ABSTRACT
Water resource designs based on flood are perfect examples of decisions under uncertainty. Three types of uncertainties may exist in a flood design problem - when the classical method is used. These are natural, parameter and model uncertainties. To reduce the parameter and model uncertainties, the Bayesian procedure has been used. Three common probability distributions (normal, log-normal, and exceedance) were considered. To reduce the parameter uncertainty, the Bayesian procedure was applied to these three distributions to give the corresponding Bayesian probability models. To achieve this objective, flood discharge data collected from Gassol gaging station we re taken as the sample information while flood discharge data from eleven other selected gaging stations in the same homogeneous area as Gassol in Upper Benue River Basin were considered to be the prior (regional) information. Natural uncertainty is also reduced by this proceedure. To reduce the model uncertainty, Bayesian procedure was also applied to estimate the marginal likelihoods of the three models. Results showed that only the exceedance model fits the flood discharge data. Comparison of the classical method with Bayesian method shows that Bayesian method predicts higher flood discharges than the classical method. Also comparison of the regional (prior) method (only in the case of exceedance model) shows that the Bayesian method predicts lower discharges than the regional information.