Metacommunity structure reflects the interplay of various processes, including niche filtering, extinction/colonization and interspecific interactions. Spatial patterns of species distributions are often analyzed to infer these processes. However, such inferences rely on often unrealistic equilibrium assumptions, and remain ambiguous, as different processes can produce similar patterns. Temporal data may improve these inferences. For example, stochastic species turnover may occur in local communities, while, on the long run, temporal changes are kept within limits set by locally available niches. Our objective is to explore how the joint analysis of spatial and temporal patterns can clarify the contribution of different processes to metacommunity structure. We recorded the occurrences of 21 freshwater mollusc species, and environmental data, in 250 sites over 17 successive years in a network of ponds in Guadeloupe (Lesser Antilles). We analyzed variation in α and β-diversities in space and time, and used a joint-species distribution mode to characterize species–environment and species–species relationships. Local communities showed pronounced temporal variation reflecting both imperfect species detection and true stochastic species turnover. On the long term however, local communities were largely controlled by niche filtering along two main environmental gradients, one driven by site connectivity, the other by hydrological stability and aquatic vegetation. Two gastropod clades, caenogastropods and pulmonates, showed contrasted spatio–temporal distributions resulting from different responses to these gradients, and these distributions seemed little altered by interspecific competition. Our study illustrates the benefit of using spatiotemporal metacommunity data to discern long-term impacts of niche filtering and species interactions behind short-term stochasticity.