Thanks Curtis - that is a good idea. I should add that (unfortunatly) these elevation rasters are in integer feet, which tends to cause some "stair stepping" in the output derivative rasters (aspect, plan/prof curvature, hillshade, etc.). So for example, even the aspect pixels of one of the facets of the "crystal" does not entirely end up being a uniform value like you might expect. Had these been float rasters some of these detection methods would be quite a bit more discriminating I think. Best I have come up with is to add some additional masks - so for example, make sure it is shrubby/forested landcover & a low aspect value diversity & not a flat slope & relatively planar hillslope. Getting a bit better, but not a lot of hope for this one.