In #635 i described a technique that can be used to conservatively transform data from one depth grid to another. The idea is simply to multiply the concentrations by the original cell depth, then transform, and finally divide by the new cell depth.
I think this would be a nice example to include in the transform docs.