Announcement Detail
Tuesday, December 9, 2025
8:00 AM PST
Join via Zoom: https://us06web.zoom.us/j/89805920868?pwd=UaxOZeVHZAs9BeKkjqRWbNadZYCIZo.1
Energy & Earth Systems TTA Webinar
Multifidelity uncertainty quantification for ice sheet simulations
Nicole Aretz, The University of Texas at Austin
Abstract:
Simulations of ice mass loss in Greenland and Antarctica have a central role when choosing policies to combat climate change, but suffer from vast parametric uncertainties that propagate into ice mass loss projections. For example, the basal friction field and geothermal heat flux are typically inferred in an ill-posed inverse problem, but their variation has significant long-term effects on ice velocity and temperature simulations. Quantifying the propagated uncertainties in the projections is of utmost importance to enable judicious decision-making, but high-fidelity simulations are too expensive to allow Monte Carlo approximations. Less expensive but also less accurate low-fidelity models are readily available — e.g., coarser meshes, approximated physics, physics-based reduced-order models, machine-learning methods, interpolation, and extrapolation — but the replacement of the entire high-fidelity model for a low-fidelity surrogate introduces model bias. Multifidelity methods, in contrast, keep the high-fidelity model in place but expand the estimator to shift the computational burden onto the low-fidelity models while still guaranteeing an unbiased estimate. Through this exploit of the model hierarchy, the multifidelity estimates guarantee a smaller statistical error than Monte Carlo sampling for the same computational budget. In this talk, we demonstrate and compare three multifidelity estimators — multifidelity Monte Carlo, multilevel Monte Carlo, or the multilevel best linear unbiased estimator technique — on a model of the 2015-2050 Greenland ice mass loss. For a target accuracy equivalent to 1 mm sea level rise contribution at 95 % confidence, the multifidelity estimators achieve computational speedups of two orders of magnitude.
The results presented in this talk are based on:
[1] Nicole Aretz, Max Gunzburger, Mathieu Morlighem, Karen Willcox. Multifidelity uncertainty quantification for ice sheet simulations. Comput Geosci 29, 5 (2025). https://doi.org/10.
Bio: