It's probably something like, here are the environments where we've done comprehensive surveys, here are the kind of different situations where we expect to find different species (decomposers of various types, mycorrhizal, within plants, within animals, on surfaces, specialists, generalists, climates, etc). Multiply the species from places where we've probably found most of them by the number of places where we've only found the most obvious fungi. However it works it's going to have big error bars, reflected in the fact that 12M species is the upper end of a range starting at 2.2M.