One important goal of operational high-resolution meso-b scale numerical weather prediction models like the ``Deutschland-Modell'' (DM) of the DWD is the detailed simulation of the orographic modification of precipitation. Orography impacts the precipitation pattern due to several reasons (Fig. 1) such as large-scale upslope precipitation, the enhancement of rainfall over small hills or the triggering of convection. These processes are considered in the model, too, but perhaps not detailed enough. Additionally, several numerical procedures like horizontal diffusion or the use of a reflective upper boundary condition may impact the precipitation pattern artificially. Thus a careful comparison between the observed and modelled precipitation pattern in mountainous terrain based on high-resolution data helps in estimating the quality of the model.
Model: The operational DM has a meshsize of 0.125 ° ( ~ 14 km) and 20 layers. The precipitation for the period +6 up to +30 hours of each 00 UTC forecast is collected for the seasons winter 93/94 (Dec. 93, Jan., Feb. 94), spring 94 (Mar., Apr., May), summer 94 (Jun., Jul., Aug.) and autumn 94 (Sep., Oct., Nov.).
Observation: The climatological precipitation network consists of some 4300 raingauges in Germany, i.e. the average station density is some 10 km. The temporal resolution is 24 hours (approximately 06-06 UTC). These data have been quality-checked and gridded to a 1 km-net by Mueller-Westermeier (Climatology Dep.) and then averaged to the DM grid by Link (Research Dep.). The number of DM gridpoints in Germany is around 1800.
Table 1 compares the area-mean precipitation for Germany and the four seasons. Whereas the model verifies well for spring, it underestimates precipitation by up to 20% in the summer of 1994. The latter was probably due to an overestimation of the entrainment rate in convection and an underestimation of the transpiration by plants. Both have been corrected in spring 1995 and preliminary results indicate a clear improvement of the convective precipitation in the model.
| Season | Obs(mm) | DM(mm) | DM/Obs(%) |
| Winter 93/94 | 280 | 237 | 85 |
| Spring 94 | 262 | 262 | 100 |
| Summer 94 | 220 | 175 | 80 |
| Autummn 94 | 199 | 180 | 90 |
Table 1 Observed (Obs) and predicted (DM) precipitation in Germany for the four seasons 93/94.
| Reality (R.B. Smith, 1979) | Model (DM, D ~ 14 km) | Impact on precipitation | |
| Large-scale upslope precipitation | 
| Included, but `correct' orogra phy? | Maximum rainfall near or slightly upwind of steepest sur face slope. |
| Enhancement of rainfall over small hills | 
| Included; seeder-feeder; but enough orographic detail at 14 km? | Maximum rainfall on hill top. |
| Orographic-convective showers | 
| Slopes not included in radiation, elevated heat source; but cor rect closure for convection? | Maximum rainfall downwind
of top. |
| Föhn effect | 
| Included; but mountain wave in fluenced by UBC and hydrstatic assumption. | Suppression of precipitation in the lee. |
| Wind drift of snow | Not included.
Truncation errors. | Increase of snowfall in lee.
? |
Figure 1 Impact of orography on precipitation, reality vs. model.
Figure 2 Top row: Orography of the DM and the Laplacian
of orography for a region near the Harz Mountains.
Bottom row, left: Observed precipitation for the winter season
93/94.
Middle: Predicted by DM, period 06 up to 30 hours of each 00 UTC
forecast.
Right: Ratio model to observed.
Concerning the spatial distribution, the area near the Harz Mountains is presented in Fig. 2 for the winter 93/94. Observed precipitation maxima of more than 500 mm are closely connected to the mountain slope with shading effects visible east of the major peak. The model is able to pick up this spatial structure but overpredicts the impact of orography on the precipitation pattern, e.g. up to 50% more precipitation than observed at the Harz Mountains and 50% underestimation in the lee. This deformation of the precipitation pattern in the DM can be mostly attributed to an overprediction of the hydrostatic mountain waves, especially the downslope winds in the lee. Increasing the roughness length in mountainous terrain considerably reduces the amplitude of the mountain waves and improves the spatial distribution of precipitation.
A second prominent model error is the strong dependence of precipitation amounts on the Laplacian of the terrain height whereas observed precipitation tends to be more dependent on the terrain height itself. In the model, each peak of the orography introduces a pronounced local maximum of precipitation, even hills of some 50 to 100 m height in Northern Germany increase the amount by a factor of three, which is not supported by the observations which suggest a factor of two at most. The close correlation between precipitation and the Laplacian of the orography in the DM is probably due to the linear fourth order horizontal diffusion on model layers which tends to moisten the mountain peaks at the expense of the valleys. Preliminary test forecasts using a significantly reduced horizontal diffusion coefficient in the lower troposphere improved the spatial distribution of the precipitation in the DM considerably, but the impact on forecast quality has to be more carefully evalutated prior to an operational implementation.
The DM is able to catch the impact of orography on the precipitation
field in Germany in some detail but has a clear tendency to overestimate
the contrasts between the upwind and lee sides of the mountains
and to concentrate the precipitation maxima on the tops of the
hills. These systematic errors of the model have to be corrected
in the near future to provide reliable precipitation forecasts
for hydrological applications.
MAP Data Centre - April '05 - MAP WebMaster