Inversion for the Basal Sliding Coefficient Field under Uncertainty for the Humboldt Glacier.
Abstract
We consider the Bayesian inference of the unknown basal sliding coefficient in the presence of additional model uncertainty in the nonlinear first-order Stokes ice sheet model MALI. The model uncertainty stems from an imperfectly known viscosity weakening parameter field which describes impurities and damage in the model equations. To account for this nuisance parameter, we employ the Bayesian Approximation Error (BAE) approach. With BAE, the nuisance parameter and measurement noise are approximately premarginalized. This results in a posterior distribution for the basal sliding coefficient field that accounts for the model uncertainty. We discuss an efficient approximation strategy for the first and second order statistical moments of the model discrepancy, which is a critical component of the BAE framework. The resulting Gaussian model discrepancy probability density is used for the inference of the basal sliding coefficient from noisy surface velocity measurements on the Humboldt glacier. The results show that the BAE approach avoids overly confident and erroneous inference which can occur when model uncertainty is neglected entirely.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2021
- Bibcode:
- 2021AGUFM.C12B..03H