The sensitivity of Ellenberg indicator values to the completeness of vegetation relevés

Ordination and calibration (indicator species analysis) are based on the taxonomic composition and richness of communities. How strongly does the performance of a widely used method like weighted averaging of Ellenberg indicator values depend on the completeness of samples? Based on a stratified random sample of coincident phytosociological relevés of forest understorey and environmental measurements (overstorey cover, elevation, moisture index, pH and C/N-ratio in topsoil) indicator values for light, temperature, moisture, soil acidity and nitrogen were tested. To simulate reduced sampling effort and uncomplete representation of plant diversity, the original compositional matrix was reduced by randomly deleting 1, 10, 20, 40 and 80% of species records with low abundance. The relationship between indicator values based on the full matrix and environmental variables was closest for temperature (R² = 0.53) and lowest for soil nitrogen (R² = 0.28). Deletion of low-abundance species records affected the correlations only weakly. With as little as 20% of the original species records (or 40% of gamma- and 20% of alpha-diversity) explained variance was still in a range of 0.26 to 0.37. The overall multivariate association between species composition and environment, as measured by Mantel statistics, was more strongly affected by the ommission of species records than the Ellenberg method. The relative resilience of Ellenberg indicator values to incomplete sampling is attributed to its predominant reliance on coarse structural information, i.e. the dominance pattern of relatively few plant species. Finely-tuned local indicator systems perform better by exploiting the idiosyncratic information of rare species. Choosing an optimal sampling scale involves a tradeoff between noise and completeness, an issue that has been largely neglected by environment-composition studies.