Deep learning models to predict flood events in fast-flowing watersheds.

Area of Science:

• Hydrology and Water Resources
• Artificial Intelligence in Environmental Science
• Climate Change Adaptation

Background:

• Traditional physics-based flood models often require extensive data and parameter estimation, limiting their application in data-scarce or complex small basin scenarios.
• Accurate flood forecasting is crucial for effective territorial management, especially amidst increasing climatic changes.
• The Arno River basin, with its densely populated cities and cultural heritage, presents a critical area for reliable flood prediction.

Purpose of the Study:

• To evaluate the reliability of deep learning models, particularly Long-Short Term Memory (LSTM) networks, for flood warning forecasts.
• To assess the applicability of these models for predicting short-duration flood events (lasting a few hours) using hydrometric data.
• To establish a foundation for using data-driven models where physical modeling is challenging due to data limitations or basin complexity.

Main Methods:

• Utilized Long-Short Term Memory (LSTM) deep learning architecture for flood event prediction.
• Employed hydrometric control station data for forecasting flood events in the Arno River basin.
• Tested model stability and robustness by simulating missing or erroneous input data.

Main Results:

• Deep learning models demonstrated acceptable error margins for flood event identification with several hours' notice.
• The LSTM models exhibited significant stability, performing well even when substantial amounts of input data were missing.
• Achieved excellent results, comparable to physics-based models, with a less data-intensive approach.

Conclusions:

• Deep learning, specifically LSTM, offers a reliable and robust method for flood warning systems, particularly in data-limited environments.
• This data-driven approach provides a valuable alternative to physics-based models for flood prediction in complex or small river basins.
• The study lays the groundwork for advanced hydrological forecasting, enhancing territorial management and climate change adaptation strategies.