Computational analysis of reef structure and benthic composition

Structural complexity is vital to coral reef systems and is threatened by direct and indi- rect anthropogenic acts, resulting in a global decline of reef health. As such, these systems are regularly assessed and monitored. The use of computational techniques in surveys is becoming more commonplace, but is an intensive process and often only covers small areas of a reef. Non-coral reef systems are rarely considered in terms of their structural com- plexity and do not have standardised metrics or methodologies, nor are structural links to health known. They are monitored in terms of biodiversity, but often structure and ben- thos are excluded from these assessments. To improve the capabilities of computational analysis of benthic environments, this thesis first developed and tested a multi-camera array for surveying of coral reefs, then utilised an array in a survey of a distinct rocky reef environment. The surveying method demonstrated a rapid approach to data collection that successfully generated 3D modelled environments that enabled successful data ex- traction. The complexity metrics used, however, appeared to be disconnected from in-situ observations of the rocky reef environment. From this, a novel assessment of relief and the use of tailored fractal dimension was developed to contextualise rocky reef complexity and link it to commercially relevant associated species. From this, links between structural complexity and associated species were found, and relief proved to be a distinct form of complexity on the rocky reef. Additionally, the development of the ImageCLEFcoral task is outlined: a global benchmarking competition which aimed to further the capabilities of automatic annotation of coral reef substrates. A series of submitted runs were produced for the 2021 edition of the task and are also presented here. Though the annotation ap- proaches are not yet successful enough for practical use, the annual task has continued to develop and will likely improve year on year. When fully utilised, computational analysis would enable researchers to focus on furthering knowledge rather than surveying restric- tions and repetitive data extraction. While the approaches presented here do require in-situ data collection and researcher interaction, the benefits lend themselves to widely expanded analysis capabilities and shareable data stores. Multi-camera arrays allow for data collection to span larger areas in less time, while automatic annotation increases the data volume without manual intervention. When used appropriately, research can become more efficient, cost-effective, replicable, and accurate.