Title: Accounting for model error in global and regional ensemble data assimilation systems

Authors: Laure Raynaud (Meteo-France)
Loik Berre (Meteo-France)
Gerald Desroziers (Meteo-France)

Ensemble data assimilation (EDA) is now a widespread technique to sample analysis and background uncertainties. A variational EDA system has been running operationally at Meteo-France since july 2008 for the global Arpege model. It is used to calculate flow-dependent background-error statistics, and it also provides perturbed initial states for the global ensemble prediction system. Another important area of ongoing research at Meteo-France is the experimentation of ensemble variational assimilation with the high resolution regional model Arome (with a 2.5 km resolution).

An important aspect in both systems is the representation of model error uncertainty. An adaptive multiplicative inflation has first been investigated in the Arpege EDA. An observation-based estimate of background-error variances is extracted from diagnostics relative to the minimum of the variational cost function. This can be compared to ensemble-based variances in order to estimate the contribution of model errors to background-error variances. This model-error information is then used to implement a multiplicative inflation of background perturbations after each 6h forecast step. This approach leads to a more realistic ensemble spread and the new ensemble-based background-error statistics have a positive impact on the forecast skill. It is thus considered to implement this approach operationally in the near future. Similar techniques are being tested in the regional EDA. First results will be presented.

GMAO Head: Michele Rienecker Global Modeling and Assimilation Office NASA Goddard Space Flight Center

Curator: Nikki Privé Last Updated: May 27 2011