The German Weather Service (Deutscher Wetterdienst) has recently developed a new operational global numerical weather prediction model, named GME, based on an almost uniform icosahedral-hexagonal grid. The GME gridpoint approach avoids the disadvantages of spectral techniques as well as the pole problem in latitudelongitude grids and provides a data structure extremely well suited to high efficiency on distributed memory parallel computers. The formulation of the discrete operators for this grid is described and evaluations that demonstrate their second-order accuracy are provided. These operators are derived for local basis functions that are orthogonal and conform perfectly to the spherical surface. The local basis functions, unique for each grid point, are the latitude and longitude of a spherical coordinate system whose equator and zero meridian intersect at the grid point. The prognostic equations for horizontal velocities, temperature, and surface pressure are solved using a semi-implicit Eulerian approach and for two moisture fields using a semi-Lagrangian scheme to ensure monotonicity and positivity. In the vertical direction, finite differences are applied in a hybrid (sigma pressure) coordinate system to all prognostic variables. The semi-implicit treatment of gravity waves presented here leads to a 3D Helmholtz equation that is diagonalized into a set of 2D Helmholtz equations that are solved by successive relaxation. Most of the same physical parameterizations used in the authors' previous operational regional model, named EM, are employed in GME. Some results from the verification process for GME are provided and GME performance statistics on a Cray T3E1200 as well as on the ECMWF Fujitsu VPP5000 systems are summarized. For the case of the severe Christmas 1999 storm over France and Germany the pronounced sensitivity of the model with respect to the initial state is discussed. Finally, a test case is shown where it is currently possible, though not yet operationally practical, to run GME at 15-km resolution on the VPP5000.

The most common option for numerical models of the atmosphere is to use model layers following the surface of the earth, using a terrain-following vertical coordinate. The present paper investigates the forecast of clouds and precipitation using the z-coordinate nonhydrostatic version of the Lokalmodell (LM-z). This model uses model layers that are parallel to the surface of the sphere and consequently intersect the orography. Physical processes are computed on a special grid, allowing adequate grid spacing even over high mountains. In other respects the model is identical to the nonhydrostatic terrain-following version of the LM, which in a number of European countries is used for operational mesoscale forecasting. The terrain-following version of the LM (LM-tf) is used for comparison with the forecasts of the LM-z. Terrain-following coordinates are accurate when the orography is shallow and smooth, while z-coordinate models need not satisfy this condition. Because the condition of smooth orography is rarely satisfied in reality, z-coordinate models should lead to a better representation of the atmospheric flow near mountains and consequently to a better representation of fog, low stratus, and precipitation. A number of real-data cases, computed with a grid spacing of 7 and 14 km, are investigated. A total of 39 real-data cases have been used to evaluate forecast scores. A rather systematic improvement of precipitation forecasts resulted in a substantial increase of threat scores. Furthermore, RMS verification against radiosondes showed an improvement of the 24-h forecast, both for wind and temperature. To investigate the possibility of flow separation at mountain tops, the flow in the lee of southern Italy was investigated.