Vultures and artificial intelligence(s) as death detectors: High-tech approach for wildlife research and conservation


In order to use remote locations to record and assess the behaviour of wildlife and environmental conditions, the GAIA Initiative developed an artificial intelligence (AI) algorithm that reliably and automatically classifies behaviours of white-backed vultures using animal tag data. As scavengers, vultures always look for the next carcass. With the help of tagged animals and a second AI algorithm, the scientists can now automatically locate carcasses across vast landscapes. The algorithms described in a recently published article in the Journal of Applied Ecology are therefore key components of an early warning system that can be used to quickly and reliably recognise critical changes or incidents in the environment such as droughts, disease outbreaks or the illegal killing of wildlife.

The GAIA Initiative is an alliance of research institutes, conservation organisations and enterprises with the aim of creating a high-tech early warning system for environmental changes and critical ecological incidents. The new AI algorithms were developed by the Leibniz Institute for Zoo and Wildlife Research (Leibniz-IZW) in cooperation with the Fraunhofer Institute for Integrated Circuits IIS and the Tierpark Berlin.

The death of wildlife is an important process in ecosystems — regardless whether this is a regular case, such as the successful hunt of a predator, or an exceptional case caused by the outbreak of a wildlife disease, the contamination of the landscape with environmental toxins or illegal killing by people. For the investigation of mammalian species communities and ecosystems it is therefore important to systematically record and analyse these regular and exceptional cases of mortality. In order to achieve this, the GAIA Initiative makes use of the natural abilities of white-backed vultures (Gyps africanus) in combination with highly developed biologging technologies and artificial intelligence. “This combination of three forms of intelligence – animal, human and artificial — is the core of our new approach with which we aim to make use of the impressive knowledge that wildlife has about ecosystems,” says Dr Jörg Melzheimer, GAIA project head and scientist at the Leibniz-IZW.

Vultures are perfectly adapted by millions of years of evolution to detect carcasses across vast landscapes quickly and reliably. They have outstanding eye-vision and sophisticated communication that allows them to monitor very large areas of land when many individuals work together. Vultures thus fulfil an important ecological role by cleaning landscapes of carrion and containing the spread of wildlife diseases. “For us as wildlife conservation scientists, the knowledge and skills of vultures as sentinels are very helpful to be able to quickly recognise problematic exceptional cases of mortality and initiate appropriate responses,” says Dr Ortwin Aschenborn, GAIA project head alongside Melzheimer at the Leibniz-IZW. “In order to use vulture knowledge, we need an interface — and at GAIA, this interface is created by combining animal tags with artificial intelligence.”

The animal tags with which GAIA equipped white-backed vultures in Namibia record two groups of data. The GPS sensor provides the exact location of the tagged individual at a specific point in time. The so-called ACC sensor (ACC is short for acceleration) stores detailed movement profiles of the tag — and thus of the animal — along the three spatial axes at the exact same time. Both groups of data are used by the artificial intelligence algorithms developed at the Leibniz-IZW. “Every behaviour is represented by specific acceleration patterns and thus creates specific signatures in the ACC data of the sensors,” explains wildlife biologist and AI specialist Wanja Rast from the Leibniz-IZW. “In order to recognise these signatures and reliably assign them to specific behaviours, we trained an AI using reference data. These reference data come from two white-backed vultures that we fitted with tags at Tierpark Berlin and from 27 wild vultures fitted with tags in Namibia.” In addition to the ACC data from the tags, the scientists recorded data on the behaviour of the animals — in the zoo through video recordings and in the field by observing the animals after they had been tagged. “In this way, we obtained around 15,000 data points of ACC signatures ascribed to a verified, specific vulture behaviour. These included active flight, gliding, lying, feeding and standing. This data set enabled us to train a so-called support vector machine, an AI algorithm that assigns ACC data to specific behaviours with a high degree of reliability,” explains Rast.

In a second step, the scientists combined the behaviour thus classified with the GPS data from the tags. Using algorithms for spatial clustering, they identified locations where certain behaviours occurred more frequently. In this way, they obtained spatially and temporally finely resolved locations where vultures fed. “The GAIA field scientists and their partners in the field were able to verify more than 500 of suspected carcass locations derived from the sensor data, as well as more than 1300 clusters of other non-carcass behaviours,” says Aschenborn. The field-verified carcass locations ultimately served to establish vulture feeding site signatures in the scientists’ final AI training dataset — this algorithm indicates with high precision locations where an animal has most likely died and a carcass is on the ground. “We could predict carcass locations with an impressive 92 percent probability and so demonstrated that a system which combines vulture behaviour, animal tags and AI is very useful for large-scale monitoring of animal mortality,” says Aschenborn.

This AI-based behaviour classification, carcass detection and carcass localisation are key components of the GAIA early warning system for critical changes or incidents in the environment. “Until now, this methodological step has been carried out in the GAIA data lab at the Leibniz-IZW in Berlin,” says Melzheimer. “But with the new generation of animal tags developed by our consortium, AI analyses are implemented directly on the tag. This will provide reliable information on whether and where an animal carcass is located without prior data transfer in real time without any loss of time.” The transfer of all GPS and ACC raw data is no longer necessary, allowing data communication with a significantly lower bandwidth to transmit the relevant information. This makes it possible to use a satellite connection instead of terrestrial GSM networks, which guarantees coverage even in remote wilderness regions completely independent of local infrastructure. Even at the most remote locations, critical changes or incidents in the environment — such as disease outbreaks, droughts or illegal killing of wildlife — could then be recognised without delay.

In recent decades, the populations of many vulture species declined sharply and are now acutely threatened with extinction. The main causes are the loss of habitat and food in landscapes shaped by humans as well as a high number of direct or indirect incidents of poisoning. The population of the white-backed vulture, for example, declined by around 90 percent in just three generations — equivalent to an average decline of 4 percent per year. “Owing to their ecological importance and rapid decline, it is essential to significantly improve our knowledge and understanding of vultures in order to protect them,” says Aschenborn. “Our research using AI-based analysis methods will not only provide us with insights into ecosystems. It will also increase our knowledge of how vultures communicate, interact and cooperate, forage for food, breed, rear their young and pass on knowledge from one generation to the next.” GAIA has so far fitted more than 130 vultures in different parts of Africa with tags, most of them in Namibia. Until today, the scientists analysed more than 95 million GPS data points and 13 billion ACC records.

Leave a Reply

Your email address will not be published. Required fields are marked *