Bottom-up digital soil mapping. I. Soil layer classes

Abstract

Great advances have been made in quantitative soil classification and mapping in the past few decades, and these have enabled the development of continuous soil classes that are not necessarily mutually exclusive. This is a more natural a better representation of the nature of the soil continuum than the traditional mutually-exclusive soil classes. Attempts at continuous soil classification have–perhaps subconsciously–used bottom-up classification techniques that are based on the quantification of similarity between soil profiles. Approaches that use fuzzy clustering techniques are a good example of this. The intention of this work is to draw attention to continuous bottom-up soil classification in its own right, rather than just as a means to an end as has often been the case in previous digital soil mapping exercises. Our efforts are invested in developing a bottom-up soil classification system, and we begin in Part I of this paper by developing a set of continuous soil layer classes. We created a set of continuous soil classes based on measured soil properties and soil properties estimated using soil mid-infrared spectra. The soil layers came from a set of 262 soil profiles sampled from a study area in the lower Hunter Valley in New South Wales, Australia. A fuzzy k-means clustering algorithm was used to carry out the cluster analysis. The results of the cluster analysis suggested that 7 layer classes were optimal for our soil layers. Each observed soil layer was allocated to the continuous layer class to which it had the highest membership value. The confusion index for the soil layers tended to be low which gave confidence in the effectiveness of the continuous classification system.