HR: 17:15h
AN: H24A-06 INVITED    [Abstracts]
TI: Bayesian Uncertainty Analysis for Computationally Expensive Hydrologic Simulation Models
AU: * Shoemaker, C A
EM: cas12@cornell.edu
AF: Cornell University, Civil and Environmental Engineering, Ithaca, NY 14853, United States
AB: This article presents a new computationally efficient method for statistically rigorous assessment of uncertainty in parameters and model output when the model is calibrated to field data. The Bayesian method be general and is here applied to water resource problems The innovative aspect of this procedure is that an optimization method is first used to find the maximum likelihood estimator and then the costly simulation done during the optimization are re used to build a response surface model of the likelihood function. Markov chain Monte Carlo is applied then to the response surface model to obtain the posterior distributions of the model parameters and the appropriate transformations to correct for non normal error. The computational effort to obtain roughly the same accuracy of solution is 150 model simulations for the response surface method versus 10,000 simulations for conventional MCMC analysis, which is a 60 fold reduction in computational effort
DE: 1805 Computational hydrology
SC: Hydrology [H]
MN: 2007 Fall Meeting