[R-sig-Geo] nonseparable space/time covariance functions in an LGCP?

[Seth Flaxman]

Hi list.

I'm looking to fit a spatiotemporal log-Gaussian Cox Process (LGCP)
with a nonseparable covariance function. Before I do the
implementation from scratch (actually I'm thinking about using Stan),
does anyone know of a package or code in any language
(R/WinBugs/matlab/C++) that does something similar? I've looked at
INLA and Benjamin Taylor's lgcp package in R. I've also looked at GPML
and GPStuff. The latter two come the closest; I think for those I'd
just need to implement the nonseparable covariance.

And while I'm writing, I'm curious if there's some other model class /
method that comes to mind--the problem I'm tackling is one where I'm
specifically interested in comparing a model with a separable
covariance function (i.e. space and time decompose multiplicatively or
additively) to a model with a nonseparable covariance function, also
known as "space/time" interaction. I know about space/time interaction
tests like Knox and Mantel, and I know there are some tests (Fuentes
has one) for nonseparability, but I want something model based.

Thanks,
Seth