A COMPARISON OF KRIGING, CO-KRIGING AND KRIGING COMBINED WITH REGRESSION FOR SPATIAL INTERPOLATION OF HORIZON DEPTH WITH CENSORED OBSERVATIONS

We compared the performances of kriging and two methods of interpolation which allow to account for an auxiliary variable: co-kriging, and kriging combined with regression. These two methods were applied to improve the interpolation of the soft layers depth (D-sl) measured by augering, using the bulk soil electrical conductivity (EC(a)) measured by an electromagnetic instrument as auxiliary variable, We predicted D-sl from observations of EC(a) using an exponential regression model. These predictions were treated as uncertain measurements of D-sl in kriging. Results of this type of kriging were compared with those of co-kriging for grids of various spacing for D-sl. As the analysis of the spatial structure showed the presence of a drift of degree 2, IRF-2 kriging using generalized covariance functions was applied. Soft layers were not detected within the augering depth of 1.20 m at 10% of the locations. We paid special attention to these so-called censored observations of D-sl in the fitting of the regression model and in the interpolations. Kriging combined with regression gave better results than co-kriging. Moreover, in kriging combined with regression fewer model parameters needed to be estimated. This would be even more advantageous if two or more auxiliary variables were used.