Study On Bayesian Forecasting System Of Deteministic Hydrologic Models

The hydrologic process is thought as a complex phenomena influenced by both deterministic and stochastic factors. Some stochastic characters exist in the input and the parameters of hydrologic model. So the output of hydrologic model is uncertain. The hydrologic forecasts based on the uncertainty are more scientifically than those based on deterministic forecasts.BFS (Bayesian Forecasting System) is one of theoretic frames to produce probabilistic hydrological forecasts. The AM-MCMC based on AGA (Accelerating Genetic Algorithm) was proposed, i.e. AGA-AM-MCMC. Under the BFS, the BFS of Nash model based on the AGA-AM-MCMC and the BFS of XAJ model were built.The effects on hydrologic forecasts by the uncertainties were researched with the BFS of Nash model based on AGA-AM-MCMC at flow concentration of Yanduhe basin and channel routing of Xiangyang-Huangzhuang reach. Then, the probabilistic forecasting was performed. The result of cases shows that mean of flood discharge could be calculated by the model, varicance and the specified probabilistic confidence interval, which quantified the uncertainty of forecast, also were given. There is obvious correlation between the two parameters of Nash model in both routing of watershed and channel routing. Furthermore, the extent of effect on forecast caued by the input uncertainty was larger than that by the parameters uncertainty in channel routing.The AGA-BP was built to characterize the prior density and likelihood function, and it overcomes the difficulties found in getting them practically. Then, on the basis of the AGA-BP, the BFS of XAJ model based on AGA-AM-MCMC was built. The application to rainfall-runoff forecast in Yanduhe basin showed that the probabilistic forecasting could be performed; the mean forecasting and the specified probabilistic confidence interval of flood discharge were produced and the later characterized the extent of uncertainty. The precision of mean forecasts processes was higher than the forecasts of XAJ model. The probablisitc forecasting can form an important basis for decision making of flood control.