Estimation of the Basal Sliding Parameter in Ice Sheet Models in the Presence of Errors and Uncertainties
Please login to view abstract download link
Basal sliding strongly influences ice-sheet flow but is generally impossible to observe directly. Instead, it is commonly estimated based on available surface data (e.g., surface velocities) via a PDE-constrained inverse problem. However, such approaches can easily become computationally intractable for large-scale problems due to the complexity and scale of the associated PDE models. Moreover, additional parameters in the models, such as rheological parameters, are unknown or only roughly known at best, which can result in severely misleading estimates. In this work we consider the use of surrogate models within a Bayesian framework to estimate the (spatially varying) basal sliding parameter. The surrogate models could involve reduced order models, data-driven models, or simplified physics/fixed parameters. Our approach allows for systematic incorporation and quantification of uncertainties, including those induced by the use of the surrogate model. Furthermore, our proposed approach allows for application of scalable, adjoint-based approaches to solve the inverse problem. The goal is a workflow that delivers both basal sliding estimates and calibrated uncertainty quantification for more reliable ice-sheet predictions.
