A Bioacoustic Analysis of North Atlantic Populations
Introduction
Understanding the movement of species across the vast oceans is incredibly important for humans to comprehend their behaviours better and ultimately inform the implementation of public policies geared towards their protection. Tracking the movements of marine mammals such as dolphins and whales, which traverse vast underwater seascapes, is often an incredibly challenging task that requires the costly collection and processing of large amounts of data. While direct and indirect visual observations have been used successfully for many years on land to monitor species presence and to track the distribution and composition of animal populations, using visual observation in marine environments is often difficult. This difficulty is due to the vast habitat range that some marine animals cover, especially when their feeding patterns involve seasonal and long oceanic migrations. To monitor these populations, such as the North Atlantic humpback whales, passive acoustic monitoring has been employed by ecologists either exclusively or in conjunction with direct visual monitoring. This technique involves the recording of many hours of cetacean vocalisations, which helps scientists understand not only these animals' movements but also what is typically referred to as cultural transmission, which is crucial to understanding these animals’ behaviour.
For this large amount of data to become useful knowledge for scientists, the use of emerging technologies such as AI in recent years has offered the scientific community the opportunity to speed up the processing and interpretation of large amounts of data. This aids in achieving a more accurate and timely understanding of the social and environmental interactions of the most intelligent marine species, such as baleen whales.
Bringing this innovation to marine research is the ambition of the project funded through the DOTphin initiative.
Thanks to the generous grant from the Polkadot Treasury from August 2024 to March 2026 Aquasearch in partnership with SNI, will be undertaking an R&D project leveraging AI technology for the interpretation of acoustic data from male humpback whales belonging to the North Atlantic Ocean subpopulations during their seasonal migrations.
Since its inception, SNI has been a believer in and supporter of emerging technologies' potential to reconfigure the human relationship with the non-human world. This project has the potential to allow us to relate anew to these highly intelligent mammal species by understanding how to use their communication to monitor their movements to avoid disturbance to their lives.
The scientific study will entail the analysis and comparison of the specific vocal sequences of these whales’ songs, visualised as spectrograms, to understand their social interactions and migratory trajectory. This study uses innovative technology to explore how their songs evolve and potentially incorporate cultural elements during their journey from cooler feeding waters to the warmer equatorial regions where they reproduce.
Understanding the movement of species across the vast oceans is incredibly important for humans to comprehend their behaviours better and ultimately inform the implementation of public policies geared towards their protection. Tracking the movements of marine mammals such as dolphins and whales, which traverse vast underwater seascapes, is often an incredibly challenging task that requires the costly collection and processing of large amounts of data. While direct and indirect visual observations have been used successfully for many years on land to monitor species presence and to track the distribution and composition of animal populations, using visual observation in marine environments is often difficult. This difficulty is due to the vast habitat range that some marine animals cover, especially when their feeding patterns involve seasonal and long oceanic migrations. To monitor these populations, such as the North Atlantic humpback whales, passive acoustic monitoring has been employed by ecologists either exclusively or in conjunction with direct visual monitoring. This technique involves the recording of many hours of cetacean vocalisations, which helps scientists understand not only these animals' movements but also what is typically referred to as cultural transmission, which is crucial to understanding these animals’ behaviour.
For this large amount of data to become useful knowledge for scientists, the use of emerging technologies such as AI in recent years has offered the scientific community the opportunity to speed up the processing and interpretation of large amounts of data. This aids in achieving a more accurate and timely understanding of the social and environmental interactions of the most intelligent marine species, such as baleen whales.
Bringing this innovation to marine research is the ambition of the project funded through the DOTphin initiative.
Thanks to the generous grant from the Polkadot Treasury from August 2024 to March 2026 Aquasearch in partnership with SNI, will be undertaking an R&D project leveraging AI technology for the interpretation of acoustic data from male humpback whales belonging to the North Atlantic Ocean subpopulations during their seasonal migrations.
Since its inception, SNI has been a believer in and supporter of emerging technologies' potential to reconfigure the human relationship with the non-human world. This project has the potential to allow us to relate anew to these highly intelligent mammal species by understanding how to use their communication to monitor their movements to avoid disturbance to their lives.
The scientific study will entail the analysis and comparison of the specific vocal sequences of these whales’ songs, visualised as spectrograms, to understand their social interactions and migratory trajectory. This study uses innovative technology to explore how their songs evolve and potentially incorporate cultural elements during their journey from cooler feeding waters to the warmer equatorial regions where they reproduce.
Aquasearch scientists installing a hydrophone
Let’s get Eco-Technical
The project leverages the YOLOv5 model, an extension of the original YOLO (You Only Look Once) system, enhanced for its application in bioacoustics to analyse humpback whale vocalisations. Unlike its primary use in visual object detection, YOLOv5 has been adapted to process spectrogram data, which visually represents sound frequencies over time. This adaptation enables detailed analysis and classification of complex whale songs, capturing their unique sequences and variations.
Utilizing convolutional neural networks, the AI system identifies and classifies distinct vocal patterns, facilitating a deeper understanding of whale behaviour and social structures through their vocal communications. This non-invasive approach broadens our scientific understanding of cetaceans and by providing insights into their migratory patterns and social interactions may aid the development of conservation strategies.
Data collection covers several key locations, including breeding grounds from the warm waters of Martinique and feeding areas in the colder waters of Western Canada, Iceland, and Norway. Organisations who have agreed to cooperate with this project include OMMAG, UQAR-ISMER, M-Expertise Marine, and the Irish Whale and Dolphin Group.
Thanks to the collaborative spirit of these organisations, who share a tireless dedication to the conservation of these species, this project proposes to provide a detailed overview of complex migratory patterns of humpback whales across the Northern Atlantic, based on historical records complete with data collected during the next breeding period.
The project leverages the YOLOv5 model, an extension of the original YOLO (You Only Look Once) system, enhanced for its application in bioacoustics to analyse humpback whale vocalisations. Unlike its primary use in visual object detection, YOLOv5 has been adapted to process spectrogram data, which visually represents sound frequencies over time. This adaptation enables detailed analysis and classification of complex whale songs, capturing their unique sequences and variations.
Utilizing convolutional neural networks, the AI system identifies and classifies distinct vocal patterns, facilitating a deeper understanding of whale behaviour and social structures through their vocal communications. This non-invasive approach broadens our scientific understanding of cetaceans and by providing insights into their migratory patterns and social interactions may aid the development of conservation strategies.
Data collection covers several key locations, including breeding grounds from the warm waters of Martinique and feeding areas in the colder waters of Western Canada, Iceland, and Norway. Organisations who have agreed to cooperate with this project include OMMAG, UQAR-ISMER, M-Expertise Marine, and the Irish Whale and Dolphin Group.
Thanks to the collaborative spirit of these organisations, who share a tireless dedication to the conservation of these species, this project proposes to provide a detailed overview of complex migratory patterns of humpback whales across the Northern Atlantic, based on historical records complete with data collected during the next breeding period.
Example of spectrogram image from a male humpback whale recorded in April 2024 in Martinique
How does the study of spectrograms relate to humpback whales' interactions and movement?
Studying humpback whale songs provides valuable insights into the use of their repartition area, behaviours and social dynamics through their complex vocalizations. These songs are structured from individual sounds called units, which combine into phrases and are organized into themes, repeating in a specific sequence to form extended songs that can last up to 30 minutes and often continue for hours. The longest recorded song stretched over 24 consecutive hours.
While the reasons for changes in their songs are yet not fully understood, it is generally accepted that humpback whales exchange song segments, incorporating these into their vocal repertory through a process known as horizontal cultural transmission. This enables researchers to detect shared song patterns across different subpopulations by analysing the song’s composition units from various regions and periods. This technology-driven approach allows for the identification of migration patterns by noting when similar song elements appear among distinct populations. Moreover, changes in these songs are thought to reflect responses to behaviour, population structure, cultural transmission or anthropic activities. (Cerchio et al., 2000; Rish et al., 2012; Rebecca, 2016).
The project seeks to enhance our understanding of this cultural transmission by correlating identified song sequences, migration patterns, anthropic disturbances and environmental factors. This comprehensive approach aims to explain how humpback whales adapt their song, considering their social structure and environmental contexts.
This ambitious project kicked off on the 12th of August 2024, setting the stage for an endeavour to decipher the intricate songs of humpback whales using AI and ML technologies.
Currently, Aquasearch efforts are focused on testing the AI's capability to process these unique vocalisations, by feeding the AI with Terabytes of raw bioacoustic files recorded over the past 10 years in the Caribbean Sea. By the end of this month, we expect the AI to detect the first sequences from the analysis of spectrograms, which will be available on the dedicated project page on SNI’s REAL interface.
As we move into autumn, the spectrograms analysed by Aquasearch’s AI will expand to include historical data from Norway, Canada, Cape Verde, and Iceland. By the end of December 2024, we anticipate presenting all the songs identified by the AI, marking the first significant milestone in our project.
The new year will begin with an intensive phase of data collection during the 2025 mating season, particularly focusing on the Lesser Antilles. This fresh bioacoustic data will be processed, enriching the available dataset with the latest seasonal insights expected to be available by May 1st, 2025. Subsequent months will be dedicated to the analysis and comparison of the songs collected over the year.
The final quarter of 2025 will be a period when Aquasearch will synthesise and analyse the insights from all sequenced songs. The culmination of our efforts is expected in early 2026, when the findings are slated to be published in a prestigious scientific journal, sharing Aquasearch’s discoveries with the scientific community and policy-makers who might find the insights relevant for future regulatory interventions.
While the reasons for changes in their songs are yet not fully understood, it is generally accepted that humpback whales exchange song segments, incorporating these into their vocal repertory through a process known as horizontal cultural transmission. This enables researchers to detect shared song patterns across different subpopulations by analysing the song’s composition units from various regions and periods. This technology-driven approach allows for the identification of migration patterns by noting when similar song elements appear among distinct populations. Moreover, changes in these songs are thought to reflect responses to behaviour, population structure, cultural transmission or anthropic activities. (Cerchio et al., 2000; Rish et al., 2012; Rebecca, 2016).
The project seeks to enhance our understanding of this cultural transmission by correlating identified song sequences, migration patterns, anthropic disturbances and environmental factors. This comprehensive approach aims to explain how humpback whales adapt their song, considering their social structure and environmental contexts.
This ambitious project kicked off on the 12th of August 2024, setting the stage for an endeavour to decipher the intricate songs of humpback whales using AI and ML technologies.
Currently, Aquasearch efforts are focused on testing the AI's capability to process these unique vocalisations, by feeding the AI with Terabytes of raw bioacoustic files recorded over the past 10 years in the Caribbean Sea. By the end of this month, we expect the AI to detect the first sequences from the analysis of spectrograms, which will be available on the dedicated project page on SNI’s REAL interface.
As we move into autumn, the spectrograms analysed by Aquasearch’s AI will expand to include historical data from Norway, Canada, Cape Verde, and Iceland. By the end of December 2024, we anticipate presenting all the songs identified by the AI, marking the first significant milestone in our project.
The new year will begin with an intensive phase of data collection during the 2025 mating season, particularly focusing on the Lesser Antilles. This fresh bioacoustic data will be processed, enriching the available dataset with the latest seasonal insights expected to be available by May 1st, 2025. Subsequent months will be dedicated to the analysis and comparison of the songs collected over the year.
The final quarter of 2025 will be a period when Aquasearch will synthesise and analyse the insights from all sequenced songs. The culmination of our efforts is expected in early 2026, when the findings are slated to be published in a prestigious scientific journal, sharing Aquasearch’s discoveries with the scientific community and policy-makers who might find the insights relevant for future regulatory interventions.
Image of a humpback whale tail captured in the Lesser Antilles by Aquasearch
The Impact We Hope to Achieve
At SNI, our project is guided by a commitment to enhancing scientific knowledge and its transmission to ultimately inform and contribute to an important policy collaboration between governments responsible for the conservation of this exceptional biodiversity.
One of the aims of this project is for Aquasearch to be able to analyse the evolution of songs overtime, contribute to the growing efforts from the scientific community aimed at leveraging the use of bioacoustics to understand the migratory patterns of humpback whale populations in the North Atlantic region.
Additionally, Aquasearch will release part of the technology and methods developed during this project as open-source. This allows other organizations and researchers to apply these techniques not only to humpback whales but also to other baleen species. By making our methodologies accessible, we hope to catalyse further research and collaboration within the marine biology community.
The challenges facing these whales in the Atlantic such as noise pollution, vessel collisions, resource depletion and entanglement in fishing gear are intensifying and urging adequate interventions. By providing solid scientific knowledge on the seasonal movement of humpback whales across important feeding and reproductive grounds and how environmental changes influence such vital patterns, combined with the growing understanding of the direct and indirect threats to these species and their habitats, we hope to influence needed multilateral policy changes that could alleviate these pressures and enhance marine conservation efforts.
The results will be made available through periodic reports and a final scientific paper published in a renowned journal which will be made available through the SNI REAL portal to the growing DOTphin community which has indirectly contributed to the project funding.
This initiative stands as one of SNI’s most significant since our first funded project in partnership with the Kenya Wildlife Trust, underscoring SNI’s commitment to supporting high-quality scientific research through technical innovation with the final intention to inform international cooperation to safeguard the most vulnerable and valuable ecosystems on our planet.
At SNI, our project is guided by a commitment to enhancing scientific knowledge and its transmission to ultimately inform and contribute to an important policy collaboration between governments responsible for the conservation of this exceptional biodiversity.
One of the aims of this project is for Aquasearch to be able to analyse the evolution of songs overtime, contribute to the growing efforts from the scientific community aimed at leveraging the use of bioacoustics to understand the migratory patterns of humpback whale populations in the North Atlantic region.
Additionally, Aquasearch will release part of the technology and methods developed during this project as open-source. This allows other organizations and researchers to apply these techniques not only to humpback whales but also to other baleen species. By making our methodologies accessible, we hope to catalyse further research and collaboration within the marine biology community.
The challenges facing these whales in the Atlantic such as noise pollution, vessel collisions, resource depletion and entanglement in fishing gear are intensifying and urging adequate interventions. By providing solid scientific knowledge on the seasonal movement of humpback whales across important feeding and reproductive grounds and how environmental changes influence such vital patterns, combined with the growing understanding of the direct and indirect threats to these species and their habitats, we hope to influence needed multilateral policy changes that could alleviate these pressures and enhance marine conservation efforts.
The results will be made available through periodic reports and a final scientific paper published in a renowned journal which will be made available through the SNI REAL portal to the growing DOTphin community which has indirectly contributed to the project funding.
This initiative stands as one of SNI’s most significant since our first funded project in partnership with the Kenya Wildlife Trust, underscoring SNI’s commitment to supporting high-quality scientific research through technical innovation with the final intention to inform international cooperation to safeguard the most vulnerable and valuable ecosystems on our planet.
Make sure to have a look at SNI DOTphin project that has been a key supporter of this research.