Three-dimensional modelling of coral reefs for structural complexity analysis

Coral reefs are some of Earth’s most biodiverse and economically valuable ecosystems. Simultaneously they are among the most threatened by anthropogenic factors including global climate change. Their unique three-dimensional (3D) structural complexity is part of what enables them to provide their ecosystem services. It strongly affects species richness, abundance, and other indicators of ecosystem health. This thesis explores the relationship between coral reef 3D structural complexity and ecosystem features. It has developed a new, low-cost method for creating and analysing photogrammetric 3D models of shallow reefs from diver-held camera footage. 3D models are analysed at scales 1-175 cm in terms of point-to-point distances, linear rugosity (R), fractal dimension (D), and vector dispersion (1/k). The 3D models’ accuracy and precision were determined by comparisons with ground truths. The 3D models have root mean square errors of 1.35-1.48 cm in the X, Y and Z dimensions. Values of R from the 3D models were 86.8% accurate compared to in-situ chain-and-tape measurements. Values of D and 1/k were 86.9-99.6% accurate compared with ground truths from 3D printed objects modelled underwater. Data collected around Utila, Bay Islands, Honduras in the Caribbean showed that 3D metrics automatically calculated from the 3D models had the same predictive power for fish abundance and diversity as the more traditional Habitat Assessment Score (HAS). Like HAS, the 3D metrics explained 12-34% of variation in the fish data. A controlled experiment furthermore tested how 1/k affected sessile epibenthic organism settlement around Utila after one year. Results from approximately 200 3D printed recruitment tiles showed that 1/k significantly affected algae settlement, but not coral spat, polychaete, sponge, or bryozoan settlement. The results suggested that the surfaces of artificial reefs can be designed to minimise algal recruitment and that the availability of sheltered, reef-facing area influences epibenthic settlement more strongly than 1/k at the 1 cm scale. Finally, a convolutional neural network (CNN) learned patterns between the 3D models and fish data with just 85 data points. The CNN is a promising approach for analysing larger data sets without 3D metrics. We suggest 3D models become a standard approach for measuring reef structural complexity. Not only can they explain as much variation in fish abundance and diversity as traditional measurements, but also they can nondestructively produce a variety of 3D metrics at numerous spatial scales and keep a permanent record of reef structure over time.