Authors: Sadra Seyvani – The University of Alabama
Title: FIM-Combinator: A Neural Network-Based Integration of HEC-RAS, LISFLOOD-FP, and OWP-HAND-FIM for Enhanced Flood Inundation Mapping
Presentation Type: Poster
Abstract: The variations in topographical and geomorphological parameters across different regions lead to discrepancies in the accuracy of hydrodynamic and topography-based Flood Inundation models such as HEC-RAS, LISFLOOD-FP, and OWP HAND-FIM. Consequently, each of these models, depending on the specific topographical and geomorphological characteristics of the study area, exhibits its own unique strengths and weaknesses. Previously, the study “A Novel Algorithm Based on a Combination of CNN and Fully Connected Neural Networks in a Parallel Architecture for Selecting the Most Accurate Flood Inundation Model for Any Desired Area” introduced a trained neural network that functioned as the “brain” of the DMAF tool. This tool could predict, within a fraction of a second, the accuracy of the three models HEC-RAS, LISFLOOD-FP, and OWP HAND-FIM for any desired region. In the present study, we explore another potential application of this trained neural network. By making minor adjustments to adapt it to a new objective, we combined the three models of HEC-RAS, LISFLOOD-FP, and OWP HAND-FIM to produce a more accurate flood inundation map. Specifically, this new tool processes the flood inundation maps generated by the three models, evaluates the region’s topographical and geomorphological characteristics, and without relying on any benchmark data, decides which of the three models should be used in each local area. Initial results obtained by testing this innovative tool on the flood event data from October 2016 in North Carolina are highly promising. While the highest accuracy achieved by any of the individual models (HEC-RAS, LISFLOOD-FP, or OWP HAND-FIM) was 0.84, the flood inundation map generated by the FIM Combinator developed in this study, which integrates outputs from all three models, achieved an accuracy of 0.98, representing a remarkable improvement. This tool is particularly useful for studies where rapid flood inundation mapping is not a requirement and higher accuracy is a priority.