GMWID

GeoMermaids World Infrastructures Dataset

The world's power grids, telecom masts, pipelines, wells and water works, from OpenStreetMap, measured against authoritative open sources and completed where OpenStreetMap falls short. Every row says where it comes from, and every build publishes how OpenStreetMap compares with each source.

GeoParquet 2.0, one whole-world file per layer, rebuilt every week. Query it from a URL with DuckDB, no account, no download.

Why enrich OpenStreetMap

OpenStreetMap is already remarkable for infrastructure: in the United States it holds 97.8% of the generating capacity the Energy Information Administration reports. What it lacks is the last few percent (recent and small plants, remote wells, offshore assets), and a way to know how complete it is. GMWID adds both: the missing features, from sources that are authoritative for them, and a published measure of completeness for every layer that has a reference.

OpenStreetMap stays the base and is never overwritten. What GMWID adds is additive, labelled, and removable with one filter.

How the enrichment works

1. Pick an authoritative open source per layerA national register or inventory that is the reference for that kind of asset (EIA-860M for US power plants, the NSTA for UK offshore oil and gas), with a licence that allows open redistribution, and never a source that was itself built from OpenStreetMap.
2. Match, strictest test firstA feature counts as present in OpenStreetMap when OpenStreetMap carries its identifier (a ref:US:EIA, a UK well registration number, a pipeline PL number). Failing that, by location: inside an OpenStreetMap outline, or within a distance suited to the asset (25 m for a well, 500 m for a platform, 2 km for a power plant). Borderline cases count as present: a missed addition is better than a duplicate.
3. Add only what is missingUnmatched features become rows with the layer's exact columns, the tags a mapper would have used, and origin set to the source (eia, nsta). They have no osm_id. Where a better geometry exists (a solar farm's outline instead of a point), it is used.
4. Publish the measureEvery build writes _coverage.json: how many features each source reports, how many OpenStreetMap has (by identifier or by location), and what was added. It is shown below, live.
5. Credit, and fail loudlyEvery source is credited in the attribution file and the catalog. If a source cannot be downloaded, the build stops and last week's data stays online, rather than publishing a silently thinner dataset.

Coverage, measured every build

Read live from this week's _coverage.json. Bars are split into what OpenStreetMap already has, found by identifier or by location, and what GMWID adds.

What is inside

27 layers in four groups, with the data model of Open Infrastructure Map: voltages parsed to kV per circuit, outputs to MW, plants and substations assembled from their OpenStreetMap relations, and every OpenStreetMap tag kept. Row counts are this week's.

Layers marked OpenStreetMap only have no open authoritative reference anywhere we could find (pylons outside France, switchgear, street cabinets, pipeline markers, drinking water plants): OpenStreetMap is the best open source there is for them.

Sources and licences

The dataset in gmwid/latest/ is a derived database of OpenStreetMap and is distributed under the ODbL 1.0: free to use, commercially included, with attribution and share-alike. A source joins it only if its own licence allows the same (public domain, CC0, CC BY, government open licences). The full credits of each build are in ATTRIBUTION.txt.

Why some rows live apart

Abandoned features, kept apart

What is next

Every layer was surveyed for open authoritative sources. The next additions, in order:

1. Europe, matched by identifier: ANFR (French masts and water towers), Sandre and the EEA's Urban Waste Water Treatment reporting (wastewater plants), the UK Renewable Energy Planning Database, Germany's Marktstammdatenregister (plants and wind turbines).
2. United States: EPA NPDES (wastewater plants).
3. France's distribution grid, continued: Enedis's overhead medium-voltage lines and poles.
4. The Global South: ANEEL (Brazil's power plants), the Water Point Data Exchange (wells, mostly in Africa), HydroWASTE (wastewater plants outside Europe and the US).

Know a source we should use, or one you need measured? Tell us.

Get the data

https://parquetry.geomermaids.com/gmwid/latest/<layer>.parquet
s3://parquetry/gmwid/latest/<layer>.parquet   (endpoint s3.geomermaids.com, anonymous)

Each file is Hilbert-sorted in row groups of 32,000: a bbox filter reads only the row groups it touches, and a filter on country and state (GAUL first-level units as name slugs, texas, bayern) skips most of the others.

What GMWID added to US power plants

SELECT origin, count(*) AS plants, round(sum(output_mw_estimated)) AS mw
FROM read_parquet('https://parquetry.geomermaids.com/gmwid/latest/power_plant.parquet')
WHERE country = 'US'
GROUP BY origin;

OpenStreetMap only, whatever else is added later

SELECT * FROM read_parquet('https://parquetry.geomermaids.com/gmwid/latest/power_plant.parquet')
WHERE origin = 'osm';

UK wells, with the NSTA rows (non-commercial use only)

SELECT origin, lifecycle, count(*) AS wells FROM (
  SELECT * FROM read_parquet('https://parquetry.geomermaids.com/gmwid/latest/petroleum_well.parquet')
  WHERE country = 'GB'
  UNION ALL BY NAME
  SELECT * FROM read_parquet('https://parquetry.geomermaids.com/gmwid/nsta/petroleum_well.parquet')
)
GROUP BY ALL ORDER BY ALL;

• Browse the files: parquetry.geomermaids.com/gmwid/
• Portolan catalog (STAC, one collection per layer, column docs, agent guides): gmwid/catalog/
• On a map: GeoPQ Workbench has GMWID built in as a repository.
• Abandoned features: gmwid/latest/abandoned/, one file per layer
• Region and state slugs: _regions.json