Though oceanic and atmospheric turbulent flows have leading order effects on climate, they are computationally intractable to simulate on global scales. Instead, climate models typically rely on turbulence closures, or statistical approximations of the complex dynamics of flows, when simulating oceans and atmospheres. Improving turbulence closures for oceanic and atmospheric flows is a crucial step towards improving the accuracy of climate models. In this project, conducted for the Climate Modeling Alliance, we look to improve existing turbulence closures for large-scale flows by employing machine learning methods such as Bayesian equation discovery to add residual terms and coefficients to closures in a physically-informed way, minimizing uncertainty and maximizing model accuracy.

“I am participating in SuperUROP because I am broadly interested in stochastic processes, Bayesian inference, and modeling complex physical phenomena, all of which my project will allow me to explore in an application that is new to me. I am also excited to have a meaningful opportunity to contribute to the climate modeling research effort.”