Application of deep learning approaches to predict monthly stream flows

Area of Science:

• Hydrology
• Artificial Intelligence
• Water Resource Management

Background:

• Accurate river flow estimation is critical for hydroelectric power, flood/drought management, and water resource utilization.
• Traditional methods face challenges in capturing complex hydrological dynamics.
• Advanced computational models are needed for reliable streamflow forecasting.

Purpose of the Study:

• To evaluate the performance of Gated Recurrent Unit (GRU), Recurrent Neural Network (RNN), and Long Short-Term Memory (LSTM) models for monthly river flow prediction.
• To compare these artificial intelligence models using various statistical performance metrics.
• To identify the most effective model for streamflow forecasting in the specified regions.

Main Methods:

• Utilized monthly streamflow time series data from 1978-2015 for three stations in Erzincan, Bayburt, and Gümüshane.
• Developed and trained GRU, RNN, and LSTM artificial intelligence models using 70% training, 15% validation, and 15% test data splits.
• Assessed model performance using Correlation Coefficient, RMSE, RMSE/Std Dev ratio, Nash-Sutcliffe efficiency, Index of Agreement, and Volumetric Efficiency.

Main Results:

• The Gated Recurrent Unit (GRU) model demonstrated superior performance in estimating river streamflow compared to RNN and LSTM.
• GRU achieved efficient estimation results across the evaluated hydrological stations.
• The study confirmed GRU's potential for reliable streamflow forecasting.

Conclusions:

• GRU neural networks provide a highly effective tool for accurate river flow prediction.
• The findings support the application of GRU in water resource management and allied hydrological studies.
• GRU models offer a promising alternative for enhancing the reliability of water resource planning and operations.