### 'Availability' bias : removal models of avian singing rates

Most birds that are detected during point-count surveys are singing males. The rate at which birds sing largely determines their **availability** for detection, and therefore their **detection rate**. BAM uses a removal model (Farnsworth et al. 2002) to estimate species-specific singing rates from the subset of surveys where the initial detection of each bird encountered is recorded relative to two or more time intervals (e.g., 0-3, 3-5, 5-10 min). The singing rate is the average rate at which individuals in the population of available birds produce detectable (auditory) clues per unit time interval (1 minute). The resulting singing rate is then transformed into the probability of a bird singing at least one time during the survey (singing probabilities). This allows BAM to account for the proportion of birds that were present but missed because they were not singing during the survey.

Singing probabilities can vary with important temporal aspects of sampling. For example, survey duration (the time spent counting birds at survey points) often varies from 3-10 min among projects. This difference results in a 70% average increase in the raw counts from 3 min to 10 min surveys. Similarly, songbird singing rates tend to be highest early in the morning during the dawn chorus and highest early in the breeding season when males are actively establishing territories and attracting mates.

BAM uses singing probabilities to account for sampling effects related to survey duration, the time of day, and the time of year. We modelled singing rates with different combinations of linear and quadratic effects of time of day and time of year. Results presented on this website are the best-supported model for each species.

This figure shows the probability of a bird singing at least one time relative to the length of the counting time (sampling duration) for 72 species of songbirds.

This page last updated: Sep 10 2012