Modeling dynamic controls on ice streams: a Bayesian statistical approach

Abstract

Our main goal is to exemplify the study of ice-stream dynamics via Bayesian statistical analysis incorporating physical, though imperfectly known, models using data that are both incomplete and noisy. The physical–statistical models we propose account for these uncertainties in a coherent, hierarchical manner. The initial modeling assumption estimates basal shear stress as equal to driving stress, but subsequently includes a random corrector process to account for model error. The resulting stochastic equation is incorporated into a simple model for surface velocities. Use of Bayes' theorem allows us to make inferences on all unknowns given basal elevation, surface elevation and surface velocity. The result is a posterior distribution of possible values that can be summarized in a number of ways. For example, the posterior mean of the stress field indicates average behavior at any location in the field, and the posterior standard deviations describe associated uncertainties. We analyze data from the 'Northeast Greenland Ice Stream' and illustrate how scientific conclusions may be drawn from our Bayesian analysis.