Assimilation of Radar Data in the Mesoscale NWP-System of DWD

The limited-area model Lokal-Modell (LM) of DWD is a non-hydrostatic mesoscale model for short-range numerical weather prediction (NWP). For operational purposes, the LM gets boundary values provided by the global model (GME) of DWD and the initial state is generated by an assimilation scheme based on the nudging technique. Using conventional data like surface, radiosonde, aircraft and windprofiler measurements the focus is on the analysis of meso-alpha-scale structures. In view of the development of a very high-resolution version of LM dedicated to very short range NWP of severe weather high-resolution precipitation data derived from radar networks are introduced in the assimilation. Using the Latent Heat Nudging (LHN) technique the thermodynamic quantities of the atmospheric model are adjusted in such a way that the modeled precipitation rates resemble the observed precipitation rates. This adjustment scheme works locally, because it is based on the assumption that in a vertical column the integrated release of latent heat is proportional to the precipitation rate at the ground. The aim of this approach is to assimilate observed meso-gamma-scale structures and adjust the moist processes in order to improve the quantitative precipitation forecasting (QPF). Recent changes in model dynamics include a prognostic treatment of precipitation, which takes into account the drifting of precipitation. This is found to limit the validity of the above mentioned basic assumption of LHN. The results from real case studies show that precipitation patterns are introduced in the analysis (data assimilation mode) in good agreement, both in position and amplitude, with those observed by radar if precipitation is calculated diagnostically. The performance of LHN becomes worse if the prognostic treatment of precipitation is deployed. During the free model run (forecast mode) the impact of LHN is limited to several hours. After outlining the theory and implementation of the LHN algorithm, results of real case studies and the sensitivity of model predictions to model setup and settings of the assimilation scheme will be shown.