Strategic Zoom

I don't know the exact details on how they did it, but I have a pretty good idea and its similar to what you mentioned with "layer of detail", though it's usually called "level of detail" (LOD). It's probably similar to the way mip mapping works, but with models instead of sprites. As you zoom farther and farther out from the surface, the model level of detail decreases. Eventually, the models get so small that you can't see them anymore and symbols would be more beneficial from the UI perspective.

I imagine that the same technique could be used for the terrain maps. The map data could contain multiple layers of terrain detail generated in the map editor. At the highest resolution, you're able to see individual rocks, pits, grooves, etc. and it is a very high vertex representation. As you zoom further out, rendering those millions of verts would become GPU intensive, so you'd switch the terrain to a slightly lower poly version which renders faster but with little or no noticable decrease in quality. The terrain in SupCom isn't deformable and the light source is constant, so as a further optimization, you could even bake the shadows into the terrain texture rather than calculating them each frame. Then, the only shadows you'd have to calculate are unit and cloud shadows depending in your LOD.