Cloud coloring: Actually the color of light reaching you changes quite rapidly when increasing altitude at sunset. Take of from an airport and observe the sun color and you'll see what I mean. So the color change is not linear and only using 2 points for interpolation probably will not be correct.
Yes, I've worked that out on the way home yesterday. What should enter the color model is the length of the light through the atmosphere. We can't do that, we can do it for the local position of the airplane, but not for every pixel in the scenery.
So it has to be an effective shadow 'wedge', the projection of which on the ground is the terminator, and where you are in this 3d shadow wedge determines (aproximately) light intensity *and* color. Inside this wedge, the effective forward shift in shadowing conditions goes like sqrt(2 * R * altitude) with altitude where R is the Earth radius - if you plot that, you note that indeed that is a rapidly growing function initially.
1) The earth is not a perfect sphere (ellipsoid, thicker at the equator), so the path distance of the light rays will differ depending on latitude.
2) The thickness of the atmosphere varies. It is thickest around the equator and thinnest around the poles. Since the difference is up to 10 or more kilometers this will have an impact on scattering. So both of these effects should have an influence on the amount of scattering depending on the viewers latitude.
I suspect the variation of lighting with aerosols and water vapour is *much* more significant than that, so unless we start getting control over those, there's no need to delve into this.
The problem is, quite simply stated, that you need to compute a different color for every point in the scene - not just for the sun, or for the airplane position - for a suborbital view for an area of ~200.000 km^2 actual scenery and 3.200.000 km^2 skydome posing as scenery and all in all ~2.000.000 pixels with 20+ fps. So whatever scheme you have better be *really, really* fast, and that means approximations to geometry and plausible-looking schemes instead of actual calculations.