Show description

Heuvelink (1998) summarised the concepts and main results on this subject. Although much work has been done in the last 20 years on the different functions that are involved in soil inference system, marginal progress has been made in building functions or chains of functions that better integrate soil surveyor knowledge and numerical procedures. The state of the art of such techniques is provided in Chapter 22. 3 How best to represent the results of digital soil mapping? The proposed addition of an inference component to the current soil information systems leads to revisit the problem of user interfaces.

G. USDA, 1951). Fridland (1972) raised the idea that the within-small-scale-mapping unit variations could be described by soil patterns for which he proposed a complete descriptive framework. g. , 1982; Butler, 1982) also attempted to break the nexus between pedon and mapping unit by suggesting the notion of a toposequence. , 1994). In spite of these attempts, descriptions of soil mapping units by representative profiles – and the underlying choropleth map model (Burrough, 1993) – have been most often preferred by the users of soil maps for the sake of simplicity, and perhaps visual perception.

Personally I like rule-induction methods because they can capture and represent the way people learn from observations in the field. A consequence of predictive modelling is that it generates an explicitly stated model of the distribution of soils in landscapes. , 1999; McKenzie and Ryan, 1999; Hewitt, 1993). Nevertheless, I admit that large tree models or artificial neural networks can be difficult to interpret. Quantitative models include a quantifiable level of uncertainty, in terms of model statistics.