Well of course it's the rendering of hills, vallys, mountians ect. but what makes it better than just creating a polygon model of a mountan and rendering it normaly? maybe speed (but if broken up correctly some culling would help speed matters) so, what makes it worth it?

I can think of a few reasons:

1) you can make assumptions about terrain that you can't make about poly soups

2) source data for terrain is usually regular gridded data

3) storing gridded data instead of meshes usually takes less disk space

4) horizon-scanning mechanisms for terrain are easier to debug than BSP cutters for meshes of hundreds of thousands of polys