- Title
- Benthic habitat mapping using marine geophysics and machine learning on the continental shelf of South Africa
- Creator
- Pillay, Talicia
- Subject
- Gqeberha (South Africa)
- Subject
- Eastern Cape (South Africa)
- Subject
- Marine geophysics
- Date Issued
- 2021-04
- Date
- 2021-04
- Type
- Master's theses
- Type
- text
- Identifier
- http://hdl.handle.net/10948/52061
- Identifier
- vital:43452
- Description
- A method to map seafloor substrates using machine learning, based primarily on hydroacoustic data including multibeam bathymetry, backscatter, and side-scan sonar, has been developed. The aim was to produce a customdesigned benthic habitat classification method that digitally integrates marine geophysics and biological science data, with relevance to all elements of the local substrate, and this was the first time it was attempted in a South African context. The algorithm developed is able to produce bio-physical benthic habitat maps and this can be extended along the continental shelf of South Africa as new data setsare collected and the algorithm is supplemented. At the outset, this work has focused on broad categories of rock and detailed categories of sediment. Four study sites with varying substrate were selected to holistically build the algorithm that followed a tiered approach of machine learning: Table Bay, Clifton, Koeberg Harbour and Cape St Francis. Table Bay was used to develop a new method of physical seafloor classification, by comparing and contrasting a number of statistical algorithms and software programs. Clifton was used to test the developed clustering algorithm, and Koeberg which is 35 km to the north was used to validate the algorithm because sediment samples, along with drop camera footage, were integrated to better define the results. The resultant verified algorithm was tested at Cape St Francis, where Remotely Operated Vehicle (ROV) footage was acquired in addition to hydroacoustic data. In the first phase of the process towards developing an algorithm, a customised tool was created within ArcGIS using python scripting language to classify seafloor bathymetry, which can be applied to any area of seafloor whatsoever. The tool was based on pioneering work done by the National Oceanic and Atmospheric Administration (NOAA) on a benthic terrain modelling toolbox and adapted to include side-scan sonar data. In the second phase of work, multibeam bathymetry, backscatter and side-scan sonar data that were processed using Qimera, Fledermaus Geocoder Toolbox, and Navlog processing software, were classified using different machine learning techniques including Decision Trees, Random Forests, and k-means clustering computer algorithms. The results from these algorithms were compared to manually-digitised polygons which were created to classify the seafloor substrate distribution by identification of different textures. Integrating all results facilitated a quantitative comparison that illuminated advantages and disadvantages of each machine learning technique and ultimately the k-means clustering techniques were found to be the simplest to implement and understand and worked most efficiently based on their seafloor segmentation capabilities in Table Bay, against all three hydroacoustic data sets (multibeam bathymetry, backscatter and side-scan sonar). In the third phase of work, ground-truthed seafloor characterisation maps were produced for the two study areas of Clifton and Koeberg Harbour. This applied multibeam bathymetry and backscatter data that were collected and processed with machine learning clustering techniques.
- Description
- Thesis (MSc) -- Faculty of Science, Ocean Sciences, 2021
- Format
- computer
- Format
- online resource
- Format
- application/pdf
- Format
- 1 online resource (165 pages)
- Format
- Publisher
- Nelson Mandela University
- Publisher
- Faculty of Science
- Language
- English
- Rights
- Nelson Mandela University
- Rights
- All Rights Reserved
- Rights
- Open Access
- Hits: 958
- Visitors: 891
- Downloads: 0
Collections
NMU School of the Built Environment
| Thumbnail | File | Description | Size | Format |
|---|