Quantifying Century-Scale Uncertainties Of The Global Mean Sea Level Rise Contribution From The Amundsen Sea Sector, West Antarctica

Predictions of the Antarctic ice sheet contribution to sea-level have large uncertainties. Confidence in projections relies on multi-model agreement, probabilistic calibrations to judge simulations by consistency with observations and a solid representation of observational uncertainties. Each of these factors require statistical considerations which can be challenging and even computationally prohibitive, especially since ice sheet models are often computationally expensive. Here we show that diverging results from two structurally very different ice sheet models can be fully explained by differences in the study designs. Furthermore we find that the spatial characteristics of ice thickness change observations can be used to improve probabilistic calibrations of simulations from the high-resolution ice sheet model BISICLES in the Amundsen Sea Embayment, West Antarctica. These spatial constraints reduce the 50-year sea-level contribution uncertainty interval by nearly 40%, compared to a calibration on total mass loss, and by nearly 90% compared to using no observations at all. Finally, we build a stochastic model to construct an ensemble of plausible bedrock topographies and show that measurement and interpolation uncertainties for Pine Island Glacier topography contribute substantially to predictive uncertainties. Our most optimistic and pessimistic 100-year projections are 4.7 ±-0.87 mm and 19.4 ±-5.15 mm sea-level contribution (mean and standard deviation), where the stated uncertainties originate solely from the bedrock. Together, this work improves our understanding of limitations in modelling the future of the Amundsen Sea Embayment and helps to substantiate and reduce predictive uncertainties. Most of the uncertainty quantification methods we adapt to ice sheets in this work can be applied to large ensemble, Antarctic-wide model studies where fast but exhaustive use of available information is crucial. The stochastic approach to bedrock uncertainty is, in its current form, limited to regional applications but is nevertheless a call for caution for the use of any single bedrock topography