The Totten Glacier is one of the fastest-flowing glaciers in East Antarctica , with ice surface speeds of over 750 myr-1 at the main tributary grounding line (southernmost region) – more than 100 myr-1 greater than the corresponding ice surface speed at the Vanderford grounding line (Fig. a). Average ice discharge from 2009 to 2017 at the Totten Glacier (71.4 ± 2.6 Gtyr-1) is almost twice that at the Vanderford Glacier (36.2 ± 0.5 Gtyr-1), naturally reflecting thicker ice, a steeper surface slope, and hence greater driving stresses (Fig. c) near the Totten Glacier grounding line. Both glaciers have accelerated in recent decades: the Vanderford Glacier surface speeds increased by 31 % over the period 2000 to 2013, with little subsequent change ; the Totten Glacier experienced insignificant increases in surface speed from 2007 to 2022 , but with quasi-decadal cyclicity . Furthermore, the Vanderford Glacier grounding line retreated by ∼ 18.6 km over the period 1996 to 2020 , making it the fastest-retreating glacier in East Antarctica . Changes at both glaciers are consistent with ocean-driven ice shelf thinning and calving (Sect. ) and consequent reductions in buttressing .

Discharge rates from each glacier are related to the broad-scale flow configuration of the ASB (Fig. ). The Totten Glacier is currently fed by a catchment of 267 904 km2. The majority of discharge through Totten is channelled from the west through its main tributary, with a smaller portion channelled through the eastern flank of the glacier (Fig. b). Ice sheet models suggest ongoing grounding line retreat along the eastern flank and into the Sabrina Subglacial Basin (SSB) will occur in coming decades . Despite the fact that the undulating bed topography in the SSB lies largely below sea level (Fig. ), this portion of the Totten catchment is generally thinner, having a sea level potential that is much smaller than that of the ASB main basin .

The current Vincennes Bay catchment that feeds the Vanderford Glacier is approximately 71 329 km2 – just over a quarter of the size of the adjacent Totten catchment. Similarly to Totten, discharge from the Vanderford Glacier is channelled through a western and an eastern trunk (Fig. b). The western trunk largely originates from, and is to the west of, Highland A (compare Figs.  and ). The eastern trunk originates in the same region as the main Totten tributary, east and south of Highland A; the Elcheikh saddle point (cross in Fig. a) marks the divergence of flow to either the main Totten tributary or the eastern trunk of the Vanderford Glacier.

The geology of the ASB – in particular, the extents and types of sedimentary basins in the Totten and Vincennes Bay catchments – shows correspondence with patterns in ice sheet flow and subglacial hydrology (Sect. ). The ASB region is dominated by sedimentary basins, interpreted to represent deposition during rifting events that occurred well before the EAIS developed . Sequences of these sedimentary rocks are stacked vertically, ranging from an older Neoproterozoic to lower Paleozoic sequence and a younger sedimentary sequence dating from the Permian–Triassic to the Cenozoic . The topographic barriers of Highlands A, B, and C and the Knox Highlands typically comprise erosional remnants of the older sequence, while the topographic basins typically comprise rocks of the younger sequence (Fig. ).

The sedimentary cover in the main topographic lows reaches thicknesses of several kilometres, while the cover on the highlands is thinner. In regions of enhanced glacial erosion, the geology has been eroded to basement or possesses a mixture of basement with remnants of the younger sedimentary sequence . Areas with sedimentary basins are in general floored by more easily erodible rocks than in basement-dominated regions, and upstream basins – particularly upstream of the main Totten tributary – are likely to provide an abundance of sediment to the downstream glacier bed, facilitating till continuity and basal sliding see Sect. ;.

Sedimentary basins also permit the occurrence of a viable hydrogeology system, where the permeable bed allows fluid exchange between the subglacial hydrological system and groundwater storage . Groundwater discharge into the glacier bed varies with ice sheet retreat rate and also with the thickness and permeability of the sedimentary basins, with major potential impacts on the basal sliding of contemporary glaciers in East and West Antarctica . In the case of the Vincennes Bay catchment, the bed is dominated by the older sedimentary sequence and crystalline basement (Fig. ), which are more likely to have low permeability; here, capacity for active hydrogeology may be restricted (although these regions of hard bed are characterised by channelised subglacial hydrology and faster flow; see Sect.  and ). In contrast, the ASB main basin – which both is thick and contains relatively young sedimentary rocks – has higher potential to support active hydrogeology, with potential impacts on basal sliding processes and hence ice streaming . Although estimates for the contributions of groundwater to the subglacial hydrology system exist broadly across Antarctica , they are currently not available for the ASB specifically.

Approximately 75 % of the bed topography in the Totten and Vincennes Bay catchments is below sea level, and over 10 % is more than 1 km below sea level, particularly in the ASB main basin (Fig. ). Regions of topographic highs play a role in determining the overall ASB catchment boundary as well as the orientation of flow within the Totten and Vincennes Bay catchments. This is notably the case with the topographic barriers of Highlands A and B, the western edge of the Vanderford Glacier, Law Dome, and the coastal ridge adjacent to the Totten Glacier and the Moscow University ice shelf (Figs. , ). Ice flowing into the main Totten tributary has a similar flow trajectory to ice that flows into the eastern trunk of the Vanderford; ice from both glaciers originates in a region of generally high topography north of Lake Vostok (Vostok Highland; Fig. a). Downstream, both the Totten and Vanderford glaciers flow into the Vanderford Trench – a deeply incised subglacial channel that borders Law Dome and records the lowest known topography of the ASB (-2782 m below sea level; Fig. b). While topography provides a clear control on the overall ASB catchment boundary, there are no distinguishing features of the bed topography that clearly control the location of the drainage divide between the Totten and Vincennes Bay catchments. However, possible local controls in the Vanderford Trench are difficult to observe where very steep topography and thick ice hinder imaging of the ice base by ice-penetrating radar.

In terms of the susceptibility of each glacier to ongoing retreat, bathymetric pinning points on the Totten ice shelf, in the ∼ 20 km downstream of the main Totten tributary grounding line, provide significant buttressing of the upstream glacier . A prograde bed slope in the ∼ 20 km upstream of the present-day grounding line decreases the likelihood of substantial retreat occurring under climate change projections to 2100 . For the Vanderford Glacier, there is a high degree of uncertainty in the underlying bed topography within the Vanderford Trench; notably, uncertainties of 600 m and greater exist within the 10 km upstream of the present-day Vanderford grounding line see discussion in Sect.  and Fig. c;. Given the controlling influence of the bed elevation and slope on grounding line retreat, this has significant implications for the observed and simulated timing and rate of retreat. These lines of evidence, coupled with the current rates of retreat from each glacier, suggest that the Vanderford Glacier could retreat more rapidly, and further into the Vanderford Trench, than the Totten Glacier over the coming decades.

Observations and modelling confirm the presence of a widespread, persistent subglacial hydrological network in the ASB , consisting of a distributed “sheet-like” system in the ASB main basin and a channelised system in the few hundred kilometres upstream of the Totten and Vanderford Glacier grounding lines (Fig. a). The proximal ice–bed interface of the distributed system is well lubricated and supports little to no basal shear stress (< 20 kPa; Fig. b). Basal specularity content derived from radar is reduced within the channelised system , consistent with a rougher ice–bed interface where channels are present , and a strong bed here supports most of the basal shear stresses for each glacier (Fig. b). The calculation of the basal shear stresses in Fig. b is described in Sect. , and the striped pattern in the approach to the grounding line is consistent with previous studies e.g..

The majority of basal water in the ASB is routed towards the Totten Glacier ; output from the Glacier Drainage System (GlaDS) subglacial hydrology model estimates approximately 42 m3s-1 discharge from the Totten catchment into the ocean, compared with only 3 m3s-1 from the Vanderford Glacier . This is partly due to fewer channels developing in the approach to the Vanderford grounding line in the steady-state GlaDS simulations compared to the Totten grounding line. Nevertheless, the subglacial hydrological networks of the Totten and Vincennes Bay catchments are strongly connected. For example, ICESat surveys identified at least two “active” lakes located within the 200 km upstream of the Totten grounding line, with lake drainage and filling evidenced in surface elevation changes between 2003 and 2006 . Hydraulic potential analysis of one of these lakes shows that basal water drainage may be routed from the Totten to the Vanderford Glacier with only small changes in the ice surface elevation .

The geomorphology of the continental shelf proximal to the Vanderford Glacier – revealed in high-resolution multibeam bathymetry  – provides evidence for an active subglacial hydrology network in the Vincennes Bay region in the past. Of particular note is the presence of deeply incised bedrock channels, which are the most abundant landform on the inner continental shelf and which span depths of 450 to 2280 m below sea level. Similar bedrock channels and flat-bottomed basins are also observed in front of the Thwaites and Pine Island glaciers in West Antarctica , where hydrological modelling suggests that these are relict subglacial meltwater channels and lakes . The presence of these landforms on the inner shelf of Vincennes Bay supports the existence of an active, dominant subglacial hydrological system beneath a more extensive Vanderford Glacier, which formed over multiple glacial periods. A previous study has also documented a series of relict subglacial lakes akin to those mapped in front of Vanderford (indicated in grey in Fig. ). The network of subglacial meltwater channels and lakes would have formed during the last glacial period, or possibly over multiple glacial periods . Based on lake geometry and contemporary subglacial water transfer, these lakes likely drained episodically through the subglacial meltwater channels on decadal timescales , facilitating ice basal sliding over the rugged topography of the inner continental shelf.

In summary, the large-scale geometry of the ASB controls the current divergence of flow between the Totten and Vanderford glaciers. However, it is likely that the characteristics of, and variability in, the subglacial environment, particularly those elements that influence basal motion (e.g. the bed substrate and deformability, subglacial hydrological network, groundwater system) in the ASB, play a key role in the flow configuration. As basal water pressure and water accumulation vary into the future – e.g. associated with drawdown of the Vanderford Glacier due to grounding line retreat and subsequent thinning – basal water and ice flow piracy towards the Vanderford may also occur.

Surface mass balance (SMB) is the primary atmospheric driver of variability between the Totten and Vanderford glaciers. The atmospheric circulation over Law Dome is predominately polar easterlies , and the orography of the Dome causes relatively large snow accumulation, with a strong east–west gradient . The temporal distribution of the snow accumulation is related to the passage of synoptic-scale features and is therefore episodic, typical of coastal Antarctica .

Antarctic snow accumulation is expected to increase as water vapour increases with a warming atmosphere . While there is no clear trend in Law Dome accumulation rates over the past 2000 years , a trend of increasing total Antarctic annual snow accumulation has been observed since 1800 . While high-spatial-resolution snow accumulation observations are broadly lacking, the RACMO2.3p2 reanalysis model captures the strong east–west gradient in snow accumulation over Law Dome on annual timescales. Results averaged over the Totten and Vanderford glaciers (Fig. ) suggest that without major changes in atmospheric circulation patterns, a local increase in snow accumulation over Law Dome will amplify the east–west gradient.

In addition to the direct snow accumulation implications, there is a secondary effect on glacier flow related to the enhanced vertical advection of cool surface temperature into the interior at increased snow accumulation rates. For pseudo-steady-state surface elevations and approximately equal surface temperatures, the higher snow accumulation over the Totten will result in higher vertical velocities compared to the Vanderford. Without compensating differences in geothermal heat flow, deformational heating, or heat transport from basal hydrology processes, this increased vertical advection of cold will result in the bulk ice properties of the Totten being colder and stiffer than those of the Vanderford.

Without considering the effects of ocean-driven ice shelf melting (and associated surface lowering), the projected SMB changes in the ASB under a warming climate could lead to enhanced surface elevation and thickness increases at the Totten Glacier compared with the Vanderford Glacier. These effects could be further enhanced if they were to occur in concert with a dynamics-driven deceleration of the Totten Glacier due to cooling and stiffening ice over timescales of centuries to millennia.

Similarly to glaciers in the Amundsen Sea sector of West Antarctica, ice shelf thinning, grounding line retreat, and increased discharge at the Totten Glacier are linked to the incursion of warm modified circumpolar deep water (mCDW) through deeply incised bathymetric channels , evidenced from both airborne and shipborne measurements . Ice shelf melt rates at the Totten Glacier have been estimated to exceed 50 myr-1 near the grounding line and are among the largest observed melt rates in East Antarctica. Modelling indicates increased melt in recent decades, attributed primarily to freshening shelf water . This is despite a consistent trend of poleward migration and warming of mCDW off the shelf , which is projected to increase as the climate warms. However, observational evidence and modelling outputs are consistent with significant variability in melt rates as a result of variable mCDW supply in this region .

Links between ocean forcing and recent trends at the Vanderford Glacier are less well established than at the Totten Glacier. Data from seal dives suggest that the mCDW in Vincennes Bay is up to 0.5 ∘C and the warmest sampled to date in East Antarctica; however, temperatures at the Vanderford Glacier ice shelf of -0.5 ∘C are comparable to those at the Totten Glacier ice front of -0.4 ∘C . The presence of a deep bathymetric trough at the Vanderford ice front provides a clear pathway for mCDW incursion to the ice shelf cavity, although the bathymetric connection from this trough to the continental shelf slope has not been mapped. It is hypothesised that the ocean is driving the observed retreat and thinning of the Vanderford Glacier 2003 to 2020 , but observations and modelling are needed to establish causal relationships.

Modelling has been used to evaluate the potential for enhanced melting on the Sabrina and Knox coasts over the coming centuries . Upwelling of mCDW is positively correlated with the Southern Annular Mode (SAM), and positive SAM trends are projected to continue through the 21st century, which could lead to enhanced mCDW-driven ice shelf melt . Surface winds near the East Antarctic coastal margin are projected to intensify under climate warming , which could also increase wind-driven upwelling of mCDW to the continental shelf, and the supply of warm water to ice shelf cavities . suggested that warming surface waters linked to a decline in summer sea ice cover could dominate freshening associated with ice sheet mass loss trends in this sector. A key caveat on modelling results is the generally low spatial resolution (hence coarsely resolved bathymetry) used and the fact that previous bathymetry products do not incorporate the deep bathymetric trough in Vincennes Bay important for warm water supply to the ice shelf cavity. Nevertheless, evidence consistently points to the possibility of increasing ocean-driven ice shelf melt in this sector.

The cumulation of these factors – the warmer mCDW in Vincennes Bay, the pathway for its incursion into the Vanderford ice shelf cavity, and the greater potential for grounding line retreat of the Vanderford Glacier into the Vanderford Trench compared to the buttressed Totten Glacier – suggests a potential heightened vulnerability of the Vanderford Glacier to continued thinning in the coming decades to centuries.

The model domain covers the ASB (Fig. a) and comprises 58 938 anisotropic mesh elements. The initial bed topography, thickness, surface elevation, and ice masks are from BedMachine Antarctica v3 , and surface velocities are from MEaSUREs version 2 MEaSUREs2;. Using the shallow shelf approximation SSA; and inverse methods, we calculate the basal friction coefficient across the grounded ASB using the Budd friction law , given by 1τb=Cf2Nub, where τb (Pa) is the basal shear stress, Cf (s1/2m-1/2) is the friction coefficient, and ub (myr-1) is the basal sliding velocity. The effective pressure N (Pa) is given by N=ρicegh+ρwatergb, where ρice is the ice density (kgm-3), g is the gravitational acceleration (ms-2), h is the ice thickness (m), ρwater is the density of freshwater (kgm-3), and b is the bed elevation (m). We assume a Glen-type flow relation , where the viscosity μ (Pas) is given by 2μ=B2ε˙e1-n/n, and inverse methods over the entire domain to calculate the ice rigidity B (Pas1/n), for effective strain rate ε˙e (s-1) and n=3. The inverse method relies on minimising a cost function that includes terms for the linear and logarithmic misfit between the simulated and observed surface velocities, as well as regularisation parameters that smooth strong gradients in the rigidity and basal friction coefficient . MEaSUREs2 velocities are used at the inflow boundaries, and a free-flux boundary condition is applied at the calving front.

Holding the grounding line positions fixed, we run a 200-year spin-up simulation with monthly time stepping, at the end of which the ice surface speed and geometry are in a pseudo steady state. For this simulation we use annual average surface mass balance from RACMO2.3p2 and basal melt rates are calculated using a parameterisation that linearly decreases the basal melt from 30 myr-1 at ice shelf drafts of 400 m below sea level or deeper to 5 myr-1 at ice shelf drafts of 200 m below sea level or shallower . The final ice velocities, surface elevation, and thicknesses from this simulation are used as the initial conditions for our perturbation experiments described below.

To generate ice surface elevation increases at the Totten Glacier (expSMB), we increase the annual average SMB as follows: 3SMBnew=SMBRACMO+βSMBt, where SMBnew is the perturbation field, SMBRACMO is the observed annual average SMB from RACMO2.3p2 over the period 1979 to 2021 , and SMBt is the observed annual average SMB within the Totten catchment with zero SMB elsewhere in the ASB. Values for the constant β increase from 0.06 to 0.42, in increments of 0.06, and are chosen to represent the percentage increases in SMB that correspond to 1 ∘C increases in air surface temperature from 1 to 7 ∘C . Although we note that other surface processes influence the SMB, we assume that increases in accumulation will dominate at basin scale.

To generate ice surface elevation decreases at the Vanderford Glacier (expFriction), we decrease the basal friction coefficient within the Vincennes Bay catchment as follows: 4Cnew=Cf-σCvMα, where Cnew is the perturbation field, Cf is the basal friction coefficient calculated using inversion (Fig. A1a), and Cv is the basal friction coefficient within the Vincennes Bay catchment which is zero elsewhere in the ASB (Fig. A1b). Values for the constant σ increase from 0.1 to 0.7 in increments of 0.1. These values are chosen to yield the same area-weighted percentage changes between the simulated surface elevation compared with the control, when averaged over the Vincennes Bay catchment, as the corresponding surface elevation change generated from expSMB. M is a mask that is generated by calculating the distance from the grounding line within the Vincennes Bay catchment and then normalising the field to vary linearly from 1 at the grounding line to 0 at the furthest point (on the southern inflow catchment boundary). We choose α=20 as the exponent to M, which concentrates the change in the basal friction coefficient within the ∼ 100 km upstream of the Vanderford Glacier grounding line (see Appendix ). We reduce the basal friction coefficient in the Vincennes Bay catchment rather than modify the basal melt rate because, although not directly observable, we expect a reduction in basal traction as a consequence of increased ocean melting, retreat, and acceleration. In this way, we avoid increasing the basal melt rates at the Vanderford Glacier relative to the Totten Glacier (given the lack of evidence for such a relative increase in ocean forcing), while still capturing the effects of continued retreat at the Vanderford Glacier, which is more likely than at the Totten Glacier over the coming decades to century (Sect. ). We also emphasise that the aim here is not to model the expected evolution of this system but to generate surface change fields that are dynamically consistent with increasing SMB at the Totten Glacier and continued retreat at the Vanderford Glacier, to ascertain the impact of perturbations in the surface elevation on ice and basal water flow piracy.

In addition to the expSMB and expFriction experiments, a third experiment (expCombined) combines the SMB and basal friction coefficient fields in Eqs. () and (). Using the output geometries (thickness and surface elevation) and velocities from the pseudo-steady-state simulation, we run each of the perturbation experiments expSMB, expFriction, and expCombined for 1000 years with monthly time stepping. A 1000-year control run that uses the original SMBRACMO and Cf fields is also simulated. Figure  shows the differences between the perturbed surface elevations and the control (dSi, where the subscript i refers to SMB, F, or C) for an air surface temperature increase of 1 ∘C (i.e. for β=0.06 in Eq.  and σ=0.1 in Eq. ), hereafter denoted expSMB1, expFriction1, and expCombined1.

We investigate the impact of the changed surface elevations on the driving stresses and ice flow routing within the ASB. The driving stress p (Pa) is given by 5p=ρicegh∇Si, where ∇Si is the gradient in the perturbed surface elevation and the sum of the BedMachine surface elevation and each dSi field.

For expSMB1, the driving stress decreases through much of the interior of the ASB but increases within ∼ 10 km of the Totten Glacier grounding line. This is consistent with decreasing velocities upstream of the Totten Glacier grounding line but increasing velocities on the Totten Glacier ice shelf and the Vanderford Glacier ice shelf. The divide between the Totten and Vincennes Bay catchments migrates eastwards for expSMB1 compared with the catchments defined using the MEaSUREs2 velocities, increasing the Vincennes Bay catchment area by ∼ 1208 km2 (1.7 %; Table ). This leads to a rerouting of ice flow from the Totten Glacier towards the Vanderford Glacier, with an increase in discharge from the Vanderford Glacier of ∼ 0.02 Gtyr-1 (i.e. 1.6 % increase; Table ). Both trends increase with increasing air surface temperature change; for expSMB7, the Vincennes Bay catchment area increases by 17 % and the ice discharge increases by 11 %.

Decreasing the friction coefficient in the Vincennes Bay catchment (expFriction1) leads to immediate acceleration of the Vanderford Glacier and surface elevation lowering upstream of the grounding line (Fig. ). The driving stress also decreases close to the grounding line and on the Vanderford Glacier ice shelf, although it increases upstream of the grounding line to the Elcheikh saddle point (Fig. ). This also results in an eastwards migration of the divide (i.e. into the Totten catchment) between the Totten and Vincennes Bay catchments, increasing the Vincennes Bay catchment area by 958 km2 (1.3 %), which is less of a migration than observed in expSMB1. By contrast, the flux across the Vanderford Glacier grounding line increases by 5.1 %, which is over 3 times that of expSMB1, reflecting the greater increase in velocities in the Vincennes Bay catchment. For expFriction7, the discharge increases by 42 %, reflecting the greater sensitivity to relative increases in ice surface speed.

The driving stress changes for expCombined1, along with the changes in the Vincennes catchment areas and discharges, are close to a linear combination of expSMB1 and expFriction1 (Fig. ). The magnitude of the differences increases for increasing air surface temperature, such that in expCombined7, the increase in the Vincennes Bay catchment area is ∼ 5 % larger than expSMB7 and the discharge 15 % larger than in expFriction7.

We next examine the sensitivity of basal water routing between the Totten and Vanderford glaciers due to changes in the surface elevation. We use the MATLAB TopoToolbox to calculate the basal water accumulation across the ASB (the total upstream area in km2 that contributes to water accumulation within each individual grid cell). The upstream area is defined based on the basal water flow direction vector, which is calculated using the local slope in the effective pressure at the base of the ice sheet , and an upstream cell can provide water to only one neighbouring outlet cell. Here, rather than assuming overburden hydraulic potential, we use the parameterisation for the effective pressure NE given by , defined as 6NE=ρiceghi(1-rl)hĩmhĩm+him, where rl, h̃, and m are constants, and the fields hi are calculated by summing the BedMachine thickness and the perturbed thickness from each simulation. A summary of the derivation of this parameterisation is provided in Appendix .

The basal water accumulation calculated using the BedMachine bed topography, thickness, and surface elevation is shown in Fig. a. Currently, basal water is preferentially routed towards Totten over any other glacier in the ASB. This pattern of basal water accumulation is very similar for each of the surface change experiments under 1 to 4 ∘C warming. However, for the expSMB5 and expCombined5 experiments, basal water piracy to Vincennes Bay occurs, with the glaciers in this sector draining almost the entire region in the interior of the ASB (Fig. b, d). Under these scenarios, the only water routing towards the Totten Glacier originates north of Highland B. The pattern is consistent, with little subsequent change, for air surface temperature increases above 5 ∘C. By contrast, basal water accumulation is relatively insensitive to changes in surface elevation associated with the basal traction reduction over the Vanderford catchment imposed in our experiments, over the entire investigated range (i.e. expFriction1 to expFriction7; Fig. c).