Choices of abundance currency, community definition and diversity metric control the predictive power of macroecological models of biodiversity.

Aim This study focuses on the influence of methodological choices on the predictive performance of macroecological models ( MEMs), i.e. statistical models designed to predict patterns of biodiversity using environmental predictors. We emphasize the influence of three methodological choices: (1) the choice of the currency in which the abundance of each species is measured, i.e. numbers of individuals or biomass; (2) the rules used to define the species assemblage under focus, i.e. broad communities or refined ecological guilds; and (3) the influence of rare over common species in the biodiversity measure. Location The effects of these choices are investigated using an exhaustive dataset on the fish fauna of the continental shelf of the Barents Sea. Methods We conducted an analysis of 220 models resulting from all possible combinations of the three methodological choices. For each, we evaluated the predictive performance through an iterative cross-validation process. Results All methodological choices we investigated strongly affected the predictive performance of MEMs. High predictive performances were obtained when using biomass instead of numbers of individuals, when focusing on narrow ecological guilds composed of species sharing the same ecological traits and when using diversity measures that give high weight to rare species. Main conclusions We recommend that future projections of biodiversity pay more attention to abundance currency, ecological homogeneity of focal species assemblages and the diversity metric used, and systematically investigate these methodological choices prior to producing biodiversity forecasts. Splitting a whole set of species into ecological guilds appears to be a promising practice, leading to a selected set of MEMs with high predictive performances and more detailed forecasts on the fate of diversity. [ABSTRACT FROM AUTHOR]