Abstract:
|
Hierarchical spatial models are very widely used in a number of disciplines, including climate science. They provide a framework for combining information from disparate sources while accounting for spatial dependence and complicated error structures. Computation can remain a challenge for many such models, especially when the data sets are high-dimensional. For example, research on the West Antarctic ice sheet requires combining physical models with non-Gaussian spatial data. I will describe a new dimension-reduction strategy for speeding up Bayesian inference for hierarchical spatial models, and demonstrate how it is useful for studying ice sheets.
|