Testing the performance of beta diversity measures based on incidence data: the robustness to undersampling

Aim Researchers measuring beta diversity have rarely concerned themselves with the problems of how complete the species lists of studied communities are, and of how the varying degrees of completeness can actually change estimates of beta diversity. No comprehensive assessment has been made regarding the behaviour of most beta diversity indices when applied to incomplete samples, a situation which is more common than usually recognized. Our objective was to assess the behaviour and robustness of a number of beta diversity measures for incidence data from undersampled communities. Location Mainland Portugal and the Azorean archipelago (North Atlantic). Methods Data from intensive sampling of spiders in mainland Portugal and arthropods in Azores were collected. We examined the properties of 15 beta diversity measures developed for incidence data. We simulated varying degrees of completeness, whereas computing beta diversity for selected pairs of samples. The robustness of these beta diversity accumulation curves was assessed for the purpose of finding the best measures for undersampled communities. Results The Harrison et al.beta(-2) and the Williams beta(-3) are particularly robust to undersampling. These measures are also insensitive to differences of alpha diversity (species richness) between communities, and therefore to nestedness. Colwell & Coddington beta(cc) and the related Jaccard beta(j) and Gaston et al.beta(g) performed best of the measures sensitive to alpha diversity. They performed poorly, however, when compared communities exhibited very low values of beta diversity. In such cases, the Routledge beta(r) performed the best. Main conclusions No index was found to perform without bias in all circumstances. Overall, beta(-2), beta(-3) and beta(cc) (or related measures beta(j) and beta(g)) are recommended as they seem to be the most robust to undersampling.