Assessing the accuracy of sequential gaussian simulation through statistical testing

Abstract

Sequential Gaussian simulation is one of the most widespread algorithms for simulating regionalized variables in the earth sciences. Simplicity and flexibility of this algorithm are the most important reasons that make it popular, but its implementation is highly dependent on a screen effect approximation that allows users to use a moving neighborhood instead of a unique neighborhood. Because of this, the size of the moving neighborhood the number of conditioning data and the size of variogram range are important in the simulation process and should be chosen carefully. In this work, different synthetic and real case studies are presented to show the effect of the neighborhood size the number of conditioning data and the size of variogram range on the simulation result, with respect to the reproduction of the model first and second-order parameters. Results indicate that, in both conditional and non-conditional simulation cases, using a neighborhood with <50 conditioning data may lead to an inaccurate reproduction of the model statistics, and some cases require considering more than 200 conditioning data. It also can be understood from the result of example 3 that when the variogram range is beg compared to the simulation domain determination of inaccurate simulation program is harder.