ghattasMICDE Seminar: Omar Ghattas

Omar Ghattas is the John A. and Katherine G. Jackson Chair in Computational Geosciences, Professor of Geological Sciences and Mechanical Engineering, and Director of the Center for Computational Geosciences at the Institute for Computational Engineering and Sciences (ICES) at the University of Texas at Austin. He also holds courtesy appointments in Computer Sciences, Biomedical Engineering, and the Texas Advanced Computing Center. He earned his PhD in computational mechanics from Duke University.

He has general research interests in simulation and modeling of complex mechanical, geological, and biological systems on supercomputers, with specific interest in inverse problems and associated uncertainty quantification for large-scale systems. His center’s current research is aimed at large-scale forward and inverse modeling of whole-earth, plate-boundary-resolving mantle convection; global seismic wave propagation; dynamics of polar ice sheets and their land, atmosphere, and ocean interactions; and subsurface flows, as well as the underlying computational, mathematical, and statistical techniques for making tractable the solution and uncertainty quantification of such complex forward and inverse problems on parallel supercomputers.

He received the 2003 IEEE/ACM Gordon Bell Prize for Special Accomplishment in Supercomputing, was a finalist for the 2008, 2010, and 2012 Bell Prizes, and received the 2008 TeraGrid Capability Computing Challenge award.

Towards Bayesian Inversion for Large-Scale Antarctic Ice Sheet Flow

4 p.m., Friday, Feb. 28, 2014
Room 2211, G.G. Brown Building, 2350 Hayward

The flow of ice from the interior of polar ice sheets is the primary contributor to projected sea level rise. One of the main difficulties faced in modeling ice sheet flow is the uncertain spatially-varying boundary condition that describes the resistance to sliding at the base of the ice. Satellite observations of the surface ice flow velocity, along with a model of ice as a creeping incompressible non-Newtonian fluid, can be used to infer the uncertain basal sliding parameter field. We approach this ill-posed inverse problem using both regularization theory as well as Bayesian methods. The latter provide a systematic framework to infer not only the basal sliding parameters, but also the associated uncertainty.

However, solution of Bayesian inverse problems via conventional MCMC sampling methods remains prohibitive for expensive models and high-dimensional parameters.  Observational data, while large-scale, typically can provide only sparse information on model parameters. Based on this property we design MCMC methods that adapt to the structure of the posterior probability and exploit an effectively-reduced parameter dimension, thereby rendering Bayesian inference tractable for high-dimensional Antarctic ice sheet flow inverse problems.

This work is joint with Tobin Isaac, James Martin, Noemi Petra, and Georg Stadler.

Ghattas’ talk is co-sponsored by the Department of Mechanical Engineering.