Practical aspects of large-scale conditional simulations

All large-scale applications of conditional simulations intend to benefit from the ‘correct’ (or’ accurate’) characterization of uncertainty, adequately describing the variability observed from the data and its impact on the final objective. This paper discusses a number of relevant issues that highlight the extent to which, in reality, a model of uncertainty is dependent on the random function model chosen. Often in the presence of large number of conditioning data most available random function models will provide similar models of uncertainty. When conditioning is scarce, the resulting models of uncertainty will vary more widely, since the final output will depend mostly on the underlying assumptions of the model chosen. The application described in this paper is taken from a large porphyry Cu operation in northern Chile, and it demonstrates the impact of several variables on the resulting models of uncertainty. In particular, some of the decisions involved include the number of conditioning data and the underlying Random Function model used (Gaussian or indicator-based). Also, implementation-specific parameters must be decided upon, such as grid size used, random number generator, simple or ordinary kriging, multiple-grid simulation, and simulation of irregular three-dimensional bodies (as opposed to regular grids). Some practical guidelines are proposed to deal with these decisions.