To monitor wildlife behavior and environmental changes in remote areas, the GAIA Initiative has created an artificial intelligence (AI) algorithm that effectively and automatically categorizes the behaviors of white-backed vultures using data from animal tags. As scavengers, vultures continuously search for carcasses. With the assistance of tagged animals and another AI algorithm, scientists can now pinpoint carcasses over extensive landscapes. The algorithms, detailed in a recent publication in the Journal of Applied Ecology, are key elements of an early warning system designed to swiftly and accurately identify significant environmental changes or incidents, including droughts, disease outbreaks, or illegal wildlife killings.
The GAIA Initiative comprises a coalition of research institutions, conservation bodies, and companies focused on developing a state-of-the-art early warning system for ecological shifts and critical incidents. The newly created AI algorithms were developed in partnership by the Leibniz Institute for Zoo and Wildlife Research (Leibniz-IZW), the Fraunhofer Institute for Integrated Circuits IIS, and the Tierpark Berlin.
The mortality of wildlife serves as a crucial ecological process—whether it’s through normal events like a predator’s successful hunt or unusual occurrences caused by wildlife diseases, environmental toxins, or illegal poaching. Investigating these regular and unusual mortality cases is vital when studying mammalian communities and ecosystems. The GAIA Initiative harnesses the natural capabilities of white-backed vultures (Gyps africanus) alongside advanced biologging technologies and artificial intelligence. “Our innovative I³ approach combines three types of intelligence—animal, human, and artificial—which allows us to tap into the vital insights wildlife has regarding ecosystems,” states Dr. Jörg Melzheimer, GAIA project leader and researcher at the Leibniz-IZW.
Over millions of years, vultures have evolved to quickly and effectively detect carcasses across large landscapes. They possess exceptional eyesight and advanced communication skills, enabling them to cover vast areas when working in groups. These birds play a significant ecological role by cleaning up carrion and helping to curb the spread of wildlife diseases. “For us as wildlife conservation scientists, understanding vulture behavior as environmental sentinels helps us to promptly identify problematic mortality cases and take appropriate actions,” explains Dr. Ortwin Aschenborn, another GAIA project lead at the Leibniz-IZW. “To utilize this vulture knowledge effectively, we need an interface, which GAIA creates by linking animal tags and artificial intelligence.”
The animal tags used by GAIA to track white-backed vultures in Namibia gather two types of data. A GPS sensor records the precise location of the tagged bird at any given time, while an ACC sensor (which stands for accelerometer) captures detailed movement profiles across three spatial dimensions simultaneously. Both datasets are processed by the AI algorithms developed at the Leibniz-IZW. “Each behavior generates unique acceleration patterns, leading to specific signatures in the ACC data,” explains Wanja Rast, a wildlife biologist and AI expert at the Leibniz-IZW. “To recognize these signatures and match them to corresponding behaviors, we trained the AI using reference data. This data comes from two white-backed vultures tagged at Tierpark Berlin and from 27 wild vultures tagged in Namibia.” Along with the ACC data from the tags, the team documented the animals’ behaviors via video in the zoo and through direct observation in the field post-tagging. “This resulted in approximately 15,000 ACC signature data points linked to verified specific vulture behaviors, including active flight, gliding, resting, feeding, and standing. This extensive dataset allowed us to train a support vector machine, an AI algorithm that accurately correlates ACC data with specific behaviors,” Rast elaborates.
In the subsequent stage, the researchers integrated the classified behaviors with the GPS data from the tags. By applying spatial clustering algorithms, they identified locations with higher occurrences of certain behaviors, thereby pinpointing where vultures were likely feeding. “Field scientists from GAIA and their partners validated over 500 suspected carcass locations identified through sensor data, as well as over 1300 clusters of other non-carcass behaviors,” notes Aschenborn. These verified carcass locations ultimately contributed to defining feeding site signatures in the scientists’ final AI training dataset—allowing the algorithm to predict carcass locations with remarkable accuracy. “We achieved an impressive 92 percent success rate in predicting carcass locations, demonstrating that combining vulture behavior, animal tags, and AI is highly effective for large-scale monitoring of wildlife mortality,” asserts Aschenborn.
The AI-driven behavior classification, carcass detection, and localization are vital elements of the GAIA early warning system, designed to flag critical environmental changes or incidents. “So far, this methodological step has occurred in the GAIA I³ data lab at the Leibniz-IZW in Berlin,” reveals Melzheimer. “However, with the latest generation of animal tags developed by our consortium, AI analyses can now occur directly on the tag. This advancement will facilitate real-time, reliable information regarding whether and where an animal carcass is located without prior data transfer or delays.” The necessity for transferring all GPS and ACC raw data becomes obsolete, significantly reducing the bandwidth required for communicating relevant information. This enhancement allows for satellite connectivity instead of relying on terrestrial GSM networks, ensuring coverage even in remote wilderness areas, independent of local infrastructure. Consequently, critical ecological changes or incidents—like disease outbreaks, droughts, or unlawful wildlife killings—could be detected promptly, even in the most isolated locales.
In recent decades, numerous vulture species have faced drastic population declines and are now perilously close to extinction. Key factors include habitat loss and food scarcity from human-altered landscapes, along with numerous instances of poisoning, whether deliberate or accidental. For instance, the white-backed vulture’s population has plummeted by approximately 90 percent in just three generations, resulting in an average annual decline of 4 percent. “Given their ecological significance and rapid population drop, it is imperative to enhance our understanding of vultures to ensure their survival,” asserts Aschenborn. “Our research utilizing AI-based analysis will not only provide valuable insights into ecosystems but also deepen our understanding of vulture communication, interaction, food foraging, breeding, parenting, and the transfer of knowledge across generations.” So far, GAIA has tagged over 130 vultures across various African locations, primarily in Namibia. To date, the researchers have analyzed more than 95 million GPS data points and 13 billion ACC records.
