Twenty Years of Whale Calls
A long archive of underwater recordings becomes more useful when researchers can analyse it consistently and explain where their detector makes mistakes.
From recordings to comparable observations
Alksne and colleagues trained a deep-learning detector for five blue- and fin-whale call types in the southern California Current Ecosystem. They applied it to CalCOFI sonobuoy recordings collected quarterly from 2004 to 2024.
The work included repeated human review and testing across years, seasons and recording platforms. Performance varied among call types. Fin-whale 40 Hz calls were particularly challenging because they could be confused with overlapping humpback sounds.
After post-processing, the researchers accounted for recording effort and detection area to estimate standardised call density, with uncertainty.
Calls are not a census
A count of detected calls is not automatically a count of whales. Calling behaviour, sound propagation, background noise and recording coverage affect what a hydrophone detects. An apparent change needs interpretation before it becomes a population claim.
What a future Golfo Dulce archive would need
For a local acoustic programme, our editorial recommendation is to plan the archive before accumulating recordings. Each deployment should have a clear record of location, dates, depth, equipment, sampling settings and interruptions.
Preserve original files and document processing steps so that future researchers can revisit the same material. Keep a carefully checked set of local examples, including difficult sounds and periods without target calls, to test automated methods.
A detector trained elsewhere should be evaluated on local recordings before its results guide decisions. Published model performance is not a guarantee of accuracy in a different soundscape.
A tool that still needs people
The opportunity is to make long-term observation more manageable while preserving scientific accountability. Human review, transparent error estimates and consistent sampling remain part of that work.
This research analysed blue- and fin-whale sounds off California. It does not provide twenty years of recordings from Golfo Dulce or demonstrate a ready-to-use local humpback detector. For SaveGolfoDulce, it offers a practical reason to build future monitoring so that its evidence remains usable for years.
Source
Source checked: 28 September 2026. Hero image: AI-generated editorial illustration; not a photograph of the reported research.
Comments