Inferences based on counts adjusted for detectability represent a marked improvement over unadjusted counts, which provide no information about true population density and rely on untestable and unrealistic assumptions about constant detectability for inferring differences in density over time or space. Distance sampling is a widely used method to estimate detectability and therefore density. In the standard method, we model the probability of detecting a bird as a function of distance alone. Here, we describe methods that allow us to model probability of detection as a function of additional covariates—an approach available in DISTANCE, version 5.0 (Thomas et al. 2005) but still not widely applied. The main use of these methods is to increase the reliability of density estimates made on subsets of the whole data (e.g., estimates for different habitats, treatments, periods, or species), to increase precision of density estimates or to allow inferences about the covariates themselves. We present a case study of the use of multiple covariates in an analysis of a point-transect survey of Hawaii Amakihi (Hemignathus virens).
Amélioration des estimations de densité d’oiseaux par l’utilisation de l’échantillonnage par la distance avec covariables multiples