Marine reserve networks are increasingly implemented to conserve biodiversity and enhance the persistence and resilience of exploited species and ecosystems. However, the efficacy of marine reserve networks in frequently disturbed systems, such as coral reefs, has rarely been evaluated. Here we analyze a well‐mixed larval pool model and a spatially explicit model based on a well‐documented coral trout (Plectropomus spp.) metapopulation in the Great Barrier Reef Marine Park, Australia, to determine the effects of marine reserve coverage and placement (in relation to larval connectivity and disturbance heterogeneity) on the temporal stability of fisheries yields and population biomass in environmentally disturbed systems. We show that marine reserves can contribute to stabilizing fishery yield while increasing metapopulation persistence, irrespective of whether reserves enhance or diminish average fishery yields. However, reserve placement and the level of larval connectivity among subpopulations were important factors affecting the stability and sustainability of fisheries and fish metapopulations. Protecting a mix of disturbed and non‐disturbed reefs, rather than focusing on the least‐disturbed habitats, was the most consistently beneficial approach across a range of dispersal and reserve coverage scenarios. Placing reserves only in non‐disturbed areas was the most beneficial for biomass enhancement, but had variable results for fisheries and could potentially destabilize yields in systems with well‐mixed larval or those that are moderately fished. We also found that focusing protection on highly disturbed areas could actually increase variability in yields and biomass, especially when degraded reef reserves were distant and poorly connected to the meta‐population. Our findings have implications for the design and implementation of reserve networks in the presence of stochastic, patchy environmental disturbances.

Correctly identifying the effects of a human impact on a system is a persistent challenge in ecology, driven partly by the variable nature of natural systems. This is particularly true in many marine fishery species, which frequently experience large temporal fluctuations in recruitment that produce interannual variations in populations. This variability complicates efforts to maintain stocks at management targets or detect the effects of rebuilding efforts. We address this challenge in the context of no‐take marine reserves by exploring how variable larval recruitment could interact with the timing of reserve establishment and choice of sampling design to affect population dynamics and the detectability of reserve effects. To predict population changes in the years following a no‐take reserve implementation, we first tested for periodicity in larval recruitment in an important U.S. Pacific coast recreational fishery species (kelp bass, Paralabrax clathratus) and then included that pattern in a population model. We also used this model to determine the detectability of population increases under alternative sampling approaches and minimum age sampled. Kelp bass larval recruitment in the Channel Islands, California, peaked every about six (major) and about two (minor) years. Our model showed that establishing a reserve during a peak or trough enhanced or delayed, respectively, the post‐reserve population increases. However, establishing a reserve during a recruitment peak could obscure a failing reserve, that is, a reserve that is unable to secure longer‐term metapopulation persistence. Recruitment peaks and troughs also interacted with sampling design to affect the detectability of reserve effects. Designs that compared inside‐outside were the most robust to variable recruitment, but failed to capture whether the reserve has improved metapopulation growth. Designs that included a time element (e.g., before‐after) are more suited to assessing reserve effectiveness, but were sensitive to recruitment variation and detectability can change year‐to‐year. Notably, detectability did not always increase monotonically with reserve age; the optimal time for detectability depended on the minimum age of organisms sampled and was greatest when the cohort of a major recruitment peak first appeared in the sampling. We encourage managers to account for variable recruitment when planning monitoring and assessment programs.