Multi-observer passive acoustic recordings, analyst annotations, and automated detections for Antarctic blue and fin whale calls, Casey Station 2019

Name: Multi-observer passive acoustic recordings, analyst annotations, and automated detections for Antarctic blue and fin whale calls, Casey Station 2019
Published: 2026-04-29
License: These data are publicly available for download from the provided URL. No formal restrictions beyond the CCBY Attribution License apply. These data are under active use for ongoing analyses; we welcome collaboration and ask that you contact the data originator before beginning work that may duplicate or closely relate to analyses already underway. | Attribution 4.0 International (CC BY 4.0)

Published: 2026-04-29

This dataset supports a multi-observer capture-recapture analysis of passive acoustic detection of Antarctic blue whale (Balaenoptera musculus intermedia) ABZ-unit calls at Casey Station, East Antarctica, during 2019. ABZ-unit calls are low-frequency (~25–29 Hz) repeated tonal pulses forming part of the Antarctic blue whale's characteristic three-part song (A-, B-, and Z-units). It is associated with the publication: Miller, B.S. et al. (in press). Common ground: efficient, consistent, observer-independent bioacoustic call density estimation with adjudicated ground truth and capture-recapture detection functions. Methods in Ecology and Evolution. doi:[to be assigned]. The dataset consists of three components: Acoustic recordings. Approximately 200 hours of underwater acoustic recordings collected at Casey Station (63°48'S, 111°47'E) between 23 December 2018 and 13 December 2019, downsampled to 1000 Hz. These recordings are a subset of the full Casey 2019 dataset available in the related archive: (Miller, B.S., Milnes, M. and Whiteside, S. 2025. Long-term underwater acoustic recordings in the Southern Ocean 2013–2024, Ver. 9, Australian Antarctic Data Centre. doi:10.26179/fhsv-ft93), selected following the AAD Blue and Fin Annotated Acoustic Library (AAD-BAFAAL) protocol: approximately 2% of total recording effort, distributed approximately evenly throughout the year to be broadly representative of annual recording conditions. Recordings were made using an AAD Moored Acoustic Recorder (MAR) deployed at approximately 2700 m depth, equipped with a hydrophone with sensitivity –165.9 dB re 1 V/µPa and an analogue front-end with frequency-dependent gain (see accompanying metaDataCasey2019.m for full calibration). Recordings are in 16-bit WAV format with timestamps encoded in the filename (yyyy-mm-dd_HH-MM-SS.wav). Raven Pro annotations. Time-frequency bounding box annotations made independently by three human analysts (Analysts 1–3, identities anonymised) using Raven Pro 1.5 (Cornell Lab of Ornithology, Ithaca NY). Annotations cover the following call types, consistent with the AAD-BAFAAL protocol: Bm-Ant-A, Bm-Ant-B, Bm-Ant-Z — Antarctic blue whale song units (~25–100 Hz) Bm-D — Antarctic blue whale D-calls (~50–90 Hz, short FM downsweeps) Bp-20, Bp-20Plus — Antarctic fin whale 20 Hz song pulses Bp-Downsweep — fin whale 40 Hz downsweep calls Unidentified — unidentified low-frequency biological sounds Unid-BlueOrFin — detections of low-frequency biological sounds identified as likely blue or fin whale but not distinguishable between the two species. Annotations are provided as Raven Pro Selection Tables (tab-delimited text format), one file per analyst per call type, directly importable into Raven Pro. Multi-observer detection capture history. A comma-separated values (CSV) file consolidating detections from five independent observers: the three human analysts described above, an Ishmael spectral correlation coefficient (SCC) automated detector (Observer 4), and a Koogu deep neural network (DNN) automated detector (Observer 5). Observers 4 and 5 were applied to 250 Hz downsampled versions of the same recordings; their detections were matched to analyst annotations by temporal overlap as described in Miller et al. (in press). Each row in the capture history represents a candidate detection event. Columns record: detection flags for each observer; time and frequency bounds of each observer's annotation; an adjudicated true/false positive verdict (assigned by BSM) for a subset of rows; and per-observer signal-to-noise ratio (SNR) estimates computed using the spectrogramSlices method with noise windows placed symmetrically before and after each annotation. Full column descriptions are provided in the supplementary material of Miller et al. (in press). Note: the capture history covers ABZ-unit calls only, as these are the focus of Miller et al. (in press); annotations for all other call types are provided as Raven Pro Selection Tables only.