FUZZY CLUSTERING OF ECOLOGICAL DATA

Fuzzy sets are a suitable description of ecological communities, using Calluna vulgaris moorland data from the North York Moors National Park as an example. The clustering approach used, the fuzzy c-means algorithm (or fuzzy ISODATA), requires a starting classification that is refined by a least-squares criterion. The starting strategy used in the example was based on the division of one ordination axis. Ordination axes were also used to reduce the high level of noise present in the ecological data. Fuzzy clustering was compared with the classification produced by TWINSPAN. The first division of TWINSPAN contrasted species of wet environments (Agrostis canina, Sphagnum recurvum, Polytrichum commune and Eriophorum angustifolium) with Pohlia nutans, a moss which grows on peaty or sandy banks and often under mature Calluna vulgaris. Fuzzy c-means show a strong clustering into 2 groups, although there were indications of a substructure at 3 or perhaps 5 groups. The 2 groups contrasted sites from wet habitats and those from drier habitats. When clustering into ≥3 groups, the clusters of drier habitats may be associated with the C. vulgaris development cycle. In comparison with TWINSPAN, fuzzy clustering produces clusters which are more strongly correlated with relevant external environmental variables. This may reflect the fact that ecological communities are more similar to fuzzy sets than to ordinary sets with sharp boundaries. -from Author