These images are based upon the randomly selected time step 12UTC on 2008-05-03. The layer is always the one with the lowest altitude since this is the one subjected to downscaling.

Fig. 1: Temperature as supplied by the COSMO model: The coarse data.

Downscaled data

Due to the existence of a Schomburg rule for temperature, it can be subjected to full downscaling (fig. 3,5,5.1). Figures 2,4,4.1 deal with results from mere interpolation.

Fig. 2: Temperature after spline interpolation. Fig. 3: Temperature after full downscaling. Fig. 3.1: Detail of figure 3, showing only the area Sauerland.

In cotrast to the merely interpolated field (fig. 2), the fully downscaled temperature field (fig. 3) shows fine scale structures. These have the form of rivers and are thus not numerical fluctuations (fig. 3.1).

Fig. 4: Temperature after spline interpolation and subsequent upscaling back to the coarse grid. Fig. 5: Temperature after full downscaling and subsequent upscaling back to the coarse grid.

Fig. 4.1: Temperature bias after spline interpolation, on the coarse grid. Fig. 5.1: Temperature bias after full downscaling, on the coarse grid. The axis limits have changed.

While the other displayed fields model absolute temperature, the bias field is a difference of 2 absolute temperatures and can fall below 0. While the bias after interpolation is evenly distributed above and below 0, the bias after the full downscaling once assumes an unusually negative value, –6*10^–5 K. Still, all these values fall short of the usual temperature difference span of about 9 K. The reasonable interpretation is as numerical fluctuations.