Stochastic Modelling of Mineral Exploration Targets

• Published in: Mathematical geosciences Vol. 54; no. 3; pp. 593 - 621
• Main Authors: Talebi, Hassan, Mueller, Ute, Peeters, Luk J. M., Otto, Alex, de Caritat, Patrice, Tolosana-Delgado, Raimon, van den Boogaart, K. Gerald
• Format: Journal Article
• Language: English
• Published: Berlin/Heidelberg Springer Berlin Heidelberg 01.04.2022; Springer Nature B.V
• Subjects: Algorithms; Anomalies; Chemistry and Earth Sciences; Computer Science; Cross correlation; Earth and Environmental Science; Earth Sciences; Geological mapping; Geological processes; Geotechnical Engineering & Applied Earth Sciences; Heterogeneity; Hydrogeology; Learning algorithms; Lithology; Machine learning; Metals; Mineral deposits; Mineral exploration; Mineralization; Modelling; Multivariate analysis; Physics; Prediction models; Spatial dependencies; Special Issue; Statistical analysis; Statistical methods; Statistics for Engineering; Stochastic models; Uncertainty; Uncertainty quantification; Mineral prospectivity mapping; Geostatistical learning; Spatial data; Spatial predictive model; Machine learning
• ISSN: 1874-8961; 1874-8953
• DOI: 10.1007/s11004-021-09989-z

Mineral deposits are metal enrichment anomalies, occurring as local manifestations of the interplay between various geological processes that operate at a wide range of temporal and spatial scales. Mineral prospectivity maps are generated by integrating several proxy maps that represent critical geological processes in a mineral system conceptual model. The derivation of mineral prospectivity maps is subject to several types of uncertainty, including systematic (inadequate knowledge of mineralisation processes), stochastic (incomplete geoscience data), and model uncertainty (multiple choices for predictive models and their parameters). Traditional approaches to mineral prospectivity mapping often fail to fully appreciate different sources of uncertainty and spatiotemporal interdependencies between proxy maps associated with the mineral system components. Therefore, these traditional approaches are biased and understate the overall uncertainty. For instance, spatial proxies are mapped using univariate deterministic approaches that ignore stochastic uncertainty and spatial dependencies (i.e., auto- and cross-correlations). This study presents a multivariate stochastic model for prediction and uncertainty quantification of mineral exploration targets by combining multivariate geostatistical simulations and spatial machine learning algorithms. The spatial machine learning algorithm used in the stochastic model is a spatially aware random forests algorithm based on higher-order spatial statistics. It is demonstrated that the proposed approach can detect intrinsic heterogeneity, spatial interdependencies, and complex spatial patterns in proxy maps that are related to the mineralisation type of interest. The approach is illustrated using a synthetic case study with multiple geochemical, geophysical, and lithological attributes.