Toward data-driven methods in geophysics: the Analog Data Assimilation - Archive ouverte HAL Accéder directement au contenu
Communication Dans Un Congrès Année : 2017

Toward data-driven methods in geophysics: the Analog Data Assimilation

Résumé

The Analog Data Assimilation (AnDA) is a recently introduced data-driven methods for data assimilation where the dynamical model is learned from data, contrary to classical data assimilation where a physical model of the dynamics is needed. AnDA relies on replacing the physical dynamical model by a statistical emulator of the dynamics using analog forecasting methods. Then, the analog dynamical model is incorporated in ensemble-based data assimilation algorithms (Ensemble Kalman Filter and Smoother or Particle Filter). The relevance of the proposed AnDA is demonstrated for Lorenz-63 and Lorenz-96 chaotic dynamics. Applications in meteorology and oceanography as well as potential perspectives that are worthy of investigation are further discussed. We expect that the directions of research we suggest will help in bringing more interest in applied machine learning to geophysical sciences.
Fichier non déposé

Dates et versions

hal-01609890 , version 1 (04-10-2017)

Identifiants

  • HAL Id : hal-01609890 , version 1

Citer

Redouane Lguensat, Pierre Tandeo, Pierre Ailliot, Manuel Pulido, Ronan Fablet. Toward data-driven methods in geophysics: the Analog Data Assimilation. EGU 2017 : 19th EGU General Assembly, Apr 2017, Vienne, Austria. ⟨hal-01609890⟩
118 Consultations
0 Téléchargements

Partager

Gmail Facebook X LinkedIn More