|
Abstract:
|
In recent years there has been some debate in the spatial statistics literature over the merits (or lack thereof) of reduced rank random effects parameterizations for spatial models. These same concerns are often implicitly assumed to be present in spatio-temporal processes. However, these concerns are not necessarily warranted in dynamic spatio-temporal models (DSTMs) given that the true dynamical process often resides on a lower-dimensional manifold. But, one does have to be concerned about the type of basis functions used, and more importantly, the way in which the projection coefficients are evolved in time. Here we show the importance of interaction and the challenges associated with estimation in nonlinear DSTMs.
|
