I've been wondering, why is this not a default feature?
I don't own the game yet, but from what I've seen, flying is indeed too high and we can barely see anything below the ship.
Long story short, the terrain engine belongs to a family of algorithms called "chunked lod", in which the terrain is divided in a number of "chunks", cubic in NMS' case. As you approach a chunk, it subdivides into smaller chunks. Though there's preloading and caching going on, a chunk doesn't exist in memory until it's requested. Once it is requested it takes some time before it actually exists because it has to be either loaded or, in the case of procedural generation, generated. You can of course see this at work while you're flying around, with more detailed chunks popping around constantly, but that also means you don't have reliable info about the geometry of chunks that aren't in memory.
NMS uses voxel terrain. That means chunks are much more expensive to generate than in planetary engines using chunked lod with heightmap terrain, like kerbal or elite (jumping from quadratic to cubic complexity). Heightmap terrains also have one huge advantage, in that there's a pretty cheap way to know exactly how high the terrain at an arbitrary point is
even if you haven't actually generated the matching chunk. with voxel terrain you may have an informed estimate depending on how your generation works and based on lower detail chunks, but querying the exact height is tons more expensive and basically fuzzy unless you have the actual chunk on hand.
You don't want your player to fly into what the engine considers empty space one frame and a huge block of copper the next.
So at runtime, when you have limited resources per frame for generation, which chunks do you favor? The ones right under the player, which they usually do not see but will allow for flight collisions, and which in the case of a fast moving flying object will have become useless on the next frame already? Or the ones that are gonna be in the player's field of view, the ones the player will actually see and judge you game's visuals by, and the ones they're going towards?
Ideally you don't want the player flying at high speed low altitude anyway, because the engine won't have time to generate the chunks. Note how for example google earth in 3d mode doesn't load more detailed chunks until you slow down.
So they went with a reasonable solution, which is to limit the lower flight altitude to be above conservative terrain estimates. Then whenever you land there's time during the landing sequence to generate the high detail chunks and on foot you can't outpace the generation.
