Testing and evaluating non-extractive sampling platforms to assess deep-water rocky reef ecosystems on the continental shelf

Traditional extractive sampling methods, such as netting and trawling, to assess benthic species diversity, size and abundance are unable to sample complex hard substrates, e.g., rocky reefs. This inability led to the development of alternative non-extractive sampling platforms, such as digital stills and video cameras mounted onto stationary or moving platforms. This thesis examined two moving platforms, autonomous underwater vehicle (AUV) and towed video platform and one stationary platform, stereo baited underwater video systems (BUVS). These platforms were used as sampling tools to assess reef fish diversity, size distribution and absolute and relative abundance in complex deep-water (30-100 m) rocky reefs in temperate Australia (Tasmania). Each platform was evaluated with respect to their efficiency and reliability within a sustainable resource management framework. A novel feature extraction routine, using colour, texture, patch gap summaries and rugosity, to semi-automatically classify AUV images into habitat classes is proposed. Here, the randomForest classification tree algorithm was used to assign habitat classes to images after initial training (i.e., 500 images annotated by a trained human expert). Classifier accuracy was assessed using this human scored image set. Habitat prediction accuracy was 84% (with a kappa statistic of 0.793). The evaluation of stereo BUVS as a tool to inventory and monitor deep-water temperate reef fish diversity can inform resource managers of advantages and disadvantages of this particular sampling platform. Species richness and relative abundance across different survey sites over two years were investigated. In addition, stereophotogrammetric fish length estimation of two commercially important species, striped trumpeter (Latris lineata) and blue-throated wrasse (Notolabrus tetricus), were utilised to set a benchmark for future reference and compared to line-fishing (L. lineata) and trapping data (N. tetricus). All three platforms were compared to evaluate their ability to assess reef fish diversity and abundance. Sample variability for each tool was assessed statistically and synergy between platforms proposed. The cost-effectiveness of each platform was assessed qualitatively. An assessment of the size and abundance distribution of the ocean perch Helicolenus percoides was conducted using photographic records taken by the AUV. Stereophotogrammetric size estimates were converted into biomass and examined with respect to depth and habitat types. Habitat preferences of adult and juvenile ocean perch were also investigated. The results suggest that AUV Sirius is a mature survey platform in complex hard substratum environments. The utility of this nonextractive sampling tool in a fisheries context is discussed. Non-extractive imagery-yielding sampling platforms provide useful alternatives when sampling complex environments. Data quality, derived from imagery, is benefitting from rapidly developing technology, e.g., high-definition video and megapixel digital cameras. Non-extractive methods provide the only means to sample marine protected areas. Advantages and disadvantages of each platform are now readily accessible to advise resource management agencies.