An Ice Sheet Model Initialization Procedure for Smooth Coupling with Climate Forcing
Mauro Perego1, Stephen F Price2, Georg Stadler3, Irina Tezaur4, Andrew Salinger5, John Jakeman5, Mike Eldred5 and Charles Parsons Jackson6, (1)Sandia National Laboratories, Albuquerque, United States, (2)Los Alamos National Laboratory, Fluid Dynamics and Solid Mechanics, Los Alamos, NM, United States, (3)UT Austin, Austin, TX, United States, (4)Sandia National Laboratories, Livermore, United States, (5)Sandia National Laboratories, Albuquerque, NM, United States, (6)University of Texas, Institute for Geophysics, Austin, TX, United States
Abstract:
In order to perform transient simulations of ice sheet evolution, a spin up of several thousands of years is usually required to obtain an initial state for which the ice sheet model is close to thermo-mechanical equilibrium. Using this approach, however, the final ice sheet geometry will likely be significantly different from that of the present-day, which will strongly impact forward model simulations, in particular those targeting a short (50-100 years) period of time. To mitigate this problem, we propose an adjoint-based optimization algorithm for the ice sheet initialization, in which we minimize the mismatch with observed surface velocity data and between the surface mass balance forcing and the modeled flux divergence. We invert for basal topography and basal friction fields simultaneously, and use the first-order Stokes approximation. We provide results for the Greenland ice sheet and also compare our proposed approach to the more widely used approach of matching surface velocity data only, by inverting for basal friction parameters. Also, we compute the sensitivity of integral quantities like the total ice discharge flux with respect to spatial changes in the basal friction or the basal topography. Current and future work will focus on modeling uncertainties of the optimized basal sliding parameters using the inverse of the Hessian as an approximation for the covariance of the parameter distribution.
