Modelagem bayesiana das vazões máximas do reservatório de Furnas

Abstract

The objective of this work was to predict the maximum flow of the Furnas reservoir in the dry and wet periods. The generalized distribution of extreme values (GEV) was used with parameter estimation via Bayesian inference. Data on average daily streamflows corresponding to the years 1965 to 2017 were obtained from Hidroweb, by the National Agency for Water and Basic Sanitation (Agência Nacional de Águas e Saneamento Básico – ANA), from which maximum values were extracted, by period and in each year. Accuracy and mean error of prediction of maximum streamflows were analyzed, comparing the estimates provided by the Bayesian inference, with informative and non-informative prior distributions. Information from a series of maximum streamflows from the Camargos reservoir was used to elicit the informative prior distribution. The use of prior information provided an increase in the precision and accuracy of the maximum streamflow estimates. Thus, the GEV model, with informative a priori distribution, was used to predict the return levels of Furnas maximum streamflow with their respective high posterior density intervals considering several return times.