As part of the Marine Biodiversity Hub this study tested and evaluated imagery-
yielding samplers that are non-extractive and quantitative. An advantageous
characteristic of imagery-yielding samplers is the permanent record and auxiliary information contained in the imagery, e.g., the target species AND its environment.
Testing and evaluation was conducted under the following criteria: (1) cost-
effectiveness with respect to collecting an inventory of habitats and monitoring fragile and/or protected environments, applicability to existing needs, e.g., fisheries- independent stock assessment, MPA planning and monitoring and habitat mapping. Three different sampling tools were tested to address several challenges with respect to the current EBM challenges outlined above.
Advanced habitat mapping techniques – chapter 3 (Seiler et al., doi.10.1016/j.csr.2012.06.003)
Almost 90% of Australia’s EEZ remains to be bathymetrically mapped. However, within this 200 nm EEZ there are several geological features, such as the continental slope, seamounts, abyssal plains and canyons that are less well known than the
continental shelf. Currently, multibeam echo sounders (MBES) are the only means to efficiently fathom Australia’s continental shelves, especially areas with high conservation value such as Australia’s MPAs. Most notable areas that were mapped using scientific multibeam echo sounders are Jervis Bay, NSW (Anderson et al., 2009) and the Freycinet and Huon Commonwealth Marine Reserves, TAS (Nichol et al., 2009). However, the resolution and information provided by MBES alone is insufficient at the habitat scale – the scale at which EBM [ecosystem-based
management] operates. At the habitat scale (area that comprises ecologically
linked multi-species assemblages, such as kelp, invertebrates and fish in a kelp forest habitat), imagery-yielding samplers are the only means of mapping benthic habitats on hard substrates such as sponge gardens or kelp forests on rocky reef
(Copeland et al., 2011). However, there are examples where Regional Marine
Planning is based on geomorphic features, such as continental rise, pinnacle, canyon, terrace, trench/trough, etc, which are derived from grid-based terrain analysis using multibeam bathymetry data (Harris et al., 2003). Two dominant methods of producing marine habitat maps of hard substrates are currently in use (i) seafloor images (point samples) combined with continuous interpreted MBES data and (ii) transect or full-area photographic surveys. The former uses machine-learning algorithms to establish relationships between topographic attributes, obtained from terrain analysis using digital elevation models (MBES data), and distinct habitat classes, annotated seafloor images, to predict habitat distribution outside the photographed area (Rattray et al., 2009). The latter, photographic surveys, are solely based on imagery collected by the sampler (Singh et al., 2004b). Both methods
require annotation of imagery by a trained expert, which is time-consuming and often subjective. A widely used habitat classification scheme based on substratum
type, requires classification based on a primary (> 50% coverage) and secondary
(>20% coverage) substratum type (Greene et al., 1995). Subjectivity, or observer
bias, can cause coverage estimates to differ by 20% (personal communication Mark Green, CSIRO). However, image annotation time can be significantly reduced and observer bias eliminated using computer vision techniques. By combining several computer vision techniques, such as edge and scene detection, assigning habitat classes to images based on image content can be automated. Once the automation routine is set up it only takes a few seconds to classify additional images thereby increasing statistical power and precision (Purser et al., 2009). Chapter 3 describes a method to automatically classify seafloor images into habitats based on a training data set.
Effective sampling beyond diver’s depths – chapter 4
Underwater visual census is commonly used to monitor temperate marine protected areas (Barrett and Buxton, 2002). However, high quality optical surveys are needed to monitor MPAs beyond the range of safe SCUBA diving operations (Singh et al.,
2004a). For example, only 6% of the Great Barrier Reef Marine Park can be
safely monitored using SCUBA (Cappo et al., 2003). However, within depth ranges encountered on continental shelves, remote or tethered camera platforms are free
from depth restrictions and serve as reliable samplers in deeper depths (> 30 m).
data quality as good, or better than those provided by SCUBA divers in shallow depths. Assis et al. (2007) found that their towed video platform could assess a larger protected area with respect to number of observed elasmobranch species and individuals within the same time compared to UVC. However, Colton and Swearer (2010) found that UVC recorded more individuals (fish species), higher richness at species and family level than Baited Underwater Video Systems (BUVS). Chapter 4 tested the hypothesis whether BUVS are an equivalent to underwater visual census in deeper waters with respect to reef-fish assemblage composition, species richness and abundance and size structure. Further chapter objectives include a test whether a new relative abundance index based on stereophotogrammetric fish length measurements to identify individuals by length is superior to the current
relative abundance standard MaxN – maximum number of individuals of species
x in videoframe y and to develop a novel statistical approach to conduct power
analysis using count data, for which the common assumption of normality do not apply.
Non-extractive fisheries-independent stock assessment – chapter 5 (Seiler et al., doi:10.1016/j.fishres.2012.06.011)
Traditional fishery resource assessment methods using extractive trawl gear are unable to sample rocky substratum and are prone to underestimate the biomass
of species having partial or strong association with rocky reefs. The ocean
perch Helicolenus percoides, a species with strong association with rocky reefs was
and Bax, 2001). Non-extractive imagery-yielding alternatives, that can sample rocky substratum include manned submersibles (Yoklavich et al., 2000) and autonomous underwater vehicles (Tolimieri et al., 2008). Tolimieri et al. (2008) report rosethorn rockfish (Sebastes helvomaculatus) densities and habitat preferences over different substrata, i.e., rock, sand and mud in depth ranging from 100 – 300 m based on digital images taken by an AUV. One major advantage of camera platforms over trawl gear is the ability to observe species-habitat interactions. For example, trawl gear usually samples several kilometers of seafloor, thereby traversing several habitat types, however, the trawl catch comprises only fish and bycatch and provides no information as to where a particular fish was caught. In contrast, images capture fish in their natural environment and therefore allow
species-habitat investigations. Chapter 5 details the use of the stereo-camera
onboard the autonomous underwater vehicle Sirius to collect fisheries-independent
complementary data, such as abundance, size structure and habitat preferences of
the ocean perch Helicolenus percoides. More specifically, I tested the hypothesis
whether annotated, geo-referenced digital images taken by the AUV Sirius can
provide data required for ocean perch stock assessments under constraints of spatial autocorrelation and untrawlable terrain, i.e. rocky reefs.
Efficiency testing three non-extractive samplers – chapter 6
Several non-extractive samplers are available to resource managers to inventory and monitor protected or restricted marine areas. From a resource management perspective these samplers should be cost-effective, easy to deploy and applicable
to various management objectives. One management objective, to halt the decline of biodiversity, encompasses enumeration of individuals (abundance) and species (species richness). In order to halt or reverse biodiversity decline an understanding of fish assemblage distribution over various spatial and temporal scales is essential. BUVS studies in three marine parks in New South Wales, Australia by Malcolm et al. (2007) found that total species richness of fish assemblages did not follow the latitudinal gradient phenomenon and that the temporal component (5 yr) is small compared to the spatial component. Other imagery-yielding platforms such as towed video and AUVs are potentially useful to assess temporal and spatial differences
in fish assemblage composition. Chapter 6 tests three non-extractive imagery-
yielding samplers and their ability to efficiently assess abundance and diversity of fish assemblages on temperate rocky reefs.