Space-efficient substring occurrence estimation