Sure but youd have to wirte the entire thing yourself (in essence rewriteing entire fluid). Even so you cannot separate the sahdow form the emission of the cloud its halfway meaningless. Offcourse that depends on your definition fo the shadow.
See the problem si that theres
point1->point2->point3->point4....
in succession and each point can be in shadow or light however each point can slo be in shadow of each other point now if you just calculate the shadow contribution Now suppose youd collect oonly the info of shadow percentage of each point. And let us suppose poit2 and 4 are in shadow. Now you could collect point2shadow result + point4 shadow result. and youd end up loosing the light rays of point1 and point3 so the end result couldnt rebuild the same cloud info as the standalone since youve removed the local occlusion of the light in between. So in reality you can jsut have a meaningful lightness pass theat has BOTH light and shadow or ou end up with nothing.
Other methd would eb to save a very deep raster. Wich has both.
Im not saying its impossible im just saying its too interlocked to simply unravel with simple logic methods.
Its the same kind of thing as that while you can have shadow passes you rarely get the chance to adjust the shadow intensities for the objects reflections just the primary surfaces contribution. But gas has no primary surface and thus not a normal either.
Now a could has no normal, but let us suppose you caoudl sue some treshold as gradient, but again each point would have its own gradient, again could be of use in adeep raster but not a non deep one. Or you could suffice to say the first normal at treshold is enough.
And using deep rasters would be very cose to rerendering teh entire image its just a sort of sample cache.
see a cloud isnt a surface so theres no definitive answers for its surface it has a unlimitted amount fo themn along the visibility axis.