全球变化知识资源中心

We use PISM (refs 21, 29) to carry out regional simulations on 7 km and 10 km resolution. Our hybrid shallow approximation ensures stress transmission across the grounding line and a smooth transition between regimes of fast-flowing, sliding and slowly deforming, bedrock-frozen ice. The grounding line can freely evolve also under lower resolution owing to a local interpolation of the grounding-line position that affects the basal friction and a new driving stress scheme at the grounding line. The interpolation leads to reversible grounding-line dynamics that is consistent with full-Stokes simulations in high resolution²³. The response time to external perturbations on decadal timescales deviates from the full-Stokes simulations by a factor of around two²³. Compared with these decadal-scale perturbations, the relevant dynamics in this study occur on timescales that are one order of magnitude slower, that is, on centennial timescales and below. We therefore expect only minor influence on our results.

Model boundaries lie on present drainage divides¹⁵ and are far from the region of self-sustained retreat. Boundary velocities are zero for the drainage divides and taken from observations¹⁵ for the glaciers draining through the Transantarctic Mountains. The fixed boundaries may lead to a slight underestimation of the equilibrated ice loss from the unstable simulations because no ice can be drained from neighbouring basins.

We ran the model under constant ocean and atmosphere boundary conditions for at least 40 kyr to generate the equilibrium states that serve as starting points for our forcing experiments (Supplementary Table 1). We sampled a three-dimensional parameter space to capture the core uncertainty in ice flow and basal sliding by variation of the two enhancement factors for the shallow-ice approximation and the shallow-shelf approximation as well as the basal till pore-water pressure. The maximum deviation of ice volume from observations is less than 8% for the total domain and less than 20% for the plug region. The ice plug (Fig. 3, x axis) is defined as the difference from the observed volume and therefore the equilibrium deviation does not directly affect the determined threshold.

Subshelf melting is calculated from ocean-model temperature and salinity data solving a three-equation system³⁰ for the boundary layer beneath the ice shelf as also applied in at least two dynamic Southern Ocean models. The parametrization is detailed in the Supplementary Information.

The equilibrium simulations are forced with time-constant ocean data from the BRIOS (ref. 25) model, leading to typical temperatures of −1.8 °C at the ice–ocean interface (Supplementary Fig. 3). We enforce the ice-mass loss from the George V coast by warm water anomalies from 1.0 to 2.5 °C and periods of 200–800 years. The resulting subshelf melting, boundary-layer temperature and boundary-layer salinity are illustrated in Supplementary Fig. 4. The forcing time and strength that is sufficient to remove the ice plug depends on the ice-model parameters. Details are given in Supplementary Table 2. We integrated all forcing experiments for at least 25 kyr after the forcing ended so that a stable grounding line was reached.

As oceanic conditions in regions covered at present by land ice are unknown, we filled missing ocean data by simple diffusion along regions below sea level, leading to water properties in the Wilkes subglacial basin similar to today’s George V coast. Salinity may deviate from observations in the model data we use. However, the sensitivity of subshelf melting to salinity is relatively weak (Supplementary Fig. 5).

Atmospheric boundary conditions are taken from observations²⁴. We test the sensitivity to the potential future increase of surface accumulation by scaling the pattern observed at present with a factor of 1.3.