The solution
Solution
Finds settlements, a route, and cost.
01 · Find
Satellites find the roofs maps forgot
A study in 2025 showed that the best population maps in the world today still miss over half the population in the rural areas they cover. Instead, Hestia uses satllite imagery to identify settlements that those population maps missed.[2]
Hestia uses Microsoft building outline data as a first pass over an area. Learn more about how they are grouped in the Algorithm page.[3]
When Microsoft doesn't have data for a scan, Hestia downloads images from the Esri World Imagery Wayback, whose photos we run our custom algorithm over.[4]
- 01Find the roofsMicrosoft outlines by default, but resorts to satellite without it.
- 02Separate from surroundingsEach building is drawn on a grid to remove everything but the homes.
- 03Join neighborsNearby buildings are combined into a settlement.
- 04Trace the outlineTo display, we outline them to make them easy to see.
- 50 mlargest gap between buildings in one settlement
- 16 km²per scan, about 4 x 4 km
- 4.7people per home (average household in Nigeria)
02 · Check
Most of Hestia's settlements haven't been mapped
To be sure, Hestia compares its results with OpenStreetMap to check for neighboring homes, outputting a certainty score of 100.[5]
If OpenStreetMap doesn't return anything nearby we limit our certainty to 99 and use three signs of a settlement to build that value. Building count, compactness, and building cover (the amount of outline that looks like roof).
- On OpenStreetMap–
- Satellite only–
- Buildingsfull points at 8 buildings
- Compact shapefull points at 0.45 (a circle is 1)
- Area covered by buildingsfull points at 15% of the outline (35% roof pixels with the color backup)
That adds up to 100 points. The score stops at 99, because only an OpenStreetMap match can reach 100.
03 · Reach
Finding the cheapest route from a clinic
Through OpenStreetMap, Hestia finds different types of health facilities in a 40 km radius of the scan by default, but if nothing is found, the search area increases to 660 km. After finding a facility it builds a map of all the roads a vehicle can traverse (it skips this if no facility is found).[5]
Hestia finds the cost of a journey from the three closest health facilities found near each settlement and keeps the least expensive one. Different types of road have different costs.
In addition to cost, each trip is allotted a difficulty score, keeping track of the terrain and time needed to make the expedition.
Roads that are usually paved cost 1.5 times more if OpenStreetMap says they are dirt, gravel or sand. We chose these rates for planning. They have not been measured in the field yet.
04 · Tour
Maximizing impact, reducing cost
Plotting and following a route to each settlement means the health team would need to traverse the same road many, many times. To counteract this, Hestia plots a trip through all settlements (with high certainty scores) in a scan so the health team can minimize cost.
The clinic route is replanned every time a new scan is saved to make sure it is as optimal as we can make it.
Sources
- Hestia: saved detector scans (this project)Runs of detect-villages.js saved in web/data/scans. The scan numbers on this page are recomputed from them on every load.
- Láng-Ritter, Keskinen & Tenkanen (2025): Global gridded population datasets systematically underrepresent rural populationNature Communications 16, March 2025.
- Microsoft: Global ML Building FootprintsBuilding outlines traced from satellite imagery by a machine learning model. Open Database License.
- Esri: World ImageryOne meter or better satellite and aerial imagery for most of the world's landmass. Esri, Vantor, Earthstar Geographics and the GIS User Community.
- OpenStreetMap contributors: places, health facilities and roadsQueried through the Overpass API. Data © OpenStreetMap contributors, available under the Open Database License.