Deep Learning-Based River Flow Forecasting with MLPs: Comparative Exploratory Analysis Applied to the Tejo and the...

Area of Science:

• Hydrology
• Artificial Intelligence
• Water Resource Management

Background:

• Accurate river flow forecasting is crucial for effective water resource management, flood control, and infrastructure operation.
• Existing forecasting methods may have limitations in precision and timeliness for short-term predictions.
• Dam-controlled rivers require reliable forecasting for operational efficiency and downstream impact mitigation.

Purpose of the Study:

• To develop and demonstrate an innovative remote service for short-term river flow forecasting.
• To create and enhance Artificial Intelligence (AI) models, specifically Multilayer Perceptron (MLP), for predicting river discharge.
• To provide precise and timely river flow predictions for the next 3 days.

Main Methods:

• Utilized Multilayer Perceptron (MLP) architectures for river discharge prediction.
• Employed comprehensive hydrological data from Portugal's National Water Resources Information System (SNIRH).
• Conducted a comparative study of MLP model performance on the Tejo and Mondego river basins, detailing data preparation, model training, and forecasting.

Main Results:

• MLP models demonstrated acceptable accuracy in short-term river flow forecasts for the selected case studies.
• The models effectively captured discharge patterns and peak occurrences in the Tejo and Mondego rivers.
• Comparative analysis confirmed the models' performance across different hydrological scenarios.

Conclusions:

• MLP models offer a viable data-driven approach for enhancing short-term river flow forecasting accuracy.
• The developed AI service can significantly improve water resource management, decision-making, and flood mitigation strategies.
• These forecasts provide valuable boundary conditions for downstream hydrological and meteorological forecast systems.