The influence of transformations of an ordinal scale of a floristic gradient, applied on a TWINSPAN classification

Abstract

Summary TWINSPAN is a widely applied method for hierarchical classification. Particular options such as the minimum group size, the maximum split levels and both the amount and score of cut levels to be used are left to the judgement of the researcher. When using small and homogeneous data sets, the score assigned to the cut levels can be derived from intermediate transformations of the cover-abundance values, rather than transformations obtained with a strong weighting on presence, or an emphasis on dominance. Our research question is to know whether these coefficients have the same robustness for more heterogeneous and larger data sets, with a large amount of transition communities. A comparison was made between five different sets of cut levels in order to determine the most effective one for the classification of data on a floristic gradient. The analysed cut level sets differed in the number of cut levels and in the weight provided to the cover-abundance coefficients. The classifications resulting from the different sets of cut levels were compared, by determining the amount of identical associations that were clustered together, by visualising the results of the different cut level types through DCA, by comparing the correlations between the median values of the abiotic factors in each TWINSPAN group and DCA-values of the first axis of the different cut level types. The analysis showed that the cut level types in particular derived from the original values of the Braun-Blanquet cover-abundance coefficients, reflect best the floristic gradient and the underlying structure of the data. Results on heterogeneous data sets are complementary to the previous findings in literature where better results were achieved with a combined transformation for more homogeneous data sets.