Data
What we render from
How a course film gets built, every open-data layer behind it, and where each course line came from.
Every film on this site is built from published open data, not from satellite-imagery software: the building models, aerial photographs and elevation surveys that cities and national mapping agencies release themselves. This page is the list of exactly which datasets, so any distance, gradient or landmark position we print can be followed back to the thing it was measured from.
Nothing is rendered in satellite-imagery software: those terms forbid the use we put them to. So every layer has to be one somebody published to be used, and every one of them is below, with its licence and the date we last checked it.
- Layers registered
- 13
- Cleared for use
- 13
- Course geometries
- 2
How a course is built
Four steps between a published street list and a frame of film.
The line
An organiser's own GPX file, where one exists, is a geometric reference and nothing else. Those files usually carry no licence at all, so no coordinate from one appears in anything we publish. The reference is map-matched onto OpenStreetMap road geometry, which is explicitly commercial-friendly, and the matched line is swept for artefacts before it is drawn. Where a city publishes its own street-closure notice, that is better still: it gives the course as an ordered list of streets, and a street list is a fact rather than someone's file. Each course below carries the method that produced it, and how far it sits from its reference.
The ground
Elevation comes from a bare-earth model wherever one is published, a model that has had the buildings taken out of it. It matters more than it sounds: a global surface model reads Berlin as 197 m of climb, because it is measuring rooftops. The real figure is 35.9 m. A bare-earth model cannot see a bridge deck either, so a crossing is drawn dashed and no gradient claim rests on it.
The city
Buildings come from the city's own 3D model and the ground texture from its own aerial survey. Landmarks get built: a city model gets a building's mass right and its character wrong. The Brandenburg Gate arrives as a solid block with no colonnade and no passages, so the landmarks on a course are modelled from published dimensions and checked against them.
The claims
Corner counts, the longest straight, where the high point falls against the thinnest crowds: measured off the course line itself. Each one is marked on the race page as measured, sourced or our own reading. What we could not stand up is listed on the race page too, with the reason. A course film should not repeat something just because every guide does.
Layer register
Best local source wins. A global fallback is used where a city publishes nothing better, never as a default.
13 of 13layers verified usable. Every layer below may enter a render.
| Layer | Role | Resolution | Licence | Commercial use | Verified |
|---|---|---|---|---|---|
| Berlin LoD2 city model (LoD2-DE Berlin, CityGML)Geoportal Berlin / Senatsverwaltung für Stadtentwicklung, Bauen und Wohnen · City | Buildings | LoD2 - modelled roof forms, no facade texture | Datenlizenz Deutschland - Zero - Version 2.0 (dl-de/zero-2-0)Zero-2-0 waives attribution as a legal condition; we credit Geoportal Berlin anyway.The source CityGML carries no ParameterizedTexture at all - the buildings are untextured by design, which is why facades are procedural and the orthophoto is projected top-down only. Six 1 km tiles cover the Phase 1 scene block. | Cleared | |
| NYC 3-D Building Model (CityGML, LoD1.5/LoD2 hybrid)City of New York, Office of Technology and Innovation - NYC Open Data · City | Buildings | LoD1.5 prismatic with roof-structure detail; roughly 100 iconic buildings modelled to LoD2 | NYC Open Data - Terms of Use (no named licence)Weaker than Berlin's position, and the difference is worth stating plainly.Berlin ships under a named, machine-readable licence (dl-de/zero-2-0). New York publishes no named licence at all: the dataset metadata says only 'Open Data policies and restrictions apply', nyc.gov's general Terms of Use reserve rights over site materials without addressing datasets, and the clearest grant anywhere in the chain is the Open Data FAQ - 'Open Data belongs to all New Yorkers. There are no restrictions on the use of Open Data.' That sentence is the basis for commercialUse here, and it is a policy statement rather than a licence grant. Second constraint, unrelated to licensing: the survey is 2014 and described as a one-time capture, so a decade of Long Island City and Hudson Yards towers is missing. | Cleared | |
| Berlin true orthophoto 20 cm (truedop_2023)Geoportal Berlin · City | Imagery | 20cm/px | Datenlizenz Deutschland - Zero - Version 2.0 (dl-de/zero-2-0)Attribution is not a legal condition; we credit Geoportal Berlin anyway.This is the ground texture for every low pass: a true orthophoto, so building lean is already removed. It is projected top-down onto ground and roofs only; nadir imagery holds no facade data. The vintage is part of the layer rather than a detail of it. Berlin serves each flight as its own WMS layer, 2023 is the one every frame is drawn from, and it was chosen for the season: it is the only leaf-on flight, and the sun in the renders is solved against the light already baked into it. A later vintage is a different photograph of a different summer, so the credit names the year. Verified live against the service's own GetCapabilities on 2026-09-14: layer truedop_2023, titled 'Digitale farbige TrueOrthophotos 2023 (DOP20RGBI)', ows:Fees reading 'Für die Nutzung der Daten ist die Datenlizenz Deutschland - Zero - Version 2.0 anzuwenden. Die Lizenz ist über https://www.govdata.de/dl-de/zero-2-0 abrufbar.' and access constraints 'Es gelten keine Zugriffsbeschränkungen.' | Cleared | |
| NYC orthoimagery, 2024 flightCity of New York / New York State GIS Program Office · City | Imagery | 6 in/px (about 15 cm), 4-band, true ortho | NYC Open Data - Terms of Use (no named licence)Same unnamed-licence position as the building model.Resolution is better than Berlin's DOP20: 15 cm against 20 cm, which raises rather than lowers the ceiling on low passes. Vintage still has to be picked by season before it is picked by resolution - the Berlin lesson was that leaf-on is the decision and sharpness is the detail. New York flies biennially, so 2024, 2022 and 2020 are all available to compare. | Cleared | |
| Sentinel-2 optical imageryCopernicus / European Space Agency · Global | Imagery | 10m/px | Legal Notice on the use of Copernicus Sentinel Data and Service InformationAttribution required“Contains modified Copernicus Sentinel data”Free, full and open - commercial use included - under Regulation (EU) No 1159/2013.Altitude only. At 10 m/px it reads well from 2 km up and like porridge at 100 m, so it is the wide-shot fallback and never a city-level texture. Where a city publishes its own orthophoto, that wins. | Cleared | |
| ATKIS DGM1 - Digitales Geländemodell Berlin, 1 mGeoportal Berlin / Senatsverwaltung für Stadtentwicklung · City | Terrain | 1 m grid | Datenlizenz Deutschland - Zero - Version 2.0 (dl-de/zero-2-0)The reason a Berlin elevation profile is possible at all.This is a DGM, a bare-earth model, defined by its own abstract as excluding vegetation and buildings, where Copernicus GLO-30 is a DSM and carries every rooftop: sampled off that global model the same course reads 197 m of gain, most of it architecture. Sampled through the queryable layer c_dgm1, whose GetFeatureInfo returns a height in metres above NHN. Two things about the number that matter more than the source. Unsmoothed, even this bare-earth series sums to 115 m of gain at 50 m spacing, because at that spacing it is a picture of kerbs and ramps rather than of the ground, which is why the published 35.9 m is quoted with its 1 km smoothing window and its 200 m spacing. And a bare-earth model reads the ground UNDER a bridge, not the deck a runner is on, so canal and Spree crossings sit up to 5 m low in the raw series. Verified live against the service's own GetCapabilities on 2026-09-14: layer c_dgm1 served, ows:Fees naming dl-de/zero-2-0. | Cleared | |
| Copernicus DEM GLO-30European Space Agency / European Union (Airbus Defence and Space, DLR) · Global | Terrain | 30m | Copernicus DEM open licence (ESA, worldwide royalty-free, commercial use permitted)Attribution required“(c) DLR e.V. 2010-2014 and (c) Airbus Defence and Space GmbH 2014-2018 provided under COPERNICUS by the European Union and ESA; all rights reserved”GLO-30 is a DSM, not a bare-earth model: it already contains the buildings.It must never be draped under a separate city model, or every building is counted twice - once as terrain, once as geometry. Under buildings, use a bare-earth DGM. This is also the layer that makes raw urban elevation lie; see the Berlin gain figure, 197 m raw against 35.9 m smoothed over 1 km at 200 m spacing on a bare-earth DGM. | Cleared | |
| NOAA NYC 2017 topobathymetric lidar (bridge decks)NOAA Office for Coastal Management, National Ocean Service (NODD collection 9306) · City | Terrain | 10.24 points per square metre; ASPRS class 17 = Bridge Deck | NOAA Open Data Dissemination (public, attribution requested)Attribution required“NOAA Office for Coastal Management”The only source that can see the deck the New York Marathon actually runs across the Queensboro Bridge.That bridge carries two roadways in one cantilever truss and the race uses the LOWER one, so the 3DEP cloud cannot resolve it - the upper deck is directly overhead and nadir lidar does not see through a roadway at any density. The 2017 survey caught it from oblique scan angles and off the deck edges, and classified it. Licence, verbatim: 'NOAA data disseminated through NODD are open to the public and can be used as desired. NOAA makes data openly available to ensure maximum use of our data... NOAA requests attribution for the use or dissemination of unaltered NOAA data. However, it is not permissible to state or imply endorsement by or affiliation with NOAA. If you modify NOAA data, you may not state or imply that it is original, unaltered NOAA data.' ISO metadata access constraints: 'None.' Take NOAA's copy, not the city's: the same 2017 lidar published by New York carries licenseId: null, no licence at all, so routing through NOAA is what keeps this layer out of the city's licence debt rather than adding to it. CRS NAD83(2011) / UTM 18N with NAVD88 heights on GEOID18, in metres - note the 3DEP series it sits beside is GEOID12A; the two differ by a few centimetres over New York, which is inside the measurement's own bin width and cancels entirely from a net-rise figure, but the two must not be mixed into one absolute series without saying so. Read by COPC range request: 210 MB fetched from a four-tile footprint totalling 354 MB, no bulk download. Four things to know before reading it again. Class 17 does not separate the two decks - both carry it, six and a half metres apart in the same histogram - so the split is by z band inside an XY corridor, not by class. Tiles are NAMED in New York State Plane Long Island feet on a 2,500 ft grid while their coordinates are UTM 18N metres, and the date prefix varies per tile, so a filename-derived lookup is silently wrong and the numeric suffix must be resolved against a key index. The collection stops at State Plane E 999,999 ft, which leaves the first 118 m of the Queensboro structure, at the Queens end, with no coverage. And 9306 uses classes outside the LAS 1.4 standard set (class 76, Subway Stairs), so a strict validator will reject it. Cross-checked: measured deck separation 6.48 m over 89 stations, and the ENC's lower-level soffit of 39.9 m above Mean High Water converts to 40.49 m NAVD88, leaving 3.97 m of floor system under the measured 44.46 m running surface - the right order for a 1909 cantilever. | Cleared | |
| USGS 3DEP 1 m digital elevation model (bare earth)United States Geological Survey, 3D Elevation Program · National | Terrain | 1 m, bare earth (DTM) | Public domain (work of the United States Government)The one layer in the New York stack whose licence needs no conversation with anybody.It is also a bare-earth DTM, which is the whole point: our standing rule on elevation is that an urban surface model contains the buildings and reads them as hills, and Berlin only escaped that by finding an open ATKIS DGM1. 3DEP is the American equivalent and it is national, so this solves the terrain layer for every United States race, not only this one. SAMPLED 2026-09-11 along the full course: 857 points, none missing, 851 of them answered from 1 m data captured 2014-03-01. Six fell back to the seamless 1/3 arc-second DEM (about 10 m) in three short runs between km 32.2 and km 32.8 in the Bronx, and which raster answers near a coverage boundary is not deterministic - re-running moves the course gain by up to 0.2 m on 105 m. Both facts are recorded per run alongside the profile itself. The service reports cell size in the source raster's own units, so the test is 'is it 1', never 'is it small'. Bare earth is also its limit: over a bridge it returns the ground or the hydro-flattened water surface beneath, and 14.7 % of this course is on a deck - it reads -1.62 m across The Narrows while the Verrazzano deck is sixty metres up. Those runs are interpolated in this layer, and drawn dashed. They no longer carry no gradient claim: as of 2026-09-12 the decks are measured from 3DEP's point cloud, which still contains the deck the DEM removes - bare earth is a product, not the survey. See the 3DEP point cloud entry below, and the NOAA topobathy lidar entry for the Queensboro, which is occluded by its own upper roadway and is the one bridge 3DEP cannot see. A gradient claim must cite those, never this. | Cleared | |
| USGS 3DEP lidar point cloud (bridge decks)United States Geological Survey, 3D Elevation Program · National | Terrain | 0.7 m nominal point spacing; 0.116 m vertical accuracy at the 95th percentile | Public domain (work of the United States Government)The answer to a question the bare-earth DEM cannot be asked.3DEP's DEM removes the bridge deck; 3DEP's point cloud still contains it, unclassified rather than absent - so the deck of the New York bridges the race crosses is measurable from the same federal programme that already supplies the terrain, and this adds NOTHING to the city's licence debt. The project is NY_New_York_CMGP_SANDY_LiDAR_15 / NY_CMPG_2013 (Woolpert), flown 2013-08-06 and 2014-03-21 to 2014-04-21 over Bronx, Kings, New York, Richmond and Queens - the only 3DEP lidar project covering all five bridges. Its FGDC metadata states access constraints as 'No restrictions apply to this data.' and use constraints as 'None. However, users should be aware that temporal changes may have occurred since this dataset was collected and that some parts of these data may no longer represent actual surface conditions. Users should not use these data for critical applications without a full awareness of its limitations. Acknowledgement of the U.S. Geological Survey would be appreciated for products derived from these data.' The National Map's own terms are blunter: 'Map services and data downloaded from The National Map are free and in the public domain. There are no restrictions.' Read over HTTP as Entwine Point Tiles from https://s3-us-west-2.amazonaws.com/usgs-lidar-public/NY_NewYorkCity/ept.json - 4,755,025,996 points, source CRS EPSG:26918, vertical datum NAVD88 (GEOID12A) in metres, the SAME datum the 3DEP DEM answers in, so deck and ground need no conversion between them. Three limits. (1) ASPRS class 17 is NOT 'Bridge Deck' in this delivery - it is 'Overlap Default', and 18 is 'Overlap Ground'; 17 became Bridge Deck only in Lidar Base Specification 2020 rev. A, years after this LAS 1.2 flight. Filtering on 17 returns swath overlap near sea level and no bridge at all. The deck is found geometrically instead: drop classes 2, 7, 9, 10 and 18, then take the densest continuous band in a 14 m corridor of the course line. (2) Levels are binned at 0.5 m over 20 m stations, so a single-station gradient is quantised at plus or minus 2.5 per cent - quote a net rise between two named kilometres, never a point gradient. (3) The Queensboro's LOWER level, which is the deck the race runs, is in the upper level's shadow over the main spans and is NOT measurable from this collection; about 320 m of it returns only upper-deck and truss returns. Cross-checked against a second reading of a delivered LAZ tile (Verrazzano upper deck 79.5 m against this method's 79.93 m) and against published navigation clearances converted from Mean High Water, which corroborate but cannot confirm: NBI item 39's datum is set per navigation permit and is not recorded in the data. | Cleared | |
| Baumbestand Berlin - street and park tree cadastre (Strassenbaeume + Anlagenbaeume)Geoportal Berlin / Senatsverwaltung fuer Stadtentwicklung, Bauen und Wohnen · City | Vectors | per-tree point, with height, crown diameter, trunk circumference and genus | Datenlizenz Deutschland - Zero - Version 2.0 (dl-de/zero-2-0)Verified from the WFS 2.0 service's own GetCapabilities, not from a third-party listing.ows:Fees reads verbatim: 'Fuer die Nutzung der Daten ist die Datenlizenz Deutschland - Zero - Version 2.0 anzuwenden. Die Lizenz ist ueber https://www.govdata.de/dl-de/zero-2-0 abrufbar.' ows:AccessConstraints: 'Es gelten keine Zugriffsbeschraenkungen.' Zero-2-0 waives attribution as a legal condition; we credit Geoportal Berlin anyway. Same licence as LoD2 and DOP20, so this layer adds NO new licence surface to the Berlin stack. Our fetch re-asserts that string on every run and refuses to write a single feature if it stops naming dl-de/zero-2-0. NOT to be confused with the Gruen Berlin 'Anlagenbaeume' publication, which is a different dataset under dl-de/BY and would carry an attribution condition; the layer used here is baumbestand:anlagenbaeume served by this service. Two layers, native EPSG:25833. A GetFeature over the Phase 1 scene block's bounding box (EPSG:25833 388000,5819000 to 391000,5821000) returns 3,551 Strassenbaeume and 12,683 Anlagenbaeume, 16,234 in all; the service's BBOX filter is inclusive to about half a metre, and on an exact point-in-box test 5 of those sit 0.003-0.387 m beyond an edge, so the scene itself builds 16,229. That is the figure to quote, and it reproduces the content audit's independently derived 16,229 exactly. OpenStreetMap returns 7,461 over the same block. | Cleared | |
| GeoNamesGeoNames (Unxos GmbH) · Global | Vectors | city centroid | Creative Commons Attribution 4.0 (CC BY 4.0)Attribution required“Place coordinates (c) GeoNames, CC BY 4.0”Read from the published gazetteer dump rather than the live API, so a sweep costs 101 downloads instead of 23,000 requests.Every coordinate taken from it is a CITY CENTROID: measured against the 107 curated races whose start lines were authored by hand, a centroid lands a median 2.5 km out and 100% within 50 km. That is good enough to say how far a race is from home and is not good enough to say where anybody assembles at 8 a.m., so nothing is ever drawn from one. Attribution is required, which most of the geodata here does not require, and this record is how it is discharged. | Cleared | |
| OpenStreetMapOpenStreetMap contributors / OpenStreetMap Foundation · Global | Vectors | vector, street level | Open Database License 1.0 (ODbL)Attribution requiredShare-alike“(c) OpenStreetMap contributors”Explicitly commercial-friendly, which is why the shipped course line is OSM geometry rather than the organiser's GPX. Share-alike binds a derived database, not a rendered frame. Overpass returns HTTP 406 unless a User-Agent header is set. | Cleared |
Course geometry
- Shipped format
- geojson
- Licence
- Open Database License 1.0 (ODbL)
- Commercial use
- permitted
- Share-alike
- yes
- Attribution
- (c) OpenStreetMap contributors
- Verified on
- 2026-09-05
- Scene anchor
- E 388,000 · N 5,819,000
- Display transform
- 8× vertical · 2% apparent ribbon width
Derived with Newson & Krumm (2009) HMM map matching with Viterbi decoding (emission sigma 12 m, transition beta 15 m; cycleway/footway/path classes log-penalised -9.0, cosine-tapered to zero by 60 m from the nearest carriageway). Deviation from the reference trace: 3.5 m mean, 8.3 m at p95, 21.3 m worst case, with a length difference of 1.49%.
Reference input, not shipped geometry. Official organiser GPX (berlin.gpx), 508 trackpoints, 42,274.3 m. Licence: None stated. The file neither grants nor forbids reuse, so it carries no content licence we can rely on. Geometric reference for the map-matcher only. No coordinate from this file appears in any shipped output; every output coordinate is interpolated from OSM way geometry.
Share-alike binds derived databases, not rendered frames. Every shipped coordinate is interpolated from OpenStreetMap highway ways fetched via Overpass; 96,959 OSM edges were in the matching network.
Course geometry
New York City Marathon
42.76 km over 14,255 points, in EPSG:26918 (NAD83 / UTM zone 18N).
- Shipped format
- geojson
- Licence
- Open Database License 1.0 (ODbL)
- Commercial use
- permitted
- Share-alike
- yes
- Attribution
- (c) OpenStreetMap contributors
- Verified on
- 2026-09-11
- Scene anchor
- E 579,000 · N 4,495,000
- Display transform
- 8× vertical · 2% apparent ribbon width
Derived with Street-sequence walk: the city's own closure notice gives the course as an ordered list of streets, junctions are resolved as shared OpenStreetMap nodes, each segment routed by Dijkstra with off-street edges at 3x length, and out-and-back spurs collapsed. There is no reference trace to deviate from - the course is defined by a street list, not a line - so the one figure that means anything is the length: 1.34% over the certified distance, which is what a centreline costs against a course measured on the shortest legal line.
Reference input, not shipped geometry. City of New York street-closure notice for the TCS New York City Marathon, 1-2 November 2025: the course as an ordered list of streets with their cross-street limits. Licence: A published street list, not a geometry file. It carries facts about which roads close and in what order, and no coordinate is taken from it. Defines the sequence of ways to walk. This is a stronger position than Berlin's, where the reference was an organiser GPX with no content licence at all: here there is no third-party geometry in the chain anywhere, only OpenStreetMap.
Share-alike binds derived databases, not rendered frames. Every shipped coordinate is an OpenStreetMap highway node or an interpolation between two of them; 50,768 OSM ways were in the walked network.