[R-sig-Geo] temperature interpolation -- was: 3D kriging with gstat

[Edzer J. Pebesma]

Doug Nychka wrote:
> Prodromos,
> 
> You might also consider the altitude entering just  as a linear term 
> in elevation instead of a full spatial variable. Also don't forget that 
> aspect (whether the slope faces north or South etc.) may be an important
> covariate also. 
> 
> The easiest way to fit these kinds of models is to use back fitting ...
>
Doug, I think your linear term suggestion is identical to mine except
that we used completely different wording.

Your suggestion about back fitting triggered my curiousity though -- do
you mean using generalized additive models? If so, how then would you
deal with (residual) spatial correlation?

Best regards,
-- 
Edzer