Bayesian generation of synthetic streamflows

Abstract

Generation of synthetic streamflow traces has proven to be an extremely useful technique for the evaluation of water resource planning and management alternatives. But existing models do not account for the uncertainty in the streamflow parameters. Past research efforts have focused on obtaining ‘best’ estimates of these parameters which are then used as the ‘true’ values of the process. Bayesian methods are here used to overcome this shortcoming. By incorporating the parameter uncertainties into the generation scheme, alternatives may be evaluated under both the natural and the parameter uncertainties. This is accomplished by integrating over the probability distribution of the parameters to obtain the Bayesian, predictive, or unconditional probability distribution function (pdf) of the streamflows. Use of the Bayesian pdf for synthetic generation is shown to lead, on the average, to better designs under uncertainty conditions.