Multivariate perspectives on patch use by masked bobwhites

Models that discriminate habitat patches acceptable to wildlife assist biol ogists in managing habitat, evaluating the effects of management treatments , and selecting areas for development or preservation. We used neural netwo rk modeling to discriminate between used and random patches for the endange red masked bobwhite (Colinus virginianus ridgwayi) in Sonora, Mexico, and A rizona during 1994-96. Input variables, thought to encompass the habitat sp ace of bobwhites, were canopy coverage of woody vegetation (%). exposure of bare ground (%), exposure to ground predators (m(2)), exposure to aerial p redators (m(3)), and operative temperature (degreesC). A neural model devel oped with data from Mexico correctly classified 87.4% of patches for traini ng (n = 483) and validation data (n = 118). The model developed for Arizona correctly classified 82.3% of patches for training data (n = 265) and 78.1 % for validation data (n = 64). Mathematical transplants of Mexico bobwhite s to Arizona habitat and of Arizona bobwhites to Mexico habitat revealed th at bobwhites from Mexico (native) were adapted to a broader range of condit ions than those in Arizona (reintroduced). For masked bobwhites and probabl y other species, the contingent nature of habitat features in a multivariat e sense may permit the redress of a habitat deficiency without addressing t he perceived deficiency per se.