MULTIPLE DISCRIMINANT ANALYSIS IN SOIL SURVEY

Summary Multiple discriminant analysis can be used to distinguish to best advantage several multivariate, more or less polythetic, classes of soil. The method allows profiles to be allocated consistently and most appropriately to existing groups for which representatives have already been chosen. The method is described in geometric terms and is applied to the allocation of profiles in both well known and little known areas, South Central England and Sabah. For the areas in England good separation between soil classes was demonstrated; in Sabah the classes of the reconnaissance survey were less distinct. Representative profiles were chosen for all existing classes and the remainder of the profiles allocated using discriminant functions. In all cases a better classification resulted in that there was less dispersion within classes and hence better separation between them. Provided conformant data can be collected in routine survey the method promises to allow a good working combination of man and machine: the soil surveyor to decide the classes desired, and the computer to handle large quantities of data and make routine allocation.