Biologists seek an understanding of the biological and environmental factors determining local community diversity. Recent advances in metacommunity ecology, and neutral theory in particular, highlight the importance of dispersal processes interacting with the spatial structure of a landscape for generating spatial patterns and maintaining biodiversity. The relative spatial isolation of a community is traditionally thought to have a large influence on local diversity. However, isolation remains an elusive concept to quantify, particularly in metacommunities with complex spatial structure. We represent the metacommunity as a network of local communities, and use network centrality measures to quantify the isolation of a local community. Using spatially explicit neutral theory, we examine how node position predicts variation in alpha diversity across a metacommunity. We find that diversity increases with node centrality in the network, but only when centrality is measured on a given scale in the network that widens with increasing dispersal rates and narrows with increasing evolutionary rates. More generally, complex biodiversity patterns form only when the underlying geography has structure on this critical scale. This provides a framework for understanding the influence of spatial geographic structure on global biodiversity patterns.