I (and I suppose many) are faced with setting parameters that are a compromise between dense and sparse scenery. I can get fluid, detailed (“ultra” performance @30fps) over small cities and most all terrain. The same settings do not work over large cities and the fps drop.
How about an adaptive system that learns over time, and is specific to each user’s machine. It would dynamically adjust settings based on measured framerate, instant to instant.
When fps drops, settings are adjusted downwards (subtly, slowly, perhaps based on the approaching scenery) so as to meet a (user set) framerate at the cost of less resolution, detail, etc. They go back up when the fps exceeds the set point.
As if Asobo did not have enough tasks 
. This could be given to an intern for research into machine learning as a long-term investigation. I know a little about ML and think it could be done! In fact, if SimConnect had the appropriate information/controls I would play with the idea myself.