Subgrid Scale

Subgrid-scale (SGS) modeling aims to represent the effects of unresolved small-scale processes on larger-scale dynamics in numerical simulations, particularly in fluid dynamics and climate modeling. Current research heavily utilizes machine learning, employing neural networks (including convolutional, Fourier neural operators, and recurrent networks) to learn complex SGS relationships from high-resolution data, often incorporating techniques like convex limiting or total variation diminishing methods to ensure numerical stability and physical realism. These advancements improve the accuracy and efficiency of simulations across diverse fields, from weather prediction and climate modeling to turbulence simulations and material science, by reducing computational costs and mitigating biases stemming from simplified parameterizations.