How do locally infrequent species influence numerical classification? A simulation study

Phytosociological databases are important data sources for a broad scale of ecological investigations. Vegetation samples are traditionally managed and published in tabular format, allowing for handling of the vegetation data in various combinations. Such tables usually comprise relevés originated from the same locality, vegetation type and collected by the same investigator. Nevertheless, these relevés are usually affected by the same bias. In this paper, we demonstrate the importance of the effects acting at the level of the table (i.e., ‘locally’), using the example of species removals from groups of relevés. We examine the effect of the removal of infrequent species on community classification in relation with several data set properties using simulated plot data sampled from simulated coenoclines. A data set comprised groups of relevés (‘tables’), within which relevés are sampled from the same point of the coenocline. Classifications obtained after the removal or permutation of infrequent species occurrences from these tables, after the removal of rare species from randomised tables and without any treatment were compared to a reference classification based on gradient positions of the relevés. The results show that the removal of locally infrequent species helps to recognise the gradient pattern incorporated in the tabular arrangement of relevés if the arrangement of relevés among tables is in accordance with their gradient position. In cases when the grouping of relevés is irrelevant regarding the real underlying pattern, the species removal is disadvantageous. Testing between-table heterogeneity within a data set is an especially successful way of examination of biological relevance of the arrangement of relevés. We conclude that influence of table-level effects is mainly dependent on the pattern which is in accordance with the grouping of plots.