Spatial uncertainty estimation techniques for shallow coastal seabed mapping.

Techniques for seabed habitat mapping have developed considerably over the past 10 years, principally through advances in acoustic technologies and tools for improved spatial analysis. The increasing need for information on the distribution and structure of seabed habitats in coastal waters (< 50 m deep) has developed in Australia due to increasing pressures from development and exploitation. A clear understanding of the uncertainties in creating spatial information from marine data is required within seabed mapping projects, particularly those using acoustic methods that vary in coverage and resolution. This thesis investigates methods to explore spatial uncertainty in the techniques used to characterise shallow water seabed habitats using local and regional scale case studies ranging from interpolation of sediment cores to classifying digital elevation models generated from multi-beam acoustic data. Uncertainties are investigated in a multidiscipliary approach to habitat mapping. Broad-scale and fine-scale mapping of marine seabed habitats can provide considerable information on patterns of physical seafloor structuring and this is now achievable given recent technological advances in echosounders and backscatter analysis, digital underwater video, differential GPS and Geographic Information Systems (GIS). The uncertainties in classifying single beam acoustic data are examined by comparing data visually classified into habitat classes in real time compared to those defined through post-processing in the laboratory. The influence of a range of transect spacings on qualitative knowledge-based interpolation of shallow rocky reef acoustic data is assessed. Parameters of physical reef characteristics from single beam acoustic data are investigated which aid in improving kriging interpolation techniques. A fuzzy classification method is applied to sediment core data collected over 100s of kms in order to map the distribution of unconsolidated sediments. This technique is tested on Australia's SeaScapes data. The SeaScapes layer was recreated with classes derived from an unsupervised fuzzy clustering algorithm. A sensitivity analysis was performed which explores the optimal number of clusters and clearly shows where classes overlap, highlighting confusion and therefore potential classification errors in the data. Conditional simulation was utilised to explore uncertainty in the interpolated data layers and animations produced which provided an advanced way of visualising the surfaces. Image segmentation techniques are applied at various values of splitting and merging thresholds to identify objects in sidescan sonar imagery. Grey Level Co-occurrence Matrices (GLCM) texture measures are integrated with segmentation to identify homogeneous texture regions in an image. The segmentation and object oriented classification provide good classification results in 10-40 m water depth with accuracy values of >80 % when classifying two classes and >60% when classifying three classes. This section of research focuses on the analysis of seabed habitats through the use of algorithmic calculations at multiple scales to quantitatively delineate distinct seabed regions based on texture parameters. The textural characteristics of three habitat classes are explored and tested onsidescan sonar data in two case studies. Segmentation results are validated using underwater video transects and statistical techniques. The classified sidescan acoustic images are spatially characterised using fragmentation statistics. These results are a leap forward providing a methodology and a terminology to describe the distribution of shallow rocky reef, combined with a classified map and an uncertainty map generated by the object oriented technique. Fuzzy classification techniques are used to characterise the two dimensional structure of shallow rocky reefs from multi-beam bathymetric data. The results from two morphometric classification techniques are contrasted and compared. Many physical and biological processes acting on the seabed are highly correlated with bathymetric features, such as ridges and channels. Examples of these include the selection of habitat by many commercially fished species, such as rock lobster, abalone and reef associated fish species. These physical attributes can therefore often be key predictors of habitat uitability, community composition and species distribution and abundance. These methods greatly improve insight into classification and related uncertainties of morphometric classification. There are many potential applications of seabed habitat mapping for which estimates of uncertainty will provide additional crucial information. As habitat mapping becomes more common in Australian coastal waters and datasets build up over time, compatibility between different sets of information will be essential. Monitoring and change detection analysis requires detailed baseline data with uncertainty estimates to ensure that features that display change are reliably detected. The accuracy of marine habitat maps and their associated levels of uncertainty are extremely hard to convey visually or to quantify with existing methodologies. The new techniques developed in this research provide a rigorous alternative to current practices. The methods used in this research integrate existing techniques in a novel way to improve insight into classification and related uncertainty for seabed habitat maps which will progress and improve resource management for Australian coastal waters.