Land use/land cover classification and change detection in Umlazi township, KwaZulu-Natal, South Africa (2013–2022) using Landsat 8 and Machine Learning Algorithms

The application of Land Use and Land Cover (LULC) highlights the importance of monitoring natural resources and significantly aids decision makers in planning for urban settings with dense populations. In this context, LULC studies are crucial tools for evaluating flood risk mapping and urban growth, especially in study areas that have not yet undergone LULC changes analysis in the past few years. uMlazi is one of the townships in KwaZulu-Natal, South Africa, that have experienced climate change disasters such as floods due to spatial planning. This study compared three machine learning classification algorithms Random Forest (RF), Support Vector Machine (SVM), and Artificial Neural Network (ANN) Landsat 8 in classifying LULC in uMlazi. Furthermore, to determine the change detection between 2013 and 2022. Change distribution dynamics of each class is reflected in table 6 such as built-up areas from 1.13 km² (19.01%) in 2013 to 2.51 km² (42.14%) in 2022. Barren land saw a reduction of 65.6% due to built-up areas, 2.9% due to grassland, 9.2% due to tar roads, 0.46% due to trees, and 1% due to water bodies. RF achieving the highest accuracy and Kappa coefficient of 0.97 in 2013 and 0.989 in 2022. These findings support improved spatial planning and flood-risk mitigation in rapidly urbanizing townships. Key Words: Land Use and Land Cover; machine learning classification algorithms; Random Forest, Support Vector Machine, Artificial Neural Network and uMlazi