The mesoscale numerical weather prediction modelALADINhas been applied for downscalingERA40data onto a 10 km grid covering the complex terrain of Slovenia. The modelled wind field is compared with the time-series of observations at 11 stations. In addition to traditional scores (root-mean-square error, mean absolute error, anomaly correlation), a frequency-domain comparison is carried out in order to explore aspects of the mesoscale model performance other than that depicted by the conventional statistics. The verification period is the Special Observing Period of the Mesoscale Alpine Program (MAP-SOP), for which ECMWF reanalyses including MAP-SOP observations are available every 3 hr on a ∼40 km grid.

Traditional scores indicate that the downscaling has been successful. Scores are little dependent on the nesting strategy (direct versus two-step nesting), in spite of a ratio of horizontal resolutions between ERA40 and ALADIN as large as 12. The model performs best at mountaintop stations, characterized by over 80% of their spectral power in motions with longer than diurnal periods. A majority of stations is, however, located in the complex terrain where around 40% of the spectral wind power is contained in the subdiurnal frequency range. This part of the spectrum is significantly underestimated by the model, indicating that the downscaling is predominantly a dynamical adjustment to the new terrain. At the same time, the MAP-SOP reanalyses of the ECMWF model include relatively more power in the subdiurnal frequency range than ALADIN. However, these subdiurnal oscillations do not agree with observations and their removal improves conventional scores for the MAP-SOP wind data.

It is suggested that a frequency-domain comparison is a useful complement to the conventional statistics and it enables a more physical insight into a mesoscale model performance.