Abstract
The tailings storage facilities (TSFs) associated with abandoned mine sites are characterised by sparse vegetation, erosion features, deposition and are surrounded by contaminated water. The TSFs' abandonment results from the poor management of mine sites, which can occur due to sudden or temporary mine closures. The TSFs associated with abandoned mine sites tend to lack the development of flora owing to the toxicity of their metal content. These TSFs tend to cause environmental impacts, such as soil, water and air pollution. The contamination of soil, water and air can result in damage to aquatic life, animals, plants and threaten human health. It is not clear what the quantity of abandoned TSFs is in the Gauteng Province of South Africa. It is for this reason that it remains challenging to estimate the extent of the hazard posed by the abandoned TSFs. This study aims to identify and quantify the number of the abandoned TSFs in the Gauteng Province. The study used remote sensing techniques to understand the TSFs spatial features and classify them. Sentinel 2A and B satellite imagery were used to classify the reflected TSF spatial features and understand their distribution. The Google Earth Engine (GEE) was used to process and analyse satellite data and allowed the users to access and analyse remotely sensed imagery without the need to download the data onto local computers. The GEE computing programme was utilised to perform machine learning (ML) supervised classification to classify Land Use and Land Cover (LULC), intending to identify the different types of TSFs. Three types of TSFs were identified: namely active, dormant and abandoned TSFs. The active TSFs are marked by ongoing deposition and growing plantations, while the dormant TSFs have no active deposition from mining operations and feature mature plantations. They were found using coordinates and characteristics. The abandoned TSFs typically exhibit sparse vegetation, polluted water sources and pronounced erosion gullies. This paper shows how the integration of ML and satellite imagery data can assist in mapping and categorising TSFs in the West Rand of the Gauteng Province. The detection of active, dormant and abandoned TSFs with outstanding spatial accuracy was made possible by the combination of Sentinel-2 MSI data and ML algorithms (RF and SVM). Using shortwave infrared (SWIR) and NIR bands, the abandoned TSFs are often obscured by partial vegetation or misclassified due to spectral similarities, which were effectively identified, revealing their environmental impact and proximity to at-risk communities. In terms of overall accuracy, the RF model performed better than the SVM, especially when it came to differentiating between the various types of surrounding land cover and different TSFs. The significance of scalable, data-driven methods for environmental monitoring, adherence to regulations and rehabilitation priority is highlighted by these findings. This paper helps create a more open and proactive structure for handling South Africa's mining heritage in accordance with global tailings governance standards.