Author(s): G. Manache; C. S. Melching
Linked Author(s):
Keywords: Uncertainty Analysis; Sensitivity Analysis; Simulation; Water-Quality Modeling
Abstract: Computational models are commonly used as decision-making tools. Since many uncertainties are associated with the input variables and/or parameters of such models, sensitivity and uncertainty analyses are necessary prior to and during, respectively, model application to planning and design. Monte Carlo Simulation (MCS) and Latin Hypercube Sampling (LHS) are powerful, robust, and flexible uncertainty analysis methods. In MCS and LHS, values of uncertain inputs and parameters are selected at random from their assumed probability distributions. Dynamic simulations of the system are repeated for all sampled inputs. The output statistics, distributions, and correlation among input and output variables allow the estimation of the model output uncertainty and the identification of the parameters and input variables to which the output is most sensitive. The selection of the sample size and the probability distribution for the uncertain model input variables or parameters are important aspects of the application of simulation-based sensitivity and uncertainty analysis methods. In this paper, LHS is applied to a complex water-quality model to examine the effect of the sample size and selected probability distribution of the input parameters on the identification of the parameters significantly affecting model output and on the model output uncertainty. The results indicate that the selection of ranges and probability distributions of the model input parameters has an effect on the sensitivity and uncertainty analysis results. A sample size of 4/3 times the number of uncertain input variables is shown to adequately identify the sensitive parameters, but may not adequately estimate the model output uncertainty.
Year: 2007