For PLS regression analysis, it was found that coefficient of determination
(R2
) for top and sub soil properties range from 0.523 to 0.916 and 0.589 to 0.900,
respectively. The best predictive model of top and sub soil property was sand and Ca,
respectively. While the worst predictive model of top and sub soils was silt and N,
respectively. At the same time, three significant soil forming factors from PLS
regression analysis accordance with VIP values were used as 3 auxiliary variable of
cokriging interpolation for soil property prediction. It was found that RMSE of top
and sub soils properties range from 0.094 to 308.7 and 0.031 to 272.4, respectively. In
addition, an optimum semivariogram type of cokriging interpolation for physical and
chemical topsoil properties was Spherical model. In contrary, an optimum
semivariogram type for physical subsoil properties was Exponential model and
chemical subsoil properties was Spherical or Gaussian model. As results, it was found
that an optimum model for top and sub soil properties prediction based on least
NRMSE was PLS regression model. However, cokriging interpolation model
provided a better result for available P and K prediction of subsoil than PLS
regression model.