2024 – Estimation of TSF Decant Pond Behaviour Using Remote Sensing Data and Unsupervised Machine-learning Algorithms.

Arianna-Onate Paladines, Manuel Diaz, Muhammad Parker, Dale Hone

This paper presents a cost-effective and relatively accurate approach to estimating historical tailing storage facility (TSF) decant-pond areas and, consequently, volume. It is achieved by integrating open-source remote sensing data and employing unsupervised machine-learning techniques. The combination allows an understanding of decant-pond behaviour over time in mine sites lacking monitoring equipment or where monitoring occurs infrequently.

The study focuses on a mine site in Australia; however, the methodology is applicable to any site. Sentinel-2 spectral bands were extracted and analysed using various unsupervised clustering machine-learning algorithms to identify decant-pond extents in a TSF structure from 2020 to 2024. The k-means clustering algorithm along with Near Infrared Reflectance (NIR) imagery proved to be the most efficient in identifying and distinguishing the decant pond under different conditions from the tailings beach (wet and dry) within the TSF. Areas were estimated at a five-day temporal resolution, and the time series area estimates were validated against manual recordings, showing agreement between both datasets.

The case study also shows the method captured the decant-pond behaviour with regards to rainfall. Thus, the proposed method is not only suitable for estimating historic TSF decant-pond volumes and long-term trends but is also necessary for capturing short-term fluctuations due to rainfall response. Although the resulting decant-pond delineation depends on image quality and the absence of obstructed visibility due to cloud cover in the source data, the case study has demonstrated that this methodology could help overcome data limitations. It also provides support for more robust TSF water-balance model calibration, thereby increasing confidence in predictive water-balance models. Ultimately, this leads to more informed decisions, improved TSF risk management practices, and increased awareness.