Improving a Reconnaissance Soil Classification by Multivariate Methods

Summary An initial reconnaissance classification was difficult to use in soil survey in south‐west Sabah, East Malaysia, because the most commonly occurring classes of soil (subdivisions of Orthic Acrisols in the FAO/UNESCO Soil Map of the World Legend) could not be distinguished consistently. Principal component, similarity, and canonical variate analyses showed that the classes that were originally thought to be desirable overlapped almost completely with respect to profile characters and chemical properties determined by routine analysis. New guidelines for classification were devised, and class overlap was minimized by reallocating profiles using multiple discriminant analysis. The final classification of the Orthic Acrisols represented a gradation from profiles containing fresh rock fragments and weatherable minerals to those containing none. Some of the last occurred on strongly pre‐weathered sandstone; elsewhere they provided the main evidence for the remnants of a Pliocene land surface. To use the new classification consistently, estimates of total magnesium, potassium and phosphorus had to be included in routine chemical analyses. The revised classification was used successfully in subsequent reconnaissance survey work in south‐west Sabah.