Presentation Information
[hyhteQZr-05]Physics Informed Shallow Water Emulator
*Iuri Gorenstein1 (1. Universidade de São Paulo (Brazil))
Numerical ocean models face a trade-off: low-resolution runs miss turbulence, while
high-resolution ones require smaller time steps, making them slow and expensive. Neural
Networks (NNs) can decouple spatial and temporal resolution, integrating with larger time
steps and capturing both linear and nonlinear behaviors without explicit equations. This
flexibility accelerates simulations and allows NNs to represent processes that traditional
models might oversimplify in their higher-level description of the discretized physical
equations. Yet it raises a critical question: How can we trust physics-uninformed models to
produce stable, realistic long-term ocean simulations? This study aims to build a fully
generalizable, point-wise, physics-informed NN embedded with the Regional Ocean
Modeling System (ROMS) barotropic dynamics. Bridging the gap between traditional
numerical models and NN algorithms, our results demonstrate that the NN emulator accurately reproduces the non-linear instabilities and eddy dynamics observed in barotropic flows, comparable to the original numerical solver.
high-resolution ones require smaller time steps, making them slow and expensive. Neural
Networks (NNs) can decouple spatial and temporal resolution, integrating with larger time
steps and capturing both linear and nonlinear behaviors without explicit equations. This
flexibility accelerates simulations and allows NNs to represent processes that traditional
models might oversimplify in their higher-level description of the discretized physical
equations. Yet it raises a critical question: How can we trust physics-uninformed models to
produce stable, realistic long-term ocean simulations? This study aims to build a fully
generalizable, point-wise, physics-informed NN embedded with the Regional Ocean
Modeling System (ROMS) barotropic dynamics. Bridging the gap between traditional
numerical models and NN algorithms, our results demonstrate that the NN emulator accurately reproduces the non-linear instabilities and eddy dynamics observed in barotropic flows, comparable to the original numerical solver.
Comment
To browse or post comments, you must log in.Log in