Optimising a regional Antarctic Ice Sheet model to investigate basal conditions and initialise transient experiments

Computer models for ice sheet dynamics are the primary tools for making future predictions of ice sheet behaviour, marine ice sheet instability, and ice sheet contributions to sea level change.  Such modelling studies face a number of challenges, and we consider here two examples.  The dominant mode of flow for ice streams is sliding at the bed, and the physical processes that control sliding are hard to observe. Ice sheet models often prescribe basal resistance as a function of sliding velocity.  But laboratory experiments and real-world observations indicate that basal resistance is also dependent on the water pressure in the sub-glacial hydrologic system, a property that is hard to constrain.  Initialising an ice sheet model for future projections is usually implemented either by a multi-millennial spin up or else by optimisation simulations, both of which have significant drawbacks.  In particular, long spin-up simulations cannot easily ensure a close match to present-day ice geometry, and optimisations cannot easily ensure an overall ice sheet mass balance that matches the present-day mass balance.

Using a 3D Stokes-flow ice dynamic model, we carry out optimisations for two Antarctic catchments: The Pine Island Glacier (PIG) in West Antarctica and the Lambert-Amery Glacier System (LAGS) in East Antarctica.  We optimise both the basal resistance and flow enhancement in order to minimise discrepancy between modelled and observed (from satellite) horizontal velocities at the ice upper surface.  We use these optimised model configurations to estimate the transient mass trend and also look at the 3D velocity field, its sensitivity to choice of boundary conditions in the normal direction at upper and lower surfaces, and its implications for the 3D temperature structure.  These simulations provide an estimate of the present-day thermo-mechanical state of the PIG and LAGS.

We demonstrate that constraining only horizontal velocity in the optimisations can lead to unrealistic normal velocities at the upper surface.  We show that this can, in turn, strongly impact on the catchment’s total mass budget (through locally unconstrained thinning/thickening rates) and lead to a large-scale bias in temperatures simulated using the optimised model with the steady state assumption, due to unphysical advection of heat through the ice upper surface.

We employ the optimised model to estimate basal melt, due mainly to friction heat, and drive a subglacial hydrology model beneath the PIG, providing a model-based estimate of the distribution of basal water pressure.  We use this, along with simulated sliding velocity and basal resistance, to evaluate some commonly used sliding relations.