Main

Antarctica’s ice sheets are underlain by liquid water made possible by geothermal and frictional heat and the thick insulation provided by the ice itself1. This subglacial water facilitates the rapid sliding of the overriding ice2, transports sediment3, archives ice-sheet history4 and plays host to unique biological communities5. Subglacial water flows down hydraulic gradients determined mostly by the pressure exerted by the overlying ice6. In places, water pools in subglacial lakes, some of which fill and drain7 at times in a coordinated manner8. This variability is important as changing water discharge or its routing can change the movement of the overlying ice streams and glaciers at timescales of days to months2 to decades to centuries9. Eventually, subglacial water can reach the ice-sheet margin, emerging from beneath the grounded ice into the open ocean or a sub-ice-shelf cavity where it stimulates biological productivity10 and amplifies ice-shelf melt11.

When subglacial discharge reaches the ice-sheet–ice-shelf transition (the grounding zone) its conservative temperature (Θ) is close to its in situ freezing point (Θfp = −0.48 °C for freshwater at 650 m depth). Subglacial discharge therefore has little thermal driving (Θ − Θfp ≈ 0 °C) capable of generating ice melt on its own, but it is buoyant compared with the saline water it emerges into12. High-salinity shelf water (HSSW), which is generated by sea-ice formation at the ice-shelf front and circulates within the ice-shelf cavity, is by contrast dense and has a relatively high thermal driving (Θ – Θfp ≈ 0.5 °C at 650 m depth). The buoyancy of subglacial discharge and its ability to entrain HSSW means it can have an outsized effect on ice melt. By mixing with HSSW in the ice-shelf cavity and ascending the ice-shelf base, it can amplify ice melt11,13, in places forming deep channels in the base of ice shelves14,15,16. Channelized ice-shelf melt was first documented in Greenland17 and in the Antarctic can result in sub-ice-shelf channels, visible on the surface and often extending hundreds of kilometres14,15. This concentrated melt is thought to have both positive and negative effects on the ice shelves that buttress the flow of inland ice. Subglacial discharge is likely to disrupt the temperature-based stratification observed in the grounding-zone cavity, but this effect would be limited if discharge were confined to channels18. Modelled subglacial discharge correlates with observed ice-shelf mass loss16, which in turn reduces buttressing. Ice dynamic modelling including subglacial discharge shows that reduced buttressing and associated grounding-zone retreat can cause increased loss of grounded ice19. However, narrow channels have also been shown to reduce overall ice-shelf melt rates by concentrating melt20. Deeply incised channels can also promote fracture due to vertical flexing of the ice shelf21,22, and while ice flow may counter the effects of melt in thick ice, melt may destabilize thin ice shelves23.

Despite its importance, no direct observations of subglacial discharge have previously been made in the interior of Antarctica’s large cold-cavity ice shelves where HSSW contributes to melt12. Observations from beneath Thwaites eastern ice shelf in the Amundsen Sea Sector of West Antarctica recorded the sudden onset of subglacial discharge in mooring records24 while remotely operated vehicle observations documented subglacial discharge proximal to the mooring25. To our knowledge, the only other Antarctic observations of subglacial discharge come from East Antarctica, where subaerial discharge was observed at the ice-sheet margin26, and from Taylor Glacier, where a suspected remnant marine body discharges subaerially at Blood Falls27. Also noteworthy is the absence of subglacial discharge where it might be expected, such as close to grounding zones28,29 and at the downstream end of subglacial drainage paths18.

In this Article, we report observations of subglacial discharge crossing the grounding zone of Kamb Ice Stream (KIS) in West Antarctica (Fig. 1). KIS is one of the five main ice streams that feed ice into the Ross Ice Shelf from West Antarctica and is of particular interest as its rapid ice flow ceased approximately 190 years ago30. Currently, mass gain upstream on KIS offsets approximately a quarter of the mass loss occurring elsewhere in West Antarctica31, making the variability of KIS’s ice flow an important component of the West Antarctic Ice Sheet’s mass balance. Such stagnation and reactivation of ice streams has occurred in the past32 and has been attributed to changing basal thermal regime9, or the quantity and routing of water at the base of the ice33.

Fig. 1: Study site.
figure 1

Subglacial water concentrates beneath KIS before entering the sub-ice-shelf cavity at the study site. Modelled subglacial water routing and flux (plotted in blue on the ice surface). Ice-sheet–ice-shelf transition (the grounding zone) shown by black line35 with black arrow showing general ice-flow direction. Background imagery from ref. 50.

Source data

Full size image

To observe subglacial discharge, we drilled a hot-water borehole through the ice in the austral summer of 2021–2022 at 82.4704° S, 152.2919° W (red triangle, Fig. 1). Our borehole accessed a prominent channel incised into the base of the ice approximately 750 m downstream from where subglacial water first interacts with seawater from the Ross Ice Shelf cavity34 and approximately 5 km upstream of the grounding line35. Melting in this grounding-zone channel has resulted in an upstream migration of the channel’s surface expression of 1.5 km since 198534,36. The feature is geographically continuous with a substantial feature on the ice-shelf surface downstream14,37 (Fig. 1). Borehole access allowed us to image the channel interior and estimate the discharge of subglacial water. We combine the oceanographic observations with a sedimentary record collected from the floor of the channel and new estimates of subglacial routing and downstream ice thickness. Together our findings provide a comprehensive view of subglacial water discharge from beneath KIS and evidence for larger discharge events in recent decades.

Direct observations of subglacial discharge

Beneath approximately 500 m of ice, we used an acoustic sensor to image a 252-m high water-filled grounding-zone channel (Fig. 2a). The channel had an oblate upper section approximately 107 m high overlying a narrower slot with near-vertical side extending 145 m down to the channel floor (Figs. 2b and 3a). Current velocities were measured during six vertical profiles over a 7-hour period, and six profiles of turbidity, temperature and salinity were measured over a 12-day period (Extended Data Fig. 1). These profiles revealed a layer of turbid water between 127 m and 218 m height above the channel floor (Figs. 2b and 3d). It flowed downstream and exhibited high vertical gradients in conservative temperature (Θ) and absolute salinity (SA) (hereafter, we refer to Θ and SA as temperature and salinity). The highest rates of downstream flow reached 5 cm s−1 and coincided with the highest observed turbidity (Fig. 3b,d). Transverse flow velocities in the turbid layer were an order of magnitude lower than the longitudinal velocities (Fig. 3c). A relatively cold (Θ < –2.18 °C) and fresh (SA < 34.57 g kg–1) low-turbidity layer was observed between the top of the channel and the high-turbidity outflow (Fig. 3d–f). This clear and well-mixed layer flowed longitudinally at low velocity both up- and downstream. Beneath the turbid outflow, within the vertical-walled section of the channel, low rates of downstream flow were observed, switching to an upstream flow of relatively warm and saline water in the lowermost 75 m of the channel. Peak upstream velocities reached almost 3 cm s−1 close to the channel floor where peak Θ and SA were observed. Measured velocities in the turbid layer and lower channel were consistent with average velocity measured at fixed depths over a spring-neap tidal cycle (Fig. 3b,c).

Fig. 2: Borehole access.
figure 2

Hot-water drilling through approximately 500 m of ice accessed a water-filled channel approximately 250 m high. a, The borehole location downstream of where subglacial water (blue arrow) first interacts with ocean-cavity water (red arrow). The red parallelogram shows the location and orientation of b. b, Grounding-zone channel cross section (grey dots) imaged using a profiling acoustic sensor. Oceanographic observations, including velocity (blue vectors), turbidity, temperature and salinity, were profiled throughout the water column. Turbidity (grey–blue colour scale), temperature (red–white colour scale) and salinity (white–green colour scale) are normalized and plotted in the downstream (positive x axis) direction for illustrative purposes. Coordinates are metres in a local coordinate system (x, along channel; y, across channel; z, height above channel floor) with a vertical compression of approximately 2/1. Velocity vectors are scaled and plotted in the same coordinate system. The true bearing of the positive x axis is approximately 260°. Observations show estuarine-like flow whereby warmer saltier water flows upstream along the lower channel, and fresher cooler water travels down channel in a turbid plume in the lower half of the circular portion of the upper channel. NTU, nephelometric turbidity units.

Source data

Full size image
Fig. 3: Water-column properties.
figure 3

A plume of turbid water flows downstream in the upper portion of the grounding-zone channel. a, Channel cross section looking downstream determined using borehole altimeter profiling. Grey box denotes region of uncertainty due to lack of altimeter returns. b,c, Longitudinal (b) and transverse (c) mean current (dark blue) and individual cast current profiles overlain with 14-day fixed mooring velocities (mean and standard deviation (variability) shown in red). Currents are positive downstream and to the right when looking downstream. Legend in b shows timing of cross-section and current observations. d, Turbidity for each cast session. Legend shows timing of turbidity, temperature and salinity observations. e, Conservative temperature (Θ) of each cast session. f, Absolute salinity (SA) of each cast session. g, Volume fraction of subglacial discharge water (SGW, solid lines) and glacial meltwater (GMW, dashed lines). h, Temperature–salinity values coloured by the same colour scale as d. Dashed red line in h denotes Gade line38. Heavy dashed black line denotes freezing point at the base of the ice (channel apex), while heavy dashed grey line denotes freezing point at sea surface. Dotted black lines denote contours of potential density anomaly (σ0).

Source data

Full size image

We determined the sources of water in the channel using their distinct temperature and salinity values and a three-member mixing model13 (Extended Data Table 1, Supplementary Tables 1–5 and Extended Data Fig. 2) whereby oceanic source water is assumed to be well mixed with glacial meltwater (derived from ice melt within the grounding-zone channel) and subglacial discharge water (sourced from the subglacial network upstream) (Fig. 3g). In the case of freezing or the generation of glacial meltwater in well-mixed conditions, temperature and salinity evolve along a straight line in Θ–SA space (a Gade line38). Deviation from the Gade-line gradient of 2.4 °C (g kg–1)−1 towards warmer, fresher conditions indicates subglacial discharge (Fig. 3h). Combining the channel cross section and longitudinal flow velocities and assuming uniform cross-channel properties allowed us to estimate subglacial discharge water flux through the channel of 0.9 ± 0.4 m3 s−1 (mean and standard deviation of observations). Glacial meltwater flux through the channel was similarly estimated as 0.3 ± 0.2 m3 s−1.

Sediment record of episodic subglacial discharge

Sedimentation in the channel has preserved a proxy record of subglacial discharge. This record is unlikely to provide a continuous record of drainage through the channel, with the possibility of erosional events, and variable rates of deposition. We recovered a 0.53-m-long sediment core consisting of five distinct units (Fig. 4 and Extended Data Fig. 3). The uppermost unit (0–6 cm) exhibits cm-scale sub-units characterized by light and dark couplets in computed tomography (CT) scans (Fig. 4b) and cycles in grain-size distribution and density (Fig. 4d). The couplets lack sharp transitions, implying continuous deposition. The sand content in this uppermost unit is low, with only a few coarse particles present, indicating deposition did not result from the melt-out of basal debris39. We also note the similarity between this unit and the rhythmite deposits recovered upstream on the adjacent Whillans Ice Stream3. We interpret this upper unit as resulting from subglacial discharge similar to that recorded by our oceanographic profiling and suggest the couplets result from flow variability. At times of higher flow, coarser material is transported and settles out and fines remain suspended, resulting in the darker deposits with lower bulk density. During lower flow periods, the finer material settles, resulting in the lighter-coloured deposits with higher bulk density.

Fig. 4: Sediment record.
figure 4

Sediment core recovered from the base of the channel shows sediment deposition that is dominated by distinct events. a, Visible image of split core. b, Single CT slice with dashed line showing location of density profile shown in d. c, Provenance indicators (neodymium isotope ratio εNd in black, and 87Sr/86Sr in grey) with dashed lines denoting εNd values from Whillans Grounding Zone (εWGZ), Whillans Ice Stream (εWIS), upstream Kamb Ice Stream (εKIS) and J9 borehole (εJ9) (Supplementary Table 6). Note, εNd and 87Sr/86Sr are estimated from single sample measurements. εNd error bars are two standard deviations calculated from the external reproducibility of the JNdi standard, and 87Sr/86Sr error bars, estimated from the internal measurement error, are smaller than the symbol size. d, Sand (light brown), silt (light blue) and clay (dark brown) grain-size percentages overlain with density profile (blue line).

Source data

Full size image

Below 6 cm, two similar yet distinct units (units 2 and 3) are interpreted as deposits resulting from discrete discharge events. These deposits are bounded by abrupt contacts in CT scans (Fig. 4b), density and grain-size distribution (Fig. 4d). The contacts are depositional and show no evidence of erosion. Units 2 and 3 exhibit grain-size distributions that are indicative of higher flow speeds than the top 6 cm of the core with greater sand fractions. These units also exhibit inversely graded density profiles, commonly found in flow deposits with high sediment concentrations40.

Below 25 cm, unit 4 (25-40 cm) ranges from poorly sorted at depth to moderately sorted (Fig. 4). Geological provenance proxies from unit 4 (εNd and 87Sr/86Sr; Fig. 4c) indicate a different source geology when compared with the other units. Unit 4’s provenance is most similar to samples taken from upstream and downstream on KIS and the adjacent Whillans Ice Stream41 (Figs. 4c and 5). The poor sorting and lack of preferred clast orientation in unit 4 indicate it has not settled through a water column and is unlikely to be the direct result of the melt-out of basal debris. Given this, and the sharp basal contact, we suggest this unit resulted from a density flow related to a discharge event40.

Fig. 5: Upstream catchment and downstream impacts.
figure 5

a, Catchment probability from Monte Carlo simulations of subglacial routing indicates the highest-probability catchments include the lower portion of KIS, but lower-probability catchments (<35% of simulations) extend into the upper Kamb and cross into the upper regions of Whillans Ice Stream (WIS). Three documented subglacial lakes (blue dots labelled KT1, KT2, KT3) are present in the lower Kamb trunk catchment, and an additional 13 lakes (blue dots) lie within the 10% probability contour. The subglacial access borehole (red triangle) is located at the downstream end of KIS’s subglacial network. Locations of provenance observations (εNd) from Whillans Ice Stream (εWIS), upstream on KIS (εKIS), Bindschadler Ice Stream (εBIS), Whillans Grounding Zone (εWGZ) and the ice-shelf sites εJ9 and εHWD2 are shown as coloured boxes (Methods and Supplementary Table 2). b, The grounding-zone channel34 is geographically continuous with reduced ice thickness downstream and prominent surface features visible in Moderate Resolution Imaging Spectroradiometer (MODIS) Mosaic of Antarctica (MOA)50. Grounding-zone channel and region of reduced ice thickness shown by red arrows. The portion of the grounding line used to estimate modelled subglacial flux is shown by the orange line. In both a and b, the grounding-zone transition from grounded ice to floating ice shelf is shown by the black line35.

Source data

Full size image

The lowermost unit (5) consists of Miocene-age diatom clasts, many of which are intact, in a diatom-rich matrix, indicating this deposit is not the result of a discharge event. Miocene diatom deposits are found throughout this sector of Antarctica, and their biostratigraphic age is generally not considered to represent their age of deposition4,42. Imagery from the channel floor within 350 m of the borehole (Extended Data Fig. 4) shows decimetre- to metre-scale boulders. Fluvial transport of these boulders would require fluxes one to two orders of magnitude higher than we observe43. It is possible that this lowest unit sampled either an in situ deposit or a boulder of unknown origin.

The likely subglacial discharge origin and geological provenance of units 1–4 indicate subglacial discharge variability. Material transported from different regions of the ice-stream bed has crossed the same location at the grounding zone at different times and been preserved in the sediment record. The depositional mode records the last phase of transport, and we note that the distinct provenance of unit 4 may result from the melt-out of basal debris during a discharge event that was then remobilized and represents the source geology at the site of basal accretion. These findings have implications for the interpretation of ice-sheet proximal sedimentary sequences (for example, ref. 44) as provenance changes may occur even when ice flow remains largely unchanged.

Upstream catchment and downstream impacts

The source of subglacial discharge upstream of the channel can be constrained using probabilistic estimates of subglacial routing45 (Fig. 5a). These hydrological catchments are distinct from the more commonly used ice-flow catchments and indicate the source regions of subglacial water and fluvial sedimentary deposits originating beneath ice sheets. The most likely catchment for subglacial water emerging at our channel site encompasses 22,000 km2 on the lower KIS (90% probability; Fig. 5a). Lower-probability catchments encompass much of the upper KIS (Fig. 5a). Combining these routing and catchment estimates with melt estimates46 allows us to model subglacial water flux across our borehole site. The median modelled discharge is 7.9 m3 s−1 (5.8–28.2 m3 s−1, 25th–75th percentiles; Extended Data Fig. 5), and more than 99% of our catchment estimates result in subglacial discharge greater than the discharge we estimate using oceanographic observations. Part of the mismatch between modelled and observed discharge may be due to a more distributed subglacial network than suggested by the routing45. However, the concentration of flow routing beneath KIS (Fig. 1) and the absence of ice-shelf features indicating melt elsewhere suggest our channel observations are unlikely to be missing much discharge.

The extent of the subglacial catchment also determines the active subglacial lakes that can contribute to the time-varying component of subglacial hydrology. Three documented active subglacial lakes8,36 (KT1–KT3, Extended Data Figs. 6 and 7) are present in the lower KIS catchment, and an additional 13 active subglacial lakes7 lie within the 10% catchment probability contour (Fig. 5a). The active subglacial lakes in the lower KIS catchment have exhibited one fill–drain cycle within the satellite observation period (2003–present) (Extended Data Fig. 7).

Downstream of the grounding-zone channel, notable surface features are apparent in satellite imagery14 (Fig. 5b and Extended Data Fig. 8) and surface elevation models47. In places, these features cut across indicators of past flow (streak lines), indicating that they developed since the stagnation of the ice stream approximately 190 years ago30,37. Ice thickness observations from swath-processed airborne radar show these surface features correlate with regions of thinner ice (Fig. 5b). Both the surface features and the associated regions of thin ice are geographically continuous with the grounding-zone channel (Extended Data Fig. 8). The region’s history is complex. Before stagnation, it was an active shear margin, and after stagnation the grounding zone retreated, most likely in a stepwise manner37. Weakening and thinning within the shear margin would make the ice shelf more vulnerable to subglacial drainage-sourced basal melt48, and stepwise retreat would allow melt features to amplify. Plume theory indicates subglacial discharge events amplify melt but also decrease in impact with distance from their source11. Given the history of this part of the ice shelf, we suggest a progressive development of ice-shelf features in response to episodic subglacial drainage-induced melt exploiting the former shear margin, and grounding-zone retreat increasing the distance over which these impacts occurred.

Episodic discharge and ice-shelf evolution

We directly observed subglacial water emerging from beneath the West Antarctic Ice Sheet and interacting with cavity water beneath the Ross Ice Shelf. The observed subglacial discharge flux of 0.9 ± 0.4 m3 s−1 is smaller than the modelled long-term estimate of 7.9 m3 s−1. Much of this difference can be explained by episodic discharge events, whereby subglacial water is stored and released by the catchment upstream8,36 (Extended Data Figs. 6 and 7). Sediment deposits within the channel interpreted to result from previous discharge events have sampled distinct geological provenances, indicating the subglacial hydrologic system may be capable of spatial variability (Figs. 4 and 5a). Sediment deposits related to subglacial discharge probably all postdate the migration of the channel over this location sometime since 198534, and the most recent large event (Unit 2; Fig. 4) probably resulted from the subglacial lake drainage event observed in the satellite record (Extended Data Fig. 7).

Downstream, the impact of subglacial discharge is manifest in the record of melt left behind in the ice-shelf base and manifest on the surface34 (Fig. 5b and Supplementary Figs. 1 and 2). The subglacial discharge plume we observed was not in contact with the roof of the channel (Fig. 3). With greater discharge flux and associated enhanced buoyancy, mixing and entrainment, the discharge plume would reach the channel apex with greater melt potential11, would influence a larger region of the ice shelf19 and may have contributed to the formation of the existing keyhole-like shape of the grounding-zone channel. The increased turbulence provided by episodic discharge events has the potential to overcome stratification in the ocean-cavity water column that otherwise can reduce the basal melt of ice shelves18,24. The resulting distributed melt would help explain the regional thinning observed49 (Extended Data Fig. 9). Episodically high melt rates would also further reduce the restorative role of internal ice flow23 and result in further mechanical weakening21,22,48.

The implications of subglacial discharge for ice-shelf melt and sea-level projections have begun to be explored11,16,19. Our findings emphasize the importance of episodic subglacial discharge, which is likely to amplify the impact that subglacial water has on ice shelves and ice sheets.

Methods

Borehole access

The borehole was drilled at 82.47048° S, 152.29145° W using hot water, heated on the surface and expelled through a hose ending in a lance tipped with a high-pressure nozzle51. Borehole science was conducted between 28 December 2021 and 13 January 2022. The borehole was initially reamed to a nominal diameter of 0.3 m. A blockage of the borehole on 31 December 2021 led to repositioning of the borehole approximately 10 m downstream to 82.47040° S, 152.29192° W, and borehole science resumed on 3 January 2022. Borehole operations cycled between science operations interspersed by reaming of the borehole, which occupied approximately 6 hours every 24 hours. Borehole access timing and modelled tidal elevation52 are shown in Extended Data Fig. 1.

Channel imaging and velocity profiling

Profiling with a rotating 500 kHz altimeter (Subsea ISA500) allowed us to image the roof, walls and floor of the grounding-zone channel. The altimeter was lowered through the borehole and water column tilted at various angles from the vertical. During profiling, the line was rotated slowly. Individual range estimates have sub-centimetre accuracy. A best-fitting cross section of the channel walls was estimated by spline-fitting a polygon to the range estimates. A region of uncertainty in the lower right-hand corner was included due to a lack of altimeter returns from this region. This uncertainty does not effect our flux and melt estimates due to the low fractions of subglacial discharge water and glacial meltwater in the bottom portion of the channel.

During velocity profiling, a Nortek Aquadop current meter measured water-mass velocity and bearing, and an RBR Duet recorded temperature and pressure. The Aquadop had an averaging interval of 1 s and a sampling interval of 1 s. The current observations in the top 50 m of the channel were affected by weak backscatter and noise from the Subsea altimeter, leading to greater apparent variability in this region. At the end of the survey (13 January 2022), we deployed a mooring with Nortek Aquadop Deep Water Current meters at heights of 9, 162 and 222 m above the channel floor. These sampled at a 20 s averaging interval with a sampling interval of 30 minutes. As with the profiling data, the upper mooring location was affected by weak backscatter and contamination from the altimeter, so was omitted from our mooring–profiling comparison.

Hydrographic profiling

The water-column structure was measured by lowering an RBR Concerto CTD (conductivity, temperature, pressure and turbidity) package. Data processing included correcting the pressure record for atmospheric and tidal effects, removing erroneous observations caused by instrument equilibration and icing, and combining remaining casts in each session into a single mean profile binned at 1 m vertical intervals. Aquadop pressure, current velocity and temperature records were treated similarly but binned at 2 m intervals to account for fewer observations.

Practical salinity and in situ temperature were converted to absolute salinity (SA) and conservative temperature (Θ) following TEOS-10 (Thermodynamic Equation of Seawater 2010)53. Turbidity profiles were calibrated using paired observations from out sensor with that reported previously29. The average turbidity within the top 40 m of the channel for each profile was then subtracted to aid comparison between casts. The average within the top 40 m from all profiles was then added to preserve the absolute value.

Water-mass partitioning and flux estimation

Water masses were partitioned using temperature and salinity and a three-member mixing model following refs. 13,54,55. The source water was assumed to be HSSW and to have the properties of the deepest water body we observe. We note that our HSSW source water (Extended Data Table 1) is approximately 0.05 °C warmer and approximately 0.04 g kg−1 fresher than HSSW observed at the ice front55 and in the ocean cavity approximately 55 km away29. This suggests that the HSSW we observe retains some signature of modified circumpolar deep water. This observation does not affect our analysis, but it is of interest considering the distance from the ice front (approximately 470 km) and the absence of modified circumpolar deep water nearby.

We assumed the HSSW was well mixed under fully turbulent conditions with ice-shelf GMW and SGW. The effective conservative temperature of GMW (ΘGMW) was estimated as –92.5 °C following ref. 13, and the conservative temperature of SGW (ΘSGW = –0.48 °C) was estimated from the melting point of ice at 650 m depth, which is the approximate depth of the point where the subglacial channel meets the grounding-zone channel34.

Assuming the water column was well mixed and consisted of three known water masses (Supplementary Table 1) allowed us to estimate the proportions of each water mass (p1, p2, p3) from the observed values of Θ and SA in the following way:

$$\begin{array}{c}{p}_{1}{\Theta }_{1}+{p}_{2}{\Theta }_{2}+{p}_{3}{\Theta }_{3}=\Theta \\ {p}_{1}{S}_{\rm{{A}}_{1}}+{p}_{2}{S}_{\rm{{A}}_{2}}+{p}_{3}{S}_{\rm{{A}}_{3}}={S}_{{\rm{A}}}\\ {p}_{1}+{p}_{2}+{p}_{3}=1\\ A=\left[\begin{array}{ccc}{\Theta }_{1}&{\Theta }_{2}&{\Theta }_{3}\\ {S}_{\rm{{A}}_{1}}&{S}_{\rm{{A}}_{2}}&{S}_{\rm{{A}}_{3}}\\ 1&1&1\end{array}\right]\\ B=\left[\begin{array}{c}\Theta \\ {S}_{{\rm{A}}}\\ 1\end{array}\right]\\ x=\left[\begin{array}{c}{p}_{1}\\ {p}_{2}\\ {p}_{3}\end{array}\right]\end{array}$$

We estimated water-mass proportions (x = A−1B) from average values of Θ and SA in 1 m vertical bins for each of the six profiles presented in Fig. 3e,f. Θ and SA observations were obtained during six independent casts between 28 December 2021 and 1 January 2022. (See Extended Data Fig. 1 for experiment timing in relation to tidal state.)

To convert these water-mass proportions to flux estimates, we combined the mass-proportion estimates with independent water-column velocity estimates (Fig. 3b,c). Water-column velocity was observed during six casts (three up casts and three down casts) on 9 January 2022. The water-column velocity observations included temperature but not salinity. To allow for variability in the vertical position of the plume, we combined the observations of velocity and temperature, and temperature and salinity, on the basis of temperature. To accomplish this, we divided the water column into four distinct layers: the upper layer, the plume layer, the lower layer and the bottom layer. For layer definitions, see Supplementary Table 2 and Extended Data Fig. 2.

Layer water-mass fractions and velocities were then estimated using area weighting per bin to account for the varying channel width with depth. To account for the correlation between layer areas and mass fractions, the equivalent areas of the water-mass fractions per layer were used to estimate fluxes. The standard deviations of layer properties were used to obtain an estimate of the variability in flux during the experiment. Summary results are presented in Supplementary Tables 3–5. Our methods quantify the variability observed during our observation period and probably underestimate variability outside of our times of observation. Additional unquantified uncertainty comes from our assumption of uniform cross-channel properties implicit in our use of cross-channel area in our flux estimates.

Coring

A 0.53-m-long sediment core was recovered using a gravity corer deployed though the borehole. The corer was weighted and fitted with a sterilized 52 mm (inner diameter) polycarbonate core tube. The corer was fitted with a core catcher and lowered through the borehole and allowed to freefall from approximately 20 m above the seafloor. After recovery, the water from the core headspace was drained, and the core was packed with porous foam. The core was then sealed and stored horizontally at 4 °C. The core was CT scanned at a voxel size of 0.2734, 0.2734, 0.625 mm (x, y, z) using a GE Medical Systems Brightspeed CT scanner. Densities measured in Houndsfeld units were converted to density in kg m−3 following ref. 56. A representative density profile was estimated from the average of the central 10 × 10 voxels in each vertical slice. After CT scanning, the core was split lengthwise with a Geotek core splitter. The split core was then photographed and scanned using a hyper-spectral scanner with a spatial resolution of approximately 0.4 mm.

Grain-size analysis

Sediment samples were collected and treated with 27% hydrogen peroxide (H2O2) to remove organic material by digestion until completion of the reaction. Samples were topped up with deionized water and centrifuged three times before being decanted into glass beakers, and the surfactant Calgon (1 g l–1 Na6O18P6) was added to ensure particles remained disconnected and in suspension. Samples were sonicated in an ultrasonic bath for 20 minutes with stirrers and analysed with the Malvern Laser Sizer 3000 using the aqueous module. Samples with larger grain sizes were sieved at 1,400 μm, the practical upper limit of the Malvern Laser Sizer.

Nd and Sr isotopes

Sediment samples were prepared for Nd and Sr isotope measurements as previously described44,57. The <63 μm fraction was leached to remove authigenic coatings, and the Nd and Sr were isolated using established ion exchange chromatography methods. Neodymium isotopes were measured in the MAGIC laboratories at Imperial College London on a Nu high-resolution multi-collector inductively coupled plasma mass spectrometer. To account for instrumental mass bias, isotope ratios were corrected using an exponential law and a 146Nd/144Nd ratio of 0.7219. Interference of 144Sm on 144Nd, although negligible, was corrected for. To correct measured 143Nd/144Nd ratios to the commonly used JNdi-1 value of 0.512115 (ref. 58), bracketing standards were used. USGS (US Geological Survey) BCR-2 rock standard measurements yielded 143Nd/144Nd ratios of 0.512639 ± 0.000002 (n = 3), in very close agreement with the published ratio of 0.512638 ± 0.000015 (ref. 59). A full procedural blank was 25 pg of Nd. Final 143Nd/144Nd ratios are expressed using varepsilon notation (εNd), which denotes the deviation of a measured ratio from the modern chondritic uniform reservoir (0.512638) in parts per 10,000 (ref. 60).

Strontium isotopes were measured in the MAGIC laboratories at Imperial College London on a Triton thermal ionization mass spectrometer. Samples were loaded in 1 μl of 6 M HCl onto degassed tungsten filaments with 1 μl of TaCl5 activator. The measured 87Sr/86Sr ratios were corrected for instrumental mass bias using an exponential law and an 88Sr/86Sr ratio of 8.375. Interference of 87Rb was corrected for using an 87Rb/85Rb ratio of 0.386. Samples were corrected to the published NIST (National Institute of Standards and Technology) 987 standard reference material value of 0.710252 ± 0.000013 (ref. 59). These were completed every four unknowns, with a mean of 0.710246 ± 0.000013 (2 s.d., n = 19). Accuracy of results was confirmed using rock standard USGS BCR-2, processed with every batch of samples, which yielded 87Sr/86Sr ratios of 0.705012 ± 0.00017 (2 s.d., n = 10). This is in agreement with the published ratio of 0.705013 ± 0.00010 (ref. 59). εNd values plotted in Fig. 4a are shown in Supplementary Table 6.

Diatom age determination

The lowermost sedimentary unit (Unit 5) was determined to be Miocene in age on the basis of the diatom assemblage reported previously61. Diagnostic taxa placed the unit in the Thalassiosira praefraga Range Zone with overlap between T. nansenii and T. praefraga suggesting the unit was sourced from deposits of 18.7–18.0 Ma age42,61,62.

Subglacial routing and catchment modelling

Subglacial routing and catchments were determined using hydropotential gradients6 and followed the stochastic D8 method described previously45. This allowed probability to be quantified using a Monte Carlo ensemble with 1,000 runs that sampled Gaussian random fields to make realizations of surface elevation, bed elevation, flotation fraction (the ratio of subglacial water pressure to ice overburden pressure) and subglacial melt input. Surface elevation and bed elevation were obtained from REMA (200 m spatial resolution)47 and BedMachine (500 m spatial resolution)63, respectively. An average flotation fraction of 1.0 was assumed. The subglacial melt input is from ref. 46. The Gaussian random fields used correlation lengths of 10 km for the bed elevation, surface elevation and flotation fraction and a correlation length of 20 km for the subglacial melt. The amplitude of the bed elevation’s Gaussian random field, which is expressed as a standard deviation (σ), uses the spatially varying error field provided in the BedMachine dataset (average value of 93 m within the KIS catchment); similarly, the σ for the subglacial melt uses the standard deviation field provided in that dataset. The surface elevation σ was set to 0.6 m (ref. 47) plus 10% of the firn thickness63, and the flotation fraction σ was set to 0.03. Subglacial catchments were estimated by determining the flow paths upstream of an approximately 18-km-long portion of the KIS grounding zone (Fig. 4a). Subglacial flux was estimated similarly as the flux crossing that stretch of the grounding line (Fig. 4b and Extended Data Fig. 5). A mismatch between modelled subglacial routing and the actual channel location of approximately 3.5 km was observed. This mismatch was determined to be a model artefact due to the absence of any other grounding-zone channel away from the main channel as determined by oversnow radio echo sounding34.

Subglacial lake activity

We estimated subglacial lake activity using CryoSat-2 and ICESat-2 radar and laser altimetry data. We grid-phase unwrapped Synthetic Aperture Radar-Interferometric mode CryoSat-2 radar altimetry data (following ref. 2) for the period July 2010 to the start of ICESat-2 era (October 2018) mimicking ICESat-2’s ATL15 data product’s spatial structure and time periods (propagating time periods back at 3-month intervals to the start of the CryoSat-2 mission). During the ICESat-2 era, we used a higher-level ICESat-2 data product, ATL15 Gridded Antarctic Land Ice Height Change r003, that provides gridded estimates of ice-sheet elevation change across Antarctica at a spatial resolution of 1 km and a temporal resolution of 91 days from during the ICESat-2 era64 (October 2018 to April 2023). We used the standard deviation across a total of 53 time steps of gridded ice-surface height-change rates estimated over 3-month periods to highlight areas of inferred subglacial lake activity (Extended Data Fig. 6). According to the gridded standard deviation estimates for each mission, subglacial lakes KT1, KT2 and KT3 were active between 2 July 2010 and 30 September 2018 (Extended Data Fig. 6a) and inactive between 1 October 2018 and 2 April 2023 (Extended Data Fig. 6b).

We estimated volume change time series (following ref. 8) by calculating the mean change in surface elevation with time (dh/dt) within each published subglacial lake outline at quarterly (3-month) intervals (Extended Data Fig. 7). We corrected this height-change time series for regional secular height change by subtracting the mean height change in an area surrounding the lake extending 10 km from the lake outline for each time step (following ref. 65). We then integrated the dh/dt time series through time and multiplied the corrected time series by the area of each outline to estimate surface ice-volume displacement. This approach assumes a 1/1 ratio of ice volume to water volume displacement (for example, refs. 66,67,68); although this approach is commonly implemented, the 1/1 assumption may be an oversimplification in slow-flowing, highly viscous ice settings, such as the KIS trunk69.

Swath radar imaging

Swath radar data were collected between 20 December 2013 and 27 December 2013 using the Center for Remote Sensing and Integrated Systems eight-element array of the Multichannel Coherent Radar Depth Sounder radar system70. This swath imaging radar was installed on a Basler air frame that flew at approximately 1.8 km above the ice surface. Processing followed refs. 71,72 and involved digitizing the basal return in cross-track images. This resulted in swath widths of approximately 1 km in our study area, with nominal along-track sampling every 15 m and across-track sampling every 70 m (Fig. 4b).