We are listening to Seabirds with AI to Better Understand Biodiversity

For more than a decade, the Auk Lab has been a unique hub for seabird research in the Baltic Sea. Equipped with cameras, scales, and sensors, it has revealed how storms, heatwaves, predators, and human-created stressors impact the birds’ health and their chances of survival. Recently, and with support from RISE researchers, the lab has been “mic’d up”: 18 microphones now capture the nonstop chatter of the colony — from courting calls to warning cries.

By combining sound with video and other sensor data, scientists gain a richer, multi-sensory picture of guillemot life. This makes it possible to study how environmental change is impacting one of the Baltic’s most iconic seabirds in entirely new ways. As a top predator, and with the capacity to dive up to 200 m below the surface, the guillemot's entire life is a reflection of the ocean's health; the fish it eats and its success in raising chicks give us vital clues about the hidden world beneath the waves, from the health of fish populations to the impacts of climate change.

AI turns sound into insight

The challenge with audio is scale: thousands of hours of recordings are impossible for humans to analyze manually. This is where AI makes a difference.

Researchers at RISE have developed a new method called Adaptive Change-Point Detection (A-CPD), which can pinpoint when bird calls occur in long recordings. By learning from human feedback, the system adapts and improves, helping experts focus on the most relevant sounds. The result: faster, more accurate analysis of vast audio datasets — making it possible to detect patterns that would otherwise remain hidden.

The work was conducted by John Martinsson, with contributions from his colleague Delia Fano Yela and supervisor Olof Mogren, all at RISE.

For readers with a deeper technical interest: A-CPD works by analyzing the probability curve from a machine learning model trained on acoustic data. Instead of relying on fixed-length audio segments, it detects change points in the likelihood of a target sound event. These change points define candidate segments that are presented to a human annotator for validation. Each new annotation strengthens the model, which then refines its future predictions in an active learning loop. In experiments, A-CPD required fewer queries and produced more precise event labels compared to baseline methods, making it particularly well suited for large-scale bioacoustic monitoring.

“This technology not only speeds up the process of analyzing hours of audio but also enables us to uncover patterns in wildlife behavior that were previously invisible,” says Olof Mogren.

With synchronized sound, video, and sensor data, researchers now hope to explore entirely new questions: do the birds use their calls to identify each other, are there specific calls that signal stress or danger, and what is really happening during the order-based “dialogue” between the mother and the chick just before it sets out on its first flight? 

All components of the new audio system are being developed as open source, making them available for biodiversity monitoring worldwide. The work at Stora Karlsö shows how RISE, together with partners, is pioneering the use of AI in environmental research — helping us better understand and protect fragile ecosystems.

A collaborative effort

RISE contributes its expertise in sound data collection and machine learning-based soundscape analysis to the Baltic Seabird Project, a major research initiative led by the Swedish University of Agricultural Sciences and Stockholm University and funded by the Marcus and Marianne Wallenberg Foundation.

More information:

Links:
• Read the research article on Adaptive Change-Point Detection
• AI for Climate: Featured Projects on Climate AI Nordics. 
• The AukLab
• http://www.balticseabird.com/