Actually that’s only for one of the rock layers used in the high altitudes. Every other layer used some kind of blend modulation which was using a method similar to height lerp. here are some examples from old testmaps:
In some cases it was just simple height for rocks to poke through:
And for other layers, a more macro pattern was used to control spawning patterns which also controlled spawning of different kinds of meshes using the procedural system. In that way, when the foliage was culled out, the underlying terrain would look close enough so you could barely tell. In this image, all of the grass and heather stops rendering about halfway past the image. You could consider this procedural similar to what you are talking about.
I noticed that in your image using the layer blend, you only used 1 heightmap (that is visible in the image) whereas you used scalars for the other layers. You should be able to use different heightmaps for all of the layers to do what you want.
The only reason to use the Layer Samples vs Layer blend is just because you have more control over the layer that way, but doing so also requires using “No blend” layers instead of the default method of blended layers. In the kite demo example, the material could also make layers render where there was no painted alpha, entirely based on large tiling macro textures and share that for foliage placement. You can’t really do things like that with the layer blend method but of course you should do whatever fits your particular workflow best.
This comparison from siggraph slides shows the effect of the modulation. Without it, the blends would have looked blocky due to the resolution being only 1 meter per pixel.
Notice that the pink heather isn’t always matching up with the splat maps, because its largely controlled by tiling macro textures. The weight for it only very generally describes the shape.
