OSM vs Overture: roads

Layers

OSM: matched
OSM: unmatched
Overture: matched
Overture: unmatched
loading roads…

Road network totals

"Roads" here means real roads, motorway down to residential and service, not footpaths or tracks (those still show up in the per-class table below, just aren't matched). A road counts as matched if 60% or more of it falls within 15 m of a line in the other dataset. Lengths are measured properly (UTM 45N), not estimated.

—
OSM road length (km)
—
Overture road length (km)
—
OSM length matched to Overture
—
Overture length matched to OSM

Merged network quality

"Merged" means every road either dataset has mapped. It can't include a road neither one recorded, so this shows how much merging gains over using either dataset alone, not an absolute score. The overlap rate is the closest thing to a single "how good" number: shared length divided by total merged length.

—
merged road length: every road either dataset has
—
overlap rate: how much of the merged network both datasets agree on
—
extra road length merging adds over OSM alone
—
extra road length merging adds over Overture alone

Where the data comes from

Overture tags each road segment with where it came from. Most are sourced from OpenStreetMap itself, so agreement there mostly just measures how stale Overture's last sync is, not independent proof. The segments sourced from TomTom, a separate commercial dataset, are the real cross-check: when OSM agrees with a TomTom-sourced segment, that's two independent surveys landing on the same road.

Overture source Segments Length (km) Matched to OSM (by length)

Per road class

Grey rows are minor paths: counted, but not matched (see above).

Class OSM count OSM km Overture count Overture km OSM match % Overture match %

Merging's payoff, by road class

The city-wide numbers above average over every road class at once, but merging doesn't help evenly. Some classes, usually motorways, already agree almost completely, so merging adds little there. Others gain a lot more. This breaks the same numbers down per class so you can see where the real payoff is. Classes only one dataset has are skipped here (see the full table above for those).

Class OSM km Overture km Merged km Overlap Gain vs OSM alone Gain vs Overture alone

Distance to the nearest match

"Unmatched" is a pass or fail from the 15 m test above. This measures how far each unmatched segment actually is from the nearest road in the other dataset. A short distance usually means the road exists in both but was drawn slightly differently. A long one means it's probably missing from that dataset entirely (see methodology for how this is measured).

—
median distance from an unmatched OSM road to the nearest Overture road
—
median distance from an unmatched Overture road to the nearest OSM road
Unmatched dataset Distance to nearest counterpart Length share  

Where roads are different

The longest roads only one dataset has mapped. Click a row to jump there on the map (turn on satellite imagery first to see which one's right).

Only in Class Name Length (m) Location

Methodology & caveats

Full road shapes, not just counts. This page downloads actual OSM road geometry from Overpass, one grid cell at a time, and removes duplicates from roads that cross a cell boundary.

Matching is by length, not area. Roads are lines, so agreement is measured as how much of each line falls inside a 15 m buffer around the other dataset's lines, not shape overlap. That buffer absorbs ordinary tracing differences without treating a genuinely different road as a match.

"Unmatched" isn't proof something's wrong. It can mean a road is genuinely missing, was rebuilt since the last update, or the two datasets just split the same road at different points. Check the satellite toggle before assuming either dataset is wrong.

Minor paths are skipped. Footways, tracks, cycleways, and similar classes aren't matched, though they're still counted in the per-class table, since they dominate raw OSM data without being what most people mean by "roads."

Closeness distances are an approximation. They're measured to the nearest vertex of the other dataset, not the exact nearest point on a line, so they can slightly overstate distance on smooth, sparsely-drawn curves.