ABSTRACT
The rapid pace of human-related development and environmental instability has placed significant pressure on
global ecosystems, threatening the existence of numerous threatened wildlife species, including birds. Accurate
and detailed monitoring can play a crucial role in safeguarding biodiversity, supporting conservation efforts and enhancing ecosystem management. Traditional monitoring methods, such as direct observation and manual image analysis, are labor-intensive, prone to bias, and often inadequate for large and densely populated breeding colonies. In this study, we developed a fully automated deep-learning-based algorithm to identify, count, and map breeding seabirds in a large and densely populated mixed breeding colony containing breeding pairs of two visually similar species, the regionally-threatened Common Tern (Sterna hirundo) and the Little Tern (Sternula albifrons). Using YOLOv8 for initial object detection using remote-controlled cameras, we enhanced classification performance by integrating ecological and behavioral features, including spatial fidelity, movement patterns and size, through camera calibration techniques. Our algorithm successfully identified, counted, and mapped breeding individuals from both species with an average discrepancy of only 2 % compared to manual counts, and achieved over 90 % accuracy in correctly identifying the species of nesting individuals from both species. By providing high-resolution spatial mapping of nesting individuals, the system also offers valuable insights into habitat use and intra-colony dynamics. Additionally, incorporating behavioral tracking via video analysis allows for a more accurate differentiation between nesting and non-nesting individuals. Our methodology represents a significant advancement in automated wildlife monitoring by integrating artificial intelligence for automatic counting, mapping and classifying birds to enhance our understanding of ecological processes and to aid conservation. This study presents a robust, automated framework for seabird colony monitoring that minimizes human disturbance while maximizing accuracy and efficiency. By leveraging deep learning and ecological knowledge, this approach can revolutionize conservation monitoring, offering a scalable and cost-effective solution for tracking wildlife populations in an era of rapid environmental changes.
.