Adaptive Mesh Refinement Versus Subgrid Friction Interpolation in Simulations of Antarctic Ice Dynamics

Dynamic-adaptive mesh ice sheet shows importance of mesh spacing to capture instability of ice-sheets.

The Science

The 2007 & 2013 Intergovernmental Panel on Climate Change (IPCC) reports highlighted the need for better projections of the Antarctic contribution to sea-level rise (SLR). The West Antarctic Ice Sheet (WAIS) is vulnerable to marine forcing from warm-water incursion into subshelf cavities, causing melting and thinning of ice shelves and weakening or eliminating the buttressing exerted on feeder ice streams.  The result is accelerated ice loss and increased SLR. Theory and modeling suggest that ice-sheet models require very fine (sub-km) resolution to correctly model these effects. This study uses extreme melt forcing scenario to examine resolution requirements for millennial-scale whole-continent simulations.

The Impact

This is the first fully-resolved whole-continent millennial-scale Antarctic simulations to evaluate resolution requirements and it is broadly useful as a method to ensure confidence in model results. The Berkeley Adaptive Mesh Refinement (AMR) ice sheet model (BISICLES) fully resolves dynamically important regions like grounding lines. The use of a basic subgrid-scale friction interpolation scheme results in factor-of-two reduction in resolution requirements. Under-resolved-mesh simulations consistently underestimate SLR contributions. Even moderate under-resolution results in erroneous predictions of timing and mechanisms of WAIS collapse, while severe under-resolution results in missing WAIS collapse entirely. This is the first application of BISICLES, a state-of-the-art AMR ice sheet model, to millennial-scale whole-continent projections of Antarctic response to extreme climate forcing. AMR allows for full resolution of dynamically changing grounding lines in regional- and continental-scale simulations.


At least in conventional hydrostatic ice-sheet models, the numerical error associated with grounding line dynamics can be reduced by modifications to the discretization scheme. These involve altering the integration formulae for the basal traction and/or driving stress close to the grounding line and exhibit lower — if still first-order — error in the MISMIP3d experiments. MISMIP3d may not represent the variety of real ice streams, in that it lacks strong lateral stresses, and imposes a large basal traction at the grounding line. We study resolution sensitivity in the context of extreme forcing simulations of the entire Antarctic ice sheet, using the BISICLES adaptive mesh ice-sheet model with two schemes: the original treatment, and a scheme, which modifies the discretization of the basal traction. The second scheme does indeed improve accuracy — by around a factor of two — for a given mesh spacing, but < 1 km resolution is still necessary. For example, in coarser resolution simulations Thwaites Glacier retreats so slowly that other ice streams divert its trunk. In contrast, with < 1 km meshes, the same glacier retreats far more quickly and triggers the final phase of West Antarctic collapse a century before any such diversion can take place.

Principal Investigator(s)

Dan Martin
Lawrence Berkeley National Laboratory


Earth System Modeling, SciDAC PISCEES project


Adaptive mesh refinement versus Subgridfriction interpolation in simulations of Antarctic Ice Dynamics”, Cornford, Martin, Lee, Payne, Ng, Annals of Glaciology, 57 (73), DOI:10.1017/aog.2016.13, 2016