Avaliação de um Método de Assimilação de Erros nas Previsões de Vazões de Curto Prazo Afluentes ao Reservatório da UHE Camargos

Abstract

ABSTRACT Hydrological forecasting is an important tool in the planning and operational control of water resources systems. The purpose of data assimilation or updating procedures, as applied to short-term streamflow forecasting is to improve accuracy and provide more reliable results. In this paper, streamflow forecasting is accomplished by means of a conceptual rainfall-runoff model, named Rio Grande, which has been applied to simulate the inflows of the Rio Grande river basin to the reservoir of the Camargos hydropower plant, in the Brazilian state of Minas Gerais. This model consists of three main modules: the Production Module, which is responsible for the runoff formation over the sub-basin area, the Concentration Module, which is responsible for the flow concentration at the outlet of the sub-basin, and the Propagation Module, which is responsible for flow propagation through the river channel. A stochastic model for the simulation of errors in each sub-basin is proposed, in order to update the short-term forecasts of inflows to the Camargos reservoir. Data assimilation is based on the correction of simulated discharges by adding to them the forecasts of the stochastic model errors. The stochastic models considered here are of the p-th order autoregressive AR(p)-type and/or of the p-th and q-th order autoregressive and moving average ARMA(p,q)-type. The analysis of the results shows that the application of that procedure to the upstream control stations closest to the reservoir may considerably improve short-term inflow forecasting. Key-words: Hydrological forecasting; reservoir.