We have been doing precipitation experiments with a numerical weather prediction model. The model has been run with three different horizontal resolutions, 22, 11 and 5 km respectively. The results show that the model is able to forecast reasonable precipitation amounts in a broad sence, but it has difficulties to estimate the correct amount over small drainage basins. No significant improvements can be seen, when the higher horizontal resolutions are used. The results are sensitive to the magnitude of the horizontal diffusion, which is used in the model to control the horizontal spectra.
For orographically enhanced areas the result is much better, but also in these cases the higher resolution forecasts have to be strongly damped in order not to create noisy precipitation patterns. It is shown that smoothing the orography instead of a strong horizontal diffusion gives more realistic precipitation. It is suggested to compute the tendencies of the physical parameterization on a coarser grid than that of the dynamics.