The UI is a very leaky abstraction on the underlying models. Everything on the user interface feeds into the internal models in some way or another with very few hidden parameters. Things that are missing (like orbital parameters such as eccentricity, variable albedo for biomes, air-sea heat transfer ratios, and a host of other things) aren't accounted for in the models at all. A big hidden parameter for annual average temperature is that it's only calculated at four points on the orbit (solstices and equinoxes) and there isn't a way to get more samples to allow for slightly better calculations.

The major limitation that I couldn't overcome in the system is that the internal representation of user editing sample is straight equirectangular, meaning that individual samples are non-uniform in size and shape over the world. Each editing layer is (relatively) huge and each season would require at least its own wind pattern edit buffer (likely temperature and rainfall for best effects). Add on to that that users would need to get access to those individual buffers for seasonal editing and I'm not sure that I could get a useful (from the UI perspective) and meaningful (to users who won't read the documentation) result. In a way, it's a bit sad because these days I know very well how to do distance fields and curl noise without too much computational expense, what I don't know how to do is to describe those things in a way that wouldn't be too complicated to use.