Identifying the drivers of pond biodiversity: The agony of model selection

Ponds contribute substantially to the maintenance of regional biodiversity. Despite a growing body of literature on biotic-abiotic relationships in ponds, only few generalizations have been made. The difficulty in identifying the main drivers of pond biodiversity has been typically attributed to the heterogeneity of the local and regional conditions characterizing ponds. However, little is known on how the use of different analytical approaches and community response variables affects the results of analysis of community patterns in ponds. Here, we used a range of methods to model the response of water beetle and plant community data (species richness and composition) to a set of 12 environmental and management variables in 45 farmland ponds. The strength of biotic-abiotic relationships and the contribution of each variable to the overall explained variance in the reduced models varied substantially, for both plants and beetles, depending on the method used to analyze the data. Models of species richness included a lower number of variables and explained a larger amount of variation compared to models of species composition, reflecting the higher complexity characterizing multispecies response matrices. Only two variables were never selected by any of the model, indicative of the heterogeneity characterizing pond ecosystems, while some models failed to select important variables. Based on our findings, we recommend the use of multiple modeling approaches when attempting to identify the principal determinants of biodiversity for each response variable, including at least a non-parametric approach, as well as the use of both species richness and composition as the response variables. The results of this modeling exercise are discussed in relation to their practical use in the formulation of conservation strategies