Macklin, Gabrielle F., Hilary B. Moors‐Murphy, and Marty L. Leonard.
"Increasing Efficiency of Detection and Validation of Baleen Whale Call
Presence on Large Acoustic Datasets: A Case Study Using Sei Whale
Downsweeps Recorded in Atlantic Canada." Marine Mammal Science 42.2 (2026):
e70136. https://doi.org/10.1111/mms.70136

ABSTRACT

As passive acoustic monitoring (PAM) datasets expand, studies increasingly
use automated detectors to analyze baleen whale presence. However, detector
errors like false positives (impacting precision, P; proportion of detected
calls that are correct), and false negatives (impacting recall, R;
proportion of available calls successfully detected) necessitate manual
validation. Here, we demonstrate a tiered detection and validation approach
to improve efficiency in sei whale (Balaenoptera borealis) call presence
analysis that aims to maximize R while removing false positives. Using two
detectors on eastern Canadian datasets, we tested multiple parameter
settings, selecting one tier that captured only high-quality calls
(High-P), one tier that captured most calls present but included more false
positives (High-R), and one tier that balanced the P and R values (Mid). To
assess daily call presence, validation proceeded sequentially across tiers
from High-P to Mid to High-R. For each subsequent tier, detections were
only reviewed from days in which sei whale calls had not yet been
confirmed. For both detectors, using this tiered approach reduced the
overall manual validation effort (number of detections reviewed) when
compared to validating the highest recall original outputs alone. This
efficient strategy maintains accuracy in PAM studies and could be used for
other species.

Please reach out if you have any questions,
Cheers,
Gabrielle Macklin
_______________________________________________
MARMAM mailing list
[email protected]
https://lists.uvic.ca/mailman/listinfo/marmam

Reply via email to