Data Assimilation Meetings at Reading
Date | Meeting type | Speakers |
8 Feb 2012 |
Informal DARC meeting Room Change: GU10 |
Roel Stappers (University of Reading) Variational data assimilation without nonlinear models In this presentation I show that by linearizing the tangent linear model around an ensemble of trajectories the nonlinear growth of perturbations can be described exactly by the tangent linear model. It is shown that in a three level quasi geostrophic model the tangent linear model can be used for more than 200 days. Based on this result a new incremental 4D-VAR method is introduced that 1) Does not require the nonlinear model to update the linearization trajectory in the outer loops (these updates are responsible for a significant fraction of the computational cost in the current ECMWF implementation). 2) Does not modify the innovation vector in the outer loops. To demonstrate the advantage of this we derive the exact equations for adjoint based observation impact including the effect of using multiple outer loops in 4D-VAR. 3) Does not require the strict distinction between inner/outer loops. The linearization trajectory can be updated in the inner loops at no additional computational cost. |