Developing remote sensing methodology to distinguish urban built-up areas and bare land in Mafikeng town, South Africa

Application of remote sensing technologies in mapping urban land cover still poses a challenge among urban planners. The aim of this study was to develop remote sensing methodology for distinguishing bare surface and built-up area in Mafikeng, South Africa. Several indices were developed to depict various urban features including NDVI, NDBAI, NDISI, NDWI and NDSI using Landsat 8-OLI data. Different supervised classification algorithms were independently tested to determine their ability in extracting the urban land cover classes. Field survey was conducted to gather ground truth data for accuracy assessment. The classification results proved that KNN was effective in not only increasing the classification accuracy, but also in making the classification of urban land cover features more visible and distinguishable than the other classifiers. The results demonstrate the potential of KNN classifier and combination of several indices to accurately map urban land cover features that can be used as input to land management and urban policy planning decisions.