Documentation says, something more specific should be used. Currently making a mountain path more precise. It passes several terrains: wood, grass, scree, bare rock. Not wanting to split the path into sections, will surface ground use the terrain as value?
What do you mean with “will surface ground use the terrain as value”?
Do you expect that data consumers will split the way at the boundaries of natural=* areas to convert e.g. natural=grassland to surface=grass for the path? I don’t think that any data consumer does that and I see no need for that.
In my eyes, surface=ground means something like “the way is not maintained, it only exists because animals or people use it”. In German I would call it Trampelpfad. Often the surface changes every few meters and in that case I would not split the way. If there is a long stretch (say +100m) with a more specific surface then split.
The idea came, German speaking here, as JOSM calls it “Erde”, which I’d reverse translate to “soil” (https://en.wikipedia.org/wiki/Soil). “Erde” has a notion of organic material, perhaps though only, because one usually does not see the lower horizons. “Ground” on the other hand to me does not have this connotation, thats why I am asking on the international forum.
I rather not split due to surface alone. Will stick to ground as least common denominator then, where there is a mix; as a positive way to say unpaved. Human consumers still allowed to make their guesses 
PS: Seems, from the WP article, in ancient Greek, ground and earth are connected, both are “the fundamental stone” (bedrock?)
I would recommend splitting.
even surface=unpaved or surface=ground would be incorrect here (unless by “wood” you mean “tree roots”)