Well everything is tunable. The condition for snow is currently this: at altitude 4500m snow holds on anything below slope 0.9 (in range 0..1, flat to steepest), and at the altitude of 3000m on anything flatter than 0.0 - that means it ceases to appear even on flat surfaces (this is simplified a bit, there's also a factor of terrain curvature). Anything in between uses an interpolated value of critical slope.
Now this sounds like a made up rule, and it is. It's not that far off, actually. The ability of snow holding on to the slopes relates to the temperature, which falls with rising altitude.
Currently it's just a hack because it doesn't take into account the actual climate - average temperature and precipitation - and the season. But even if we just computed the approximate surface temperature using latitude, day of year and altitude, the result would look rather good - you'd get permanent snow in arctic areas, the snow line would be lower in mountains that are closer to the poles, and it will change with the season because sun rays fall more vertically. Actually, a very easy thing to do.
The final solution has to include average precipitation data, though. For example the Tibetan plateau will have less snow then, even though the temperature is low due to the altitude.
A weather engine would be probably only feeding the information if it is snowing right now; whether the snow stays on the ground is another thing, and that can be driven by the above mentioned model, maybe augmented by some snow level reports.
