Title: Bayesian multivariate spatial temporal functional data modelling
Authors: Montserrat Fuentes - North Carolina State University (United States) [presenting]
Abstract: It is well known that the volume and complexity of scientific data is increasing. This increase necessitates the development of flexible, and efficient, statistical methods which are capable of accurately capturing this complexity. Motivated by the analysis of hurricane trajectories and intensities we develop a Bayesian, multivariate, functional linear model with spatially varying coefficients. The model utilizes a hierarchical structure in order accommodate noisy functional covariates and allow for the inclusion of derivatives of functional covariates. In addition, tensor product basis expansions paired with appropriately structured prior distributions are used to allow for spatially adaptive coefficients. Temporal correlation within storms is modeled using an autoregressive term. Appropriate specification of the prior distributions allows Gibbs sampling to be used for the entire model. Posterior inferences are then constructed using MCMC output.