Re-tooling of regression kriging in R for improved digital mapping of soil properties

Abstract

Regression analysis and kriging are popular spatial estimation methods often used in soil science to provide soil information at different spatial resolutions and extent. Attempts have been made to combine them into a method known as regression kriging (RK). With the increasing acceptance of digital soil mapping paradigm, utilization of spatial estimation method such as RK is bound to rise. Although RK is versatile and popular, its current format has deficiencies which can hinder the quality of estimated soil properties. One of the deficiencies of RK is the failure of its regression model to recognize that natural soil occurs in groups with unique response characteristics to soil forming factors. Ideally, these groups should be represented as a family of curves when modelling the landscape. However, the current applications tend to use average models which either block/control the grouping effects or do not statistically recognize them. In this paper, mixed-effects modelling technique is shown for ingenious recognition of soil groupings and consequent improvement of RK accuracy. Mixed-effects modelling allows for simultaneous regression estimation for individual models in a group and for different groups in the landscape. Its implementation in RK has been illustrated using executable scripts in R. It gives better mapping accuracy and reliable maps than the current application in RK. The new RK and its easy implementation in R software are anticipated to provide potential for wide application and eventual contribution to improved soil mapping and application of DSM.