To first order, rock and ice have similar dielectric properties.

In other words, you can't tell them apart without high resolution, low noise data.

Since the geometry of aquisition is not SAR, but point-source radar, ESA will not be able to distinguish "ground clutter" (off-vertical reflections from rugged ground topography) from weak secondary events at depth.

In the seismic business we refer to this sort of confusion as "out-of-plane" events, or "sideswipe". Events along-track can be mathematically resolved and restored to their point of origin by aprocess of 2D migration (collapsing reflection hyperbolae into point reflectors). Events cross-track cannot, since the data is not sampled in a direction the spacecraft has not travelled.

The only cure for sideswipe is closely-spaced parallel track lines. It will take more than 3 years in orbit to aquire even the beginnings of enough data for 3D migration of surface-derived noise, although some smart tricks may be possible by using MOLA and Viking DEM data as a proxy for surface shape in the cross-line direction.

In short, this is a huge problem. The expertise exists to help solve it, but it will take time, money, and lots of data.