Passive acoustic monitoring (PAM) of marine mammals
Acoustic monitoring exploits the sounds produced by marine mammals to detect their presence, identify species and, under suitable conditions, determine their location, behavior and abundance.
Marine mammal acoustic monitoring is predominantly passive. Active acoustic monitoring is discussed in the article Underwater acoustic observation.
Passive acoustic monitoring (PAM) records sounds produced by marine mammals with underwater hydrophones. It complements visual surveys because marine mammals spend most of their time below the sea surface and can therefore remain undetected visually. PAM can detect vocalizing animals while submerged and can operate at night, in poor weather and, with autonomous instruments, continuously for months or years.
Passive acoustic monitoring can detect only animals that produce detectable sounds. Animals that are present but silent may therefore remain undetected, and interpretation of acoustic observations requires knowledge of species- and behavior-dependent vocalization rates[1].
Contents
Introduction
Pinnipeds (seals) and Cetaceans (whales) have highly developed hearing and use sound extensively for communication; toothed whales additionally use echolocation for navigation and prey detection. Moreover, numerous species of toothed whales are capable of using echolocation to detect and characterize their food, to navigate underwater and to avoid obstacles.
The efficient underwater transmission of sound enables researchers to detect and localize vocalizing marine mammals with passive acoustic methods. The acoustic characteristics of sound emissions of marine mammals can differ considerably, ranging from very short pulsed echolocation clicks to long-lasting frequency modulated songs. However, sophisticated detection and analysis techniques are needed to distinguish between sounds emitted by different animals and to isolate these often weak animal sounds from the numerous sources of contaminating background noise, including sound produced by wind, waves, precipitation, the survey vessel and other ships.[2].
Monitoring of marine mammals
Due to their mainly submerged lifestyle, it is difficult to assess the movements and habitat uses of whales and seals. To monitor their movements as well as their abundance and density in selected areas over time, it is important to choose the appropriate methods.
Although high-resolution satellite imagery has been used to monitor large whales, the vast majority of visual surveys are conducted from airplanes or research vessels. These surveys typically cover large areas, but at low frequency given their high costs and logistical constraints. Additionally, visual surveys only can be conducted in daylight and in favorable weather. Marine mammal abundance and distribution are commonly assessed by visual line-transect surveys from ships or aircraft.
Because marine mammals regularly produce sound for communication, navigation, and prey detection, and because sound generally travels efficiently underwater, towed hydrophone arrays commonly complement visual observations during shipborne surveys. Towed arrays can collect data throughout the night and in inclement weather, and are especially useful for detecting deep-diving cetacean species (e.g. beaked whales, family Ziphiidae) that spend little time at the surface. However, certain inferences about marine mammal populations that can be derived from visual observation, such as average body condition, are virtually impossible to derive with passive acoustics.[2]
Hydrophones allow detection of sounds emitted underwater by the marine mammals. The receiving hydrophones are either being towed behind a vessel or deployed as stationary hydrophones. When towed, usually two or more hydrophones are installed in an array. In a hydrophone array, differences in arrival time of the same sound at spatially separated receivers can be used to estimate the direction of the source. With a sufficiently large and suitably configured array, or by combining observations from several positions, the source can also be localized. Fixed autonomous recorders, provide high temporal resolution at one location. Mobile autonomous platforms, such as gliders, floats and drifters, trade temporal resolution for greater spatial coverage.[2]
The distance over which an animal can be detected depends not only on hydrophone sensitivity and bandwidth, but also on source level and directionality of the animal's call, ambient noise and frequency-dependent sound propagation. Low-frequency baleen-whale calls may be detected over tens of kilometers, whereas high-frequency harbor-porpoise signals are typically detectable only over much shorter distances. Consequently, the effective detection range can vary strongly between species, sites and environmental conditions. Recording systems must be adapted to the frequency range of the target species: low-frequency baleen-whale calls require much lower sampling rates than the high-frequency echolocation clicks of toothed whales and porpoises.
Repeated PAM surveys can contribute to detecting population trends, provided that changes in acoustic detections can be distinguished from changes in detection probability and vocalization rate. The number of detected calls is therefore not directly proportional to the number of animals. Moreover, the vocalization rate may change as the number of animals present changes. This further depends on behavior, season and environmental conditions. Several methods to account for these factors in animal density estimation are reviewed by Marques et al. 2013[1].
Animal-borne acoustic tags record sounds and movement of individual animals. They are particularly useful for relating vocalizations to diving and behavior and for estimating vocalization rates needed to interpret passive acoustic surveys.
Comparison of different observation methods
| Monitoring method | Main strength | Main limitation |
|---|---|---|
| Fixed autonomous hydrophone | Continuous long-term monitoring | Limited spatial coverage |
| Mobile acoustic platform | Surveys larger areas | Lower temporal resolution at any particular location |
| Towed hydrophone array | Combines acoustic and ship survey | Requires survey vessel |
| Visual survey | Direct observation and group counts | Daylight/weather/surfacing limitations |
Analysis of passive acoustic monitoring records
Long-term passive acoustic monitoring produces very large datasets that generally cannot be analyzed entirely by manual inspection. Modern systems increasingly use machine-learning methods to detect signals, extract acoustic features, identify recurring patterns and classify sounds by species or sound type. Well-trained classifiers can process recordings much faster than human observers and may achieve high detection and classification accuracy.
Their performance nevertheless depends strongly on the quality and representativeness of the labelled recordings used for training and validation. Supervised machine-learning algorithms trained on labelled recordings are limited by their reliance on costly expert annotations, which are rarely available at scale. Supervised methods can perform very well when representative labelled training data are available, but they can only recognize the categories represented in their training data and therefore depend strongly on predefined classes and expert annotations. PAM increasingly demands representation-learning approaches that organize acoustic sources without assuming fixed taxonomies.[3]
Self-supervised and unsupervised methods can make use of the much larger volume of unlabeled recordings. They learn acoustic similarities and recurring patterns directly from the data and can group vocalizations into acoustically coherent clusters without predefined species or call categories. These methods can reveal previously unrecognized acoustic groupings and recurring signal types, which may correspond to different call types, species, populations or behavioral states. The resulting acoustic clusters do not by themselves have a biological meaning: linking them to species, call types or behavioral states still requires expert interpretation or independent observations. The extraction of meaningful patterns is complicated due to overlapping calls, heterogeneous environments, subtle across-species variation and low signal-to-noise ratios.
Increasingly, large pretrained acoustic models combine both approaches: general representations are learned from extensive unlabeled datasets and subsequently adapted to particular species or monitoring questions with relatively small amounts of labelled data. The development of effective machine-learning methods for the analysis of large PAM records is an active field of research.[4]
Related articles
- Underwater acoustic observation
- General principles of optical and acoustical instruments
- Currents and turbulence by acoustic methods
- Acoustic backscatter profiling sensors (ABS)
- Acoustic point sensors (ASTM, UHCM, ADV)
- Health biomarkers in marine mammals
- Counting seabirds from ships and aircraft
- Application of data loggers to seabirds
References
- ↑ 1.0 1.1 Marques, T.A., Thomas, L., Martin, S.W., Mellinger, D.K., Ward, J.A., Moretti, D.J., Harris, D. and Tyack, P.L. 2013. Estimating animal population density using passive acoustics. Biological Reviews 88: 287–309
- ↑ 2.0 2.1 2.2 Fleishman, E., Cholewiak, D., Gillespie, D., Helble, T., Klinck, H., Nosal, E.-M. and Roch, M.A. 2023. Ecological inferences about marine mammals from passive acoustic data. Biological Reviews 98: 1633–1647
- ↑ Acs, R., Ibrahim, A., Zhuang, H. and Chérubin, L.M. 2026. Contrastive learning for passive acoustic monitoring: A framework for sound source discovery and cross-site comparison in marine soundscapes. PLoS Computational Biology 22(3), e1014005
- ↑ Schäfer-Zimmermann, J.C., Demartsev, V., Averly, B., Dhanjal-Adams, K.L., Duteil, M., Gall, G., Faiß, M., Johnson-Ulrich, L., Stowell, D., Manser, M.B., Roch, M.A. and Strandburg-Peshkin, A. 2026. animal2vec and MeerKAT: A self-supervised transformer for rare-event raw audio input and a large-scale reference dataset for bioacoustics. Methods in Ecology and Evolution 17: 875–888
Please note that others may also have edited the contents of this article.
|
