Comparing regression-based digital soil mapping and multiple-point geostatistics for the spatial extrapolation of soil data

Abstract

In this study, two approaches for spatial data extrapolation are investigated. The intention here is to predict at fine spatial resolution, total gamma radiometric counts across a large mapping extent (recipient site) on the basis of finely resolved information collected from a nearby donor site. The extrapolation methods used were a digital soil mapping (DSM) regression model approach and a multivariate multiple-point statistical (MPS) approach. Qualitative interpretation of the results from both extrapolation approaches across the recipient site in the Lower Hunter Valley, Australia (area≈220km2) shows promise in terms of highlighting known geochemical and physical variations of soils in this area. The extrapolated map was evaluated in a small portion of the study area (area≈4km2) where similar high-resolution gamma radiometric data were available. Results show comparable performance of both approaches where a root-mean-square error of 87ppm was found. A concordance correlation coefficient value of 0.04 was found for the DSM approach, but higher for the MPS approach (0.16). Under the Homosoil framework, where soil point data and mapping are sparse, either method investigated in this study would be suitable as a ‘first-cut’ approach for developing a comprehensive soil information system in those areas.