Thursday 08 Mar 2018: Statistical science seminar- Statistical models for glaciology
Giri Gopalan - University of Iceland
There are three objectives of this talk. The first is to introduce glaciology, with a particular emphasis on Icelandic glaciers and the physics of glaciers, to a statistical audience. The second objective is to describe a Bayesian hierarchical model for glacial dynamics that has been constructed based on the shallow ice approximation (SIA) partial differential equation (PDE). In particular, this model (and the corresponding model fitting methodology) is checked with exact analytical solutions to the SIA PDE from Bueler et al. (2005). The third objective is to discuss two directions that we are pursuing to make posterior computation more efficient: the first relies upon the method of emulator inference (Hooten et al. 2011) for the SIA PDE, and the second relies upon sparse matrices for evaluating the log-likelihood of the data quickly. This work stems from a doctoral project at the University of Iceland that is sponsored by the Icelandic Research Fund.