Soil mapping based on assessment of environmental similarity and selection of calculating samples

Abstract Many predictive soil mapping (PSM) methods encounter two problems. One is how to construct a similarity index that can indicate not only the similarity of an environmental covariate between sites but also the directional difference of the covariate itself between sites and predict the soil property values beyond the range of sample values. Another is how to determine which soil samples should be used when estimating the soil property value at an unknown site. We propose a new asymmetrical environmental contrast index between an unknown site and a sample site to address the first problem, based on which we developed an enhancive predictive coefficient (EPC)-based soil mapping (EPSM) method. EPC integrates the contrasts of associated environmental principal component covariates by the weights of the covariates influencing a certain soil property. A member recruiting process was programmed to determine the calculating samples for an unknown site after conducting an uncertainty threshold (ut) test for all the sample sites. We applied the EPSM method to five data groups with different numbers and distributions of sample sites for prediction. The results showed that the EPSM method performs better than the soil-land inference model (SoLIM) method regardless the value of ut and thus can be used to estimate the soil property values well at most unknown sites. The method is especially valid when the unknown sites are spatially far from the sample sites and when sample sites are limited in number or spatially distributed at a local area. Our study suggests that the EPSM method is an effective PSM method that can be widely used in soil mapping.