Previsão hidrológica por conjunto como ferramenta de decisão na operação de reservatórios de abastecimento: estudo de caso do Reservatório de Serra Azul (MG)

• Ano de defesa: 2017
• Autor(a) principal: João Augusto de Souza Pinto
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Minas Gerais; Brasil; ENG - DEPARTAMENTO DE ENGENHARIA SANITÁRIA E AMBIENTAL; Programa de Pós-Graduação em Saneamento, Meio Ambiente e Recursos Hídricos; UFMG
• Palavras-chave em Português: Previsão hidrológica; Previsão de vazões por conjunto; Verificação probabilística; El Niño - Oscilação Sul; ENOS; Reservatório de água; Engenharia sanitária; Recursos hídricos - Desenvolvimento; Reservatórios; Serra Azul (MG); Verificação probabilística de modelos; El NiNo, Corrente
• Link de acesso: http://hdl.handle.net/1843/39523

Water production and distribution systems require a variety of decisions to be taken at different time horizons, which becomes a challenge in view of the randomness of rainfall and the partial understanding about hydrological processes in the river basin. The Extended Streamflow Prediction (ESP) method allows the incorporation of these uncertainties into a probabilistic approach, whereby a deterministic rain-flow hydrological model generates a set of flow forecasts. The ESP method was applied to predict the inflow to Serra Azul reservoir (Minas Gerais, Brazil), in order to provide better information on reservoir operation and planning. The ESP method presented good performance, especially regarding probabilistic verification through reliability and ROC diagrams, considering 35 years of hindcasts. The inflow to the reservoir was predicted during a hydrological year. The results were inputted into reservoir balance model, and simulations were carried out with 10%, 50% and 90% exceedance probabilities, considering different initial reservoir stages and different adduced flows. It was concluded that the decision making in the operation of water reservoirs can be better assisted by long-term probabilistic forecasts.