2. Collection data and ecological analyses
We collected a total of 1216.5 h of audio, of which 544.5 hs (45%) came from deployment 4 and 225.5 (19%), 201.0 (17%), and 245.5 hs (20%) came from deployments 1, 2, and 3, respectively (S3). The total number of recording hours per habitat type (899.0 hs in terra firme, 317.5 hs in floodplain) was roughly proportional to the number of site-deployment combinations in each habitat type (24 vs 13). We detected a total of 15,891 tinamou vocalization events, 2,189 of which were added after the second classification pass. Our detections represent nine of the 11 species present at LACC, with data densities ranging from 4,468 detections for C. strigulosus to 26 for T. tao (S3). Two species were not detected: C. atricapillus and C. obsoletus . Both species are uncommon at Los Amigos (eBird, 2017; personal obs.), are known to have affinities for brushy edge habitats that were located away from most of the recorders (Cabot et al., 2020; Anjos, 2006), and were entirely absent from the camera trap dataset. Therefore, we suspect that their lack of detection indicates true absence from the dataset rather than poor class performance. The relative occurrence frequency of the tinamou species as measured by our audio detection pipeline differs significantly from the observation frequencies reported by eBird (χ2 = 567.4, p < 2.2e-16, Figure 5b), but notably there was no significant difference between these frequencies and camera trap capture rates for the five species represented in both datasets (χ2 = 0.037102, p > 0.1, Figure 5a).