Field Comparison of Removal and Modified Double-Observer Modeling for Estimating Detectability and Abundance of Birds

Abstract

Abstract For point-count data to reliably index bird abundance or density, estimates must be corrected for variation in detection probabilities across species, observers, and environmental conditions. Removal and double-observer modeling are two recently developed statistical techniques for estimating detection probabilities and bird abundance. We collected point-count data in north-central Indiana and used a Huggins closed-capture model in MARK to directly compare those two methods. We found that when detection probabilities were relatively high for individual observers, the two methods yielded similar estimates of density for nearly all 17 species modeled. However, when true detection probabilities for observers were relatively low, removal estimates of detectability and density were biased high and low, respectively, perhaps because of the effect of low-detection probability on the removal estimator or smaller sample sizes associated with less-skilled observers. In general, we consider removal modeling a more desirable approach than double-observer modeling because it requires half as many observers, allows more sources of variation in detectability to be modeled, and estimates abundance or density of the true population of birds. By contrast, double-observer modeling estimates only the abundance of the “apparent” population (i.e. those birds that are visually or audibly conspicuous). For species that vocalize infrequently or are otherwise elusive, the apparent population may be significantly smaller than the true population. However, double-observer modeling is more robust to violations of the assumption of population closure and may outperform removal methods when data are collected by less-experienced observers.