Forecasts of hydrological variables are extremely important for water resource management in regions that are particularly vulnerable to the effects of climate change and climate variability. The aim of this study was to develop a hybrid neural network capable of producing daily streamflow forecasts on a sub-seasonal scale. This hybrid neural network consists of an artificial neural network with inputs preprocessed by the wavelet transform (WANN). The WANN was tested in the Três Marias, Sobradinho, and Retiro Baixo reservoirs, located in the São Francisco River Basin (SFRB). The obtained results show that WANN was highly accurate in short-term forecasts (7–28 days); however, for long-term forecasts (35 and 42 days), there was a significant drop in performance, especially during the transition periods to the rainy season and in the dry months. The comparison between the performance metrics of the WANN forecasts and the National Electricity System Operator (ONS) operational models for the Três Marias, Sobradinho, and Retiro Baixo basins showed that WANN outperformed all these models. The results obtained show that WANN is a valuable tool for addressing the complex and dynamic challenges of hydrology, making it essential for decision-making on water resource management.
Castro et al. (Sun,) studied this question.