Sequential Spatial Simulation using Latin Hypercube Sampling

Abstract

An efficient method is proposed for generating realizations from an arbitrary multivariate distribution using sequential simulation and Latin hypercube sampling. In a spatial context, this efficiency entails a reduction of sampling variability in statistics of spatially distributed model outputs when the inputs are realizations of random field models. The proposed method yields an unbiased reproduction of a target semivariogram, even for a small number of realizations, and consequently can be used for enhanced uncertainty and sensitivity analysis incomplex spatially distributed models. In addition, the method is simple enough to be incorporated in virtually any geostatistical software for sequential simulation.