[4-4]

Preliminary test of an EnVar for the operational AROME-France NWP system at convective scale

T. Montmerle, E. Arbogast, P. Brousseau and Y. Michel (CNRM-GAME (Meteo-France/CNRS))

Since 5 years now, Meteo-France is running an operational NWP system at 2.5 km horizontal resolution over France, aiming in forecasting surface processes and local meteorological events such as convective systems or fog. This system, called AROME-France, is based on a 3h cycle of assimilation/forecast steps that makes use of an incremental 3DVar, allowing to assimilate a comprehensive set of observations, including Doppler radar data.

In order to optimize the use of these observations, work is ongoing to add flow dependencies in the representation of forecast errors by considering covariances sampled from an ensemble of background perturbations. The background error covariances of the variational DA system can be either fully represented by these covariances using an EnVar formulation, or be expressed as a linear combination of the latter and of the modelized climatological ones that are used operationally in the 3Dvar, following a hybrid approach. In both cases, the filtering of sampling noise in the raw covariances is essential and is classically performed through the use of a localization procedure.

In this framework, preliminary tests are ongoing considering different types of ensemble and different subsets of observation. Localization functions are adaptively tuned using an original methodology that have been recently developed at CNRM, which is based only upon quantities deduced from the ensemble. Impact on analyses and on the resulting forecasts obtained with the EnVar and with the Hybrid DA systems will be displayed and compared with the 3DVar. The dependency of the results to the ensemble type and to the localization will be in particular discussed.