You're not wrong, but the major issue is actually that most PCG terrain is based on a height map(s).
In it's simplest form the terrain can be thought of as a 2D image, where the height of each pixel is represented by a colour
(say black is lowest point and white is highest point). Or for example you give the PCG algorithm a location and it returns a height.
This works fine and can produce very detailed/interesting terrains but it has a major limitation: You cannot represent overhangs or caves. As a single
location is limited to a single height.
In it's simplest form the terrain can be thought of as a 2D image, where the height of each pixel is represented by a colour
(say black is lowest point and white is highest point). Or for example you give the PCG algorithm a location and it returns a height.
This works fine and can produce very detailed/interesting terrains but it has a major limitation: You cannot represent overhangs or caves. As a single
location is limited to a single height.