These bands are going to be separate, not a single channel shared by both. You can already find satellite cellular frequencies in current 3GPP releases. (Every other assignment will be, at this point, legally and financially so hard that I'd claim it's economically infeasible.)
So, you luckily don't have to model a "mixed" terrestrial / space channel, but can keep them separate. That's probably very advantageous, because of the near/far problem: even good satellite reception would be very easily outshadowed by terrestrial cells, even those relatively far away on the ground.
The satellite has a LOS link with the user,
That's the ideal case. But it's not always true: an indoor user might see two paths, say one reflection coming in through the door and one coming in through a window.
Out-in-the-wilds users might have good LOS, but even there, they might be subject to multipath. For comparison, one can use DVB-S satellite emissions to do bistatic radar – because mountains and ships are large enough to contribute to a significant second path for wideband signals.
How should this joint channel look like? The model should contain 3 different channels, right?
Yes, that would be my prognosis. At this point, it's honestly educated guesswork, because what 6G is and isn't going to be is still in the making. I find "Was ist 6G?" (German, What is 6G?; the authors have added English subtitles) refreshingly honest. But, combining terrestrial and satellite signals in the same band always comes with dynamic range problems; thus, at least for the sat->UE downlink, assuming separate bands is a relatively sane approach. (And, as said, there's even 5G-specified satellite bands already, so this isn't really some speculation out of – pardon my pun – thin air.)
On the LOS/NLOS aspects, that really comes down to what you want your channel to be able to do. Is it low-rate/coordinative stuff? You might get by with a narrow enough channel (or a set thereof!) that behaves sufficiently much like LOS (in terms of amplitude statistics and phase stability). If you want your satellite to also be reachable in a dense urban scenario and do wideband signals, then considering the link LOS will probably not cut it. But: that's a question of which use cases you define for that link. The second paper you cite actually does address the different usage scenarios/potential benefits quite nicely. Before hoping to nail down your channel type, you'll have to consider how much bandwidth you'll actually use, where, and in which direction. That's not per se defined – and thus, as a researcher, and that might be an upside, you get to choose an interesting problem and work on the channel model that implies.
