Methodology:

Simulations were run using the MITgcm numerical ocean model, version c65z. The model featured an ice shelf with steady thickness and a new bathymetry created using natural neighbour interpolation of 114 seismic soundings in the LCIS cavity. The bathymetry was deepened to create a minimum water-column thickness of 40 m, the thickness of two full grid cells, to allow unhindered flow wherever the ice is known to be floating. The domain included the LCIS cavity and a small area of the western Weddell Sea, with a uniform grid resolution of 20 m in the vertical, 1/20° in longitude, and variable in latitude, scaled by the cosine of the latitude, resulting in isotropic grid cells of ~2 km width. Constant diffusivities of 10 m2/s in the horizontal and 10-4 m2/s in the vertical were used, and lateral and vertical eddy viscosity coefficients of 50 m2/s and 10-3 m2/s, respectively. A three-equation model was used to parameterise melting and freezing, with a drag coefficient of 0.0022 and heat and salt transfer coefficients of 0.011 and 3.1 x 10-4, respectively. Tides were implemented by imposing velocities on open boundaries to the north, east and south from the CATS2008 inverse tidal model. Constant ocean properties were imposed on the model boundaries with a potential temperature of -1.9°C and practical salinity of 34.5, the same values as the initial conditions. Further simulations with initial and boundary condition temperatures of -1.8°C, -1.6°C, and -1.4°C were also run, as well as alternative bathymetry simulations. The marine ice fields were calculated by assuming that modelled melt/freeze rates, and ice shelf velocities are fixed in time. These fields were interpolated onto a 100 m grid and then marine ice thickness was time-stepped on this grid for 500 years using a simple upwind advection scheme.