Map Data

1. Source

Every build starts from the full OpenStreetMap planet file. Each month the build takes the earliest weekly planet file dated that month once its MD5 checksum is published, checks the download against that checksum, and records the snapshot date and checksum with the build.

A build is named osm-<snapshot date>/p<pipeline version>. The current build is …, made from the … planet snapshot with pipeline version …. The pipeline version changes whenever a tag rule, label rule or geometry parameter changes, and every value below is the one that build applied.

2. Which features are included

One osmium tags-filter pass over the planet selects candidate features with these expressions:

• …

Further tags-filter passes over that result split out tiers A and B (…), tier C (…, then …) and country polygons (…). Each split is exported as polygons with osmium export --geometry-types=polygon, keeping each element's ….

A polygon is kept only when its own tags match one of these rules. A polygon the filter returns only because a matching relation references it is dropped.

Tier A: administrative and place boundaries

• …

Tier B: historic and heritage areas

• …

Tier C: named residential areas

• …

3. What is excluded

• Every other place value, such as place=square
• Every other boundary value, including postal code boundaries
• Every other landuse value, and residential areas without both a name and a wikidata tag
• Protected areas without a name or a heritage tag
• Polygons smaller than … km², measured on the WGS 84 ellipsoid
• Shapes with no polygon left after repair
• A way that duplicates a relation: when the two differ by less than …% of their combined area (symmetric difference over union), only the relation is kept
• An area whose drawn shape cannot be brought within … KB of GeoJSON: within … tolerance doublings for an area of … km² or less, or, for a larger area, even when its outline keeps only its largest outer ring

4. How shapes are processed

Shapes are processed with GEOS, through the shapely library, in this order, and then loaded into our database.

• Repair: a shape is repaired with make_valid only when it is invalid, and only its polygon parts are kept.
• Snap the source: the repaired shape is snap-rounded to a … degree grid, repaired again if snapping left it invalid, and oriented. Details narrower than the grid, such as zero-width slits, cannot be represented and close. Every later step starts from this snapped shape.
• Simplify: topology-preserving simplification, never plain Douglas-Peucker, with a tolerance set by the area's size: …. Tolerances convert to degrees at … m per degree.
• Densify: edges are split so that none is longer than … km.
• Snap: coordinates are snap-rounded to the same grid again, and the shape is repaired again if snapping left it invalid.
• Orient: outer rings run counter-clockwise (CCW) and holes clockwise.
• Size cap: for an area of … km² or less, one processed shape is used both to find the area a stop is in and to draw it. While its GeoJSON is larger than … KB, the tolerance doubles and the steps above run again, up to … times.
• Large areas: an area larger than … km² keeps a shape processed at its base tolerance, or a smaller one under the checks below, for finding stops. It also gets a separate outline for drawing, made with the same steps except densification, with the tolerance doubling until the outline fits … KB. If it still does not fit, the outline keeps as many of its largest outer rings as fit, without holes. How far that outline strays from the full shape is recorded with the area.
• Checks: a shape used for finding stops that strays from the snapped source by more than … its tolerance (Hausdorff distance) is processed again at half the tolerance, rounded down to a whole metre, down to no simplification at all, and the first shape within that limit is kept. For an area of … km² or less that shape must also still fit … KB. With no simplification the limit is … the largest distance snapping can move a point (about … cm). A shape that still strays by more than its limit is left out of the build. A build is not published if any processed shape is invalid or has a clockwise outer ring, or if more than …% of its shapes are left out.
• Labels and countries: an area is labelled with the first of … that it has, or else with its kind and admin level. Its country comes from its own … tag, or from the prefix of the subdivision code in its … tag, or else from the smallest country or subdivision polygon containing a point inside the area.
• Loading: when a build is loaded into our database, any shape the database reports as invalid is repaired with its MakeValid function, and a shape that is not a polygon after that repair is removed.

5. Mainland China

Google’s road map of mainland China is drawn in the GCJ-02 coordinate system, which is offset from the WGS 84 coordinates OpenStreetMap uses. Unless GCJ-02 conversion is enabled, no boundary is shown for a stop in mainland China. Conversion is currently ….

When conversion is enabled, a mainland stop’s coordinate is converted from GCJ-02 to WGS 84 to find its area, and the boundary drawn for it is converted from WGS 84 to GCJ-02 point by point, so it lines up with the road map. Either way, no boundary is shown for a stop within … km of a … boundary, and coordinates in … are never converted.

6. Safety scores

Safety scores come from GeoSure, not from OpenStreetMap. Each stop is scored at its own coordinate, and the scores are kept separately from the boundary data and shown beside it.

7. License and credit

OpenStreetMap data is available under the Open Database License 1.0. See openstreetmap.org/copyright for its copyright and license terms.

Wherever Vew draws these boundaries on a map, the map credits “© OpenStreetMap contributors” and links to this page.

8. Corrections

Boundaries are never edited by hand. A missing or wrong boundary can be fixed in OpenStreetMap, and the fix reaches Vew maps with the next monthly build.