If in addition to direct variograms the cross variograms are defined for all variable pairs, then the prediction mode becomes multivariable (i.e., cokriging or co-simulations). In case of multivariable prediction, prediction error covariances from multivariable prediction [26] on a map can be specified per identifier pair with covariances( id1, id2) (example 11). In gstat, multivariable prediction comprises simple cokriging, ordinary cokriging or universal cokriging (as well as standardised cokriging, multivariable indicator or Gaussian simulation).
When, for multiple variables a linear model is specified with independent errors (no variograms are defined), and one or more of the variables' regression parameters are defined as common parameters (with the command merge, section 4.1) then the prediction mode becomes multivariable as well (cf. analysis-of-covariance models refXXchristensen96).