Advantages of the Two-Moment Microphysical Scheme for the Prediction of Deep Convection in a Numerical Weather Prediction Model

During the last years development in the German Weather Service (DWD), as well as in many other weather centers, has been aimed to improve the practical predictability of deep convection in numerical weather prediction (NWP) models. At a grid spacing of 2.8 km or finer our limited area model, COSMO, is able to resolve deep convection sufficienctly well, and therefore classic deep convection parameterizations are no longer needed. DWD's new data assimilation system, KENDA, which is based on a local ensemble transform Kalman filter (LETKF), allows for directly assimilating radar reflectivities as well as radial winds, which has the potential to reduce uncertainties in the initial conditions at the convective scale.

In this presentation we discuss the advantages of using the 2-moment microphysical scheme of Seifert and Beheng for predicting deep convection in NWP, with respect to the currently operational 1-moment scheme. Using the LETKF we are now able to construct consistent initial conditions for each microphysical scheme, a prerequisite for a fair comparison of the scheme in forecast experiments. We observe that the 2-moment scheme produces a more realistic simulated radar signal compared to the one-moment scheme. For example, high reflectivities that appear as a result of few and large hydrometeors (like graupel or hail) are underestimated or completely absent in the operational 1-moment scheme, but are well captured by the two-moment microphysical scheme. A more realistic radar signal has a positive impact for the assimilation of radar reflectivities, and thus has the potential of improving the initial conditions in the forecast. Second, we expect that the 2-moment scheme produces a more realistic convective dynamics.

We compare the performance of the 1-moment and 2-moment schemes in a set of experiments that resemble the DWD operational assimilation cycle and forecast. The experiments run for one week in May-June 2016, a period characterized by strong convective activity over Germany. In all experiments we assimilate volume scans of radar reflectivities and the corresponding radial winds of the German radar network. We concentrate on short-time forecast, up to 6 hours. We find that the 2-moment scheme produce more realistic deep convection, which results in an improved forecast skills for the evolution of convective cells on short time scales.