244 - Converting Low-Fidelity to High-Fidelity Flood Inundation Map using Super Resolution Techniques

Flooding is one of the most frequent natural hazards, and it causes more economic loss than all other natural hazards. Fast and accurate flood prediction has significance in preserving lives, minimizing economic damage, reducing environmental and public health risks. However, current methods cannot achieve speed and accuracy at the same time. Numerical methods can provide high fidelity results, but it is time-consuming, particularly when pursuing high accuracy. On the other hand, neural networks can provide results in second level, but it has shown low accuracy in flood map generation by all existing methods. This work is going to combine the strengths of numerical methods and neural networks and build a framework that can fast and accurately model the high-fidelity flood inundation map with detailed water depth information. In this paper, we employ U-Net and Generative Adversarial Networks (GAN) models to recover the lost physics and information from ultra-fast, low-resolution numerical simulations, ultimately presenting high-resolution, high-fidelity flood maps as the end results. Additionally, this approach adeptly addresses the challenge of handling highly variable input data. The low-resolution numerical method preprocesses high-variability inputs into low-variability formats, enabling neural networks to rapidly learn the relationships even with extremely limited training data. In our study area, both the U-Net and GAN models have proven their ability to reduce the computation time for generating high-fidelity results, reducing it from 7-8 hours down to a mere 1 minute. Furthermore, the accuracy of both models is notably high.