Author(s): Juan F. Farfan-Duran; Luis Cea
Linked Author(s): Juan Fernando Farfán Durán
Keywords: No Keywords
Abstract: Streamflow forecasting is crucial for water resources management, with its complexities stemming from various factors including weather patterns and basin characteristics. This study evaluates the efficiency of Deep Learning (DL) techniques, such as LSTM, GRU, CNN, and their hybrid combinations, in predicting hourly streamflow in two basins (Groba and Anllons) located in Northwest Spain, using as input the observed rainfall and discharge previous time steps. Our results indicate a context-dependent performance, with the GRU model excelling in 1 – 3-hour predictions for Groba, while all models perform comparably in Anllons. Interestingly, added complexity in hybrid models did not guarantee superior accuracy. The results highlight the challenge of doing 12-hour predictions in basins with short concentration times and underscores the significance of including the hydrological characteristics of the basin in the forecast.
Year: 2024