Project 1

Developing Data Assimilation and Inversion Techniques for Large Scales Applications

Details

​Involved:
  • Ibrahim Hoteit
  • Xiadong Luo
  • Umer Altaf
Collaborators:
  • Dinh-Tuan Pham (CNRS)
  • Bruse Cornuelle ( Scripps Institution of Occanography)
  • Chris Farmer (Unversity of Oxford)
 
Description:

The goal of this project is to explore new directions to develop advanced data assimilation techniques with focus on high dimensional and strongly nonlinear models with focus on realistic ocean analysis and prediction applications. We are interested in both the deterministic variational and the stochastic Bayesian assimilation approaches. We are working on developing new efficient Bayesian assimilation techniques based on Gaussian mixture models and particle filters. We are also developing new approaches for efficient state/parameters estimation and assimilation into coupled models. We are also interested in the development of robust ensemble filters and how this theory can be related to the Bayesian filtering theory. Adaptive methods that allow the parameters of the assimilation system to learn from the assimilated data and the evolving model dynamics are considered. From the deterministic point of view, we are interested in improving the performance of the adjoint method with chaotic models and testing new adjoint-free variational assimilation approaches. We are also working on developing new efficient reduced variational assimilation techniques.