Abstract
The mining sector is an important source of revenue for the South African economy; however, mining can have a detrimental impact on water quality. Therefore, efficient assessment and monitoring are needed to protect water bodies in mining-related environments. While remote sensing has proven to be an effective monitoring tool in various sectors, efforts must be intensified to apply it in the mining sector in order to combat the impact of mining pollution on water resources. Remote sensing techniques have been successfully used to estimate water quality parameters of inland waters, however, applications focussing on mining environments are rare. There is, therefore, a need to test the capabilities of the technology in mining areas in order to design an efficient water quality monitoring system that will allow relevant authorities to implement mitigation plans and sustain ecosystem services derived from the water bodies. This dissertation has investigated the capabilities of remote sensing in detecting and monitoring water quality parameters in a mining environment along the Mooi River, South Africa. The first objective of the dissertation sought to investigate the performances of raw hyperspectral data and simulated multispectral datasets in quantifying various water quality parameters. Seventy-eight water samples were collected from the study area. Reflectance measurements were taken from each sample using a field-spectroradiometer. The all-subsets regression technique and a support vector machine (SVM) were used to explore the relationships between 17 water quality parameters and hyperspectral datasets, as well as four simulated multispectral datasets (i.e. Landsat Operational Land Imager, Sentinel-2 Multispectral Instrument, WorldView-3 and SPOT 6). The results revealed the usefulness of combining hyperspectral and simulated datasets with different algorithms for effective water quality monitoring. Water quality parameters were estimated with high accuracy using a support vector machine (SVM), compared to the all-subsets regression approach for both datasets (raw hyperspectral and simulated). The second objective explored the accuracy of actual multispectral datasets in detecting water quality in the same river and field data utilised in the first objective mentioned above. The all-subsets regression technique that lists all possible models was applied to estimate the laboratory-measured parameters using reflectance values derived from the individual bands of Landsat OLI, Sentinel-2 MSI, ASTER and SPOT 6 data as explanatory variables. The results demonstrated the potential of multispectral reflectance data in water quality measurements...
M.Sc. (Environmental Management)