The mining industry generates billions of tonnes of waste annually, which is often stored in tailings storage facilities (TSF). This waste is generated from the extraction of ore from surface or underground mines, as well as from metallurgical processing and low-grade stockpiles. TSF can have significant environmental impacts, as they can cause acid mine drainage resulting in the leaching and transport of heavy metals into ground and surface waters. With increasing demand for critical raw material, recent studies have shown that the valorisation of mine waste can be a potential secondary source of critical raw materials. The valorisation of mine waste is possible when the waste is accurately characterised.A novel method that uses multispectral satellite remote sensing and machine learning to estimate the mineral resource in a defunct TSF in the Witwatersrand Basin, South Africa is proposed in this research. Four machine learning models: 1) random forest (RF); 2) adaptive boosting (AB); 3) extra trees (ET); and 4) k-nearest neighbours are developed using supervised machine learning. The models are trained using training data acquired from a TSF with known gold concentration located 3 kilometres from the TSF and deployed on the TSF to predict the gold grades. The results of the machine learning model predictions indicates that machine learning models had high performances for predicting gold grades in the TSF. The AB, RF and ET, models performed best. Their performances were evaluated using the coefficient of determination (R2) value. The R2 values for the machine learning models were 0.95, 0.92, 0.87 and 0.70 for AB, ET, RF and kNN respectively. The mean gold grade predicted was 0.44 g/t by all machine learning models. This was compared to a 2D surficial geostatistical model which estimated 0.35g/t gold in the TSF using ordinary kriging and a 2D vertically averaged geostatistical model with an estimated 0.4 g/t mean gold grade. The short-wave infrared (SWIR) - band 11 at a 20 m spatial resolution had the highest correlation with the reflectance of gold in the TSF. This study demonstrated the value of leveraging multi-spectral remote sensing data and machine learning to perform mineral resource estimation in defunct TSF.
| Identifer | oai:union.ndltd.org:UPSALLA1/oai:DiVA.org:uu-496119 |
| Date | January 2023 |
| Creators | Agard, Shenelle |
| Publisher | Uppsala universitet, Institutionen för geovetenskaper |
| Source Sets | DiVA Archive at Upsalla University |
| Language | English |
| Detected Language | English |
| Type | Student thesis, info:eu-repo/semantics/bachelorThesis, text |
| Format | application/pdf |
| Rights | info:eu-repo/semantics/openAccess |
| Relation | Examensarbete vid Institutionen för geovetenskaper, 1650-6553 ; 588 |
Page generated in 0.0026 seconds