Presently, the code that interpolates from one resolution to another has a bit of a hack when it comes to interpolating the balance coefficients. In my view, I am missing the adjoint of the spline routines being used to perform the interpolation. The consequence of the present formulation is that it doesn't do too well when interpolating from low to high vertical levels - the same would be case horizontally, but typically, I do not employ interpolation horizontally.