2024 CSDMS meeting-094

From CSDMS



(if you haven't already)




Log in (or create account for non-CSDMS members)
Forgot username? Search or email:CSDMSweb@colorado.edu


Browse  abstracts


A Deep-Learning Model for Continuous Flood Extent and Water Depth Mapping


Mohamed Fathi Said, Villanova University Villanova Pennsylvania, United States. msaid@villanova.edu
Zihan Liu, Villanova University Villanova Pennsylvania, United States. zliu5@villanova.edu
Anjali M. Fernandes, Denison University Granville Ohio, United States. fernandesa@denison.edu
Michael T. Hren, University of Connecticut Storrs Connecticut, United States. michael.hren@uconn.edu
Dennis O. Terry, Jr., Temple University Philadelphia Pennsylvania, United States. doterry@temple.edu
C. Nataraj, Villanova University Villanova Pennsylvania, United States. c.nataraj@villanova.edu
Virginia Smith, Villanova University Villanova Pennsylvania, United States. virginia.smith@villanova.edu



Predicting river hydrodynamics through computational models is critical for advancing science and engineering practices to manage rivers and floodplains. Traditional hydrodynamic models pose computational challenges, often demanding extensive processing time for large-scale 2D flood simulations. While data-driven algorithms have shown promise in improving simulation efficiency, existing efforts have primarily concentrated on generating inundation maps only at event peaks. In this research, we introduce a novel deep learning model designed to provide accurate and rapid simulation of the temporal evolution of floods, providing 2D prediction of both water depth and flood inundation maps across an entire event. We trained and evaluated this model based on a dataset that was developed using HEC-RAS, a physics-based model, for a segment of Ninnescah River, in Kansas. This was done using a deep learning model to integrate the spatial advantages of a convolutional neural network along with the temporal sequence capabilities of a long-short term memory network. The hybrid model demonstrates remarkable proficiency in capturing the dynamic nature of flood events. Evaluation of the inundation maps, at the highest testing peak, exhibited exceptional performance, with precision exceeding 0.99 and an F1-score approaching 0.98. Moreover, this hybrid model showed robust performance in predicting water depth maps, with RMSE values of 0.03 m on average during testing and 0.08 m at the highest peak time-step. This study represents a significant advancement in our ability to conduct long-term simulations of hydrodynamics and sediment transport.