Machine learning for accelerating <scp>2D</scp> flood models: Potential and challenges

Two-dimensional hydrodynamic models numerically solve full Shallow Water Equations (SWEs). Despite their high accuracy, these have long simulation run times and therefore are of limited use for exploratory or real-time flood predictions. We investigated the possibility improving modelling speed using Machine Learning (ML). propose a new method that replaces computationally expensive parts with simple efficient data-driven approximations. Our hypothesis is by integrating ML physics-based numerical methods, we can achieve improved generalization performance: is, trained model one case study be used in other studies without need training. tested two approaches: first, integrated curve fitting, and, second, artificial neural networks (ANN) finite volume scheme to local inertial approximation SWEs. The approximated Momentum Equation, which explicitly solved time derivative flow rates. depths were then updated applying water balance equation. also different training datasets: simulated dataset, generated from results model, random directly solving momentum equation on randomly sampled input data. Various combinations features, example, slope depth, explored. proposed small hypothetical real studies. Results showed curve-fitting implemented successfully, given sufficient ANN dataset was substantially more accurate than dataset. However, it not successful resulted better performance increased 23%. Future research should test terms an increase stable step size