Streamflow forecasting with deep learning models: A side-by-side comparison in Northwest Spain

• DOI: 10.1007/s12145-024-01454-9
• Publication Date: 2024-08-23T22:15:53Z
• Authors: Juan F. Farfán-Durán; Luis Cea
• Keywords: 0208 environmental biotechnology; 02 engineering and technology

Abstract Accurate hourly streamflow prediction is crucial for managing water resources, particularly in smaller basins with short response times. This study evaluates six deep learning (DL) models, including Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), Convolutional Neural Network (CNN), and their hybrids (CNN-LSTM, CNN-GRU, CNN-Recurrent (RNN)), across two Northwest Spain over a ten-year period. Findings reveal that GRU models excel, achieving Nash-Sutcliffe Efficiency (NSE) scores of approximately 0.96 0.98 the Groba Anllóns catchments, respectively, at 1-hour lead Hybrid did not enhance performance, which declines longer times due to basin-specific characteristics such as area slope, where NSE dropped from 0.969 0.24. The inclusion future rainfall data input sequences has improved results, especially 0.24 0.70 basin 0.81 0.92 12-hour time. research provides foundation exploration DL forecasting, other sources model structures can be utilized.