Stochastic closure for partially-resolved turbulence
Résumé
To better constraint and explain data-based algorithms (e.g., data assimilation, Bayesian inference, machine learning), we tie priors to physics. Although statistical physics propose stochastic processes to be used as priors, many real systems hardly accommodate with some usual statistical simplifications (e.g., stationarity, delta-correlation). Therefore, our framework stands halfway between analytic stochastic descriptions and heavy numerical simulations. We rely on turbulence multiscale decomposition and stochastic transport. This framework is applied to wave-turbulence interactions and model errors quantification for data assimilation in geophysical fluid mechanics and reduced order models.
| Origine | Fichiers produits par l'(les) auteur(s) |
|---|
