It was applied at the enter point of the walking algorithm, which
was the camera when it was inside a cloud layer.
Now it is applied at the first found segment, which is still not
optimal but better.
The bruneton model was also fixed to not produce black results for
aerial perspective exactly at the camera location.
