AI-enhanced clustering of mine tailings using Geostatistical data augmentation and Gaussian mixture models.

The reprocessing of mine tailings presents a valuable opportunity to recover critical raw materials essential for advancing green technologies and achieving sustainable resource management. However, conventional mineral resource estimation techniques-designed for primary ore deposits with well-defined geological domains-are difficult to apply to tailings, which typically lack geological structure and exhibit highly irregular spatial patterns. To address this challenge, we propose an artificial intelligence-based framework that integrates geostatistical data augmentation with Gaussian Mixture Models (GMM) to define compact, spatially contiguous estimation domains in mine tailings. The original geochemical data obtained from exploratory drillholes at a tailing deposit in East Kazakhstan were spatially augmented using Ordinary Kriging over a regular 3D grid, enhancing both spatial resolution and continuity of the dataset. Multiple GMM covariance structures are evaluated using spatial and statistical performance metrics. Results from this tailing dataset demonstrate that the augmented GMM approach, particularly with tied covariance, yields substantial improvements in spatial coherence (Moran's I = 0.52 vs. 0.29) and cluster compactness (Silhouette Index = 0.59 vs. 0.41), compared to conventional borehole-only clustering. The proposed method offers a robust and scalable solution for AI-assisted domain modeling in Earth system residues and supports the design of selective reprocessing strategies aligned with sustainable resource development.