Optimizing flood predictions by integrating LSTM and physical-based models with mixed historical and simulated data

Area of Science:

• Hydrology and Water Resources
• Artificial Intelligence in Environmental Science
• Geospatial Analysis and Flood Modeling

Background:

• Traditional flood forecasting models struggle with data scarcity and high measurement costs, limiting prediction robustness.
• Existing methods often fail to fully leverage the potential of both physical process simulations and data-driven approaches.
• Effective flood prediction is critical for mitigating risks in vulnerable river basins.

Purpose of the Study:

• To develop and evaluate a novel hybrid flood forecasting model integrating physical-based simulations with Long Short-Term Memory (LSTM) networks.
• To assess the impact of different data combinations (measured, simulated, mixed) on LSTM model performance.
• To optimize LSTM hyperparameters for accurate flood prediction in a specific river basin.

Main Methods:

• Employed the NAM hydrological and HD hydraulic models for flood process simulation.
• Utilized Long Short-Term Memory (LSTM) networks trained on measured, simulated, and mixed datasets.
• Evaluated model performance using Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and Peak-relative Error (PRE) in the Jinhua basin.

Main Results:

• LSTM models trained on mixed datasets, particularly with a simulated-to-measured data ratio below 2:1, showed superior performance.
• These hybrid models achieved significantly lower RMSE and MAE compared to models trained on other data ratios.
• The findings underscore the benefit of synergistically integrating simulated and measured data for enhanced flood forecasting accuracy.

Conclusions:

• The hybrid approach offers a robust solution to data scarcity and computational efficiency challenges in flood forecasting.
• Integrating physical model outputs with LSTM networks provides a powerful framework for improving prediction accuracy.
• This methodology holds promise for real-time flood prediction and risk management in various flood-prone regions.