I would have assumed that nowadays using a small neural net to approximate a CFD model of a flame would be the easiest.