Search for a command to run...
The health of marine ecosystems is under unprecedented threat from human-induced pressures such as habitat degradation overfishing and climate change. Monitoring fish biodiversity is essential for assessing ecosystem status and informing conservation efforts, particularly within Marine Protected Areas (MPAs). However, traditional underwater visual census (UVC) methods are labour-intensive, limited by diver availability, and subject to observer bias. This body of work presents an integrated series of studies aimed at automating underwater biodiversity assessments using deep learning and computer vision techniques applied to diver-operated video (DOV) data.The first study evaluates the feasibility of fully automated species detection using a YOLOv7-based object detection pipeline trained on Mediterranean fish species, demonstrating strong agreement with manual analysis and UVC data. The second study extends this by developing novel fish counting approaches that significantly improve upon the commonly used Nmax method, offering reliable abundance estimates across different ecological niches. The third study applies these automated tools to multi-year datasets across various conservation zones in the French Riviera, assessing spatio-temporal trends in species presence and abundance under different environmental conditions and protection levels. Building on the core pipeline, the fourth study applies these methods to a different ecological challenge, the detection and monitoring of non-indigenous species (NIS). Using annotated video transects from the Kaş-Kekova Marine Protected Area a different model was trained to distinguish native and invasive fish species. This allowed for the automated estimation of NIS percentages across habitats and provided early insights into the extent and spread of biological invasions across sites and substrates.Together, these studies establish a robust, scalable framework for ecological monitoring that minimises human bias, enhances data reproducibility, and enables high-resolution, long-term biodiversity assessments. The approach offers a transformative step towards modernising marine monitoring programs, supporting data-driven conservation and sustainable management of vulnerable marine ecosystems.