Validating a structure-based population estimate against survey data

A structure count and a building-height layer will get you most of the way to a population number for a district between censuses. They will not get you a number you can defend in front of a technical committee on their own. Before that figure goes into a planning document or a budget allocation formula, it needs to survive contact with whatever household survey data you already have for the area.

This is the step a lot of teams skip, usually because the imagery-derived estimate looks clean and the survey data is old, patchy, or covers a different boundary. Skip it anyway and you're publishing a number nobody can reconcile with the last DHS round or the most recent LSMS cluster in that province.

Why the structure count needs an outside check

A structure-based estimate works by multiplying a building count (or built-up footprint, adjusted for height) by an assumed occupancy rate, usually an average household size pulled from the last census or a recent survey. That multiplication is where the uncertainty lives. The structure count itself, from high-resolution imagery, is usually solid. What you're less sure of is whether the household-size assumption still holds three, five, or eight years after it was measured, in a district that's been adding peri-urban housing faster than anyone can survey it.

Household surveys don't give you a population count for the district either, usually, but they give you two things the imagery can't: a current average household size for a comparable cluster, and an occupancy rate that reflects vacancy, which matters in areas with a lot of unfinished construction or seasonal housing.

Running the ratio estimation check

The actual check is a ratio estimation exercise, and it's simpler than it sounds if you've got any overlapping survey clusters.

Pick enumeration areas or survey clusters that fall inside, or close enough to, your area of interest. For each one, you already have a survey-based population count. Run your structure count and height layer over the exact same boundary and generate your own estimate for that cluster, using whatever household-size assumption you're planning to apply district-wide.

Compare the two. The ratio of survey count to structure-based count for each cluster is your correction factor. If it's close to 1 across clusters, you've got reasonable grounds to trust the district-wide extrapolation. If it's consistently above or below 1, by a similar margin cluster to cluster, that's not noise. That's a systematic bias in your occupancy assumption, and you apply the correction factor to the full estimate rather than treating each cluster's gap as a one-off.

What you're watching for is the spread, not just the average. A correction factor that's 1.1 in every cluster is a good sign; you've found a stable adjustment and can document it as such. A correction factor that swings from 0.7 to 1.4 across clusters in the same district tells you the household-size assumption isn't holding uniformly, which usually means informal settlements, student housing, or new construction with lower-than-average occupancy are skewing one part of the area.

What a mismatch is actually telling you

A gap between the structure-based number and the survey number is rarely a sign the imagery work is wrong. The building count and height layer are measuring something real and current. A gap usually means the occupancy assumption needs to be stratified rather than applied as a single district-wide figure, or that the survey clusters you're using for comparison aren't representative of the growth you're trying to capture, which happens a lot in fast-growing peripheries where the last survey round predates most of the new construction.

Document the correction factor and the clusters it's based on alongside the final number. A technical committee reviewing an intercensal estimate wants to see that the structure count was checked against something, not just that it exists.

If you're building a structure-based estimate for a district that's outgrown its last census and need the building count, height layer, and extraction work done over your own boundary, that's the core of what Population Estimation sets out to do. Get in touch if you want to see how the pipeline handles your area of interest.