Abstract
M.A. (Geography)
Water is a scarce resource in South Africa and the world at large. The quality of freshwater systems in South Africa is deteriorating as a result of anthropogenic influences. To ensure the sustainable use of water resources, directives from various authorities have recognised the importance of promoting good water quality. Traditionally water quality assessments have relied on in-situ observations. These are relatively expensive, time-consuming and labour-intensive, resulting in spatially and temporally inadequate databases. There is a need for the implementation and integration of sophisticated technologies and methodologies in water quality monitoring programmes. Remote sensing techniques have been applied to characterising the quantity and quality, as well as to monitoring geographical distribution of aquatic environments.
This dissertation aims to use remote sensing products and techniques in assessing and monitoring the quality of resources. The first objective of the dissertation was to explore the utility of hyperspectral and simulated multispectral data in discriminating water hyacinth from adjacent grass cover in the Hartbeespoort Dam of South Africa. Continuum-removal analysis was performed on the water hyacinth and grass reflectance spectra to quantify the absorption features of both water hyacinth and grass. Random Forest classification was used to discriminate water hyacinth from grass. The results indicated that the continuum-removal approach was able to characterise the absorption features in water hyacinth and grass spectra. The Random Forest classification method proved to be effective in discriminating water hyacinth from grass using all and key relevant wavelengths as well as simulated datasets. The second objective was to investigate the utility of Random Forest regression along with hyperspectral, simulated Landsat 8 Operational Land Imager (OLI) and actual Landsat 8 OLI datasets in estimating chlorophyll-a (chl-a) and turbidity based on in-situ sampling regime in the Hartbeespoort Dam of South Africa. Seventy in-situ samples of surface water were collected from the study site. Measurements of chl-a, turbidity and reflectance were taken from each sample using a hand-held spectrometer. The Random Forest regression method was used to build relationships between chl-a and a range of datasets including hyperspectral, simulated Landsat 8 OLI and actual Landsat 8 OLI. The results demonstrated the potential of...