Marine reserves hold promise for maintaining biodiversity and sustainable fishery management, but studies supporting them have not addressed a crucial aspect of sustainability: the reduction in viability of populations with planktonic larvae dispersing along a coastal habitat with noncontiguous marine reserves. We show how sustainability depends on the fraction of natural larval settlement (FNLS) remaining after reserves are implemented, which in turn depends on reserve configuration and larval dispersal distance. Sustainability requires FNLS to be greater than an empir‐ically determined minimum. Maintaining an adequate value for all species requires either a large, unlikely fraction (> 35%) of coastline in reserves, or reserves that are larger than the mean larval dispersal distance of the target species. FNLS is greater for species dispersing shorter distances, which implies reserves can lead to: (1) changes in community composition and (2) genetic selection for shorter dispersal distance. Dependence of sustainability on dispersal distance is a new source of uncertainty.

Habitat destruction is a critical factor that affects persistence in several taxa, including Pacific salmon. Salmon are noted for their ability to home to their natal streams for reproduction. Since straying (i.e., spawners reproducing in nonnatal streams) is typically low in salmon, its effects have not been appreciated. In this article, we develop both a general analytical model and a simple simulation model describing structured metapopulations to study how weak connections between subpopulations affect the ability of a species to tolerate habitat destruction and/or declines in habitat quality. Our goals are to develop general principles and to relate these principles to salmon population dynamics. The analytical model describes the dynamics of two density-dependent subpopulations, connected by dispersal, whose growth rates fluctuate in response to environmental and demographic stochasticity. We find that, for moderate levels of environmental variability, small dispersal rates can significantly increase mean extinction times. This effect declines with increasing habitat quality, increasing temporal correlation, and increasing spatial correlation, but it is still significant for realistic parameter values. The simulation model shows there is a threshold rate of dispersal that minimizes extinction probabilities. These results cannot be seen in classical metapopulation models and provide new insights into the rescue effect.