… is still an open issue and a fundamental “gotcha” of the interaction between visible 3D geometry and the physics engine (calculating aircraft damage)
There is a lot of background info in the archived thread.
The problem occurs when the aircraft is still parked on the ground
… the drone moves far away
… the 3 scenery around the aircraft gets removed from the GPU
… because the drone is looking at a far away place
… and then the aircraft “crashes” (or collides with an object, etc) while parked
The “crash” in the last step depends on how the 3D scenary (LOD) gets reduced at the aircraft location.
Back home, on Hawaii, I also managed to crash into the ocean, while “parked” on the airport.
… but something seems to have changed “a little” … perhaps here is more “caching” of 3D landscape /elevation ? I need to check again at some of my “crash-while-parked” test sites.