Radar forward operator for data assimilation and model verification for the COSMO-model

The new weather radar network of the German Weather Service (DWD) comprises 16 C-Band dual polarisation Doppler radar systems evenly distributed throughout Germany for complete coverage. They provide unique information about cloud structure and precipitation in three dimensions and high resolution. Up to now these data are not used in the operational COSMO-model of DWD (a non-hydrostatic limited area numerical weather prediction model), except within the framework of the latent heat nudging and for a simple nudging method of the radial wind. Future applications are however planned to make better use of radar data within an upcoming new LETKF data assimilation system, which will be based on a convection-allowing high resoluton ensemble forecasting system. Here, the use of weather radar data, which is able to provide measurements of dynamical and microphysical characteristics of precipitating clouds at high temporal and spatial resolutions, is a promising means for improvements of short-term precipitation forecasts.

However, the observations (reflectivity, radial velocity, polarisation parameters) are not directly comparable to the prognostic variables of the model. In order to, on one hand, enable radar data assimilation in the framework of the above-mentioned LETKF-assimilation system and, on the other hand, to facilitate comparisons of numerical simulations with radar observations in the context of cloud microphysics verification, a comprehensive modular radar forward operator is under development. This operator calculates the radar observables reflectivity, radial wind and (some) polarisation parameters from the prognostic model output. Given the planned range of applications, it has to be applicable on supercomputer architectures in an operational environment and efficiency is a major design criterion, which requires for good parallelization and vectorization properties of the code.

The operator consists of several modules, each of which handles a special physical process (scattering, extinction, microwave propagation, etc.). Each of these modules offers different formulations to choose from, which enables variable possibilities to configure the forward operator with respect to the often conflicting constraints of physical accuracy and computational costs. Here, a main goal is to find in some sense “optimal” configurations for both applications (data assimilation and model verification). For example: 1) the radar beam can be considered to propagate as a simple ray or with the actual volume averaging characteristics (beam function). 2) The beam bending can be either derived from a 4/3 earth radius concept or from the actual simulated vertical gradient of the refractive index of air. 3) Radar reflectivity may be calculated from the full Mie-theory or from various (more efficient) approximations, and attenuation effects may be taken into account or not. The strategy is to start from the "full detailed" operator and, by subsequent neglect, simplification and/or approximation of single physical effects, the "best" balance between computational efficiency and physical accuracy for each of the two applications is investigated. Later on, impact studies on radar data assimilation and studies on model microphysics verification will round out the project.

Currently the project is in the stage of finishing the development of the "full" operator. The presentation will show first results.