Abstract :
[en] CGMS currently operates on observed station data. Switching from this classical approach to a modelled data from European Centre for Medium-range Weather Forecast, ECMWF, has to be tested before applying it operationally. In this study, we concentrated on the impact of the downscaling of meteorological data with grid sizes ranging from 1.875 to 0.35 degrees to the new CGMS grid size of 25 X 25 km.
Four different grid sizes corresponding respectively to the dimensions of the Operational, EPS, Monthly and Seasonal ECMWF Models were checked. The control was done on daily data of 25 stations selected on a 2-year period in a window covering the South of Germany and Czech Republic, a large part of Austria and the North of Italia in order to analyse the downscaling impact on plains, mountainous and coastal zones.
For each grid size, four different downscaling methods were applied: the reference method that uses the classical CGMS interpolation procedure, the nearest neighbour approach, and two more complicated interpolation techniques using the Model Output Statistics developed by Meteo Consult (MC-MOS). The analysis showed that the accuracy of the downscaling procedure is largely influenced by the input data grid size. RMSE increases between the smallest and the largest grids are respectively 59%, 51%, 33%, and 33% for Tmax, Tmin, Wind Speed and Radiation fields. Rainfall does not seem to be affected by the downscaling process but the RMSE is high in all cases. In most cases, the best interpolation method is also the more complex one and the one that requires more computer time to be calculated. RMSE decrease of 22%, 50%, 57% and 28% respectively for Tmax, Tmin, Wind speed and Radiation fields when we compare the best interpolation method results with the reference approach. An exception in this general rule is for the rainfall rate estimation whose accuracy is not always best with the most complex interpolation technique.
