Bayesian Inference and Markov Chain Monte Carlo Methods Applied to Streamflow Forecasting

In this work we propose a Bayesian approach for the parameter estimation problem of stochastic autoregressive models of order p, AR(p), applied to the streamflow forecasting problem. Procedures for model selection, forecasting and robustness evaluation through Monte Carlo Markov Chain (MCMC) simulation techniques are also presented. The proposed approach is compared with the classical one by Box-Jenkins (maximum likelihood estimation) on a monthly streamflow time series from Furnas reservoir. We conclude that the use of Bayesian statistics and MCMC simulation gives more flexibility and powerful results than those obtained from the classical approach.