Parcel types

A parceltype identifies types of parcels included in sales data samples.

Vacant land

We train our models on sales of vacant and “mostly” vacant properties.

This is by design: PLACES-FMV (CONUS) was designed to support conservation planning, and most land conservation transactions seek to protect land without buildings.

Excluding sales with buildings also avoids the challenge of accounting for heterogeneity in building value. Such heterogeneity can be substantial, and it is often shaped by attributes that we cannot observe consistently for the entire contiguous United States.

Vacant parcels

Vacant parcels (v) are parcels which appear to have no buildings at the time of sale.

Vacant parcels need to pass the following filters:

• No detected building footprint p_bld_fp

• No recorded positive building assessed value (val_b_za) in linked tax assessor data.

• No recorded building market value (mv_b_za) in linked tax assessor data.

• No standardized property land use code (bld_code) that indicates the presence of a building (e.g. “single-family home”)

“Mostly” vacant parcels

“Mostly” vacant parcels (mv) are parcels that are either fully vacant or that only have buildings of small size and/or value.

“Mostly” vacant parcels need to pass the following filters:

• Detected building footprint (p_bld_fp) covers <0.1% of the parcel area.

• Recorded building assessed value (val_b_za) is ≤0.1% of the total assessed value (val_t_za)

• Recorded building market value (mv_b_za) is ≤0.1% of the total assessed value (mv_t_za)

In addition, samples of “mostly” vacant parcels need to contain ≥50% fully vacant parcels (min_frac_vacant).

Urban-rural gradients

Proximity to urban centers and resorts is a major driver of land value in the United States.

Conservation goals often differ between urban and remote locations. For organizations focused on urban parks or remote wildlife conservation, it is useful to know the drivers of land prices and associated uncertainties in their landscapes of interest.

We use our “population gravity” raster (bld_pop_exp_c4) to distinguish urban, exurban, and rural settings. Our cutoffs are: 0.09 to separate urban from exurban areas, and 0.008 to separate exurban from rural areas.

We fit separate models for each setting, as well as their combinations: urbex (urban + exurban) and exrur (exurban + rural).

General parcel type filters

• All samples exclude properties smaller than one acre.

• All samples exclude properties with known conservation easements (p_e > 20%, to avoid including encumbered purchases).

• All samples exclude properties that were part of a publicly financed land acquisition included in the validation data (ct_p > 20%, to avoid including validation data in the training sample).

• Some model samples apply custom filters to predictor data, e.g. to fit models specific to floodplains, wetlands, coastlines, forests, croplands, and pasture (see table below for details).

Parcel type specifications

	label	buildings	min_frac_vacant	bld_pop_class	query
v	vacant	vacant
v-rural	vacant, rural	vacant		rural
v-exrur	vacant, exurban/rural	vacant		exurban,rural
v-exurban	vacant, exurban	vacant		exurban
v-urbex	vacant, urban/exurban	vacant		urban,exurban
v-urban	vacant, urban	vacant		urban
v-d	vacant (include hexagon dummies)	vacant
mv	mostly vacant	mostly-vacant	0.5
mv-all	mostly vacant, no balancing	mostly-vacant
mv-rural	mostly vacant, rural	mostly-vacant	0.5	rural
mv-exrur	mostly vacant, exurban/rural	mostly-vacant	0.5	exurban,rural
mv-exurban	mostly vacant, exurban	mostly-vacant	0.5	exurban
mv-urbex	mostly vacant, urban/exurban	mostly-vacant	0.5	urban,exurban
mv-urban	mostly vacant, urban	mostly-vacant	0.5	urban
v-floodplain	vacant, floodplain	vacant			fld_fr_fath_f100 > 0
mv-floodplain	mostly vacant, floodplain	mostly-vacant	0.5		fld_fr_fath_f100 > 0
v-coast	vacant, coast	vacant			cst_2500 > 0
mv-coast	mostly vacant, coast	mostly-vacant	0.5		cst_2500 > 0
v-crops	vacant, cropland	vacant			p_crops > 20
mv-crops	mostly vacant, cropland	mostly-vacant	0.5		p_crops > 20
v-forest	vacant, forest	vacant			p_forest > 20
mv-forest	mostly vacant, forest	mostly-vacant	0.5		p_forest > 20
v-wet	vacant, wetland	vacant			p_wet > 20
mv-wet	mostly vacant, wetland	mostly-vacant	0.5		p_wet > 20
v-pasture	vacant, pasture	vacant			p_pasture > 20
mv-pasture	mostly vacant, pasture	mostly-vacant	0.5		p_pasture > 20
v-barren	vacant, barren	vacant			p_barren > 20
mv-barren	mostly vacant, barren	mostly-vacant	0.5		p_barren > 20
v-grassland	vacant, grassland	vacant			p_grassland > 20
mv-grassland	mostly vacant, grassland	mostly-vacant	0.5		p_grassland > 20