Death detector: Using vultures as sentinels to detect carcasses by combining bio‐logging and machine learning

• Published in: The Journal of applied ecology 2024-12, Vol.61 (12), p.2936-2945
• Main Authors: Rast, Wanja, Portas, Rubén, Shatumbu, Gabriel Iita, Berger, Anne, Cloete, Claudine, Curk, Teja, Götz, Theresa, Aschenborn, Ortwin, Melzheimer, Jörg
• Format: Article
• Language: English
• Subjects: accelerometry; Algorithms; Artificial intelligence; behaviour classification; carcass detection; Carcasses; Clustering; Data logging; Datasets; Environmental monitoring; Feeding behavior; feeding sites; Food poisoning; Food resources; Food sources; Global positioning systems; GPS; gyps africanus; Logging; Machine learning; Marking; Marking behavior; Pest outbreaks; random forest; Recall; Remote monitoring; Spatial data; support vector machine; Support vector machines; Wildlife
• ISSN: 0021-8901; 1365-2664
• DOI: 10.1111/1365-2664.14810

Bio‐logging technologies allow scientists to remotely monitor animal behaviour and the environment. In this study, we used the combination of natural abilities of African white‐backed vultures Gyps africanus and state‐of‐the‐art bio‐logging technology for detecting and locating carcasses in a vast landscape. We used data from two captive and 27 wild vultures to create a reference data set for the training of a support vector machine to distinguish between six behaviour classes based on acceleration data. Next, we combined the classified behaviour of the initial 27 and 7 additional vultures with GPS data and used the ‘Density‐Based Spatial Clustering of Applications with Noise’ algorithm to cluster all GPS data to get a position of potential feeding locations. Finally, we used the clustered data set to train a Random Forest algorithm to distinguish between clusters with and without a carcass. The behaviour classifier was trained on 14,682 samples for all behaviour classes, which were classified with a high performance (overall precision: 0.95, recall: 0.89). This enabled a ground team to examine 1900 clusters between May 2022 and March 2023 in the field, 580 linked to a carcass and 1320 without a carcass. The cluster classifier trained on this data set was able to correctly distinguish between carcass and no carcass clusters with high performance (overall precision: 0.92, recall: 0.89). Synthesis and applications. We showed that a carcass detection system using vultures, loggers and artificial intelligence (AI) can be used to monitor the mortality of numerous species in a vast landscape. This method has broad applications, such as studying the feeding ecology of vultures, detecting and monitoring of disease outbreaks, environmental poisoning or illegal killing of wildlife. Similar to vultures and carcasses, our methodological framework can be applied to other species to locate their respective food resources. It could also be applied to other types of resources like temporary water sources, roosting sites and to other behaviours such as marking to locate marking sites. We showed that a carcass detection system using vultures, loggers and artificial intelligence (AI) can be used to monitor the mortality of numerous species in a vast landscape. This method has broad applications, such as studying the feeding ecology of vultures, detecting and monitoring of disease outbreaks, environmental poisoning or illegal killing of wildlife. Similar to vultures and carca