Major Ice Sheet Change in the Weddell Sea Sector of West Antarctica Over the Last 5,000 Years

Until recently, little was known about the Weddell Sea sector of the West Antarctic Ice Sheet. In the last 10 years, a variety of expeditions and numerical modelling experiments have improved knowledge of its glaciology, glacial geology and tectonic setting. Two of the sector's largest ice streams rest on a steep reverse‐sloping bed yet, despite being vulnerable to change, satellite observations show contemporary stability. There is clear evidence for major ice sheet reconfiguration in the last few thousand years, however. Knowing precisely how and when the ice sheet has changed in the past would allow us to better understand whether it is now at risk. Two competing hypotheses have been established for this glacial history. In one, the ice sheet retreated and thinned progressively from its Last Glacial Maximum position. Retreat stopped at, or very near, the present position in the Late Holocene. Alternatively, in the Late Holocene, the ice sheet retreated significantly upstream of its present grounding line and then advanced to today's location due to glacial isostatic adjustment and ice shelf and ice rise buttressing. Both hypotheses point to data and theory in their support yet neither has been unequivocally tested or falsified. Here we review geophysical evidence to determine how each hypothesis has been formed, where there are inconsistencies in the respective glacial histories, how they may be tested or reconciled, and what new evidence is required to reach a common model for the Late Holocene ice sheet history of the Weddell Sea sector of West Antarctica.


Introduction
Quantification of future sea level rise is hindered by uncertainty regarding how the world's polar ice sheets will change under global warming scenarios (Intergovernmental Panel on Climate Change, 2013). While satellite observations have allowed an appreciation of how ice sheets are changing today (and over the last 40 years; Shepherd, 2018;Rignot et al., 2018) key to understanding how they will change in the future is knowledge of their most recent significant alterations. However, this knowledge requires that we have proficient geophysical measurements of ice sheet form and flow, and of past change, in the most vulnerable ice sheet regions.
The Weddell Sea (WS) sector of the West Antarctic Ice Sheet (WAIS; Figure 1) was, until 10 years ago, one of the poorest understood regions of Antarctica. Since then there has been a plethora of airborne and ground based geophysics (including ice-penetrating radar and solid Earth measurements), glacial geological and Marginal Basins, where it is overlain with highly deformed Palaeozoic metasediments . Aeromagnetic anomalies indicate that Palaeozoic metasediments and~13-km-thick Cambrian igneous rocks overlie the Proterozoic basement (2,500-541 Ma) in several sectors of the Ellsworth domain (Jordan et al., 2013).
The bed topography of the WS sector is both rough (over mountains and exposed bedrock) and smooth (e.g., across sediment-filled regions; . The deepest parts of the Robin Subglacial Basin (Figure 1a) are anomalously rough, marking the edge of a sedimentary drape (Siegert et al., 2016), while the subglacial topography of the region between the Robin Subglacial Basin and the Pirrit and Martin-Nash Hills is relatively flat, smooth, and gently sloping. Rose et al. (2015) defined this flat region as a  (Fretwell et al., 2013); (b) Ice velocity (Rignot et al., 2011); and (c) Ice surface elevation (Fretwell et al., 2013). (d) MODIS ice surface morphology (Haran et al., 2005;Scambos et al., 2007). The modern grounding line (Haran et al., 2005)  bedrock planation surface: a geomorphological feature that might be attributable by marine and/or fluvial erosion (Sugden & Jamieson, 2018). A series of large subparallel subglacial bed channels, thought to have been formed by the flow of basal water, lie between the MIS and FIS (Figure 1b; Rose et al., 2014). As these channels are presently located beneath slow moving and cold-based ice, the channels must be ancient, possibly formed at a time when the ice sheet was warm-based in this region and surface melting was prevalent in West Antarctica. Field evidence and modelling suggests that such conditions may have last existed in the mid-Miocene . In other parts of the WS sector higher bed elevations allow the RFIS to ground, creating ice surface features known as ice rises and rumples (Matsuoka et al., 2015), the largest of which include the BIR and HIR (Figure 1).

Glaciology
The SWSE is composed of three major ice sheet outlets (Figure 1): the IIS, MIS, and FIS, respectively, feeding ice to the Ronne-Filchner Ice Shelf, the second largest ice shelf in Antarctica after the Ross Ice Shelf. Subglacial topography and geology control the position and dynamics of these ice streams. (The downstream ice shelf is also a major control on dynamics and grounding line position.) The IIS and MIS grounding lines sit atop a steep reverse bed slope , similar in scale to that measured for upstream Thwaites Glacier. The bed slopes inland to a~2-km-deep basin (the Robin Subglacial Basin), which is divided into two sections, with few obvious significant ice sheet pinning points .
The IIS has three tributaries to the south and west of the Ellsworth Mountains, occupying the Horseshoe Valley, Independence Trough, and Ellsworth Trough (Bamber et al., 2000;Winter et al., 2015). The Horseshoe Valley trough, around 20-km wide and 1.3-km below sea level at its deepest point, is located downstream of the steep mountains of the Heritage Range. A subglacial ridge is located between the mouth of the Horseshoe Valley trough and the main trunk of the IIS (Winter et al., 2015). The Independence Trough is subparallel to the Horseshoe Valley trough, separated by the 1.4-km high Independence Hills. The trough is~22-km wide and over 1-km below sea level at its deepest point. It is characterized by two distinctive plateaus (~6-km wide each) on either side of the trough, which are aligned alongside the main trough axis. Presently, ice flows eastward through the Independence Trough for~54 km before it shifts to a northward direction, where the trough widens to 50 km at its connection with the main IIS. The Ellsworth Trough  is aligned with the Independence Trough; both are orthogonal to the orientation of the Amundsen-Weddell ice divide, dissecting the Ellsworth Subglacial Highlands northwest to southeast. The Ellsworth Trough is considered to be the largest and deepest trough-controlled tributary of the IIS Winter et al., 2015), measuring~34 km in width and~260 km in length. At its deepest point, the Ellsworth Trough extends~2 km below sea level. The trough contains the~10-to 15km-long Ellsworth Subglacial Lake Woodward et al., 2010). The Ellsworth Trough is intersected by several smaller valleys aligned perpendicularly to the main axis, which are relic landforms from a past small dynamic ice mass (predating the WAIS in its present configuration; . The MIS is smaller and slower than IIS and is supplied by ice from the upstream IIS tributary (Mouginot et al., 2017). A bifurcation south of the BIR sends most of the ice into the IIS trunk and a smaller fraction into the MIS. The shear margins of the two ice streams interact differently with the BIR. The IIS-BIR shear margin is controlled thermally; the IIS bed is wet while the BIR is frozen at its base (Siegert et al., 2016). The MIS-BIR margin is controlled tectonically by the Pagano Shear Zone (see below), which has created a subglacial trough, currently occupied by the MIS.
The trunk of the FIS occupies a very deep (~2 km) trench, fed by ice from both the WAIS (via an upstream tributary of the FIS), and the East Antarctic Ice Sheet (through Academy Glacier). The grounding line of the FIS rests on a horizontal bed and has been shown to oscillate across the bed by~20 km as a consequence of tidal forcing, where the trunk of the ice stream is floated at high tide and grounded at low tide (Jeofry, Ross, Le Brocq et al., 2018).
The flow of subglacial water is critical to the ice dynamics of the WS sector of the WAIS. Jeofry, Ross, Le Brocq, et al. (2018) calculated the pathways of basal water flow using recently constructed bed and surface DEMs. This study reveals how water flows beneath the tributaries and into the trunks of the ice streams. It also shows how each of the ice streams interacts distinctly with this water. For IIS, the greatest surface ice flow velocities are recorded across a well-defined sedimentary drape, which is predominantly fed by 10.1029/2019RG000651 Reviews of Geophysics water from the upstream IIS catchment. This water also feeds the Robin Subglacial Basin, where it exits the ice sheet as a point source, etching a channel into the ice shelf above (Le Brocq et al., 2013). This water acts to weaken the sediment, so allowing the fastest ice velocities above it (Siegert et al., 2016). For FIS, significant levels of water are provided by Academy Glacier (i.e., from East Antarctica), in which over a dozen "active" subglacial lakes exist all the way upstream to the South Pole (Wright & Siegert, 2012;Jordan et al., 2018;Smith et al., 2009). Across the floor of the ice stream trunk, flow-parallel hard-rock landforms channel water so that it too exits as a point source, leading to an ice shelf channel >130 km in length (Jeofry, Ross, Le Brocq, et al., 2018).

Sensitivity of the Region to Future Change
The discovery that the IIS and MIS grounding lines are perched on the top of a major reverse bed slope revealed that they are potentially at a threshold of substantial glaciological change through marine ice sheet instability . This concern was added to by numerical ocean modelling, which predicted enhanced flow of warm water to the WS grounding lines by the end of this century as a consequence of wind-pattern changes offshore (Hellmer et al., 2012), suggesting the potential for future ice losses from this sector equivalent to~0.3 m of sea level rise. To investigate the sensitivity of the WS sector ice streams to future ocean-driven warming, Wright et al. (2014) used the BISICLES ice sheet model (which includes appropriate grounding line physics) to determine grounding line dynamics of WS ice streams under ocean warming scenarios, leading to an appreciation of the conditions that deliver grounded ice retreat and an understanding of the regions most sensitive to such change ( Figure 2). The model was first run using best estimates of basal ice sheet parameters, ice accumulation and grounding line melt rates. The first "control" experiment ran this model for 2,000 years, with the hope of replicating the present-day ice sheet configuration. However, in this run, the IIS began to lose mass by around 200 years and experienced accelerating mass loss in the subsequent 1,800 years. By the end of the model run, it was losing around 8 km 3 of ice annually. Similarly, MIS lost mass at a rate of 1-2 km 3 per year. This behavior suggests that the two ice streams are not in equilibrium with the imposed forcing fields, so either these fields are wrong or the ice streams are on the brink of major change. However, as Wright et al. (2014) state, the first run should "be viewed solely as a neutral starting point from which to assess the relative sensitivity of the ice streams to change, and not as a prediction." The model was then used in a series of experiments that adjusted the levels of ice accumulation and basal iceshelf melting. Approximately, "stable" ice sheet configurations were obtained for runs where basal melting is reduced compared to the reference run, by as much as half the estimated values of today. However, when small increases were applied to the model, the IIS and MIS experienced mass loss and ice sheet retreat ( Figure 2). Indeed, in all runs where increased melting was applied these two ice streams lost mass unabated throughout the 2,000 years of the experiment. Other ice streams, such as Rutford Ice Stream and Carlson Inlet, were relatively unaffected in the experiments, suggesting they are insensitive to small levels of increased rates of basal melting. However, when values around 25% more than today's are used in the model, these glaciers also experience mass loss and ice retreat. Clearly, the model reveals that the IIS and MIS are the most sensitive ice streams in the WS to future oceandriven change. Indeed, the IIS appears to be the most sensitive of the entire WAIS in this regard [Cornford et al., 2015]. Their retreat is halted when the entire Robin Subglacial Basin is deglaciated, and the bed slope returns to "normal" (as opposed to reverse). Hillenbrand et al. (2014) compiled ice core data (from Berkner Island and EPICA-Dronning Maud Land), marine geophysical measurements of former glacial landforms, marine sediment core information between the ice-shelf margin and the continental-shelf break, and onshore cosmogenic nuclide dating of rock exposures, to assemble two possible scenarios (MAX and MIN) describing the retreat of the WS sector between 20,000 years ago (i.e., the LGM) and 5,000 years ago (the mid-Holocene). The two scenarios were based on different interpretations of dated material and on the style of glaciation.

LGM Ice Sheet Configuration and LGM to Mid-Holocene Change
Scenario MIN (Figure 3) is based largely on the onshore cosmogenic record (Bentley et al., 2010;Whitehouse, Bentley, & Le Brocq, 2012) and assumes that the oldest marine sediment dates represent minimum ages for post-LGM ice sheet retreat. It also assumes that subglacial landforms on the present-

10.1029/2019RG000651
Reviews of Geophysics day seafloor are pre-LGM in age. Scenario MIN also utilizes an ice sheet model that matches where former ice surface elevation and extent have been constrained using dated onshore exposures. Scenario MAX ( Figure 3) assumes that dates from sedimentary cores represent a mixture of maximum and minimum ages for the LGM and its retreat, and that the style of glaciation here was similar to that interpreted elsewhere in Antarctica at the LGM, where ice advanced to the continental shelf break. Scenario MAX requires that the vertical limits of fresh erratics around the WS embayment do not record the LGM limit Clark, 2011) and that the ice was substantially thicker without leaving a detectable geomorphological record. Importantly, both scenarios are consistent with low excess ice volumes during the LGM suggesting the region could only make a minor contribution to global meltwater pulses during the last deglaciation.
Scenario MIN at the LGM involves grounded ice extending across what is now the central RFIS to the continental shelf, and floating ice extending across the channels to the east and west (e.g., the Thiel Trough). At 15,000 years ago, scenario MIN accommodates modest expansion of grounded ice across the continental shelf, where ice sheet extent is limited seaward of the KIR and HIR, and the Thiel Trough is largely free of grounded ice (i.e., modest advance of the FIS). Grounded ice is predicted north of Berkner Island, but absent to the east and west of this lobe. By 10,000 years ago, grounded ice continuously covers the area between the present grounding line and the KIR and HIR. Grounded ice around Berkner Island is more extensive than today but is now cut off from the grounded WAIS. By 5,000 years ago, the WAIS margin was similar to today, Berkner Island still contained more grounded ice than today and the KIR and HIR were joined but separated from the WAIS.
Scenario MAX starts with the grounded ice sheet fully expanded to the continental shelf, with continuous grounded ice across what is now the FRIS. By 15,000 years ago, only modest retreat from this maximum position has occurred, and by 10,000 years, continuous ice cover remains between Berkner Island, the Thiel Trough, and KIR/HIR. By 5,000 years ago, the extent of grounded ice remains substantial; indeed, greater in some places than the LGM configuration in scenario MIN.

Background to Radio-Echo Sounding
The two post-LGM scenarios discussed above are obviously in apparent conflict, although both feature a retreat from far out on the continental shelf at the LGM back toward the present-day configuration (Hillenbrand et al., 2014). Our review now concentrates on reconstructing the ice sheet history from the mid-Holocene (~5,000 years ago), using a combination of glaciological data derived from glacier-geophysical investigations, with support from cosmogenic data in locations where both are in close proximity. The use of geophysical data from the ice sheet itself is possible and appropriate, as the ice itself is aged up to and older than 5,000 years (accounting for accumulation, vertical deformation, and lateral transport); hence, changes to ice flow conditions around this time are recorded englacially. While we cannot comment on which of the two LGM scenarios are most accurate, we will comment on which of the two 5,000-year reconstructions best fit with these observations. The prime data set used here is radio-echo sounding (RES), which can be mounted on aircraft or pulled by snowmobile. The RES technique operates through the emission of a radio wave which propagates through air and into the ice sheet and is reflected off any boundaries separating media with differing dielectric properties (Dowdeswell & Evans, 2004;Siegert, 1999). Radio wave reflections can be identified in RES data from the ice surface and the subglacial bed. In this way, information about Antarctic ice thickness, bed conditions, LGM ice thickness (m) difference relative to present Johnson et al., 2019;Nichols et al., 2019). GPS sites are shown by the blue triangles. Black areas are nunataks; grey line is the modern grounding line. The background is the Landsat Image Mosaic of Antarctica (LIMA) (Bindschadler et al., 2008) in combination with a hill shade of the Weddell Sea bathymetry from Bedmap2 (Fretwell et al., 2013). and the distribution of subglacial water and lakes has been established (Dowdeswell & Evans, 2004). In addition, Antarctic RES data show a number of strong reflections from within the ice sheet. The most widely noticeable of these englacial reflections are known as internal layers (Siegert, 1999).
Because a radio wave reflection will occur at a boundary of dielectric contrast, the origin of internal layers must be a response to dielectric boundaries within the ice sheet column. There are two possible causes of dielectric variation within the ice sheet: (1) ice density and/or fabric changes and (2) chemical properties of ice (see Siegert, 1999). Observations of density variation within ice cores indicate that below about 500 m of ice depth it will be too small to affect significant dielectric boundaries. Thus, an ice density origin for internal layering is only valid for the uppermost layers of the ice. However, variation in the chemistry of ice and crystal fabric can occur throughout the ice column. The most commonly held view is that dielectric variation is best explained by acidic layering within the ice sheet. Acidic layers originate from an aerosol product derived from volcanic eruptions that subsequently precipitates on the ice surface during snowfall. Burial of this snow eventually causes discrete layers of acid-laden ice at depth within the ice sheet. As ice flows within the ice sheet, these acid layers will deform accordingly. This deformation should result in folding of layers but not necessarily cause discontinuities. Thus, internal RES layering below 500 m is expected to be continuous over several tens to hundreds of kilometers within central slow flowing regions of Antarctica (Siegert, 1999). Such internal layers obey the law of superposition and represent isochronous surfaces. However, where ice flows more quickly, internal layers buckle and become significantly disrupted and discontinuous. The transition between continuous and buckled layers mark shear zones of ice streams or flow over major subglacial obstacles. If ice flow conditions have changed, the record of former ice flow is thus contained within englacial layers but may be advected downstream or overprinted by subsequent accumulation. Hence, internal layer conditions are especially useful in determining ice-flow modifications in the last few thousand years.
Other englacial layers have been established close to the beds of ice sheets. Where englacial stress is enhanced as ice flows around large subglacial obstacles, crystal fabrics can develop, leading to permittivity layers. Fracture and crevassing of the ice at depth may also lead to distinctive changes in dielectric properties. In addition, layers resulting from accretion of ice from the bed, potentially also involving ice fabric development, have been seen in both slow (Bell et al., 2011;Wrona et al., 2018), enhanced (Wolovick et al., 2014) and in transitional (Bons et al., 2016;Ross & Siegert, 2014) regions of the ice sheet. In such cases, basal "accretion" layers mark a boundary between isochronous ice (conformable with the basal layer, not necessarily the bed) and a basal ice unit with few noticeable layers or features.
Further englacial reflectors may result from basal sediment inclusion (Siegert et al., 2013;Winter et al., 2019) and from where ice is, or has been, floating to form basal crevasses (Kingslake et al., 2018;Wearing & Kingslake, 2019). In all cases, interpreting past flow from internal layers requires consideration of how changes in time may affect layer arrangements. This becomes especially challenging where more than one type of change has occurred, leading to secondary episodes of deformation that may be difficult to distinguish. Nonetheless, analysis of internal layers and structures has proven very useful in determining the recent flow history of both the WAIS (Catania et al., 2012;Conway et al., 1999;Holschuh et al., 2018;Martıń et al., 2006; and EAIS .

Glacial History of the SWSE Sector From Internal Layering and Basal Features
We summarize englacial structures across the SWSE starting from the ice divide, across the transition to fast flowing ice and onto ice shelves and ice rises. The center of the WAIS contains some of the clearest and bestpreserved internal layers on the continent (Siegert & Payne, 2004). Two sets of evidence support the idea that the ice-divid position is unlikely to have changed significantly between the LGM and today. First, by calculating former rates of ice accumulation necessary for the measured distribution of isochronous layers in a transect across the ice divide, it seems likely that the spatial pattern of accumulation has been unchanged in the last 20,000 years (Siegert & Payne, 2004). If the ice divide had changed, the modification to the orographic influence on ice accumulation would lead to an accumulation distribution different in the past compared with today, as observed in Taylor (Morse et al., 1998) and Talos (Siegert & Leysinger Vieli, 2007) ice domes. This is not observed across the WAIS, however, indicating ice-divide stability. Second, small folds within internal layers, caused by ice flow across a rough bed, track downstream with ice flow. By tracing the troughs and peaks of internal layer folds, a pathway of former ice flow can be established, the date of which begins from when the furthest fold from the rough topography was positioned over it. Ross et al. (2011) calculate stability of the WAIS divide based on an assessment of internal layer structures over Lake Ellsworth experiencing no detectable change from the modern ice flow direction. A similar match between layer structures and modern ice flow is found on the WS side of the ice divide (Ross & Siegert, 2014) and in the Siple Coast (Holschuh et al., 2018). Elsewhere on the Siple Coast side of the divide, however, ice flow direction has been shown to vary by around 30°due to ice relaxation and retreat of grounded ice across the Ross Shelf since the LGM .
Further downstream, ice flows into the trunk of IIS through multiple subglacial valleys. Winter et al. (2015) analyzed the internal layers within the Independence, Ellsworth, and Horseshoe Valley troughs, concluding that while the pattern of continuous versus disrupted layers is consistent with modern ice flow, supporting little ice flow direction change, evidence exists for former enhanced flow in these tributaries. Horseshoe Valley also contains basal ice structures associated with subglacial hills and pinnacles that are best explained as inclusions of basal sediment (Siegert et al., 2013;Winter et al., 2019). The feature can be tracked downstream to the trunk of IIS, but no internal layers exist within it, preventing further tracking ( Figure 4).
The next location to consider is the BIR, adjacent to the IIS and MIS, which contains key glaciological data constraining the Holocene glacial history of the SWSE. Presently, the BIR has only very slow flowing ice. Most stable ice rises that are frozen at their beds have well-established "Raymond arches" beneath their ice divides, which form as a consequence of non-linear variation in ice rheology with strain rate and stable flow over one characteristic time; the time taken to accumulate the full ice thickness (Martıń et al., 2009;Raymond, 1983). Ice rises that do not contain Raymond arches are either too young or have experienced ice divide migration. The BIR contains a small Raymond arch beneath its divide, suggesting recent stability, but not long enough for a major arch to develop (Martıń et al., 2009). Satellite imagery reveals the BIR to have relic flow stripes across its surface ( Figures 5 and 6), indicating the direction of former ice flow was similar to the present flow direction of the adjacent MIS (Siegert et al., 2013). Beneath the surface, radar has measured buckled internal layers under continuous layers, indicative of ice-dynamic processes being suddenly replaced by stagnant flow (Figure 6). One conclusion to be drawn from these two data sets is that past ice flow across the BIR was aligned with the surface stripes at enhanced flow speeds. Flow would have tracked across what is now the trunk of IIS, emanating from the three tributaries the other side of it (i.e., Ellsworth, Independence, and Horseshoe valleys) (Winter et al., 2015). Siegert et al. (2013) used a simple one-dimensional ice compression model and the modern rate of ice accumulation, to estimate the age of the transition between continuous and buckled layers at around 4,000 years ago.
The radar data also show the bed of BIR to be extremely smooth and continuous between it and the IIS trunk. Hence, the shear margin between the IIS and BIR should be controlled by basal hydrology and thermomechanics; on one side, the IIS is wet based and on the other the BIR is cold based ( Figure 6). The shear margin between BIR and MIS is tectonically influenced by the Pagano Shear Zone (Jordan et al., 2013) and may represent a more stable border between fast and slow flow. In summary, these data show that the IIS used to once track across the BIR and over what is now its hydrologically controlled shear margin.
Another interesting englacial feature across BIR is an apparent basal layer (Siegert et al., 2013; Figure 6), which looks remarkably similar in appearance to the sediment inclusion observed in Horseshoe Valley (Winter et al., 2019). It can be traced across the entire BIR along a former flow band (i.e., flow stripes). Interpolation upstream (across what is now the IIS trunk) links it to the feature in Horseshoe Valley, though without drilling through the ice and sampling it is not possible to be certain whether this is the same feature. An alternative explanation for the BIR basal feature is that it represents crevasses along a former shear margin, that have been subsequently filled with salt water.
To the north of BIR, adjacent to the ice plain that separates the grounded ice from the ice shelf, englacial folds can be observed, which are consistent with recent localized advance (i.e., 10-20 km) of the grounding line and expansion of the ice rise. Further downstream, embedded within the Ronne Ice Shelf, are the KIR and HIR and the Doake Ice Rumples (Figure 7). Ice rises are independent flow features with ice flow emanating from their center to the grounding line surrounding them. In contrast, at ice rumples, an ice shelf becomes temporarily grounded and slows significantly before becoming afloat at the downstream margin.
The flow of the ice shelf is heavily modified by the presence of the relatively slow-flowing ice rises and rumples. HIR is approximately 100-km northeast of BIR. Ice from the MIS flows pasts its eastern flank, ice from IIS flows past HIR to the west over Doake Ice Rumples, and ice flowing off the front of BIR flows very slowly toward HIR before being diverted westward at HIR's southern flank. KIR lies approximately 200 km to the west of HIR, with ice from IIS flowing past both flanks.
In terms of glaciological setting and planform size, HIR and KIR are similar. Ground-based, 4-MHz RES surveys conducted in 2013/2014 and 2014/2015 showed that they differ significantly in thickness and in englacial structure. KIR consists of a single ice divide, approximately 150-km long, running approximately SW-NE. KIR is approximately 600-m thick at its summit. The bed of KIR is flat and smooth and englacial layers are smooth and conformable everywhere that we have RES data. Beneath the ice divide are prominent Raymond Arches running the length of the ice divide (see Kingslake et al., 2016).
HIR consists of two ridges, each approximately 100-km long, extending to the north and southwest from a peak in surface elevation in the southeast. In contrast to the surface topography, a basal high point is  (c), highlighting the basal topography (brown), lower basal ice unit (B1) and upper basal ice unit (B2; as described by Winter et al., 2015) as well as visible englacial stratigraphic layers (black). The secondary axis is the satellite-derived surface ice flow speed from MEaSUREs version 2, shown in the figure as an orange line (Rignot et al., 2011). located at the northern end of the northern ridge (Wearing & Kingslake, 2019). HIR is 400-to 800-m thick.
There are only small, underdeveloped Raymond Arches in HIR, and these are restricted to the northern end of the northern ridge. Elsewhere, Raymond Arches are undetectable. This suggests ice flow has not been stable here for the characteristic time (Martıń et al., 2009) of approximately 4.5 kyr (estimated from the radar-derived measured ice thickness and accumulation rates from satellite data (Arthern et al., 2006) and modelling ). From the ground-based RES survey, continuous regions of bright tilted reflectors are observed at the base of HIR up to 12-km inland from the grounding line of the northward pointing promontory (Figure 7). These reflectors are distinct from the ubiquitous internal layers also observed in HIR. Individual dipping reflectors extend vertically 150-200 m into the ice from the base and span up to 2 km horizontally. The isochronal layers intersected by the dipping reflectors

10.1029/2019RG000651
Reviews of Geophysics decrease in age as one moves toward the grounding line (Wearing & Kingslake, 2019). This is consistent with these reflectors being relic basal crevasses, formed in a previous flow configuration, when the grounding line was upstream of its present location. Basal crevasses likely formed as vertical features at former grounding lines and filled with seawater which froze, leading to the high dielectric contrast that generates the reflection in radar data. The ice rise grounding line advanced through this region and, once grounded, basal crevasses were deformed by vertical shear leading to their present day tilt.
Also found within these regions of bright reflectors are troughs in internal layering and layers that intercept the bed. These features are continuous across radar lines. It is thought that these features form as a result of concentrated basal melting at former grounding lines, as observed at former shear margins on the Siple Coast (2) the flat bed beneath BIR; (3) the steep reverse slope of IIS; (4) buckled RES layers beneath flat layers (solid arrow); and (5) a tilting englacial reflection emanating from the bed (dashed arrow). (Catania et al., 2006). The spatial distribution of relic basal crevasses and melt synclines (Figure 7) suggests ungrounding of the northern basal high, as part of a complete or only partial ungrounding of HIR (Wearing & Kingslake, 2019). We expect ice rises to remain grounded over their high points if they form during a period of ice sheet retreat (Favier & Pattyn, 2015), so these observations suggest complete ungrounding, but we cannot rule out partial ungrounding.
The timing of the grounding line advance surrounding HIR toward its present-day position is estimated to be 6 ± 2kyr (Wearing & Kingslake, 2019); based on a model that simulates the vertical deformation of basal crevasses from their height at formation to the height as observed with RES, while accounting for uncertainty in accumulation, firn density, RES-derived depth, ice-thickening history, initial crevasse height and glacial isostatic adjustment (GIA).
Overall, ground-based RES surveys of HIR and KIR have highlighted stark differences in their internal structure that has been interpreted to be reflective of the two features having contrasting recent flow histories.

Cosmogenic Nuclide Methods
In situ cosmogenic nuclide dating of formerly exposed rock and sediment is the only feasible method to date the timing and trajectory of past ice thickness changes in Antarctica. Cosmogenic nuclides are formed in material at the Earth's surface through interaction with secondary cosmic radiation. The flux of cosmic radiation decreases exponentially in the subsurface as particles interact with the solid Earth. Many cosmogenic nuclides are produced through this interaction, and their concentration in surface rocks can be used as a chronometer indicating the duration of time spent at or within the upper few meters of Earth's surface (Lal, 1991). The method is widely used to reconstruct ice histories because glaciers are highly erosive and generally effective at eroding their beds deeper than a few meters. Therefore, erratics, moraine boulders, or the ice-molded bedrock left behind as the glacier recedes are usually "fresh," with the cosmogenic nuclide inventory having been fully reset. In this way, given knowledge of production rates, the measured concentration of cosmogenic nuclides that have built up subsequently in the dated surface can be used to determine an apparent "exposure age" for that surface. The apparent exposure age may closely date ice retreat or thinning from that position providing the surface has subsequently remained continuously exposed with minimal rock surface erosion. The exposure dating method has been used extensively in Antarctica to constrain past ice sheet elevation change by dating erratics deposited on the flanks of nunataks as the ice sheet thinned. In this way, the nunataks act as dipsticks that record the changing elevation of the ice sheet through time, and the approach has proved highly effective at many locations in Antarctica Jones et al., 2015;Balco, 2011;Mackintosh et al., 2007;Stone et al., 2003).
Exposure dating in Antarctica is sometimes complicated by ineffective erosion beneath and burial by nonerosive ice. In this case, the cosmogenic nuclide inventory is not fully reset between periods of glacial cover. Where erratics and sediment have been buried by cold-based ice, the assumption of continuous exposure since deposition is violated, and exposure dates from these surfaces will not date the last thinning of the ice sheet. These and other processes can introduce geological scatter into cosmogenic nuclide data sets, which complicates the interpretation of exposure ages in terms of ice sheet thinning. Measurement of multiple cosmogenic nuclides with different half-lives can be used to assess whether a rock surface has experienced a simple exposure history involving continuous surface exposure or a complex exposure history involving periods of burial under ice. In the latter case, the shorter-lived isotope will preferentially decay when buried by ice, causing a detectable change in the isotopic ratio (Lal, 1991). Most studies to date have used the cosmogenic 26 Al/ 10 Be pairing, since both isotopes are produced in quartz and are easily measured in the same rock. However, the half-life of 26 Al (0.705 Ma) is just half that of 10 Be (1.389 Ma), so only very long periods of burial on the order of hundreds of thousands of years can be detected using this method. More recently, measurement of in situ 14 C in quartz has become possible, and its short half-life (5,730 years) makes it highly sensitive to exposure and burial over the past~25 ka. For example, 14 C paired with a longerlived isotope, such as 10 Be, can detect burial periods as short as a few thousand years, making it highly useful for unravelling ice sheet change over the Late-glacial to Holocene period (Goehring et al., 2011;White et al., 2011).
Most cosmogenic nuclide studies to date have focused on measuring 10 Be in subglacially derived, sometimes striated cobbles deposited on bedrock on the flanks of Antarctic nunataks. The cobbles are preferred because they often retain a subangular to subrounded shape that is typical of subglacial sediment transported by erosive, warm-based ice. These erosive conditions should produce fresh surfaces with minimal nuclide inheritance. The preservation of striations also indicates minimal erosion of the rock surface after deposition. Exposure dates from locations where these sample characteristics are common can produce clear thinning histories with little geological scatter in exposure age results (e.g., Jones et al., 2015). In other settings, the success of the method is highly variable. For example, in the Pensacola Mountains, Balco et al. (2016) and Bentley et al. (2017) independently sampled altitude profiles in the Williams Hills that indicated at least 450 m of thinning occurred from the LGM through the Holocene, while a similar altitude profile from the Schmidt Hills, just tens of kilometers downstream, yielded 10 Be exposure ages that predate the last glacial cycle, and imply no thinning at all, which is hard to reconcile with the data from Williams Hills. The contrast is likely to do with specific flowlines of ice streams and tributary glaciers, as well as transitions between warm-and cold-based ice . It is situations like these where the measurement of in situ 14 C can be advantageous (e.g., Balco et al., 2016). Recent analysis of in situ 14 C from the Schmidt Hills indicates at least 800 m of thinning occurred after the LGM . Likewise, in situ 14 C analysis of rocks from the Shackleton Range indicate the Slessor Glacier thinned by at least 300 m, and perhaps as much as 650 m, following the LGM; this result contrasts with long-lived isotopes that imply no thinning at all and demonstrates the utility of in situ 14 C at sites where nuclide inheritance is problematic (

Data From the SWSE Sector and Surroundings
Changes in the Holocene elevation of the ice sheet surface in the heart of the SWSE have been studied in the southernmost Ellsworth Mountains, notably the Patriot Hills, Marble Hills and Independence Hills (Bentley et al., 2010;Johnson et al., 2019). Analysis has also been carried out on material from the Pensacola Mountains, to reveal the thinning history at Mount Bragg and Mount Harper, recording past elevation change of the Academy Glacier, and the Williams Hills and Thomas Hills adjacent to the FIS (Balco et al., 2016;Bentley et al., 2017). In all areas, the pattern is similar but with minor differences in timing, namely that the altitude profiles show ice sheet thinning that reached the level of the present-day ice in the Late Holocene ( Figure 8). The data are consistent with the two Holocene ice history scenarios noted previously but currently cannot distinguish between them. The first scenario is that the ice reached present-day ice level in the Late Holocene and thinned no further, staying at the same level since that time to present (scenario A in Figure 8a). A second scenario is where the ice thinned below the present-day level in the Late Holocene and only subsequently thickened to present levels (scenario B in Figure 8a).
In the Ellsworth Mountains, the more recent work relies on cosmogenic nuclide 26 Al/ 10 Be pairing on quartz and 36 Cl from limestone samples of unweathered and often striated glacial erratics (Hein, Marrero et al., Figure 8. Cosmogenic nuclide measurements of Holocene ice sheet behavior. (a) Hypothetical age-elevation plot with two groups of scenarios. In both cases, the ice sheet thins from a former (LGM) thicker configuration to reach the present ice sheet elevation in the Late Holocene. In scenario "A," the ice sheet then stays at the present-day level for a few ka to the present day. In scenario "B," the ice sheet thins to below present elevations in the Late Holocene and rethickens to the present-day.  Bentley et al. (2017). Two samples emerging from actively ice-cored moraines have been excluded from the plot because they may record buried ice ablation rather than ice sheet thinning.

10.1029/2019RG000651
Reviews of Geophysics 2016). Field observations and airborne RES show that these erratics have been derived from the glacier base and been brought to the ice surface in a blue-iced ablation area. On the mountain slopes above the present ice surface such erratics have been deposited as a veneer of perched blocks. The upper limit of such fresh erratics demonstrates the maximum thickness of the Weddell Sea sector of the ice sheet during the LGM; a state that persisted until the beginning of the Holocene. Cosmogenic exposure-age transects from the upper limit to the present ice record the history of thinning of the ice during the Holocene at a location close to the present grounding line in Hercules Inlet.
The upper limit of unweathered erratics shows that the ice was up to~475 m thicker than present in the southernmost Ellsworth Mountains during the LGM (Bentley et al., 2010;. The range of dates at the upper limit of 49-10 ka encompasses the peak of the LGM at~21 ka and could imply that thick ice persisted throughout the warming of the late-glacial (~18-10 ka). It seems reasonable to argue that this persistence reflects the increased snowfall and ice accumulation associated with warming as shown in the WAIS ice core (WAIS Drilling Project, 2013). Importantly, the direction of flow of the thicker ice sheet was eastward toward BIR (cf. Siegert et al., 2013). This is shown by erosional landforms, erratic trains of known lithology and the orientation of a blue-iced moraine immediately north of Independence Hills.
The vertical transects, taken together, show that the ice sheet surface in the Ellsworth Mountains began to fall about 10 ka ago and had reached the present level by 3.5 ka. Moreover, there was a pulse of rapid thinning of~400 m at 6.5-3.5 ka that if extrapolated around the Weddell Sea Embayment, would have contributed 1.4-2 m to global sea level rise . Interestingly, the geomorphology, changes in direction of ice flow of glaciers into Horseshoe Valley, and the disruption of blue-ice moraines, all linked to the eastward flow of thicker ice, accompanied or followed this sharp pulse of thinning. The most likely reason is that the migration of the grounding line into the troughs of IIS and Hercules Inlet captured the eastward ice flow toward BIR from the Horseshoe Valley.
In the Pensacola Mountains, the ice sheet history is consistent with that inferred for the southern Ellsworth Mountains. The Academy Glacier thinned similarly rapidly through the Holocene and reached present levels by 2.5 ka, while a few 10s of kilometers away, the FIS had thinned at least a few 100 m to within 40 m of present levels in the Williams Hills by 5.2 ka Balco et al., 2016;Nichols et al., 2019). A mid-Holocene pulse of thinning of similar scale and magnitude has also been reported from the Lassiter Coast in the western WSE based on in situ 14 C ages . Here the ice elevation fell from 385 to 138 m above modern ice between 6.0 ± 0.5 and 7.5 ± 0.7 ka before present.

Ice Dynamics, Solid Earth Deformation, and Sea Level Change
The evidence reviewed above does not allow us to distinguish between the two scenarios proposed for the recent glacial history of the SWSE. Therefore, in an attempt to fill some of the spatial and temporal gaps in the data, we turn to the insight that can be provided by ice sheet and solid-Earth modelling. Numerical ice sheet models, driven by changes in climate, can be used to predict changes in ice flow and thickness which can be compared with evidence from RES and cosmogenic exposure dating. The solid Earth and sea level response to these changes in ice loading can be predicted using a GIA model and compared with GPS-derived estimates of present-day surface deformation and geological evidence for relative sea level change.
Feedbacks between the ice sheet, the solid Earth and sea level change likely played an important part in controlling the behavior of the SWSE during the late Holocene. Such feedbacks have the potential to stabilize (Gomez et al., 2010) or destabilize (Siegert, 2001) the ice sheet, and both types of behavior have been reproduced by numerical models. Studying the strength of the various feedbacks, via modelling and model-data comparisons, helps us to determine which processes governed the evolution of the WS sector during the late Holocene.
Marine ice sheet instability describes the process whereby grounding line retreat on a reverse bed slope leads to an increase in ice thickness at the grounding line (where the bed deepens upstream) and, hence, an increase in ice flux across the grounding line (e.g., Joughin et al., 2014). If this increase in flux is not balanced by an increase in accumulation, this can lead to runaway retreat until the bed begins to shallow, and ice thickness reduces, again upstream. Along many of the ice streams in the SWSE the bed deepens inland, so if the marine ice sheet instability was the dominant process controlling ice sheet dynamics, we might expect the grounding line to have retreated continuously from an LGM position somewhere on the continental shelf to a stable position upstream of the present grounding line. Instead, the present-day grounding line is perched part-way along the reverse bed slope that extends from the continental shelf to regions such as the Robin Subglacial Basin . The reason for the apparently unstable position of the present-day grounding line is likely the presence of the RFIS, which exerts a significant buttressing effect on the grounded portion of the ice sheet, and the presence of large ice rises (Matsuoka et al., 2015). Indeed, Whitehouse et al. (2017) used a numerical flowline model to demonstrate that ice shelf collapse would lead to grounding line retreat behind present but that ice shelf regrowth would rapidly allow the grounding line to readvance to its observed position along the FIS. Kingslake et al. (2018) modeled similar behaviour in the central SWSE ( Figure 9).
In addition to the buttressing provided by the RFIS and its ice rises, isostatic uplift in response to ice loss may also have helped to stabilize the WS sector of the ice sheet during the late Holocene. As an ice sheet thins or retreats, the net reduction in surface loading, after accounting for the flux of ocean water into the space left by a retreating ice sheet, triggers viscoelastic rebound. The wavelength of the rebound will depend on the thickness of the lithosphere, while the magnitude and time dependence will be controlled by the viscosity structure of the mantle (Nield et al., 2018). In tandem with the uplift of the bed, the decreasing mass of the ice sheet leads to a decrease in the gravitational attraction exerted by the ice sheet on the ocean, and hence a drop in the local height of the geoid, that is, the sea surface (Gomez et al., 2010). The net effect of these two GIA-related processes is a decrease in the water depth at the grounding line, which, if large enough, has been hypothesized to have the potential to slow, or even reverse, unstable grounding line retreat (Gomez et al., 2015;Kingslake et al., 2018;Konrad et al., 2015). In addition to reducing water depth at the grounding line, the reduction in water depth offshore can help to promote ice rise formation, thus increasing the buttressing effect of the ice shelf (Favier & Pattyn, 2015).
The spatiotemporal history of isostatic rebound during the late Holocene and the degree to which it has played a role in governing the evolution of the SWSE is poorly known. The spatial pattern of rebound will depend on the history of both ice thickness change and grounding line retreat across the Weddell Sea embayment, a region that is currently covered by the RFIS. Ice sheet modelling provides some insight into possible configurations of past retreat (e.g., Golledge et al., 2014;Maris et al., 2014;Pollard et al., 2017), with large-scale constraints on past ice thickness and grounding line extent available from ice cores recovered from ice rises (e.g., Mulvaney et al., 2002).
The second factor that controls the rate of isostatic rebound is the rheology of the solid Earth. Mantle viscosity is hypothesized to be relatively high across the southern part of the SWSE (~10 21 Pa s; Whitehouse et al., 2019), but there is limited information on the rheology beneath the outer part of the embayment due to poor coverage by seismic stations (Shen et al., 2018). High mantle viscosity can be related to a long decay time  (McConnell, 1965) and, hence, rebound in response to post-LGM ice loss is likely to be ongoing today across the southern portion of the SWSE, with the caveat that the timing of post-LGM ice loss is poorly known. When seeking to reconstruct past ice sheet change using contemporary rates of solid Earth deformation, the trade-off with Earth rheology leads to significant non-uniqueness (Whitehouse, 2018), with an additional complication being that the loss of ice close to flotation will trigger a limited isostatic response. Feedbacks between ice dynamics and isostatic rebound will be greatest in regions where there has been a significant decrease in the thickness of ice above flotation.
In summary, isostatic rebound may have reduced the rate of grounding line retreat, promoted the likelihood of ice rise formation, or even triggered grounding line readvance across the SWSE and wider WS sector during the late Holocene. If readvance did take place and was accompanied by sufficient ice thickening above flotation, this will have damped the rebound process, and contemporary uplift rates would be smaller than expected in a scenario of monotonic ice loss and retreat. The following two sections briefly outline what is known about isostatic rebound across the WS sector, and how these observations compare with recent modelling efforts.

Data From the SWSE and Surrounding
Records of isostatic rebound across the SWSE are limited to a small number of GPS sites that record contemporary solid Earth deformation. The traditional approach of reconstructing millennial-scale solid Earth deformation via the interpretation of relative sea level records is not possible due to the complete lack of exposed coastline in the region. In other regions, relative sea level records have traditionally been used to determine the relaxation time and, hence, the viscosity of the mantle (e.g., Mitrovica & Forte, 2004). The short time span of the GPS data in the SWSE provides no insight into the relaxation time of the mantle, and this exacerbates the issue of non-uniqueness when seeking to use the GPS rates to infer the history of past ice sheet change.
The sparse spatial coverage of bedrock GPS sites is related to the lack of outcrops in the region; currently, no techniques exist for measuring isostatic rebound beneath regions of grounded or floating ice. It is logistically challenging to regularly access the bedrock GPS sites to download data and service the instruments. This is changing as the majority of instruments now have the capacity to transmit data and state-of-health information in real time, but there are significant historical data gaps in many records, and these are reflected in the large uncertainties associated with deformation rates at many sites. Published GPS rates across the region span −2.20 to +7.04 mm/year but typically have uncertainties on the order of 1-2.5 mm/year (Bradley et al., 2015;Wolstencroft et al., 2015). Extending the temporal and spatial coverage of measurements that reflect the contemporary pattern of solid Earth deformation across the WS sector would provide invaluable insight into recent ice sheet change and the stability of the current ice sheet configuration. Spatially continuous estimates of isostatic rebound are beginning to be derived via data inversion, whereby satellite gravimetry and altimetry data are combined to simultaneously invert for contemporary ice mass loss and the GIA signal, often also incorporating GPS observations (Engels et al., 2018;Gunter et al., 2014;Sasgen et al., 2017). Uncertainties are large, due to assumptions related to the interpretation of satellite data, but at least they can be formally quantified in this independent, data-driven approach.

Modelling and Plausible Ice Sheet Reconstructions
GIA modelling can be used to predict the pattern and magnitude of isostatic deformation across the WS sector for the post-LGM period. A "forward" GIA model requires inputs of ice history and Earth rheology, and this permits the calculation of spatially varying sea level change and solid Earth deformation (e.g., Argus et al., 2014;Briggs et al., 2014;Ivins et al., 2013;Whitehouse, Bentley, Milne et al., 2012). To better understand the impact of isostatic rebound on ice dynamics, GIA model output relating to the uplift history of the WS sector could be prescribed as an a priori boundary condition for ice sheet modelling. However, there is currently considerable uncertainty regarding the pattern of GIA-related solid Earth deformation across the wider WS sector.
GIA model predictions of present-day uplift rates across the WS sector vary significantly in both spatial pattern and magnitude (Whitehouse et al., 2019). The majority of models predict maximum present-day uplift rates in the region of the HIR, KIR, and BIR, with predicted rates peaking at >10 mm/year (Argus et al., 2014;Whitehouse, Bentley, Milne, et al., 2012). Models that incorporate 3-D variations in mantle rheology may predict both faster (Geruo et al., 2013) and slower (van der Wal et al., 2015) uplift rates compared with calculations that combine the same ice history with a simple 1-D profile of mantle viscosity. The predictions of Ivins et al. (2013) have a very different spatial pattern, with maximum uplift rates of only~5 mm/year predicted for the southwest Weddell Sea region and negligible (<2 mm/year) rates predicted for the central SWSE. The history of post-LGM deformation will also vary considerably between models. These differences between GIA model predictions arise primarily due to a lack of constraints on post-LGM ice sheet configuration, which has led to there being significant differences between the ice history models adopted in the different GIA models. None of the models provide a good fit to the regional uplift pattern documented by GPS .
Most Antarctic GIA modelling studies are motivated by the desire to test a new data-constrained or modelling constrained reconstruction of past ice sheet change within a GIA framework (e.g., Whitehouse, Bentley, & Le Brocq, 2012;Whitehouse, Bentley, Milne, et al., 2012). In contrast, Bradley et al. (2015) sought to explain the surprising observation that a GPS site in the Ellsworth Mountains, a region expected to be uplifting in response to post-LGM ice mass loss, was subsiding. They calculated the isostatic response to a suite of hypothesized ice history scenarios, including an approximation of the two scenarios proposed by Siegert et al. (2013) and found that the GPS observations could not be explained by monotonic post-LGM ice loss, that is, scenario A. Instead, the data required there to have been a period of either ice thickening or re-advance during the late Holocene (Bradley et al., 2015). The study was unable to provide detailed insight into the position of the grounding line during the late Holocene, primarily due to the fact that any advance or retreat of grounded ice close to flotation would have left a negligible isostatic fingerprint.
There is clearly a significant range in model predictions of isostatic rebound across the WS sector leading to uncertainty in efforts to quantify the impact of this rebound on ice sheet dynamics during the late Holocene. Many ice sheet models include an isostatic component (e.g., Bueler et al., 2007), but they are unable to account for the second feedback associated with GIA, which is the local drop in the height of the sea surface associated with a decrease in the mass of the ice sheet. This can only be determined by running a GIA model that accounts for the global redistribution of meltwater across the ocean. The most self-consistent approach to determining feedbacks between GIA and ice dynamics is therefore to run a coupled model (e.g., de Boer et al., 2017;Gomez et al., 2013). However, a significant limitation of this approach, aside from the uncertainty associated with climate forcing, is the need to know the correct Earth rheology. Upper mantle viscosity across Antarctica is thought to vary by >3 orders of magnitude (Hay et al., 2017;van der Wal et al., 2015;Whitehouse et al., 2019). The associated range in mantle relaxation times significantly perturbs the strength of the feedback between isostatic deformation and ice dynamics in different regions of Antarctica (Gomez et al., 2015;Konrad et al., 2015;Pollard et al., 2017) and motivates the need for a coupled ice sheet-GIA model that incorporates 3-D Earth structure. Gomez et al. (2018) use such a model to quantify the effect of incorporating 3-D variations in mantle viscosity compared with a more common 1-D radial profile of mantle viscosity. Differences in predicted ice thickness in the southern Weddell Sea region are on the order of several hundred meters during the Holocene, with implications for grounding line position. Similarly, they predict differences in relative sea level and contemporary uplift rates between the 1-D and 3-D models on the order of tens of meters during deglaciation, and several millimeters per year at the present day, respectively. Such models are rapidly enhancing our ability to explore the drivers of ice sheet change in the WS sector, but a crucial outstanding issue is the need to simultaneously represent feedbacks between ocean forcing and ice shelf evolution (Asay-Davis et al., 2016). The buttressing provided by ice shelves plays a significant role in controlling the stability of the ice sheet (Wright et al., 2014), and ice shelf thickness directly determines whether ice rises are able to form as a result of isostatic rebound.
We close this section by reviewing the results of ice sheet modelling studies that are not coupled to a GIA model but that do incorporate some component of solid Earth rebound. It is interesting to note that model-derived ice sheet reconstructions predict grounding line retreat behind present, and subsequent readvance, at some point during the Holocene, at a range of locations around the southern Weddell Sea (Golledge et al., 2014;Kingslake et al., 2018;Maris et al., 2014;Pollard et al., 2017). These studies use different numerical ice sheet models, and they differ in their approach to climate forcing, ice sheet basal conditions, and ice shelf dynamics, but all represent an attempt to reconstruct the history of the WAIS since the LGM. This is not to say that all modelling attempts reproduce grounding line retreat behind present; the ensemble analysis of Briggs et al. (2014) does not appear to predict Late Holocene readvance in the wider WS sector, while in the studies of Pollard et al. (2017) and Kingslake et al. (2018) model scenarios that involve faster rates of rebound predict slower rates of ice loss and no retreat behind present. This is due to the damping effect of the rapid rebound.
In this section, we have summarized the ice sheet and solid Earth modelling efforts that provide insight into the processes controlling recent ice sheet change in the SWSE and wider WS sector. As may be expected, standalone and coupled ice sheet models are able to reproduce a range of scenarios that encompass the retreat histories suggested by the glaciological and geological data documented above. However, it appears that the rate of isostatic rebound potentially plays a crucial role in determining the details of the retreat. On one hand, isostatic rebound can act to slow the rate of grounding line retreat, which would be compatible with the hypothesis that the grounding line is currently in its most retreated position since the LGM (scenario A). Alternatively, isostatic rebound may have helped trigger ice rise formation and grounding line readvance subsequent to retreat behind present at some point earlier in the Holocene (scenario B). Ultimately, additional data are needed to resolve the details of recent ice sheet change and provide further insight into the strength of the feedbacks operating between the ice sheet, the solid Earth, and the ocean.

Ice Flow at and Before Mid-Holocene
Geophysical data from the ice sheet and exposed rocks point consistently to major ice sheet change in the WS sector at around 5,000 years. Buckled layering in the BIR, beneath continuous layers dated to around 4,000 years, coupled with palaeoflow stripes across the ice rise, are clear indicators that the ice location was once covered by fast flowing ice. The flow stripes point to an origin from the IIS tributaries, parallel to the modern flow of MIS. This is also consistent with basal features that can tentatively be correlated between BIR and Horseshoe Valley. We are unable to explain whether the ice across BIR was floating (i. e., a fast-flowing section of an ice shelf emanating from upstream IIS) or an ice stream grounded across the Robin Subglacial Basin and BIR, however.
Beneath the ice divide on KIR a prominent Raymond Arch suggests stable flow here during at least the past 2-3 kyr (Kingslake et al., 2016). Further north along the KIR divide, polarmetric phase-sensitive radar and active seismic surveys reveal a transition in englacial fabric at around 200-to 230-m depth (Brisbourne et al., 2019). These authors interpret the fabric beneath this depth as evidence of a change in past flow conditions. Specifically, the observations can be explained by previous flow aligned approximately along the ice divide, away from IIS. Tracing stratigraphy back to the divide used to date the formation of Raymond arches (Kingslake et al., 2016) indicates that the change from this previous SW-NE flow regime occurred around the time of Raymond Arch formation (2-3 kyr). There is no englacial stratigraphic evidence for past fast flow at KIR and because 2-3 kyr is not long enough for such evidence to flow out of KIR, we can conclude that at least immediately prior to the onset of ice divide flow (and the formation of the Raymond arches), flow was not fast over KIR. This does not rule out a previous period of fast flow over KIR and even regrounding of the ice shelf in this location if it occurred long enough in the past for any englacial evidence of fast flow, as is seen in BIR and HIR, to advect out of the region surveyed by Kingslake et al. (2016).
The lack of mature Raymond Arches on HIR suggest that ice flow has not been stable here for at least the characteristic time of the ice rise (Martıń et al., 2009), approximately 4.5 kyr. The complex englacial structure of HIR can be explained if the grounded area of the ice rise was smaller in the past and readvanced toward its current position at around 6 kyr. Without a previously smaller ice rise and a subsequent advance of the grounding line to its present location, it is difficult to explain the spatial pattern of the interception between isochrones and steeply dipping englacial layers interpreted as marine ice-filled relic basal fractures (Wearing & Kingslake, 2019).
Two interpretations are possible as to the minimum extent of HIR during the Holocene. In one interpretation the ice rise ungrounded completely; this can explain all the unusual englacial structures observed. Alternatively, the ice rise ungrounded from the northern subglacial topographic high but remained grounded over deeper regions. Based on our limited understanding of ice rise formation and subsequent evolution, this appears unlikely; it is inconsistent with recent idealized modelling of ice-rise dynamics (Favier & Pattyn, 2015). While a smaller than present HIR that nonetheless persisted throughout the Holocene cannot be ruled out, until a deeper understanding of ice-rise flow dynamics and independent constraints on the past spatial extent of the ice rise are obtained, the interpretation that the ice rise ungrounded completely during initial post-LGM ice sheet retreat and later regrounded, perhaps due to GIA(section 6), is consistent with this argument. Some modelling studies pertain to the question of whether ice rises in the Weddell Sea ungrounded during the Holocene (section 6.3). For example, Kingslake et al. (2018) examined the hypothesis of a complete Holocene ungrounding of HIR and its implications using an ice sheet model that incorporates simplified GIA physics. They were able to reproduce the situation where BIR, KIR and HIR ungrounded after the LGM. Despite limitations related to resolution and bed topography, this suggests ungrounding was possible, but depends sensitively on, among other factors, the solid Earth and ice properties prescribed in the model, the climate, and the LGM extent of the ice sheet.

Ice relaxation Model: Mid-Holocene to Today
In this model, gradual retreat of the grounding line to its position 5,000 years ago, with associated ice thickness reduction and crustal uplift, led to relatively thin ice in the locations of the ice rises. This caused rerouting of water from the palaeo IIS trunk into the Robin Subglacial Basin, and both slow cold-based flow over the BIR and the creation of today's fast flowing IIS along the subglacial basin. Evidence in support of this idea comes from the shear margin between IIS and BIR, which is a hydrological one, and from the Siple Coast, where subglacial water migration is thought to be key in the shutting down and activation of ice streams (Anandakrishnan & Alley, 1997). It is possible that either Scenario MAX or Scenario MIN of Hillenbrand et al. (2014) at 5,000 years is consistent with this model (Figure 3).

Ice sheet Retreat and Readvance Model: Mid-Holocene to Today
An alternative explanation is that the ice margin retreated upstream of today's position, deglaciating the Robin Subglacial Basin. During this time, an ice shelf persisted across the current location of the BIR/KIR/ HIR, and it became grounded sometime during the middle-to-late Holocene due to post-LGM isostatic uplift. This grounding led to the buildup of the ice rises, blocking fast-flowing ice, and leading to the rerouting of the IIS to the west of BIR, and MIS to its east. Evidence in support of this includes the englacial stratigraphy discovered within HIR and the buckled internal layers at the northern edge of BIR. In addition, the low GPS uplift rates recorded in the Ellsworth Mountains have been interpreted to reflect the damping effect of ice sheet readvance on post-LGM rebound (Bradley et al., 2015). Models that incorporate both ice sheet dynamics and glacial rebound reveal how this scenario is possible (e.g., Kingslake et al., 2018), but the details depend on poorly quantified factors such as Earth rheology, seafloor topography, and ocean-ice shelf interactions.

Cosmogenic Nuclide Evidence: Mid-Holocene to Today
Cosmogenic nuclide evidence indicates that the current ice surface is close to that first achieved at 3.5 (Ellsworths) or 2.5 ka (Pensacola; Figure 8). The data also point to major ice flow direction change consistent with former flow across what is now the BIR. However, the data do not favor one or other hypothesis (relaxation vs. readvance) concerning the history of ice changes in the last few thousand years. There is nothing from the geomorphology or from cosmogenic nuclide dating to rule out thinning to below the 3.5 ka (2.5 ka) level. On the other hand, the geomorphological evidence of blue-iced moraine stability  and the early Holocene age of some blue ice itself  rules out extensive deglaciation with open fjords.

Hypotheses Synthesis
On the face of it, the two scenarios for Late Holocene change in ice sheet flow seem irreconcilable. One involves a gradual relaxation of the ice sheet, with ice dynamics and flow modified by the rerouting of basal water. The other relates to deglaciation upstream of the modern grounding line, followed by regrounding through postglacial uplift of relatively high ground to form ice rises. However, there is a consistent message, from multiple RES data sets and cosmogenic analyses, for major ice-flow dynamic change and ice sheet reconfiguration within the SWSE, and the wider WS sector of the WAIS, around the mid-Holocene. Buckled RES layers within ice rises point to former enhanced flow where the ice now flows slowly; palaeoflow stripes on the BIR indicate the former direction of that enhanced flow, and cosmogenic data reveal how upstream ice sheet flow has changed direction from flowing east toward the BIR, to flowing west into the modern IIS. RES also reveals Raymond bumps within the KIR, HIR and BIR that are relatively recently formed, suggesting ice rise formation in the last few thousand years.
The major difference between the two scenarios is that, in one, the ice sheet grounding line progressively retreated from its LGM position to its present grounding line, whereas in the other the grounding line retreated upstream of present limits and subsequently readvanced to the present. While numerical modelling demonstrates how this latter scenario is possible, there are no conclusive data for either.
While this disagreement might seem challenging to reconcile, the two scenarios have one important point of agreement beyond an acknowledgement of mid-Holocene ice sheet change; that is, they can both accommodate a level of post-LGM isostatic uplift. In the former scenario, such uplift could lead to topographic influences on basal water flow and ice sheet dynamics; in the latter it would lead to regrounding of an ice shelf and backfilling of grounded ice upstream. This agreement stems from the fact that the WS sector is a particularly sensitive part of the ice sheet to relative sea level (Wright et al., 2014). Indeed, the grounding line is known to oscillate by several kilometers each day as a consequence of tides lifting up otherwise grounded ice during high tide and dropping it to the bed on low tide. This process has been observed over the BIR (Brunt et al., 2011) and the FIS (Jeofry, Ross, Le Brocq, et al., 2018).
Given that the Robin Subglacial Basin is occupied by grounded ice that is relatively close to flotation , if it were to deglaciate (i.e., the ice sheet was to lift off the bed), it is unlikely there would be any major isostatic signal. Hence, there may simply be no conclusive isostatic record in favor of one scenario.
It is also unlikely that RES or more cosmogenic data from exposed surfaces will provide a resolution to the problem. However, if we could obtain samples from key regions of the subglacial environment, and date them, we may be able to provide a solution. Conceptually, the best way to use cosmogenic nuclides to distinguish between the different scenarios is to drill the margins of some of the glaciers or ice streams where we have ice sheet thinning histories. Retrieval of bedrock from below the margins will allow a test of the hypothesis that the ice has been thinner than present in the Holocene; any bedrock exposed in the Holocene and then recovered by ice will still contain a significant inventory of in situ 14 C isotopes. If the bedrock had been exposed in a previous thinning/interglacial, then the 14 C would have decayed (as its half-life = 5730 years). Drilling at various depths to find the boundary between 14 C-containing bedrock and that at lower elevation with no 14 C would enable determination of the minimum thickness reached. In addition, if subglacial access to the Robin Subglacial Basin or beneath the BIR, KIR or HIR is possible, where soft marine sediments deposited when the ice sheet was not present are thought to exist (Siegert et al., 2016), it may be possible to date the material by radiocarbon methods-certainly obtaining a Holocene date from such a sample would be a significant piece of information in support of the readvance hypothesis. Sampling the ice from a borehole drilled to access the bed (either as a full core or as chippings from a rapid drilling system) may permit isotopic analysis to reveal signals associated with either scenario; for example, if HIR was completely ungrounded during the Holocene, its deep ice would have been deposited on the surface while on the main ice sheet at higher elevations than is plausible in an ice rise. This would leave an isotopic signature. Finally, sampling ice from the hypothesized relic basal crevasses and measuring borehole temperatures on HIR and BIR may be key to determining the grounding history of these crucial areas.

Summary
The WS sector of the WAIS is known to be one of the most sensitive regions of the ice sheet to ocean-driven melting and grounding line recession, with the SWSE being particularly vulnerable. However, despite the SWSE being at a threshold of change, owing to the present grounding line of IIS being perched at the top of a major steep reverse bed slope, satellite altimetry shows that the region is, at present, not experiencing significant change like in the Amundsen Bay region of WAIS. However, satellite imagery of surface flow stripes, coupled with RES evidence of buckled internal layers and corroborated by cosmogenic dating of exposed surfaces, point to recent (mid-Holocene) and substantial ice-dynamic change in the SWSE.
Two scenarios to explain this change have emerged. First, the ice sheet gradually relaxed from its LGM position involving fast-flowing ice across the BIR. Because of ice sheet thinning, water and ice flow were routed around what is now BIR to set up the ice sheet as we now see it. Alternatively, ice retreated inland of the present margin, forming an ice shelf over the Robin Subglacial Basin and the BIR/KIR/HIR, whereupon isostatic uplift led to regrounding over the ice rises and the present flow regime.
While available glaciological and geophysical information cannot distinguish between these scenarios, they do have some areas of commonality. First, all evidence points to mid-Holocene change to the flow and form of the SWSE. Second, both scenarios can accommodate postglacial uplift to drive either ice dynamic alteration on one hand, or grounding on the other. Third, it is also possible that both scenarios may have acted together, with uplift leading to regrounding and ice direction change. Such a situation may account for buried crevasses within HIR, thought to be formed within an ice shelf and captured as a consequence of grounding, as well as the shear margin between IIS and BIR, which appears to be controlled solely by ice sheet thermodynamics and the supply of water, rather than topography. Fourth, we should appreciate that the SWSE experiences significant (>10 km) grounding line change as a consequence of tides, and much of the region is relatively close to the level of flotation now; hence, migration of the grounding line upstream of modern limits could easily occur without much isostatic signal today.
It is curious that the SWSE currently appears to show no sign of significant mass loss. The recent readvance hypothesis may be consistent with this, as the ice sheet would have effectively experienced expansion to its present limits. As ice sheet modelling has shown, however, this is unlikely to make the sector more stable to the influx of warm ocean water; if this happened, we still expect, under either scenario, the SWSE to experience grounding line retreat across the Robin Subglacial Basin.

Glossary
LGM LGM values Scenario B Holocene retreat upstream of today's grounded limit with subsequent ice expansion MISI Marine Ice Sheet Instability-associated with ice retreat across retrograde bed slope below sea level DEM Digital Elevation Model Nunatak exposed pinnacle of rock surrounding by thick ice exposure age the time since material has been uncovered ice rise a dome of ice with higher elevation and slower flow speed than surrounding ice RFIS Ronne Filchner Ice Shelf

Data Availability Statement
Ground-based ice-penetrating radar data can be obtained from the UK Polar Data Center (http://doi.org/ 99d). A MATLAB script for viewing the raw radar data is also provided at this link. The digital elevation model of the SWSE, and radar data used to build it, are available at (https://doi.org/10.5194/essd-10-711-2018).