Efficient sampling techniques for uncertainties in risk analysis

Abstract

Abstract Risk and policy analysis involves consideration of uncertainties. A quantitative probabilistic method for uncertainty analysis includes: (1) quantifying and assigning probabilistic distributions to the input uncertainties, (2) sampling the distributions of these uncertain parameters in an iterative fashion using Monte Carlo methods, (3) propagating the effects of uncertainties through the model, and (4) predicting the outcomes in terms of probabilistic measures like mean, variance, and fractiles. However, the results of the probabilistic analysis depend on the number of samples chosen. The sample size required for a particular analysis depends on various factors such as type of model, the random number generator used, type of distributions, and the output probabilistic measure and cannot be universally defined. The general tendency is to reduce the samples as much as possible without realizing the effect on decisions. For example, the mean of the output requires a number of samples that is an order of magnitude less compared to the variance. Therefore, it is desirable to use a sampling technique that can predict the output probabilistic measure accurately with the minimum number of samples. In this work, we present new sampling techniques based on Quasi‐Monte Carlo sequences and Latin hypercube sampling. © 2004 American Institute of Chemical Engineers Environ Prog, 23: 141–157, 2004