‘Gives us a more complete picture’

A team of researchers at École Polytechnique Fédérale de Lausanne — or the Swiss Federal Institute of Technology in Lausanne — recently created a database to help scientists monitor wildlife without ever entering their habitats. Called MammAlps, the “richly annotated, multi-view, multimodal wildlife behavior dataset” is a way for scientists to better understand wildlife behavior, according to an EPFL article published on Phys.org.

To create the dataset, researchers set up nine cameras across the Swiss National Park, capturing more than 43 hours of raw footage. They then trimmed down this footage into 8 1/2 hours of highly detailed, labeled clips with the help of artificial intelligence. 

Each clip includes in-the-moment annotations of wildlife behaviors across two levels: broader activities such as “foraging” or “playing,” and more detailed actions such as “sniffing” or “running.” As an animal’s behavior changes during a clip, the labels change accordingly, creating a real-time behavioral timeline.

However, the database doesn’t just catalog video. It also includes audio recordings, environmental maps of natural features such as rivers and vegetation, and even documentation of weather conditions at the time.

“This multi-modal approach gives us a more complete picture of wildlife behavior,” EPFL professor and project supervisor Alexander Mathis said in the article.

While this detailed dataset could be a breakthrough in conservation, it’s worth noting that the use of AI comes at an environmental cost. Data centers powering AI require large amounts of electricity — often from dirty energy sources — and billions of gallons of water for cooling. AI hardware also relies on rare minerals and elements, which are often mined unethically and unsustainably, per the United Nations Environment Programme.

The growing demands of AI are expected to contribute to the doubling of data center power consumption from 2022 to 2026. 

Still, researchers believe MammAlps could help conservationists to quickly identify significant wildlife behaviors in hundreds of hours of wildlife footage. By training AI on the dataset, the technology could automatically scan new recordings and extract the most relevant segments. This could help spot changes in animal behavior caused by climate shifts, human activity, or even disease, leading to more effective protection strategies — all without disturbing animals in the wild.

The data collection isn’t over, either. The team is currently processing data collected in 2024 and plans to conduct more observations in 2025. These additional surveys will help expand recordings for rare species, including alpine hares and lynx. Additionally, researchers hope to analyze wildlife behavior over multiple seasons by expanding their observations.

Join our free newsletter for weekly updates on the latest innovations improving our lives and shaping our future, and don’t miss this cool list of easy ways to help yourself while helping the planet.