Refining predictions of metacommunity dynamics by modelling species non-independence.

Predicting the dynamics of biotic communities is difficult because species' environmental responses are not independent, but covary due to shared or contrasting ecological strategies and the influence of species interactions. We used latent-variable joint species distribution models to analyse paired historical and contemporary inventories of 585 vascular plant species on 471 islands in the south-west Finnish archipelago. Larger, more heterogeneous islands were characterized by higher colonisation rates and lower extinction rates. Ecological and taxonomical species groups explained small but detectable proportions of variance in species' environmental responses. To assess the potential influence of species interactions on community dynamics, we estimated species associations as species-to-species residual correlations for historical occurrences, for colonisations, and for extinctions. Historical species associations could to some extent predict joint colonisation patterns, but the overall estimated influence of species associations on community dynamics was weak. These results illustrate the benefits of considering metacommunity dynamics within a joint framework, but also suggest that any influence of species interactions on community dynamics may be hard to detect from observational data.