Skip to content

Hyper-parametrization of Oceanic Eddies: new strategy and methodology for a long-standing problem

Earth and planetary sciences are characterized by ‘Big Data’ capturing processes and products at a large range of temporal (milliseconds to billions of years) and spatial (microns to light years) scales. Our goal is to use cutting edge data science, machine learning and deep learning to better characterize and predict complex planetary processes in order to address societal problems such as climate change and sourcing clean energy.

Machine Learning (ML) attracted huge and ever-growing interest over the last years for providing novel modelling frameworks. In the ocean modelling, ML has been used for many purposes (e.g., observational data analysis, active flow control and shape optimization, reduced-order modelling), but its promising applications for parameterizing mesoscale eddies remain in infancy, although with huge research potential. Our project enriches ML methods with ideas from dynamical systems to demonstrate the power and utility of the hyper-parameterization approach — a new paradigm for modelling turbulent ocean flows.

Led by Professor Pavel Berloff.


Our research