Glacial erratics and glacially striated bedrock were collected adjacent to the David Glacier both upstream and downstream of the present-day grounding line. Downstream sites, Hughes Bluff and D'Urville Wall, are expected to show the most recent and dramatic changes in ice thickness, while smaller and longer-term changes are expected at the upstream site, Mt Kring. (Fig. ) . Using the structure-from-motion photogrammetry technique , we use helicopter-based photography to construct a high-resolution digital elevation model of each field site, allowing integration of glacial geological field surveys with the regional and local geomorphology (Figs. S3 and S4).

We primarily sampled glacial erratic cobbles from local glacial deposits and glacially moulded bedrock surfaces along vertical transects perpendicular to the modern glacial flow direction. The cosmogenic nuclide inventories in cobbles and bedrock comprise signals of both exposure since the most recent deglaciation and nuclides accumulated during any prior exposure (known as “inheritance”). For long-lived nuclides such as 10Be, this inherited component can be significant . Inheritance of cosmogenic nuclides is widespread in Antarctica as the cold-based nature of some ice acts to cover (preventing build-up of cosmogenic nuclides) without causing significant erosion . However, since cosmic rays attenuate rapidly with depth into rock, subglacial erosion of more than a few metres can reset the cosmic-ray signal. As such, we primarily focused on sampling faceted and striated glacial erratics because they indicate a subglacial origin which reduces the likelihood of a prior exposure history, and these intact features indicate minimal post-depositional erosion. Exposure ages from these samples are interpreted to track glacier thinning at the site because such samples would have been brought to the glacier surface by upward-flowing ice before being deposited at the ice margin .

We use cosmogenic nuclides 10Be and 3He to determine surface exposure ages on 45 samples. For 10Be analysis, quartz is the target mineral and is present in granites and sandstones. For 3He, pyroxene is the target mineral and present in dolerites. In the field, we preferentially sampled on isolated, local topographic high points, distant from areas of blue ice, snow drifts or local till deposits. While it is the focus of this study to use glacial erratics to track glacier thinning, we also collected bedrock samples to better understand complex exposure histories due to burial and the non-erosive nature of cold-based ice .

Samples were processed at the cosmogenic nuclide facilities at Te Herenga Waka Victoria University of Wellington. For 10Be, we separated quartz by crushing and sieving to extract the 250–500 mum grain size fraction. Magnetic minerals were separated using a Frantz isodynamic magnetic separator. For granitic rocks, feldspars were removed by froth flotation. We etched samples for 1 d in 10 % hydrochloric acid, for 1 d in 2.5 % hydrofluoric acid (HF) and for 3 d in 1 % HF to further purify the quartz separates. Beryllium was extracted from quartz following an established method including the addition of a 9Be carrier, enhanced quartz etching and digestion followed by ion exchange chromatography and BeOH precipitation . Targets of BeO were packed and sent to the Purdue Rare Isotope Measurement Laboratory (PRIME lab) for analysis using accelerated mass spectroscopy (AMS). For 3He, we targeted pyroxene from the 125–250 mum grain size fraction. Pure pyroxene was obtained using an established HF etching method, and 3He / 4He ratios were measured at the Berkeley Geochronology Center noble gas spectrometer . We supplemented this data set with eight samples collected prior to the 2016–2017 austral summer, and those samples were processed for10Be or 3He using methods outlined in and . Exposure ages are calculated by converting the AMS-derived 10Be / 9Be ratio to 10Be concentration by subtracting a known amount of 9Be carrier added during chemical processing. Final exposure ages are calculated using 10Be or 3He concentration, field information (elevation, shielding and sample thickness) and production rate scaling method (LSDn) . For 10Be, we employed systematics (http://ice-tea.org/en/, last access: 21 October 2021), and for 3He, we use systematics (https://hess.ess.washington.edu/, last access: 21 October 2021). All relevant data used to calculate exposure ages are served on the ICE-D online database within the David Glacier region (http://antarctica.ice-d.org/, last access: 21 October 2021) and in Supplement Tables S1–S5. The resulting glacier thinning chronologies derived from surface exposure ages inform our glacier modelling approach by providing geometric targets for evaluating the simulated maximum thickness of ice and the subsequent magnitude and duration of glacier thinning.

To understand potential controls on grounding-line migration and onshore thinning, we apply a glacier flowline model to the David Glacier. Originally designed to track grounding-line migration using a moving grid, the flowline model we employ has been described fully elsewhere e.g.. The underlying theory and governing equations are outlined in , and here we describe the components pertinent to our research questions.

Our glacier modelling focuses on a ∼ 1600 km long flowline from Dome C to the western Ross Sea shelf edge. The domain consists of varying flow regimes and extends to areas where the ice stream converged with grounded ice in the western Ross Embayment during the LGM. Ice flow is governed by fundamental Eqs. ()–().

Ice flow along the grounded portion of the model domain, u, is calculated using the shallow ice approximation, which assumes ice flow only from internal ice deformation and horizontal velocity defined as 1u=CδsδxHn+1, where n(=3) is Glen's flow law exponent, s is ice surface elevation, H is ice thickness and C is a constant given by 2C=2A(ρig)nn+2, where A is the temperature-dependent rate factor (for -20 ∘C), ρi is ice density (0.910 g/cm3) and g is gravitational acceleration (9.8 m/s2).

As ice reaches flotation, ice is assumed to spread unidirectionally along the flowline, and ice flow is balanced by 32δδxHvδuδx=ρighδsδx, where and v is the vertically average effective viscosity defined as 4v=A-1nδuδx2(1-n)/(2n), where δuδx is the effective strain rate.

For an ice stream, Eq. () is modified by including a basal traction coefficient β which is linearly related to the horizontal velocity u and is given by 52δδxhvδuδx-βu=ρigHδsδx.

We also incorporate a width term, including flat, a lateral buttressing factor, to account for lateral buttressing along a coupled ice stream shelf. Thus, as in previous applications of this model () the final ice thickness (h) and width (W) averaged ice flow (u) is computed using the following equation: 62δδxHvδuδx-uβ|u|m-1-HW52AWflat1n|u|1n-1u=ρigHδsδx.

flat controls the strength of lateral drag exerted at the ice stream margins. For all experiments in this study, flat is only altered where the ice stream is floating. When flat equals 1, full effects of lateral drag are felt; when values of flat are greater than 1, the magnitude of buttressing applied to the margins of the ice stream is reduced non-linearly.

The inclusion of the width term allows modelling of changing offshore trough width, which is common in palaeo-ice streams and outlet glaciers and is further supported by geomorphic interpretations using high-resolution bathymetry data . To determine offshore trough width, we map all glacial geomorphic features using the global multi-resolution topography data set within GeoMapApp (http://www.geomapapp.org, last access: 21 October 2021) based on analogues and prior regional mapping .

improved the ice shelf dynamics to incorporate the treatment of large horizontal grounding-line movements that are expected for the palaeo-David Glacier based on the long-term presence of the Drygalski Ice Tongue and evidence of extensive sub-ice-shelf conditions inferred from marine sedimentary cores . In the model, the ice shelf geometry evolves with time, and variations in ice shelf thickness and extent are fed into lateral drag calculations . Improved treatment of ice shelf dynamics allows exploration of the grounding-line sensitivity to ice shelf feedbacks. Importantly for this study, a reduction in lateral buttressing from adjacent, coalesced ice is expected as the expanded David Glacier and grounded ice in the Ross Sea decouple, which has been suggested from interpretation of submarine geomorphic features in the western Ross Sea .

Accumulation in the model is applied using modelled modern rates at the central flowline and multiplied to account for the ice stream width (W). Accumulation is then scaled to represent warmer or cooler conditions in our experiments (see Sect. 2.2.2). Where an ice shelf is present, sub-ice-shelf melt rates (SIMRs) are applied as a linear function of the depth of the ice shelf draught. From a minimum rate of 0.1 m/yr at 0 m (the ocean surface), SIMR increases to a maximum value at a depth of 500 m, and where ice shelf draught is deeper than 500 m, the maximum value is applied.

Mt Kring, situated in the interior of the EAIS, flanks the streaming ice of David Glacier draining Dome C. The nunatak, composed of layered dolerite, stands ∼ 300 m above the modern ice surface, henceforth “above the ice”. Of the 24 samples processed, 3 are dolerite bedrock and 21 are glacial erratics of mixed lithologies (Figs.  and S3). The bedrock surface at the peak of Mt Kring lacks glacial striations or erratics and has an apparent 3He bedrock exposure age of 554 ± 91 ka. A discontinuous, steep ridge line between ∼ 200–300 m above the ice was not sampled due to inaccessibility. At ∼ 200 m above the ice, the bedrock is striated parallel to the modern ice flow direction and has an apparent 3He exposure age of 550 ± 82 ka. The highest-elevation erratic is found at 171 m above the ice with increasing erratic abundance culminating in a drift sheet covering nearly all bedrock lower than ∼ 110 m above the ice. Striated, bulleted cobbles and boulders of dolerite, basalt and sandstone are common in the thin, patchy drift covering the bedrock. Based on our erratic exposure dating, the glacial drift here is a composite of three age populations (Fig. a). The younger population spans 7.4–5.5 ka (10Be, n=5) (Fig. b). An older population spans 51–25 ka (10Be, n=8 and 3He, n=4). The oldest erratic age population shows scattered evidence of glacial behaviour between 123–63 ka (10Be, n=2, 3He, n=2). The older populations are likely an artefact of repeated burial by overriding ice and multiple periods of exposure (i.e. inheritance).

Hughes Bluff, situated on the Scott Coast, is a granite outcrop along the southern flank of the David Glacier. The outcrop is glacially scoured exhibiting spectacular rochés moutonnées, crescentic gouge marks and striations parallel to modern flow directions (Figs.  and S4). Scattered glacial erratics blanket the entire outcrop. 10Be exposure ages from 15 erratics span 6.7–4.3 ka (Fig. ). The majority of glacial erratics are dated to between 6.7 and 6.2 ka. Two bedrock surface exposure ages sampled from rochés moutonnées at the highest and lowest outcrops (20.55 ka ± 2.10, 5.5 ± 0.47 ka) suggest significant wet-based glacial erosion since the LGM. At 20 m above the ice, similar bedrock and erratic ages provide evidence of recent emergence of this low-lying outcrop since 5.5 ka. The fact that well-developed landforms of glacial erosion occur at the highest outcrops at Hughes Bluff, including evidence of abundant basal sliding and plucking, indicate that ice thickness during the LGM was considerably greater than 230 m, the maximum elevation at Hughes Bluff . While the onset and magnitude of thinning prior to 6.7 ka is not constrained, the Hughes Bluff chronology clearly indicates a period of rapid ice surface lowering during the mid-Holocene.

In an effort to identify higher-elevation glacial activity and long-term erosion history, fieldwork was undertaken along the northern flank of David Glacier from the D'Urville Wall area (Mt Neumayer to Cape Phillipi) (Fig. S5). The D'Urville Wall is a steep, high-elevation (> 400 m above the ice) granite outcrop delimiting the northern flank of David Fjord. The geomorphology and exposure ages from two field sites situated high above the David Glacier (Mt Neumayer and Cape Phillipi) limit the reconstruction of the glacier surface during the LGM. Mt Neumayer extending above the D'Urville Wall forms a rounded summit with faint striations sub-parallel to modern ice flow and few scattered erratics . In areas 400 m above the ice, bedrock samples contain weathering rinds and deep weathering pits filled with erratics . Bedrock exposure ages from Mt Neumayer (649 m above the ice, 642 ± 61 ka), a terrace on top of the D'Urville Wall (418 above the ice, 116 ± 10 ka) and Cape Phillipi (∼ 300 m above the ice, 532 ± 52 and 957 ± 98 ka) suggest either a thin cover of cold-based ice or ice-free conditions through the LGM. Bedrock exposure ages from the D'Urville Wall area (including Mt Neumayer and Cape Phillipi) do not allow a precise estimate of the past ice surface along the northern flank of David Glacier. Supported by geomorphic evidence from Hughes Bluff, which indicates ice thicker than 230 m, and the LGM limit of ∼ 400 m a.s.l. derived from drift deposits in TNB, we suggest the past ice surface was between 300 and 649 m higher than today in this area (Fig. S2).

Using the high-resolution chronology from Hughes Bluff and Mt Kring, we derive a mean estimate of palaeo-thinning rates along David Glacier using a weighted least squares regression scheme within the iceTEA plotting tools (http://ice-tea.org/en/) . These reconstructed glacier thinning rates are compared to modern thinning rates derived from satellite data and continental-scale ice sheet models . At Mt Kring, we reconstruct a 2 ka thinning event from 7.5–5.5 ka of up to 0.19 m/yr (0.06–0.19, 2σ) thinning rate (Fig. A). At Hughes Bluff, we reconstruct a period of rapid thinning from 6.7–6.2 ka of up to 2 m/yr (0.19–2.06 m/yr, 2σ) followed by ∼ 4 kyr period of minimal thinning (Fig. b). The reconstructed palaeo-thinning along the David Glacier during the mid-Holocene is synchronous with rapid thinning reconstructed at a number of sites in the Weddell embayment , Amundsen Sea sector , Marie Byrd Land and the Transantarctic Mountains . However, the rate of palaeo-thinning reconstructed at Hughes Bluff is one of the highest rates recorded in Antarctica, comparable to a reconstruction from nearby Mackay and Mawson glaciers .

Previous work from inland sites in a similar setting as Mt Kring suggested that such sites changed little or may have thickened during the Holocene due to accumulation increases . In contrast, the synchronous timing between thinning at Hughes Bluff and Mt Kring but lower thinning rate observed at Mt Kring suggests that this thinning event was driven by ocean–ice-sheet interactions at the coast that rapidly propagated inland. Such dynamic thinning, where increased ice shelf basal melting and grounding-line retreat leads to accelerated flow and inland thinning, is well documented along modern marine-terminating margins of the Greenland and Antarctic ice sheets , yet identification in the geologic record is rare, primarily due to a lack of exposed bedrock in the upper reaches of glacier catchments. The thinning history from David Glacier allows for a unique comparison with the broad pattern of dynamic thinning derived from the modern satellite record and suggests dynamic thinning can occur > 100 km into the interior of the EAIS and can persist over multiple millennia.

During the LGM, ice core records and numerical model outputs suggest that the EAIS experienced widespread thinning in its interior, while coastal sites experienced extensive thickening . Our data from Mt Kring show that the ice sheet was thicker than present at this site during the LGM and recorded at least 171 m of thinning during the Holocene. This provides a critical tie point between the high-elevation, low-accumulation ice domes where ice cores are drilled and the low-elevation, high-accumulation coastal sites with more abundant geologic data. Mt Kring is the first site along the high-elevation margin of the EAIS to constrain LGM ice sheet behaviour, and it represents a unique site to compare with other ice sheet reconstructions and continental-scale ice sheet modelling experiments.

The chronologies from David Glacier provide critical insights to the rate, timing, style and magnitude of thinning from a marine-based outlet glacier. The thinning histories presented here provide context to modelled continental-scale ice sheet reconstructions. Using a data–model comparison software, http://dmc.ice-d.org/, last access: 21 October 2021, we extracted a modelled elevation history for David Glacier from five different ice sheet models (four continental-scale and one regional-catchment scale) and one ice sheet reconstruction created to drive a GIA model . To overcome differences in spatial resolution, the scheme extracts an interpolated ice elevation history from the model grid cell containing the field site and its neighbouring model grid cells. The suite of ice sheet models in which we compare against our geologic data represent a variation in model grid size, flow approximations, model physics used, forcing and boundary conditions. We use them to provide a first-order approximation for expected results from the ice sheet modelling community.

The resulting data–model comparisons for Hughes Bluff and Mt Kring reveal a noticeable mismatch in time during the main phase of thinning (Fig. ). For Mt Kring, ice sheet models indicate a phase of thinning that precedes our thinning history by ∼ 4–7 kyr. For Hughes Bluff, the timing lag is comparable to Mt Kring with one notable exception being the deglacial model of ICE-6G . This improved match is likely because ICE-6G is constrained by multiple relative-sea-level curves along the Scott Coast . The style of thinning is variable between models with noticeable short-lived pauses in thinning, mainly during the Antarctic Cold Reversal (∼ 15–13 ka) . Relative to Hughes Bluff, the magnitude of modelled elevation changes at Mt Kring geometrically fit better with the surface exposure age data. At Mt Kring, the average modelled thickness change throughout the deglacial phase for all models in Fig. a is 190 m ± 117 m, which compares well with the 173 m thickness change derived from our ice thinning chronology. In contrast, at Hughes Bluff, we capture only 171 m of thickness change over the deglacial period, and the average modelled thickness change for all models in Fig. b is 623 m ± 142 m. These comparisons suggest that our surface exposure ages capture more of the overall ice thinning at Mt Kring than at Hughes Bluff. Overall, the data–model mismatch may be related to (1) individual topographic features/outcrops not being spatially resolved in the models, (2) limited constraints on ocean/atmosphere forcing e.g. and (3) poorly constrained model parameters that influence basal sliding, isostatic adjustment and ice flow/rheology, which impact the rate of ice sheet response to a climate forcing .

Continental-scale ice sheet model reconstructions and available geological data suggest that the glacier thinning profile at Hughes Bluff likely only captures the final 200 m of the approximately 400+ m thinning since the LGM (Fig. S2). In total, marine geological evidence, ice modelling and new results in this study suggest that at the LGM the expanded David glacier had an ice surface profile above Hughes Bluff and a grounding-line location near Coulman Island . Flowline modelling experiments starting from this ice extent and thickness capture both the magnitude and rate of onshore thinning derived from surface exposure data.

The new geological reconstruction of ice surface elevation changes at Hughes Bluff shows a rapid lowering of the David Glacier at 6.5 ka and a period of slow thinning from ∼ 6–4 ka. Given the marked slowdown in thinning rate from ∼ 6 ka at Hughes Bluff and its location downstream of the present-day grounding line (Fig. b), we suggest that stabilisation of the Drygalski Ice Tongue occurred after ∼ 6 ka. This finding is consistent with and , who mapped a series of raised beaches along the TNB coastline that mark beach depositional processes in an open ocean setting (i.e. no grounded ice) initiating at 7.2 ka. show that the modern Drygalski Ice Tongue is essential for the development of the modern TNB polynya. Thus, if the Drygalski Ice Tongue formed at ∼ 6 ka as our exposure chronology from Hughes Bluff and the raised beach chronology from TNB suggest, it is likely that the TNB polynya has also existed since this time.

General agreement between palaeo-thinning rates derived from surface exposure ages of glacial deposits and those from our modelling experiments provides confidence in the approach used in this study. In Table , we show that in the combined forcing experiments we achieve a close match between data-driven and glacier modelled thinning rates. For Hughes Bluff, a maximum thinning rate of 2.06 m/yr compares well with the average modelled thinning rate of 2.27 ± 0.29 m/yr (experiments MS1–3). At Mt Kring, the maximum thinning rate is 0.19 m/yr, and, for experiments MS1–3, the average modelled thinning rate is 0.36 m/yr ± 0.02 m/yr.

Our modelling demonstrates that glacier thinning at Mt Kring is sensitive to grounding-line migration during ice retreat from the outer to the inner continental shelf. Mt Kring is part of a broad bedrock platform along the northern flank of the major glacial trough dissecting the TAM (Fig. S1). The platform is comprised of three high-elevation outcrops protruding ∼ 200 m above the ice, with Mt Kring nearest to the zone of streaming ice (e.g. ice surface velocities > 100 m/yr). Geophysical characterisation of the northern TAM suggests the presence of individual tectonic blocks bounded by faults which likely serve as zones of relatively weak rock strength allowing preferential ice flow and glacial erosion . In West Antarctica, dynamic thinning of the inland ice sheet has been linked to underlying tectonic controls . At David Glacier, similar tectonic controls may have conditioned the spatial pattern of dynamic thinning during the Holocene.

Along the upper reaches of the David Fjord, there are no outcrops in the zone of highest ice surface velocity (> 100 m/yr). Mount Kring lies 40 km away from the modelled flowline, in an area of relatively slow flowing ice. Modern ice surface velocities near Mt Kring are ∼ 15 % of the ice surface velocities at the projected flowline position . On modern ice sheets experiencing dynamic thinning, satellite-derived thinning estimates are largest in the centre of the ice streams or outlet glaciers and become progressively smaller at lower velocity sites further from the central flowline. For example, the central parts of Greenland’s outlet glaciers are currently thinning at rates of ∼ 0.84 m/yr, while marginal areas with slower ice velocities are thinning at 0.12 m/yr . Therefore, the Mt Kring site likely reflects ice stream marginal thinning (maximum of 0.19 m/yr, derived from surface exposure data) rather than the greater thinning rate that was experienced in the centre of an ice stream (modelled average maximum 0.36 m/yr ± 0.02 m/yr). Taken together, the Hughes Bluff and Mt Kring chronologies suggest that ∼ 2 kyr of dynamic thinning occurred at David Glacier and that this thinning propagated significantly into the ice sheet interior.

Overall, the style of modelled grounding-line retreat is controlled by marine ice sheet instability, a positive feedback where grounding-line retreat into overdeepened subglacial basins leads to progressively enhanced ice discharge and ice sheet thinning . Experiments designed to simulate glacier–ice-sheet decoupling show that once independent, the David Glacier grounding line rapidly retreats through the Drygalski Trough and stabilises on a bathymetric sill at the mouth of the David Fjord.

This study provides insights into the processes that occurred as a large grounded section of an ice sheet retreated into discrete outlet glaciers. Previous descriptions of this retreat have focused on grounded ice in the Ross Sea as a whole . However, at the scale of David Glacier, retreat was likely influenced by local processes including interactions with adjacent glaciers and other ice bodies. For example, a recent ice sheet retreat reconstruction of suggests that ice lingered on higher-elevation banks as the grounding line retreated in adjacent, large bathymetric troughs. Modelled grounding-line retreat may have been slower in reality due to lateral buttressing from outlet glaciers on the TAM coast and stagnant ice on bathymetric highs immediately seaward of the expanded David Glacier, and such lingering ice is not accounted for in our simplified modelling approach.

Knowledge of this complex lingering ice history is poorly constrained as the majority of research has focused on trough axes , whereas offshore Mawson Glacier, south of David Glacier, submarine geomorphology reveals a complex record of lingering ice . Analysis of marine sediments, primarily from trough axes, indicates that a calving bay environment formed during grounding-line retreat . In this scenario, grounding-line retreat is rapid along the trough axis, while lingering ice remains grounded along the lateral margins of the ice stream/glacier. In addition to sedimentary evidence, the abundant large iceberg keel marks seaward of the Coulman Island grounding zone wedge (GZW) and smaller keel marks within TNB provide evidence for a calving bay during deglaciation. Short-term grounding-line stagnation may have been facilitated by complex interaction with other outlet glaciers (Reeves and Priestley glaciers together forming proto-Nansen Ice Sheet/Shelf) of TNB or grounded ice lingering on the banks surrounding the Drygalski Trough, or both.

During the first ∼ 500 model years, the modelled grounding line initially retreats to the location of a large grounding zone wedge . A lack of modelled grounding-line stability at other, smaller GZWs suggests that the grounding line may only have experienced short pauses in these locations. Further, in the case of the much smaller GZWs observed in the deepest portion of the Drygalski Trough, the elongate morphology of these smaller GZWs may reflect a point source versus a line or zone source associated with more classically defined sheet-like GZWs. Regardless, the small mounds in the trough axis are likely to reflect short-lived grounding line pauses during overall retreat.

Generally, once initiated, the modelled David Glacier grounding line retreats rapidly and pauses at a prominent sill at the mouth of David Fjord for up to 5 kyr before subsequent grounding-line retreat to its modern configuration. This simulated two-phase grounding-line retreat aligns with phases of enhanced (at ∼ 6.5 ka) and reduced thinning (from 6 ka) identified in our onshore reconstructions at Mt Kring and Hughes Bluff, both in terms of timing and rates of past glacier thinning. Models forced by moderate SIMR and LBR (MS1–3) fit best with onshore geologic constraints.

Figure  synthesises the results of the terrestrial thinning chronology, modelled glacier flowline behaviour and the existing regional marine retreat chronologies. This synthesis suggests that beginning at 7.5 ka, the grounding line unpins from the sill at the mouth of the David Fjord, the David Glacier and proto-Nansen Ice Shelf decouple, and widespread onshore thinning is initiated. Between 6–5.5 ka, the grounding line retreats to near its modern position, thinning slows significantly and open marine conditions prevail regionally. In summary, geological evidence from offshore David Glacier indicates that retreat of grounded ice through the Drygalski Trough and the formation of open marine conditions similar to today occurred immediately prior to the dynamic thinning of David Glacier recorded in this study . Our onshore surface exposure data record the final stages of glacial thinning and retreat along the David Glacier.