[R-sig-Geo] different spatial structures for different variables

[Paolo Piras]

Dear all,
my question is some-what related to a previous post that received an answer. 
However, this time I want to ask you an advice about the fact that when fitting multivariate-multiple gls model.... different responses and-or different predictors could have DIFFERENT correlation structures (i.e. gaus, exp or spher.).
How then one can fit a unique model?
Maybe one can perform a gls without response for ANY spatially correlated variable by applying the correct corStructure and successively use the residuals of such models in a unique model.
However, I'm not really sure about this.
Thankyou in advance for any advice
Best regards
Paolo Piras