Neural Netwok can be used to encode geophysical fields such as temperature, salinity or velocity.
Machine Learning
We use machine learning to reconstruct or emulate geophysical fields
Implicit neural representations and physics-informed neural networks (PINNs) are extremely promising and versatile tools for reconstructing and encoding geophysical fields by combining observational data with physical constraints. In the oceanographic context, they offer a valuable opportunity for deep flow reconstruction. While surface ocean currents can be reconstructed with relative confidence thanks to the abundance of available observations, reconstructing abyssal flows is far more uncertain due to the scarcity of data at depth. Addressing this challenge requires a multidisciplinary approach that integrates oceanography, machine learning, and inverse modeling.
This project is funded by ANR PINOT, in collaboration with A. Combette and N. Pustelnik (LP ENS Lyon), E. Riccietti (LIP ENS Lyon) B. Deremble and E. Cosme (IGE Grenoble), E. Simonnet (INPHYNI Nice).