i have an idea about geomipterrain.
currently the geomipterrain has 1:1 ratio of heightfield pixel and vertices and LOD 0, and each 4 pixels form a quad which is divided into 2 triangles.
when the height/brightness of the 4 vertices are far away from coplanar, for example 0 0 0 1, the quad has obvious division line.
if 1 more vertex, of brightness 0.5 is added in the middle of the quad, and the quad is divided into 4 triangles, the quad will look smoother. and in this way the heightfield image need not be enlarged.
so in the expense of 1 vertex and 2 triangles in a quad, the terrain looks smoother where it’s most uneven.
on the other hand, where the terrain is very even in the area of more than 1 quad, these quads can be combined without much loss of detail. that is, if per-pixel lighting is used on the terrain, or no directional lighting.
these can be done using geom editing after the geomipterrain has been generated. but that’s effected only before the terrain is updated, or bruteforce is on.
if the geomipterrain class has these optimization inside, that’s even better.
i don’t know whether there are existing similar terrain engines already.