Population
× Transit

Train Station Density

Population density & transit connectivity around train stations.

Population

Transit mode

Other layers

Transit Population Jobs Coverage

Station type

Basemap

Country

Loading map data…

Legend

Train

Metro

LRT

Tram

Stations pop. density (/km²)

Bus daily trips · ≤10 min · ≥72/day

Buildings population density

Population hexes per km² · 1 km² hexes

Job hexes jobs per km² · 1 km² hexes

Train Station Density

About

Rapid-transit stations in Israel, the Netherlands, Belgium, Paris (Île-de-France), Madrid (Comunidad de Madrid), Portugal and Switzerland (Kanton Zürich + Léman) sized and coloured by the population density of their walking-distance catchment. The Transit mode filter shows any combination of train, metro/subway, LRT (light rail) and tram stations, each drawing its own line network + stations. Each mode uses its own pedestrian-shed radius (train 1 km · metro 800 m · LRT 500 m · tram 350 m) and its own catchment tessellation, so densities compare like-for-like within a mode. Only operational networks are shown (e.g. Israel has light rail but no metro yet). Trams are off by default (dense); the Chart tab compares the networks head-to-head.

The Country switcher picks a network; the stats card's National / Metro toggle re-scopes it. National = the whole country. Metro reveals that country's metropolitan areas — "All metros" (aggregated) or one to focus the map + stats on. Multi-metro countries: Israel (Tel Aviv · Jerusalem · Haifa), Netherlands (Amsterdam · Rotterdam · The Hague · Utrecht), Belgium (Brussels · Antwerp · Ghent), Portugal (Lisbon · Porto). France (Île-de-France) and Spain (Comunidad de Madrid) are single-metro regions, so they show Metro only and are excluded from the National comparison in the charts. Switzerland is a two-metro region — Kanton Zürich and the Léman arc (Genève + Vaud) — Metro-only and excluded from National for the same reason.

Station typology

The Station type filter groups stations by three location/role axes — setting (inside a metropolitan area or not), distance to the nearest city centre, and hub level — never by density, so density can be compared within a type rather than baked into it. The axes are mode-aware: metro/LRT/tram use tighter distance bands and read hub level from the number of lines meeting at the stop (an interchange count) rather than nearby bus volume.

How much to trust it: distance-to-centre dominates the result — the “Central” band alone tracks most of the classic density-falls-with-distance gradient — so the genuinely extra signal is the setting axis (suburban vs rural at the same distance). Read the type→density link within one mode (it is strong for metro/LRT, weak for trams) and as “consistent with”, not a validation of, the scheme.

Type	Setting	Distance to centre	Hub level
Central	any	core / inner (≤10 km)	any
Suburban	metro	outer / far (>10 km)	any
Regional Hub	regional	outer / far	major / connector
Rural Halt	regional	outer / far	local

Distance is straight-line to the single nearest city centre (one rule everywhere: every agglomeration above ~400k gets a centre point; satellites of a bigger one don't), so a station near a secondary subcentre can read "outer". The hub axis only separates Regional Hub from Rural Halt outside the metros; in the centre a major interchange is flagged as a popup badge rather than its own type. The map's two top-right cards and the per-station charts re-scope to the selected type.

Data sources

CartoDB Positron / Dark Matter
OpenStreetMap
Israel Central Bureau of Statistics — statistical areas 2022
CBS Wijk- en Buurtkaart 2024
Eurostat GEOSTAT 2021
IRIS contours · INSEE Base infracommunale 2021
Israel Ministry of Transport GTFS
OVapi / OpenOV national GTFS
iRail · STIB-MIVB Open Data
Île-de-France Mobilités (IDFM)
Renfe Cercanías GTFS · CRTM Madrid (Metro, EMT, Metro Ligero, Interurban)
Comboios de Portugal (CP) GTFS · Transporlis
INE Portugal — BGRI Censos 2021 (subsecção level)
Swiss BFS — STATPOP 2024 · STATENT 2023 (100 m hectare grids) · opentransportdata.swiss GTFS · swissBOUNDARIES 3D (gg25)
Observable Plot
Leaflet

Caveats

Residential only. Population comes from census zones and reflects residents, not employment or daytime population. The building layer ships all OSM building=* tag values (no residential-tag filter) — industrial / commercial sheds therefore appear in the Buildings overlay, even though they contribute no residents to the per-station numbers.
Census resolution varies. Per-station population areal-interpolates census zones onto each catchment. It is fine-grained for the Netherlands, Portugal and Paris, but Belgium and Madrid use a 1 km² grid, so their core densities read slightly low against the finer networks — though cross-network rankings barely move.
Air distance. All buffers are straight-line. A station close to a highway, fence, or river shows "coverage" that isn't actually reachable on foot.
Routes, not stops. Transit counts use route geometry passing within 100 m — a route that passes near a station without stopping there still counts.
Two density definitions, by design. Train reach % assigns each resident to their nearest station only, so it never exceeds 100%. Pop / station and the map's density ramp use each station's full disk, so a resident between two stations counts toward both — describing "what's around a typical station". Density denominators are clipped to land, so coastal and border stations aren't diluted by sea or off-map area.
Modes are never summed. Each mode is measured in its own radius (train 1 km · metro 800 m · LRT 500 m · tram 350 m) and compared like-for-like within that mode; the per-mode densities aren't added together. The Coverage chart is the one place modes combine, counting each resident once. Headline metrics (reach %, pop/station, stations/1M) are heavy-rail only.
Israel's population hexes look hotter than they are. The hex layer only draws cells that hold people, and Israel's census (CBS statistical areas) maps localities only — so its grid covers about 28% of the country, against 86–103% where the census tiles the whole territory (Netherlands, Belgium, Paris). The empty Negev simply isn't drawn, so nothing cools the map down and every visible Israeli hex reads dense. The totals are unaffected (the grid sums to the national figure) and Israel's per-station densities are ordinary — a median of about 2,600 people/km², essentially the Dutch figure and well below Madrid's. Compare stations, not the extent of the colour.
Single sample day. All transit counts reflect one weekday of the GTFS feed.
Coverage chart. Buffers rail 1 km, metro 800 m, LRT/tram 400 m and frequent bus 200 m. The bus layer buffers route lines (not stops), so it slightly overstates bus reach; the headline train reach is point-based and unaffected.
Portugal merges several operators. CP rail plus the Lisbon and Porto urban + metro networks; some suburban routes appear in coverage but not the high-frequency Transit overlay.
Station type is independent of density. Types come from setting, distance-to-centre and hub level — never density — so density can be compared within a type. Distance-to-centre dominates, so read the type→density link within one mode (strong for metro/LRT, weak for trams) as "consistent with", not proof. The "vs peers abroad" popup line is directional only.
Hub level. Heavy-rail connectivity uses feeder-bus volume within 100 m. Portugal is the gap: its bus feeds cover only Lisbon and Porto, so 316 of 408 stations have no bus data at all — they're marked unmeasured rather than given a misleading zero, and are kept out of the peer comparisons. The measure is also blind to tram-fed hubs — Zürich HB honestly reads zero because its feeders are trams, which count as their own mode — so a busy interchange can still miss its major-interchange badge. Elsewhere a zero is a real zero: no bus within 100 m on the sampled weekday. Metro/LRT use the number of lines meeting at the stop instead.
The four types are roughly a density ladder. Central ≈ 11k > Suburban ≈ 5.5k > Regional Hub ≈ 3k > Rural Halt ≈ 0.8k people/km²; within one mode or a thin network the middle two can swap. A major interchange shows as a badge on Central stations, not its own type. Regional Hub is structurally absent in the three all-metro regions (Paris, Madrid and Switzerland — every one of their stations sits inside a metropolitan area, so the setting axis can't separate suburban from rural), where its facet auto-hides.
Buildings are rail-only. The Buildings / Heatmap overlays show the population around train catchments and are disabled when train mode is hidden — metro/LRT are deliberately left off these overlays to keep the map light.
Paris, Madrid and Switzerland are regions, not countries. Their reach % is measured against a compact, dense metro region, so it isn't comparable to the whole-country rows (IL, NL, BE, PT) — they're excluded from the National charts. Treat them as their own category.
Switzerland — finest data, densest transit. Two regions (Zürich + Léman) on BFS 100 m hectare grids for both residents and jobs — the finest data here — so read its rows as what a system can look like, not a national benchmark. Only operational (not under-construction) lines are shown.
Workplace jobs — Paris, Netherlands, Belgium, Portugal, Madrid & Switzerland. A Jobs hex layer and the Density tab's Population / Jobs / Integrated selector show job density where it ships; the rest of the app stays residential. Sources are official workplace statistics at the finest open geography per network — Paris and Switzerland resolve to building / 100 m-hectare level, the others to municipality (Madrid city split into barrios). Definitions differ a little (e.g. the Netherlands counts employee jobs only; Portugal excludes "no fixed workplace"), so cross-network job comparisons are directional. The Jobs hexes reuse the population grid, so jobs in unpopulated cells (industrial / business parks) don't render there — the station figures and charts use the full data. A Raw data button (where shipped) shows the source zones. Israel's Jobs layer is greyed: no open Israeli statistic reports employment at place of work at any sub-district geography, and a point-of-interest proxy was built, measured and rejected — it ranked stations well but put ~1.3–1.8× too much job density in a typical station's catchment, which is precisely the number this map would be read for.

For questions or comments, contact us at contact@nsa-hub.com.

Summary