MSc Student Technion - Israel Institute of Technology, Hefa, Israel
Abstract Submission: Floods are among the most frequent natural disasters, necessitating accurate and efficient forecasting models to minimize damage. Traditional hydrodynamic models, while providing detailed simulations of flood dynamics, are often computationally intensive and impractical for real-time, large-scale applications. This research investigates the potential of deep learning as an alternative approach for hydrodynamic flood modeling, focusing on predicting future flood states across diverse terrains, initial and boundary conditions, spatial domain sizes, and temporal snapshots.
A deep learning-based model was developed to generalize flood predictions at the patch level and was further integrated into a closure framework designed to extend these predictions across larger spatial domains. The deep learning model at the patch level demonstrated robust performance across various datasets, effectively capturing critical spatial features essential for accurate flood state prediction. Overall, this study demonstrates that deep learning offers a viable path toward efficient and scalable flood modeling, paving the way for real-world applications in flood forecasting and management.
