Inference methods for spatial variation in species richness and community composition when not all species are detected

Inferences about spatial variation in species richness and community compos ition are important both to ecological hypotheses about the structure and f unction of communities and to community-level conservation and management. Fete sampling programs for animal communities provide censuses, and usually some species in surveyed areas are not defected. Thus, counts of species d etected underestimate the number of species present. We present estimators useful for drawing inferences about comparative species richness and compos ition between different sampling locations when not all species are detecte d in sampling efforts. Based on capture-recapture models using the robust d esign, our methods estimate relative species richness, proportion of specie s in one location that are also found in another, and number of species fou nd in one location but not in another The methods use data on the presence or absence of each species at different sampling occasions (or locations) t o estimate the number of species not detected nt any occasions (or location s). This approach permits estimation of the number of species in the sample d community and in subsets of the community useful for estimating the fract ion of species shared by two communities. We provide an illustration of our estimation methods by comparing bird species richness and composition in t wo locations sampled by routes of the North American Breeding Bird Survey. In this example analysis, the two locations (and associated bird communitie s) represented different levels of urbanization. Estimates of relative rich ness, proportion of shared species, and number of species present on one ro ute but not the other indicated that the route with the smaller fraction of urban area had greater richness and a larger number of species that were n ot found on the more urban route than vice versa. We developed a software p ack age, COMDYN, for computing estimates based on these methods. Because th ese estimation methods explicitly deal with sampling in which not all speci es are detected, we recommend their use for addressing questions about spec ies richness and community composition.