Last updated: 12 Jan 2023
The Working Group No. 1 (WG1) on data assimilation takes care of the data assimilation system for the limited-area mode of ICON, i.e. ICON-LAM. For the time being, it also provides support on the data assimilation for the COSMO model, and a small amout of the development work is still carried out with COSMO.
At several COSMO members, the KENDA (Kilometre-scale Ensemble-based Data Assimilation) system based on the LETKF (Local Ensemble Transform Kalman Filter), often complemented by latent heat nudging for the assimilation of radar-derived precipitation rates, is deployed for the operational model configurations. While the first operational applications of KENDA started in 2016 and 2017 for the COSMO model and in 2021 for ICON-LAM, the previous continuous data assimilation based on observation nudging is still in operations for COSMO at some members. Observation nudging is not available for ICON-LAM.
Hence, the research and development activities in WG1 focus on the KENDA system and its further development. Most of this work takes place within the COSMO priority project KENDAscope which is aimed to further develop and refine the KENDA system. On the one hand, this includes new algorithmic options in addition to LETKF, namely variants of EnVar (Ensemble-Variational data assimilation) and Particle Filters, and on the other hand the possibility to assimilate additional observation types such as all-sky satellite radiances. A comprehensive description of the project KENDAscope can be found here. This also includes a background and motivation section which explains the main decisions for the further development of the algorithms and the use of observations.
An even slightly broader view compared to the KENDAscope page is given in the 'WG1 Guidelines'. This does not only cover the activities of the project KENDAscope and its motivation, but also addresses further aspects.
It is noted that since observation nudging is not available for ICON-LAM, a cheap variant of EnVar is also being developed as an option to avoid the need of running an ensemble data assimilation at convective scale when only a deterministic analysis is required: CEnVar uses the background error covariances from a 'c'oarse-scale ensemble, e.g. ICON-EU-EPS or (the global) ICON-EPS, and requires to run only a deterministic assimilation cycle at high resolution.