Presenter: Ibrahim Hoteit (KAUST)

Description:
The Background Error Covariance (BEC) is a critical element in any data assimilation (DA) system as it spreads the observations information between model variables. Ensemble Kalman Filter (EnKF) DA systems provide an efficient framework to update the BEC based on the current model dynamics and observations, so-called flow-dependent BEC. The robustness of the EnKF BEC strongly depends on the ensemble size, i.e. number of samples used to describe the ocean state distribution (mean and covariance in an EnKF) . In real-time applications only limited ensembles (~1-100 members) can be however afforded. Large Ensemble experiments (LEEs) can provide robust BEC that may help devising better BECs by for example revealing missing information from the small ensembles or improving/tuning covariance localizations and inflations techniques, used to compensate for large ensemble. EnKFs further assume Gaussian state distributions, which may not be valid in highly nonlinear ocean regimes. LEEs provide enough samples for better assessment of the ocean state distributions and their types. We conducted a series of 1-year-long LEEs, starting from 50 to 5000 members, using the Red Sea ensemble data assimilation system that uses a 4km MITgcm forced with ensembles of ECMWF atmospheric fields for forecasting and assimilates real observations of sea surface temperature, sea surface height and temperature and salinity profiles. Results are analyzed under three different scenarios: (i) ensembles are integrated freely without assimilation, and ensembles evolve while assimilating observations (ii) with and (iii) without the covariance localization technique. The resulting ensemble statistics will be presented and discussed.