Scholarlink

<p>Geological domain delineation is a critical step in mineral deposit modeling, yet it often depends on subjective interpretation and sparse data. This study proposes an objective workflow to delineate geological domains using unsupervised learning, quantify assignment uncertainty, and evaluate the cut-off sensitivity of contained gold in a gold deposit in northern Peru. A total of 2,129 8-m composites from diamond drilling were analyzed using spatial coordinates and winsorized gold grades, complemented by neighborhood-derived attributes: local mean, local standard deviation, and local vertical trend, which capture local background, heterogeneity, and a smooth local vertical trend. Three clustering approaches were compared: K-Means, Gaussian Mixture Models (GMM), and Agglomerative Clustering (AC), with K = 2 selected as the most consistent partition based on both the Elbow and Silhouette methods. K-Means provided the most operationally interpretable and statistically contrasted domaining, achieving Silhouette = 0.285, Davies-Bouldin Index (DBI = 1.564), and Calinski-Harabasz Index (CHI = 559.3), and the strongest grade separation (Kruskal–Wallis = 664.51, <i>p</i> &lt; 0.001). GMM and AC produced less favorable trade-offs between domain balance and discrimination. Uncertainty mapping highlighted coherent domain cores and transitional zones, supporting soft-boundary interpretation. Finally, cut-off screening showed that the higher-grade domain consistently concentrates most of the contained gold above threshold grades, reinforcing the practical relevance of the domaining derived for early-stage decision support. Overall, the results show that, for early-stage exploration and mine-planning applications, a simple centroid-based approach such as K-Means can outperform more complex probabilistic models by providing stable, interpretable, and operationally meaningful geological domains.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Geological domain delineation, uncertainty quantification, and cut-off-based contained gold screening using unsupervised learning in a Peruvian gold deposit

  • 期刊论文

  • 发表日期：

    2026年4月17日

查看全文
  • Marco A. Cotrina-Teatino,
  • Jairo J. Marquina Araujo

摘要

Geological domain delineation is a critical step in mineral deposit modeling, yet it often depends on subjective interpretation and sparse data. This study proposes an objective workflow to delineate geological domains using unsupervised learning, quantify assignment uncertainty, and evaluate the cut-off sensitivity of contained gold in a gold deposit in northern Peru. A total of 2,129 8-m composites from diamond drilling were analyzed using spatial coordinates and winsorized gold grades, complemented by neighborhood-derived attributes: local mean, local standard deviation, and local vertical trend, which capture local background, heterogeneity, and a smooth local vertical trend. Three clustering approaches were compared: K-Means, Gaussian Mixture Models (GMM), and Agglomerative Clustering (AC), with K = 2 selected as the most consistent partition based on both the Elbow and Silhouette methods. K-Means provided the most operationally interpretable and statistically contrasted domaining, achieving Silhouette = 0.285, Davies-Bouldin Index (DBI = 1.564), and Calinski-Harabasz Index (CHI = 559.3), and the strongest grade separation (Kruskal–Wallis = 664.51, p < 0.001). GMM and AC produced less favorable trade-offs between domain balance and discrimination. Uncertainty mapping highlighted coherent domain cores and transitional zones, supporting soft-boundary interpretation. Finally, cut-off screening showed that the higher-grade domain consistently concentrates most of the contained gold above threshold grades, reinforcing the practical relevance of the domaining derived for early-stage decision support. Overall, the results show that, for early-stage exploration and mine-planning applications, a simple centroid-based approach such as K-Means can outperform more complex probabilistic models by providing stable, interpretable, and operationally meaningful geological domains.