Restoring forest to hundreds of millions of hectares of degraded land has become a centerpiece of international plans to sequester carbon and conserve biodiversity. Forest landscape restoration will require scaling up ecological knowledge of secondary succession from small-scale field studies to predict forest recovery rates in heterogeneous landscapes. However, ecological field studies reveal widely divergent times to forest recovery, in part due to landscape features that are difficult to replicate in empirical studies. Seed rain can determine reforestation rate and depends on landscape features that are beyond the scale of most field studies. We develop mathematical models to quantify how landscape configuration affects seed rain and forest regrowth in degraded patches. The models show how landscape features can alter the successional trajectories of otherwise identical patches, thus providing insight into why some empirical studies reveal a strong effect of seed rain on secondary succession, while others do not. We show that seed rain will strongly limit reforestation rate when patches are near a threshold for arrested succession, when positive feedbacks between tree canopy cover and seed rain occur during early succession, and when directed dispersal leads to between-patch interactions. In contrast, seed rain has weak effects on reforestation rate over a wide range of conditions, including when landscape-scale seed availability is either very high or very low. Our modeling framework incorporates growth and survival parameters that are commonly estimated in field studies of reforestation. We demonstrate how mathematical models can inform forest landscape restoration by allowing land managers to predict where natural regeneration will be sufficient to restore tree cover. Translating quantitative forecasts into spatially targeted interventions for forest landscape restoration could support target goals of restoring millions of hectares of degraded land and help mitigate global climate change.