I built a full Earth scene graph for a virtual world and military training purposes a while back.
It is very hard, and it is a LOT of data. The earth is HUGE. Also, the graphics cards and physics engines for games typically use 32-bit floating point values, which have an error of about 1.5 meters at the surface of the earth if you stay with the origin as the center of the earth. Thus, you need to translate all your data to be camera-centric or close-to-camera-centric, and keep all the actual (authoritative) data in 64-bit floating point.
If you want gameplay on Earth, you’re faced with at least two problems:
- It’s a lot of data to model. Even if you import it from satellite imagery (a la Google Earth,) it’s really messy data, and you won’t be getting good vegetation and buildings from a ground perspective. You’ll see shadows painted straight into the ground texture, etc. It’s a mess.
- It would take a normal person many thousands of years to walk/travel to all places on Earth. Meanwhile, you may be aiming for a hundred hours of gameplay.