Sub-ice-shelf bathymetry of the Eastern Ice Shelf and other ice shelves in the Amundsen Sea was modeled through inversion of airborne gravity data by . While gravity data can be collected efficiently on a regional scale, they have relatively low spatial resolution (5 km), so only topographic features with wavelengths down to about 5 km are well resolved . With a width of about 15 km, the general pattern of Thwaites pinning point is therefore resolved reasonably well, but absolute values remain uncertain (100 m standard deviation). Where ice is grounded on the seafloor, however, ice thickness measurements from airborne radar surveys can be used to further constrain the gravity-derived bathymetry. We compare two data products to the airborne radar data, the 1000 m resolution bathymetry from and the 500 m BedMachine v2 bed-elevation product from , which was calculated from the mass conservation method where ice is grounded and gravity inversion where ice is floating .

The airborne radar data we used were collected using the Multichannel Coherent Radar Depth Sounder MCoRDS; in 2009 and a 600–900 MHz accumulation radar in 2019 Fig. a. The ice base was picked from a series of synthetic aperture radar-processed radargrams in a semi-automated fashion. Where ice is grounded on the seafloor, a change in the strength of the basal radar reflection occurs . From the surveys on Thwaites Glacier, only two transects cross the edges of the pinning point in 2009, and only one crosses its central part in 2019. Figure  shows two airborne radar transects that reveal a heavily crevassed ice base, suggesting that the Eastern Ice Shelf is far from reaching an equilibrium with the underlying ocean system. If basal melting underneath the floating ice were homogeneous, the ice geometry would be expected to be more uniform .

We compare ice thickness from airborne radar measurements over the pinning point to two bathymetry data sets: gravity-derived bathymetry for the Thwaites ice shelves of and BedMachine Antarctica Version 2 . Directly subtracting these bathymetry estimates from the height of the ice base across the pinning point results in average differences of 200±49 and -150±82 m, respectively. We attribute these differences to the original gravity inversions not being constrained with the radar ice thickness at this isolated location. We therefore fit a quadratic plane to these differences and vertically adjust the bathymetry where ice is grounded according to the 2011 grounding line. As a result of this adjustment, the mismatches between estimated and observed bathymetry are reduced to 0±13 and -74±121 m, respectively (Fig. S2). Because of the large residual error in the BedMachine bathymetry for the eastern portion of the pinning point, we only use the adjusted bathymetry map that is based on for further calculations outlined below.

Ocean tides as well as the elastic deformation of the Earth’s crust underneath the moving water masses (referred to as load tides) influence ice-shelf surface elevation. The predictions of two tide models were validated with available GPS records from the Eastern Ice Shelf: the regional barotropic Circum-Antarctic Tidal Solution (CATS2008) model developed by and the fully global barotropic assimilation model (TPXO9) from Oregon State University developed by . Tidal loading is accounted for in TPXO9 and was added to the CATS2008 model predictions. In addition to the tidal oscillation underneath the floating ice, a 1 hPa increase of barometric pressure on the ice surface causes an isostatic -1 cm response of the ice shelf . To correct for this inverse barometric effect, we used barometric pressure measured by an automatic weather station (AWS) on Thurston Island (approx. 500 km away across the Amundsen Sea). We validate this record with separate barometric pressure measured on the Eastern Ice Shelf over a 176 d period and find a very good correlation between the measured anomalies (R2=0.94). We note that the Thurston Island AWS also correlates well (R2=0.86) with a much closer AWS located on Bear Island (approx. 200 km away) between 2012 to 2019, before the latter stopped reporting data. The inclusion of the inverse barometric effect does not improve the fit to the GPS data (Fig. S3), suggesting that storm surges and local wind-forced effects on small-basin geometry could be significant (Laurie Padman, personal communication, 2021). The resulting tide correction generally underestimates the vertical displacement measured at the Eastern Ice Shelf within ±17 cm, which is relatively poor compared to the Antarctic-wide accuracy of tide models in coastal areas of about ±10 cm . The inaccuracy of the tidal correction translates to about ±3 cm yr-1 in terms of the derived surface-lowering rates between 2014 and 2020. We attribute this inaccuracy to inaccurate tidal constituents in both tide models, errors in the bathymetry used by the tide models, grounding-line location, and inaccurate ice-shelf thickness, as well as to insufficient knowledge of the ice-water drag coefficient .

Along a narrow band oceanward of the grounding line, ice is not freely floating because longitudinal stresses within the ice limit ice-shelf flexure. We numerically model the vertical displacement due to this tidal flexure with the well-known elastic approximation as formulated by : 1kw+∇2D∇2w=q,2D=EH312(1-λ2),3q=ρswgA-w, where w is elastic vertical deflection, ∇2 the 2D Laplace operator, and k a spring constant of the foundation, which is zero for the floating part. D is vertically integrated ice-shelf stiffness while undergoing tidal bending p. 443, with E=1.5 GPa the effective Young’s modulus, H the ice thickness given by BedMachine, and λ Poisson’s ratio of a Maxwell model accounting for transverse deformation due to longitudinal strain. The tidal force underneath the floating ice, q, is given by the tidal amplitude A=1 m, the density of ocean water ρsw, and gravitational acceleration g. Symbols used in this study are summarized in Table A1. For boundary conditions, we implement a fulcrum at the grounding line (w=0) and anchor the upstream boundaries of the model domain of the grounded portion rigidly (w=0, ∇2w=0). Generally, the elastic flexure pattern is time-invariant and only dependent on ice-shelf stiffness . We therefore normalize the finite-element solution for the deflection, w, with the applied tidal forcing to derive tide-deflection ratio throughout the area α map,Fig. S4. By extracting the α for a given location, tide model output from freely floating parts of the ice shelf can thus be directly scaled to include the effects of 2D elastic tidal flexure near grounding lines . The computationally expensive task of running the finite-element model over time is thus circumvented. By using the tide-deflection ratio, inaccuracies in the α map due to a possibly incorrect ice-shelf stiffness are always smaller than inaccuracies in the utilized tidal forcing. We estimate these within the reported uncertainty of the effective Young’s modulus from tiltmeter data elsewhere ±0.69 GPa; to a maximum of ±0.08 m. In areas where ice is grounded, α values close to or equal to zero prevent the application of any tidal correction of GPS and ICESat-2 measurements.

We use height above flotation as a proxy to assess changes in the degree of grounding. Ice-surface elevations are given by the continent-wide and time-stamped REMA data set . REMA data in our study area were mostly acquired in 2014 with feathered strip edges in the vicinity of our region of interest where we expect the largest inaccuracies. To detect ice-surface elevation change since the acquisition of REMA, we use the ICESat-2 L3A Land Ice Height, Version 3, data set , which gives absolute ice-surface elevations with under 3 cm vertical and 9 cm horizontal accuracies . We removed about 16 % of the ICESat-2 points with the provided quality summary flag. ICESat-2 data in our study region were acquired in 2019/20 and allow us to calculate changes in ice-surface elevation since 2014 compared to REMA.

We first translate ice-shelf surface elevations from both REMA and ICESat-2 data to freeboard using the EIGEN6c4 geoid model as the mean sea level. We neglect the effect of mean dynamic topography as it is not directly measurable on the ice-shelf surface and invert freeboard, zf, to floating ice thickness, Hf, based on hydrostatic equilibrium principles. Here, we take advantage of the two available airborne radar transects and use all traces acquired on the freely floating part of the Eastern Ice Shelf to derive a single mean ice-column density (ρ¯) for this inversion from each of the radargrams (893±21 kg m-3 for 2009 and 871±10 kg m-3 for 2019/20, respectively). We attribute the decline in mean ice-column density of the Eastern Ice Shelf to the interplay of a prolonged erosion of dense ice at the ice-shelf base where ρice=917 kg m-3, with an increase of relatively soft snow accumulation over the last decade on the surface where ρfirn=430±66 kg m-3 from snow-pit measurements. Modeled surface mass balance further supports our observation of a decline in mean ice-column density . This trend is independently confirmed by comparing REMA surface elevations and BedMachine ice thickness data from 2014 to field measurements on 38 sites using GPS and phase-sensitive radar in the 2019/20 season (883±7 and 865±12 kg m-3, respectively). On grounded areas including the pinning point, we then directly subtract the adjusted seafloor depths from REMA and ICESat-2 surface elevations to calculate absolute ice thickness, Ha, in 2014 and 2019/20, respectively. Lastly, the difference between absolute and our hydrostatic ice thickness estimate is calculated and expressed as height above flotation, zf: 4zf=(Hf-Ha)ρsw-ρ¯ρsw. We then use the transition between grounded areas that always feature a positive height above flotation to floating areas (zf=0 m a.f.), to manually delineate the location of the grounding line in 2014 and in 2019/20. These estimates did not prove sensitive to the reduced value of mean ice-column density beyond the inaccuracies arising from manually picking the grounding line. We estimate this uncertainty to about ±0.1 km yr-1 in terms of grounding-line retreat rates. We calculate retreat rates from comparison with the 2004 grounding line from , which uses 1999–2003 Landsat and 2003–2009 ICESat-1 data, the InSAR-derived 2017 grounding lines from , and, where these are unavailable, such as at the pinning point, with the 2011 grounding line from .

A Lagrangian framework tracks ice parcels with time and thus corrects for the effects of ice-flow advection during change calculation. This is necessary as the uneven ice base of the Eastern Ice Shelf (Fig. ) combined with relatively fast ice flow induces an advection signal that may spuriously be misinterpreted as rapid surface-height change if left uncorrected for. We use the annual velocity field composite from Landsat-8 feature tracking between 2013 and 2020 and migrate the ICESat-2 point cloud back in time to where each point would have been when REMA data were acquired. We therefore capture ice-dynamical changes such as an observed counterclockwise rotation of the ice flow and the associated change in ice-flow divergence in the migration. No seasonality of flow direction nor any inter-annual change in speed was detected in the velocity record. We estimate accuracy by validating our velocity record with GPS measurements at 38 sites to 28 m yr-1, or 4.5 % of the average flow speed.

We then use the α map in combination with tide modeling to correct for elastic tidal flexure as ICESat-2 points acquired at different times migrate along streamlines across the grounding zone. As a result, rates of surface lowering are valid on both grounded and floating ice. We then rasterized the migrated ICESat-2 point cloud using a 2D Gaussian kernel to obtain an approximately 68 m resolution map of surface-lowering rates since 2014, when REMA data were acquired. Overall accuracy of these rates is estimated to be the squared sums of errors from REMA, ICESat-2, the applied tidal correction, and the accuracy of our velocity record, divided by the 5–6 years of time difference to ICESat-2 data acquisition. This conservative estimate results in an accuracy of the derived surface-lowering rates of ±1.25 m yr-1 at a maximum, which is well below the detected change signal. We note that an area-wide uncertainty of about 6 m in REMA is the largest contribution to this accuracy estimate and originates mainly from a narrow area where REMA strip edges have been feathered to form a seamless mosaic. This narrow area lies largely outside of our region of interest (Fig. b), which increases our confidence in accuracy within the boundaries of individual REMA strips.

We use a widely used ice-sheet model to investigate causes of the observed changes in ice dynamics. If the recent and present changes can be replicated well, numerical modeling can be used to provide insights into likely scenarios of future change and sensitivity to various forcings. The Ice-Sheet and Sea-level System Model ISSM; is a thermomechanically coupled model capable of calculating higher order stress components of ice dynamics that relies on the classical conservation laws of physics such as the balance of stresses and the incompressibility of ice. Deforming ice is represented in the model equations with a purely viscous, nonlinear rheology following the Nye–Glen isotropic relation : 5σij=12Bϵ˙e1n-1ϵ˙ij, where σij is the deviatoric stress tensor and ϵ˙e the effective strain rate of deformation defined as the second invariant of the strain rate tensor, ϵ˙ij. Our model simulations employ a depth-averaged rheological parameter, B, with a creep exponent of n=3. Rheology is dependent on ice temperature and crystal fabric, which are both not explicitly implemented in the model, so fields for B are inferred from surface velocity data. Higher values of B indicate colder/stiffer ice, whereas lower values indicate warmer/damaged ice. Where ice is grounded, friction at the ice base exerts a drag force, τb, opposing the ice flow that is directly proportional to the basal velocity, vb . We use a linear, Budd-type friction law : 6τb=-β2Nvb, where N is the effective water pressure at the ice base given by N=g(ρiceH+ρswzb), which becomes zero for floating ice to eliminate vertical gradients of horizontal velocities within ice shelves. This formulation assumes a perfect hydrological connection between the ocean and the grounded ice base and neglects dynamic subglacial water flow, such as lake drainage or basal discharge . The basal friction coefficient, β2, cannot be measured directly. On grounded ice, positive values of β2 enable horizontal shear within the ice column, which can be used to infer properties of the bed itself by comparing observed and modeled velocity fields on the ice surface. Here we use the Shallow Shelf (or Shelfy Stream) Approximation SSA; to infer both B and β2 in a series of inverse steps by minimizing the mismatch between observed and modeled velocity fields (Appendix A and Fig. S5). In the SSA, horizontal velocity does not vary with depth in the ice column, and vertical gradients in shear stress are hence neglected. As a consequence, the derived fields of B are vertically integrated, and β2 is directly derived from the surface velocity. We then solve the stress balance using the SSA in a series of diagnostic model runs. The finite-element mesh is optimized using static anisotropic adaptation techniques to provide fine resolution near grounding lines and narrow shear zones as guided by surface velocity fields during model initialization . We use a Bidimensional Anisotropic Mesh Generator to create a final mesh consisting of 44 168 unstructured elements with a minimum edge length of 100 m and a maximum edge length of 450 m (Fig. S6).

Two different velocity data sets are compared in our modeling work to invert for B over the entire domain and β2 over grounded ice. To compute an unperturbed reference state, we use the Making Earth System Data Records for Use in Research Environments product MEaSUREs;, which originates from interferometric synthetic aperture radar data acquired between 2007 and 2009 when the Western Glacier Tongue was largely intact. We compare MEaSUREs with a recent velocity field in 2019, which we derived from Landsat-8 imagery collected after the breakup of the Western Glacier Tongue. Bed elevation and ice thickness are prescribed by our adjusted bathymetry and the REMA surface elevation. We outline the model domain guided by the drainage basins feeding into the floating parts of Thwaites Glacier and the location of the ice front. For the perturbation experiments, we first delineate the domain using visible satellite imagery before and after breakup of the Western Glacier Tongue and later modify the grounded mask at the pinning point as guided by our analysis of height above flotation. We impose a Dirichlet boundary condition (v=vobs) along the upstream limit of the domain using observed velocity and a Neumann stress boundary condition (σ⋅n=-pswn) along the ice-shelf front prescribed by the hydrostatic pressure psw=-ρswgz of the water column (Fig. S6). The ice surface is assumed stress-free. We note that the diagnostic model does not solve for the temporal evolution of the ice geometry as a response to the applied environmental perturbations but proves valuable to identify the main drivers behind the observed ice-dynamical changes between the two velocity fields by solving the instantaneous stress balance.

To investigate the effects of the breakup of the Western Glacier Tongue and a vanishing pinning point on the regional ice-flow configuration, we perform a series of perturbation experiments (Table 1) to detect the underlying mechanisms behind the observed changes in ice dynamics: (i) counterclockwise rotation of the ice flow since breakup of the Western Glacier Tongue and (ii) ice funneling through the two remaining portions of the pinning point. To address (i), we run the model with and without the full extent of the Western Glacier Tongue (Fig. ). Then, (ii) is addressed by reducing the spatial extent of the pinning point as guided by our analysis of height above flotation and comparison with model solutions when the pinning point has its full 2011 extent. Finally, we entirely unpin the Eastern Ice Shelf from the pinning point and recalculate the stress balance to model the regional response of the ice-shelf system on the predicted ungrounding. All other ISSM simulations use the 2011 location of the grounding line.

Integrating data from airborne radar surveys with the REMA digital elevation model and recent ICESat-2 measurements reveals temporal changes in height above flotation from 2014 to 2020. Grounding-line retreat of Thwaites Glacier is known to be heterogeneous, with retreat rates averaging to about 0.8 and peaking up to 1.2 km yr-1 between 2011 and 2017 along preferential channels in the bathymetry since the last assessment . Our results show continued grounding-line retreat along a 45 km stretch of Thwaites Glacier’s grounding line, with sporadically accelerating retreat rates.

While a number of new ice rises emerged during recent deglaciation on bathymetric bumps, several new cavities are forming or have already opened up to the intrusion of seawater. To highlight this heterogeneity we present 10 areas of interest that show different retreat patterns (Fig. c and d): (I) Retreat on the eastern wing of what is referred to as the butterfly region in the literature has slowed down from 1.2 km yr-1 between 2011 and 2017 to about 0.6 km yr-1 between 2017 and 2020. The maximum inland reach of its western wing remained stable since 2017 as this cavity is enclosed by a well-grounded bathymetric ridge that is currently preventing further retreat. Height above flotation indicates that the majority of the butterfly region was already floating or at least close to flotation in 2014, suggesting that ungrounding likely occurred between 2011 and 2014 with retreat rates up to 1.9 km yr-1. (II) To the west of the butterfly region, a newly formed cavity is rapidly expanding, with retreat rates up to 1.5 km yr-1 since 2017. Further retreat along this narrow bathymetric low would cut off a newly formed ice rise. However this is unlikely in the near future as the remaining 2.5 km is well grounded with zf>15 m. (III) Grounding-line retreat rates within this narrow embayment steadily decelerated since 2011, which is in conjunction with the bed elevation rising in the inland direction as observed in the bathymetry data (Fig. c) and matches the 0.3 km yr-1 between 2011 and 2017 from . (IV) A large grounded protrusion at this location has been retreating constantly at an accelerating rate of 0.4–0.6 km yr-1 within the central part of the grounding line, and it has now formed three well-grounded ice rises with zf>18 m. (V) Here, rapid retreat of 1 km yr-1 between 2011 and 2014 slowed down to about 0.1 km yr-1 due to a bathymetric bump. (VI) Similar to area II, a new ice rise will likely emerge within only a few years as the remaining 1.6 km between the current grounding line and an inland bathymetric depression ungrounds at a rate of 0.5 km yr-1. (VII) Grounding-line retreat of up to 0.5 km yr-1 is observed compared to the 2017 grounding line along a cavity that stretches into the main trunk of Thwaites Glacier. However, our analysis of height above flotation reveals that much of this cavity was already close to flotation in 2014, which would change this rate to 0.8 km yr-1 between 2014 and 2020. Further inland, we detect an isolated area that is already at its level of flotation but is currently shielded from warm water intrusion by a seafloor saddle. (VIII) Here, the grounding line remained almost stable since 2011 due to the locally elevated seafloor that is separating the eastern from the western branch of Thwaites Glacier, feeding into the former Western Glacier Tongue. (IX) Grounding-line retreat accelerated up to 1.6 km yr-1 along this deep seafloor trough where Thwaites Glacier flows the fastest. (X) Further offshore, where the Eastern Ice Shelf is confined by the pinning point, the two remaining eastern and western portions are separating at an accelerated rate of 0.3 km yr-1 between 2014 and 2020, compared to 0.2 km yr-1 between 2011 and 2014. This opening now allows the ice to funnel over a bathymetric saddle in between the portions of the pinning point. The temporal evolution of the observed grounding-line retreat rates is summarized in Fig. d.

Here we summarize the relevant processes that directly impact stability of the pinning point. An in-depth study on the history of dynamical changes at both the Eastern Ice Shelf and Western Glacier Tongue during the satellite measurement era is published in .

We use the two velocity fields and detect significant changes in ice-flow divergence over the last decade (Fig. ). Before the breakup of the Western Glacier Tongue, ice flowed mainly through the calving front to the east. We interpret this as evidence of significant buttressing of the Eastern Ice Shelf through the pinning point. In the years following the breakup of the Western Glacier Tongue in 2009, (i) the ice flow on the main floating part of the Eastern Ice Shelf has rotated counterclockwise, (ii) streamlines are now funneling through a narrow gap between the eastern and western portions of the pinning point (white streamlines in Figs. b and b), and (iii) the shear zone upstream of the pinning point weakened and migrated up to 10 km upstream. hypothesize that this counterclockwise rotation is attributed to the breakup of the Western Glacier Tongue, following years of increased lateral drag.

Along the coast, flow across the eastern half of Thwaites Glacier’s grounding line has remained largely unchanged. This indicates the sustained ice-shelf buttressing of this area by the Eastern Ice Shelf. The western half, in turn, exhibits increased absolute values of flow divergence, which is either a result of accelerating ice discharge across the grounding line as the grounded ice is still adjusting to the removal of the Western Glacier Tongue (Figs.  and ) or triggered by reduced ice-shelf buttressing and subsequent new ice divergence because of shrinking of the pinning point (Fig. ).

We validate these remote-sensing observations with ground-truth data from a 12 d GPS repeat-survey collected in the 2019/20 season (Fig. ). GPS data were processed using the Precise Point Positioning kinematic solutions from the Canadian Geodetic Survey (https://webapp.geod.nrcan.gc.ca/geod/tools-outils/ppp.php, last access: 15 June 2020). Errors were estimated from the standard deviation of the solutions. At 38 sites distributed on the freely floating part of the Eastern Ice Shelf, our GPS measurements support the counterclockwise rotation of the ice flow observed with satellite data from 89±8 in 2009 to 101±6∘ in 2019/20 from east with an average flow direction of 99±6∘ in the GPS observation. Simultaneously, average flow velocities at these sites decreased from 661±79 in 2009 to 597±23 m yr-1 in 2019, which fits well with 625±28 m yr-1 of the GPS observation. The spatial distribution of these changes shows that the largest counterclockwise rotation of ice flow occurred just a few kilometers upstream of the pinning point, where the ice flow had accelerated since 2009 (Fig. a). Because our velocity field derived from Landsat-8 feature tracking captures the measured velocity field at these 38 sites within the errors, we trust the validity of satellite-derived streamlines in areas beyond the field survey, such as near the pinning point and the grounding line where a series of rifts and crevasses prevented the collection of further ground-truth data.

The bathymetry product of shows that the elevation of the seafloor ridge on which the pinning point is located is -498±72 m. Airborne radar data from 2009 and 2019 reveal that underneath the eastern portion of the pinning point, the ridge is several hundred meters higher at -348±74 m compared to the BedMap2 data set from . The average height above flotation of the eastern pinning point is 7.6±6.3 m in 2019/20, indicating that only modest ice-column thinning is required to detach from the seafloor ridge underneath (Fig. ).

Rates of surface-height change also reveal a thickening signal upstream of the smaller western portion of the pinning point of up to 5 m yr-1 (area of interest XI in Fig. c). The additional basal shear stress from localized grounding induces compressional stresses, and thus thickening, upstream. A similar thickening signal, although of smaller magnitude (3 m yr-1), is observed upstream of the eastern edge of the pinning point. Validation of the lowering rates from satellite data with kinematic GPS data acquired within the area of safe field operations in the 2019/20 field season shows a similar pattern as observed with ICESat-2 (Fig. b). Differences between the derived lowering rates are -0.28±0.66 m yr-1 with ICESat-2, slightly underestimating the GPS measurements. We attribute this small mismatch to moving wind-sculpted snow features (sastrugi) on the ice-shelf surface. We also identify a series of alternating thickening and thinning signals oblique to ice flow, which we interpret as the ice shelf’s response to heterogeneous basal melt patterns and subsequent small-scale hydrostatic adjustment.

With the derived information about stability and rate of destabilization of the pinning point, it is possible to estimate when the Eastern Ice Shelf will likely unpin entirely from the seafloor ridge. Average height above flotation on both remaining portions of the pinning point is about 8.0±6.5 m (Fig. b). Assuming that recent surface-lowering rates between 2014 and 2019/20 of -0.3±1.5 m yr-1 persist into the future, unpinning will happen within the next 20 to 30 years. There is, however, the potential to detach within the next decade if the true height above flotation is on the lower end and the true surface-lowering rates are on the higher end of the reported range. Since ice-shelf thinning involves a multitude of highly nonlinear processes that speed up the ungrounding, we hypothesize that entire unpinning within the next decade is very likely.

We compare two model configurations with the aim to evaluate the net effect of the weakening shear margin on the influence of the Western Glacier Tongue. For example, rheologically weaker ice in the shear margin will progressively decouple the Eastern Ice Shelf from the fast-flowing Western Glacier Tongue. The derived fields of the rheological parameters capture the weakening of the shear margin between 2009 and 2019 (Fig. ), which further supports our hypothesis of an ongoing reconfiguration of regional ice dynamics.

The basal friction coefficients as derived from the two velocity fields show relatively low to moderate values around 10–100 (Pa yr m-1)1/2 for the majority of the grounded areas (Fig. A2). This is in general agreement with full Stokesinverse modeling efforts from , with values between 0–50 (Pa yr m-1)1/2. Low basal friction coefficients indicate that sliding over the ice–bed interface is more likely than a no-slip type of internal deformation, when ice is frozen to the bed. Lubricated subglacial conditions, in turn, increase the radar basal reflectivity, which led to the conclusion that much of Thwaites Glacier’s bed is thawed . Our inferred fields of the basal friction coefficients support this, with grounded parts underneath Thwaites Glacier's main trunk showing large patches of slippery subglacial conditions indicated by very low basal friction coefficients (Fig. A2). At the pinning point and isolated bathymetric highs along the 2011 grounding line, we required basal friction coefficients to significantly increase to up to 1000 (Pa yr m-1)1/2 to match the satellite observed velocity fields. These high values were necessary to slow down the modeled ice flow. In reality, localized straining and fracturing in this area further lowers the effective ice viscosity, which reduces longitudinal stresses on the pinning point and thus inferred basal drag.

The control run that consists of the MEaSUREs velocity field, a model domain that includes the Western Glacier Tongue, and the full extent of the 2011 pinning point shows that the inversion is successful in finding a basal friction coefficient field that reproduces the observed velocity field (Fig. ). Velocity misfits are -2±100 m yr-1 over the entire domain, 1±43 on the Eastern Ice Shelf, and -1±100 m yr-1 on the Western Glacier Tongue (Fig. S5a). We then remove the Western Glacier Tongue from the computational domain and solve the stress balance again for the modeled velocity field, keeping the basal friction coefficients from the control run. The modeled streamlines are almost perpendicular to the orientation of the pinning point and represent a counterclockwise rotation not observed in the MEaSUREs velocity field (Fig. S7b). This experiment suggests that the Western Glacier Tongue was still affecting ice dynamics on the Eastern Ice Shelf when data for the MEaSUREs product were acquired. However, relative to the field-acquired GPS vectors (Fig. a), the modeled streamlines slightly overestimate the influence of the Western Glacier Tongue.

To further test the hypothesis of whether the counterclockwise rotation of ice flow is a consequence of the breakup of the Western Glacier Tongue, we repeat the perturbation of removing the Western Glacier Tongue from the model domain but replace the MEaSUREs velocity field from 2009 with our Landsat-8 record from 2019. The modeled streamlines of the second control run match the satellite-derived streamlines near the pinning point but slightly overestimate the counterclockwise rotation of ice flow on the Eastern Ice Shelf (Fig. S7c). Velocity misfits are -10±282 m yr-1 over the entire domain, -2±100 on the Eastern Ice Shelf and -10±282 m yr-1 on the Western Glacier Tongue (Fig. S5b). This mismatch is a consequence of including the Western Glacier Tongue in the model domain, although ice rheology in the shear zone weakened (Fig. ). Removing the Western Glacier Tongue entirely from the model domain results in modeled streamlines that also overestimate the counterclockwise rotation of ice flow (Fig. b), which confirms the response of the first perturbation experiment. Both simulations suggest that although lateral drag across the shear margin is reduced, the Western Glacier Tongue still affects ice flow on the Eastern Ice Shelf, particularly in areas closer to the grounding line, where closely spaced icebergs to the west still transmit lateral shear stresses.

Finally, we investigate the effect of continued weakening of the pinning point on the regional ice flow pattern (simulations e and f) by reducing the spatial extent of the pinning point as guided by our analysis of height above flotation from ICESat-2. This requires the grounded mask in ISSM to be modified by designating model elements between the eastern and western portion of the pinning point as floating. ISSM then solves the stress balance again diagnostically that is directly compared to the simulation based on the full 2011 extent of the pinning point (simulation c). Differences between the modeled stress balances show increased ice funneling across a narrow bathymetric saddle between the eastern and western portion of the pinning point (Fig. a) that is also evident in visible satellite imagery (Fig. S8). The induced weakening leads to an increase of ice flow velocities that reaches from the floating parts of the Eastern Ice Shelf upstream and across the grounding line. The western half of the Eastern Ice Shelf is more affected than its eastern half, with a mean increase in ice flow velocities of 124±66 m yr-1 across the grounding line or about 9±5 % of the 2019 velocity from Landsat-8 feature tracking. Between the two remaining portions of the pinning point, ice velocities almost double with the removal of basal shear stress, underlining the funneling effect between the eastern and western portions of the pinning point. Entire ungrounding from the pinning point (Fig. b) causes a widespread reconfiguration of regional ice dynamics, with the Eastern Ice Shelf doubling in speed according to the stress balance. We interpret this as the model’s representation of a Western Glacier Tongue-like breakup following the unpinning. Mean ice-flow velocities across the grounding line increase by 154±63 m yr-1, corresponding to 10±3 % of the 2019 velocity.