Data Assimilation Meetings at Reading

Date Meeting type Speakers
13 June 2012 DARC seminar David Fairbairn (Uni. of Surrey)
Coupling 4D-Var with 4D-En-Var (PDF)
Four data assimilation methods are compared using toy models for their ability to produce a deterministic analysis: (i) 4D-Var with a perfect climatological background error covariance matrix; (ii) An ensemble of 4D-Ensemble-Vars (abbreviated to 4D-En-Var), which provides a least squares fit between the observations and an ensemble in an assimilation window; (iii) 4D-Var-Ben, which is 4D-Var that uses the flow-dependent background error covariance matrix from 4D-En-Var; (iv) DEnkf, which is an approximation of the ensemble square root filter. We show that the time correlation of the observations in 4D-En-Var and the DEnkf is negatively affected by localization, since the background error covariance matrix is localized after it has been propagated by the nonlinear model, but the localization function and the nonlinear model do not commute. The 4D-Var-Ben provides time correlation of the observations using the tangent linear model and performs significantly better than 4D-En-Var when severe localization is required i.e. for a small ensemble size.

Page navigation