Abstract
Species communities vary as a result of many processes: filtering of species by environments, interspecific interactions, dispersal, and stochastic processes. Empiricists often try to evaluate their respective roles based on the traces they leave on the spatial structure of communities. Unfortunately these traces are often ambiguous. Temporal data (surveys repeated through time) convey historical information that may help resolve ambiguities. In this thesis, I explore and develop different methods to analyze spatio-temporal data in order to understand the dynamics of sets of communities linked by dispersal within fragmented landscapes (metacommunities). These methods are applied to three long-term datasets from freshwater metacommunities (snails in the Guadeloupe and Martinique islands and Daphnia spp. in Finland). I first analyze the metacommunity of Guadeloupe snails using both pattern-based approaches (using successive years as replicates) and joint species distribution models explicitly relating time t to time t-1. This approach documents how species interactions, filtering by local environmental conditions, and stochastic colonizations and extinctions shape spatio-temporal variation in community richness and composition. Second, I develop explicit multi-species metapopulation models to analyze two- or three-species systems (taken in Guadeloupe and Finland). The fitted parameters are used in simulations to analyze how competition and niche differences contribute to species coexistence. Remarkably, in both cases, although competition significantly limits species co-occurrence at a local scale, it is far for being strong enough to threaten coexistence at the landscape level. Regional exclusion seems to require uncommonly strong competition, in the presence of highly stochastic colonization-extinction dynamics. Such conditions (strong interactions and weak stochasticity) may characterize the Martinique dataset (a set of 11 clonal, related invasive taxa inhabiting permanent rivers). In this case, I develop an approach inspired by population genetics, inferring one- or two-dimensional competitive hierarchies among taxa and how they vary in space and time to explain exclusion and/or coexistence patterns. Overall, this work shows that the temporal dimension carries a lot of information on community processes acting both at the local and regional scales. An even more integrative view may, in the future, arise from metacommunity models explicitly coupling the dynamics of local and regional species abundances.