Events / Media
Data Assimilation Seminar Series
Prof. Pierre Tandeo (Feb. 16, 2018, 15:30-)
| Affiliation | IMT Atlantique |
|---|---|
| Title | The analog data assimilation: method, applications and implementation |
| Abstract | Nowadays, oceanic and atmospheric sciences face a deluge of data pouring from space, in situ monitoring and numerical simulations. The availability of these different data sources offer opportunities, still largely underexploited, to improve the understanding, modeling and reconstruction of geophysical dynamics. Classically, we use data assimilation methods to reconstruct the space-time dynamics from noisy and partial observations. In practice, this requires simulations based on explicit dynamical models with large computational costs, and with possible limitations such as inconsistencies between the model and the observed data, and substantial modeling uncertainties. In this presentation, I present an alternative data-driven approach, using a representative catalog of historical data: the Analog Data Assimilation (AnDA) method is used to learn the local relationships between state variables and provide realistic forecasts from the analog method, without online evaluation of a physical model. Numerical experiments are examined for two chaotic dynamical systems: the Lorenz-63 and Lorenz-96 systems. The performance of the AnDA method is discussed with respect to classical model-driven assimilation. Then, I present some applications in spatial oceanography, using archive of satellite data, for the reconstruction of ocean surface. Finally, I make a demonstration of the AnDA Python code and explain how it can be used for any data-driven assimilation problems (e.g. where no model is available but historical observations are abundant). |
| Link to PDF file |
