Title: Information-based data selection for ensemble data assimilation

Authors: Stefano Migliorini (Department of Meteorology, University of Reading)

A well known limitation of ensemble filtering is the inability of providing a sufficiently accurate representation of the posterior density function in the presence of a large number of sources of observational information. This problem is usually dealt with through procedures -- e.g., Schur-product localization and local ensemble-transform Kalman filtering (LETKF) -- that limit the number of measurements that are allowed to constrain the analysis at a given location. One of the drawbacks of the Schur-product approach is that the radius of localization should be large enough not to disturb the balances that act at given spatial scales and that are well represented by the ensemble error covariance. Also, in the LETKF case, the radius of localization should be large enough to include enough observations to provide a meaningful analysis. A radius of localization that is too large may limit the ability of current localization procedures to reduce the amount of observational information, particularly over data-dense areas. Another limitation of localization procedures is in dealing with non-local observations (e.g., satellite radiances), particularly in the vertical.

In this talk we will first show results illustrating the effects of not using localization at all for a case study where the forecast ensemble is determined with an ETKF at convective scale, with and without radar data. Also, we will discuss the theoretical basis of a new localization procedure based on the information content of the measurements and show results of a number of ensemble data assimilation experiments with a two-dimensional (horizontal and vertical) linear advection model using both local and non-local observations.

GMAO Head: Michele Rienecker Global Modeling and Assimilation Office NASA Goddard Space Flight Center

Curator: Nikki Privé Last Updated: May 27 2011