Bayesian Analysis for Evaluation of Uncertainty in Conceptual Rainfall-Runoff Models

Abstract

This paper presents a Bayesian approach for the evaluation of uncertainty in rainfall-runoff modeling employing Markov Chain Monte Carlo (MCMC) methods. The Bayesian approach via MCMC provides the full posterior distribution of parameters and any function of them, including the distribution of simulated flows, allowing one to employ hydrologic model results in decision making process under a risk analysis framework. A basin located in the State of Ceará, in the northeast semiarid region, was selected to be used as an example for this study. The paper presents the posterior distributions of model parameters, as well as the standard deviation of the model error, and a description of uncertainty based on the 95% credible interval. Results illustrate the potential of the Bayesian approach in exploring the uncertainties in hydrologic modeling, although it is still necessary to explore the role prior and proposal distributions in the generation of the Markov chains.