Events and Meetings of Italian Statistical Society, Advances in Latent Variables - Methods, Models and Applications

Font Size: 
Advances in estimation of large hierarchical spatio-temporal models
Håvard Rue

Last modified: 2013-06-17


In this talk I will discuss our work towards flexible spatio-temporal models that we can do full Bayesian analysis of in pratice. Starting with spatial models, the concept of using stochastic partial differential equations (SPDEs) to represent Gaussian fields has been very successful. We automatically achieve valid covariance functions and a Markov representation of the continuously indexed field. Although non-separable spatio-temporal covariance functions are easy to construct, the aim is to extend the SPDEs to time-dependent SPDEs providing more "realistic" non-separable models. I will discuss how can be done within our current framework, the computational challenges we meet and will meet following this path, and our current work to overcome them.