I also think you are inferring you that you can use the subject property for exxtraction which you cannot. The reason is the same for an income property; you don't derive the GRM from the subject by dividing the sale price by the rent. It is an incestuous conclusion.
IF your estimate of contributory value of the improvements for the subject property is accurate (we can cover the chances of that in another thread as well, but hang with me for a moment), and IF you have accurately determined the market value of the subject property, then the difference between the two HAS to be an accurate assessment of the subject's site value, does it not? It's simple logic. If A=market value, and B=depreciated value of improvements, and C=estimate of site value, then it necessarily follows that A-B=C. Again, and as I've said MULTIPLE times, the validity of the extraction results is obviously strengthened by the use of multiple properties, but that does not negate the fact that A-B=C.