What is Parcel Data? Parcel Numbers, Boundaries, Sources

Parcel data is the map of land ownership: one polygon for every piece of land, with the records that describe it. Each parcel carries a boundary, parcel number, owner, land use code, zoning district, area, and assessed value. Counties maintain parcel data to tax property and record who owns it. Insurers, utilities, lenders, farmers, and AI agents use it to tie an address, a building, or a pixel to a legal unit of land. The claim tested here: parcel data fills the fields that name a parcel and its owner, and leaves many others empty. Before a join, count filled fields, match buildings to parcels by center point, and merge records on one polygon.

Key takeaways

• Parcel data pairs a boundary polygon with property records: parcel number, owner, land use, zoning, area, and value.
• The parcel number, often called the assessor's parcel number (APN), identifies a parcel within its county.
• Counties create parcel data. State portals and nationwide providers map it to one schema.
• Parcel boundaries come from tax maps. A licensed survey sets the legal line.

What parcel data contains

Parcel data has two parts: a polygon that outlines the parcel, and the columns of the assessor's record.

Attribute group	Typical fields
Identifiers	Parcel number (APN), alternate IDs, county FIPS (Federal Information Processing Standards) code
Ownership	Owner name, mailing address, owner type
Location	Situs (property) address, city, ZIP code, centroid coordinates
Land use and zoning	County use code, zoning district, standardized land use
Area	Recorded acreage, and acreage and square feet calculated by a geographic information system (GIS)
Value and tax	Land value, improvement value, total assessed value, tax amount
Sales	Sale price, sale date
Structures	Year built, number of stories, units, building count
Legal description	Plat, book, page, block, lot, subdivision, Public Land Survey System (PLSS) township, range, and section

Land use records what a parcel is used for today. Zoning records what the city or county allows on it. Counties write their own use codes, so nationwide datasets add a standard. Regrid maps local codes to the American Planning Association's Land-Based Classification Standards (LBCS), which classify each parcel along five dimensions: activity, function, structure, site, and ownership. Regrid's LBCS columns carry a 4-digit code and a description for each dimension.

Regrid maps over 100 county-provided columns into one standard schema and keeps each county's extra columns.

What is a parcel number?

A parcel number is the identifier a local taxing authority assigns to each parcel. Counties that use the term assessor's parcel number abbreviate it APN. Other jurisdictions say parcel ID, property identification number (PIN), assessor's identification number (AIN), or tax ID. The number is unique within its jurisdiction and can repeat across jurisdictions. Nationwide datasets pair it with a county code or a provider key.

The format encodes how the county maps land. San Francisco combines a block and a lot: parcel 3590-035 is lot 035 in block 3590. New York State builds its parcel numbers from section, block, and lot. Its tax bills call the identifier the section, block, and lot (SBL) number. The SanGIS parcel metadata for San Diego County keys each polygon to an APN and to a separate polygon identifier, PARCELID.

A parcel is the unit the assessor taxes. A lot is a unit on a recorded plat, and one parcel can contain several lots.

Legal descriptions in metes and bounds, PLSS, and lot and block

The polygon in a parcel layer is a drawing. The legal boundary is the description in the deed, written as metes and bounds, PLSS, or lot and block.

Metes and bounds describes a boundary as a walk. Start at a point of beginning, follow a bearing (a compass direction) for a distance to the next corner, and close back at the start. Each step, or call, can name a monument such as a stone or a creek, as the Bureau of Land Management's metes-and-bounds study guide describes. It is the main system in the original 13 colonies and Texas, which lie outside the public land survey.

The Public Land Survey System (PLSS) describes land by its place in a national grid. The Land Ordinance of 1785, adopted by the Continental Congress on 20 May 1785, ordered the western territory surveyed into townships 6 miles square. Each township was split into 36 sections of 1 square mile, or 640 acres. A PLSS description reads like an address in that grid: the northeast quarter of section 12, township 3 south, range 2 west, from a named principal meridian. The Bureau of Land Management (BLM) publishes the grid as the PLSS CadNSDI dataset.

Lot and block describes parcels in recorded subdivisions by their plat: lot 7, block 3 of a named tract, filed in a plat book at the county recorder.

Cadastre vs parcel fabric vs parcel layer

A cadastre is the official land record of a jurisdiction. The International Federation of Surveyors (FIG) published a Statement on the Cadastre in 1995. It defines a cadastre as "normally a parcel based, and up-to-date land information system containing a record of interests in land (e.g. rights, restrictions and responsibilities)". The statement adds that a cadastre usually links a geometric description of each parcel to records of ownership and value.

A parcel fabric is a data model that stores parcels as a connected network of points, lines, and polygons. Each line carries the recorded bearings and distances from the source document. Esri's parcel fabric organizes parcels by the record that created them and enforces topology rules, which keep shared edges free of gaps and overlaps. A change to one boundary moves the neighboring parcels with it.

A parcel layer is the polygon dataset counties and providers distribute: one feature per parcel with attributes, and no stored relationship between neighbors.

Where parcel data comes from

Parcel data starts at the county assessor, recorder, or GIS office, or in New England at the city or town. Each office keeps its own parcel map, often linked to a computer-assisted mass appraisal (CAMA) system that stores the values.

County portals, state portals, and nationwide providers republish those maps:

1. County and city portals. Many counties publish a parcel viewer and downloads. San Francisco publishes recorded parcel geography back to 1995 under a public domain license. San Diego County's parcels come from SanGIS, the regional GIS source for the City and County of San Diego.
2. State portals. States such as North Carolina (NC OneMap), Utah (the Utah Geospatial Resource Center, UGRC), Maryland, and Massachusetts (MassGIS) combine county parcels into statewide layers.
3. Nationwide providers. Regrid, LightBox, and ReportAll compile county parcels into standardized national datasets. Esri offers Regrid parcels in ArcGIS: over 158 million parcel records across the United States and Canada, sourced from all 3,143 US counties.

Parcel records are public in many US jurisdictions. Terms vary by county: some charge for bulk files, and some withhold owner names or sale prices. Each county updates on its own schedule. Regrid reports that on average 94% of its parcels were refreshed in the last 12 months.

Parcel boundaries and property lines

A parcel boundary is the line on the tax map. A property line is the legal boundary in the deed. The two often differ by a few feet, because tax maps are compiled from deeds, plats, and older maps for assessment.

Parcel lines also shift against aerial imagery and can cross roofs or fall into streets. Regrid leaves parcels unaligned to imagery and sends large shifts back to the county. For a fence, a dispute, or a building permit, a licensed land surveyor sets the legal line.

Parcel data formats and coordinate systems

Counties publish parcels as Shapefile, File Geodatabase, GeoJSON, or CSV with a geometry column. Shapefile truncates column names to 10 characters, which breaks long schema names such as parcelnumb_no_formatting.

County data usually arrives in a local projection, such as a State Plane coordinate reference system in feet. Nationwide datasets reproject everything to WGS 84 (EPSG:4326) longitude and latitude so counties line up.

Parcel data quality

Topology errors, slivers, update lag, and mixed county schemas each change the result of a join.

• Topology gaps and overlaps. Neighboring parcels should share edges with no gap and no overlap. Tax maps digitized from paper leave gaps and overlapping strips. Condominiums add stacked parcels: several records on one footprint.
• Slivers. A sliver is a thin polygon with almost no area. Michael Goodchild studied slivers in "Statistical aspects of the polygon overlay problem" (1978, Harvard Papers on Geographic Information Systems). He showed that overlaying two versions of the same line produces spurious slivers in numbers that grow with line complexity. Delafontaine et al. (2009) studied how to detect and assess slivers, noting that some thin polygons, such as road and canal parcels, are real.
• Update lag. A split reaches the parcel layer after the deed is recorded and the map redrawn, and a nationwide copy lags again by its refresh cycle.
• Mixed county schemas. More than 3,000 US counties choose their own use codes, value fields, and owner formats, and some withhold zoning or sale prices. Nolte et al. (2024), in Land Economics, list the resulting problems: identifying arm's-length sales, geolocating parcels and buildings, telling property types apart, and handling missing housing attributes.

Pullar and Donaldson (2022) found Queensland cadastral boundaries 3 to 5 m off fence lines in places on 0.1 m orthoimagery, aerial photos corrected to map geometry. Fetai et al. (2022) compared surveyed and digitized boundaries in North Macedonia and Slovenia. The root mean square error (RMSE) of position was 48 cm and 56 cm in two study areas.

Parcel data examples and uses

Teams use parcel data to answer "what piece of land is this, and who owns it?":

• Real estate and lending. Appraisers and lenders pull sales, assessed values, and zoning for comparable properties.
• Insurance underwriting. Underwriters attach property value, land use, and building counts to each policy location, then overlay flood and wildfire layers for catastrophe modeling and probable maximum loss.
• Utilities. Utilities match service addresses to parcels and contact owners along a line before vegetation work.
• Agriculture. Farm parcels link ownership and value to field boundaries, crop maps, and land cover classification.
• Site selection. Developers screen parcels by size, zoning, slope, and distance to the grid for solar farms and data center sites.

In package tracking, "parcel data received" means the carrier has the label's electronic record and has not scanned the box yet, as Ship24's tracking help explains.

Parcel data in Wherobots

A Regrid parcel sample is ready to query in Wherobots open data. It holds 7,935,620 parcels in Los Angeles, Orange, Riverside, San Bernardino, and San Diego counties in California and Maricopa County in Arizona.

The walkthrough below uses one box over Escondido, in northern San Diego County: longitude -117.11 to -117.04 and latitude 33.095 to 33.15, about 6.5 by 6.1 km. A parcel belongs to the box when its centroid, the center point of its polygon, falls inside it.

The walkthrough reads two Apache Iceberg tables with a geometry column:

Source	Table	Size	What the queries read
Regrid parcels	wherobots_open_data.partner_samples.regrid_parcels	7,935,620 parcel polygons; 139 columns, including parcel number, owner, use code, zoning, values, sale and legal description fields, the five LBCS dimensions, a building count, the majority Cropland Data Layer class, flood zone from the Federal Emergency Management Agency (FEMA), a refresh date, and geometry	23,275 San Diego County records with a centroid in the box: owner, use code, LBCS activity, zoning, assessed value, year built, sale date, legal description, refresh date, ll_bldg_count, and the polygon
Overture buildings	wherobots_open_data.overture_maps_foundation.buildings_building	Building polygons; full row count not counted here	Buildings whose bounding box overlaps a slightly larger box, and their centroids: 26,635 centroids fall in the box

Three WherobotsDB SQL queries ran in 14 to 18 seconds each: one on parcels alone, two joining parcels to buildings.

Profile the fields

The first query counts how many records fill each field, when they were refreshed, and how many are tiny or thin. It ran in 15.7 seconds.

WITH p AS (
  SELECT owner, usedesc, lbcs_activity_desc, zoning, parval, yearbuilt, saledate, legaldesc, ll_last_refresh,
         ST_AreaSpheroid(geometry) AS m2,
         ST_Perimeter(geometry, true) AS perim_m
  FROM wherobots_open_data.partner_samples.regrid_parcels
  WHERE county = 'san-diego'
    AND ST_Intersects(geometry, ST_PolygonFromEnvelope(-117.11, 33.095, -117.04, 33.15))
    AND ST_Contains(ST_PolygonFromEnvelope(-117.11, 33.095, -117.04, 33.15), ST_Centroid(geometry))
)
SELECT COUNT(*) AS parcels,
       COUNT(NULLIF(TRIM(owner), '')) AS with_owner,
       COUNT(NULLIF(TRIM(usedesc), '')) AS with_county_use,
       COUNT(NULLIF(TRIM(lbcs_activity_desc), '')) AS with_lbcs_activity,
       COUNT(NULLIF(TRIM(zoning), '')) AS with_zoning,
       COUNT_IF(parval > 0) AS with_value,
       COUNT_IF(yearbuilt > 0) AS with_year_built,
       COUNT(saledate) AS with_sale_date,
       COUNT(NULLIF(TRIM(legaldesc), '')) AS with_legal_desc,
       MIN(ll_last_refresh) AS oldest_refresh,
       MAX(ll_last_refresh) AS newest_refresh,
       ROUND(PERCENTILE_APPROX(m2, 0.5), 1) AS median_m2,
       ROUND(SUM(m2) / 1e6, 3) AS total_km2,
       COUNT_IF(m2 < 50) AS under_50_m2,
       COUNT_IF(m2 < 100 AND 4 * PI() * m2 / (perim_m * perim_m) < 0.1) AS slivers
FROM p

The box holds 23,275 parcel records. Owner names fill 23,267 of them, county use codes 23,274, legal descriptions 23,264, and Regrid's LBCS activity 22,870. A positive assessed value appears on 21,883. Zoning fills 1,747 records, 7.5%, and no record carries a year built or a sale date. Every record was last refreshed on 13 February 2024. The median parcel covers 985.2 m².

20 parcels are under 50 m². 8 are slivers: under 100 m², with a Polsby-Popper compactness score (4πA/P²) below 0.1. The score compares area with perimeter; a thin strip scores near 0. The parcel areas sum to 273.942 km², and the box covers 39.85 km² (ST_AreaSpheroid of the box envelope). Many records must share the same ground.

Land use and a first building join

The second query adds land use and joins Overture buildings to parcels. It matches each building footprint by its centroid to the parcel that contains it, with ST_Contains. A building that crosses a lot line intersects two parcels, so a polygon intersection join would count it twice. The query plan (EXPLAIN FORMATTED) shows an indexed join (BroadcastIndexJoin) with an R-tree spatial index on the building side, and no Cartesian product (every parcel paired with every building). The query ran in 14.1 seconds.

WITH parcels AS (
  SELECT ll_uuid, ll_bldg_count, geometry,
         COALESCE(NULLIF(TRIM(lbcs_activity_desc), ''), '(no LBCS activity)') AS activity,
         ST_AreaSpheroid(geometry) AS m2,
         ST_Contains(ST_PolygonFromEnvelope(-117.11, 33.095, -117.04, 33.15), ST_Centroid(geometry)) AS in_box
  FROM wherobots_open_data.partner_samples.regrid_parcels
  WHERE county = 'san-diego'
    AND ST_Intersects(geometry, ST_PolygonFromEnvelope(-117.11, 33.095, -117.04, 33.15))
),
buildings AS (
  SELECT id, ST_Centroid(geometry) AS centroid,
         ST_Contains(ST_PolygonFromEnvelope(-117.11, 33.095, -117.04, 33.15), ST_Centroid(geometry)) AS in_box
  FROM wherobots_open_data.overture_maps_foundation.buildings_building
  WHERE bbox.xmax >= -117.12 AND bbox.xmin <= -117.03
    AND bbox.ymax >= 33.085 AND bbox.ymin <= 33.16
),
pairs AS (
  SELECT p.ll_uuid, b.id
  FROM parcels p JOIN buildings b ON ST_Contains(p.geometry, b.centroid)
),
per_parcel AS (
  SELECT p.ll_uuid, p.activity, p.ll_bldg_count, p.m2, COUNT(pr.id) AS n
  FROM parcels p LEFT JOIN pairs pr ON p.ll_uuid = pr.ll_uuid
  WHERE p.in_box
  GROUP BY p.ll_uuid, p.activity, p.ll_bldg_count, p.m2
),
per_building AS (
  SELECT b.id, COUNT(pr.ll_uuid) AS n
  FROM buildings b LEFT JOIN pairs pr ON b.id = pr.id
  WHERE b.in_box
  GROUP BY b.id
)
SELECT COALESCE(activity, 'ALL PARCELS') AS grp,
       COUNT(*) AS total,
       COUNT_IF(n = 0) AS zero,
       COUNT_IF(n = 1) AS one,
       COUNT_IF(n >= 2) AS two_plus,
       SUM(n) AS buildings_matched,
       COUNT_IF(n = ll_bldg_count) AS agrees_with_ll_bldg_count,
       ROUND(PERCENTILE_APPROX(m2, 0.5), 1) AS median_m2,
       ROUND(MAX(m2) / 1e6, 3) AS max_km2
FROM per_parcel
GROUP BY ROLLUP(activity)
UNION ALL
SELECT 'OVERTURE BUILDINGS (parcels containing the centroid)', COUNT(*), COUNT_IF(n = 0), COUNT_IF(n = 1), COUNT_IF(n >= 2),
       SUM(n), NULL, NULL, NULL
FROM per_building
ORDER BY total DESC

Household use covers 20,819 of the 23,275 records, followed by Shopping, business, or trade (537), Shopping (370), industrial uses (310), and Office (156). Of the 405 records with no LBCS activity, 365 contain no building.

The building join explains the area total. Of 26,635 Overture buildings in the box, only 23 have a centroid outside every parcel, so the layer has almost no gaps under buildings. 23,093 fall in one parcel record and 3,519 in two or more. The 472,643 building-to-record matches come from parcel records stacked on one polygon. Regrid's own building count, ll_bldg_count, agrees with the Overture count on 13,079 records.

Deduplicate stacked parcels, then check topology

The third query merges records with an identical polygon into one footprint. It finds them by hashing each polygon's well-known binary (WKB) form with md5. It counts building centroids per footprint and joins footprints to each other to find overlaps. The plan shows two indexed joins (BroadcastIndexJoin and RangeJoin) and no Cartesian product. It ran in 17.7 seconds.

WITH p AS (
  SELECT ST_AsBinary(geometry) AS wkb
  FROM wherobots_open_data.partner_samples.regrid_parcels
  WHERE county = 'san-diego'
    AND ST_Intersects(geometry, ST_PolygonFromEnvelope(-117.11, 33.095, -117.04, 33.15))
    AND ST_Contains(ST_PolygonFromEnvelope(-117.11, 33.095, -117.04, 33.15), ST_Centroid(geometry))
),
fp AS (
  SELECT md5(wkb) AS k, COUNT(*) AS records, FIRST(wkb) AS wkb
  FROM p GROUP BY md5(wkb)
),
fpa AS (
  SELECT k, records, g, ST_AreaSpheroid(g) AS m2
  FROM (SELECT k, records, ST_MakeValid(ST_GeomFromWKB(wkb)) AS g FROM fp)
),
overlaps AS (
  SELECT a.m2 AS ma, b.m2 AS mb, ST_AreaSpheroid(ST_Intersection(a.g, b.g)) AS o
  FROM fpa a JOIN fpa b ON ST_Intersects(a.g, b.g) AND a.k < b.k
),
buildings AS (
  SELECT id, ST_Centroid(geometry) AS centroid
  FROM wherobots_open_data.overture_maps_foundation.buildings_building
  WHERE bbox.xmax >= -117.12 AND bbox.xmin <= -117.03
    AND bbox.ymax >= 33.085 AND bbox.ymin <= 33.16
),
per_fp AS (
  SELECT f.k, f.records, COUNT(b.id) AS n
  FROM fpa f LEFT JOIN buildings b ON ST_Contains(f.g, b.centroid)
  GROUP BY f.k, f.records
)
SELECT 'A footprints' AS section, IF(records > 1, 'stacked (2+ records)', 'single record') AS label,
       COUNT(*) AS n, SUM(records) AS records, MAX(records) AS extra
FROM fp GROUP BY 2
UNION ALL
SELECT 'B buildings per footprint', CASE WHEN n = 0 THEN '0' WHEN n = 1 THEN '1' WHEN n = 2 THEN '2'
                                         WHEN n <= 5 THEN '3-5' ELSE '6+' END,
       COUNT(*), SUM(records), SUM(n)
FROM per_fp GROUP BY 2
UNION ALL
SELECT 'C footprint pairs that intersect',
       CASE WHEN o >= 0.99 * LEAST(ma, mb) THEN '1 one inside the other'
            WHEN o >= 10 THEN '2 overlap of 10 m2 or more'
            WHEN o >= 1 THEN '3 overlap of 1 to 10 m2'
            WHEN o > 0 THEN '4 overlap under 1 m2'
            ELSE '5 shared edge or corner only' END,
       COUNT(*), NULL, ROUND(SUM(o))
FROM overlaps GROUP BY 2
ORDER BY section, label

The records collapse to 17,782 distinct footprints. Of those, 17,482 carry one record each, and 300 stacked footprints carry the other 5,793 records. The largest stack holds 339 records on one polygon. Condominiums produce this pattern: each unit carries its own APN, mapped to the shared lot. Counted once per footprint, 12,498 footprints (70%) hold one building, 2,349 hold two, 987 hold three to five, and 284 hold six or more. 1,664 footprints hold no building.

The overlap check finds clean topology. 33,830 footprint pairs touch only at an edge or a corner. 60 pairs overlap by under 1 m² and one pair by 2 m², consistent with small digitizing errors. No pair overlaps by 10 m² or more, and no footprint sits inside another once exact duplicates are merged.

The same joins at state scale

One Wherobots project scored every building in Colorado, 2,771,126 Overture buildings on their Regrid parcels for five perils. The result is the Colorado property risk explorer. The same join can add flooded Sentinel-2 pixels or elevation, because WherobotsDB handles raster and vector data in one engine. Most stacks pick one.

The Model Context Protocol (MCP) is a protocol that lets AI agents call data tools. An agent connected to parcel data through the Wherobots MCP server can answer "which parcels over 15 acres sit within a mile of a substation?" with a spatial query.

Parcel data research and what changes at scale

From land registers to the multipurpose cadastre

Deed registration, the common US model, records documents and leaves the buyer to trace the chain of title. South Australia's Real Property Act 1858, designed by Robert Torrens, replaced that chain with a central public register. Title passes by registration. Many countries have since adopted Torrens registration and extended it to mortgages, leases, and strata titles.

The National Research Council (NRC) published Need for a Multipurpose Cadastre in 1980. It called for a parcel-level land information system with five components. Three are a geodetic reference frame (the network of surveyed control points), current large-scale base maps, and a cadastral overlay of every parcel. The other two are a unique identifying number for each parcel and land data files keyed to that number. The NRC returned to the topic in National Land Parcel Data: A Vision for the Future (2007), which concluded that nationally integrated parcel data was necessary, feasible, and affordable.

FGDC, FIG, and LADM cadastral standards

The International Association of Assessing Officers' Standard on Digital Cadastral Maps and Parcel Identifiers sets mapping and identifier practice for US assessors. The Federal Geographic Data Committee (FGDC) publishes the Cadastral Data Content Standard (version 1.4, 2008), maintained by the BLM. It defines the entities of publicly available land records: survey measurements, transactions on interests in land, property descriptions, and boundary and corner evidence.

Kaufmann and Steudler's Cadastre 2014 (FIG, 1998) predicted a shift from drawing cadastral maps to modeling land objects in databases. The International Organization for Standardization (ISO) made that shift formal in ISO 19152:2012, the Land Administration Domain Model (LADM). It is a conceptual model of parties, rights, restrictions and responsibilities, and spatial units. Lemmen, van Oosterom, and Bennett (2015) describe it in Land Use Policy.

Parcel data as research data

Nolte (2020), in PNAS, built parcel-level land value maps for the contiguous United States. It showed that earlier cost proxies underestimated the budgets for floodplain protection by a factor of 2 and for climate-driven species conservation planning by a factor of 37.5. Uhl et al. (2021) used Zillow's ZTRAX assessor and transaction database to map built-up properties in the conterminous US from 1810 to 2015 at 250 m resolution.

What scale changes

A nationwide parcel dataset holds over 158 million records, and joins to buildings, flood zones, and pixels multiply that. Yu, Wu, and Sarwat (2015) introduced GeoSpark, now Apache Sedona, which partitions both sides of a join by space and runs it in parallel across a cluster. WherobotsDB runs the queries above on that design.

Open problems

• Identity over time. Splits and merges retire parcel numbers, and a new parcel number carries no link to the numbers it replaced.
• Legal vs mapped boundaries. Rebuilding tax maps from survey measurements is slow, county by county.
• Three-dimensional parcels. Condominiums, air rights, and subsurface rights need 3D spatial units, which LADM defines and a 2D layer stores as stacked records.
• Completeness of attributes. Zoning, sale prices, and building details can be missing across a whole extract, as in Escondido.

Conclusions

In Escondido, parcel data records where each parcel sits and who owns it, with thin attributes and stacked records that every join must handle.

1. Parcel data is near complete on ownership and use, and thin on detail. Owner, county use code, and legal description fill over 99.9% of 23,275 records. Assessed value fills 94%, zoning 7.5%, and year built and sale date none. If a question needs a thin field, source it elsewhere.
2. A parcel layer covers almost every building. Of 26,635 buildings, 23 fell outside every parcel. 3,519 matched two or more stacked records. Match building centroids to merged footprints.
3. One parcel polygon can carry many records. 23,275 records collapsed to 17,782 footprints, one carrying 339. Parcel areas that sum past the study area (273.942 km² in 39.85 km²) point to stacking. Merge identical geometries before joins and counts.
4. Parcel lines assign features, and a survey sets the boundary. No Escondido footprint pair overlapped by 10 m² or more. A Queensland study put cadastral lines 3 to 5 m off fence lines in places.
5. Building counts in parcel data are uncertain (one hedge). Regrid's ll_bldg_count matched the Overture count on 13,079 records. Their sources and dates differ, so this walkthrough cannot say which is right.

Related terms

• Building footprint: the outline of each building on a parcel
• Spatial join: matching parcels to buildings, imagery, and hazards by location
• Catastrophe modeling: estimating losses across a portfolio of properties
• Flood risk data: hazard layers joined to parcels
• Data center site selection: screening parcels for power, water, and size
• Coordinate reference system: the projections parcels arrive in
• Physical AI: AI that works with real places

Read more from Wherobots

• How to score every building in a state for catastrophe risk
• El Niño 2026, atmospheric rivers, and California's burn scars: one SQL engine, two data models
• Learning Wherobots by building a national AI data center suitability report

Join parcels to buildings, imagery, and hazard layers with a Wherobots free trial at cloud.wherobots.com.