Modelled ice sheet sensitivity to basal friction parameterizations is controlled by the amount of buttressing

The basal friction parameterization is often mentioned as key source of uncertainty when using ice sheet models to project future evolution of the ice sheet. Previous work suggests that parameterizations with an exponential relationship between friction and basal velocity (power laws) predict lower sea level rise than ‘Coulomb-style’ friction laws. For Coulomb laws, the basal friction asymptotes for high velocities and is effectively independent of velocity for fast-flowing ice. We use the Community Ice Sheet Model (CISM) for two kinds of simulations: one with present-day climate forcing, including sub-shelf ocean temperatures kept constant and one with 1-degree ocean warming in the Ross Sea, both matching the current rate of ice thickness changes. In the constant scenario, Thwaites Glacier and Pine Island Glacier collapse, creating a huge, laterally bounded ice shelf. In the scenario with Ross Sea warming, a large part of the Ross Ice Shelf disappears, allowing Siple Coast glaciers to flow freely into the ocean. For Thwaites and Pine Island Glaciers, there are competing processes causing increases or decreases in ice flux across the grounding line, ice shelf thickness, buttressing, and ice velocities, once the glaciers are collapsing. These processes work in the opposite direction of the differences caused by choosing different basal friction parameterizations. Therefore, in our model runs, the choice of basal friction parameterization has little effect on the collapse of Thwaites and Pine Island glacier. In the unbuttressed Siple Coast case, we confirm earlier results: Coulomb friction leads to more ice mass loss and sea level rise. We conclude that unbuttressed glaciers are more sensitive to the choice of basal friction parameterizations than are heavily buttressed glaciers, and that the presence of a large buttressing ice shelf decreases the sensitivity of glacier dynamics to different basal friction parameterizations.