A central challenge of conservation biology is using limited data to predict rare species occurrence and identify conservation areas that play a disproportionate role in regional persistence. Where species occupy discrete patches in a landscape, such predictions require data about environmental quality of individual patches and the connectivity among high quality patches. We present a novel extension to species occupancy modeling that blends traditional predictions of individual patch environmental quality with network analysis to estimate connectivity characteristics using limited survey data. We demonstrate this approach using environmental and geospatial attributes to predict observed occupancy patterns of the Yosemite toad (Anaxyrus (= Bufo) canorus) across >2,500 meadows in Yosemite National Park (USA). A. canorus, a Federal Proposed Species, breeds in shallow water associated with meadows. Our generalized linear model (GLM) accurately predicted 84% of true presence-absence data on a subset of data withheld for testing. The predicted environmental quality of each meadow was iteratively ‘boosted’ by the quality of neighbors within dispersal distance. We used this park-wide meadow connectivity network to estimate the relative influence of an individual Meadow’s ‘environmental quality’ versus its ‘network quality’ to predict: a) clusters of high quality breeding meadows potentially linked by dispersal, b) breeding meadows with high environmental quality that are isolated from other such meadows, c) breeding meadows with lower environmental quality where long-term persistence may critically depend on the network neighborhood, and d) breeding meadows with the biggest impact on park-wide breeding patterns. Combined with targeted data on dispersal, genetics, disease, and other potential stressors, these results can guide designation of core conservation areas for A. canorus in Yosemite National Park.