Improved spatial prediction of soil properties and soil types combining semi-automated landform classification, geostatistics and mixed effect modelling

Abstract

Landforms classifications using a range of techniques from pure automatic classification to hybrid semiautomated model / expert opinion was tested as a base for a robust modelling of soil characteristics spatial distribution. The current approaches for producing soil maps use a single model which either blocks/controls the grouping effects or do not statistically recognize the natural landscape groupings. This study tested mixed-effects modelling technique for ingenious recognition of soil groupings and consequent improvement of the accuracy of the resultant soil maps. It further tested the various landscape classification for soil mapping. Mixed-effects modelling is a form of regression analysis for simultaneous modelling of the average landscape characteristics and individual units within the landscape. Hence, it can model a family of curves and potentially remove inadequacies inherent in the current models for soil mapping. Its potential in regression kriging of continuous and categorical soil attributes has been shown in this paper. Compared to the current application of a single model in regression kriging, mixed-effects modelling produced about five times improvement of the mapping accuracy. It is anticipated that its adoption will contribute to improved soil mapping.