P-3

Impact of different sources of uncertainty in convective-scale EPS

C. Keil (LMU), F. Harnisch (LMU), C. Kuehnlein (ECMWF)

The effective representation of the sources of uncertainty in convective-scale EPSs is a crucial but largely open issue. A common approach to account for initial condition (IC) uncertainty in high-resolution EPSs is based on dynamical downscaling of an ensemble of coarser-resolution driving (global) model forecasts. The downscaling approach is attractive due to its relative simplicity and practicality of implementation (and is used in COSMO-DE-EPS at Deutscher Wetterdienst). However, the downscaling approach cannot comprehensively address IC uncertainties at the scales represented by the high-resolution EPSs.

One novel approach to account for IC uncertainty in convective-scale EPSs offers the combination with ensemble data assimilation methods. The novel kilometer-scale ensemble data assimilation (KENDA) system for the COSMO model is based on a Local Ensemble Transform Kalman Filter and provides, in principal, an analysis ensemble that can be used to initialise convection-permitting ensemble forecasts. Error sources in the model formulation inevitably result in a systematic underestimation of variances cycled in ensemble data assimilation. Inflation methods and stochastic physics represent different techniques to enhance ensemble variance and improve forecast quality.

Results gained with operational COSMO-DE-EPS forecasts and experimental ensemble forecasts based on KENDA-COSMO address these issues and highlight the role of different realizations of uncertainty.