Sure, I could talk about a method I came up with for unbiased prediction of a large number of vegetation classes of very different prevalences. Used for mapping vegetation in PNW national parks, done in R/random forests. Project also includes a semi-innovative way of improving tractability of giant prediction tasks (3-meter resolution pixels across huge national parks) using standard desktop computers.
Eric
---------------------------------------------------------------------
Eric Nielsen, geospatial scientist/mapmaker
Institute for Natural Resources, Portland State University