Reliability of Singing Bird Surveys: Effects of Song Phenology during the Breeding Season

Abstract

Singing bird surveys are frequently used to estimate the difference in density between two or more populations of a species detected almost entirely by vocalizations. Typically, the statistical analysis assumes that individuals in the populations have the same average probability of being detected. We estimated how much difference in average detectabilities might occur owing to variation in singing rates during the breeding season. Presence or absence of song proved to be the best measure of detectability. Among 125 House Wrens (Troglodytes aedon) in central Ohio, the probability of singing at least once during a 3-min period was 0.7 before mating, 0.5-0.6 from mating to the completion of egg-laying, 0.7 during incubation, and 0.5 or less thereafter. These probabilities were not detectably affected by season (1 June-15 July), time of day (first 5 h of daylight), or number of other nearby males (1 to 6). Our analysis shows that random phenological differences, such as an earlier season in one year as compared to another, might cause an error of up to 25% in the estimate of relative density. If several years of data are available, then the estimates of relative density will usually be more accurate. Substantial errors may still occur if changes in song phenology are density-dependent. Several methods have been designed to overcome the problem of unequal detectabilities, but most of them require the assumption that all birds close to the observer are detected; our study shows this assumption to be seriously in error for House Wrens. The study indicates that, in most cases, singing bird surveys should not be relied upon to produce accurate estimates of relative density.