You run a regression on the value of all the improvements, then subtract that from the sale price to get the value of the unimproved land. Smooth that over the per-square-foot value of neighboring properties and you can have a pretty good estimate of for the unimproved value of the land.

If you have a hundred thousand properties to collect data on, you can get a pretty good idea of "a 3BR unit increases the value of the property this much, a 2BR unit increases it by this much, X feet in height result in this change in value, per parking space the value increases by this much". Sophisticated real-estate investors and the banks that fund them are running this calculation anyway, that's how they know which properties are good deals. Once you have a formula for this, you plug in the stats of the property in question to figure out the value of the improvements, and then the unimproved value of the land is everything left over.