Moisture deficits caused by groundwater extraction are land qualities that need to be interpolated for certain areas of land. Because these data are difficult to collect, and may be correlated with basic soil data, cokriging is used. Standard cokriging routines, however, do not allow the existence of a trend. Therefore, cokriging has been formulated for predicting multivariate and non-stationary data. Attention is focused on multivariate random functions. An important type of non-stationarity is defined in terms of multivariate increments that are stationary of order k. The covariance between increments of different variables is modelled by the pseudo-cross-covariance function. Cokriging equations and the induced cokriging equations are given. The procedures have been applied in a study of moisture deficits in the Netherlands. Parameters of the pseudo-cross-covariance function are estimated using Restricted Maximum Likelihood. Universal cokriging resulted in an improvement of the predictions in terms of the mean kriging error for a set consisting of 100 test data with 20–30%. No reduction, however, in the mean squared error between predictions and observations (MSE) was observed.