Reconciling statistical rigor and biological inference in models of butterfly occupancy

Use of occupancy models to characterize a species’ probability of occurrence does not require individuals to be captured or marked. However, application of occupancy models to many species within a butterfly assemblage rather than to a small number of target species is complicated by variation in phenology. We explored two methods for estimating detection probability: sampling each site multiple times on each date, and relaxing the closure assumption. We then explored the extent to which occupancy of butterflies in the Chesapeake Bay Lowlands, Great Basin, and southwestern Sierra Nevada could be explained on the basis of covariates including vegetation structure and composition, sugars in nectar sources, and topography. Detection probabilities of a majority of species were associated with abundance of nectar or mud. We found considerable temporal variation in whether individual covariates were associated with detection probability or occupancy of a given species. Vegetation measures associated with occupancy in the Chesapeake Bay Lowlands may reflect intensity of browsing by white-tailed deer. Elevation and precipitation were prominent in occupancy models for butterflies in the Great Basin. In the southwestern Sierra Nevada, we documented associations between occupancy and the number of inflorescences or sugar mass. Our work elucidated trade–offs of applying occupancy models to butterflies and highlighted ecological relations, especially the extent to which detection probabilities may relate to ephemeral resources.