Group seminar on 27. April, 14:15 CET
NWP and Data Assimilation developments at Deutscher Wetterdienst
In 2015 the ICON model became the operational numerical forecast system at Deutscher Wetterdienst generating a remarkable increase of forecast accuracy. But also changes in the data assimilation system (DA) are contributing to the enhanced predictability. Many new types of observations have been included. Moreover, the local ensemble transform Kalman filter (LETKF) provides the ensemble of initial states for the global ICON-EPS and by the way enables the use of flow dependent background error covariances in the high resolution EnVar DA for the deterministic run. At the moment we are working on an operational 4DEnVar system for short assimilation windows (3h) which become more and more important in very short range weather forecasting when connecting statistical extrapolation (nowcasting) to NWP.
For the longer ranges it has been shown to be of advantage to run a 4DVar data assimilation in a longer time window (12h to 24h). The 4DVar adapts a free model forecast to the observations by modifying the background state at the beginning of the assimilation window. This requires a linear version of the nonlinear forecast model to run forward linear and backward in time adjoint integrations. In the ICON modelling framework there is no linear model version and consequently, we cannot run a classical 4DVar. The same holds for the singular vector (SV) technique. In the SV context we are developing a new mathematical algorithm based on the Arnoldi method which replaces the classical Lanczos scheme and exclusively runs full non-linear model forecasts forward in time. Interestingly, the SV technique not only provides estimations of the fastest growing modes but also of the fastest shrinking ones what opens up the possibility of using these modes in DA. This talk will discuss the potential use of SV in data assimilation showing first results from a Lorenz96 model.