Title: Dealing with convergence problems when accounting for correlated observation errors in image assimilation.

Author: Chabot Vincent (Météo-France)
Nodet Maëlle (Université de Grenoble)
Vidard Arthur (INRIA)

In the last decades, satellite image sequences have increased in both quantity and quality. Compared to conventional observations, satellite images have some advantages and drawbacks. On the one hand, satellites can provide regularly dense observations of any region on earth. On the other hand, those images are affected by spatially correlated errors. Most of the time, observation error correlations are ignored in data assimilation methods. In order to incorporate consistent information, the no correlation hypothesis is partly compensated by observation thinning methods in association with variance inflation. Those two operations result in discarding a huge part of the information content of satellite image sequences.

In this presentation, we investigate a method based on a multi-scale (spectral, wavelet, curvelet) transform in order to represent (at a cheap cost) some of the observation error correlation in a variational data assimilation context. The proposed approach consists in considering an image sequence in a multi-scale space instead of pixel space. We show that the diagonal of the covariance matrix in such spaces (if well chosen) is able to represent an important part of the error covariance (in pixel space). The benefit of this approach is demonstrated in twin experiments involving a 2D-shallow water model and synthetic observations. However, in some cases, taking into account the error correlation leads to a convergence problem. Such a case is presented, as well as some solutions (involving the multi-scale aspect of our observations) to improve the convergence rate.

GMAO Head: Steven Pawson Global Modeling and Assimilation Office NASA Goddard Space Flight Center

Curator: Nikki Privé Last Updated: Feb 9 2015