Hydrologic uncertainty for Bayesian probabilistic forecasting model based on BP ANN

Abstract

The Bayesian forecasting system (BFS) consists of three components which can be deal with independently. Considering the fact that the quantitative rainfall forecasting has not been fully developed in all catchment areas in China, the emphasis is given to the hydrologic uncertainty for Bayesian probabilistic forecasting. The procedure of determining the prior density and likelihood functions associated with hydrologic uncertainty is very complicated and there is a requirement to assume a linear and normal distribution within the framework of BFS. These pose severe limitation to its practical application to real-life situations. In this paper, a new prior density and likelihood function model is developed with BP artificial neural network (ANN) to study the hydrologic uncertainty of short-term reservoir stage forecasts based on the BFS framework. Markov chain Monte Carlo (MCMC) method is employed to solve the posterior distribution and statistics of reservoir stage. A case study is presented to investigate and illustrate these approaches using 3 hours rainfall-runoff data from the ShuangPai Reservoir in China. The results show that Bayesian probabilistic forecasting model based on BP ANN not only increases forecasting precision greatly but also offers more information for flood control, which makes it possible for decision makers consider the uncertainty of hydrologic forecasting during decisionmaking and estimate risks of different decisions quantitatively.