A Buyer's Guide: How to Choose a US Building Footprint Dataset

On this page

1. The dataset in numbers
2. What does "building" mean in this dataset?
3. How the options compare at a glance
4. Why precise building shapes matter
5. Coverage: every county, and a measured gap in every state compared
6. What's in a record
7. Accuracy: what we publish, and what we don't
8. Who buys it, and for what
9. What you can do with each license
10. The value of a finished building dataset
11. How to buy
12. How can our data help your project?

A dataset’s value goes beyond its footprint count: what can you reliably do with each building, parcel, and address?

This guide compares public US datasets with EarthDefine across five practical questions:

• How are buildings split?
• Are heights measured or estimated?
• What attributes come with each polygon?
• How current is the data?
• What does the license allow you to build?

It highlights where EarthDefine’s parcel-level approach and 3D attributes add value, where public datasets may be sufficient, and the cost and engineering effort of combining separate public sources into a building-footprint layer.

For details on how EarthDefine’s data is built, verified, and delivered, see the Building Footprints 3D technical specification.

Data as of June 30, 2026, refreshed quarterly.

EarthDefine Building Footprints is the largest US dataset we have identified, public or private, with a published count. Against the largest public alternative, Overture Maps, the margin is over 35 million structures on the same counting basis.

EarthDefine’s footprints exist at two levels: a structure [one contiguous physical building mass], and a building [that same structure split by parcel boundary]. Microsoft’s US Building Footprints, Overture Maps’ Buildings theme, and FEMA’s USA Structures generally provide structures/footprints rather than EarthDefine’s parcel-split building records. A connected townhome row or strip mall may therefore be represented as a single or fewer footprint records rather than one record per parcel. Overture also provides building-part features, but these represent building components rather than EarthDefine’s parcel-based subdivision. So a townhome row in their dataset will show up as one building spanning several parcels.

EarthDefine reports both structures and buildings because the latter are split at parcel boundaries.

• Use the structure count (193,362,521) when comparing against Microsoft, Overture, or FEMA.
• Use the building count (which are split by parcel boundaries; 202,285,763), when working with the delivered parcel-split records.

EarthDefine against three public US datasets: Microsoft US Building Footprints, Overture Maps Buildings, and FEMA USA Structures.

To assess the shape and fidelity of EarthDefine's data and to compare them with one of the alternatives, use the layer viewer below.

For some projects, a simplified outline is enough. If you’re counting buildings, checking whether a parcel has a structure, or assigning a rooftop point, you may not need an outline that follows the roof. In some cases, a centroid will do.

Precise outlines matter when small differences in location or shape affect the result. A simplified polygon can misstate distances, overlaps, or usable area, changing downstream calculations. For example:

• Rooftop solar: Panel layout and energy estimates depend on usable roof area and shape. A rectangle drawn around a cross-gabled house may include space where panels cannot fit.
• Insurance assessments: Whether a building overlaps a mapped flood zone, or how close it is to trees, wildfire fuels, or neighboring buildings, can affect an assessment.
• Roof replacement and claims estimates: Material quantities depend on the roof’s area and shape.
• Municipal stormwater programs: Parcel-level estimates of building and other impervious area can change when building outlines are inaccurate.

For these applications, polygon accuracy is a data requirement. A missing building is often easy to spot; an inaccurate outline can be harder to catch because it may still produce a plausible measurement. In the accuracy section below, we explain how public datasets create their outlines and how EarthDefine produces and reviews its own.

EarthDefine covers 3,227 US counties, rural and urban, in every state and territory¹. We recommend that you test any coverage claims on every state, not just the national total. Nationally, the gaps run to tens of millions.

Below, we have provided a comprehensive comparison table (based on structures, since that is the unit available in these datasets).

Microsoft also publishes a separate machine-learning footprint dataset covering geographies worldwide. This guide compares the dedicated US Building Footprints release, not Microsoft’s Global ML Building Footprints dataset, which has different licensing.

¹ Remote interior Alaska outside populated areas is partially covered, and sites where the federal government restricts aerial imagery may carry older or missing data.

Every EarthDefine building comes as a polygon plus one or more address points that join on the same identifiers. Below is the point schema from the current data dictionary, grouped by what it’s for. Note: (a) 3D fields are available with the 3D license tier; the 2D footprint license does not include these attributes, (b) the parcel and land-use fields derive from Regrid parcel data.

Polygon schema: str_UUID, bld_UUID, Area_SqFt. A contiguous structure keeps one str_UUID; each parcel-split building under it gets its own bld_UUID. This allows you to work at either level without losing the link between them.

On a parcel with a single building, our addresses typically match what a good parcel dataset would already give you. A parcel-only approach can be insufficient when multiple structures occupy the same parcel because it assigns every structure the same address. EarthDefine addresses each building individually, so multiple buildings on one parcel can carry different addresses. We also provide addresses where no parcel exists at all.

The underlying sources have different specifications, so we do not combine them into a single product-wide accuracy number. Here is what we publish instead, and what you can get for any area of interest.

• Independent production and review. In many cities, EarthDefine starts from the municipality’s own planimetric building dataset, then reviews and edits it against newer aerial imagery, adding buildings the source missed and removing ones that no longer exist. Elsewhere, footprints come from EarthDefine’s own processing of public LiDAR and aerial imagery. EarthDefine does not use Microsoft, OpenStreetMap, Overture, or FEMA footprint geometry as an input to its own footprint layer. Overture’s Buildings theme, by contrast, incorporates multiple source datasets.
• Height accuracy. Heights are measured from LiDAR point clouds rather than estimated from imagery, and delivered as multiple values per building (maximum, mean, maximum-with-clutter) with ground elevations and roof slopes alongside. We do not publish a single product-wide height-accuracy figure; the published vertical specification of the underlying LiDAR collect is available for any area of interest.
• Positional accuracy. Footprints derive from three sources (NAIP imagery, state orthoimagery, LiDAR), each with its own published horizontal specification, so we do not quote a single blended figure. The source specification for any area of interest is available on request.
• Address confidence. Every address point carries Add_Conf, a low, medium, or high confidence score based on the number of corroborating sources and the point’s positional relationship to the building.
• Human review. Output is not shipped straight from the model. Record counts are reconciled at every production stage; each state passes a release checklist (id consistency, no null addresses, elevation sanity, source-date verification, tile-edge dissolve, spot checks against independent map services); counties are reviewed one by one with named-analyst sign-off; and the review work itself is audited, with errors classified as omissions, commissions, incomplete annotations, or mislabeling and worked examples issued back each round.
• Footprint accuracy. On our own evaluation set we measure a mean Intersection over Union (IoU) of 0.91 against manually digitized reference outlines. Microsoft publishes 0.86 for its US Building Footprints. The two figures come from different samples against different references, so read each as a provider’s account of its own data.
• Shape fidelity. Machine extraction from imagery often includes a regularization step that favors dominant orientations and right angles (Microsoft describes its polygonization as imposing “a priori building properties” and reports dominant-angle rotation error). Simplified building outlines can miss wings, bays, and inside corners. Our outlines follow the roof edges more closely, preserving these details for measurements and spatial analysis.

• Insurance and property risk teams need roof pitch, height, and footprint measurements for each insured parcel, along with nearby grade information for assessing flood exposure. Relevant fields include MeanSlope, MaxSlope, MaxHeight, LAG, HAG, and the parcel-level building splits identified by bld_UUID.
• Property intelligence and real estate teams need current building records that can be queried by address, with one building associated with each parcel. Relevant fields include bld_UUID, ll_uuid, Address, Add_Conf, and SourceDate.
• Telecom and RF planning teams need measured building heights and volumes to model signal propagation and line of sight. Relevant fields include MaxHeight, MeanHeight, Volume, and MeanElev.
• Municipal, stormwater, and planning teams need building area by parcel, land-use information, and complete coverage across the jurisdiction. Relevant data includes Area_SqFt, lbcs_* fields, and county-complete coverage.
• GIS and data teams need stable identifiers, standard file formats, and an API for looking up buildings. Relevant options include str_UUID, bld_UUID, WGS84 files, and the Buildings API.

This is a description of the license, not legal advice; the question of what a derivative database is in your specific architecture belongs with your counsel. What the table shows is that these public options are free to download and build on, but ODbL imposes ongoing attribution and share-alike obligations when the data is incorporated into a derivative database. Those obligations can affect how proprietary data is combined, distributed, and protected, potentially creating additional engineering and operational requirements for a commercial product.

Free building footprints can be a good fit when their coverage, detail, and attributes meet your needs. When they fall short, the work shifts to your team: finding missing structures, correcting outlines, splitting buildings at parcel boundaries, matching addresses, and processing LiDAR for heights and roof slopes. Those tasks require time and expertise, and they must be repeated as source data changes.

EarthDefine does that work as part of producing and maintaining the dataset. Detailed outlines, parcel-split records, addresses, and source dates come together in a consistent schema, with LiDAR-derived attributes available through the 3D product. Analyst review and quarterly releases support ongoing use, while commercial licensing provides a path for proprietary applications without share-alike obligations.

The purchasing decision is therefore about the total cost of getting the building information your application requires and keeping it useful over time. Compare the license price with the engineering, review, and maintenance work you would otherwise take on, then test the data in an area you know. The strongest case for EarthDefine is the work it saves and the analysis it enables.

See the difference on your own ground. Ask for a sample over a market you know, or request a quote for your area of interest. If you’d rather start now, you get 50 free tokens to evaluate our Building Footprints API that returns any address, coordinate, or region as JSON.