DEMARCATION OF COMMUNITIES IN LARGE DATAB ASES

Different methods were tested in order to classify vegetation in large releve-based data sets. These methods were applied to 3907 releves of oatgrass meadows (Arrhenatherion) and Gentian-hairgrass swards (Mesob romion) in central Germany Four methods were used: minimum variance cl ustering (ORLOCI 1967), number of character species in the Braun-Blanq uet system, Twinspan (HILL 1979), and the species group method develop ed in Gottingen (BRUELHEIDE 1995). Although some details differed, all four methods yielded satisfactory assignment results for both allianc es. Differential species which make up the assignment criteria for the resulting vegetation units are obtained by the last two methods only. The species group method uses more species as differential criteria t han Twinspan, and therefore assigns only well-characterized stands to a vegetation unit. The species selected by Twinspan result in a higher degree of assignment but have the disadvantage that more releves are misclassified. As a result, the species group method is especially sui ted for processing large databases and for extracting unambiguous assi gnment criteria for various uses like mapping vegetation in the field.