Abstract
Nine chemical properties of soils in the Aizu Basin in the northern part of Japan, were predicted from sample data using two different methods. The predictors were the universal kriging estimator and a simplified smoothing spline, referred to as nonlinear optimization method. The performance of the methods was evaluated and compared using cross-validation.
The nonlinear optimization method is based on the concept that optimal grid data satisfy two requirements: smooth interpolation, and minimum differences between predicted and observed values at the sampling points. This method contrasts with kriging in that it is a global estimate, whereas kriging is a local one.
The kriging estimator provided a better estimate than the other method for most of the nine attributes: contents of total carbon and nitrogen, free Fe2O3 , and exchangeable Ca, Mg, K, and Na. However, it was not as effective for the phosphate adsorption coefficient (PAC) and CEC. For the phosphate adsorption coefficient, the low estimate of kriging was affected by the low estimate of the semivariogram due to a sawtooth-like pattern in the sample semivariogram. This pattern was ascribed to the presence of Ando soils in the vicinity of the Aizu Basin. For the CEC, the low estimate of kriging was due to a pure nugget effect in the semivariogram. The application of fertilizers may widen the range of semivariograms for exchangeable cations.