Ivan Kasanický, Institute of Computer Science, The Czech Academy of Sciences

Hybrid EnKS/4DVar assimilation in the WRF model 

Jan Mandel, Kryštof Eben, Pavel Juruš, Aimé Fournier, Serge Gratton, Selime Gürol


The EnKS-4DVAR method, using the ensemble Kalman smoother (EnKS) as a linear least-squares solver in the Gauss–Newton method for weak-constraint 4DVAR, has been proposed recently. Similarly to some other method combining ensemble and variational approaches, there is no need for tangent or adjoint operators which is the greatest asset of the algorithm. The EnKS-4DVAR, however, converges to the 4DVAR iterations as the EnKS becomes more accurate as a linear solver. An additional Tikhonov regularization term, added to the cost function, ensures the convergence of the Gauss-Newton method by turning it into Levenberg-Marquardt method, and it is implemented as an additional independent observation within the EnKS. The Weather Research and Forecasting (WRF) model is used to perform the standard pseudo observation test. Its outcome together with preliminary results of classical twin model experiments will be presented.

The research has been partially supported by NSF grant DMS-1216481 and Czech Science Foundation grant GA13-34856S.

File available here