Exploration of an ice-cliff grounding zone in Antarctica reveals frozen-on meltwater and high productivity

Ice fluxes across the grounding zone affect global ice-sheet mass loss and sea level rise. Although recent changes in ice fluxes are well constrained by remote sensing, future projections remain uncertain, because key environments affecting ice-sheet dynamics – the ice-sheet bed and grounding zone – are largely unknown. Here, we used a remotely operated vehicle to explore the grounding zone of a Weddell Sea tidewater ice cliff. At 148 m we found a 0.3–0.5 m gap between the ice and the seafloor and a 0.4 m clear facies of debris- and bubble-free basal ice, suggesting freeze-on of meltwater in the distal marine portion of the ice sheet over the last 400 yr. Ploughmarks and low epifauna cover reveal recent grounding line retreat, as corroborated by satellite remote sensing. We found dense algal tufts on the ice cliff and high phytoplankton pigment concentrations, suggesting high productivity fuelled by nutrients from ice melt. As grounded tidewater ice cliffs rim 38% of the Antarctic continent, sinking and downwelling of organic matter along with low benthic turnover may contribute to enhanced carbon sequestration, providing a potentially important feedback to climate. Direct observations at the grounding zone of a tidewater glacier ice cliff in the Weddell Sea, obtained by remotely operated vehicle, reveal evidence for a freeze-on of meltwater above the seafloor, grounding line retreat and high productivity.

T he largest repository of fresh water on the planet, equivalent to 58 m global sea-level rise, is locked up in the Antarctic Ice Sheet, with a mass balance governed by the processes at the interface between ice, atmosphere, ice-sheet bed, and ocean 1 . Remote sensing has greatly improved observational records over the last decades, and concomitant advances in numerical modelling now provide plausible predictions on the response of the ice sheet to climate change 2 . Whereas the large-scale atmospheric and oceanographic drivers affecting ice-sheet dynamics appear well constrained, large uncertainties still remain in parametrizing the small-scale processes at the boundary between the ice, the icesheet bed, and the ocean 3 , such as the presence and movement of liquid subglacial water and the feedbacks of subglacial hydrology, temperature, and pressure on basal sliding at the ice-bed interface 4 . Observational data are needed to improve our understanding of these critical processes, reduce the uncertainties in the models, and achieve better predictions. However, groundtruthing is a logistical challenge because the base of the ice sheet is buried under hundreds to thousands of metres of ice, and for much of the Antarctic, extensive ice shelves impede access to the grounding zone 5 . Major drilling expeditions have been carried out to access the base of the ice sheet and the grounding zone in some of the large ice shelves and glaciers. They yielded ice cores with laminated ice facies 6,7 , samples of subglacial water 8 as well as remotely operated vehicle (ROV) 9 , and borehole camera images of the seafloor showing a strong biological connection between the glacial and the open-water marine environment [10][11][12] .
Grounding zones are located up-ice of the calving front along ice shelves, but in areas with low ice-flow rates relative to rates of marine melting and calving, the grounded ice sheet ends abruptly as a vertical cliff or tidewater terminus 13 . Whereas tidewater cliffs offer direct access to the grounding zone, ice fall from an unstable calving front is a threat that has so far discouraged ice-or ship-based explorations. During RV Polarstern cruise PS111, we were able to approach to less than 50 m of the Coats Land coast (Fig. 1), one of the few sectors in the Weddell Sea that features a tidewater ice cliff, providing a rare opportunity to explore the grounding zone. Here, we combine ROV observations of the ice cliff and seafloor, hydrographic measurements, and available ice sheet and topography data to provide an integrated glacio-marine and ecological perspective of the grounding zone of an Antarctic tidewater ice cliff.

Results and discussion
Grounding zone. A ROV, equipped with high-definition cameras and oceanographic sensors, was deployed along a tidewater ice cliff. The entire face of the white meteoric ice showed the typical scalloped melt cups 14 . At a depth of 148 m, we reached the base of the cliff marking the grounding zone-the junction between the grounded ice, the seafloor, and the ocean 15 (Fig. 2a). Photogrammetric reconstruction of a 4 × 1 × 1 m section of the grounding zone showed a 0.3-0.5 m gap between the base of the ice and the seafloor (Figs. 2 and 4b).
A striking discovery was a 0.4 m facies of transparent ice frozen to the base of the white meteoric ice (Fig. 2). Video close-ups showed it was bubble-free and largely devoid of debris (Supplementary Movie 1). The abrupt transition and absence of bubbles clearly identifies the basal ice as being derived from freeze-on 16 . The lack of debris in the clear facies indicates a subglacial hydraulic system providing sufficient pressure to separate the ice from the bed. Its homogeneity and thickness implies that water supply and pressure were sustained for prolonged periods of time. Transparent ice facies, albeit of lesser thickness, have been found in slow- 16 and fast-moving ice streams 17,18 . However, an important difference in our study is the lack of multilayered basal ice: we found only the transparent facies but no alternating facies of clear and debris-laden ice 7,19 . The only incidence of debris in 90 m of inspected ice front is a small spot in the meteoric ice, i.e., above the transparent facies ( Supplementary Fig. 5).
The base of the transparent ice was square-edged, and its surface on the underside and side lacked the scalloped melt cusps of the meteoric ice. The interface separating the meteoric from the transparent ice facies showed undulations that could be the line of contact between the meteoric ice and the bed, prior to freeze-on (Fig. 2a). In contrast to the undulations in the interface between the ice facies, the bottom of the basal ice was level, causing a total reflection of the seafloor like a mirror ( Supplementary Fig. 2). The flat surface and absence of melt cusps suggests that the basal ice is harder than the meteoric ice and/or is less affected by turbulent flow 14 .
Dynamics of the calving front. Coats Land bed topography acts like a barrier buttressing the ice sheet, with mountains at the coast (Fig. 1b). Ice-flow velocities are extremely low: inverse modelling shows that it takes the ice more than 3500 yr to travel the final 18 km to the calving front ( Fig. 3b and Supplementary Data 5). The observed increase in velocity toward the terminus causes thinning of the terminal 6-7 km of the ice sheet to about half of its former thickness (Fig. 3a). For a tidewater ice cliff in hydrostatic equilibrium at a seawater density of 1028 kg m −3 ( Supplementary  Fig. 3), a thickness of 170 m, and a freeboard of 20.8 ± 0.7 m ( Supplementary Fig. 4), we calculate an average density of meteoric ice of 901 kg m −3 . This is much more than the average density of tabular icebergs in the Weddell Sea (845 kg m −3 ; n = 24, ref. 20 ) and implies compaction of the ice in this slow-moving part of the ice sheet. If the reported ice density value 20 is correct, however, the difference between ice thickness and floatation thickness suggests that the ice front can accommodate nonhydrostatic water pressures and was depressed below hydrostatic equilibrium. Finite element analysis (FEA) of ice sheet deflection ( Supplementary Fig. 8) shows a maximum spring tidal flexure at the calving front of only 2 mm-much less than the observed gap beneath the ice (0.3-0.5 m) and the high tide at the time of our study (0.54 m), as determined from the Tide Model Driver (TMD 2.5) 21,22 (Supplementary Fig. 1 and Supplementary Data 6). Unless the integrity of the ice sheet is compromised by fracture, we thus expect no tidal variations in grounding line position and, hence, no tidal pumping, at the tidewater terminus. The final kilometre features deep crescent shaped crevasses parallel to the ice front (Fig. 1c, d). We propose calving of ice bands according to the spacing of crevasses (~150 m) in the terminal section of the ice sheet (Fig. 1d). As the ice front advances 29.8 m yr −1 (Fig. 1c), calving is triggered, on average, every 5 yr. Figure 1d shows the satellite image acquired on 24 February 2018, 5 days prior to our study, with (i) the position of the ice front, digitized from a satellite image taken roughly 2 yr before on the 31 December 2015 (white line) and (ii) the modelled ice front at the time of our study, derived from the 31 Dec 2015 position and the ice flow data (blue line) [23][24][25] (Supplementary Data 3 and 4). The blue line shows a near perfect fit with the calving front in the southern two thirds of Fig. 1d, showing that the advance of the calving front is predicted very well by the flow model. The upper section of the figure shows that the calving front trails the blue line by 300 m. This part of the tidewater terminus must have disintegrated in the period between the two satellite images. The 300-m-wide grounded iceberg off the calving front is likely to have originated north-east of our study site (orange contour, Fig. 1d).
Basal ice origin. Two preconditions have to be met for basal freeze-on to occur: the availability of subglacial meltwater and temperatures below the pressure-melting point. We derive a maximum ice-sheet thickness along the path of the ice flow of 510 m (red line in Fig. 3b), corresponding to a −0.34 ∘ C change in melting point 26 . Ice flow (~10 m yr −1 , Fig. 1c) and associated deformation stress were assumed to be too low to have a significant effect on internal temperature increase 27 . Whether clean or dirty ice facies is formed depends on the relation between hydrostatic pressure of the meltwater and ice load: If the effective stress p 0 (i.e. the difference between ice and water pressure) is lower than the regelation criterion (ζ = 100 kPa), clean facies is formed at rates of 1-4 mm yr −1 (ref. 19 ). If we discount the unknown portion of clean facies that was lost through melting in the grounding zone, the 0.4 m facies of clean ice we encountered corresponds to up to 400 yr of accretion, suggesting a groundwater source feeding the subglacial water system at a pressure allowing sustained freeze-on. We used an ice-sheet model to determine whether there is a sufficient large basal zone below the freezing point to produce the observed basal freeze-on layer. Friction was calculated as a function of shear stress, normal stress, and ice velocity, assuming ice density as a function of ice thickness (vertically homogeneous), and a linear friction law, where the friction factor was determined as a function of slope angle and ice velocity 28 . Cumulative ice growth (blue line in Fig. 3c) was calculated assuming an accretion rate of 1 mm yr −1 . As the ice descends the seaward slope and reaches the ocean, it is buoyed up and loses friction with the seabed. We find that freeze-on was initiated about 6.5 km upstream, corresponding to the subglacial bed at sea level (Fig. 3c). Our model assumes the availability of subglacial meltwater in the marine portion of the ice sheet, which could be due to a dynamic equilibrium between a groundwater head supplied by meltwater and static seawater 29 .  Temperatures in the water column near the ice cliff are well above the freezing temperature of −1.93 ∘ C for the range of salinities and pressures in our study, and we see that the data in the upper 10-40 m are closely aligned with the ratio predicted from melting ice into seawater (dashed line) on the S A -Θ diagram ( Supplementary Fig. 7c) 30,31 . The salinities in the immediate vicinity of the ice cliff (black line, Supplementary Fig. 7b) are at or above the salinities measured in open water 250 m away 32 , but the temperatures are consistently lower in the top 40 m of the water column (black line, Supplementary Fig. 7a). Whereas ice melt is also demonstrated by the melt cusps in the ice cliff, the lack of a salinity gradient suggests no substantial discharge of freshwater at the grounding line. We could not find any indication (runnels, suspended matter) 9 suggesting subglacial discharge along the 90 m of grounding line we inspected.
Seafloor. To assess how the decadal waxing and waning of the calving front affects the seafloor, we carried out a 230 m ROV transect normal to the grounding zone. We found a heavily scoured seafloor with half-buried boulders and clasts on the coarse-sediment surface. The slope of the seafloor (2.5%) was identical to the seaward slope under the ice sheet, suggesting a glacially levelled seafloor. Erosional planing of the seafloor is supported by observation and modelling. Deep subparallel scours running in a NWW direction along the ROV transect (Figs. 1e and 4d) mark previous advances of the ice terminus. FEA confirms an ice terminus advancing with insignificant lift or bending, consistent with a flat seafloor planed off by the advancing ice front ( Supplementary Fig. 9). In order to determine origin and transport paths of the clasts encountered on the seafloor in the grounding zone, we analysed their shape after Powers 33 . The predominantly subrounded shape of the clasts (Supplementary Table 1) suggests that the clasts are derived from a basal zone of shearing and crushing 34,35 . A supraglacial origin can be ruled out by the lack of angular clasts and by the absence of nunataks or valley sides in the study region (Fig. 1). Subbottom acoustic profiling data showed strong surface reflectors, low penetration and no subbottom reflectors ( Supplementary Fig. 6), confirming the visual impression of a glacially compacted till which is typical for shelf diamicton in the area 36 . At and near the calving front, sedentary fauna were rare: a large ascidian (Molgula sp.) was found between clasts under the calving front (Fig. 4b), occasional stalked ascidians and sponges (Homaxinella balfourensis) were found perched between clasts, polychaete tubes and bryozoans were scattered on the sides (but not: tops) of the clasts, covering less than 5% of the available surface (Supplementary Movie 1). All of the above pioneer species are fast-growing early colonizers of recent scour marks (stage 0) 37,38 . Mobile fauna included fish (adult Pagothenia borchgrevinki, occasional channichthyids), ophiuroids, and crinoids. At some 50 m distance from the calving front, the first patches of post-scour growth appeared (stage 1) 37,38 , featuring loose stands of H. balfourensis and bryozoans, bordering abruptly with barrens caused by iceberg scour (Fig. 4c). At the end of the transect, the ROV transited into the next postscour successional stage (light green area, Fig. 1e, stage 2) 37,38 , featuring the appearance of the first gorgonians (Thouarella spp.), the first lollipop sponges (Stylocordyla chupachups), denser stands of H. balfourensis and epizooic holothurians on the upper parts of the gorgonians and rocks (Fig. 4d). The appearance of stage 2, which also harboured a denser fish fauna, corresponds to the maximum extent of ice at the end of 2015 (white line, Fig. 1e). We propose that this line is less frequently exceeded by the advancing calving front, so that it allows for more time of settlement by benthic fauna.
Ice cliff and water column biota. A surprising new finding, to the best of our knowledge, is the abundance of algal tufts lining the ridges separating the scalloped melt cusps of the meteoric ice (Fig. 4a), particularly in the upper marine part of the ice cliff (Fig. 5a). The ice itself showed no trace of pigmentation, suggesting that drifting phytoplankton (not sympagic algae) seeded the growth of algal tufts on the submarine ice cliff face. Given the dominance of chain-forming diatoms in Weddell Sea phytoplankton blooms 39 , we propose that chain-forming diatoms drifting with the currents along the ice front get caught up in the open pores of the elevated ridges of the melting ice, where they proliferate under the relatively more favourable light, flow and nutrient environment. Particularly iron, a trace element limiting diatom growth in the Antarctic 40 , is known to be released from melting meteoric ice 41 encouraging diatom "superblooms" near the calving front 39,42 . These conditions may have favoured the proliferation of algal tufts in our study. Chlorophyll a concentrations in the water column show that a summer bloom was underway (Fig. 5b) with peak values (>1 mg m −3 ) at the thermo-and halocline at 20 m ( Supplementary Fig. 3), corresponding to the highest levels that can be sustained in summer in open waters of the Weddell Sea 43 . The phytoplankton maximum coincides with the maximum of the algal tufts (Fig. 5a) and contributes to the maximum in oxygen concentration (Supplementary Fig. 3) due to photosynthesis 44 . The lower margin of the algal tufts at 90 m (Fig. 5a) marks the compensation depth 44 , where algal growth equals respiration, while phytoplankton concentrations remain high beyond the compensation depth down to the seabed, showing that the critical depth for the phytoplankton 45 exceeded 148 m at the calving front. Because phytoplankton were abundant beyond the compensation depth, it shows also that the algal tufts are due to biological proliferation,   (Fig. 4a, inset), supporting earlier ice shelf observations 46 . Whereas these carnivores are probably not consuming the algae, they are likely to take advantage of invertebrate prey 47 associated with the tufts.
Ice sheet-ocean and carbon pump. Coastal polynyas open up regularly around the Antarctic Ice Sheet, supporting one of the highest primary production rates in the Southern Ocean 48 , fuelled by glacial inputs of limiting iron 41 . The export of this biomass by sinking and downwelling in the coastal Southern Ocean is considered a global hotspot of carbon sequestration 49 . The flux of organic matter correlates well with the abundance of benthic suspension feeders 50,51 , while deposition and burial of food particles appears to be inversely related to their species richness [51][52][53] . Iceberg keel scouring reduces the biomass of benthic consumers 54 and supports carbon deposition and burial [51][52][53] . A given seafloor location on the Antarctic continental shelf is disturbed by icebergs, on average, every 340 yr 55,56 . Scouring is much higher at our study site (once every 5 or 10 yr) leading to an extreme paucity of benthic consumers by Weddell Sea standards 38 , but also low relative to Mackay Glacier seafloor community in the Ross Sea, the only other Antarctic grounding zone study available 9,57,58 . We therefore postulate that a reduced mineralization of organic matter in combination with a high flux of phytodetritus from phytoplankton and ice cliff algae enhance organic carbon export and burial in the grounding zone 49,53 . As tidewater ice cliffs fringe 38% of the Antarctic coastline 5 , our findings have potential importance for carbon sequestration.
Implications for glaciology and biology. Our ROV dive to the tidewater ice cliff grounding zone of Coats Land in the Weddell Sea revealed a basal ice layer of exceptional transparency, an impoverished benthic fauna affected by scouring of the seafloor, and a rich flora on the calving front and in the water column, likely fuelled by glacial inputs of iron (Fig. 6). We show that the basal ice was formed by slow centuries-long freeze-on to an ice sheet lifted from its bed by liquid subglacial meltwater, most likely in the marine portion of the ice sheet (Figs. 3c and 6), which could be due to a meltwater head replacing denser seawater 29 . Traces of basal debris found in the meteoric ice above the clear ice layer support the hypothesis that the meteoric ice must have been in direct contact with the ice-sheet bed, before being wedged apart by basal freeze-on. Calving of icebergs occurs on average every 5-10 yr. The advancing ice front leads to heavy scouring of the seafloor, preventing the build-up of benthic biomass. Meltderived nutrients and a favourable light environment support high growth rates of the phytoplankton and the ice cliff algae. Phytodetritus from both sources accumulates near the grounding zone in the absence of benthic consumers, increasing the availability of organic carbon for export and burial. By contributing to the control of the amount of the greenhouse gas CO 2 , the biological pump in grounding zones may play an important part in the global carbon cycle and in regulating climate 49 .
Determination of freeboard height. Freeboard height was derived from ROV video footage. The ROV was hovering over Polarstern's working deck perfectly balanced (pitch 0 ∘ ) 5.6(±0.2) m above sea level, prior to deployment. Hence, distance between image centre and water surface was equal to ROV's elevation. Freeboard of the ice front was approximated according to this reference (Supplementary Fig. 4).
Two-dimensional scaling. Reference lasers marked a distance of 10 cm vertically and horizontally. In cases where the lasers were not well reflected, i.e. on the ice's surface, an altimeter was used to measure acoustically the distance from camera to the observed object. Length and areal scales were calibrated on images where lasers Fig. 6 Main findings of the Coats Land grounding zone study. Basal meltwater contributes to basal freeze-on in the marine portion of the ice sheet, where the ice is buoyed up, loses friction with the ice bed and speeds up. Continuous accretion of basal ice over 400 yr (6.5 km) contribute to the transparent facies of 400 mm at the calving front (blue layer at ice base). Basal debris picked up by the meteoric ice in the terrestrial part of the ice sheet melts out at the calving front. Melting releases dust-borne iron (Fe, green arrow) that fuels the growth of algae on the ice front and in the water column. Downwelling of High Salinity Shelf Water (HSSW) associated with sea-ice formation and sinking phytodetritus (black arrows), along with low mineralization by an impoverished benthos lead to a high export of organic matter. Debris, algae, and basal-ice layer not drawn to scale.
are clearly visible and no obstacle blocked the acoustic measurements.
Image width (w) was determined as a function of altimeter distance (x, in metres). Image height h was calculated as where a = 16/9 represents the aspect ratio.
Tuft algae cover. Relative cover of algae attached to the ice front was derived from rgb (red green blue) standardized video images. The video footage was converted into a frame sequence (one frame per second, total 1158 images) at 4k resolution (3840 by 2160 px) using FFmpeg 3.4.8. by Fabrice Bellard. The changing illumination, from predominantly natural light near the surface to almost exclusively artificial light (LED of the ROV) at depth, required colour normalization of the images. This was done by substracting the overall average rgb value from the stack from the rgb value of the respective image. In order to reduce horizontal variations caused by spotlight illumination, towards the left and right border of each frame, this method was applied columnwise to the pixel matrix. Colour values were restricted to 8 bit integers with a lower and upper cut-off at 0 and 255. To differentiate algae from underlying ice, rgb images were converted into 8 bit hsv (hue saturation value) integers. Hue values less than 132 were considered as algae; hence, relative cover is the number of pixels for hue less than 132 divided by the number of image pixels. All calculations and conversions were made with scilab 6.0.2 and the IPCV 4.1.2 (Image Processing and Computer Vision) toolbox.
Three-dimensional reconstruction. Video footage was used for stereophotogrammetry. FFmpeg was used to convert the video files into 25 frames per second at 4k resolution (3840 × 2160). Four metres of grounding zone were reconstructed from 57 frames in Metashape v1.6.2(64 bit) from Agisoft (Supplementary file "GLineMeshImg.zip"). Frames were aligned sequentially with high accuracy, a key point limit of 1.5 M, and 50 k tie points. A dense cloud was built in medium quality and mild depth filtering. The mesh was reconstructed from the dense cloud with high face count setting. The resulting three-dimensional model was scaled to conform with the distance of 10 cm given by reference lasers. Local measurements could be taken on the scale-corrected model without distortions of images perspectives (Fig. 2b).

Ice-sheet modelling
FE analysis of ice sheet flexure. Numerical FEA was used to determine the degree of ice-sheet bending caused by ice depressed below neutral buoyancy. Ice thickness was taken from the IBCSO bathymetric model 59 along a cross section at the ROV site parallel to the ice flow. Water depth was derived from the IBCSO bathymetric bed model 59 and adapted for a high-water and low-water spring tide level. Buoyancy of ice below sea level and weight of ice above sea level were calculated as a function of ice thickness H derived from ice densities presented by Orheim 20 (Eq. (4)). Seawater density was taken from CTD data (Supplementary Fig. 3 and Supplementary Data 2) assuming a mixed water column ρ water = 1028 kg m −3 . Difference of ice weight and replaced water is the effective load deflecting the ice (see Supplementary Figs. 8b and 9c). Load was applied in 50 m integrated intervals. Geometry of the ice sheet was built in FreeCAD version 0.19, meshed with Netgen, and solved by Calculix. Mesher and Solver are implemented in the FEA module of FreeCAD. Tetrahedra elements were generated using user defined settings of two segments per edge with a growth factor of 0.3. Differential equations were solved using the linear static method of Calculix, assuming the following values for Young's modulus and Poisson's ratio, respectively: (E = 9 GPa, ν = 0.3; ref. 60 ).
Ice density ρ ice was expressed as a function of ice thickness H derived from data given by Orheim 20 : ρ ice ¼ 527:062H 0:090 ðR 2 ¼ 0:976Þ ð 4Þ Ice thickness and water depths were determined as linearly regressed function along the ice cross section x. Origin of x was chosen at the ice front. Relation of ice thickness H (Eq. (5)), water depth D h at high-water spring tide (Eq. (6)), and water depth D l at low-water spring tide (Eq. (6)) was regressed over the last 1000 m of the ice sheet terminus.
Analytical beam validation of FEA. Analytical validation of the numerical approach was simplified as a beam.
Assuming a very rigid beam, the integral is zero where negative and positive moments cancel each other out.
The total moment consequently takes effect on the region between x M 0 and 150 m. The total force F of the moment is then, Location x of the effective force F is then (Supplementary Fig. 10) Material properties are given above. Coordinate centre is point F (the limit of flexure) in Supplementary Fig. 8a and x = 150 m represents the calving front.
Basal friction. A general friction law that brings basal shear stress (τ) and effective normal stress (N) in relation to the ice flow velocity (u) is given by 28 Considering an ice column of 1 m 2 , the effective shear stress τ equals friction force R. Assuming velocity u constant at a given location, the sum of all forces at this point is zero; thus, downhill-slope force D is equal to friction force R. When τ = R and R = D, with D ¼ sin α F g (F g : weight of ice), τ ¼ sin α F g . Neglecting basal-water pressure, normal force applied on the slope is N ¼ cos αF g . Eq. (15) is then sin α F g ¼ C cos α F g u 1=m ð16Þ Assuming ice velocity has a linear effect (m = 1), ice is incompressible, mass flux is constant, we can solve Eq. (16) for C to obtain the friction coefficient, where sin=cos ¼ tan: As the ice velocity is effected by thinning, we modified the original Eq. (15) 28 by replacing velocity u with the ice flux (flux = uH). By introducing velocity and flux into the general friction law, friction coefficient C was replaced by the friction factor f with the unit [s m −2 ].
Ice front extrapolation. The shape of the 2015 ice front was digitized from satellite image TDX1-SAR © DLR, 31 December 2015 using QGIS. The basis for motion interpolation of the ice front was given by the ice flow vector field from Rignot et al. and Mouginot et al. [23][24][25] (Fig. 1c). Flow components of the 2015 ice front were determined by the radial basis function (rbf) and linear interpolation from the scientific python (scipy) module. Every point of the 2015 ice front was extrapolated 25.5 months (31 Dec 15 to 18 Feb 18) into the future (light blue line in Fig. 1d). We found implausible data at the boundary. In order to avoid potential contamination and bias from boundary effects, we removed all boundary data.
Clast shape distribution. We analysed the shape of the clasts on 22 randomly selected images (n = 1026) using Powers' semi-quantitative roundness chart (Supplementary Table 1) 33 . Observed roundness was compared to published "basal debris" and "rockfall" roundness distributions 35 .

Data availability
Owsianowski

Code availability
Code will be made available by the corresponding author upon reasonable request.