Generating terrain

So to me, at first glance, it looks like this guy is implementing a sphere-cube map, and is (obviously) using chunked LOD (essentially, each face of the sphere cube is a quadtree, and you subdivide a quad when the camera is close enough to need more detail), to minimize the detail at which terrain needs to be generated.

Way back when I first started experimenting with planet mapping, I was working on implementing a sphere cube, but some time ago I decided that it would be a better idea to switch to using large polygons made of arcs that represent continent cratons or other large structures, and instead of the sphere cube (which has some nasty length warping as you go from cube surface to sphere and back) we'd use an icosahedral tessellation of the sphere (project an icosahedron onto the surface of the sphere) to partition the polygons for collision detection. Never got this implemented, so maybe I'll change my mind again, dunno.

Anyway, once we start wanting to generate static planetary surface meshes on a large scale, we will certainly want to do LOD, and will probably project everything onto a sphere-cube because it's simpler for static meshes (the distance warping doesn't matter so much when continents don't move around, since we don't have to worry about preventing the distance warping from changing their shapes if they don't move).