a) By high-resolution calculations with the numerical model MM5 driven by boundaries from the ECMWF reanalysis.

b) By a multiple linear regression model that calculates precipitation values on a regular grid from observations.

Both methods agree on the basic pattern of the precipitation and they do not show large differences in precipitation values. The linear regression model is computationally inexpensive and can be run at a very high resolution with low cost. Consequently, it can be run to show patterns attributed to small scale topography. The numerical simulations give on the other hand more valuable information in data-sparse mountainous regions. The linear model serves to validate the numerical simulations for downscaling of future general circulation scenarios.