Leveraging Computer Vision for Efficient and Scalable Biodiversity Monitoring in Marine Ecosystems: A Multi-Year Study on 3 Ecologically Important Fishes

Marine environments face multiple threats, including climate change, overfishing, and shoreline artificialisation leading to a great loss of biodiversity. Therefore, fast and efficient biodiversity assessments are crucial to understand the current and future state of marine ecosystems. Traditionally, underwater visual censuses (UVCs) have served as the primary method for assessing biodiversity, with divers recording species presence, abundance, and other biological indicators along fixed transects. In this study, we collected video transects recorded by camera-equipped divers, resulting in 41h of video footage that enhanced the data collection but also increasing the volume of material requiring tedious manual analysis. This manual analysis of video footage is in a lot of cases financially and logistically unfeasible. Here, we present a comprehensive multi-year computer-vision-based pipeline applied to all recorded video transects to extract ecological information on three ecologically and economically important species (Epinephelus marginatus, Sciaena umbra and Diplodus vulgaris). Using a fully automated computer vision pipeline that enables researchers rapid and less-biased data extraction from large datasets, we confirmed known depth and substrate preferences of the three target species. Additionally, we identified seasonal variations in D. vulgaris that may be disrupted by climate change. Our results demonstrate that computer-vision enables rapid and less-biased data extraction from large datasets, facilitating efficient biodiversity monitoring. We anticipate that this study will serve as a foundation to expand computer-vision applications to other marine regions, species, and environmental scenarios. Increased efficiency in data processing will make more frequent and extensive biodiversity assessments possible, strengthening conservation efforts through improved ecological insights.