layers.

Arcuate ridges of several kilometres in length and up to 35 m in height are present in Hoffmann, Albert and Caird troughs (Figs. 2a, 3b, c). These ridges are transverse to the long axis of the troughs, typically symmetric and steep-sided in cross profile (Fig. 3e). There is no acoustic penetration of these features on sub-bottom profiles, suggesting that they consist of coarse, diamictic and/or compacted material. The most extensive ridges occur at the western end of the trough basins (e.g. Hoffmann, Albert and the Lerchenfeld/Schweitzer glacier troughs; Figs. 2a, 3c, 7c). The dimensions, morphologies and locations of these ridges, transverse to the former ice-flow direction, indicate that they represent terminal push moraines formed during an advance or readvance of grounded ice in the troughs (e.g. Ottesen et al., 2005). The smaller ridges found at topographic highs and/or narrow trough sections between trough basins (e.g. Hoffmann and Caird troughs; Figs. 2a, 3b) most likely represent recessional moraines formed during minor still stands of the retreating grounding line.

A variety of elongate streamlined landforms also occur in the troughs. These range from long, low amplitude, straight ridges, to arcuate ridges separated by depressions (e.g. Dawson-Lambton and Albert troughs; Fig. 3a, c); dimensions are 1–3 km in length, 5–10 m in height and 200–500 m in width with elongation ratios of up to 10 : 1. The subdued morphology of the ridges and rare sub-bottom reflections in TOPAS profiles indicate that they consist of unlithified sediments. We interpret these trough-parallel landforms as subglacial lineations formed by fast glacier flow over a deformable bed (see King et al., 2009; Ó Cofaigh et al., 2005). Shorter streamlined landforms consist of rounded to blunt-nosed hills with a tapered (seaward) end up to 70 m high and 1.8 km long (elongation ratios < 5 : 1), usually in the inner and middle parts of the troughs (Figs. 2d, 3c). Sub-bottom profiles do not show penetration through acoustic basement, indicating that these features are likely formed in bedrock. These landforms are interpreted as subglacially moulded rôche moutonnée, whaleback forms and small crag and tails (see Livingstone et al., 2015; Nitsche et al., 2013). In Albert Trough, glacial lineations extend throughout the trough and curve northwards, where the trough widens into the Weddell Sea Embayment (Figs. 4c, 7d), suggesting that past ice flow was deflected northwards where the Albert Glacier joined the pFIS in Filchner Trough.

Relatively wide (2–40 km), subdued, asymmetric ridges and topographic highs occur offshore of the Caird and Albert troughs (Fig. 7) and inshore within the Dawson-Lambton and Albert troughs (Fig. 4b, c). These features are up to 100 m high and have the steep seaward-facing slopes and gentler landward slopes typical of GZWs (Dowdeswell and Fugelli, 2012; Graham et al., 2009). These typically form at the grounding line of a retreating ice mass when it pauses for periods of several decades up to a few thousand years (Alley et al., 1989; Dowdeswell and Fugelli, 2012). The widest GZWs or recessional moraines are found > 20 km north-west of Caird Trough, marking a series of at least five grounding line positions (Fig. 7a). There are similar GZWs, marking at least three grounding line positions north of Albert Trough (Figs. 4c, 7d), the northernmost two of which (Figs. 7d, 1 and 2) are oriented transverse to past ice flow in Filchner Trough (Fig. 7d, red arrow).

The GZW within the Dawson-Lambton Trough was studied in detail. Its gently dipping upstream side (∼ 1∘) and much steeper downstream side (7–9∘) (Fig. 4b) indicate that it was deposited by a seaward-thinning glacier, and the presence of lineations and notable absence of plough marks suggest that ice was fast flowing and probably already near to floatation (see Batchelor and Dowdeswell, 2015). An along-strike seismic profile shows multiple architectural units. These include dipping reflections in a bank at the western side of the wedge consistent with development through lateral progradation. Along-trough progradation can be inferred from the top sets and discrete units in the internal architecture of the GZW (Fig. 8a). The base of a sediment core recovered from the crest of the GZW (GC630, Fig. 9a) consists of a sandy mud containing large clasts, calcareous marine micro-fossils. Its low water content and high shear strength are consistent with glaciomarine sediments. This is overlain by a thin layer of structureless diamicton, interpreted as a subglacial till formed by grounded ice. The upper unit of the core consists of gravelly sands and sandy muds. A second core on the seaward side of the GZW (GC626, Fig. 9b) is characterised by a sequence of sandy mud and normally graded gravel to muddy sand, interpreted as debris flows, likely generated from glaciomarine sediments reworked by ice grounded on the crest of GZW.

A topographic high occurs to the west of the Brunt Basin (Figs. 4a, 5). Named the McDonald Bank, it forms part of a 100–400 m high ridge of glacial sediments with internal truncated reflectors (seismic profile 5; Elverhøi and Maisey, 1983) rising to within ca. 220 m of present sea level. Surface altimetry of the BIS shows that the McDonald Bank deflects the flow of the SWGT to the north and BIS to the south. Back stresses create a series of upstream concentric pressure waves in the ice shelf. This area of deformation is known as the McDonald Ice Rumples (75∘28′ S, 26∘18′ W), which cover an area of approximately 3 × 3 km2 (Fig. 10). Apart from the Lyddan Ice Rise (north of the SWGT, Fig. 1c), the deformation of the BIS around the McDonald Ice Rumples is the only evidence in the study area of ice being in contact with the bed beyond the grounding line. This can be attributed to differences in water depth, with the McDonald Bank being less than 220 m from the surface while the topographic highs formed by the GZWs in Dawson-Lambton and Albert troughs are at 380 and 450 m, respectively (Fig. 4b, c). These differences in water depth constrain the potential of these topographic highs to act as pinning points for advancing ice streams and glacier tongues.

The timing of ice advance and retreat is not well constrained by our data.

Mixed microfossil ages of ca. 34 140 cal. yr BP (at 80 cm core depth) and > 43 500 14C yr BP (at 60 cm depth; Table 1) were found in the gravity-flow deposits sampled at the base of the GZW in Dawson-Lambton Trough (GC626, Fig. 9b). This suggests that glaciomarine sediments from Marine Isotope Stage 3 (MIS3), and predating the last ice advance, were present and being reworked during the progradation of the GZW.

A radiocarbon age of 24 931 cal. yr BP (Table 1) from a calcareous microfossil in the glaciomarine sediments, immediately underlying the diamicton on the crest of the GZW in Dawson-Lambton Trough (100 cm in GC630, Fig. 9a), provides a minimum age constraint before the onset of the last ice advance. The older microfossil age of ca. 33 837 cal. yr BP is found in the overlying sandy mud at 40 cm core depth can be attributed to the continued reworking and progradation of MIS3 glaciomarine sediments.

In terms of ice retreat, a basal date from GC569 in the deep basin in Albert Trough (Fig. 6) gave a bulk acid-insoluble organic age of 13 382 cal. yr BP (Table 1). Subtracting the 4899 yr surface age of the core yields a minimum corrected age of 8483 cal. yr BP. The presence of mixed ages (ranging from 3693 to 6550 cal. yr BP) in the diatomaceous ooze above 3.88 m indicates that this date must be treated with caution due to sediment mixing and/or varying contributions of reworked fossil carbon.

The MIS3 marine microfossils incorporated in the reworked sediments of Albert and Dawson-Lambton troughs and elsewhere in the Weddell Sea (Hillenbrand et al., 2014) suggest widespread seasonally open water before the last main ice advance. This, in addition to previous advances, has contributed to the erosion of the 450–900 deep glacial troughs and trough basins along the Coats Land ice margin. The streamlined landforms in bedrock, moraine complexes trending NNW from Albert Trough (Fig. 4c), GZWs north of Albert Trough (Fig. 7c) and the NNW-trending trough under the BIS–SWGT show that ice flow from Coats Land was redirected by the pFIS, contributing to a regional ice advance in the Filchner Trough sometime after 24 931 cal. yr BP (GC630). In Filchner Trough, this ice advance has previously been dated to sometime after 11 800 cal. yr BP (Arndt et al., 2017). It followed evidence of an earlier ice advance prior to 27 500 cal. yr BP (Arndt et al., 2017), which was characterised by subglacial lineations and drumlins extending to within approximately 40 km of the shelf break (Fig. 1c) (Gales et al., 2012; Hillenbrand et al., 2014; Larter et al., 2012). Combined, the evidence suggests that streaming ice in the Coats Land troughs, together with the major ice streams feeding the pFIS (Bailey Ice Stream, Slessor and Recovery glaciers) and lower relative sea levels, may have contributed to the drawdown of the interior ice sheet. This would provide a mechanism for the minor thickening, lack of change or even thinning seen in the hinterland (Bentley et al., 2010; Hein et al., 2011; Hodgson et al., 2012; Parrenin et al., 2004; Whitehouse et al., 2012a).

The large volumes of glacial sediments in the Filchner Trough and downstream of the higher-velocity Coats Land glaciers have been reworked into a number of landforms that document the retreat of grounded ice. The first group of landforms includes features oriented transverse to the pFIS, the best examples being the previously described GZWs in Filchner Trough (Arndt et al., 2017; Larter et al., 2012) and beyond Albert Trough (this study, Fig. 7d). The second group of landforms is oriented transverse to the ice streams, glaciers and ice shelves discharging from Coats Land. These include the GZWs beyond Caird Trough (Fig. 7a) and Lerchenfeld Glacier (Fig. 7c), terminal moraines at the seaward end of the deep basins and at trough narrowings, and the within-trough GZWs. These landform assemblages suggest at least four phases in the ice retreat across the inner shelf.

The first phase was the southward migration of the pFIS ice front in the eastern Weddell Sea and the progressive loss of contact between the Filchner Ice Shelf and Coats Land glaciers from 12 848 to 8351 cal. yr BP. The timing of this southward migration is consistent with “scenario B” of Hillenbrand et al. (2014) for regional deglaciation, with sediment age constraints from this, and other studies, tracking an approximate north-to-south migration of the Filchner Ice Shelf (Table 1). This began with core G2 off the Dawson-Lambton Trough at 12 848 cal. yr BP, followed by core 3-6-1 off Caird Trough at 9555 cal. yr BP, core GC569 in Albert Trough at 8483 cal. yr BP (this study with the surface correction applied), core G7 off the Lerchenfeld and Schweitzer glaciers at 8676 cal. yr BP and core PS1621 south of Vahsel Bay near the Filchner Ice Shelf front at 8351 cal. yr BP (all core locations in Fig. 1c). At some point during this southward migration the pFIS lifted off the bed and became an ice shelf, resulting in the preservation of subglacial lineations and drumlins in Filchner Trough (Fig. 1b; Larter et al., 2012). Landforms associated with grounded ice and quasi-stable ice-margin positions, such as GZWs, are therefore rare in the main Filchner Trough; the exceptions are one GZW landward of linear iceberg furrows on the outer shelf (Larter et al., 2012), and one stacked GZW on the outer shelf, deposited after 11 800 cal. yr BP. (Arndt et al., 2017). However, chronological control and survey data are exceedingly sparse over most of the trough (as shown by Fig. 1b, c).

The southward-migrating Filchner Ice Shelf front resulted in a loss of buttressing and a shift in the orientation of the Coats Land glaciers that initiated their retreat. This second phase of grounded ice retreat was punctuated by grounding line halts at the trough narrowings (e.g. Fig. 3b and c) and topographic highs, the formation of GZWs and moraines at the seaward end of the deep basins in the troughs (e.g. Fig. 3e). The period of grounding line stability provided by the progradation of the GZWs was likely short-lived, possibly in the region of a few hundreds to thousands of years (based on estimates elsewhere: Batchelor and Dowdeswell, 2015; Dowdeswell and Fugelli, 2012; Jakobsson et al., 2012), and would have been dependent on sediment flux rates (Graham et al., 2010; Livingstone et al., 2016) and water depth.

The third phase involved development of subglacial cavities, loss of contact with the bed and the formation of ice shelves along the Coats Land ice margin. The absence of iceberg plough marks and recessional moraines in the troughs with soft sedimentary substrates suggests that this phase of ice retreat was controlled by thinning to form ice shelves and not retreat of grounded calving ice fronts.

Based on the landform assemblages in the troughs and modern analogues around the nearby Dronning Maud Land ice margin we infer that, during the fourth phase, the newly formed ice shelves advanced to the shallower, outer parts of the troughs. Here, at least some of them would have re-grounded on topographic highs and on inner trough GZWs. In the latter case, the re-grounding may have reworked at least part of the sediments comprising the GZWs – for example the prograded structures on the surface of the GZW in Dawson-Lambton Trough (Fig. 8a), and the gravelly sands and sandy muds overlaying the structureless diamicton in GC630 (Fig. 9a). However, once the ice shelves thinned to the point where mechanical contact with the bed was fully lost, the reverse-bed slope and consequent absence of further pinning points, likely resulted in their rapid retreat back towards the modern coastal grounding line where a rising bed stabilises the ice sheet. These processes, in particular the loss of contact with the bed, explain the absence of large extant ice shelves between 75.5 and 77∘ S.

The submarine geomorphology, ice-sheet geometry, and deglacial history of the Coats Land ice margin provide constraints on the future behaviour of the ice sheet in this region.

First, it is clear that loss of the connection between the ice shelves and the bed had a threshold impact on ice-shelf stability. The geomorphological evidence suggests that current absence of ice shelves south of the BIS (75.5–77∘ S) can be attributed to loss of contact with the bed, with the reverse-bed slopes (Figs. 2e, 3d) depriving retreating ice shelves of further pinning points.

This failure to reconnect with the bed is exemplified by the behaviour of the SWGT and the Dawson-Lambton Ice Stream tongues, which are essentially “failed ice shelves”. Historical records show successive extensions of these glaciers over the last 100 years have been of insufficient extent or thickness to re-establish contact with the topographic highs at the distal ends of their troughs. These have included a temporary > 10 km long ice tongue in front of the Dawson-Lambton Ice Stream surveyed during the 1958 International Geophysical Year (Admiralty Charts, British Antarctic Survey Archives), and an extension, then substantial calving, of the SWGT sometime between 1915 and 1955 (Thomas, 1973). The latter shows that the SWGT remains predisposed to large-scale calving events despite potential lateral buttressing from the Lyddan Ice Rise (Fig. 1c) and McDonald Ice Rumples (Fig. 10).

In this context the BIS can be considered close to failure as connection with the bed is limited to a very small (3 × 3 km) part of the McDonald Bank. This represents about 0.03 % of the surface area of the BIS–SWGT using the 27 950 km2 estimate of Humbert et al. (2009). Here the limited contact with the bed makes the ice shelf particularly vulnerable to only moderate changes in ice draught. A surge event in 1971 (Thomas, 1973) is consistent with a thinning of the ice shelf and a temporary loss of contact with the bed. As a result, ice-shelf velocities accelerated from 400 to > 700 m a-1 between 1968 and 1982 (Simmons and Rouse, 1984), immediately preceded by the formation of rifts upstream of the ice rumples in 1968 (Thomas, 1973). This behaviour is supported by an ice-flow model which shows that temporary loss of mechanical contact with the bed simulated for the 1971 local calving event can explain both a near-instantaneous 2-fold increase in ice velocities over a large section of the ice shelf and a subsequent decrease once contact was re-established with the McDonald Bank (Gudmundsson et al., 2016).

Second, the extant glacier tongues and ice shelves consist of ice which loses its structural integrity when flowing across the steep grounding line (King et al., 2018). For example, at the grounding lines of the Stancomb-Wills Glacier and the unnamed glaciers feeding the BIS, the marked change in surface gradient forms a “hinge zone” at the transition between the interior grounded ice and floating ice masses. This change in gradient known as the Brunt Icefalls (75∘55′ S, 25∘0′ W; Fig. 5) extends for about 80 km. Here, the ice-sheet fragments into a melange of icebergs forward of the grounding line (Anderson et al., 2014). These meteoric icebergs then “fuse” together with sea ice and with falling and drifting snow layers bridging the space between them to form a structurally heterogeneous ice shelf. Although sea ice has plastic (and therefore stabilising) properties (Holland et al., 2015), the meteoric ice units are more prone to fracture (Kulessa et al., 2014). However, observations on the BIS have shown that rifts preferentially skirt the edges of the incorporated icebergs and slow their rate of propagation when forced by the stress field to break through them (King et al., 2018). It is still not known whether the heterogeneous nature of these ice shelves will make them more or less vulnerable to break-up once detached from the bed.

A relatively stable ice-sheet configuration is reached once the calving fronts of ice streams and glaciers have retreated to the coastal grounding line. Here the bedrock rises above sea level within ca. 10 km from the coast, and the ice streams and glaciers calve directly into the Weddell Sea. This stable configuration is consistent with an ice-sheet model that suggests stability of the Coats Land ice margin during the last interglacial (DeConto and Pollard, 2016).