Gordon Inverarity, Met Office
The development of a hybrid 4D-ensemble variational assimilation system
CoauthorsNeill Bowler, Adam Clayton, Mohamed Jardak, Peter Jermey, Andrew Lorenc, Marek Wlasak, Dale Barker, Gordon Inverarity and Richard Swinbank
Abstract: The Met Office has been developing a hybrid 4DEnVar (four-dimensional ensemble variational) data assimilation (DA) system since 2011 for global weather forecasting. The hybrid covariances are obtained as a weighted sum of the stationary covariances used in our operational hybrid 4DVar (four-dimensional variational) DA system and an ensemble component, implemented as a so-called alpha control variable. The stationary covariance component is not evolved in time but the ensemble component naturally varies with time. The resulting increments are applied to the forecast using a 4D incremental analysis update method.
Experiments have been run on supercomputers at the Met Office, KMA (Korea Meteorological Administration) and ECMWF (European Centre for Medium-Range Weather Forecasts) exploring the use of small (23 then 44 member) and large (176 then 200 member) ensembles. Results will be presented examining the impact of changing the ensemble information used in hybrid 4DEnVar. Specifically, we have examined the use of different ensemble sizes, contrasted ensembles generated using the ETKF (ensemble transform Kalman filter) with the ensemble of 4DEnVars system described by Bowler in this meeting and explored the effects of applying different weights to the stationary and ensemble covariance components.
Using an En-4DEnVar ensemble and increasing both the number of members and the weight applied to the ensemble has benefited both hybrid 4DVar and hybrid 4DEnVar DA techniques. Nevertheless, hybrid 4DVar still outperforms hybrid 4DEnVar when these changes are applied to both methods but there is evidence that further work on tuning localisation settings may narrow the performance gap.