Breeding bird atlases providing distribution data at a regional scale are becoming increasingly common. To assess the ability of such data to develop bread-scale bird-habitat models, we used data from a breeding bird atlas and landscape variables obtained From a geographic information system (GIS) to study the distribution of seven woodpecker species in the Jura, France: the Black (Dryocopus martins), Green (Picus viridis). Grey-headed (P. canus), Great Spotted (Dendrocopos major), Middle Spotted (D. medius), and Lesser Spotted (D. minor) Woodpeckers, and the Wryneck (Jynx torquilla). We used logistic regression to develop predictive models from variables that described each 575-ha atlas cell in terms of forest composition, forest class richness, edge density, and elevation. For all seven species, prediction rates were better than chance; however, improvements over chance classification varied from 14-39%, indicating that predictive ability was species-specific. From our study, we identified limitations inherent to working with gridded data, including grid positioning problems and inability to compute spatial variables. In spite of these limitations, our models could be used for simulations, to improve the atlas itself, and to identify potential suitable habitat.
© 1999, University of California Press. View original published article in JSTOR.