Stochastic Representation of Flow-dependent Model Errors in the GME Ensemble Prediction System

A stochastic kinetic energy backscatter scheme is used to represent the flow-dependent model errors in an Ensemble Prediction System (EPS) using the global model GME of the Deutscher Wetterdienst. The back-scatter scheme also serves as an alternative to represent the model errors in the Local Ensemble Transform Kalman Filter (LETKF) assimilation system of the GME EPS, since it reflects the errors due to dissipation or damping on the small scales by numerical diffusion and parameterization schemes. In our GME-EPS experiments, a Cellular Automated (CA) scheme and a power-law noise scheme are used to introduce stochasticity in the back-scatter scheme. Kinetic energy spectra of the forecast shows the double cascade, which indicates that the back-scatter in fact compensates for reduced energy up-scaling due to the over-dissipation, and seems to compensate for the absence of the inverse cascade. Our ensemble results show that the back-scatter significantly improves the forecast up to 6/7 days if it is used along with random white noise in the LETKF assimilation system. Our results also show the backscatter significantly improves the model forecast up to 4/5 days even without the assimilation system.