Particle-size fractions (psf) of mineral soils and, hence, soil texture, are the most important attributes affecting physical and chemical processes in the soil. More often, psf data are available only at a few locations for a given area and, therefore, require some form of interpolation or spatial prediction. However, psf data are compositional and, therefore, require special treatment before spatial prediction. This includes ensuring positive definiteness and a constant sum of interpolated values at a given location, error minimization, and lack of bias. In order to meet these requirements, this study applied two methods of data transformation prior to kriging of the psf of soils in two regions of eastern Australia. The two methods are additive log-ratio transformation of the psf (ALROK) and modified log-ratio transformation (m ALROK). The performance of the transformed values by ordinary kriging was compared with the spatial prediction of the untransformed psf data using ordinary kriging, compositional kriging (CK) (UTOK), and cokriging, based on the criteriaprediction bias or mean error (ME) and precision (root mean square error (RMSE)), and validity of textural classification. ALROK and m ALROK outperformed UTOK and CK in terms of prediction ME and RMSE. Because of the closure effect on the psf data, UTOK, and, to a lesser extent, CK, did not meet all of the requirements for spatially predicting compositional data and, therefore, performed poorly. m ALROK outperformed all of the interpolation methods in terms of misclassification of soils into textural classes. The results show that without considering the special requirements of compositional data, spatial interpolation of psf data will necessarily produce uncertain and unreliable interpolated psf values.
