IECM 12.0 beta User Manual > Using the IECM > Analysis Tools > Uncertainty > Configure > Sampling Method >

Latin Hypercube Sampling (LHS)

Latin Hypercube sampling (LHS) is a stratified sampling method that divides the sampling space into equally probable intervals, or strata. For each input variable, the method samples each interval in a random order. When the samples from each input variable are combined, one resultant output is determined. This process is repeated m times, forming a final result of m output values. These m output values contain the uncertainty of the output variable, based on all the uncertainties of the entire set of input variables. The value m is referred to as the sample size.

The model contains two variations of Latin Hypercube sampling: Random and Median . Random LHS samples each strata randomly, while Median LHS samples each strata by its median value. Median LHS is the default sampling method.

Both forms of Latin Hypercube have the advantage of sampling more uniformly over the input distributions relative to Monte Carlo sampling, resulting in less noise in the final distribution. Another advantage is the reduced number of samples that must be taken to satisfy a given precision. Latin Hypercube has the drawback that the precision is more difficult to calculate using statistical methods. Finally, the output is random but not independent.