Downscaling Neural Network for Coastal Simulations

Researchers have developed a neural network that enhances the resolution of coastal ocean simulations. It takes coarse, low-detail models and generates high-resolution visualizations of water surface elevation and velocity. By incorporating physics-informed training, the model reduces prediction errors by 24% compared to data-only methods. This allows for more accurate and faster forecasting of flooding from tsunamis and storm surges, outperforming current state-of-the-art techniques.