Parcel-Derived Land Metrics: Lot Size, Coverage, and Assembly Patterns

Every land metric a lender relies on is a quotient. Coverage is a footprint over a parcel. Floor area ratio is a floor area over a parcel. Land value share is an assessed land value over an assessed total. Even lot size, which looks like a single measurement, is usually a quotient or a product of two other fields. The numerators and denominators come from different offices, on different cycles, under different law, and the arithmetic hides all of it. A metric computed from public parcel records is therefore only as honest as the account you can give of where each of its terms came from.

That account is worth giving because these metrics carry real weight. The Interagency Guidelines for Real Estate Lending Policies set a supervisory loan-to-value limit of 65 percent on raw land and 75 percent on land development, against 85 percent on improved property (12 CFR part 365, subpart A, Appendix A, eCFR text in force August 20, 2026). The gap between 65 and 85 turns on which category a site sits in, and that classification is argued from land use codes, lot area, coverage and improvement value, all of them parcel-derived. A number that moves a supervisory limit by 20 points deserves a documented recipe.

This piece defines six parcel-derived land metrics precisely, gives the computation recipe for each, names the failure mode that breaks it, and works one screening problem end to end on named public layers. Every figure below is either quoted from a document read on August 24, 2026 or computed here from a public data service on that date, with the county named and the method stated. It sits inside the wider library of CRE benchmarks built from public data, and it assumes the source diagnostic has already been done: the companion study of parcel record availability across the country covers who publishes what, and the parcel data options piece covers where a lending team gets a national layer at all.

The counterintuitive result arrives early, in the first metric. Lot size, the simplest number on the list, is in most tax rolls not a measurement of the parcel. It is the product of two other recorded fields, or one of those fields is the quotient of the other two. The three numbers a site screen treats as independent facts are, in the two large public systems examined here, two facts and one piece of arithmetic, and the systems disagree about which is which.

What a parcel-derived land metric is made of

Three separate products have to be joined before any of these metrics exists.

The first is the parcel polygon, a GIS representation of the assessment unit, drawn by a county mapping or land information office from plats, deeds and survey control. The second is the assessment roll, maintained by an assessor or appraisal district for the purpose of levying tax, which carries the parcel identifier, a land use or property class code, assessed values, land area, and in many jurisdictions a set of building attributes. The third, needed for coverage and for any honest statement about what is built on a site, is a building footprint layer, which almost never comes from either of the first two offices.

Utah makes the split visible in the record itself. Every parcel in the Land Information Records layer published through the state geospatial program carries two lineage fields: an assessor source and a boundary source. For Salt Lake County, all 394,610 records name the county assessor for the attributes and the county recorder for the boundary (Utah Geospatial Resource Center, Salt Lake County LIR parcels feature service, read August 24, 2026). Two offices, two production cycles, one row. Nothing in the row tells you that the two were current on the same day, and nothing requires them to have been.

The practical constraint is not the geometry. It is whether the attribute you need was populated at all. Across the 29 Utah counties that publish a Land Information Records layer, covering 1,845,811 parcels read on August 24, 2026, the parcel acreage field is populated on 90.7 percent of records and the property class on 91.3 percent, but building square footage falls to 69.0 percent, floor count to 65.5 percent and construction material to 63.2 percent (MMCG tabulation from the 29 county LIR feature services, read August 24, 2026). The distribution behind those averages matters more than the averages. Summit County publishes 37,294 parcels with a building square footage on none of them. Six counties publish no floor count at all. Sanpete County carries an acreage on 25.8 percent of its parcels and a building area on 14.3 percent.

Read that as an availability envelope rather than a quality complaint. A lot size metric is computable across essentially the whole state. A coverage or floor area ratio metric built on assessor building area is computable across roughly two thirds of it, and the missing third is not random: it is the rural counties, which is where raw land and land development lending concentrates. The metric you can compute nationally and the metric you want are not the same metric, and the honest report says which one it is showing. The companion census of building attributes from assessor rolls works that availability problem at national scale.

Why land metrics enter the credit file

The Interagency Guidelines for Real Estate Lending Policies, codified for state non-member banks at 12 CFR part 365, subpart A, Appendix A, with parallel text at 12 CFR part 34, subpart D, Appendix A and 12 CFR part 208, Appendix C, instruct institutions to set internal loan-to-value limits that do not exceed a published supervisory schedule. Raw land is 65 percent. Land development is 75 percent. Construction of commercial, multifamily and other nonresidential property is 80 percent. One-to-four-family residential construction is 85 percent, as is improved property. For owner-occupied one-to-four-family and home equity lending no limit is set, with credit enhancement expected at or above 90 percent (eCFR text in force August 20, 2026).

Two features of that schedule drive the analytics. First, the categories are defined by the state of the land, not by the borrower or the loan structure. The guidelines define a land development loan as an extension of credit for improving unimproved real property before structures are erected, and an improved property loan as one secured by completed commercial property or other completed income-producing property available for occupancy and use. Second, and less often noticed, the guidelines do not define raw land anywhere in the definitions section. The term appears once, in the table, carrying the tightest limit in the schedule. The classification is left to the institution, and in practice the evidence offered is a land use code, an improvement value of zero, an absence of building attributes, and a lot area.

The consequences of getting it wrong are not confined to one file. The same appendix caps the aggregate of all loans exceeding the supervisory limits at 100 percent of total capital, and within that caps loans on commercial, agricultural, multifamily and other non-one-to-four-family property at 30 percent of total capital, with increased supervisory scrutiny as those levels approach. A land metric that misclassifies collateral does not simply misprice one advance, it misstates a portfolio exception bucket that examiners read. That is the connection between a parcel field and a concentration report, and it is why CRE concentration monitoring built on public data and the parcel layer end up drawing on the same records.

None of this makes an analytics provider a party to the credit decision. MMCG Analytics supplies data and analytics with source and vintage provenance carried on displayed values; the classification, the limit and the exception belong to the lender. What the analytics can do is make the classification argument auditable, which means showing the field, the office that produced it, the date it was current, and the arithmetic applied to it.

Lot size: the metric that is usually not a measurement

Definition. Lot size is the ground area of the legal parcel, expressed in square feet or acres. Two public sources appear to offer it: the area of the GIS polygon, computed from the geometry, and the area field on the assessment roll, variously labelled lot area, parcel acres or deeded acreage.

Recipe. Compute the polygon area in an equal-area or local projected coordinate system, never in a web mapping projection, where area scales with the square of the secant of latitude and a national layer will be wrong by tens of percent at the extremes. Then read the roll's area field. Then compare them, parcel by parcel, and report the distribution of the difference rather than picking one.

Failure mode, and the reason the comparison is not a quality check. In New York City, the Department of Finance calculates lot area by multiplying lot frontage by lot depth for every tax lot not flagged as irregularly shaped; for irregular lots it calculates the area from the Digital Tax Map instead (New York City Department of City Planning, PLUTO Data Dictionary, August 2022 edition). The roll's lot area is therefore not an independent observation of the parcel. For the large majority of lots it is a rectangle assumption written into a field.

The evidence is in the data. Across Kings County, New York, in PLUTO version 26v1, 275,305 tax lots carry a positive frontage, a positive depth and a positive lot area. On 258,166 of them, or 93.8 percent, frontage multiplied by depth reproduces the recorded lot area to within 5 percent. Restrict the same test to the 30,144 lots the Department of Finance flags as irregularly shaped and the agreement collapses to 14,780 lots, or 49.0 percent (MMCG tabulation from PLUTO 26v1, read August 24, 2026). The rectangle holds exactly where the agency says it built the number by rectangle, and fails where the agency says it went to the map.

Now the counterintuitive part. Salt Lake County runs the same three fields in the opposite direction. Its assessor's land record documentation defines effective frontage as the average frontage for valuation purposes, square feet as the recorded area of a land segment, and depth as a derived quantity: the depth is determined by dividing the square feet by the effective frontage (Salt Lake County Assessor, Land Record Field Descriptions, read August 24, 2026). The county also keeps a separate legal frontage field for the actual frontage or width of the parcel, which is not the same field used in valuation.

So two large, well-run, fully public systems both publish lot area, frontage and depth, and in neither system are all three measured. In one, area is the product of the other two. In the other, depth is the quotient of the other two. A screen that treats frontage, depth and lot area as three independent facts about a site is double counting a single measurement, and it is doing so in opposite directions in the two jurisdictions. This is not sloppiness by either assessor. Both are transparent about it in their own documentation. It is a property of mass appraisal records that an analytics layer has to carry forward rather than average away.

The second failure mode is distributional. Lot size is not a quantity with a useful average. In Kings County, of 276,311 tax lots, 167,039 are under 2,500 square feet and a further 83,311 fall between 2,500 and 5,000 square feet: 90.6 percent of the county's lots sit below 5,000 square feet, while 343 lots of five acres or more carry a disproportionate share of the land. Restrict to lots carrying an office, retail, warehouse, factory, garage, loft, hotel or educational building class and the shape changes completely: of 17,222 such lots, 5,295 are under 2,500 square feet and 818 are an acre or more. A mean lot size across either population describes no property in it.

The third failure mode is unit and vocabulary drift across jurisdictions. Kings County publishes square feet on a tax lot. Salt Lake County publishes acres on a parcel, and its 26,046 commercial and industrial parcels distribute across acreage bands with 10,919 under a quarter acre and 449 at 20 acres or more (MMCG tabulation from the Salt Lake County LIR parcels feature service, read August 24, 2026). Comparing the two requires converting units, reconciling two different definitions of the assessment unit, and accepting that a New York tax lot and a Utah parcel are not the same object. Condominium regimes make this sharper still: a single building can generate hundreds of tax lots with no separate ground area, which is why 226 Kings County records carry no lot area at all and 59 carry a lot area of zero.

Frontage and depth: what a corridor screen may claim

Definition. Frontage is the length of the parcel boundary along a street; depth is the perpendicular distance from that street line to the rear boundary. Both are used constantly in site screening: minimum frontage governs curb cuts and access, depth governs whether a prototype building and its parking fit.

Recipe. Compute frontage from geometry by intersecting the parcel boundary with a road centerline buffer or a street right-of-way polygon, and report the result as a geometric estimate. Read the roll's frontage and depth fields separately, and treat them as a second, differently constructed estimate rather than as confirmation.

Failure modes are three. First, the dependency described above: in a system where lot area is frontage multiplied by depth, the depth field can absorb every irregularity in the parcel and become a fiction that makes the product come out right. Second, corner and multi-frontage lots. The New York dictionary notes that when a lot fronts on more than one street, the building address in the source system often determines which side is used for the frontage calculation. A corner site with 200 feet on an arterial and 90 feet on a side street may be recorded at 90. Third, missing values and irregular geometry. In Kings County, 971 lots carry no frontage and 989 carry no depth, and 30,223 lots, 10.9 percent of the county, are flagged irregular.

What follows for practice is narrow and useful. A frontage screen run on roll fields is a filter, not a finding. It is legitimate to use it to shrink a corridor of several thousand parcels to a few dozen candidates. It is not legitimate to report a frontage figure for a specific site in a credit file without a geometric measurement or a survey behind it, and the same caution applies to every dimension that feeds a parcel-to-buildable analysis, where setbacks and yards compound whatever error the base dimensions carry.

Lot coverage: the metric that needs a second layer

Definition. Lot coverage is the ground area of building footprints on a parcel divided by the parcel area, expressed as a percentage. It is the metric zoning codes regulate directly, and it is the one metric on this list that cannot be computed from a parcel file alone, because the assessment roll records floor area and dimensions, not footprints.

Recipe. Join a building footprint layer to the parcel layer spatially, sum footprint area within each parcel, divide by parcel area. Report the footprint layer by name, its vintage and its confidence attributes, and report the share of parcels where the join produced no footprint at all.

The public footprint options are federal. USA Structures, produced by FEMA with Oak Ridge National Laboratory, published a derived feature set covering more than 131 million buildings in the United States (Scientific Data, 2025). Read through its public feature service on August 24, 2026, the layer carried 135,321,228 structures. Its occupancy classification is where the caution belongs: 114,549,184 structures are classed Residential, 5,761,065 Commercial, 4,229,321 Agriculture and 1,353,406 Industrial, while 6,730,545 are Unclassified (MMCG tabulation from the USA Structures feature service, read August 24, 2026). There are more unclassified structures in the layer than commercial ones.

The unclassified share is also geographically uneven, which matters for any national coverage benchmark. Among states with at least one million structures, the unclassified share runs from 1.0 percent in Minnesota and 1.4 percent in Maryland to 12.2 percent in Arkansas, 18.6 percent in Louisiana and 25.9 percent in Puerto Rico. The commercial share of structures moves in a much narrower band, from 2.8 percent in Minnesota to 5.8 percent in Wisconsin, which is a reminder that the classification is derived rather than surveyed and that its density says as much about the classification pass as about the built environment.

The second federal option is the USACE National Structure Inventory, a point inventory rather than a footprint layer, whose 2026 technical documentation describes the base dataset as a nationally consistent modeled inventory intended as a practical starting point where more detailed local inventories are unavailable or infeasible. Its schema is unusually candid about lineage: a source field records whether a structure originated from parcel data or another inventory, a footprint identifier records which footprint record was used to estimate square footage and story count, and a separate footprint source field records where that footprint came from, which is not one provider. A layer that names the provenance of each estimate is a layer you can report honestly. A layer that does not is one you should not put a coverage ratio on.

The failure mode that catches most teams is the temptation to skip the footprint join and proxy coverage from assessor dimensions. The arithmetic looks reasonable: multiply building frontage by building depth by building count, or divide total floor area by the number of floors, and divide either by lot area. In Kings County, that calculation can be attempted on 14,883 of the 17,222 commercial and industrial lots, because the rest lack a usable building dimension. Of those 14,883, the frontage-by-depth proxy returns a coverage ratio above 1.0 on 2,220 lots, 14.9 percent of the attempts (MMCG tabulation from PLUTO 26v1, read August 24, 2026). A coverage above 100 percent is not a result, it is a diagnostic: the assessor's building dimensions describe the primary structure at its widest points, they are not a footprint, and 1,828 of those lots carry more than one building. Use the proxy to rank candidates. Do not put its output in a file as a coverage figure.

Floor area ratio from assessor building area

Definition. Floor area ratio is gross building floor area divided by lot area. Built FAR describes what stands; allowable FAR describes what the zoning district permits. The difference between them is the development capacity of a site, and it is the single most useful parcel-derived number in early screening.

Recipe. Built FAR is the roll's total building floor area over the roll's lot area. Allowable FAR is not in the parcel file at all: it comes from the zoning layer, joined by geometry, and it is a district attribute rather than a parcel attribute. New York City is unusual in publishing both on the same record, with maximum allowable residential, commercial and community facility FAR assigned from the zoning district occupying the greatest share of the lot, exclusive of bonuses. Most jurisdictions require a separate join, and coverage of machine-readable zoning is far thinner than coverage of parcels, which is the subject of the companion work on the sources and coverage of U.S. zoning data.

The identity holds where the inputs exist. In Kings County, of 265,752 lots with a positive lot area, floor area and built FAR, 262,006, or 98.6 percent, reproduce the published built FAR from floor area divided by lot area to within 1 percent. That is a useful validation: the field is what it says it is. The caution is in the numerator. New York's total building floor area is described by its own dictionary as a rough measurement, and its source is recorded per lot in a separate code: of the 276,311 Kings County lots, 265,904 take it from the property tax system, 10,081 from the mass appraisal system, 104 from a vacant-class rule that sets the area to zero, 33 from a calculation using the primary building's dimensions and floor count, 82 from no available source, and 107 carry no source code at all. In 16,229 lots, 5.9 percent of the county, the city's planning department edited the record. Floor area is a reconstruction, and the file should say which reconstruction.

Failure modes: basements and mezzanines are treated inconsistently across rolls; floor counts are null wherever the assessor did not record them, which breaks the floors-based footprint proxy; and a single FAR figure across a split-zoned lot is an average of two entitlements. For any site where the FAR gap drives the decision, the parcel-derived figure is the screen and the zoning text is the answer.

Land value share: the metric that breaks at the state line

Definition. Land value share is assessed land value divided by assessed total value. It is the cheapest available signal for whether a site is underimproved relative to its dirt, which is the economic condition that precedes redevelopment and assemblage.

Recipe. Sum assessed land value and assessed total value within a class, a district or a corridor, and divide. Never average the parcel-level ratios: the distribution is bounded at 1.0, heavily massed near it for vacant land, and a mean of ratios is dominated by small parcels.

Computed within one roll, the metric behaves exactly as theory says it should. In Salt Lake County, land accounts for 99.2 percent of assessed value on parcels classed Vacant and 100.0 percent on parcels classed Undeveloped, which is the sanity check. Among improved classes it ranges from 22.7 percent on commercial apartment and condominium parcels and 26.1 percent on industrial parcels to 40.0 percent on general commercial, 45.0 percent on commercial industrial and 55.9 percent on commercial retail (MMCG tabulation from the Salt Lake County LIR parcels feature service, read August 24, 2026). On retail parcels in that county, more than half of the assessed value is ground. That is a real and actionable ordering, and it holds within the roll that produced it.

Cross the state line and the ordering stops meaning what it appears to mean. In New York City, assessed land value is defined as the assessor's estimate of full market land value as if vacant and unimproved, multiplied by a uniform percentage set for the property's tax class, and assessed total value is full market value multiplied by that same class percentage (New York City Department of City Planning, PLUTO Data Dictionary, August 2022 edition). Computed on PLUTO 26v1, the land share of assessed value is 20.0 percent in New York County, 17.2 percent in Kings County, 17.0 percent in Bronx County, 28.2 percent in Queens County and 38.5 percent in Richmond County. The most expensive land in the country sits in the borough with nearly the lowest land share, and the outer borough with the cheapest land posts the highest. Within Kings County, the land share is 13.5 percent across commercial and office lots, 17.6 percent across industrial and manufacturing lots, 73.7 percent across parking facilities and 99.9 percent across vacant land.

Set the two rolls side by side and the comparability problem is unmistakable: commercial retail land share of 55.9 percent in Salt Lake County against a commercial and office land share of 13.5 percent in Kings County. Nothing about the underlying land economics produces a spread that wide. Assessment class ratios, statutory caps on assessment growth, differing treatment of exemptions and different revaluation cycles produce it. Land value share is a within-roll metric. Used across rolls without a stated adjustment it is not a benchmark, it is an artifact, and the same discipline that governs attaching a source and a date to every number is what keeps it from travelling further than it should.

Assembly patterns: inference, not fact

Definition. An assembly is a set of adjacent parcels under common control, held or acquired so they can be developed as one site. Public parcel records do not record control. They record an owner name for tax billing, and in many rolls a mailing address. Everything else is inference.

Recipe, stated as steps so it can be audited. Normalize owner names by upper-casing, stripping punctuation and collapsing whitespace. Group the normalized names. Restrict to groups whose parcels share a tax block, or better, whose polygons touch. Classify the surviving groups by owner type. Report every step's count, and report the classification rule.

Step one already changes the answer. In Kings County, the raw owner name field yields 241,764 distinct strings, of which 17,384 appear on more than one lot, covering 51,688 lots. After a single normalization pass the distinct strings fall to 239,285, the multi-lot names rise to 18,833 and the lots they cover rise to 55,616 (MMCG tabulation from PLUTO 26v1, read August 24, 2026). One deterministic cleaning step, no fuzzy matching, no judgment, moves the population of commonly owned lots by 7.6 percent. Two analysts running the same query on the same file will disagree by more than that if one of them skips the step.

Step two shows why county-wide name matching is the wrong screen. Of those 55,616 commonly owned lots, only 24,864 sit on the same tax block as their match, in 9,802 groups. Common ownership across a county is a portfolio fact. Adjacency is the development fact, and the county-wide match overstates the adjacent population by a factor of 2.2. The groups that survive are small: 7,414 of the 9,802 are two-lot pairs, 1,370 are three lots, and only 80 groups reach ten lots or more.

Step three is where the method's real bias appears. Take those 80 largest same-block clusters, 1,232 lots in all, and classify them by owner. Twenty-nine are public agencies and authorities, holding 490 lots: a bi-state port authority, the city parks department in fourteen separate blocks, the police department, sanitation, homeless services, citywide administrative services and the housing authority. Twelve are subsidized housing entities and development fund companies, holding 219 lots. Two are institutions. Thirteen, holding 181 lots, are not an owner at all: they are the literal placeholder string "UNAVAILABLE OWNER", which appears on 3,600 Kings County lots county-wide. That leaves 24 clusters, holding 319 lots, in ordinary private hands.

Owner-name matching, in other words, is very good at finding the public estate and its own missing data, and comparatively poor at finding private assemblage. The reason is structural. Public owners reuse one normalized name across hundreds of parcels; New York's planning department even normalizes them deliberately, folding several spellings of the parks department into one string. Private acquirers do the opposite. In Kings County, 50,063 lots carry an owner name containing the token LLC, spread across 43,992 distinct normalized names, and 40,394 of those names, 91.8 percent, appear on exactly one lot.

Whether that pattern reflects deliberate single-purpose entity structuring or simply the ordinary way small properties are titled cannot be settled from the parcel file, and this piece does not claim it can. What can be documented is that the public record offers no route behind the name. FinCEN's final rule of August 14, 2026, adopting and extending its interim rule of March 26, 2025, exempts reporting companies from reporting the beneficial ownership information of U.S. person beneficial owners and exempts U.S. person company applicants entirely (91 FR 52508). The one federal reporting regime that reaches real estate transfers to legal entities and trusts, the Anti-Money Laundering Regulations for Residential Real Estate Transfers, is by its terms residential and its reports go to Treasury rather than to a public record (89 FR 70258, published August 29, 2024). Commercial transfers to entities sit outside it.

The practical consequence is a reporting standard rather than a technique. An assembly finding from parcel data is a hypothesis with a stated recall problem. Write it as one: these parcels share a normalized owner name and a block, this many candidates were found, this classification rule was applied, and the method cannot see separately named entities under common control. Mailing address joins, where a roll publishes them, raise recall and introduce their own error, since agents, attorneys and property managers collect mail for unrelated owners. Neither method converts inference into fact.

A worked screen: underbuilt commercial land in one county

The metrics above are worth defining because they combine into decisions. Here is one, run end to end on a single named public file, with every intermediate count reported so the result can be reproduced or contradicted. The question: where in Kings County, New York, is there commercially zoned land that is materially underbuilt relative to its entitlement and plausibly assemblable? The file: PLUTO version 26v1, published by the New York City Department of City Planning through the city's open data portal, queried through its public API on August 24, 2026. No bulk download was taken and no extract was retained.

Start with 276,311 tax lots. Restrict to lots whose maximum allowable commercial floor area ratio is above zero, which is the file's own encoding of commercial entitlement: 18,645 lots. Restrict to lots of at least 2,500 square feet, on the grounds that a smaller site cannot absorb a commercial prototype: 11,318 lots. Restrict to lots built to half or less of their allowable commercial FAR: 4,982 lots, carrying 225,605,919 square feet of land.

That number is where an unexamined screen would stop, and it would be badly wrong. Rank those 4,982 lots by land area and the largest is a single parcel of 52,591,959 square feet owned by the National Park Service, zoned M1-1, with a built FAR of zero. The second is 23,108,842 square feet held by the city's Department of Small Business Services. Of the 4,982 lots, 489 belong to public agencies and authorities and carry 67,410,631 square feet, and 35 more, carrying 2,655,057 square feet, are owned by the placeholder string. Removing both leaves 4,458 lots and 155,540,231 square feet.

One more cut is needed, and it is the one most screens omit. Capping lot area at ten acres, on the grounds that a site above that size is a different transaction with different diligence, removes 24 lots and 87,466,234 square feet. Twenty-four lots out of 4,458 were carrying 56 percent of the land in the private set. The screen finishes at 4,434 lots and 68,073,997 square feet across 1,293 tax blocks, of which 610 blocks hold three or more screened lots.

Overlay ownership last, not first. Of the 4,434 surviving lots, 1,609, or 36.3 percent, sit on a block where the same normalized owner name already holds at least one other lot. Those are the assemblage candidates, and the count is the honest one: not 4,982, not 55,616, but roughly sixteen hundred lots that pass a stated entitlement test, a stated size test, a stated ownership-exclusion rule and a stated adjacency rule, in one county, on one file version, on one date.

What the output is: a ranked candidate list that shortens a corridor review from weeks to an afternoon, with a documented lineage for every filter. What it is not: a valuation, a confirmation that any parcel is for sale, a statement that any owner intends to assemble, or a substitute for a title search and a survey. A lender using it in a pre-term-sheet site screen should carry the filters and their counts into the file alongside the result, because a screen whose thresholds are undocumented is not reproducible and therefore not evidence.

What breaks when you cross a jurisdiction line

Four things break, in a predictable order.

Vocabulary breaks first. Salt Lake County classifies parcels into Residential, Commercial, Commercial Office Space, Commercial Retail, Commercial Industrial, Industrial, Vacant, Undeveloped, Greenbelt, Recreational and several exempt categories. New York City assigns each of its building classes to one of eleven land use categories and separately publishes the building class itself. There is no crosswalk published by either and no way to map "Commercial Retail" onto "Commercial and Office Buildings" without a judgment that should be written down. Nor is there a federal standard to appeal to. The Geographic Information Framework Data Content Standard, Part 1: Cadastral (FGDC-STD-014.1-2008, May 2008), the federal content standard for parcel data, says in its own scope that it is not intended to support real estate records or other application-based information and that it includes only the minimum data necessary to locate parcel-level information and identify its source. Its data dictionary runs to a parcel geometry, an owner type, a source and a collection wrapper. The phrase "land use" does not appear in the standard at all. Any multi-jurisdiction land metric is therefore carrying a hand-built crosswalk, and that crosswalk is an assumption, not a source.

Standards break second. The 2025 Geospatial Maturity Assessment, published by the National States Geographic Information Council on August 14, 2026 from results collected in the second half of 2025, found that of the 37 statewide parcel programs it graded, 20 apply a parcel standard that includes verification and quality control, 9 apply a standard without verification, 5 collect on a best effort basis and 3 accept county data exactly as received. Where the aggregating state accepts data as received, the schema in the statewide file is a union of county schemas, and a null in a field can mean the attribute is absent, the attribute is zero, or the county never sent that column.

Vintage breaks third, and it breaks inside a single state file rather than between files. Utah's Land Information Records layers align to the finalization of the property tax roll by county assessors in May of each year, and each county layer carries its own current-as-of date. Read on August 24, 2026, those dates ranged from March 26, 2026 for Salt Lake County to July 17, 2020 for Box Elder County, a spread of 68 months within one statewide program (MMCG tabulation from the 29 county LIR feature services, read August 24, 2026). New York City is on a different clock again: assessed values update twice a year, with tentative values released in mid-January and final values around May 25, so the same field means a tentative roll or a final roll depending on when the file was cut. Program refresh policy varies as widely as the data: Montana provides cadastral data monthly for each county, Wisconsin publishes an annual statewide version series whose twelfth release was collected from counties in the first half of 2026, and Texas states that it will attempt an annual refresh from each appraisal district at a rate that varies across the state (Montana State Library, Wisconsin State Cartographer's Office and Wisconsin Land Information Program, and Texas Geographic Information Office program pages, all read in August 2026). A national land metric with a single "as of" date on it is not describing the data it was built from. The broader case for treating each layer's date honestly is made in the work on open analytical atlases from public data.

Identifiers break fourth. The parcel identifier that joins the roll to the polygon is a local construction, is reused after splits and merges in many jurisdictions, and is not stable across annual vintages. Any time series built on parcel identifiers has to be reconciled at every refresh, and any address-based join carries the separate error budget documented in the work on geocoding accuracy against commercial services.

Reporting standards for parcel-derived land metrics

The metrics on this list are all computable and all defensible, provided the report carries six things. This is the disclosure set that turns a parcel-derived figure into evidence rather than an assertion.

Name the file and its version. Not "parcel data" but the publisher, the layer and the version string, because a version is what makes a number reproducible. PLUTO 26v1 and the Utah LIR layer for a named county with its current-as-of date are examples of the standard; a national layer with no version is not.

Name the office behind each term of the quotient. Coverage joins two offices, sometimes three. Land value share is one office. Lot size may be one office pretending to be two. When the terms come from different producers, say so, because that is where the vintage mismatch lives.

State the construction rule where the publisher states one. If the roll's lot area is frontage multiplied by depth, the report says so. If depth is square feet divided by frontage, the report says so. Both are documented facts published by the assessors themselves, and omitting them lets a reader believe three independent measurements exist.

Report the denominator population and the null rate, not only the metric. A coverage figure computed on the 69 percent of parcels that carry a building area is not a coverage figure for the county. Report both the value and the share of the population it was computable on.

Report the filters and their counts for any screen. The worked example above moves from 276,311 lots to 4,434 in six documented steps. A reader who disagrees with the ten-acre cap or the half-of-allowable-FAR threshold can re-run it. A reader given only the final number cannot.

Label inference as inference. Assembly detection, land classification for a supervisory limit, and any coverage figure built from a proxy are inferences. They belong in a file with the method attached and the recall problem stated. The same standard governs every other public-data benchmark, from supply per capita metrics to the share of commercial land in the floodplain, and it is the difference between analytics a credit committee can interrogate and analytics it has to take on faith. Teams weighing what public sources cover before they buy will find the same tests applied in the audit of free federal data against paid subscriptions.

Frequently asked questions

How do I calculate lot coverage from parcel data?

Lot coverage is building footprint area divided by parcel area, and it requires a footprint layer that the parcel file does not contain. Join a footprint layer such as FEMA and Oak Ridge National Laboratory's USA Structures to the parcel polygons, sum footprint area within each parcel, and divide. Report the footprint layer, its vintage and the share of parcels where the join found nothing. Proxying coverage from assessor building dimensions fails often enough to disqualify it for reporting: on Kings County, New York commercial and industrial lots, the building frontage by depth proxy returns a coverage above 100 percent on 14.9 percent of the lots where it can be attempted.

Why does the acreage on the tax roll differ from the GIS parcel area?

Because in many jurisdictions they are not two measurements of the same thing. New York City's Department of Finance computes lot area by multiplying lot frontage by lot depth for every lot not flagged irregular, and computes it from the digital tax map only for irregular lots. Salt Lake County's assessor computes depth by dividing square feet by effective frontage. Deeded acreage also comes from a legal description that may predate any survey control, while the polygon is compiled from plats and imagery. Report both, report the difference, and never assume the roll's figure is an independent check on the map.

Can parcel data show me who is assembling land?

It can produce candidates, not answers. Normalize owner names, group them, and keep only groups whose parcels share a block or touch. In Kings County, New York, that method returns 9,802 same-block groups covering 24,864 lots, but of the 80 largest groups, 29 are public agencies, 13 are the placeholder string "UNAVAILABLE OWNER", 12 are subsidized housing entities and only 24 are ordinary private owners. Separately named entities under common control are invisible to the method, and no public federal record closes the gap: FinCEN's final rule of August 14, 2026 exempts U.S. person beneficial owners from beneficial ownership reporting.

What is a good land value share for a commercial site?

There is no cross-jurisdiction benchmark, and any figure presented as one should be treated with suspicion. Within a single roll the metric is informative: in Salt Lake County, land is 55.9 percent of assessed value on commercial retail parcels, 45.0 percent on commercial industrial and 22.7 percent on commercial apartment and condominium parcels. In Kings County, New York, land is 13.5 percent of assessed value on commercial and office lots. The spread reflects assessment class ratios, statutory caps and revaluation cycles, not land economics. Compute it within the roll, compare it to peers in the same roll, and state the assessment regime.

Which loan-to-value limit applies to a land loan?

Under the Interagency Guidelines for Real Estate Lending Policies, the supervisory limits are 65 percent for raw land, 75 percent for land development, 80 percent for construction of commercial, multifamily and other nonresidential property, and 85 percent for improved property and one-to-four-family residential construction. The guidelines define a land development loan and an improved property loan but do not define raw land, so the classification rests on the institution's own evidence, which in practice means the land use code, the improvement value and the parcel attributes. Loans exceeding the limits are capped in aggregate at 100 percent of total capital, with a 30 percent sub-limit for non-one-to-four-family property.

How current is parcel data, really?

It is a distribution rather than a date, and the spread sits inside single state programs. Across Utah's 29 Land Information Records county layers read on August 24, 2026, the current-as-of dates ran from March 26, 2026 to July 17, 2020, a spread of 68 months. New York City updates assessed values twice a year, releasing tentative values in mid-January and final values around May 25, so the same field carries a different meaning depending on when the file was cut. Any national parcel metric should publish the vintage distribution, not a single date.

What should a parcel-derived figure disclose before it enters a credit file?

Six things: the file and its version string; the office behind each term of the calculation; the publisher's own construction rule where one is documented; the population the metric was computable on and the null rate; the filters and their counts for any screen; and an explicit label wherever the figure is an inference rather than a measurement. A figure carrying all six can be interrogated and reproduced. A figure carrying none of them is an assertion.

Sources

1. 12 CFR part 365, subpart A, Appendix A, Interagency Guidelines for Real Estate Lending Policies (FDIC), with parallel codifications at 12 CFR part 34, subpart D, Appendix A (OCC) and 12 CFR part 208, Appendix C (Federal Reserve): supervisory loan-to-value limits, the definitions of a land development loan and an improved property loan, and the exception caps at 100 percent and 30 percent of total capital. eCFR text in force August 20, 2026, retrieved August 24, 2026. https://www.ecfr.gov/current/title-12/chapter-III/subchapter-B/part-365/subpart-A/appendix-Appendix%20A%20to%20Subpart%20A%20of%20Part%20365
2. New York City Department of City Planning, Primary Land Use Tax Lot Output (PLUTO), version 26v1, published through NYC Open Data, dataset updated May 28, 2026; 858,602 tax lot records, queried through the public API on August 24, 2026. https://data.cityofnewyork.us/City-Government/Primary-Land-Use-Tax-Lot-Output-PLUTO-/64uk-42ks
3. New York City Department of City Planning, PLUTO Data Dictionary, August 2022 edition (22v2): definitions of lot area, lot frontage, lot depth, irregular lot code, total building floor area, total building floor area source code, built floor area ratio, maximum allowable commercial floor area ratio, owner name, type of ownership code, land use category, assessed land value and assessed total value. Read August 24, 2026. https://www.nyc.gov/assets/planning/download/pdf/data-maps/open-data/PLUTODD.pdf
4. Utah Geospatial Resource Center, Utah Parcels, SGID cadastre category, including the Land Information Records parcel layers and their per county update dates, page read August 24, 2026. https://gis.utah.gov/products/sgid/cadastre/parcels/
5. Utah Geospatial Resource Center, Land Information Records parcel feature services for the 29 participating Utah counties, including the assessor source, boundary source, disclaimer and current-as-of fields and the statement that the Land Information Records product is aligned with the finalization of the property tax roll by county assessors on May 22 of each year; queried August 24, 2026 (Salt Lake County service shown). https://services1.arcgis.com/99lidPhWCzftIe9K/arcgis/rest/services/Parcels_SaltLake_LIR/FeatureServer/0
6. Salt Lake County Assessor, Land Record Field Descriptions: definitions of lot use, lot type, effective frontage, legal frontage, depth, square feet and acres, including the rule that depth is determined by dividing the square feet by the effective frontage. Page read August 24, 2026. https://apps.saltlakecounty.gov/assessor/new/FieldDescriptions/landRecord.html
7. FEMA and Oak Ridge National Laboratory, USA Structures, public feature service: 135,321,228 structures with occupancy classification, square footage, height, image date, production date and validation method fields, the occupancy classification following the FEMA data standard of July 31, 2018; counts by occupancy class and by state queried August 24, 2026. https://services2.arcgis.com/FiaPA4ga0iQKduv3/arcgis/rest/services/USA_Structures_View/FeatureServer/0
8. Scientific Data, the derived feature set published for USA Structures covering more than 131 million buildings in the United States, 2025. https://www.nature.com/articles/s41597-025-05925-6
9. U.S. Army Corps of Engineers, National Structure Inventory, 2026 technical documentation: the description of the 2026 base dataset as a nationally consistent modeled inventory and the field definitions for structure source, estimated square footage, footprint identifier and footprint source. Read August 24, 2026. https://www.hec.usace.army.mil/confluence/nsi/technicalreferences/2026/technical-documentation
10. U.S. Army Corps of Engineers, National Structure Inventory downloads, state GeoPackages and county GeoJSON through the inventory API, page read August 24, 2026. https://nsi.sec.usace.army.mil/downloads/
11. Financial Crimes Enforcement Network, Beneficial Ownership Information Reporting Requirement Revision, final rule, 91 FR 52508, published and effective August 14, 2026, adopting as final the interim final rule issued March 26, 2025 and exempting reporting companies from reporting the beneficial ownership information of U.S. person beneficial owners. https://www.federalregister.gov/documents/2026/08/14/2026-16576/beneficial-ownership-information-reporting-requirement-revision
12. Financial Crimes Enforcement Network, Anti-Money Laundering Regulations for Residential Real Estate Transfers, final rule, 89 FR 70258, published August 29, 2024: reporting and recordkeeping on certain non-financed transfers of residential real property to specified legal entities and trusts. https://www.federalregister.gov/documents/2024/08/29/2024-19198/anti-money-laundering-regulations-for-residential-real-estate-transfers
13. National States Geographic Information Council, 2025 Geospatial Maturity Assessment, Full Report, published August 14, 2026 from results collected in the second half of 2025; Cadastre (Parcels) theme, including the parcel standard applied by each statewide program. https://nsgic.org/wp-content/uploads/2026/08/2025-GMA-Full-Report-20260814.pdf
14. National States Geographic Information Council, 2023 Geospatial Maturity Assessment report, December 2023; cadastre theme summary and statewide program tables. https://nsgic.org/wp-content/uploads/2024/02/2023GMAReportFinal.pdf
15. Federal Geographic Data Committee, Cadastral Subcommittee, coordination of cadastral data among federal, state, tribal and local governments and standardization work with public domain states, page read August 2026. https://www.fgdc.gov/organization/working-groups-subcommittees/cadastral/index_html
16. Wisconsin State Cartographer's Office and Wisconsin Land Information Program, Statewide Parcel Map Initiative, version 12 with 3.56 million records collected from counties in the first half of 2026, page read August 2026. https://www.sco.wisc.edu/parcels/data/
17. Montana State Library, Montana Spatial Data Infrastructure cadastral framework, monthly county cadastral downloads and the Department of Revenue computer assisted mass appraisal data, page read August 2026. https://msl.mt.gov/geoinfo/msdi/cadastral/
18. Texas Geographic Information Office, Texas Water Development Board, StratMap Land Parcels, county availability, the annual refresh statement and the fitness for use statement, page read August 2026. https://geographic.texas.gov/stratmap/land-parcels.html
19. Federal Geographic Data Committee, Geographic Information Framework Data Content Standard, Part 1: Cadastral, FGDC-STD-014.1-2008, May 2008: the scope statement, the parcel application schema and the data dictionary. https://www.fgdc.gov/standards/projects/framework-data-standard/GI_FrameworkDataStandard_Part1_Cadastral.pdf
20. MMCG Research, Parcel-Derived Land Metrics Series: lot area band counts for Kings County, New York and acreage band counts for Salt Lake County, Utah; frontage by depth reconciliation rates overall and on irregular lots; coverage proxy failure rates; the built floor area ratio identity check; land value share by property class, by borough and by land use category; owner name normalization, same-block clustering and cluster classification counts; attribute population rates and current-as-of dates across the 29 Utah Land Information Records counties; and structure counts by occupancy class and by state. Computed August 24, 2026 from sources 2, 5 and 7; MMCG database, 2026. https://mmcganalytics.com/methodology/