The influence of data stratification on vegetation classification

KUZELOVA I. & TICHY L.

Department of Botany, Masaryk University, Kotlarska 2, CZ-611 37 Brno, Czech Republic
Email: ikuzel@sci.muni.cz, tichy@sci.muni.cz

Processing of large phytosociological data sets for vegetation classification can be unduly influenced by bias inherent to large databases. Even in European countries with long phytosociological tradition, available relevés have irregular geographical distribution; some areas are over-sampled and there are some areas with no relevés available. In addition, vegetation types differ in their proportional representation in the databases; this fact may be caused either by rarity or commonness in the area or by preferential sampling of certain vegetation types by phytosociologists. Such disproportions can influence classification results. Therefore we used various types of stratified data sets, selected from a large phytosociological database, and compared resulting classifications.

Relevés of semi-natural meadows from the Czech National Phytosociological Database were used for compilation of eight variously stratified data sets: (1) In geographical stratification, one relevé was chosen from each square of a geographical grid; (2) In numerical stratification, groups of relevés were created by the cluster analysis and then an a priori defined number of relevés per cluster was randomly selected; (3) In association-based stratification, assignment to phytosociological associations according to original papers was used to classify the data set and then an a priori defined number of relevés per each association was used; (4) In diagnostic-species stratification, we used a newly developed method for exclusion of relevés of transitional vegetation types, which were defined by absence of species with narrow ecological ranges. Finally we compiled data sets using combinations of the above stratifications. These data sets were classified by the cluster analysis and applicability of particular types of stratification was evaluated.