The objective of this work was to evaluate the ordinary co-kriging algorithm, using data from elevation and distance from the sea, compared to ordinary kriging, in the rainfall spatialization associated to wet, dry and annual periods, for the state of Espírito Santo, Brazil. Data of altitude and distance from the sea were obtained in sampling sites on regular and irregular grids. Data from 108 rain gauge stations were used. The evaluation of methods and variables based on cross-validation was performed, considering the errors of the predicted values and the fit of linear regression models for observed and predicted values. The regular grid sampling of the covariates showed the best prediction accuracy compared to the irregular grid. Cokriging produced more accurate results than kriging, checked by small differences in mean absolute errors which were able to produce statistically different maps. Cokriging interpolation and use of regular grids for sampling are preferable, mainly if the covariates are easy to obtain and inexpensive.
precipitation distribution; geostatistic; coastline; spatial prediction
