COS 100-5 - Can You Hear Me Now? Autonomous Recording Units Yield Improved Occupancy Models compared to Point Counts for Declining Forest Bird Species

Abstract: Autonomous recording units (ARUs) are increasingly used to monitor declining or at-risk animal populations, especially at large spatiotemporal scales. Although habitat restoration successes for birds are typically evaluated using traditional point count surveys combined with occupancy modeling, point counts are costly and time consuming when employed for large scale projects. ARUs are a promising alternative to traditional surveys because they are relatively cheap, can record the entire community, and provide recordings that can identify species presence using automated classifiers. Here we investigated the differences between occupancy models fit to detection histories generated from ARUs and point count surveys for two species of eastern forest birds: Cerulean Warbler (CERW, Setophaga cerulea) and Wood Thrush (WOTH, Hylocichla mustelina). We deployed AudioMoth acoustic recorders and performed two point counts within six study sites containing 669 sampling points from May 15th to June 30th, 2020 and 2021 in the central Appalachian region of Pennsylvania. To detect species within acoustic recordings, we created a convolutional neural network automated birdsong classifier and used it to score recorded clips. We manually listened to the top scoring clip from each point across 10 days to verify song presence and create a daily detection history from the ARU recordings. We generated and compared single-season occupancy models using detection histories from both ARUs and point counts. WOTH were detected at 384/669 (57%) of points using ARUs and 318/669 (48%) of points using point counts. In the ARU models for WOTH, tall ( >1.5m high) stem density and basal area were both statistically significant predictors of WOTH presence. In the point count models for WOTH, no habitat covariates were statistically significant predictors of occupancy. Additionally, the point count data were unable to precisely estimate baseline occupancy at 3 of our 6 sites, while the ARUs had no similar limitation. At one site where CERW were found in sufficient numbers for modeling, CERW were detected at 33/153 (22%) of points using ARUs and 18/153 (12%) of points using point counts. No occupancy covariates for CERW were significant in either the ARU or point count models. Overall, we find that point counts failed to detect statistically significant habitat relationships that were detected using ARU surveys and models. The extended detection histories available from ARUs deployed over large spatial scales may be able to detect relationships between species occupancy and habitat that are otherwise hidden by point counts.