Prediction Simulation Study for Cokriging via Variable Transformation

In general, cokriging is used to estimate the observation value of the primary variable using the secondary variables. One of the problems that can arise in cokriging is that when the correlation between primary and secondary variable is low, the secondary variable has no role as a secondary variable and can not be use in cokriging. But we can not say that it is not meaningful, as it may be a special function structure (e.g. exponential). Therefore, to settle this problem we propose a method that carry out cokriging after transforming the variable in case there is a secondary variable that has lower correlation with a primary variable. To show validation of proposed method, we perform simulation study that simulates the locations, primary variable, and two secondary variables related with primary variable. We assume that the two secondary variables (V, W) has correlation with a primary variable (U). Let the variables U and W has high correlation value while variables U and W has low correlation value. We perform a cokriging for above two cases and compare the result based on the PRESS statistic and use as a criterion for the cross-validation of the proposed method. The simulation result for cokriging between U and W variable is superior than that for cokriging between U and W variable.