I'm not sure why ice would matter given that it should weigh the same as in the liquid form, but...

How do you measure a known quantity of water to within a few tens of parts per billion? Do you specify the proportion of light to heavy water? (For 'normal' water about 1 in 41 million.) What about contaminants in the water? There are so many uncertainties that it's impossible for all practical purposes.

The whole point of using 'rare' silicon is that we already have commercial processes for producing ultrapure, defectless, monocrystals of the stuff (for use in computer chips), so it's easy to reuse that technology for defining a kilogram.

The prototype kilogram was made of platinum-iridium because they are very nonreactive and fairly hard, so once the prototype was made it would be unlikely to be accidentally modified.

If you tried to use water ice as a standard, you'd find that its mass was changing all the time as the ice sublimated away, or humidity condensed onto its surface. It would be impossible to make precise measurements.

Because you don't want that lump of mass to change in any way while it is being used as a reference.

Are you aware of all the matter exchange processes that are continuously going on at the surface of water ice under normal conditions?

Necroposting, I know, but under what definition is silicon a "rare metal"?