The problem is the following:
The HRES model, because of the high resolution, is able to distinguish local phenomena such as glaciers. But the Ensemble (ENS), due to it's lower resolution, does not resolve some glaciers (for example over the Alps).

What the model does:

• The glaciers are represented on the HRES with 10m snow depth
• The HRES is used to initialise the ENS, and the ENS is at lower resolution than the HRES
• In the ENS, at day 10 with the change of resolution (legA/legB) there is a further interpolation that will smooth the snow fields

Thus an interpolation takes place that results in flattening the glaciers by reducing the snow depth during interpolation.

ECMWF therefore advises users that the snow depth on these areas should be used with caution due the issues with the treatment of glaciers and interpolation.