Lilo Bach, Meteorological Institute, University of Bonn

Ensemble data assimilation for regional reanalyses

Coauthors
Christoph Schraff and Andreas Hense

Abstract:

Data assimilation is the key ingredient of regional reanalysis systems. A newer development in the field of regional reanalyses is the incorporation of uncertainty estimation. In the framework of the FP-7 funded project Uncertainties in Ensembles of Regional Reanalyses (UERRA) we seek to identify a method which is cost-efficient, competitive in accuracy and allows for a comprehensive uncertainty estimation of both the ensemble of reanalyses and the subsequent reforecasts. We present experiments performed with a system developed for the production of the reanalysis COSMO-EN-REA12 within UERRA. It is based on a COSMO-EU set-up adapted to a grid spacing of 12 km and includes external snow, SST and soil moisture analyses. The employed ensemble data assimilation scheme is ensemble nudging. Thereby, independent realizations of nudging (e.g. Schraff,1997) runs are generated which assimilate observations perturbed by random samples of observation error. Next to observation uncertainty, model error and uncertainties in the lateral boundary conditions lead to error growth in regional reanalyses. In a first step, we assess the relative impact of these different sources of uncertainties by running nudging in combination with (1)perturbation of the assimilated observations (ensemble nudging) to account for observation uncertainty, (2)a perturbed physics ensemble similar to COSMO-LEPS and/or stochastic perturbation of physical tendencies to incorporate model error and (3)an ensemble of lateral boundary conditions provided by the global reanalysis ERA-5 or ICON. Based on these results, combinations of the three methods are tested to identify a method that yields the best average spread-skill ratio. To provide guidance for potential future implementations of regional reanalysis systems comparisons to similar experiments with a Local Ensemble Transform Kalman filter (Schraff et. al, 2016, accepted) are planned.