Subglacial features form beneath permanently grounded ice that is thick enough to offset buoyant forces exerted by the ocean. MSGLs (Fig. 2b), the most common subglacial features on the Antarctic continental shelf, are streamlined features with high parallel conformity (Clark, 1993). While the actual formation process for MSGLs is still debated (e.g. Tulaczyk et al., 2001; Shaw et al., 2008; Ó Cofaigh et al., 2008; Fowler, 2010), they are interpreted as having formed under streaming ice due to their association with modern ice streams (King et al., 2009) and their occurrence within palaeo-glacial troughs (Anderson, 1999; Livingstone et al., 2012). The streamlined nature of MSGLs makes them excellent indicators of ice-flow direction (Clark, 1993; Stokes and Clark, 1999; Shipp et al., 1999; Ó Cofaigh et al., 2002; Dowdeswell et al., 2004; Spagnolo et al., 2014). Previous studies have shown that MSGLs are associated with a massive seismic facies interpreted as the deformation till layer deposited above the latest glacial unconformity (Shipp et al., 1999; Ó Cofaigh et al., 2002, 2005; Heroy and Anderson, 2005). Most MSGLs in the Ross Sea have amplitudes of 1–9 m and lengths of about 1–10 km. As these features are associated with deforming till, MSGL amplitudes should not be greater than the thickness of the deforming till layer.

Smaller-scale streamlined features, with lengths of hundreds of metres to a few kilometres, comprise a number of landform classes such as drumlins, crag and tails, and megaflutes. We group these landforms here as a single class of drumlinoids. While their internal composition can be difficult to determine in the marine environment, and their formation mechanisms remain uncertain, this family of landforms is widely and most simply taken to record the former ice flow direction (Benn and Evans, 2010). In Antarctica, drumlinoids are most often observed at the transition between crystalline bedrock and sedimentary deposits (Wellner et al., 2001, 2006; Graham et al., 2009).

Subglacial meltwater channels have been reported from a number of locations on the Antarctic continental shelf, though almost exclusively incising crystalline bedrock on the inner shelf (e.g. Lowe and Anderson, 2003; Anderson and Fretwell, 2008; Smith et al., 2009; Nitsche et al., 2013; Witus et al., 2014). Channels in sedimentary substrates are rare, but have been previously observed on the Ross Sea continental shelf by Alonso et al. (1992), Wellner et al. (2006), and Greenwood et al. (2012), though their origin and link to subglacial meltwater is not evident.

Ice-marginal features form at the grounding line, marking the transition from permanently grounded ice to ice that has decoupled from its bed to become a floating ice shelf or a calving ice cliff. They include GZWs, marginal moraines, and linear iceberg furrows.

GZWs are depositional features (Fig. 2f, g) characterised by relatively steep foreset slopes that result in asymmetrical morphologies, broadly indicating ice-flow direction during GZW deposition. GZWs are formed during periods of stability of the grounding line. They grow as sediment is delivered to the grounding line through subglacial bed deformation and basal debris melt out (e.g. Alley et al., 1986, 1989, 2007; Anderson, 1999; Anandakrishnan et al., 2007; Dowdeswell et al., 2008, Dowdeswell and Fugelli, 2012; Batchelor and Dowdeswell, 2015). The growth of GZWs can stabilise an ice sheet against small-scale relative sea-level rise and ice-sheet thinning by reducing the minimum ice thicknesses necessary to counter buoyancy effects (Alley et al., 2007). Large GZWs can imply longer episodes of stability of the ice margin (Alley et al., 2007; Dowdeswell and Fugelli, 2012). Here, GZWs are grouped into three categories: small-scale, intermediate-scale, and large-scale. Small-scale GZWs have heights less than 10 m, cannot be traced across an entire trough width, and generally are only observable in high-resolution multibeam and side-scan sonar data (e.g. Shipp et al., 1999; Jakobsson et al., 2012; Simkins et al., 2016). Intermediate-scale GZWs range from 10 to 50 m heights and often display very sinuous fronts. Large-scale GZWs (Fig. 2g) have heights exceeding 50 m and extend across the entire trough width. The internal structure of large GZWs is occasionally detectable in low-frequency seismic data and includes distinct foreset beds indicative of GZW progradation (e.g. Anderson, 1999; Heroy and Anderson, 2005; Dowdeswell and Fugelli, 2012), but more often internal reflectors are not resolved in seismic data (e.g. Mosola and Anderson, 2006; Batchelor and Dowdeswell, 2015).

Ice-marginal moraines (Fig. 2e) are often symmetric in cross section (Dowdeswell and Fugelli, 2012), but can also display a slightly asymmetric shape (Winkelmann et al., 2010; Klages et al., 2013). They are generally believed to be formed by push processes (e.g. Batchelor and Dowdeswell, 2015). Low-amplitude (< 5 m) features with similar characteristics are sometimes interpreted as De Geer moraines, whose development is influenced by seasonal or cyclic processes (Hoppe, 1959; Lindén and Möller, 2005; Todd et al., 2007) or transverse ridges (Dowdeswell et al., 2008), which does not imply seasonal formation. Due to their limited amplitudes, these features are best resolved with high-resolution bathymetric mapping techniques, such as the EM122 multibeam system and side-scan sonar (e.g. Shipp et al., 1999; Jakobsson et al., 2011; Simkins et al., 2016).

Linear iceberg furrows exhibit high parallel conformity, often display a geomorphic expression similar to MSGLs, and are consistent with MSGL orientations. These linear furrows, however, are erosional features, whereas MSGLs are interpreted as either depositional or deformational features. The margins of marine-based ice sheets with low-slope profiles are particularly susceptible to tidal fluctuations, causing large areas of the ice sheet to intermittently contact the sea floor (Fricker and Padman, 2006; Brunt et al., 2010). We interpret linear furrows to form within this diffuse grounding zone, where ice is hovering at the buoyancy limit and cyclically contacting the sea floor. Alternatively (or additionally), linear furrows may be associated with ice-shelf break-up events when icebergs near the grounding line are held upright within an iceberg armada (MacAyeal et al., 2003; Jakobsson et al., 2011; Larter et al., 2012). Small repeating corrugation ridges have been observed within fields of iceberg furrows or within individual iceberg furrows (Fig. 2h, k; Anderson, 1999; Jakobsson et al., 2011; Klages et al., 2015). Although the exact mechanism for their formation remains somewhat controversial, corrugation ridges are thought to form as icebergs move vertically with tides, causing iceberg keels to intermittently contact the bed (Jakobsson et al., 2011; Graham et al., 2013). Their association with vertical tidal movement is based on the occurrence of identical features in proglacial arcuate iceberg furrows (Anderson, 1999) and comparison of corrugation amplitude and spacing with tidal modelling results (Jakobsson et al., 2011). Similarly, a deep keel capable of ploughing a linear furrow once ice has fractured could also have existed as an irregularity at the ice base prior to calving, forming linear furrows in a diffusive grounding zone. Both mechanisms for linear furrow formation signify that ice was still moving as a coherent body in contact with the sea floor. We argue that the linear forward motion that ploughs these furrows is caused by upstream ice flow; therefore, their significance for ice-flow direction is the same as MSGLs. For this reason, linear furrows and MSGLs are grouped into the inclusive term of “linear features.”

Shelf-edge gullies on high-latitude continental margins occur where streaming ice reached the continental shelf break (Fig. 2i). Although their origin remains uncertain, they have been attributed to many formative processes, including point sources of sediment-dense meltwater from the grounding line when it was situated at the shelf break (Anderson, 1999; Evans et al., 2005) or from small-scale slope failure due to accumulation of proglacial sediment (Gales et al., 2012). Both mechanisms imply proximity to the grounding line. Ross Sea shelf-edge gullies have not been extensively surveyed; however, a lack of significant sediment infilling suggests that they were active during the LGM (Shipp et al., 1999).

Arcuate iceberg furrows (Fig. 2j) are common features near continental shelf margins and on bank tops, overprinting any potentially pre-existing landforms. These are clearly proglacial features formed by freely moving icebergs that drifted under the influence of ocean currents and winds. Corrugation ridges have been observed within arcuate iceberg furrows, which is the most compelling evidence that these ridges result from tidal motion (Anderson, 1999).

The Drygalski Trough is the deepest region of the Ross Sea with water depths over 1000 m. Within this trough, the most seaward geomorphic expression of the ice-sheet grounding line is a large-scale GZW north of Coulman Island (D1, Fig. 3). This is consistent with previous interpretations of the maximum grounding line location (Licht et al., 1999; Shipp et al., 1999). A prominent set of MSGLs extends continuously from the Drygalski Ice Tongue to the approximate latitude of Coulman Island. A few small GZWs occur along the flanks of the trough; otherwise the MSGLs are not overprinted by recessional features. North of Coulman Island, both linear and arcuate iceberg furrows overprint GZW D1. The outermost shelf is covered by extensive arcuate iceberg furrows, which could have overprinted any older features.

Multibeam data are scarce in the southern Drygalski Trough, a key area for reconstructing the final phase of deglaciation in the WRS. Available data show a field of closely spaced, small-scale GZWs that back-step up the southern margin of Crary Bank, and a set of discrete intermediate-scale GZWs and lineations offshore of Mackay Glacier that record westward grounding line retreat (Greenwood et al., 2012; Anderson et al., 2014).

The JOIDES Trough is slightly fore-deepened on the outer shelf, relatively flat on the middle shelf, and slopes steeply into the deep inner shelf of the Central Basin (Figs. 1, 3). The outer portion of the JOIDES Trough is mostly devoid of linear features, with the exception of one group of straight furrows. High-frequency CHIRP data show a 4–8 m thick layer of acoustically laminated and draped glacimarine sediments on the outer shelf (Fig. 3a). LGM-age carbonates occur on outer shelf banks on both sides of the JOIDES Trough (Taviani et al., 1993; Fig. 3), precluding the presence of grounded ice at those locations. A large-scale, mid-shelf GZW (J1, Fig. 3) is seismically resolved (Shipp et al., 1999), although the GZW crest lacks clear expression in multibeam data. This mid-shelf GZW is separated from the next intermediate-scale GZW near the southern end of Crary Bank (J2, Fig. 3) by an extensive field of iceberg furrows followed by a continuous field of marginal moraines and small-scale GZWs. The southern JOIDES Trough is characterised by a series of meltwater channels associated with GZW erosional notches, observed at GZW J2 and ice-marginal features south of J2. The channels are incised into till deposited above the LGM unconformity and are occasionally overprinted by marginal moraines, therefore they are interpreted as subglacial channels that were active during the most recent glacial recession.

Pennell Bank and Ross Bank are linked across the Pennell Trough by a bathymetric high (referred to here as Pennell Saddle), separating the outer shelf part of the Pennell Trough from the deep Central Basin (Fig. 3). A large-scale GZW (P1) occurs at the northern margin of the Pennell Saddle and a thick (up to 14 m) package of layered glacimarine sediments extends northward from beneath the toe of the P1 GZW (Fig. 3b, c). Small-scale sinuous GZWs and relatively straight-crested moraines record the grounding line back-stepping from atop Pennell Saddle southward into the Central Basin. These recessional features overprint a large subglacial meltwater channel within the saddle (Fig. 2d).

The Central Basin is a bathymetric low that reaches water depths of over 1000 m, situated south of all three WRS troughs. It contains multiple generations of poorly preserved linear features, suggesting phases of large-scale ice stream flow reorganisation through the basin and McMurdo Sound (Greenwood et al., 2012). Numerous pockets of small, marginal moraines are found throughout the Central Basin and do not seem to be oriented parallel to depth contours.

The ERS contains three major troughs (Glomar Challenger Basin, Whales Deep, and Little America Basin) separated by low-relief ridges that are thought to have separated three palaeo-ice streams (Mosola and Anderson, 2006). Linear features dominate the ERS sea floor and extend to the continental shelf break (Fig. 3). They are associated with GZWs that are large enough to be identified in low-frequency seismic reflection data (Mosola and Anderson, 2006), (Fig. 2g; Fig. 3). Small- and intermediate-scale GZWs and moraines are confined to a few locations and no subglacial channels have been observed in the ERS. Shelf-edge gullies occur at the continental shelf break, implying the delivery of sediment and meltwater to a shelf-break grounding-line position.

The only drumlinoids observed in the Ross Sea occur on the inner shelf of Glomar Challenger Basin (Fig. 2c), covering ∼ 300 km2, and are associated with a near-surface occurrence of crystalline bedrock (Anderson, 1999; Shipp et al., 1999). Because these features are moulded predominantly from bedrock, they most likely formed over multiple glacial cycles. They do, however, exhibit highly uniform orientations (Fig. 2c) that are consistent with MSGL orientations seaward of the drumlinoids, indicating that the most recent phase of ice flow was likely responsible for the final drumlinoid shape. Extensive linear features occur throughout Glomar Challenger Basin (Figs. 2b, 3). They exhibit both trough-parallel and sub-parallel orientations, and are partitioned into discrete clusters based on orientation (see below). Legacy high-frequency CHIRP data in outer Glomar Challenger Basin show thin glacimarine sediments (Fig. 3d) and sediment cores sampled tills that typically occur within 1 to 2 m of the sea floor. Two closely spaced large-scale GZWs exist at the continental shelf break, observed in low-frequency seismic lines (G1 and G2, Fig. 3). These GZWs are wide and long but relatively thin so that they are not clearly observable in multibeam bathymetry. Two large-scale composite GZWs on the mid-to-inner shelf (G3, G4) are observed in both low-frequency seismic and multibeam data (Bart and Cone, 2012).

Whales Deep also contains a large-scale GZW at the continental shelf break (W1, Fig. 3), observed only in seismic data, as well as a mid-shelf GZW observable in both low-frequency seismic and multibeam records (W2, Fig. 3). A well-developed field of linear features extends from beneath the mid-shelf (W2) GZW to the continental shelf break. Linear features are notably absent south of W2. Little America Basin, like Glomar Challenger Basin, exhibits extensive linear features that extend across the entire trough to the shelf break. Three large-scale GZWs (L1-3, Fig. 3) are identified from low-frequency seismic data (Mosola and Anderson, 2006), but are too relatively thin to be observed in the legacy multibeam bathymetry, which has limited coverage and quality in this area.

We interpret the LGM grounding line in the outer Drygalski Trough to have been situated just north of Coulman Island, marked by the outermost GZW (cf. Shipp et al., 1999; D1, Fig. 3). Between Coulman Island and Drygalski Ice Tongue, a prominent cluster of MSGLs indicates trough-parallel flow (Fig. 3). Therefore, we interpret northward flow at the LGM from at least as far south as the David Glacier outlet (Drygalski Ice Tongue) to a grounding line north of Coulman Island.

In the JOIDES Trough, maximum ice extent is suggested to be recorded by the large-scale GZW (J1) on the mid-outer-shelf (Fig. 3). We base this hypothesis primarily on the presence of up to 8 m of draped glacimarine sediments in the outer trough shown in high-frequency seismic data (Fig. 3a). The observation of LGM-age carbonates on surrounding banks (Taviani et al., 1993) and the presence of LGM-age tephra layers in glacimarine sediments on the outer shelf (Licht et al., 1999) further support this interpretation. Straight furrows that occur seaward of this LGM limit are interpreted as iceberg furrows formed seaward of the LGM grounding line, rather than linear furrows, based on orientations that lack parallel conformity.

The LGM limit in the Pennell Trough coincides with the large-scale GZW (P1, Fig. 3), located ∼ 120 km landward of the shelf break (Howat and Domack, 2003). High-frequency seismic data show that this GZW prograded across thick glacimarine sediments that fill the outer trough (Fig. 3b, c).

Large-scale GZWs at the shelf break in each ERS trough (Fig. 3), linear features that extend across the outer shelf, and extensive shelf-edge gullies (Gales et al., 2012) indicate that grounded ice likely reached the shelf break in the ERS (Shipp et al., 1999; Mosola and Anderson, 2006). Thin glacimarine sediments on the outer shelf suggest a relatively shorter period of ice-free conditions than in the WRS, and would be consistent with a shelf-break LGM position.

Figure 3 shows the interpreted LGM grounding line and palaeo-flow directions derived from the seaward-most linear features that are assumed to represent LGM flow conditions. Generally, linear features delineate trough-parallel flow, which is consistent with previous LGM flow reconstructions (Shipp et al., 1999; Mosola and Anderson, 2006; Anderson et al., 2014).

With the exception of the Drygalski Trough, the WRS contains sparse and isolated patches of linear features, providing only glimpses of subglacial flow behaviour and direction despite extensive multibeam bathymetric coverage. Therefore, most palaeo-drainage interpretations in the WRS are based on ice-marginal features.

In the Drygalski Trough, the ice sheet decoupled from the sea floor and back-stepped rapidly from its LGM position near Coulman Island to a mid-shelf position at Drygalski Ice Tongue, as evidenced by the pristine nature of MSGLs and lack of overprinting ice-marginal landforms. South of Drygalski Ice Tongue, sparse data with poor quality results from the typical presence of pervasive sea ice. The most prominent deglacial features are a series of intermediate-size GZWs that back-step westward towards Mackay Glacier from a location north of Ross Island (Greenwood et al., 2012).

In the JOIDES and Pennell troughs, fields of closely spaced, small-scale GZWs and marginal moraines (Fig. 2d–f) dominate the sea floor, indicating that ice remained in contact with the sea floor during retreat. This implies that overall deglaciation was punctuated by pauses that were long enough to form a small recessional feature before retreating and forming another recessional feature. Retreat slowed and the grounding line stabilised in the southernmost part of the JOIDES Trough at an intermediate-scale GZW (J2, Fig. 3). A subglacial meltwater channel extending from GZW J2 to the south was likely linked to a large meltwater system that was active during deglaciation. We observe meltwater channels in the southern JOIDES and Pennell troughs, which are associated with retreat of the grounding line from positions of stability (J2, and the Pennell Saddle), leading to final rapid deglaciation of grounded ice in the two troughs. The effect of channelised subglacial meltwater on grounding-line stability is still under investigation.

Ice in the deep Central Basin appears to have retreated quickly, leaving only isolated clusters of recessional moraines. Based on the orientations of these moraines, we interpret a grounding-line embayment that opened over the Central Basin, followed by grounding-line retreat toward the east and west (Fig. 5). Fields of closely spaced, small-scale ice-marginal features in the Central Basin indicate that ice remained in frequent contact with the bed during deglaciation of this area. Because ice did not lift off from the deep sea floor first, we infer that retreat behaviour was controlled by a steep ice profile rather than physiography, as the ice did not decouple concentrically according to depth contours.

North of the Central Basin, extensive fields of small-scale GZWs and moraines record grounding-line retreat onto banks (Figs. 6a–b; 7), indicating that ice remained grounded on WRS banks during deglaciation. The presence of GZWs implies that ice was actively flowing across the banks and mobilising sediment in order to deposit these marginal features. Thus, WRS banks housed semi-independent ice rises during the late stages of ice sheet retreat from the WRS (Shipp et al., 1999; Anderson et al., 2014; Matsuoka et al., 2015). These findings are supported by modelling results that indicate the presence of independent, detached ice rises on WRS banks late in deglaciation (Golledge et al., 2014). Additionally, Yokoyama et al. (2016) argue that a grounded ice shelf remained pinned on WRS banks until the late Holocene.

Linear features on the ERS sea floor are overprinted only by large-scale GZWs (Fig. 3). These large-scale GZWs likely record periods of grounding-line stabilisation, punctuated by episodes of ice-sheet decoupling and grounding-line retreat that back-stepped tens to hundreds of kilometres in distance and preserved linear features.

We propose two alternative scenarios to explain the observed changes in flow orientation in the ERS. The first scenario (dynamic flow-switching model) is characterized by alternating regional flow direction throughout the LGM, followed by north–south recession of the grounding line (Fig. 8a). In the second scenario (embayment scenario), the ice stream occupying Whales Deep experienced extensive retreat, forming a large grounding-line embayment in the ERS (Fig. 8b).

The dynamic flow-switching scenario requires significant flow reorganisation with westward ice flow out of Marie Byrd Land (d1, Fig. 8a) followed by eastward flow across the inter-ice-stream ridge between Whales Deep and Glomar Challenger Basin (d2, Fig. 8a). Trough-parallel flow was then established (d3, Fig. 8a) and ice then began to retreat landward from the continental shelf in all ERS basins, interrupted by phases of grounding-line stabilisation and formation of the large GZWs in Whales Deep and Glomar Challenger Basin (d4, Fig. 8a). Different generations of MGSLs are preserved as the grounding-line retreats, but we would not expect them to be preserved if a readvance or major new episode of streaming had occurred, remoulding the bedform field. Although there have been examples of preserved flow fabrics during events of flow-switching (Stokes et al., 2009; Winsborrow et al., 2012) or at localised patches of basal friction (Stokes et al., 2007; Kleman and Glasser, 2007), the preservation of such extensive flow fabrics throughout three different ice flow configurations is unlikely.

The embayment scenario proposes the formation of an embayment over Whales Deep, based on the presence of large flow sets in surrounding basins that flow across neighbouring inter-ice-stream ridges into Whales Deep (flow sets g, k, Fig. 4). Trough-parallel flow likely occurred first (e1, Fig. 8b), as evidenced by the relatively undisturbed trough-parallel flow set in the outermost part of Whales Deep. During trough-parallel flow, ice grounded on inter-ice-stream ridges was likely sluggish and strongly coupled to the bed (Klages et al., 2013). An embayment in the Whales Deep grounding line formed (e2, Fig. 8b), drawing flow from outer Glomar Challenger Basin across the inter-ice-stream ridge into Whales Deep and depositing a large-scale GZW on the mid-shelf (W2, Fig. 3). The grounding-line embayment then retreated further towards the Whales Deep inner shelf (e3, Fig. 8b), drawing ice from Glomar Challenger Basin and Little America Basin, and prompting flow across the inter-ice-stream ridges into Whales Deep. The ice stream feeding Whales Deep at the LGM may have experienced stagnation or outrun its inner-shelf ice source, destabilising grounded ice on the outer shelf and causing an embayment to form. Modern Siple Coast ice streams have been observed to slow and stagnate (Anandakrishnan and Alley, 1997; Joughin and Tulaczyk, 2002), suggesting dynamic behaviour in the past.

Shipp et al. (1999) identify the inter-ice-stream ridges in the ERS as aggradational features, meaning that they were centres of focused sedimentation. Embayment grounding lines would have stabilised on the edges of the inter-ice-stream ridges on either side of Whales Deep, transporting sediment to these bathymetric features and aggrading the inter-ice-stream ridges. A large embayment over the ERS is also compatible with the interpreted WRS deglaciation pattern, where a steep EAIS profile is inferred. The formation of an embayment in the ERS is consistent with grounding-line recession in the ERS prior to the WRS (Fig. 6c), followed by east-to-west deglaciation of the WRS.

The two retreat models described here imply a succession of events that can be tested. Greater coverage of high-resolution multibeam data in outer Glomar Challenger Basin, illuminating cross-cutting relationships between flow sets, is crucial for establishing a relative chronology of cross-trough vs. trough-parallel flow. Additional multibeam surveys of inter-ice-stream ridges would also provide a better understanding of their role in directing the general flow pattern (cf. Klages et al., 2013). Furthermore, reliable marine radiocarbon dates constraining grounding-line retreat on the Whales Deep inner shelf might provide evidence for early retreat and the formation of a long-lived grounding-line embayment. Based on the available data in this study, the embayment scenario is favoured, due to the landform preservation issues inherent to the dynamic flow-switching model.

The Ross Sea geomorphological record permits us to reconstruct the pattern of ice flow and retreat independently of a radiocarbon chronology and the associated problems therein. Figure 7 presents reconstructed steps in grounding-line retreat that illustrate deglacial patterns across the Ross Sea. Observed grounding lines are linked together to form discrete episodes of deglaciation. These linkages are based on similar morphologies of observed GZWs, extension of grounding line orientations along bathymetric depth contours, and interpretation of local (albeit qualitative) retreat rates based on geomorphic features. The southern Drygalski Trough was the last area in the WRS to experience grounding-line retreat, as outlet glaciers (e.g. Mackay Glacier and David Glacier flowing into the Drygalski Ice Tongue) receded toward the west and north, leaving fields of moraines and GZWs (Greenwood et al., 2012; Anderson et al., 2014, Fig. 7). Drainage from the EAIS flowed into the Ross Embayment until the last stage of deglaciation (Fig. 7, steps 7–8). We infer a steep EAIS ice profile over the WRS throughout deglaciation, based on the contribution of EAIS ice through the Transantarctic Mountains and grounding-line recession unaffected by topography in the central WRS (Fig. 5).

Currently, there are two very different published retreat scenarios for the Ross Sea (Fig. 9). One of these models, the often cited “swinging gate” model (e.g. McKay et al., 2008; Hall et al., 2013), calls for a linear grounding line retreat across the Ross Sea, hinged just north of Roosevelt Island and extending to the Transantarctic Mountains (Conway et al., 1999). This model is constrained by the initiation of ice-divide flow over Roosevelt Island, and two locations along the Transantarctic Mountains with ages from ice-free coastlines, and it indicates deglaciation of the WRS at a faster rate than the ERS. The swinging gate model implies that controls on ice-sheet dynamics were the same throughout the Ross Sea and that physiography had little influence on ice retreat. This model also implies that ice-sheet retreat from the Ross Sea was controlled mainly by changes in the WAIS catchment, suggesting very high rates of southward retreat along the coast of the Transantarctic Mountains (Conway et al., 1999; Hall et al., 2013, 2015). Alternatively, the “saloon door” model proposes early retreat in the ERS with a grounding-line embayment in the central Ross Sea (Ackert, 2008). The implied drainage pattern of the saloon door model requires significant inputs from both the EAIS and the WAIS. This model is supported by cosmogenic exposure ages indicating a thinner ice-sheet profile in the central Ross Sea than at the margins of the WAIS (Parizek and Alley, 2004; Waddington et al., 2005; Anderson et al., 2014).

This study and previous marine studies (Licht et al., 1996; Cunningham et al., 1999; Anderson et al., 2014) suggest early grounding-line retreat within the northern Drygalski Trough, consistent with the swinging gate model. However, reconstructed grounding-line retreat on the remaining Ross Sea continental shelf contrasts with the swinging gate model (Fig. 7). In particular, our marine-based reconstruction suggests persistent EAIS drainage into the WRS throughout deglaciation, and indicates significant regional variations in grounding-line behaviour between troughs and across banks. Our reconstruction supports the presence of a grounding-line embayment in the ERS, similar to the saloon door model. Here, grounding-line recession in Glomar Challenger Basin is interpreted to precede retreat in the WRS (Fig. 6c), destabilising grounded ice in the southern Pennell Trough and the Central Basin. Deglaciation of the ERS prior to the WRS supports the observation of the EAIS as a persistent feature in the WRS throughout deglaciation.

Neither the swinging gate nor the saloon door model incorporate observations from the continental shelf, and, as we show here, are not able to fully capture the complexity of grounding-line retreat across the Ross Sea. Our new marine-based model (Fig. 7), reconstructed from comprehensive mapping of sea floor geomorphic features that directly record grounding-line retreat, can now be used to interpret more detailed Ross Sea palaeo-ice sheet behaviour and identify regional differences in deglacial behaviour. Our glacial geomorphic reconstruction independently converges with recent numerical modelling. Model results demonstrate significant EAIS and WAIS contributions to ice flow in the Ross Sea, and suggest that deglaciation was initiated in Ross Sea troughs and influenced by bedrock highs (Golledge et al., 2014; DeConto and Pollard, 2016; McKay et al., 2016). Additionally, DeConto and Pollard (2016) reproduce an early ERS grounding-line embayment confined to Whales Deep and a WRS Central Basin embayment receding to the east and west, while Golledge et al. (2014) simulate repeated occupation of WRS banks by semi-independent ice rises. Regional reconstructions between models and geologic observations are therefore becoming more and more consistent; however, smaller-scale patterns of grounding-line retreat are not yet reproduced at the resolution of modern numerical models. These localised retreat patterns are important for understanding grounding-line dynamics and smaller-scale processes that drive regional ice behaviour. A key target for further refining such efforts must undoubtedly be a robust and reliable radiocarbon chronology.

Many cycles of glacial erosion and deposition have led to Antarctic continental shelves characterised by a fore-deepened shelf profile with exposed bedrock on the inner shelf and thicker sediments on the outer shelf (Anderson, 1999). Runaway grounding-line retreat can occur as ice retreats from the outer to inner shelf due to a lack of pinning points to stabilise the grounding line (e.g. Mercer, 1978; Jamieson et al., 2012). In the WRS, banks and volcanic seamounts provided stable pinning points during deglaciation (Anderson et al., 2014; Simkins et al., 2016), and bathymetric highs continue to stabilise the modern Siple Coast ice sheet and ice shelf (Matsuoka et al., 2015). Ice-marginal features are observed to back-step up onto WRS banks (Fig. 6), demonstrating a strong physiographic control on grounding line behaviour. These banks served as pinning points for retreating ice streams and likely evolved into semi-independent ice rises during deglaciation. WRS banks supported an extensive ice shelf that buttressed WRS grounding lines and contributed to the long-lived presence of the EAIS in the WRS (Anderson et al., 2014; Yokoyama et al., 2016). While slight bottlenecking of ERS inter-ice-stream ridges may have played a role in determining positions of grounding-line stability and the formation of large-scale GZWs (Mosola and Anderson, 2006), the ERS sea floor is much more topographically subdued than in the WRS. A lack of high-relief banks and troughs permitted more variable flow in the ERS, but did not allow for pinning and stabilisation of ice streams as occurred in the WRS.

In addition to physiography, sea floor substrate has also been argued to exert a fundamental control on ice behaviour, as indicated by variations in geomorphic features across different substrates (e.g. Wellner et al., 2001; Larter et al., 2009; Graham et al., 2009). In Antarctica, studies have shown that ice streams flowing across soft, deformable sedimentary beds are characterized by MSGLs (e.g. Wellner et al., 2001, 2006; Ó Cofaigh et al., 2002, 2005; Graham et al., 2009). Ice flowing over unconsolidated beds can mobilise subglacial sediments and develop a thick layer of pervasive deformation till, facilitating faster ice flow than is possible by internal ice deformation (Alley et al., 1989). By contrast, crystalline bedrock or older and more consolidated strata outcropping on the sea floor are more resistant to glacial erosion, preventing the development of deforming till underneath a flowing ice stream, and are associated with bedrock erosional features such as drumlinoids that indicate slower ice-flow velocities and stick–slip motion. An excellent example is the field of drumlinoids in inner Glomar Challenger Basin that corresponds to a localised area of outcropping bedrock (Fig. 2c, Fig. 10). At the point where sedimentary deposits lap onto bedrock, these drumlinoids transition seaward into MSGLs (Anderson, 1999).

The compilation of geological data in Fig. 10 shows the strata beneath the most recent observable glacial erosional surface, representing the substrate that ice flowing across the continental shelf at the LGM would have encountered. The degree of consolidation of these strata is derived from information obtained from drill cores collected during Deep Sea Drilling Project Leg 28, extrapolated to high-resolution seismic stratigraphic correlations across the Ross Sea (Alonso et al., 1992; Anderson and Bartek, 1992; Anderson, 1999; Bart et al., 2000). The WRS is characterized by more variable geology and by older substrate, while mostly unconsolidated Plio-Pleistocene sediments blanket the ERS shelf. Thick and extensive unconsolidated sediments likely contributed to a pervasive layer of deformation till in the ERS (Mosola and Anderson, 2006). This thick layer of deformation till facilitated fast flowing ice and transported sediment to large-scale GZWs through a classic till conveyor-belt mechanism. Fast-flowing ice likely contributed to a low-profile ice sheet that episodically decoupled from the sea floor during retreat from the continental shelf (Mosola and Anderson, 2006). More consolidated strata outcropping in the WRS may have limited such pervasive subglacial deformation, potentially causing slower ice stream velocities in WRS troughs. This characteristic sea floor geology, coupled with numerous pinning points, was conducive to a higher profile ice sheet that remained in contact with the sea floor throughout much of its retreat from the continental shelf.

Grounded ice in Little America Basin flowed over its eastern bank and converged with an outlet glacier draining Marie Byrd Land (flow set n, Fig. 4). This flow pattern implies that at one point, Little America Basin was not able to drain all of the ice flowing into it and therefore some of that ice was forced eastward out of the trough. During the LGM, Little America Basin ice streams flowed across late Oligocene and Miocene sedimentary rocks (Fig. 10). Thus, it was more resistant to ductile subglacial deformation than the substrates encountered by other ice streams flowing across the ERS. Resulting flow velocities were therefore not high enough to transport all of the ice entering Little America Basin outlet, some of which was captured and funnelled into the neighbouring outlet.

Physiography exerts a first-order control on regional ice stream flow and retreat dynamics, and sea floor geology plays an important subsidiary role in controlling ice behaviour. These controls influence regional retreat patterns; more localised ice behaviour is still under investigation. Numerous other processes affect glacial dynamics, such as ice-shelf buttressing, sediment shear strength, and ice-bed coupling, and subglacial meltwater (e.g. Boulton et al., 2001; Dupont and Alley, 2005; Stearns et al., 2008). External forcings such as tidal effects, circumpolar deep water incursion and under-melting of ice shelves, as well as atmospheric effects are also influential (e.g. Rignot, 1998; Zwally et al., 2002; Arneborg et al., 2012; Walker et al., 2013). Ross Sea retreat was asynchronous between troughs, suggesting differential responses to these processes. Ongoing work on characterising Ross Sea glacial geomorphology highlights the effect of these forcings on local grounding-line stability.