Insights into detection probability of forest bird surveys from comparison of in-person and passive acoustic monitoring point counts

Abstract Historically, research and monitoring of bird populations has been conducted with in-person point counts; however, passive acoustic monitoring (PAM) recordings transcribed by experts (“PAM point counts”) are rapidly replacing in-person surveys as an approach for counting birds. We reviewed the literature and used case-matched datasets from North America’s boreal forest to show that despite similarities in data structure, in-person and PAM point counts have fundamental differences in the detection process that lead to differences in detectability estimates from distance and removal sampling models. Observer effects on cue rate were more pronounced in PAM than in-person point counts. Cue rate estimates from PAM were significantly higher than from point counts due to earlier time of first detection in PAM point counts, the availability of data with higher-resolution time intervals in removal models, and the exclusion of visual detections. In contrast, exclusion of visual detections from point-count data estimates resulted in lower estimates of perceptibility. Cumulative lower values of detectability estimates from in-person point counts resulted in 15% higher density estimates on average when applied as statistical offsets. We suggest some of the differences in detectability estimates between survey methods are due to human error and/or failure of statistical assumptions and that availability estimates for species that are at least 90% aurally-detected should be preferentially derived from PAM data to reduce bias in density estimates for conservation applications and facilitate continued integration of historic and contemporary point-count datasets. Future research should focus on alternatives to removal modeling and understanding the detection process of automated classifiers to continue to improve detectability estimates and maximize the ability to integrate datasets across data types.