Advanced Ice Sheet Modeling: Scalable Stokes Solver and Inversion for Basal Sliding Parameters
Tobin Isaac, University of Texas
We present a parallel, adaptive mesh, high-order finite element solver for the 3-D Stokes equations for ice sheet dynamics. The adaptive mesh capabilities allow for efficiently capturing the wide range of length scales with localized features present in ice sheet dynamics. We solve the equations using a globalized Newton-Krylov method with block, multilevel preconditioning. We carry out realistic simulations using SeaRISE datasets. Numerical results from these simulations demonstrate scalability of the algorithm and the implementation for realistic full-continent ice sheet models. Additionally, we formulate an inverse problem to infer the basal sliding parameters from observations of surface flows. For this purpose, we minimize the misfit between observed and modeled surface flow velocities. The resulting least squares minimization problem is solved using an adjoint-based inexact Newton method. Numerical inversion studies demonstrate the influence of prior knowledge on the model parameters for addressing ill-posedness of the inverse problem and for treating noise present in the observations.
Abstract Author(s): Tobin Isaac, Noemi Petra, Hongyu Zhu, Georg Stadler, Omar Ghattas