Seismic surveying: fin whales stop vocalising from the first airgun shots

What happens when an oil prospecting campaign deploys its airguns over a migration corridor used by fin whales? A team from the University of Southampton was able to measure it off Galicia, thanks to a dataset that offered a rare comparison. The result: detections of vocalisations fall by 70% during the shooting, and still by 52% once the most troublesome bias has been set aside.

What an airgun is

Marine seismic surveying is used to map the geological structures of the subsurface, in particular to look there for oil and gas. The principle: a vessel tows guns that abruptly release compressed air at regular intervals. The wave crosses the water column, penetrates the seabed and comes back as an echo, which allows an image of the deep layers to be reconstructed.

It is one of the most powerful human-made sounds in the ocean, and it can be detected more than three thousand kilometres from the vessel producing it. The difficulty lies in the frequency: the energy of these shots is concentrated in the low end, precisely the band occupied by the calls of the large mysticetes, among the lowest-pitched sounds in the animal kingdom. Our page on anthropogenic ocean noise sets out these sources in detail.

A natural experiment provided by a breakdown

Measuring the effect of a noise requires a period without that noise, at the same place and in the same season. That is rarely available. Here, the survey vessel returned to port for repairs several times during the 2013 campaign, cutting the recording into four alternating periods, with and without shooting.

The listening set-up consisted of recorders placed on the seabed, at three sites, over 63 days of continuous recording.

That leaves the matter of counting the calls in 63 days of sound. The authors trained a convolutional neural network on 50 hours of hand-annotated recordings, then turned it loose on the whole corpus. Performance is documented, which is not always the case: 85.6% accuracy and an area under the curve of 0.93, an index that measures the model’s ability to separate true calls from the rest.

Results

Fall in detections during the shooting	70% on average across the three instruments
Fall after correction for masking	52%, least favourable assumption
Time for the effect to set in	1 to 2 days after the shooting began
Statistical significance	p < 0.0001, mixed-effects negative binomial model

The speed of the response is worth underlining: the collapse in detections sets in within one to two days. This is not a slow drift, it is a reaction.

The correction that gives the study its value

Here is the methodological point that sets this work apart from a mere observation, and that has to be understood if the figure of 70% is not to be over-read.

A recorder that detects fewer calls during the shooting may do so for two opposite reasons. Either the animals really are calling less. Or they are calling just as much, but the din of the guns masks their vocalisations, which become undetectable. Since the two signals occupy the same low-frequency band, this risk is far from theoretical.

The authors therefore estimated the share attributable to masking, adopting the assumption least favourable to their own result. Even then, a fall of 52% remains. It is this value, and not the 70%, that constitutes the robust result.

Silent or gone

The authors are explicit on this point, and it is to their credit: they do not know whether the fin whales fell silent or left. A hydrophone on the seabed does not distinguish an animal that is present and silent from an animal that is absent. The two mechanisms may in any case combine.

The distinction is not academic. Falling silent for several weeks in a migration corridor compromises exchanges that are associated with breeding and with social life. Moving away imposes a detour, and so an energy cost and the temporary abandonment of an area. The consequences differ, and neither is trivial.

The authors also write that they cannot infer from this dataset the wider ecological implications. What they establish is a short-term behavioural response, measured, significant, and so far very poorly documented.

What this suggests for regulation

The team puts forward three avenues, bearing less on prohibition than on adjustment: exclusion zones around areas that matter for cetaceans, seasonal restrictions matched to the periods when the animals pass through, and the use of quieter survey technologies, which exist but remain little deployed.

Applying these measures requires knowing where and when the animals are present. That is exactly what maps of seasonal distribution and networks of passive acoustic monitoring produce, and this work incidentally illustrates a second use for them: they serve not only to count the animals, but also to measure what human noise does to them.

The comparison with another recent study is instructive. Off California, six years of continuous listening showed that whale song tracks the abundance of their prey. Here, it tracks the presence of an industrial noise. In both cases the same underlying point: a whale’s silence is data, provided one knows what produced it.

Reference: Edwards E. A., Feakes A. M., Olcay A. A. and Minshull T. A., “Seismic survey noise reduces fin whale vocalisations offshore northwestern Spain”, Scientific Reports, 25 February 2026, open access under a CC BY 4.0 licence. Press release from the University of Southampton. Work co-led by Amy Feakes, a doctoral student, and Elodie Edwards, then a master’s student.