Planning an Adaptive Numerical Classification

Abstract

Although the application of numerical classification methods is becoming common, there remains considerable confusion over the value of the results obtained. Thus in comparing numerical methods with the tabular sorting of the Braun-Blanquet system, Stanek (1973), Adam et al. (1975) and Kortekaas, van der Maarel & Beeftink (1976) stress the similarities of the results obtained with preference for the numerical approach, whereas Moore & O’Sullivan (1970), Moore et al. (1970), and Coetzee & Werger (1973, 1975) stress discrepancies and prefer the traditional approach. Such differences might simply reflect differences in the data used, as noted by Hogeweg (1976) but, in my view, there is a more fundamental reason for the difference. Present numerical methods rigidly formalise one part of the analytic process; specifically that concerned with the organisation of floristic data using intrinsic criteria. In contrast, the human classifier is considerably more flexible, primarily because he works in a wider context and can, therefore, choose between alternative procedures during the course of his analysis. As Mackay (1969) phrases it, patterns are for agents and the human agent can select, reject and reconsider patterns as he chooses. In this paper I want to examine some few means of similarly increasing the flexibility of numerical methods, although I shall also indicate in passing a larger number of alternatives which might be fruitful.

Contribution to the Symposium on Plant species and plant communities, held at Nijmegen, 11–12 November 1976, on the occasion of the 60th birthday of Professor Victor Westhoff.