The study of fidelity measures in the context of using them as a threshold criterion in the allocation of diagnostic species

Abstract

The paper is devoted to a comparative study of species fidelity measures from the viewpoint of using them as threshold criteria for distinguishing diagnostic species in the vegetation classification. The computational experiment was carried out on two test data sets on vegetation of the Dnieper floodplain in the forest-steppe zone of Ukraine (316 relevés) and forest vegetation of the city of Kyiv and suburbs (832 relevés). Classification of relevés was obtained using cluster analysis by the flexible-beta method on the Bray-Curtis distance matrix and the optimal number of clusters was determined using the Optimclass approach. In all fidelity calculations, classification of relevés has remained constant for comparability reasons. We compared the distribution properties of fidelity values using the most common fidelity indices, including phi-coefficient, IndVal, Ochiai index, chi-square statistic, Bruelheide's u, Fisher test, G-statistic, and TCR. Group-equalized and non-equalized modifications of fidelity indices were considered as separate measures. Each species was accounted in the cluster with the highest fidelity value. Such indicators were estimated as range and quartiles of values, evenness of distribution, variation of shares of diagnostic species with a variation of fidelity thresholds and vice versa, shares of species exceeding fidelity threshold in more than one cluster.