Method of generating believable random maps by multivariate analysis of known geospatial features including: Altitude, production, population anchors, and historical political boundaries

By analysis of historical maps we can use image processing to identify areas of high information density (well known) and low information density (poorly known).  First order business to take care of is non-sense and here lies beasties type stuff that obviously signifies the cartographer knows fuck-all about that area.  Second order is to measure differences in segment geometry length and line complexity in a vectorized map element.

I need to find a dataset that can meet my requirements.  Initial first thoughts are open data initiatives in England.