Authors: Supath Dhital – The University of Alabama
Title: Towards post processing enhancement of NOAA operational Flood Inundation Mapping (OWP HAND-FIM) through Surrogate Modeling
Presentation Type: Poster
Abstract: Reliable Flood Inundation Mapping (FIM) is a key component in the flood forecasting and analysis framework. FIM prediction solvers involve a tradeoff between computational costs and accuracy. Physically based models are accurate but costly, while simpler terrain-based solvers offer efficiency and scalability but lack accuracy. The NOAA Office of Water Predictions (OWP) operational FIM framework is based on the Height Above Nearest Drainage (HAND) approach. This study explores a Surrogate Model (SM) based on Convolutional Neural Network (CNN), specifically U-Net, to enhance the OWP HAND FIM (LF-FIM). The model is trained using physics-based HEC-RAS simulations and various predictors (DEM, TWI, etc.) alongside LF-FIM, which provides a foundational representation of FIM, aiding extrapolation and enhancing SM transferability across different spaces and time. Model performance and generalizability are tested quantitatively across different sites around the United States by benchmarking high-resolution Remote Sensing (RS) and HEC-RAS-based flood maps. All results show a significant improvement on LF-FIM, which will be a great integration for operational purposes. For this, a FIMserv tool is used to generate LF-FIM, which uses the OWP HAND-FIM framework, and a trained SM is applied along with other predictors to have an enhanced FIM, supporting early responders and forecasters by providing improved FIM in an efficient way for robust, rapid decision-making in operational settings.