Abstract
Accurately defining estimation domains is a crucial step in Mineral Resource Estimation (MRE) since it determines the location of the mineralization under the ground. The identification of groups of ore with high concentrations of the target metal helps to select the appropriate mining method that considers the necessary selectivity to extract the ore without diluting it with host rocks.
In this study, I utilized K-means, Gaussian Mixture Models (GMM), and Geostatistical Clustering (GC) techniques to cluster isotopic data containing concentrations of Ni, Zn, Cu, Cr, Pb, and As. The objective is to achieve spatially continuous or connected clusterings where elements within each group were as similar as possible, and the groups themselves were as different as possible. Those are the required conditions to define estimation domains (clusters or groups) in mineral resource estimation (MRE).
To determine the optimal number of clusters, I utilized the Elbow method. I evaluated the quality of the clusters by measuring the variance within and between clusters and the total cluster entropy. The results revealed that GMM provided the highest quality clusters among all the methods used. However, it should be noted that GMM is not a spatial clustering technique and may not consistently outperform other methods always.
Surprisingly, despite being a spatial clustering technique, the Geostatistical Clustering method performed poorly. This may be attributed to the fact that this implementation only considers a single continuous variable. It may be necessary to modify the method to account for multiple variables to improve its performance.
In this study, I utilized K-means, Gaussian Mixture Models (GMM), and Geostatistical Clustering (GC) techniques to cluster isotopic data containing concentrations of Ni, Zn, Cu, Cr, Pb, and As. The objective is to achieve spatially continuous or connected clusterings where elements within each group were as similar as possible, and the groups themselves were as different as possible. Those are the required conditions to define estimation domains (clusters or groups) in mineral resource estimation (MRE).
To determine the optimal number of clusters, I utilized the Elbow method. I evaluated the quality of the clusters by measuring the variance within and between clusters and the total cluster entropy. The results revealed that GMM provided the highest quality clusters among all the methods used. However, it should be noted that GMM is not a spatial clustering technique and may not consistently outperform other methods always.
Surprisingly, despite being a spatial clustering technique, the Geostatistical Clustering method performed poorly. This may be attributed to the fact that this implementation only considers a single continuous variable. It may be necessary to modify the method to account for multiple variables to improve its performance.