A0879
Title: Anisotropic Gaussian random fields with identifiable parameters and penalized-complexity priors
Authors: Liam Llamazares - University of Edinburgh (United Kingdom) [presenting]
Finn Lindgren - University of Edinburgh (United Kingdom)
Jonas Latz - Heriot-Watt University (United Kingdom)
Abstract: Gaussian random fields (GFs) are key in spatial modeling and can be represented efficiently as solutions to stochastic partial differential equations (SPDEs). These SPDEs depend on specific parameters which can be estimated using Bayesian inference. However, likelihood often provides limited insights under in-fill asymptotics, necessitating the use of priors. A parameterization of a non-stationary GF is introduced via its correlation length and diffusion matrix and constructs penalized complexity priors. Both the stationary case, where the parameters are constant in space, and the non-stationary case, where the parameters vary across the domain are addressed.
