Blog·
Estimating housing units per acre from satellite imagery

You're scoping a candidate site, the demographic file hasn't come back yet, and the pro forma already wants a density number. So you open the imagery and start counting rooftops yourself. It works, within limits, and it's worth knowing where those limits sit before you put a units-per-acre figure in a memo.
Start with the footprint, not the parcel
Parcel boundaries lie to you about density. A single parcel can hold one house or a 40-unit garden complex, and the GIS layer won't tell you which. What you actually need is the building footprint: how much of the lot is covered, how many distinct roof shapes sit on it, and whether those shapes repeat in a grid pattern (townhomes, duplexes) or stand alone (detached single-family).
At typical high-res satellite resolution, detached rooftops are countable one by one in most suburban settings. Attached product is harder. A rowhouse block or a garden-style apartment building often reads as one long rectangular footprint, and you have to infer unit count from roofline breaks, driveway cuts, or repeating window bays rather than counting separate roofs. This is the step that eats the most analyst time, because it's visual judgment, not a lookup.
Rooftop counting by housing type
The rough method most analysts use, imagery in hand:
- Detached single-family: count rooftops directly, divide by the buildable acreage of the area you're measuring.
- Townhome/rowhouse: count distinct roofline segments or party-wall shadows along the structure, since the block often reads as a single footprint from above until you look for those seams.
- Garden apartments: estimate unit count from building footprint size divided by a typical unit footprint for that building type, then sanity-check against parking stall count, which scales with units more reliably than roofline does.
- Mobile home parks: count pads, not structures. Pads are usually more consistent and easier to tile-count than rooflines.
None of this gives you a precise unit count. It gives you a defensible estimate, which is what a density-as-demand-proxy exercise needs at the screening stage.
Where the count goes wrong
Tree canopy is the biggest source of error. Mature tree cover over a cul-de-sac can hide half the rooftops in a neighborhood that's otherwise easy to read elsewhere on the same image. Flag canopy-heavy tiles rather than guessing a count through foliage.
Parking lots are the second trap. A surface lot next to a low-rise building can make the site look more intensively developed than it is, especially at a glance, because the eye reads "lots of pavement" as "lots of density." Pavement area doesn't tell you unit count. Check it against footprint size instead.
Off-nadir or older imagery can also throw off footprint measurement, since a building shot at an angle casts its roof outline in a different place than its ground footprint. If the imagery date or angle looks off, don't trust footprint-based area math from that tile without a second look.
Units per acre as a demand proxy, not a census figure
For site selection, the count matters less for precision than for what density correlates with: rooftop-driven trade area demand near a candidate site, ahead of synthetic demographic estimates that haven't caught up to what's been built. A corridor that was farmland in the last demographic vintage but now shows 300 new rooftops on current imagery is telling you something the population estimate hasn't caught up to yet. Units per acre, read directly off the imagery, is a way to get that signal early.
Doing this tile by tile, site by site, by eye, is exactly the kind of read that takes an analyst half a day per candidate and doesn't scale across a pipeline of twenty sites. Trade Area Profiling reads the high-res imagery around a candidate site on request and returns a numbered brief: housing type mix, a density read, land use breakdown, built directly off what's standing on the ground rather than what the last demographic vintage assumed.
If you're running this rooftop count by hand on every candidate in your pipeline, it might be worth seeing what a per-site brief looks like instead.