Andrew - It sounds like your goal is to reproduce the local shape of your surface over a region that lies inside the bounds of your data. Gridfit can do this, if you are willing to expand that grid so that it fully contains the data. Simply add some nodes that extend to the boundaries of your data in both x and y. The nodes used by gridfit need not be equally spaced, so this should not be a problem, but to retain the essential shape of the surface, you may need to add a few such nodes in each direction. Once the fit is obtained, then you can always extract only that portion of the surface that you wish to retain.
If this is not what you are looking for, then feel free to contact me directly, perhaps with some data so that I can better understand your problem.
This is an amazing program. I do have one question. I have a scattered grid of latitude(Y) and longitude(X) coordinates corresponding to brightness(Z) values. I've been using GRIDDATA to interpolate to a new grid that is not scattered and within a smaller range of latitude and longitude values, but GRIDDATA smooths the data so much that it becomes useless.
I understand that GRIDFIT is not an interpolant, and from my reading of the help, it does not seem that it was meant to be used in this fashion, but I just want to make sure.