If you search "bayesian skyline plot", you will find its original paper. Briefly, it is a method to infer the historical effective population size given multiple haploid genomes without recombinations. Naively, it gets the size information from the density of branches in a coalescent tree. The Bayesian version does not rely on a single tree. Bayesian skyline plot is a very good method. IIRC, it is implemented in BEAST.