Variations in ploughmark morphology reflect the type of iceberg that formed them. Narrow, single V-shaped ploughmarks are characteristic of scours produced by small, individual icebergs calved from a marine terminating ice margin that is probably grounded, e.g., refs. ^{53, 54–55}. This morphology of iceberg ploughmark is present in all of the HR3D seismic datasets examined. The second class of ploughmark is broader and has a flatter base ( > 200 m wide) reflecting the imprint of a blocky iceberg with large dimensions or the grounded corner of a tabular berg, e.g., ref. ⁵⁶. At between 300–2350 m wide, the subtle multi-keeled grooves which comprise the third class of ploughmark are one to two orders of magnitude wider than the other types observed in the HR3D seismic data and are only observed within the deeper waters of the Witch Ground Basin and west of the Norwegian Channel (Fig. 1a). The distinctive parallel groove morphology of these features reflects either ploughing by drifting tabular icebergs with massive dimensions (Fig. 4), e.g., refs. ^57,58 or incision by drifting sea ice pressure ridges, e.g., ref. ⁵⁹. These formative mechanisms can be distinguished simply by considering the palaeo water depths at which the ploughmarks were incised, as the keels of sea ice rarely exceed 20 m deep even where pressure ridging has occurred^34,35,59. Palaeo-sea level reconstructions that account for eustasy and isostasy^60,61 demonstrate that water depths in the region of the central North Sea where the multi-keeled grooves are buried were between ~50–180 m during the growth and retreat of the last BIIS (31–15 ka). This suggests that large icebergs, rather than thinner sea-ice floes, were responsible for the seafloor scouring. Given both the tendency for the ploughmarks to change direction over short ( < 200 m) distances (e.g., Fig. 5a), and the absence of multi-keeled ploughmarks further south in the basin where water depths would have been too shallow to accommodate large icebergs (Fig. 6)⁶⁰, we conclude that the grounding of tabular icebergs was the formative mechanism of the multi-keeled grooves observed in the HR3D seismic data (Fig. 5).Fig. 4

a Multi-keeled iceberg ploughmarks on the Weddell Sea continental shelf. b Cross-cutting multi-keeled iceberg ploughmarks, up to 5 km wide, on the seafloor of the Barents Sea⁵⁸. c Cross-sectional profiles of multi-keeled icebergs in the Weddell, Barents and North seas.

The geomorphological imprints of modern iceberg interactions with the seafloor are challenging to observe in action and, as a result, it is difficult to infer iceberg dimensions from the palaeo-record. One of the most informative examples capable of relating iceberg size to ploughmark dimensions exists on the mid-Norwegian margin where a 2 km wide multi-keeled iceberg ploughmark has been associated with a nearby planed-off debris-flow lobe, suggesting that the iceberg responsible for forming the ploughmarks was at least twice as wide as the seafloor imprint it left behind³². In other formerly glaciated regions, multi-keeled iceberg ploughmarks are commonly 0.3–5 km wide (Fig. 4)^{53,57,58,65,66}.

Modern tabular icebergs calved from Antarctic ice shelves with thicknesses similar to those inferred to have existed in the North Sea can be larger than 1200 km² in area and more than 50 km wide, such as iceberg A-74 which calved from the Brunt Ice Shelf in 2021 (1270 km²). On the modern seafloor of the Weddell Sea, multi-keeled iceberg ploughmarks up to 1.7 km wide have been observed in contemporary water depths of ~400 m (Fig. 4a)⁶⁶. Tabular icebergs calved in this region between 1984 and 2001 had an average width of 2.4 km, although much larger tabular bergs, such as the 170 × 25 km A-76 (which calved from the Ronne Ice Shelf in 2021), were also produced sporadically^27,67. Given that present-day water depths on Antarctic continental shelves where multi-keeled iceberg ploughmarks are observed are typically ~300–800 m⁶⁸, it is likely that only the largest contemporary icebergs, or potentially even larger icebergs calved from a thicker ice sheet during the last glacial period⁶⁹, are responsible for forming the multi-keeled ploughmarks in the Weddell Sea given their sparsity in this region and elsewhere around Antarctica³⁶. Furthermore, even the largest multi-keeled ploughmarks observed in Pine Island Trough, West Antarctica, are less than 500 m wide despite the glaciers in this region regularly producing tabular icebergs that are up to 1000 s of km² in planimetric area and 10s–100 km wide³⁶.

Whilst it is challenging to estimate the original dimensions of icebergs from the seafloor ploughmarks they produce during drift—especially when most contemporary analogs are located in far deeper waters than those present in the North Sea during the last glaciation—a general observation is that multi-keeled ploughmarks tend to be several times smaller than the width of the iceberg which formed them^32,36,53,66. Using palaeo-water depths⁶⁰ to infer the thickness of the North Sea icebergs, assuming hydrostatic equilibrium, e.g., ref. ⁷⁰, and typical freeboard-width ratios of modern Antarctic tabular icebergs⁶⁷, we estimate that feasible widths of the North Sea tabular icebergs could have ranged from ~500–10,000 m; this range is roughly in agreement with the width of the multi-keeled ploughmarks mapped in the HR3D seismic data. Consequently, given the dimensions of the multi-keeled iceberg ploughmarks observed in the central North Sea and those of their ancient and modern counterparts, the largest tabular icebergs calved from the BIIS during the last glacial period were at least 2350 m wide (the widest observed multi-keeled ploughmark), but their true size was probably significantly larger at 5–10 s km in width.

Modern rates of iceberg calving are primarily set by structural and internal stresses within an ice shelf created as ice flows towards the ice margin^8,12,23. Increased fracturing or thinning due to substantial changes in ice flow or ocean-induced basal melting can diminish the structural integrity of an ice shelf, potentially resulting in a calving regime transition that may drive rapid changes to terminus position, or make an ice shelf more susceptible to external environmental drivers that could precondition or trigger rapid collapse cf. ^8,12. Most rapid disintegration-style events observed to date in the observational era were ultimately triggered by enhanced surface and/or basal melting^{4,7,9,10,13,71}.

The multi-keeled iceberg ploughmarks observed in the Witch Ground Basin are situated directly above glacial lineations ( ~ 35 m b.s.f.) formed by the flow of grounded ice just prior to the retreat of the last BIIS from this part of the central North Sea, an event that has been dated to ~20 ka^51,52. At shallower burial depths within our data at this location ( ~ 12 m b.s.f.), evidence of multi-keeled iceberg ploughmarks disappears and stratigraphic horizons dated largely to 17–14 ka are instead ploughed by smaller single V-shaped grooves; these in turn are buried by postglacial Holocene muds and silts^52,72. This switch in the type of iceberg ploughmark demonstrates that a calving regime shift from broad parallel comb-like to single iceberg ploughmarks occurred sometime between ~19–17 ka, when the sporadic calving of large tabular icebergs ceased and instead switched to the production of many smaller single-keeled icebergs.

In contemporary settings, similar transitions in iceberg production have been associated with ice shelf collapse through the process of repeated iceberg fracture, detachment, and a corresponding acceleration in mass loss for the grounded portions of marine-based ice sheets²². For example, the Larsen B Ice Shelf collapsed at a rate of ~10 km week⁻¹ once the calving of large tabular icebergs ceased and mass was instead removed via the loss of numerous, yet relatively small ( < 250 m wide), pieces of the ice shelf^8,71. Once lost, the reduction in buttressing from ice shelves has an almost instantaneous effect on ice flow^3,73, resulting in rapid increases in ice flux over the grounding line that can propagate far inland with significant implications for ice sheet stability⁸. We therefore interpret the regime shift in iceberg production to represent a significant event in the mass loss from the BIIS: the widespread disintegration of its ice shelves in the North Sea between ~19–17 ka. In this broader context, it is notable that there is a coincidence between the timing of disintegration of the BIIS ice shelves and the saddle collapse of the grounded portions of the BIIS and FIS across the North Sea⁴⁰(Fig. 1a).

An important question, therefore, is whether ice shelf breakup was the main driver of accelerated grounded ice retreat across the North Sea, or was disintegration merely a symptom of enhanced ice sheet mass loss that was already underway when the ice shelves collapsed? Almost all modern ice shelves serve to buttress inland ice⁸, providing a ‘safety band’ which slows the flow of grounded ice towards the ocean^1,2. The efficacy of buttressing is significantly reduced, however, when ice shelves are laterally unconfined by the surrounding ice or bedrock¹⁴, as is the case for several East Antarctic ice shelves². The BIIS’s ice shelves are modeled to be largest in embayments where the margins of the ice shelves are laterally confined, particularly at 20 ka and 19 ka where grounded ice is pinned around the rim of the Witch Ground Basin (Fig. 6). Accordingly, the former ice shelves fringing the BIIS likely restrained the flow of upstream grounded ice towards several of the BIIS’s marine terminating margins. It then follows that any reduction in ice shelf area would reduce buttressing and result in accelerated grounded ice retreat across the relatively smooth low-gradient bed of the central North Sea.

Ice shelves are sensitive to environmental changes in both the atmosphere and the ocean^10,74,75. Their destablization can arise via surface thinning through melting, hydrofracture, lake ponding and drainage^4,19, or through weakening from beneath via basal melting, undercutting, grounding line retreat, or in response to increased ocean swell^18,20,30,76. The thermal limit of extant Antarctic ice shelves has been observed to follow the migration of the –9 °C mean annual air temperature isotherm¹¹ because above this temperature extensive melt ponds form in summer which lead to large scale ice shelf fragmentation and collapse driven by hydrofracturing⁷. Recently, however, it has been argued that the melt-over-accumulation ratio plays a more important role than annual mean temperature for setting the threshold of Antarctic ice shelf stability⁷⁷. Although reconstructions of the demise of the BIIS show that retreat initiated at around 22 ka—thousands of years prior to deglacial warming and CO₂ rise—changes in ice sheet volume appear to mirror variability in June insolation⁴⁰. This suggests that the earlier start of deglaciation may have been triggered by orbitally induced increasing levels of summer solar radiation that produced greater surface melting^40,78. Accordingly, it is feasible that the magnitude of surface melting reached a threshold between ~19–17 ka, causing the ice shelves fringing the BIIS to undergo widespread hydrofracture and collapse, accelerating ice losses around its marine margins. A shift in surface meltwater presence has also been linked to the rapid deposition of thick grounding-line proximal glacimarine muds that were deposited in the Witch Ground Basin at around this time⁷⁹. Alternatively, changes to internal ice sheet dynamics, potentially driven by the effects of sea-level rise and/or atmospheric or ocean warming, could also have altered the calving regime through facilitating different rifting patterns and calving frequencies.

Regime transitions in iceberg calving from modern Antarctic ice shelves currently represent some of the largest uncertainties in sea-level rise projections, yet very few examples of complete ice shelf loss have been observed, and humanity is yet to witness the longer-timescale equilibration of an ice sheet to the loss of its ice shelves⁸. Our observations document that such a regime transition occurred previously in the North Sea, with the style of ice shelf calving shifting from the full thickness calving of tabular icebergs to the rapid calving of numerous smaller ice blocks as the BIIS’s ice shelves disintegrated—an event that coincided with the collapse of the BIIS’s marine-based sectors. The scarce number of ice shelf collapse events that have been documented in the satellite era suggests that similar transitions in iceberg calving regime are often associated with rapid ice shelf loss^4,8.

Such events represent a mere blink of an eye in the long-term lifecycle of an ice sheet and its corresponding sedimentary legacy. Consequently, advances in the resolution and precision of chronological constraints in the North Sea are presently required to decipher whether ice shelf loss triggered the collapse of the marine sectors of the BIIS, or whether disintegration was merely a symptom of wider ice sheet losses that were already underway when its ice shelves collapsed. However, the impacts of such a shift were likely far longer-lasting and more substantial than the duration of the event itself, especially in the case of the relatively flat-bedded areas of the former BIIS. This is because recent work has highlighted how ice shelf thinning, iceberg calving and/or sea-level rise can trigger pulses of very rapid grounding line retreat at rates of at least 2 km yr^-1 80,81, and potentially up to 600 m day ^-182, across relatively shallow-gradient ( < 0.1°) beds after ice contact with stabilizing topographic high points is lost. As much of the regional gradient of the isostatically rebounded central North Sea also falls below this threshold⁶⁰, it is possible that similarly high rates of ice retreat were promoted across the central North Sea as the grounding line unpinned from the rim of the Witch Ground Basin in response to ice shelf disintegration. The North Sea ice shelves documented here may therefore have played a significant role in promoting the stability of the BIIS in response to early deglacial warming. Given that extensive areas of the contemporary Antarctic Ice Sheet are grounded on similarly low or retrograde gradient beds⁸³, further assessment of how the transition in calving regime and subsequent loss of ice shelves documented by our data preconditioned, or even triggered, the demise of the BIIS may provide crucial insights on future sea-level rise trajectories from the Earth’s remaining ice sheets.

We examined the morphology of the former North Sea seafloor using high-resolution 3D (HR3D) seismic data, which reveal intricate morphological structures of glacial origin that cannot be resolved using conventional 3D seismic methods^{84, 85–86}. Seven HR3D seismic datasets were examined in the central North Sea (Fig. 1a) covering a combined area of ~67 km². The data were acquired using two 1200 m-long streamers towed 3 m beneath the sea surface with 96 hydrophone groups at 12.5 m spacing with a response that varied by <±0.5 dB over frequencies between 2–500 Hz, a 6.25 m shot interval and a 1-ms sample rate⁸⁷. The seismic source was two 160-inch³ sleeve airgun clusters with a 20–250 Hz effective frequency band. Data processing included swell noise attenuation, tide correction, multiple suppression, two passes of velocity analysis run at 250 × 250  m intervals, normal moveout correction and bandpass filtering. The final processed datasets consist of time migrated 3D stacks with a 1 ms sample rate, a 6.25 × 6.25 m bin size, a vertical resolution of ~4 m, and a detection limit along individual reflectors of ~0.5 m^84,88. Palaeo-seafloor morphology was examined using sequential horizontal timeslices from the HR3D seismic volumes in S&P Global Kingdom software. Each former seafloor was digitized in 3D as an individual seismic horizon. The mapped seismic horizons were converted from two-way travel time (TWT) to depth using a sound velocity of 1850 m s⁻¹⁸⁹.

Multibeam bathymetric data from the Weddell Sea were acquired from RV Polarstern with a 15.5 kHz ATLAS Hydrosweep DS3 system during cruise PS96 in 2015–2016^66,90. Data were gridded at a 25 m resolution, and iceberg ploughmarks on the seafloor were mapped in ArcGIS software.