A research team from the University of Massachusetts Amherst, U.S., has proposed a new, data-driven approach that uses machine learning techniques and widely available satellite images to identify roofs that have the maximum potential to produce cost-effective solar power.