Optimally Partitioning Soil Transects

Summary Two methods are described for locating soil boundaries on transects from multivariate data. One method (split moving‐window, SMW) examines a transect through a window which is split at its mid point and the two halves compared. The window is moved along the transect and a boundary inferred wherever the difference between the two halves attains a local maximum. The other (maximum level‐variance, MLV) examines all possible partitions of a transect that can be produced by a given number of boundaries or less and places boundaries in those positions that minimize the within‐segment sum of squared deviations from the segment means (and hence maximize the between‐segment sum of squares). The performances of the two methods are compared for two soil transects in Oxfordshire. MLV identified all major changes even where gradual, but was less reliable for small changes. SMW identified all sharp boundaries, even between soil types that are fairly similar. In its original form SMW did not identify gradual changes reliably, but its performance was improved both by omitting data near the centre of the window and by concentrating the discriminating power of the data beforehand as in MLV. SWM requires less computing than MLV.