Satellite record reveals 1960s acceleration of Totten Ice Shelf in East Antarctica

The Antarctic Ice Sheet has experienced accelerated mass loss in recent decades, with the largest losses from West Antarctica and the Antarctic Peninsula. Mass balance in East Antarctica remains uncertain overall, but Wilkes Land, and in particular Totten Glacier (TG), stands out as a region of substantial mass loss since 1989. A major limitation in understanding the long-term response of TG and its contribution to sea level rise is the scarcity of pre-1989 satellite observations, impeding reconstruction of historical mass balance. TG is a large, marine-based glacier with a retrograde bed slope and susceptibility to marine ice sheet instability. Prior studies report increased basal melting, grounding line retreat, and ice flow acceleration since the late 20th century. This study aims to quantify earlier ice dynamics and mass balance by reconstructing TG’s ice velocity fields from 1963 to 1989 using declassified and early satellite imagery, linking historical variability to recent trends, and identifying the drivers of observed acceleration.

Previous satellite-based assessments showed that the Antarctic Ice Sheet has been losing mass, with reconciled estimates indicating small net change in East Antarctica but high uncertainty. Wilkes Land departs from this picture, with reported mass loss of approximately 51 ± 13 Gt/y. TG has undergone consistent mass loss since 1989, totaling about −175 Gt (≈0.5 mm sea level rise) at 7 ± 2 Gt/y from 1989 to 2015. TG’s ice shelf has experienced enhanced basal melting, increasing from about 9.1 m/y (1992–2007) to roughly 18 m/y (2005–2011), and its grounding line retreated by up to ~3 km between 1996 and 2013. Ice flow acceleration up to ~18% was also observed between 1989 and 2015. However, knowledge of pre-1989 dynamics is sparse due to limited data availability from InSAR, altimetry, and gravimetry. Early optical satellite imagery (ARGON, Landsat) has been shown to allow reconstruction of historical ice flow with uncertainties comparable to recent velocity products, enabling estimation of mass balance through input–output methods.

• Data and periods: Historical optical images from ARGON (1960s) and Landsat MSS/TM (1970s–1980s) were used to reconstruct TG velocity fields for 1963–1973, 1973–1989, and 1989 (8 months). An overall 1963–1989 velocity map (500 m grid) was generated by weight-averaging within a natural neighbor framework and supplemented by a regional velocity map for peripheral coverage.
• Image processing and velocity derivation: A hierarchical matching and network densification method tracked ice features in image pairs. Processing addressed film deformation, large-format lens distortion, and large georeferencing errors (>10 km). Damaged fiducial marks on ARGON films were recognized and measured semi-automatically. Exterior orientation parameters were initialized from ephemerides and ground features, then refined by bundle adjustment.
• Overestimation (OE) correction: Long-timespan velocity maps can overestimate velocities where acceleration occurs. An innovative Lagrangian velocity-based OE correction, not requiring field data, was applied. OE was concentrated along the main trunk, grounding zone, and shelf front; average correction was 50 ± 39 m/y for the 1963–1973 and 1973–1989 maps. The short 1989 span required no OE correction.
• Uncertainty estimation: Velocity uncertainties (4–79 m/y) were computed via error propagation using orthorectification error, feature identification error, matching error, and timespan. Orthorectification errors were quantified using check-point displacements. Feature identification and matching errors were set to 0.5 pixels (ARGON identification set to 2 pixels; MSS matching 1 pixel for 1963–1973 due to lower image quality).
• Mass balance (input–output): Surface mass balance (SMB) from RACMO2.3 p2 is used for 1979 onward. The reference SMB (67.7 ± 4.1 Gt/y; average 1979–2016) fills 1963–1978. Ice thickness from BedMachine Antarctica informed flux computation across an improved flux gate (moved to reduce thickness uncertainty from 61 m to 32 m). Discharge across the grounding line equals flux plus a correction for SMB between the flux gate and the grounding line. Mass balance (MB) per period is SMB minus discharge. Cumulative discharge, SMB, and MB were integrated relative to the reference SMB.
• Ocean modeling (basal melting): A regional ROMS configuration including ice–ocean interaction (three-equation parameterization) simulated basal melt from 1960 to 2007 with COREv2 forcings. The domain (104.5°E–130°E; 60°S–68°S) used RTopo bathymetry with inclusion of the Totten shelf cavity and western trough entrance. Surface fluxes combined SSM/I sea ice formation with COREv2 fields, enabling long-term polynya activity estimates.
• Calving response modeling: The Ice Sheet System Model (ISSM) was used to simulate velocity response to calving-front retreat. The model was initialized with 1963–1973 velocity and geometry; inversions estimated ice rigidity (B) within the shelf and basal friction coefficient (C) for grounded ice. Regularization in the shelf inversion was tuned by L-curve analysis. The model simulated the 1985 velocity field using the 1985 shelf front to assess dynamic response to the 1973–1985 retreat.

• Long-term acceleration and discharge: TG exhibits a persistent increase in ice discharge over nearly six decades (1963–2018), with a long-term ice discharge rate of 68 ± 1 Gt/y and an acceleration of 0.17 ± 0.02 Gt/y². Total discharge over 1963–2018 is 3830 ± 31 Gt.
• Shelf-wide acceleration tied to calving: Between 1963–1973 and 1973–1989, floating ice accelerated by 60 ± 11 m/y on average, with increases up to 135 ± 9 m/y near the shelf front. A major calving-front retreat during 1973–1985 caused a shelf area loss of about −645 km² (12%), including loss of ~104 km² (−2%) of active shelf ice beyond the passive shelf-ice boundary. Velocity near the shelf front (Box 1) peaked during 1973–1989; its long-term variability correlates with shelf-front area change (R² ≈ 0.8, timespan weighted).
• Grounding line acceleration linked to basal melt: Near the grounding line (Box 2), velocity was initially low in 1963–1973 (123 ± 21 m/y, 15% below long-term average) and shows a sustained increasing trend through 2018. After removing the linear trend, residuals correlate with shelf-front area changes (R² ≈ 0.6). Modeled basal melt rates are elevated near the grounding line (~11 m/y) and along margins (~5 m/y), consistent with acceleration in these regions, while shelf-front melt is low (~2 m/y).
• Oceanographic context: CTD profiles from 1995–1996 indicate presence of warm modified Circumpolar Deep Water (mCDW) (>−0 °C, salinity >34.5 PSU) at ~300–600 m depth near the continental shelf break/front of TIS, supporting the mechanism of mCDW intrusion into the cavity driving basal melt and acceleration.
• Modeled dynamic response to calving: ISSM simulations show that the 1973–1985 shelf-front retreat and loss of lateral margin contact induced an instantaneous speed-up >300 m/y near the western margin and a shelf-wide average increase of ~84 m/y. The modeled response at the grounding line (<10 m/y) is smaller than observed, indicating calving alone cannot explain grounding-line acceleration.
• Mass balance (period estimates): For 1963–1973, SMB 67.7 ± 4.1 Gt/y, discharge 64.7 ± 3.0 Gt/y, MB +3.0 ± 5.0 Gt/y. For 1973–1989, SMB 69.5 ± 4.2 Gt/y, discharge 66.1 ± 1.9 Gt/y, MB +3.4 ± 4.6 Gt/y. For 1989 (8 months), SMB 67.1 ± 4.0 Gt/y, discharge 67.4 ± 10.9 Gt/y, MB −0.4 ± 11.6 Gt/y.
• Cumulative balance transition: The TG basin shifted from a net mass gain of 130 ± 27 Gt over 1963–1989 to a net mass loss of 136 ± 36 Gt over 1990–2018. The basin discharged on average 66 ± 1 Gt/y during the first 27 years, compensating high SMB by ~54 ± 12 Gt and reaching equilibrium around 1989; an additional 92 ± 28 Gt of excess discharge occurred in the subsequent 29 years, accelerating mass loss.
• Regional significance: TG is identified as the greatest contributor to sea level rise in East Antarctica over the study period.

The reconstructed velocity fields demonstrate that TG’s dynamic state has been responding to two primary forcings: (1) episodic calving-front retreat that modulates ice-shelf flow and propagates dynamic changes across the shelf, and (2) persistent basal melting near the grounding line, likely driven by mCDW intrusions, which underpins the long-term acceleration and discharge increase. Correlations between shelf-front area change and velocity near the front indicate calving controls short-to-decadal shelf-front speed variability. In contrast, grounded ice near the grounding line exhibits a steady multi-decadal acceleration, consistent with increased basal melt inferred from ocean-ice modeling and limited CTD observations showing mCDW presence on the continental shelf. Modeling confirms that calving-induced changes can instantaneously speed up the shelf, but cannot fully account for observed grounding-line acceleration, implicating mCDW-driven basal melt as the dominant driver of long-term discharge increase. TG’s current evolution parallels that of Pine Island Glacier, where decades of basal-melt-driven retreat and speed-up preceded extensive calving-front retreat. Given the retrograde bed and observed retreat along the eastern lobe, continued mCDW access to the cavity is expected to further destabilize the system, increasing the likelihood of more frequent or extensive calving and accelerating mass loss.

Using first-generation satellite imagery (ARGON, Landsat-1 and -4), the study reconstructs TG velocity fields from 1963 to 1989 and links them to modern observations to establish a nearly 60-year dynamic record. New processing and correction techniques enabled accurate historical velocity retrievals and mitigation of long-timespan overestimation. Findings show shelf-wide acceleration in 1963–1989, with shelf-front speed-up driven by a large 1973–1985 calving-front retreat and grounding-line acceleration driven primarily by mCDW-induced basal melt. Over 1963–2018, TG experienced a persistent increase in discharge (acceleration ~0.17 ± 0.02 Gt/y²), transitioning from modest mass gain pre-1989 to net mass loss thereafter, making TG the largest contributor to sea level rise in East Antarctica. The acceleration and mass-loss trend likely began in the 1960s. The study recommends intensified monitoring of ice–air–ocean interactions in the Totten region and continued integration of historical and modern datasets to improve constraints on future sea level contributions.

• Sparse oceanographic observations: Available CTD profiles are few, collected mostly in 1995–1996, and constrained by fast ice and grounded icebergs, preventing deployment close to the shelf front. This limits direct observational validation of mCDW intrusions during the historical period.
• Temporal resolution: Historical velocity reconstructions span multi-year periods (up to 16 years), which preclude resolving annual to interannual variability; an overestimation correction was required to address acceleration bias in long-timespan image pairs.
• SMB gap and assumptions: SMB before 1979 is filled using the reference average (1979–2016), introducing uncertainty in early-period mass balance estimates.
• Ice thickness and gate placement: Although the flux-gate placement reduces ice thickness uncertainty (to ~32 m), remaining thickness and geometry uncertainties propagate into discharge estimates.
• Modeling simplifications and forcings: Ocean and ice-sheet models rely on reanalysis forcings and parameterizations; cavity geometry and basal processes are idealized to some extent, and inversion regularization choices may affect the modeled responses.
• Feature tracking limitations: Fast-flowing shelf-front regions over long timespans lose trackable features, requiring short-span data (e.g., 1989 map) to fill gaps, with higher associated uncertainties (~79 m/y).