Skip to content

How site outlines are made from map data

Satellite signals measure a place: a parking lot, a tank farm, a plant, a lake, a stretch of motorway. To measure a place, the service needs its outline. This page describes how the outline is made from OpenStreetMap data and every change made to that data on the way. It is the method we offer under section 4.6 of the Open Database License.

An outline is a mask. For each satellite image, the service reads the pixels inside the outline and, for most signals, the pixels in a ring around it. It turns those pixels into one number for that image. The outline decides which pixels are read. It is not part of the result.

The tables you read contain the measurements and three facts about the site on every row:

  • site_source: where the outline came from, and the site’s name when the map has one.
  • site_area_km2: the site’s area in square kilometres.
  • site_offset_km: how far the point you asked about is from the site. For the radar signals (parking, oil_storage, site_change, stored_vehicles) it is measured to the centre of the site. For plant_heat, water_area and truck_traffic on a mapped site it is measured to the site’s nearest edge, so it is 0 when the point is inside the site.

The tables do not contain the outlines.

The map data is OpenStreetMap data as distributed by the Overture Maps Foundation in its public GeoParquet releases. The service reads four Overture themes: base/infrastructure, base/land_use, base/water and transportation/segment. These themes are available under the Open Database License (ODbL) 1.0. Some road segments in transportation/segment come from TomTom, under the same licence.

The service reads the newest Overture release in which all four themes are published. It checks for a newer release at most once every 7 days, and it records the release it used. In its default configuration it does not query the OpenStreetMap servers.

Every answer that used map data carries this line:

© OpenStreetMap contributors. Contains information from OpenStreetMap (https://www.openstreetmap.org/copyright) and Overture Maps Foundation (https://overturemaps.org), which is made available under the Open Database License (ODbL) (https://opendatacommons.org/licenses/odbl/1-0/).

When a road outline used segments that Overture sourced from TomTom, the answer also carries:

Data from TomTom (via Overture Maps, ODbL).

When the outline was changed from the map data by any step on this page, the answer also carries:

Site outlines were adjusted from the map data by our snapping and merging method, described at https://mostlyright.md/docs/satellite-signals/method.

Each kind of site is chosen by its original OpenStreetMap tags, which Overture carries with each feature. When a feature carries no OpenStreetMap tags, Overture’s own class is used instead: parking for parking lots, storage_tank for tanks, industrial for industrial areas.

Kind of site Signals that use it Features that count Features left out
Parking lot parking amenity=parking parking= multi-storey, underground, rooftop, lane, street_side, on_kerb, half_on_kerb, shoulder, garage, garages or carports; location= underground, rooftop or roof; any building tag other than no; a negative layer
Tank farm oil_storage man_made=storage_tank or man_made=tank; landuse=industrial with industrial= tank_farm, oil_storage, fuel_storage, oil or petroleum_terminal none
Solar farm site_change power=plant or power=generator, or power=construction with construction=plant or generator, when plant:source or generator:source contains solar or generator:method=photovoltaic; landuse=industrial or landuse=energy when industrial contains solar or a source tag above says solar generator:place= roof or rooftop; location= roof or rooftop
Industrial site plant_heat, site_change landuse=industrial, man_made=works none
Port none at present landuse=port, landuse=harbour, industrial=port none
Lake water_area Overture water features of subtype or class lake or reservoir water= river, canal, stream, ditch, drain, riverbank, rapids, wastewater, lock or moat; any feature with a waterway tag
Motorway truck_traffic Overture road segments of class motorway or trunk link (slip) roads: subclass link or the is_link road flag

Polygons are kept as published. A polygon that is not valid geometry is repaired, and only its polygon parts are kept. Points are used only for tanks, and lines only for roads. Every other point or line is ignored.

These steps run before a site is chosen. Distances are in metres in the local UTM zone of the site.

  1. Tanks mapped as points become discs. The radius is half the number in the tank’s diameter tag, read as metres. Without a usable diameter tag the radius is 15 m. No disc is smaller than 3 m in radius.
  2. Tanks are merged into tank farms. Every tank and tank-farm area is grown by 60 m. The grown shapes are joined, and the result is shrunk by 60 m. Tanks and tank-farm areas less than 120 m apart become one farm. So that a farm is not cut at the edge of the area read, the area is read 2 km beyond the search radius. The margin is doubled, up to 16 km, while a farm within the search radius still touches the edge.
  3. Road centre lines become strips. Each motorway centre line is buffered to 15 m on each side, and each trunk road centre line to 10 m, with flat ends. The strips are used to find the road near the point. The road mask itself is drawn from the centre lines (see Outlines become masks).

A request names a point (lat, lon) and, optionally, a search radius (radius_km). The distance from the point to a feature is measured to the feature’s nearest edge, and it is 0 when the point is inside the feature. A feature counts when that distance is no more than the search radius. Features are kept whole: a feature that reaches into the search radius is not clipped.

The point is never moved, and the site does not have to contain it. parking and oil_storage measure every matching feature within the search radius, merged into one site. The other signals measure one feature, chosen by the rules in the table below, and it is not always the nearest one. site_offset_km records how far the point is from the site. Requests carry a point, not a site name, so names are not matched.

Signal Default search radius How the site is chosen
parking 0.8 km Every parking lot within the search radius. The lots are merged into one site (their union).
oil_storage 3 km Every tank farm within the search radius, after step 2. The farms are merged into one site.
site_change 2 km, fixed Solar farms and industrial sites of at least 0.05 km². The largest one that contains the point wins. Otherwise the nearest one within 2 km wins, and any within 0.5 km of that nearest distance compete: the largest of them wins. One feature is measured, never a merged group.
plant_heat 1 km Industrial sites and works up to 40 km². Among those that contain the point, a named man_made=works outline wins, then the largest. Otherwise the nearest one wins.
water_area 2 km Lakes. The smallest lake that contains the point wins. Otherwise the nearest one wins.
truck_traffic 5 km (0.5 to 25 km allowed) The nearest motorway or trunk strip must be within 1 km of the point. If that segment has a road number (the first entry of its route references), every segment with the same number is used. Otherwise every unnumbered segment of the same class that connects to it is used: two segments connect when their strips come within 1 m of each other. The centre lines are clipped to a circle of the search radius around the point.

stored_vehicles uses no map data. It always measures a circle around the point (0.9 km by default).

The outline is drawn onto the pixel grid of the images, in the local UTM zone of the site.

Radar signals (parking, oil_storage, site_change, stored_vehicles) use a 10 m grid that covers the site and its ring plus 100 m.

  • For parking, the site mask is eroded by one pixel: the outermost 10 m pixels of each lot are dropped, so kerbs, trees and buildings at the edge do not dominate. The other radar signals are not eroded.
  • At least 0.02 km² of the site must remain after erosion. Sites larger than 100 km² are refused.
  • The ring is the band between two distances outside the site, minus any pixel of the site itself. Other sites of the same kind are cut out of the ring, each grown by the inner distance first:
Signal Ring around a mapped site What is cut out of the ring
parking 10 m to 600 m outside the lots Parking lots found beyond the search radius but within the search radius plus 0.8 km, each grown by 10 m
oil_storage 150 m to 1,500 m outside the farms Tank farms found beyond the search radius but within the search radius plus 1.7 km, each grown by 150 m
site_change 150 m to 1,500 m outside the site Every other solar farm and industrial site within 5.2 km of the point (with a radius_km above 1.25 km: 4 times the radius plus 0.2 km), each grown by 150 m. If less than 40% of the uncut band remains, the outer edge moves to 3,000 m

plant_heat drops any holes in the outline and marks every 20 m pixel that the outline touches. The site must cover at least 1 km².

water_area drops any holes in the lake outline (islands are land in every image), marks every 20 m pixel that the outline touches, and grows the mask by 60 m (3 pixels) to include the shore. For very large lakes a coarser grid is used, and the mask still grows by at least one pixel. Lakes of more than 3,000 km² are refused, and lakes of less than 0.05 km² are refused.

truck_traffic draws the clipped centre lines on a 10 m grid, marking every pixel a line touches. The road mask is that line grown by one pixel, to cover both carriageways. A strip grown by five pixels around the line is the area read around the road. The road must be at least 4 km long. A dual carriageway is mapped as two one-way centre lines, so one-way lines count half towards that length and two-way lines count in full.

The service never replaces a missing outline with a shape of its own without saying so. When it measures a circle, site_source says so, for example circle 0.5 km. A circle holds no map data.

  • parking, oil_storage and plant_heat: with no matching feature within the search radius, the request is refused with NO_FOOTPRINT_FOUND. If you passed radius_km, a circle of that radius around the point is measured instead. For parking and oil_storage, the circle’s ring runs from 1.3 to 2.2 times the radius, with nothing cut out.
  • site_change: with no solar farm or industrial site chosen, a circle around the point is measured (1 km, or radius_km). Its ring runs from 1.5 to 3 times the radius from the point. Every solar farm and industrial site found is cut out of that ring, grown by 150 m. If less than 40% of the uncut ring remains, the outer edge moves to 4 times the radius.
  • water_area: with no lake mapped and radius_km given, the water body is found in the images instead of the map: the pixels within the circle that were water in at least 10% of clear images, connected to the point or nearest to it. Without radius_km the request is refused.
  • truck_traffic: with no motorway or trunk road within 1 km, the request is refused.

If you pass your own outline, the service measures it and uses no map data for the site.

For every signal that uses map data, the first site that resolves for a place is stored. Every later request with the same signal, point and radius_km reuses it, including after the service is updated. A newer map release that redraws or removes a feature therefore cannot change what is measured at that place. For the radar signals, a stored site that no longer passes the service’s current checks is resolved again. There are two exceptions. An outline you pass is measured exactly as you sent it on every request, with no map lookup. A refusal is not stored, so the next request looks again.

For plant_heat, water_area and truck_traffic, site_source names the Overture release that was current when the site was first resolved.

For every signal, when a place that answered before is refused because of the place, the service returns the last full answer for it with a note, instead of the refusal.

To receive the outline used for a site under the Open Database License 1.0, or to ask about this method, write to team@mostlyright.md. Name the signal and the request: lat, lon and radius_km if you set it.