Abstract

This work aims at comparing the performance of three algorithms for the geostatistical simulation of coregionalized variables in ore deposits, which is relevant for quantifying the uncertainty in the mineral resources and ore reserves. The first algorithm is sequential Gaussian simulation, where the variables of interest are simulated one after the other. The second one is turning bands simulation based on the decomposition of the variables into factors with no spatial cross-correlation. The third one is also a turning bands algorithm, but relies on a spectral decomposition of the variables rather than a factorization. All three algorithms are applied to a synthetic case study and to a real case study consisting of an iron ore deposit, in which seven variables are of interest for characterizing the mineral resources: iron grade, silica grade, alumina grade, manganese grade, phosphorus grade, loss on ignition, and granulometric fraction above 6.3 mm. The comparison of results indicate that sequential Gaussian simulation is slower than turning bands based on factorization, which is itself slower than spectralturning bands. Furthermore, the reproduction of the spatial correlation structure (cross-variograms) is poorer with sequential simulation than with turning bands simulation (in both variants), due to the inherent approximations needed to run the sequential algorithm. This illustrates that, in multivariate cases, the choice of the simulation algorithm is relevant, in terms of computational efficiency and accuracy.