Mapping Red Dust Dispersion in Saldanha Bay: Machine Learning Classification Using Open-Source Geospatial Tools

Abstract. Saldanha Bay, situated on South Africa’s West Coast, faces persistent environmental challenges due to red dust pollution from industrial activities, including operations at Transnet’s iron ore terminal and the former Saldanha Steel Plant. This study examined the spatial and temporal distribution of iron ore dust between June 2017 and December 2024 using Sentinel-2 satellite imagery and machine learning classification through the Dzetsaka plugin in QGIS. By applying Random Forest (RF) classification across five distinct land cover classes, this study identified significant seasonal variations in dust dispersion, driven by climatic factors such as strong coastal winds and the Mediterranean climate’s dry summers and wet winters. Although RF was the primary classification method employed in this study, its selection was informed by extensive literature highlighting its robust performance in classifying complex land cover types using Sentinel-2 imagery. Compared to other classifiers such as Support Vector Machines (SVM) and k-Nearest Neighbour (k-NN), RF is recognised for its resilience to overfitting, robustness to noisy data, and ability to generalise well in heterogeneous urban-industrial landscapes. These documented advantages in recent studies support the adoption of RF as the most suitable classifier for this research context, particularly given the diverse spectral characteristics of the Saldanha Bay region. The findings highlight the environmental impact of industrial activities and demonstrate the practical application of Free-and-Open-Source GIS (FOSSGIS) tools in environmental monitoring, contributing to effective industrial pollution control in coastal regions.