Modelagem de séries temporais para previsão de vazões em rios com monitoramento hidrológico

Abstract

Water resources play a vital role in a country's development. Streamflow is the primary variable monitored at gauging stations, however measuring this variable involves significant risks and high operational costs. The Brazilian National Hydrometeorological Network (RHN), operated by the Geological Survey of Brazil (SGB) in partnership with the National Water Agency (ANA), provides essential historical datasets for streamflow forecasting. This study aims to develop time series models to forecast monthly river discharges, using as case studies the gauging stations of Santa Maria da Vitória (1977–2022) and Batalha (2015–2023), both located in Bahia. The research employs ARIMA models, implemented in the R Core Team (2025), to evaluate the accuracy and applicability of forecasts at various temporal horizons. The results demonstrate that the ARIMA (3,0,3)(3,1,1) model achieved a satisfactory fit, with robust error metrics (MAPE, RMSE, and MAE). Statistical tests confirmed stationarity. The forecasts closely followed the historical series’ behavior, providing consistent estimates up to 12 months ahead. It is concluded that time series modeling is a promising tool to complement the traditional rating curves, enabling greater simplicity and reliability in streamflow predictions. This approach can further support water resources management and hydrological alert systems in critical events such as droughts and floods. Additionally, it contributes to the design of hydraulic structures and water allocation processes, strengthening the use of quantitative data and statistical techniques as decision-support tools in hydrology.