Ruderal communities of the railway station ceska trebova eastern bohemia czechoslovakia remarks on the application of classical and numerical methods of classification
Kovar, P.; Leps, J.
Preslia 58(2): 141-163
ISSN/ISBN: 0032-7786 Accession: 006358249
Zurich-Montpellier classification of a set of 61 phytosociological releves of ruderal plant communities was compared with results of different numerical treatments. Various agglomerative methods, the relocation method and one polythetic divisive method (TWINSPAN) were used, each with various transformations of primary data. Possible biases of classification due to differences in subjective estimation of species abundances were evaluated by introducing random errors into primary data. Particular classifications were mutually compared using the coefficient of Goodman et Kruskal (1954). A high degree of congruence characterized the group of classifications formed by the classical Zurich-Montpellier method, the Ward's clustering method and the relocation method, the latter two based on data after ordinal transformation. These three classifications appear to describe optimally the similarity structure of our phytosociological table. On the contrary, the poorest results were obtained using the presence-absence data only. Bias resulting from random perturbation of estimates of species abundances was negligible in comparison with differences between results of particular numerical treatments. Results of classifications were further compared with DCA ordination.