The thing about even a coarse heightmap is that the dataset is huge.
My point is that we should not care about the shape of mountain x, only that it is one of many mountains in the area.
If we are to generate heightmaps, then we would have to put a lot of work into going "backwards", distilling a heightmap into data about drainage basins, mountainous areas, etc.
Luckily, we don't have to do that. Many, maybe even most, methods for creating realistic-looking terrain involve lots of steps with conditionals carefully tweaked, so, for example, you don't get fjords in the middle of a flat desert.
Outerra doesn't help, because you have to get the original dataset from somewhere, and it will only be as realistic as your generation method. I've essentially detailed a physically-based planetary-generation method above, and the fact that it is physically-based is important, as not only does that mean that we can get a realistic result for an earth-like planet, but we can get realistic results even with different parameters.
I've read through dozens of papers on terrain generation, and almost all of them focus on realism in the small scale (ie, mountains which look like mountains, etc), but neglect realism in the large scale, which is what we actually need for this game. We do not need mountains which look good. I would love to have mountains which look good, I would love to work with you on making them happen, but they are not needed for the actual simulation that forms the core of the game.
So what I'm saying is that what you thought the game would do, and what I'm suggesting the game do, are (almost entirely) non-conflicting. My algorithm above would spit out a bunch of info, saying such things as:
- This area over here (which coincidentally is the size of France) is a bunch of cold sand dune desert.
- This area over here has these species in these amounts.
- This area over here has a lot of mountains which are about this eroded, with fjords and a lot of volcanism.
From that data, we can generate terrain for rendering, on the fly, possibly with some precomputation if necessary, and only for the area which our species will be exploring for this round. And we can do so using whatever methods you like.
Sorry if I sound snippy, btw, I just escaped a heated argument IRL. I could edit this for tone once I've chilled out a bit.