Baku’s bus network,
under the lens.

06 / Make it reproducible

How it was measured

Routes, stops and route lines come from AYNA's public map API, taken on ; the data is identical, row for row, to a copy taken on . Defects in the map data were detected and repaired automatically, as listed above. Residents come from the Kontur population grid (November 2023, hexagons of about 0.74 km²) and boardings from the daily counts BakuBus publishes on opendata.az (, weekday average). Bicycle lanes and parking points are AYNA's, from its opendata.az datasets (files of 24 September and 4 May 2026, CC0).

Direct route

In both directions the path is at most 1.4 times the straight line between the end stops. Indirect: 1.4 to 2 times, very indirect: 2 to 3 times, loop: more than 3 times, or ends less than 1 km apart.

Detour

A stretch where the bus travels at least 2.5 times the straight line between two of its stops, and at least 1.5 km more (12 stops at most). The extra minutes are estimates from speeds assumed by type of service.

Shorter path

A shortest graph path between a flagged segment’s ends, using stop-to-stop links in the snapshot and short connectors between nearby stops. Shown when at least 0.5 km shorter. The modelled time difference includes assumed speeds and 20 seconds per net intermediate stop avoided; it is not a verified saving or a road-feasibility check.

Near-duplicate

At least 85% of the stops of a route with 10 or more stops are also served by one other route. Express routes are left out: skipping stops is what they are for.

Corridor

A stretch of street of at least 1 km where every link between consecutive stops carries at least 8 routes.

Stop spacing

A stop is flagged by the spacing scenario when, on every route that uses it, it lies less than 250 m along the route from a neighbouring stop and removing it leaves no gap over 700 m. Terminals and interchange hubs are kept, and stops served by fewer routes go first. This geometric rule does not establish that a stop should be removed.

Reach

The stops on the routes that stop at a given stop; with one change, also the stops of every route that stops within 300 m of those. It counts connections, not minutes.

Travel time

Estimated minutes by bus from a stop, from the paper's travel model: half a route's headway of waiting at each boarding, the ride at speeds set by type of service with 20 seconds lost at each stop, and walks of up to 300 m between stops to change buses, which are avoided as if each change cost 5 minutes. Headways are estimated for 200 of the 201 routes, so the minutes are estimates.

Carriers

The routes each carrier runs, as AYNA's map assigns them, with their figures added up. AYNA plans the routes, so the figures describe the routes, not how well a carrier runs them. Names that differ only in spelling are counted once, which gives the paper's 39 carriers.

Coverage

Estimated population within a straight-line radius of a served stop, using the November 2023 Kontur grid. Each hexagon is represented by seven equally weighted points. The study area is the served-stop bounding box plus 2 km, not Baku’s administrative boundary. The default radius is 400 m; the coverage lab uses five recorded radii from 300 m to 1 km.

Bicycles

AYNA's existing bicycle lanes and its bicycle parking points, as published on opendata.az. A stop is within 100 m of a lane in a straight line; a route runs along a lane where its path lies within 40 m of one, sampled every 20 m, so that a lane on one side of a dual carriageway counts for a bus on the other. The parking file also holds 649 points named MPS whose purpose it does not state; they are drawn small and blue. People within 400 m of a lane are counted as coverage counts them.

The study has no comprehensive observed timetables, vehicle positions or origin–destination records. The captured API omits eight routes present in the September 2026 boarding records. Two identical snapshots establish equality on those dates; they do not establish that the API was unchanged throughout the interval.

Directness and overlap alone do not measure service quality: indirect routes can serve important destinations, and shared routes can offer useful frequency or different connections. A shorter path in the graph is not a verified drivable alternative; legal turns, one-way streets, skipped stops and rider impacts need checking.

The paper also explores a redesign under a fixed modelled bus-hour budget. Its results depend on estimated headways and demand; the route paths shown here do not establish real-world time, fleet or fuel savings. The preprint reports the assumptions and sensitivity checks.