Geovariances is to talk for the first time at DIGMIN 2025

Multivariate geostatistics and machine learning: complementary tools in modeling mineral resources

This presentation explores how geostatistics and machine learning complement each other. Through a real-world case study, Pedram will highlight multivariate geostatistical techniques, including Gaussian Mixture Models (GMM), imputation, cosimulations, and the Probabilistic Prediction Model (PPM). He will demonstrate how combining these approaches with machine learning can create new opportunities in resource modeling.

Abstract – Theories of geostatistics and machine learning were both developed in the 20th century. The former originates from earth sciences schools, while the latter has its roots in statistics and mathematics. Geostatistics offers a numerical framework for modeling mineral grades and zones through variographic analysis, kriging interpolation, and geostatistical simulations. It focuses on interpolating mineral grades in the (x, y, z) space, conditioned on samples and a variogram model. In contrast, machine learning estimates target (output) variables from input features by identifying and leveraging statistical relationships in the data.
While geostatistics has long been a standard in mineral resource modeling, machine learning is rapidly gaining adoption. Rather than viewing them as rivals, this presentation highlights their complementarity through a case study. It also introduces multivariate geostatistical techniques such as Gaussian Mixture Model (GMM), imputation, cosimulations, and the Projection Pursuit Multivariate Transform (PPMT).