Drilling results and seismic reflection profiles indicate that the Antarctic Peninsula Ice Sheet has advanced to its western continental shelf edge many times since the late Miocene (Larter et al., 1997; Barker and Camerlenghi, 2002). Through repeated glaciations the ice sheet has eroded and over-deepened the inner shelf, extended the outer shelf through progradation, and delivered large volumes of sediment to the deep ocean (Barker and Camerlenghi, 2002; Bart and Iwai, 2012; Hernández-Molina et al., 2017). The AHT, a 140 km long by 50 km wide cross-shelf trough (Fig. 1), was a recurring ice stream pathway during glacial maxima (Larter and Cunningham, 1993; Larter and Vanneste, 1995). The most recent grounding zone advance to the shelf edge along the AHT occurred during the LGM (Pudsey et al., 1994; Heroy and Anderson, 2005; Ó Cofaigh et al., 2014). To the southeast of the trough, the inner shelf is incised by an erosional basin, Palmer Deep (PD), that measures 26 km east to west and 10 km north to south at the 800 m depth contour (Domack et al., 2006; Fig. 1). PD has a maximum depth >1400 m, yet to both the north and south there are small islands within 12 km of its axis and there is a bank directly to its west that rises to <200 m below sea level. Rebesco et al. (1998) and Domack et al. (2006) argued that in colder temperatures than today and with a lower sea level – e.g. at the start of Marine Isotope Stage 2 (29 ka) – ice encroached towards PD from the nearby land areas and local ice caps formed on emergent platforms around the present-day islands near the basin. These authors further hypothesized that continued glacial development led to the PD basin becoming completely encircled by grounded glacial ice and to formation of an ice shelf over it, trapping a subglacial lake. Based on multibeam bathymetry data from the inner shelf, Domack et al. (2006) described channels crossing the deepest part of the sill separating the western end of PD from the AHT that are 200–500 m wide and 100–300 m deep and exhibit reversals in their longitudinal profiles. On the basis of these characteristics and similarities to channels in Pine Island Bay and to the Labyrinth channels in Wright Valley in the Transantarctic Mountains that had previously been interpreted as having been eroded by subglacial water flow (e.g. Lowe and Anderson, 2003; Nitsche et al., 2013; Lewis et al., 2006), Domack et al. (2006) interpreted the channels as having been eroded by outflow from a subglacial lake in PD. More recently, geochemical analysis of pore waters from sediments in one of the basins within the channel network in Pine Island Bay confirmed that it had been a subglacial lake (Kuhn et al., 2017).

Extensive new data were collected on RRS James Clark Ross cruise JR284 in January 2014 using a 1∘ × 1∘ Kongsberg EM122 system with 432 beams and a transmission frequency in the range 11.25–12.75 kHz (Supplement Fig. S1). Beam raypaths and seabed depths were calculated in near real time using sound velocity profiles derived from conductivity–temperature–depth and expendable bathythermograph casts made during the cruise. Processing consisted of rejecting outlying values, replacing the sound velocity profile applied during acquisition with a more relevant one for some data, and gridding to isometric 30 m cells using a Gaussian weighted mean filter algorithm in MB-System software (Caress and Chayes, 1996; Caress et al., 2018). Pre-existing multibeam data, mostly collected on RVIB Nathaniel B. Palmer and previous cruises of RRS James Clark Ross (Anderson, 2005; Domack, 2005; Lavoie et al., 2015), and data along a few tracks collected more recently on HMS Protector using a Kongsberg EM710 system (70–100 kHz), were included in the grid (Supplement Tables S1, S2, and Fig. S1). Acoustic sub-bottom echo-sounding profiles were collected along all JR284 survey lines with a Kongsberg TOPAS PS018 parametric system using a 15 ms chirp transmission pulse with secondary frequencies ranging from 1.3 to 5 kHz. Vessel motion and GPS navigation data on cruise JR284 were collected using a Seatex Seapath 200 system.

MCS data used were collected in the 1980s on RRS Discovery cruises D154 and D172 (Larter and Cunningham, 1993; Larter et al., 1997). On D154, Line AMG845-03 was collected with a 2400 m long hydrophone streamer, whereas on D172 the streamer used to collect data on Line BAS878-11 was 800 m in length. In both cases the streamer was towed at 8–10 m depth and data were recorded from 50 m long groups with a sampling interval of 4 ms. The seismic source consisted of four airguns with total volumes of 8.5 L on D154 and 15.8 L on D172, respectively, and data were processed to a common midpoint stack using standard procedures. Very high-resolution seismic data were collected using a Huntec DTB, towed within 100 m of the seabed, on RRS James Clark Ross cruise JR01 in 1992 (Larter and Vanneste, 1995; Vanneste and Larter, 1995). This system transmitted a broadband pulse with frequencies of 0.8–10 kHz and a cycle time of 0.9 s. Data were recorded with a 100 µs sampling interval from a 1 m long hydrophone trailed behind the towed vehicle. The DTB system is capable of resolving sedimentary layers <1 m in thickness and also achieves very high spatial resolution due to the proximity of the source and receiver to the sea bed.

Integration of the new multibeam bathymetry data with pre-existing data provides nearly continuous coverage of the AHT from PD to beyond the continental shelf edge, with the new data spanning the full width of the trough on the middle shelf (Fig. 1). They also include coverage of the confluence with a tributary trough that joins the AHT from the east on the middle shelf. We will refer to this tributary by the informal name “Perrier Trough”, as it originates offshore from Perrier Bay, Anvers Island. The data reveal extensive areas of MSGLs and drumlins, which are characteristic of ice stream beds and show the pattern of palaeo-ice flow (Stokes and Clark, 1999; King et al., 2009; Graham et al., 2009). Fields of drumlins, with elongation ratios between 2.5 and 6:1, occur over a broad transition zone between the rugged inner shelf and smoother mid-shelf part of the AHT, and where the AHT crosses a structural high that separates middle and outer shelf areas (Larter et al., 1997). MSGLs in the mid-shelf part of the AHT have elongation ratios between 12 and 17:1, whereas some on the outer shelf have elongation ratios up to 80:1. The data also confirm the occurrence of several grounding zone wedges (Fig. 1), some of which had been identified previously, indicating positions where the grounding zone paused during retreat from its LGM position (Larter and Vanneste, 1995; Vanneste and Larter, 1995; Batchelor and Dowdeswell, 2015). Here, however, we focus on two specific areas in which the landforms observed have a bearing on the role of subglacial hydrology in facilitating and controlling ice stream flow.

In the southern part of the AHT there is a marked along-trough change in landforms across a line where a MCS profile (AMG845-03) shows that a mid-shelf sedimentary basin pinches out (Figs. 2a and 3), with acoustic basement cropping out to the southeast (Larter et al., 1997). The acoustic basement most likely represents Palaeozoic–Mesozoic igneous and metasedimentary rocks similar to those that crop out on the nearby islands (Storey and Garrett, 1985; Leat et al., 1995). The sedimentary strata in the basin are of unknown age, but it has previously been inferred that the youngest layers could be no younger than middle Miocene and the oldest layers may be early Tertiary or Late Cretaceous in age (Larter et al., 1997). Crescentic scours around the “upstream” ends of bathymetric highs and fields of anastomosing channels are observed in the area where the acoustic basement crops out (Fig. 2a). Among the anastomosing channels, the largest are up to 30 m deep and 250 m wide, although many are smaller (Fig. 2b).

In the axis of the AHT directly north of this zone, incised into the edge of the sedimentary basin, the new data reveal a set of northward-shoaling and narrowing valleys spaced 2–3 km apart (Fig. 2a). Individual valleys are up to 1100 m wide and 60 m deep at their southern ends (Fig. 2c), but become narrower and shallower northwards (Fig. 2c, d), ultimately petering out over a distance of <5 km. The slope angles along the steepest parts of the channel axes are ∼2∘ (Fig. 2d). In detail, the southern part of each valley exhibits a v-shaped deeper section incised into a u-shaped upper section, the v-shaped sections being up to 350 m wide and 45 m deep (Fig. 2c). MSGLs start directly north of these valleys (Figs. 2a and 4) and cover most of the seabed in the trough between this point and the continental shelf edge.

Two acoustic sub-bottom profiles that run transverse to the trough about 8 and 10 km north of the northern tips of the valleys show that the MSGLs were formed in the surface of an acoustically transparent layer that is draped by a 4 m thick layer of younger sediments (Fig. 4). Where such acoustically transparent layers have been cored elsewhere on the West Antarctic shelf they have been shown to consist of relatively low shear strength, deformed till, dubbed “soft till”, and the high-amplitude reflector below this layer has been shown to correlate with the top of a higher-shear-strength “stiff till” (e.g. Ó Cofaigh et al., 2005a; Reinardy et al., 2011). A homogenous, terrigenous diamicton with moderate shear strength, which was recovered in marine sediment cores from AHT, documents the presence of the soft till there (Pudsey et al., 1994; Heroy and Anderson, 2005). Each of the sub-bottom profiles shows three depressions in this high amplitude reflector that are 8–10 m deep and 400–800 m wide and contain an acoustically transparent fill, above which a reflector is usually observed separating the fill from the overlying soft till layer (Fig. 4).

We interpret the anastomosing channels and crescentic scours incised into the hard substrate on the inner shelf as having been eroded by subglacial water flow at times when grounded ice extended further offshore, as previous authors have interpreted similar features elsewhere (Dreimanis, 1993; Lowe and Anderson, 2003; Nitsche et al., 2013; Lewis et al., 2006; Graham and Hogan, 2016). This origin is also consistent with the interpretation that a subglacial lake was present in PD during the last glacial period, and channels incising the sill at its western end were eroded by outflow from the lake (Domack et al., 2006). Considering the scale of such features here, as well as in other areas around Antarctica, and the nature of the material they are eroded into, they probably developed progressively through multiple glacial cycles (Smith et al., 2009; Nitsche et al., 2013). The scale of the features also implies that water flow rates fast enough to drive erosion can only have been achieved through subglacial storage of water and episodic discharge (rather than continuous flow), even if it is assumed that the upper parts of scours and channels were filled with ice (Nitsche et al., 2013; Kirkham et al., 2019). This is, once again, consistent with the interpretation of a subglacial lake in PD, and also with a semi-quantitative model that implies outbursts from trapped subglacial lakes in such settings and of these approximate dimensions are likely to occur with repeat periods of the order of a few centuries (Alley et al., 2006).

A conservative estimate for the volume of the PD subglacial lake stated by Domack et al. (2006) was 20 km3. However, the full volume of the basin deeper than the sill depth of ∼500 m is about 110 km3. If this were filled the lake would have been nearly 2 orders of magnitude greater in volume than subglacial Lake Ellsworth (1.37 km3; Woodward et al., 2010), but still nearly 2 orders of magnitude smaller than Lake Vostock (5400±1600 km3; Studinger et al., 2004). The length and width of PD are similar to Lake Engelhardt, the largest of a number of connected shallow lakes beneath the lower Mercer and Whillans ice streams, from which remote-sensing data show ∼2 km3 of water drained between October 2003 and June 2006 (Fricker et al., 2007). However, water depth in these lakes likely rarely exceeds 10 m (Christianson et al., 2012), so their volumes are small compared to the potential size of lakes in deep basins such as PD.

Considering that the northward-shoaling valleys are located in the axis of the AHT directly north of an area containing anastomosing channels and crescentic scours, we interpret them as also having been eroded by subglacial water flow. We interpret the upper u-shaped sections of the valleys (Fig. 2c) as having been widened by glacial erosion, implying that the valleys have been overridden by ice since they were first carved. This is consistent with the suggestion that many features eroded into bedrock on Antarctic continental shelves developed through multiple glaciations (Graham et al., 2009). Even at times when there was active water flow, ice may also have filled a large part of the v-shaped lower sections. Palaeo-ice flow paths indicated by streamlined bedforms show that the area in the axis of the AHT where the valleys occur lies directly down flow from the sill at the western end of PD. MSGLs directly north of the valleys indicate that there was fast ice flow in this part of the trough, likely facilitated by the soft till layer that is seen as an acoustically transparent layer in sub-bottom profiles (see Alley et al., 1986; Ó Cofaigh et al., 2005a; Reinardy et al., 2011). The coincidence of the onset of MSGLs with northward disappearance of the valleys suggests that water supplied through them was important in lubricating and dilating the till, thus reducing its shear strength and making it more prone to deform under stress. Thus we infer that the northward terminations of the valleys were associated with a transition from channelized to distributed water flow at the ice bed. The shallow, filled depressions observed in the sub-bottom profiles (Fig. 4) suggest that the valleys once continued further to the north, before their distal reaches were filled and the overlying soft till in which MSGLs are formed accumulated beneath fast-flowing ice. The sequence of units observed in the profiles could have resulted from upstream migration of the onset of sediment-lubricated streaming flow during the last glaciation.

In the area of the confluence between AHT and Perrier Trough, an area of east–west-aligned MSGLs terminates abruptly along a line parallel to their trend on the southern flank of the Perrier Trough (Fig. 5a). The eastern limit of the area covered by MSGLs in AHT is more irregular, but lies within a band no more than 3 km wide on the eastern flank of the trough. Streamlined bedforms are absent from the area between the two troughs, but several steep-sided bathymetric basins up to 1500 m wide and 40 m deep are observed (Fig. 5a–c). The central parts of most basins are flat or gently dipping so that cross sections exhibit box-shaped profiles (Fig. 5b, c). A 500 m wide and 8 m high mound occurs on the northwestern side of one of the largest basins (Fig. 5c). A few of the basins span the boundary of MSGLs on the southern flank of Perrier Trough. Furthermore, about 3 km north of the boundary of MSGL and to the east of the other basins, linear features connect a small basin ∼300 m in diameter with a similarly sized mound 1.6 km to its WNW (Fig. 5a).

We interpret the well-defined southern lateral boundary of MSGL in Perrier Trough as indicating the marginal shear zone position when the palaeo-ice stream reached its maximum width during the LGM. Similarly, the lateral limit of MSGL on the eastern flank of the AHT lies within a band that is no more than 3 km wide, indicating the approximate position of the shear zone at the margin of the palaeo-ice stream occupying this trough when it was at its widest. Such clear expressions of ice stream shear margin positions in bed morphology have proved elusive beneath the modern ice sheet, partly because the chaotic structure of shear zones causes scattering of ice-penetrating radar signals that is observed as “clutter” in survey data (Shabtaie and Bentley, 1987). The steep-sided basins in the area between the two troughs are similar to the holes of hill–hole pairs and scars resulting from detachment of sediment rafts observed on the northeastern part of the Amundsen Sea continental shelf (Klages et al., 2013, 2015, 2016). The small mound and basin in the Perrier Trough 3 km north of the boundary of MSGLs appear to be a clear example of a hill–hole pair. Such features are generally regarded as characteristic of erosion and deformation beneath dry-based ice cover (e.g. Ottesen et al., 2005; Evans et al., 2006). The mound to the northwest of one of the largest basins may represent a corresponding hill (Fig. 5a, c), although as its cross-sectional area is smaller than that of the adjacent hole it cannot contain all of the excavated sediment. The absence of mounds close to some of the other basins may be explained by their close proximity to the palaeo-ice stream confluence, as the excavated material would only need to have been transported a short distance into the path of one of the ice streams to be entrained by faster flow. Evidence of hill–hole pairs having been overridden and eroded by ice after their formation has previously been reported from the Norwegian continental shelf, where some hills are observed to have streamlined tails (Rise et al., 2016). The pristine form of the hill–hole pair within the Perrier Trough indicates that it must have formed after ice stream flow had stagnated.

A DTB profile runs across the bathymetric basin that lies closest to the confluence of the two troughs and, beneath thin, patchily distributed younger sediments, it shows an acoustically transparent layer up to 25 ms (∼20 m) thick (Figs. 5a, 6). This layer has a minimum thickness of <3 m beneath the southeastern edge of the basin floor, and it thickens progressively to the northwest across the basin. In places short segments of truncated, dipping reflectors can be seen beneath the strong reflector at the base of the acoustically transparent layer (Fig. 6). We interpret the acoustically transparent layer as consisting of Quaternary diamictons overlying an unconformity above mid-shelf basin sedimentary strata that are represented by the truncated, dipping reflectors. The reduced thickness of the acoustically transparent layer across the basin suggests that the “hole” was formed by erosion of Quaternary diamictons and that the older sedimentary strata were unaffected during its formation. As mentioned above, such holes are generally regarded as characteristic of erosion beneath dry-based ice cover, and therefore the restriction of erosion to the Quaternary diamictons is consistent with the shear stress threshold for brittle failure in them likely being significantly lower than in the underlying consolidated strata (Evans et al., 2006). At the seabed near the foot of the steep southeast flank of the basin, a unit that is 180 m across and 7 ms (∼5.5 m) thick containing southeast-dipping reflectors is observed. We interpret this unit as a rotated slide block that has originated from the flank of the hole after its excavation. The most prominent reflector within the block is most likely to be from the boundary between diamicton and postglacial sediment as observed to the NW of the basin (Fig. 6), in which case the displacement of the block did not occur until after a layer of postglacial sediments several metres in thickness had accumulated.

A MCS profile that runs obliquely across Perrier Trough and continues over the inter-trough area shows mid-shelf basin sedimentary strata underlying the seabed along the entire line (Figs. 5a, 7). As noted above, the oldest strata in this basin may be as old as Late Cretaceous, but most of the basin fill is likely of Tertiary age (Larter et al., 1997). The southern lateral limit of MSGL in Perrier Trough lies within 1 km of the position where a unit of younger strata with a distinct seismic facies character (labelled “later mid-shelf basin” in Fig. 7) pinches out, but is not coincident with this boundary.