Especially when 36 years of yearly National Land Cover Dataset data comes out, I can only imagine the possibilities of machine learning and image direction extraction to estimate likely directions of sprawl. While certainly things like town policies, slope and zoning play a roll, I bet you could learn a lot just by looking at land use change over the years and extrapolating out growth in the direction of existing growth. I’m sure there is a way that can be done either with image processing or GIS packages in R.