| description abstract | The Global Industry Standard on Tailings Management (GISTM) promotes performance-based approaches in geotechnical assessments. Hence, characterizing the spatial variability of deposited tailings is expected to be a key input for some tailings storage facilities (TSFs); however, it has seldom been investigated. In this study, we assess the spatial variability of thickened and conventional tailings, which have been deposited into the same TSF, providing a unique opportunity to investigate two tailings technologies. A dense array of 15 cone penetration tests (CPTus) with an average offset of 1.5 m has been conducted to collect data. In addition to evaluating the spatial variability, the collected information is also used to assess the potential of machine learning (ML) for detrending when deriving random fields. Using a new proposed stationarity score, we find that an ML-based detrending outperforms traditional procedures for most scenarios. In terms of correlation lengths, we find similar ranges for thickened and conventional tailings (vertical: δwv=0.2–0.6 m, horizontal δwh=1.5–4.5 m) and similar distributions, likely influenced by the depositional processes. In contrast, the variance in the conventional tailings is higher, which we attribute to its segregating nature. Finally, by inspecting previous studies on natural soils, we find that the variability of mine tailings (δwh/δwv=2–21) resembles that observed in alluvial deposits, which we attribute to the parallels in the depositional processes. | |
