Terrain and Slope at National Scale: Screening Buildable Land

Terrain is the one national data layer in American commercial real estate that is essentially finished. At the end of fiscal year 2025, 99% of the nation had baseline elevation data available or in progress meeting the specifications of the U.S. Geological Survey's 3D Elevation Program (U.S. Geological Survey, 2025). Parcel records are a patchwork, zoning is a patchwork, and building attributes are a patchwork. Elevation is not. Anyone can read the ground under any site in the country, for nothing, in seconds.

Which makes the state of national slope screening slightly absurd. The layer is complete, the arithmetic is a division, and the screens built on it are routinely wrong in a way that is not random.

Here is the finding, measured rather than asserted. Four hillside transects were read for this article from the USGS elevation service on 24 August 2026, each a 300 meter line across developed ground in Pittsburgh, Cincinnati, Los Angeles and Seattle, each read from three different 3DEP products covering the identical span. Mean slope was almost invariant to resolution: coarsening from the 1-meter product to the 1 arc-second product moved it by at most 3.9 percentage points across the whole set. Maximum slope was not: it fell by 78%, 71%, 30% and 80% respectively. Averaging a surface can only reduce its steepest gradient, never increase it, so the error introduced by a coarse elevation model has a direction. A national screen built on the 10 meter or 30 meter product does not make random mistakes about steep sites. It systematically passes them.

The consequence is concrete. On the 1-meter surface, 17.7% of the Magnolia Bluff transect in Seattle sits at 25% slope or steeper, which is the gradient at which San Diego's steep hillside regulations attach and the line that county planning guidance in Pennsylvania reports as the common divide between moderately steep and very steep. On the 1 arc-second surface covering the same 300 meters, none of it does. Not less of it. None.

What follows is a method rather than a survey of one: how to build a national slope screen from public data, which product to query at which stage, where the thresholds come from and who published them, how the parcel join changes the answer, and the six ways the screen fails. It is the terrain layer of the wider work on building open analytical atlases from public data, and it assumes throughout that a slope value without a product name, a cell size and a collection year is not a number a credit file can carry.

What a slope screen is actually deciding

Slope is the cheapest early disqualifier in commercial real estate because it is the only physical site constraint that is simultaneously national in coverage, free, quantitative, and directly convertible into a quantity of work. Flood zones tell you about insurance and about a regulatory line. Wetlands tell you about a permitting path whose length nobody can predict. Slope tells you how many cubic yards of dirt stand between the site as it is and the site as the site plan draws it, and it tells you that before anyone has been engaged.

Four distinct exposures sit behind the number. The first is cut and fill. A level pad on sloping ground is a subtraction and an addition, and the geometry is not linear in anything a reader would expect. For a square pad of side L cut into a uniform slope s, with finished grade set at the mean ground elevation so cut balances fill, the cut volume is s multiplied by L cubed, divided by eight. Doubling the slope doubles the earthwork. Doubling the pad multiplies it by eight. That single asymmetry explains why a 400 foot big box footprint on a 5% grade moves 14,815 cubic yards of cut while a 200 foot pad on a 25% grade moves 9,259, and why footprint growth during design is a terrain event as much as a leasing one.

The second exposure is access. Grades on the drive aisles, the loading approach and the accessible route are not a matter of preference. They are fixed by published standards, and they bind on every commercial property in the country. The third is stormwater and erosion: steeper ground concentrates flow, shortens time of concentration, and pushes a site toward detention volume and slope stabilization that a flat parcel would not need. The fourth is regulatory: a large share of American jurisdictions treat land above a stated gradient as land that cannot be counted, disturbed, or built on without a discretionary approval.

Only the first of those four is really about dirt. The other three are about whether the site plan is legal and whether the drainage works. A screen that reports one number, average slope, and stops there answers none of them.

The reason to run the screen first, before ordering anything, is sequencing. The terrain question is answerable in minutes from public data at national scale, the answer is rarely ambiguous at the extremes, and a site that fails it fails before any third party has been engaged. That is the same logic that governs the wider 30 minute pre-term-sheet site screen, and terrain sits alongside flood, wetlands and zoning in the same first pass. What terrain adds that the others do not is a quantity. The flood layer tells you which side of a line the site is on. The terrain layer tells you how much work the site is.

The input stack, and which product answers which question

Every national slope screen in the United States rests on one program. The U.S. Geological Survey's 3D Elevation Program acquires the elevation data, publishes the derived surfaces free of charge and without use restrictions, and distributes them through The National Map. At the end of fiscal year 2025, 99% of the nation had baseline data available or in progress meeting 3DEP specifications (U.S. Geological Survey, 3D Elevation Program, 2025). Earlier readings of the same program, on measures that are not identical, put 3DEP-quality data at 48% of the nation as available or in production in May 2018 and acquisition at 89% national coverage by the end of 2022 (U.S. Geological Survey, 2018 and 2023). Those three figures count different things and should never be strung into one series without saying so. What they establish jointly is that terrain, unlike parcels or zoning, is not a coverage problem any more.

Resolution is the problem. 3DEP is not one surface; it is a ladder of them, and the rungs are not interchangeable. USGS publishes a 1-meter seamless DEM whose production began in mid-2025, project-based 1-meter DEMs that have existed since 2015 and expand as new lidar is acquired, a project-based 1/9 arc-second product covering about 25% of the conterminous United States, and seamless 1/3, 1 and 2 arc-second products. The 1/3 arc-second surface is the one with complete seamless national coverage of the 48 conterminous states, Alaska, Hawaii and the U.S. territories, at approximately 10 meters north to south, with east to west spacing varying by latitude as the meridians converge. The 1 arc-second surface runs at approximately 30 meters and covers the conterminous United States and Alaska. Over Alaska, 2 arc-second and 5-meter interferometric radar products fill in where lidar does not exist (U.S. Geological Survey, About 3DEP Products and Services, read 24 August 2026).

Read the ladder as a count rather than as a distance and the consequence is immediate. An acre is 4,046.86 square meters. On the 1-meter product an acre holds about 4,047 elevation measurements. On the 1/3 arc-second product it holds about 40. On the 1 arc-second product it holds four. Across a 100 foot frontage, which is 30.48 meters, the 1-meter surface gives you thirty samples, the 1/3 arc-second surface gives you three, and the 1 arc-second surface gives you one. A single elevation value cannot describe a grade. It cannot even establish that there is one.

This is the point at which this piece and its companion divide. The companion article on USGS 3DEP terrain data treats the program as a source: what it is, how it is funded, how the products are built and where to get them. What follows here treats it as an input to a method, which is a different question with a different failure surface. The program itself makes the case in dollars: USGS reports that the 3D Nation Study identified potential annual benefits of $7.6 billion for the next generation of 3DEP (U.S. Geological Survey, What is 3DEP, read 24 August 2026). Those benefits accrue only to users who pick the right rung.

Quality level is the specification behind the pixel

A 1-meter DEM does not exist because someone chose a cell size. It exists because a lidar collection met a written specification. The requirements for collections under 3DEP live in the Lidar Base Specification, whose current edition is LBS 2025 rev. A, released in June 2025 (U.S. Geological Survey, National Geospatial Program, 2025). The specification defines quality levels, and the quality levels are what an analyst should quote when asked what a terrain screen rests on.

The current table is short enough to memorize. Quality level 0 requires 5 centimeters of vertical accuracy expressed as root mean square error and at least 8 points per square meter, supporting a 0.5 meter cell. Quality level 1 relaxes vertical accuracy to 10 centimeters at the same density and the same cell. Quality level 2 requires 10 centimeters and at least 2 points per square meter, and supports the 1 meter cell. Quality level 3 requires 20 centimeters and 0.5 points per square meter at a 2 meter cell. Quality level 5 describes interferometric synthetic aperture radar over Alaska at 185 centimeters and a 5 meter cell (U.S. Geological Survey, Topographic Data Quality Levels, read 24 August 2026).

Quality level 2 is the practical national floor, and the reason is a date. USGS states that most lidar collected in 2014 and later meets 3DEP specifications for quality level 2 nominal pulse spacing and vertical accuracy. That sentence is doing more work than it looks. It means that for most of the country, the question is not whether a 1-meter surface can be built, but whether the collection over a given site happened before or after the middle of the last decade, and whether the derived 1-meter product has been published for that project area. Those are two separate questions with two separate answers, and a screen that assumes the second follows from the first will silently fall back to a 10 meter surface without telling anyone.

The ratio between the top and the bottom of the ladder is worth holding in mind when a screen crosses into Alaska. Quality level 5 carries 18.5 times the vertical error of quality level 2. A 185 centimeter root mean square error is roughly six feet. On a 200 foot run, six feet of vertical uncertainty is three percentage points of slope, which is larger than the entire ADA cross slope allowance. Terrain screening in Alaska is a different exercise from terrain screening in Pennsylvania, and the honest way to report that is to name the quality level rather than to report a slope to one decimal place and hope.

How slope comes out of a grid, and what the grid costs you

A digital elevation model is a raster: a regular grid of cells, each holding one elevation. Slope is not stored in it. Slope is derived, cell by cell, and the derivation is rise over run on the grid.

The simplest form is the two-cell difference. Take a cell and its downhill neighbour, subtract the elevations, divide by the distance between the cell centres, and you have a gradient along that axis. Expressed as a percentage it is that ratio times 100; expressed in degrees it is the arctangent of the ratio. Production implementations do something more careful: they fit a plane to the eight cells surrounding the target cell, weighting the four orthogonal neighbours more heavily than the four diagonal ones, then take the steepest descent of that plane. The published third-order finite difference approach is the default in most tools, and a second-order variant that fits a quadratic surface is the common alternative. The choice between them moves a slope value by a fraction of a percentage point on ordinary terrain and matters far less than the choice of cell size. USGS exposes the derivation directly: its dynamic 3DEP elevation service publishes slope, aspect, hillshade and contour functions computed on the fly from the underlying DEM (U.S. Geological Survey, 3DEP Elevation dynamic image service, service description read 24 August 2026).

The cell size is where the analysis actually lives, and the arithmetic of why is worth stating plainly. A DEM cell holds one elevation for the whole cell. Whatever happens inside that footprint has been averaged away before the slope calculation begins. On a 1 meter cell, a retaining wall, a swale, a bench or a 3 foot drop shows up as a step between adjacent cells. On a 10 meter cell, the same feature is one hundredth of the cell area and disappears into the mean. On a 30 meter cell, the cell footprint is 900 square meters, roughly a fifth of an acre, and a whole building pad can sit inside one cell without moving its value.

The consequence has a direction, and this is the part that gets missed. Averaging is a smoothing operation, and smoothing a surface cannot increase its maximum gradient. It can only reduce it or leave it unchanged. So the error introduced by a coarse DEM is not noise around the truth. It is a one-sided bias: the coarser the product, the lower the maximum slope it reports, always. A national screen built on the coarse product does not make random mistakes about steep sites. It systematically passes them.

Mean slope behaves differently, and understanding why makes the whole thing intuitive. Mean slope over a stretch of ground is approximately the total elevation change divided by the total distance, and both of those are anchored at the endpoints. Every product sees the same endpoints. So the mean survives coarsening almost intact while the maximum collapses. Which means the statistic a coarse screen reports most reliably is exactly the one that matters least for cost, and the statistic that drives earthwork, retaining structures and ordinance triggers is the one it degrades most.

The same ground, three products, three answers

The argument above is geometry, so it should be testable, and it is. USGS publishes a dynamic elevation service that mosaics every 3DEP DEM it holds and lets a caller lock a query to a specific source raster by catalogue identifier. That makes a controlled experiment possible: take one line of ground, read it from each product covering that line, sample each product at its own native cell spacing across the identical span, and compare.

Four transects were run for this article on 24 August 2026, each a straight 300 meter line across a developed hillside: the South Side Slopes in Pittsburgh (40.4257 N, 79.9805 W, running north to south), Mount Adams in Cincinnati (39.1088 N, 84.4938 W), Beachwood Canyon in Los Angeles (34.1230 N, 118.3210 W) and Magnolia Bluff in Seattle (47.6446 N, 122.4050 W), the last three running east to west. Each was read from the 1-meter product, the 1/3 arc-second product at 10.31 meters at these latitudes, and the 1 arc-second product at 30.92 meters. Slope is the absolute elevation difference between consecutive samples divided by the spacing between them, which is what a slope raster derived from that product would compute along that axis.

Mean slope barely moves. Across a thirty-fold change in cell size, the mean fell from 14.5% to 11.6% at the South Side Slopes, from 24.8% to 22.6% at Mount Adams, from 32.0% to 28.1% at Beachwood Canyon and from 14.6% to 13.6% at Magnolia Bluff. The largest movement anywhere in the set was 3.9 percentage points. If mean slope is the number a screen reports, resolution looks like a detail.

Maximum slope is a different story. At the South Side Slopes it fell from 206.0% on the 1-meter surface to 45.6% on the 1 arc-second surface, a drop of 78%. At Magnolia Bluff it fell from 116.0% to 23.5%, a drop of 80%. At Mount Adams the fall was 71% and at Beachwood Canyon, the steepest and most uniform of the four, it was still 30%. Every site fell, and every site fell monotonically as the cell grew, which is what the smoothing argument predicts and what the one-sided nature of the error requires.

The threshold consequence is the part a lender should care about. Take 25%, the gradient at which San Diego's environmentally sensitive lands regulations attach and the line Chester County reports as the common divide between moderately steep and very steep. On the 1-meter surface, 17.7% of the Magnolia Bluff transect sits at 25% or steeper. On the 1 arc-second surface, none of it does. The coarse product does not report the bluff as marginal. It reports it as absent. That is the finding this article is built on: a national slope screen running on the coarse product is not noisy about steep sites, it is systematically permissive about them, and the direction of the error is knowable in advance from the arithmetic of averaging.

Two honest caveats belong with those numbers. First, a maximum computed cell to cell on a 1-meter surface will pick up walls, cut faces, steps and benches, and some of the three-digit maxima above are exactly that: real vertical features on a developed hillside, correctly measured, that no engineer would call a slope. That is an argument for reading the distribution rather than the single peak, not an argument for the coarse product, which reports neither. Second, the products differ in vintage as well as in resolution. At Beachwood Canyon the service holds two 1-meter collections at the same coordinate, from 2016 and from 2023, and they disagree: mean slope of 30.8% against 32.0% and maximum slope of 124.0% against 108.0% on the identical line. Same resolution, same ground, seven years apart, and a 16 percentage point difference in the maximum. Resolution is the first question to ask of a terrain reading. Vintage is the second, and it is not far behind.

Thresholds, and where every one of them comes from

A slope screen is worthless without a threshold, and thresholds are where terrain analysis most often stops being analysis. The literature of site selection is full of round numbers presented without provenance: 5% is easy, 10% is workable, 15% is difficult, 25% is out. Some of those numbers are close to the ones in the documents. None of them, as usually quoted, has a document behind it. The discipline this piece argues for is simple: every threshold in a screen carries a named source, and thresholds from different kinds of source are never mixed into one scale.

There are two kinds. The first is federal, applies everywhere, and is written as a ratio. Under the ADA Accessibility Standards, the running slope of a walking surface on an accessible route shall not be steeper than 1:20 and the cross slope shall not be steeper than 1:48 (section 403.3). A ramp run shall not be steeper than 1:12 (section 405.2), and the standards define a ramp as a walking surface steeper than 1:20, which is what makes 1:20 the boundary rather than a preference. In existing sites, buildings and facilities, Table 405.2 permits slopes steeper than 1:12 only where space limitations require it, capping a run at 1:10 for a 6 inch rise and 1:8 for a 3 inch rise, with the table legend stating that a slope steeper than 1:8 is prohibited. Parking spaces and the access aisles serving them permit slopes no steeper than 1:48 in any direction (section 502.4), and counter slopes of gutters and road surfaces immediately adjacent to a curb ramp shall not be steeper than 1:20 (section 406.2). Converted, that set reads 5%, 2.08%, 8.33%, 12.5% and 2.08%.

The second kind is local, applies only where adopted, and is written as a percentage. The City of Summit, New Jersey requires every lot proposed for subdivision, site plan or construction permit to be evaluated for the presence of slopes greater than 15%, directs that platting and siting avoid disturbance of slopes of 15% or greater, requires each platted lot to hold at least 25% of its area in a non-steep-slope condition, imposes a 10 foot separation between any structure foundation and the steep slope area, and provides that proposed finished grades shall not exceed 3 to 1, which is 33.3% (City of Summit Land Development Ordinance, Article 9, sections 35-9.3 and 35-9.6, as amended by Ordinance 09-2870). San Diego takes a different cut: its Environmentally Sensitive Lands Regulations attach where a site contains a natural gradient of at least 25%, stated in the guidelines as 25 feet of vertical distance for every 100 feet of horizontal distance, together with a vertical elevation of at least 50 feet, or a gradient of at least 200% with a vertical elevation of at least 10 feet (San Diego Municipal Code section 143.0110, as explained in the City of San Diego Land Development Manual Steep Hillside Guidelines).

Those two ordinances are not outliers. Chester County, Pennsylvania, whose planning commission publishes model guidance for its municipalities, reports that steep slope percentage thresholds for conservation vary by municipality but generally begin at 15%, that many municipalities distinguish moderately steep slopes of 15% to 25% from very steep slopes above 25%, and that stricter limits may apply where geographic or geological constraints justify them (Chester County Planning Commission, Steep Slope Protection, read 24 August 2026). The pattern nationally is 15% and 25%, adopted locally, defined locally, and measured by whatever method the local ordinance names.

Two disciplines follow. First, never convert between percent and degrees implicitly. A 25% grade is 14.04 degrees, and a screen that reads 25 from a degrees raster and compares it to a percent threshold has just declared a 46.6% slope acceptable. Second, do not treat the access standards and the ordinance thresholds as points on one scale. The ADA numbers govern surfaces that will be built, after grading. The ordinance numbers govern ground as it exists, before grading. A site with 30% natural slope can carry a compliant 2% parking field once it has been cut, at a price in earthwork that the first number tells you nothing about and the geometry in this article's opening section tells you exactly.

The screen, stage by stage

A national slope screen that survives scrutiny runs in four stages, and each stage answers one question before the next one is allowed to start.

Stage one, resolve the geometry. Slope is computed on ground, and ground is identified by a parcel boundary or by a site polygon. Before any elevation is read, the analyst needs a boundary in a projected coordinate system whose units are meters or feet, not degrees. Screens that sample in geographic coordinates and then divide by a degree of longitude produce slope values that are wrong by a factor that changes with latitude, and the error is invisible because the output still looks like a number. Where the boundary comes from is the subject of the companion work on the openness and quality of parcel records by state, and the geometry itself is only as good as that record.

Stage two, establish what product is under the site. This is the stage most screens skip, and it is the one that determines whether the answer means anything. USGS publishes a dynamic elevation service that mosaics all 3DEP DEM data and answers point and small-area queries directly. Its catalogue can be interrogated: an identify request at a coordinate returns not only the elevation but the list of source rasters covering that point, each with its name and its native pixel size. At a single point in Pittsburgh, that query returns four sources: a 1 meter DEM from the Western Pennsylvania 2019 project, a 1/9 arc-second product derived from a 2006 southwest Pennsylvania collection at 3.44 meters, and 1/3 and 1 arc-second seamless tiles at 10.31 and 30.92 meters respectively. The service reports that the data it holds reflects all 3DEP DEM data published as of 20 July 2026. A screen that records which of those four it used has provenance. A screen that takes the best available value and reports a slope has an answer with no address.

Stage three, read the elevations. For a small number of points, a point query is the right instrument. USGS operates an Elevation Point Query Service for exactly this purpose, though it should be treated as a service with availability rather than as a guarantee: on 24 August 2026 it returned a gateway timeout to every request made in the course of this research, and the dynamic elevation service answered instead. For a transect or a parcel, a multipoint sample request against a specific source raster is both faster and more honest, because it lets the analyst pin the product rather than accept whatever the mosaic prefers. Requests should be small. Batches of twenty to forty points answer reliably; batches of four hundred returned gateway timeouts during this work. None of this requires downloading an elevation raster, and for a national screen it should not: the point of a service is that the archive stays where it is.

Stage four, compute and report a distribution, not a number. Slope over a parcel is a distribution, and the useful summary is at least three statistics: the area-weighted mean, the maximum, and the share of the parcel above whatever local threshold applies. Report each of them with the product they came from and the vintage of that product. Where the mean and the maximum diverge sharply, the site is benched or terraced and the mean is meaningless. Where they converge, the site is a plane and one number will do. That is a screening output a credit file can carry, and it is the discipline the companion piece on the provenance standard argues for across every layer.

Escalate to the 1-meter product whenever the screen is near a threshold, whenever the parcel is under about five acres, and whenever the decision in front of the reader is a quantity rather than a yes or no. The coarse products are for triage across a portfolio or a market. They are not for a site.

The parcel join, and what it does to the number

A slope raster is national and continuous. A parcel is local, discrete and defined by a record whose quality varies by county. Joining the two is where a terrain atlas becomes a terrain screen, and it introduces four errors that have nothing to do with elevation.

The first is boundary accuracy. Assessor parcel geometry is a cadastral representation, not a survey, and the offsets between the mapped boundary and the surveyed line are routinely on the order of feet and occasionally tens of feet. On flat ground that error is harmless. On a 30% slope, a 20 foot horizontal offset moves the sampled elevation by 6 feet, and a screen clipping to the mapped boundary can pick up or drop an entire bench. The state-by-state picture of what parcel records exist and how good they are is documented in the National States Geographic Information Council's Geospatial Maturity Assessment: in the 2023 cycle, 98% of the 44 reporting states had GIS parcels covering 80% to 100% of their counties, 22 published 80% to 100% of parcels publicly, and five kept them internal (NSGIC, 2023 Geospatial Maturity Assessment). The 2025 cycle, published 14 August 2026, updates that picture and is tabulated in detail in the companion article.

The second is edge effects. Slope at a cell is computed from its neighbours, so slope inside a parcel depends on elevations outside it. Clip the DEM to the parcel and the boundary cells lose neighbours, and the slope values on the ring closest to the boundary are computed from a truncated window. The fix is trivial and routinely omitted: sample a buffer beyond the parcel, compute slope on the buffered surface, then clip the slope raster rather than the elevation raster. San Diego wrote the same logic into its ordinance from the other direction, requiring an off-site analysis of adjacent property where a site contains steep hillsides but does not itself reach 50 feet of vertical elevation, because the slope system does not stop at the property line even when the record does.

The third is the choice of statistic. Mean parcel slope answers a question about the whole parcel. Cost is driven by the pad, which is a fraction of the parcel, and by the maximum within it. A 10 acre parcel with a 3% mean slope and a 40% ravine across one corner is a good site with a lost corner; the same parcel with 3% mean and 8% maximum is a good site. Mean alone cannot tell them apart. The buildable envelope work makes the same point from the code side: what matters is which constraint binds, and the companion piece on estimating buildable area from a parcel carries the setback, coverage, height and floor area ratio arithmetic that turns a terrain-adjusted developable area into a footprint.

The fourth is the other layers. Slope is one subtraction among several, and the order matters less than the fact that they overlap. FEMA's National Flood Hazard Layer covers more than 90% of the United States population (FEMA, page last updated 3 April 2025), and the Fish and Wildlife Service's Wetlands Data Layer holds more than 37 million wetland and deepwater features and is released twice a year (USFWS, National Wetlands Inventory, 2026). The flood layer, the wetlands layer and the terrain layer frequently remove the same acre, and a screen that sums their areas will double count. This is exactly the geometry the wetlands constraint map handles at national scale, and the intersection is the reason a serious screen computes a single unconstrained polygon rather than four independent percentages. Land-metric work on lot size, coverage and assembly patterns builds on the same joined geometry, as the study of parcel-derived land metrics sets out.

The cross-check the soil survey gives you free

Every national screen needs a second opinion that does not share the first one's failure modes, and for slope that second opinion already exists. The Soil Survey Geographic Database carries, for every soil map unit in the country, an acre-weighted average slope gradient in percent. It was not derived from a DEM. It came from field mapping and photointerpretation, on a different schedule, by a different agency, with different errors. That makes it genuinely independent, which is rare enough in public data to be worth using.

The Department of Agriculture's Natural Resources Conservation Service publishes it through Soil Data Access, which answers structured queries without any download. Joining the map unit aggregated attribute table to map unit acreage and banding the result produces a slope profile for a survey area in a single request. Run for four survey areas covering the sites used in this article, the profiles differ more than most readers would guess (MMCG tabulation from USDA NRCS Soil Data Access, queried 24 August 2026).

Allegheny County, Pennsylvania holds 45.2% of its mapped soil acres at 15% slope or steeper and 22.7% above 40%, which is to say that nearly a quarter of the county containing Pittsburgh is mapped as ground where a level pad is an earthwork project before it is anything else. Hamilton County, Ohio comes in at 24.7% above 15%, with more than half its acreage in the gentle 3% to 8% band. The southeastern Los Angeles County survey, which covers the basin and the hills behind it, reads 61.1% in the flattest band and 14.0% above 15%, the signature of a flat plain with steep edges. The King County Area survey in Washington puts 47.7% of its acres in the 8% to 15% band, ground that is neither flat nor regulated as steep, which is the most operationally awkward category there is.

Two cautions govern the use of this layer. It is a map unit attribute, so it describes a polygon of mapped soil, not a point on a parcel, and its resolution is the soil map's, not a DEM's. And it excludes what it does not classify: across the four survey areas, 32,979 acres carry no slope value at all, mostly water and made land, and made land is precisely where commercial development happens. Used as a cross-check, it is excellent. Used as the screen, it would be a category error.

The value of the cross-check is diagnostic. Where the DEM-derived profile and the soils profile agree, the terrain reading is probably sound. Where they diverge sharply, something has changed on the ground since one of the two was made, and that is a signal to look rather than a reason to pick a winner.

Failure modes, ranked by how often they bite

One, resolution mismatch. Covered above and first for a reason: it is the only failure mode with a known direction. Coarse products understate maximum slope, so a coarse screen passes steep sites rather than failing flat ones. Every other error on this list is at least symmetric.

Two, vintage against reality. A DEM records the ground on the day the sensor flew. The 1 meter product under the Pittsburgh site used here comes from a 2019 collection; the 1/9 arc-second product at the same coordinate comes from a 2006 one. Anything graded, filled, quarried, benched or built since then is invisible to the surface and present on the site. For infill and redevelopment, where the site has almost certainly been disturbed, elevation vintage is the single most under-reported caveat in terrain analysis. The fix is to state the collection year of the product used, every time, and to treat a screen on a site that has been graded since collection as a screen on a site that no longer exists.

Three, hydro-flattening and the artifacts of a bare-earth surface. The Lidar Base Specification requires that waterbodies of 0.8 hectare, about 2 acres, or larger be flattened to a single water surface elevation, that streams and rivers of 30 meters nominal width or greater be flattened bank to bank, that permanent islands of 0.4 hectare or larger be delineated, and that bridges in all their forms be removed from the bare-earth surface, with stream channels breaking at culverts while the roadway over the culvert stays intact. It further requires that bare-earth points close to any breakline be reclassified as ignored ground and excluded from DEM generation, at a proximity threshold left to the producer but generally no more than twice the nominal pulse spacing (U.S. Geological Survey, Lidar Base Specification 2025 rev. A, DEM Surface Treatments). Every one of those rules is correct for hydrology and wrong for a site screen that happens to sit on a stream bank: the DEM shows a flat water surface where a real channel with real banks exists, no bridge where a bridge is, and a thin ring of interpolated ground along every breakline.

Four, the parcel edge. Slope computed on a clipped elevation raster is wrong on the boundary ring, and the boundary ring is where retaining walls, access drives and setback encroachments live.

Five, units and datum. Percent against degrees is the classic. Vertical datum is the quieter one: 3DEP products are published in a defined vertical reference, and mixing them with a site survey on a local datum produces elevation differences that are real numbers describing nothing.

Six, the mean. A single average slope for a parcel is the statistic most likely to be reported and least likely to answer the question in front of the reader.

Where the screen stops, and who takes it from there

A national slope screen is a triage instrument. It is reproducible, cheap, national, and dated, and those four properties are exactly what make it useful early and insufficient late. The boundary is not a matter of professional caution. It is written into the ordinances themselves.

Summit, New Jersey requires a steep slope disturbance plan drawn at a scale between one inch to thirty feet and one inch to ten feet, showing existing and proposed topography at two-foot contour intervals, and states that topographic information can only be supplied by a licensed land surveyor registered in the State of New Jersey. No DEM satisfies that requirement, and no reading of a DEM ever will. The 1 meter product carries 10 centimeters of vertical root mean square error under quality level 2, which is about four inches, against a two-foot contour interval. It is in the right range to be suggestive and nowhere near the right range to be a submission.

The same boundary applies to everything downstream of the grade. Slope stability, bearing capacity, the presence of rock, the depth to groundwater and the behavior of fill are geotechnical questions that no elevation surface addresses at any resolution. A screen can tell you that a site has 28% maximum slope on a 2019 surface and that the local ordinance attaches at 25%. It cannot tell you whether the slope is stable, what the cut will expose, or what the jurisdiction will approve. MMCG Analytics supplies the data and the analytics, with source and vintage carried on every displayed value; the credit decision and the professional judgments behind it rest with the lender and the licensed professionals it engages.

What the screen buys is sequence. It tells a lender, before a term sheet, that a site is worth the cost of the work that comes next, or that it is not. It tells a developer which of five candidate parcels to survey first. It tells a credit analyst reviewing a construction budget whether the earthwork line is plausible against the geometry of the site, which is a question the analyst can otherwise only answer by trusting the sponsor. Terrain sits in the platform alongside parcels, flood, wetlands, wind risk, traffic and demographics as one of the public-record layers, and its job in that stack is to be the fastest honest no.

For asset classes where the grade drives the whole program, the screen stops being triage and becomes a primary filter. Large-footprint industrial, distribution and data centers, whose power, water and fiber constraints in public records already narrow the candidate set severely, are pad-driven by definition: the cube law in the opening section means a 400 foot by 400 foot pad on a 5% grade is already a 14,815 cubic yard earthwork before anything else is considered. In those markets the terrain layer is not a screen at all. It is a site selection criterion with a threshold set by the program rather than by an ordinance.

Frequently asked questions

How do you build a slope map for development screening?

Resolve the site boundary in a projected coordinate system whose units are meters or feet, then establish which USGS 3DEP product covers the site and at what native cell size, then read elevations from that specific product rather than from a best available mosaic, then compute slope as rise over run on the grid and report a distribution: area-weighted mean, maximum, and the share of the site above the local threshold. Each of those four steps has to be recorded, because a slope value without a product name and a collection year cannot be checked by anyone. USGS distributes the elevation data free of charge and without use restrictions, and its dynamic elevation service answers small point and multipoint queries directly, so no elevation raster needs to be downloaded for a screen.

What slope is too steep to build on?

There is no national answer, and any source that gives one without a citation is inventing it. Local ordinances set the operative thresholds and they cluster at 15% and 25%. The City of Summit, New Jersey requires evaluation of slopes greater than 15% and directs that development avoid disturbance of slopes of 15% or greater. San Diego attaches its Environmentally Sensitive Lands Regulations where a natural gradient of at least 25% coincides with at least 50 feet of vertical elevation. The Chester County Planning Commission reports that municipal thresholds generally begin at 15%, with 15% to 25% treated as moderately steep and above 25% as very steep. Separately, the ADA Accessibility Standards fix built grades everywhere in the country: 1:20 for a walking surface, 1:12 for a ramp run, 1:48 for cross slopes and for parking spaces and access aisles.

Where can I get free slope data for a property in the United States?

The U.S. Geological Survey's 3D Elevation Program is the source, and it is free of charge and without use restrictions. Elevation products are distributed through The National Map, and USGS also runs a dynamic 3DEP elevation image service that publishes slope, aspect, hillshade and contour functions computed on the fly, plus an Elevation Point Query Service for single points. At the end of fiscal year 2025, 99% of the nation had baseline data available or in progress meeting 3DEP specifications. What is free is the elevation surface; the analysis, the threshold sourcing and the parcel join are the work.

What DEM resolution do I need for site screening?

Use the 1-meter product for anything site-specific and treat 1/3 arc-second, at roughly 10 meters, as the coarsest acceptable input for portfolio triage. The reason is arithmetic rather than preference: an acre holds about 4,047 cells on a 1-meter surface, about 40 on a 1/3 arc-second surface and about four on a 1 arc-second surface, and across a 100 foot frontage the 1 arc-second product supplies a single elevation. Averaging is a smoothing operation, so a coarser cell can only reduce the maximum slope it reports, never increase it. That makes the error one-sided: a screen run on the coarse product systematically passes steep sites.

Is slope measured in percent or in degrees?

Both are in use, and mixing them is one of the most common silent failures in terrain screening. Percent grade is rise divided by run times 100. Degrees is the arctangent of the same ratio. A 25% grade is 14.04 degrees, a 15% grade is 8.53 degrees, and a 1:12 ramp at 8.33% is 4.76 degrees. A screen that reads a value from a slope raster in degrees and compares it to a percent threshold of 25 has just accepted a 46.6% slope. Every threshold, every raster and every reported value should carry its unit explicitly.

How accurate is USGS 3DEP elevation data?

Accuracy is specified by quality level rather than by product name. Quality level 2, which most lidar collected in 2014 and later meets, requires vertical accuracy of 10 centimeters expressed as root mean square error and at least 2 points per square meter, and supports a 1 meter DEM cell. Quality level 1 and quality level 0 tighten density to at least 8 points per square meter and, for quality level 0, vertical accuracy to 5 centimeters, both supporting a 0.5 meter cell. Quality level 3 allows 20 centimeters at a 2 meter cell. Quality level 5, the interferometric radar product over Alaska, carries 185 centimeters at a 5 meter cell, which is 18.5 times the quality level 2 vertical error and changes what a screen can claim there.

Can a public slope screen replace a topographic survey?

No, and the ordinances say so directly. Summit, New Jersey requires a steep slope disturbance plan showing existing and proposed topography at two-foot contour intervals and states that topographic information can only be supplied by a licensed land surveyor registered in the state. A 1 meter DEM built to quality level 2 carries about four inches of vertical root mean square error against that two-foot interval, which is enough to inform a decision about whether to proceed and not enough to be a submission. Slope stability, bearing capacity, rock and groundwater are geotechnical questions that no elevation surface answers at any resolution.

Should a parcel slope screen report average slope or maximum slope?

Both, plus the share of the parcel above the applicable threshold, because they answer different questions and they behave differently under coarsening. Mean slope is anchored by the elevations at the ends of a run, so it survives a coarse DEM nearly intact. Maximum slope is the statistic that drives cut and fill volume, retaining structures and ordinance triggers, and it is the statistic a coarse DEM degrades most. A parcel with a 3% mean and a 40% ravine in one corner and a parcel with a 3% mean and an 8% maximum are different sites, and mean slope alone cannot distinguish them.

Sources

1. U.S. Geological Survey, 3D Elevation Program, About 3DEP Products and Services, read 24 August 2026: the 1-meter seamless DEM with production from mid-2025, project-based 1-meter DEMs since 2015, the 1/9 arc-second product covering about 25 percent of the conterminous United States, the 1/3 arc-second product at approximately 10 meters north to south with complete seamless national coverage, the 1 arc-second product at approximately 30 meters, the 2 arc-second and 5-meter IfSAR products over Alaska, and the statement that most lidar collected in 2014 and later meets 3DEP specifications for quality level 2. https://www.usgs.gov/3d-elevation-program/about-3dep-products-services
2. U.S. Geological Survey, 3D Elevation Program, What is 3DEP, read 24 August 2026: 99 percent of the nation with baseline data available or in progress meeting 3DEP specifications at the end of fiscal year 2025, and the 3D Nation Study finding of potential annual benefits of $7.6 billion for the next generation of 3DEP. https://www.usgs.gov/3d-elevation-program/what-3dep
3. U.S. Geological Survey, 3D Elevation Program, program page, 2026: elevation data acquired through partnership funding and distributed free of charge and without use restrictions. https://www.usgs.gov/3d-elevation-program
4. U.S. Geological Survey, 3D Elevation Program, Topographic Data Quality Levels, read 24 August 2026: quality levels 0, 1, 2, 3 and 5 with vertical accuracy expressed as RMSEz, aggregate nominal pulse density and the DEM cell size each supports. https://www.usgs.gov/3d-elevation-program/topographic-data-quality-levels-qls
5. U.S. Geological Survey, National Geospatial Program, Lidar Base Specification Online, 2026: the specification is the source of requirements for collections under 3DEP, current edition LBS 2025 rev. A released June 2025, with the version history back to 2018. https://www.usgs.gov/ngp-standards-and-specifications/lidar-base-specification-online
6. U.S. Geological Survey, Lidar Base Specification 2025 rev. A, Digital Elevation Model Surface Treatments, read 24 August 2026: hydroflattening of waterbodies of 0.8 hectare or greater and streams and rivers of 30 meters nominal width or greater, delineation of permanent islands of 0.4 hectare or larger, removal of bridges from the bare-earth surface, the culvert rule, and the reclassification of bare-earth points near breaklines as ignored ground class 20. https://www.usgs.gov/core-science-systems/ngp/ss/lidar-base-specification-digital-elevation-model-surface-treatments
7. U.S. Geological Survey, 3DEP Elevation dynamic image service, service description read 24 August 2026: a multi-resolution bare-earth DEM service with hillshade, aspect, slope and contour functions, WMS and WCS interfaces, and data reflecting all 3DEP DEM data published as of 20 July 2026. https://elevation.nationalmap.gov/arcgis/rest/services/3DEPElevation/ImageServer
8. U.S. Geological Survey, The National Map Downloader, read 24 August 2026: the distribution point for 3DEP elevation products and lidar point clouds. https://apps.nationalmap.gov/downloader/
9. U.S. Geological Survey, Elevation Point Query Service, version 1 JSON endpoint, read 24 August 2026: the single-point elevation service, which returned a gateway timeout to every request made during this research on that date. https://epqs.nationalmap.gov/v1/json
10. United States Access Board, ADA Accessibility Standards, read 24 August 2026: section 403.3 (running slope of walking surfaces not steeper than 1:20, cross slope not steeper than 1:48), section 405.2 with Table 405.2 (ramp runs not steeper than 1:12, existing site allowances to 1:10 and 1:8, and the prohibition on slopes steeper than 1:8), section 405.3 (ramp cross slope not steeper than 1:48), section 406.2 (counter slopes of adjoining gutters and road surfaces not steeper than 1:20) and section 502.4 (parking spaces and access aisles, slopes not steeper than 1:48). https://www.access-board.gov/ada/
11. City of Summit, New Jersey, Land Development Ordinance, Article 9, Regulations for Development Within Steep Slope Areas, sections 35-9.1 through 35-9.8, as amended by Ordinance 09-2870 (2009), read 24 August 2026: evaluation of slopes greater than 15 percent, the requirement that development avoid disturbance of slopes of 15 percent or greater, the requirement that each platted lot hold at least 25 percent of its area in a non-steep-slope condition, the 10 foot foundation separation, the 3 to 1 cap on proposed finished grades, the 1,000 square foot de minimis area, and the requirement for two-foot contour topography supplied by a licensed New Jersey land surveyor. https://www.cityofsummit.org/DocumentCenter/View/121
12. City of San Diego, Development Services Department, Land Development Manual, Steep Hillside Guidelines, read 24 August 2026: San Diego Municipal Code section 143.0110 Environmentally Sensitive Lands Regulations apply where a site contains a natural gradient of at least 25 percent with a vertical elevation of at least 50 feet, or a gradient of at least 200 percent with a vertical elevation of at least 10 feet, with vertical elevation measured across the entire slope system and an off-site analysis required where the 50 foot elevation is not met on site. https://www.sandiego.gov/sites/default/files/legacy/development-services/pdf/industry/landdevmanual/ldmsteephillsides.pdf
13. Chester County Planning Commission, Planning eTools, Steep Slope Protection, read 24 August 2026: municipal steep slope thresholds generally begin at 15 percent, with 15 to 25 percent treated as moderately steep and slopes greater than 25 percent treated as very steep, prohibitive or severely steep. https://www.chescoplanning.org/MuniCorner/eTools/43-SteepSlopes.cfm
14. U.S. Department of Agriculture, Natural Resources Conservation Service, Soil Data Access, SSURGO map unit aggregated attribute table joined to map unit acreage, queried 24 August 2026 for soil survey areas PA003 (Allegheny County, Pennsylvania), OH061 (Hamilton County, Ohio), CA696 (Los Angeles County, California, Southeastern Part) and WA633 (King County Area, Washington). https://sdmdataaccess.nrcs.usda.gov/
15. U.S. Department of Agriculture, Natural Resources Conservation Service, Web Soil Survey, read 24 August 2026: the public interface to the Soil Survey Geographic Database. https://websoilsurvey.nrcs.usda.gov/
16. National States Geographic Information Council, 2023 Geospatial Maturity Assessment, December 2023: 44 states reporting, 98 percent with GIS parcels covering 80 to 100 percent of their counties, 22 publishing 80 to 100 percent of parcels and 5 keeping them internal. https://nsgic.org/wp-content/uploads/2024/02/2023GMAReportFinal.pdf
17. National States Geographic Information Council, 2025 Geospatial Maturity Assessment, Full Report, published 14 August 2026 from results collected in the second half of 2025. https://nsgic.org/wp-content/uploads/2026/08/2025-GMA-Full-Report-20260814.pdf
18. Federal Emergency Management Agency, National Flood Hazard Layer, page last updated 3 April 2025: digital flood hazard data covering more than 90 percent of the United States population. https://www.fema.gov/flood-maps/national-flood-hazard-layer
19. U.S. Fish and Wildlife Service, National Wetlands Inventory, Wetlands Data Layer, 2026: more than 37 million wetland and deepwater features, released twice a year. https://www.fws.gov/program/national-wetlands-inventory/wetlands-data
20. MMCG Research, National Terrain Screening Series, 2026: the four transect readings and their per-product slope statistics computed from the USGS 3DEP dynamic elevation service on 24 August 2026; the SSURGO slope band shares for the four soil survey areas; the cells-per-acre, samples-per-frontage and degree conversions; and the balanced cut-and-fill pad geometry. All computed 24 August 2026 from sources 1 to 19; MMCG database, 2026. https://mmcganalytics.com/methodology/