Latin Hypercube Sampling (LHS) Of Input Parameters For Stochastic Simulation

Stochastic simulation has been routinely used for performing risk and sensitivity analyses in a variety of fields. The commonly used means of generating the simulations has been the Monte Carlo approach, which usually employs a Simple Random Sampling (SRS) scheme of the variable distributions. This method is prone to sample clustering and requires a large number of samples to represent the distribution faithfully. This can become problematic for simulating complex processes where it is impossible to run an arbitrarily large number of replicate simulations. The Latin Hypercube Sampling (LHS) method uses a stratified sampling approach that assures representation of the input distributions for any given number of samples. In this paper the LHS and SRS approaches are compared, using a petroleum resource estimation example.