Mark Baumgartner is a senior scientist at Woods Hole Oceanographic Institution. Back in 2011 he and Sarah Mussoline published a paper on how to use computer science to classify whales from sound recordings. This episode is about an evolution of that software by Atle Borsholm from NV5 Geospatial (https://www.nv5geospatialsoftware.com).

At the beginning of the episode you will hear a sound recording. Take a listen and attempt to classify the whale making the calls in it. In my comments at the end of this I’ll tell you what whale that was. It was an easy one to process - the recording was just 7 seconds long and a whale sound was present immediately. Consider if I instead gave you another one lasting a day and most of it was just ocean noise. This leads to the challenge discussed in today’s episode.

Imagine I have 100 microphones under the ocean. These are called hydrophones. These hydrophones record for 24 days. The scientists managing them will say they now have 2400 days of data to process. Well, that is nowhere near how big of a challenge this episode is about. The National Oceanic and Atmospheric Administration had 217,197 days of hydrophone recordings going back 20 years. Whilst Mark’s software was available, it needed to be made more efficient to process all this data.

It was worth the effort. It turned out that only 24,000, or 11%, of those days involved a whale being detected. All of it is shown on NOAA’s Passive Acoustic Cetacean Map. Atle starts by discussing a web application, Mark’s Robots4Whales website, in our episode today.

Links for further reading:

NV5 article on Harnessing Data Science for Marine Conservation: https://www.nv5geospatialsoftware.com/Learn/Case-Studies/Case-Studies-Detail/harnessing-data-science-for-marine-conservation
NV5 article on IDL® Software Extract Meaningful Visualizations From Complex Numerical Data: https://www.nv5geospatialsoftware.com/Products/IDL
Slocum glider from Teledyne Marine: https://www.teledynemarine.com/brands/webb-research/slocum-glider
Woods Hole article on whale sound detection and classification: https://www.fisheries.noaa.gov/science-blog/listening-sounds-gulf

The whale at the start? It was a humpback whale. Only Gemini managed to classify it correctly. Claude and ChatGPT got it wrong.