While playing the game (ok I admit I'm a little hooked now) I had another idea. It looks like the maps are seed generated chaos, with a half dozen having been hand selected, named, and set as presets?
If so I assume an ~infinite number of potential maps already exist within the system. Amongst them is "Europe" - waiting to be discovered.
So train the routine to recognise how close to Europe each random seed is, e.g. giving a score based on pixel matches to the reference image. Then let it run on loop for a week, and turn over a few quadrillion tries. Keep a record of the top 100 scoring maps. Manually review them at the end of the week and select a winner. Use that seed as "Europe". Total additional filesize: a few bytes. Why not look for other famous maps at the same time - e.g. The World, Westeros, etc ...