At time, the problem is the limitation in the definition for a heightfield.
I am thinking about establishing a new heightfield definition, which would be able to show caves as well, and much more structures in high sloped areas.
But because GeoControl does not offer high qualtiy rendering, it is up to the rendersoftware developers, to start up a dialog, where such new definition are discussed, to get a new standard.
Here is my idea:
The new heightfield definition is build up on a normal heightfield, like it is used now. A second byte heightfield contains a table, where for all points of the heightfield it is stored, how many heights are assigned to this point. A cave, simple would have here one entry, what means, there are all over two heights assigned to one point. A third heightfield now holds the heights, in the typical format, in the order of second heightfield. So, if there is a point, with four heights assigned, in the third heightfield, these four points are followed directly.
Such a construction would allow use -terrain generator developers- to create new algorithm, for more realistic and more detailrich terrains.
I agree, the rivers are still a problem in heightfield generation. But be sure, we, the developers of terrain generators, are working on this problem.