Nice work Thorsten! I'd really love to know how you did this one.
Hint - think of a situation in which the clouds light up for a moment at a certain location. It virtually took no additional effort, all the technology was there.
So does a light now indeed illuminate its environment (as in ground, buildings, clouds, other aircrafts)?
I suppose we need to get this out of the way.
In real-time 3d rendering, there are no lights which illuminate anything really. There's no light source in the simulation from which rays are shot, interact with the environment and then hit something in the scene as in raytracing.
Try to put a Rembrandt light into an enclosed box - it's going to shine right through the box, because it really isn't a light. Try to put it into heavy fog, and it's neither going to illuminate the fog, not going to be attenuated - because it's not really a light.
All there is is a shader responsible for a triangle (which shader this is is driven by the effect), and that shader has access to the local surface properties plus whatever you tell it about the more global situation. From the perspective of the shader, our Sun is nothing but a uniform struct containing position and color information which shaders can access (or not). But what counts in the end is whatever color operations the shader decides to do.
Here's no localized light source in the scene at all - the shader is instructed to treat light like a liquid, it fills the low places and slowly creeps higher as the day progresses:
Looking at such a picture, you can dispense with the fiction that there really are light sources in rendering.
It's all a bag of illusions. It's all different ways of handing global lighting information to the shader.
Rembrandt does this by placing artificial geometry into the scene and utilizing buffers. Which makes it deceptively easy for model designers to use it and pretend to actually put a light into the scene, because by and large it will be taken care of automagically. Of course this convenience for the designer costs the user in terms of framerate, because an additional pass over the scene is an additional pass, you touch every vertex twice.
But for many situations, there are other (much cheaper and better) ways to communicate global lighting and shadow information. Take a cockpit interior which is largely static in-sim lit by a ceiling light. You can compute this in Rembrandt every frame, and the result will be the same every frame (because it's static) and there won't ever be any shadows or reflected lightspots (because it's not a real light but just a quick trick).
You can also compute this situation offline with a raytracer and simply use the result in-sim (can take as long as it takes), and dump it to a table. You can have a high-order of reflections that way, objects shadow parts of the light, you can have the resulting distribution of the ambience,... and simply read the table in-sim - because it's a static problem. Then you use a lightmap and your cockpit lighting even casts the right shadows and reflections. You can even let the whole lit panel appear in the canopy glass (a raytracer does that easily) - at the simple expense of reading a texture, which is infinitely cheaper than a Rembrandt lighting pass.
So it's not a light you're seeing - it's a mixed bag of situation-specific tricks. Not as convenient for the aircraft designer, but usually more framerate friendly and sometimes even superior in visuals.