...which is already subtracted out along with the rest of the improvements' contribution to land values. You could argue that it ain't a perfectly accurate subtraction - and that'd be a valid and reasonable argument - but that margin of error cuts both ways (i.e. overestimating improvement value instead of underestimating it).
> Lower cost is not necessarily lower value.
Land cost is land rental value times some amount of time (usually based on the expected cap rate).
> There are, for instance, empty parcels of land in Idaho that now go for more than whole apartment buildings in Detroit
Are the two parcels the same size? Are they representative of surrounding parcels and not outliers? Are they the same distance (in transit time/cost) as the nearest town/city center?
