Abstract

The numerical classification of vegetation is frequently faced with the problem of sparse data matrices and the low occurrence of many different species of plant in only a few of the sampling sites. Gower's maximal predictive classification is a reticulate classification that defines the optimum class number using a data set where absence is common. This classification was applied to a data set from Shapwick Heath in Somerset. The classification was very successful and seven vegetation classes were clearly defined. Interpretation was simplified as the statistically more important species for each class were listed in the program output. The seven classes were noted to respond well to the stages in the vegetation succession present on the site.