Spatial deformation in a Bayesian spatiotemporal model for matrix-variate responses

Abstract

In this work we propose a matrix-variate spatiotemporal model to fit two or more response variables, measured at different points in time and distinct locations in space. Here we relax the hypothesis of isotropy in space, since this assumption is not reasonable in various real situations. We treat the anisotropy of the spatial process using the concept of spatial deformation and we deal with temporal structure via Bayesian dynamic models. Through simulation studies and also with two examples, we show that the incorporation of spatial deformation provides better interpolations in some anisotropic scenarios. We also present a solution to deal with missing values in diferente response variables.