Abstract

Constructing an interpretive geological model of the subsurface based on site-specific geological, geophysical and geochemical sampling information is a primary step in the prospection, exploration and modeling of ore deposits, in order to understand geological heterogeneity and to predict the rock properties in the subsurface, which is critical for all the following stages of the mining process. To improve the accuracy of such geological models, a geostatistical approach based on the conditional simulation of quantitative variables followed with their classification into geological categories is proposed. The novelty is the use of coregionalization analysis that allows decomposing the quantitative information into uncorrelated components associated with different spatial scales and filtering the undesired components from the simulation outcomes, in particular, the nugget effect that represents small-scale variability and measurement errors. The methodology is illustrated with the predictive geological mapping in an iron ore deposit hosted by banded iron formations, based on a set of physical and geochemical variables such as metal grades available from geochemical assays and rock types available from geological core logging. First, the quantitative variables are simulated conditionally to exploration drill hole data, then classified into rock types. The simulation in which the nugget effect has been removed provides more regular boundaries between rock type domains and improves the predictability of the rock type, with an 90.1% cross-validation accuracy score against a 79.9% for the simulation that reproduces the spatial variability without any filtering. Despite its small amplitude, the small-scale variability in the quantitative variables considerably deteriorates the rock type classification.