FORMING SPATIALLY COHERENT REGIONS BY CLASSIFICATION OF MULTI-VARIATEDATA - AN EXAMPLE FROM THE ANALYSIS OF MAPS OF CROP YIELD

A method is proposed for the generation, from multi-variate data, of c lasses with a spatially coherent distribution. The procedure is more a ppropriate for large data sets than is spatially constrained clusterin g based on modification of a similarity matrix. It is based on fuzzy c lustering of the data, followed by spatially weighted averaging of the class memberships within a local neighbourhood. A coherence index is proposed which may be used to select an appropriate neighbourhood for this smoothing procedure. The effects of smoothing on the uniformity o f the classes with respect to the variables used to define them may be measured by Wilks's criterion, and examination of the class means. Th e method is demonstrated on a multi-variate data set.