Incorporação da incerteza paramétrica na geração de séries sintéticas de vazões através de reamostragem

Abstract

Streamflow generation models are very useful in simulation studies of river regulation. However there is always an uncertainty in their parameters due to fact that time period of the streamflow records are always short. The Bayesian and Classic methods for consideration of parameter uncertainty show many dependence on the structure of the adopted model. This thesis develops a general procedure for incorporation parameters uncertainty into streamflow generation models, avoiding modeling errors. The procedure is based on a Resampling scheme (similar to Bootstrap), that can be used in series that show time dependence, as the natural streamflow series. Based on historical series several "Bootstrap samples" are obtained by resampling "blocks of observations" which are independent by hypothesis. From these "Bootstrap samples" others possible parameters values are obtained. The procedure is applied to a monthly streamflow generation model given by an annual (AR-1) proccess which is disaggregated into monthly values by the scheme of Meija e Rousselle (1976). An application in regulation studies for Furnas Reservoir in the Grande river, showed that the consideration of the uncertainty in the parameters model has an effect with the same magnitude that the effect obtained when the analisys' time interval is changed from annual to monthly.