Using null model analysis of species co‐occurrences to deconstruct biodiversity patterns and select indicator species

Aim  Using total species richness to characterize biodiversity may mask multiple response patterns of species. We propose a null model analysis of species co‐occurrence‐based classification to identify sets of species that may have similar (within‐groups) and distinct (between groups) response patterns to their environment. The classification should also provide an explicit framework for selecting indicator species with characteristic co‐occurrence patterns to predict overall species richness. Location  Côte‐Nord, Québec, Canada. Methods  We combined null‐model of species co‐occurrence and cluster analysis to identify species groups within diverse assemblages of ground‐dwelling and flying beetles of stands in a boreal forest mosaic; we then examined their co‐occurrence and response patterns to habitat characteristics. Best subset regressions were used to select indicator species of richness within each group, from which indicators of total species richness were selected. Results  The identified species groups appeared to display contrasting co‐occurrence and response patterns to at least one of the stand‐level habitat characteristics. Among flying beetles, for example, richness increased with stand‐level heterogeneity for two groups and decreased for two other groups, but the relationship was non‐significant for the total richness. We identified 28 indicator species that explained > 80% (validated by bootstrap analysis) of the variation in total species richness. Predictive performance of indicators was higher than when their co‐occurrence were reshuffled, even under a highly constrained null model, indicating that co‐occurrence patterns contributed to their predictive performance. Main conclusions  Co‐occurrence‐based classification appears as a promising and effective tool for deconstructing biodiversity into species groups which reflect their ecological commonalities and differences, thus reducing the risk of making faulty inferences about the causes underlying overall diversity patterns. The method provides an explicit framework for selecting indicator species representing different species groups that may reflect the multiple responses of species co‐occurring with them. Indicator species can be effective for predicting overall species richness.