A new approach to functional interpretation of vegetation data

In this paper we present a new approach to the simultaneous analysis of a species composition data set, an environmental gradient data set and a functional attribute data set. We demonstrate its advantages in terms of statistical modelling including model development and assessment as well as subsequent prediction. Our method is applied to a set of case data deriving from experimental wetland microcosms including 20 species, 12 treatment combinations and a classification of species into functional groups. Acknowledging that lack of independence between samples and over-interpretation of data may lead to overly optimistic assessment of model performance. we use cross-validation with different subsets of data to obtain realistic model performance measures. It is shown that although the outcome of the wetland experiment is predictable in terms of experimental treatments and taxonomic species. the functional groups cannot be used to explain the variation in species frequencies in the experiment. We compare the method with recently published approaches to the functional analysis of vegetation data, and discuss its applied perspectives.