Optimal transformation of species cover for vegetation classification

Abstract

AbstractAimsVegetation‐plot sampling usually involves estimating species cover. For classifying plots to vegetation types, covers are often transformed to decrease the effect of dominant species. However, it remains unclear which transformation is optimal. We suggest that for vegetation classification, optimal is such transformation that contributes to creating clusters of plots in an unsupervised classification that are most similar to the widely accepted vegetation types, e.g., phytosociological associations. Here our aim is to find and recommend such optimal transformation by testing a range of transformation options against the national vegetation classifications of three European countries.LocationCzech Republic, The Netherlands, Great Britain.MethodsThree national datasets of vegetation plots with species cover information, classified to associations or community types of the respective national vegetation classification systems, were analysed. From each dataset, multiple subsets of plots were selected randomly, each subset representing a vegetation‐plot table containing several similar associations/community types. Species cover values in these subsets were subjected to various transformations (power transformation, logarithmic transformation and pseudo‐species cut levels). Then each subset was classified by an agglomerative classification method (beta‐flexible clustering with different beta values), and the classification was compared with the units of the national vegetation classification using the adjusted Rand index.ResultsPower transformations of percentage covers with an exponent between 0.3 and 0.6 produced the best match between the unsupervised classifications and the national vegetation classifications. This result did not depend on the classification method used. A similar degree of matching was achieved with some cut levels of pseudo‐species and with logarithmic transformation of percentage cover.ConclusionsIf an unsupervised classification of vegetation plots aims at defining vegetation types that are close to the phytosociological associations accepted in national vegetation classifications, the best transformation is close to the square‐root of percentage cover (i.e., power transformation with exponent 0.5).