Nguyen et al. (2025) Toward real-time high-resolution fluvial flood forecasting: A robust surrogate approach based on overland flow models

Identification

• Journal: Environmental Modelling & Software
• Year: 2025
• Date: 2025-10-10
• Authors: Giang V. Nguyen, Chien Pham Van, Vinh Ngoc Tran, Linh Nguyen Van, Giha Lee
• DOI: 10.1016/j.envsoft.2025.106716

Research Groups

• School of Advanced Science and Technology Convergence, Kyungpook National University, Sangju-si, South Korea
• Thuyloi University, Ha Noi, Viet Nam
• Department of Civil and Environmental Engineering, University of Michigan, Ann Arbor, USA

Short Summary

This study presents a hybrid framework integrating machine learning with physics-based hydrodynamic models to enable efficient real-time high-resolution fluvial flood forecasting. It demonstrates that ML-based surrogate models, trained on TELEMAC outputs, achieve substantial computational efficiency while preserving accuracy for flood inundation prediction in the Cambodia floodplain.

Objective

• To develop and evaluate a robust hybrid framework that integrates machine learning with physics-based hydrodynamic models to overcome the computational limitations of traditional models for real-time high-resolution fluvial flood forecasting.

Study Configuration

• Spatial Scale: Cambodia floodplain.
• Temporal Scale: Real-time forecasting, seasonal flooding.

Methodology and Data

• Models used: TELEMAC (physics-based hydrodynamic model), Machine Learning (ML) models (surrogate predictors), Explainable AI (GeoXAI).
• Data sources: Outputs from TELEMAC simulations (for training ML models).

Main Results

• The hybrid approach achieved substantial computational efficiency while preserving accuracy.
• The best surrogate model attained a coefficient of determination (R) of 0.97 and Kling-Gupta Efficiency (KGE) of 0.91.
• The surrogate models reduced simulation time by over 70-fold compared with TELEMAC.
• Incorporating geographic features such as latitude and longitude further enhanced predictive skill, particularly in flat floodplain settings.

Contributions

• Presents a novel hybrid framework combining physics-based models with ML surrogates for real-time, high-resolution flood forecasting, addressing the computational bottleneck of traditional hydrodynamic models.
• Demonstrates significant computational speedup (over 70-fold) while maintaining high accuracy (R=0.97, KGE=0.91).
• Highlights the utility of Explainable AI (GeoXAI) for interpreting model decision-making in flood forecasting.
• Shows the benefit of integrating geographic features into ML models for improved flood prediction in complex terrains like floodplains.

## Funding -

Citation

@article{Nguyen2025Toward,
  author = {Nguyen, Giang V. and Van, Chien Pham and Tran, Vinh Ngoc and Van, Linh Nguyen and Lee, Giha},
  title = {Toward real-time high-resolution fluvial flood forecasting: A robust surrogate approach based on overland flow models},
  journal = {Environmental Modelling & Software},
  year = {2025},
  doi = {10.1016/j.envsoft.2025.106716},
  url = {https://doi.org/10.1016/j.envsoft.2025.106716}
}