| Scope of study. Many hydrologic and water quality models have been developed. Virtually all hydrologic designs, more or less, are based on the results of applying a hydrologic and water quality model. The traditional parameter estimation methods generally assumed that there exists a set of "true" parameters for a model. In reality, model parameters should be considered as random variables rather than fixed since uncertainties are involved in natural resources data, models, and even parameter estimation techniques. The Bayesian parameter estimation technique assumes that model parameters are random variables and provide not only point estimates but also probability density functions for the calibrated parameters. To evaluate the efficiency of the Bayesian parameter estimation technique, the point estimates by the Bayesian method were compared with those obtained by Least Squares. Due to uncertainty of model parameters, the model predictions will be uncertain. A method was developed to capture the uncertainty in model predictions. The Agricultural Non-Point Source pollution model (AGNPS) was used on four watersheds in Arkansas.; Findings and conclusions. The Bayesian parameter estimation method is just as efficient as Least Squares and has the advantage of providing probability density functions for the calibrated parameters. The uncertainty in model predictions can be evaluated by using the uncertain parameters estimated by the Bayesian parameter estimation method. The uncertainty in model predictions could be reduced when the prior information for parameters are correctly specified. However, miss specification of the prior information of parameters will not reduce the uncertainty in model predictions but will lead to worse or even false model predictions. Therefore, more caution needs to be taken in specifying the prior information for model parameters. |
