Constraining a large ensemble of Antarctic ice sheet simulations over the last glacial cycle using observations of past ice sheet evolution

Abstract:

To better understand the observed contemporaneous change of the Antarctic ice sheet and to make improved predictions of its future sea-level contribution, reconstructions of past ice sheet evolution are required with probabilistically self-consistent uncertainty estimates. Glaciological modeling is an effective tool to generate continental-scale reconstructions over glacial cycles, but simulations depend on parameterisations to account for deficiencies (e.g., missing physics, unresolved sub-grid processes, uncertain boundary conditions) inherent in any numerical model. These parameters, considered together, form a parameter phase space from which sets of parameters can be sampled; each set corresponds to an ice sheet reconstruction. We explore an updated Glacial Systems Model (GSM) of the Antarctic ice sheet over the last glacial cycle by performing a data-constrained large ensemble analysis. This work continues toward a Bayesian calibration, which involves a thorough sampling of model uncertainties against fits to observational data through Markov Chain Monte Carlo methods using Bayesian artificial neural network emulators of the full GSM.

A Bayesian calibration of an ice sheet and glacial isostatic adjustment model has been proven to be a powerful tool to generate explicit and well-defined confidence intervals on ice sheet and sea-level evolution. An observational constraint database is applied to calibrate the GSM. The updated database termed AntICEdat consists of observations of past ice extent, ice thickness, relative sea level, and ice core borehole temperature profiles. In addition to paleo-observations, the present-day ice sheet geometry and surface ice velocities are applied as further constraints. The large ensemble results presented here consist of over 10,000 data-constrained transient ice sheet reconstructions. Toward the final Bayesian calibration, preliminary results in the form of nominal confidence intervals of Antarctic metrics are shown. The key metrics of interest are the Antarctic equivalent sea-level contribution to the last interglacial highstand, last glacial maximum, meltwater pulse 1a, and present-day Antarctic glacial isostatic adjustment estimates.