I think Lucas brings it all together well.  You need to find the stable high and stable low areas along with knowing the unstable (less predictable) areas as well.  Then with your other data layers you can begin to verify what might be contributing to those areas.

It seems that areas more blessed with corn production ability seem to always produce better than other areas even if the high yielding areas are showing some mining effects.  The biggest question is how much better can they do if nothing is limited and can that be paid for by cutting inputs from the poorer areas that never make field average.

A comment to the composite subsoil nitrogen credits.  Here in the Central Platte Natural Rescource District (a local natural resource taxing government entity) there has been great success in utilizing composite DEEP nitrate samples to really cut back on N rates, to protect ground water, and not have yield suffer.  Granted much of that success comes from excessive use of N prior to this which has contributed to high test values of ground water.  Now that N management is played a lot closer to the cuff a new challange comes to play with variable rate N.  N is a pretty unstable beast and I think if any variable rate success exists it will be during the side dress window of application.  Anything prior to that would be base rate stuff and then only in the spring.

If you can find a surrogote data layer to tie ground truthed OM I think that would be better than using a grid to establish OM data.  I think it is too complicated out there to just use a corse grid for N management.......yet as I say that the corse composite sampling for residual N has worked very well BUT again probably because of the over use of N way above thresholds anyway.  Now that we have been doing this for many years we tend to stub our toe more often when we make an application mistake as evidenced by aerial photos.  It use to be we would never see that because we over applied N year after year.  Yield data is your best high resolution data set and other high resolution data sets (perhaps not invented yet) is what I think will be needed to make VRN success more common.  We have just started farming a small dry land piece of ground last year that is highly variable and I hope to learn a lot from that about this very subject.  Anytime my scout and I have driven to specific areas of relatively high and low yield areas of our more consistent irrigated fields we have not found any correlation of residual N and yield data on an annual basis.  After another year of yield data from this dry land field I hope to learn a lot more and hopefully join the VRN club of success.

Arnie