Data Sources & Attribution
OverWatts is built on open data. Every dataset and live feed it draws on is listed below with its publisher and licence. Seed infrastructure values are approximate and the map is not for operational use.
1 behind3 update due12 current15 live3 reference6 undated
4 of our sources have published more recently than the release we ship. Each is named below. Row counts from the build of 2026-07-31.
Infrastructure layers
- Power plants (32,429 sites ≥100 MW; status + commissioning years)Current— Global Energy Monitor — Global Integrated Power, March 2026CC BY 4.0 — attribution required · data modifiedSource release 2026-03 · 5 months old · 29,849 rows shippedUnit-level data aggregated to sites (sum capacity, dominant fuel, earliest start year, nuclear unit → reactor count). Major nuclear stations carry curated reactor detail merged on top. Modified by OverWatts. WRI Global Power Plant Database (CC BY 4.0) remains a fallback ingest.
- LNG terminals (1,198)Update due— Global Energy Monitor — Global Gas Infrastructure Tracker (LNG), September 2025CC BY 4.0 — attribution required · data modifiedSource release 2025-09 · 11 months old · 1,198 rows shippedImport/export capacity (Mtpa), status, start/retire years, owner. Fields mapped + coordinates rounded by OverWatts.
- Oil & gas pipelines (4,951 segments, real routes)Behind— Global Energy Monitor — Gas Pipelines (Nov 2025) + Oil/NGL Pipelines (Mar 2025)CC BY 4.0 — attribution required · data modifiedSource release 2025-03 · 17 months old · 4,951 rows shippedRoute geometry from GEM's GeoJSON, decimated to ≤120 vertices/feature and rounded to ~11 m. Fuel, status, length, start year retained.
- Oil & gas fields (7,055)Current— Global Energy Monitor — Global Oil & Gas Extraction Tracker (GOGET), March 2026CC BY 4.0 — attribution required · data modifiedSource release 2026-03 · 5 months old · 7,055 rows shippedStatus, discovery/FID/production-start years, basin, on/offshore, owner. Fields mapped + coordinates rounded by OverWatts.
- CC BY 4.0 — attribution required · data modifiedSource release 2026-05 · 3 months old · 5,382 rows shippedProduction (Mtpa), met/thermal split, reported methane, owner, opening/closing years. Fields mapped + coordinates rounded by OverWatts.
- Coal export terminals (519)Update due— Global Energy Monitor — Global Coal Terminals Tracker, December 2024CC BY 4.0 — attribution required · data modifiedSource release 2024-12 · 20 months old · 519 rows shippedThroughput (Mt/yr), product type, status, start/retire years. Fields mapped + coordinates rounded by OverWatts.
- Methane plumes (3,473 satellite super-emitters)Update due— Global Energy Monitor — Global Methane Emitters Tracker (GMET V3), December 2025CC BY 4.0 — attribution required · data modifiedSource release 2025-12 · 8 months old · 3,473 rows shippedPer-plume emission rate (kg/hr), observation date, attributed infrastructure type. Fields mapped + coordinates rounded by OverWatts.
- Refineries · hydrogen · CCS · storage (curated seeds)Undated— Global Energy Monitor (curated seed, full tracker ingest pending)Seed: approximate, hand-entered · GEM basis: CC BY 4.0Flagship facilities hand-entered from GEM trackers; not yet replaced by a full tracker ingest in this release.
- ODbL — attribution + share-alike on derived databasesIngested 2026-07-31 (upstream is edited continuously, so this is the vintage) · 3,269 rows shippedBreadth over depth: names, coords, operators; capacity rarely mapped. Kept separable from CC BY layers so exports carry the right per-source licence.
- CC0 — no restriction (Wikidata) · alias keys derived from OSM tags (ODbL)Source release 2026-07-31 · 0 days oldOSM's free-text operator tag carried 694 distinct strings for this layer, including five spellings of STACK Infrastructure — so grouping or counting by operator was meaningless. Every operator:wikidata QID in the corpus is resolved against Wikidata for a canonical name, parent organisation (P749) and country. Aliases are then **learned** from the sites tagged with both a QID and free text, and replayed over the sites that have only free text. Nothing is fuzzy-matched: an operator that matches no alias keeps its as-mapped spelling, and a parent with no English label is dropped rather than shown as a bare QID. Hand-curated corrections (src/data/operatorOverrides.json) supply parents Wikidata lacks — CoreSite→American Tower, Interxion→Digital Realty — each citing the operator's own investor material. Refresh with npm run ingest:operators.
- AI compute clusters (power, owner, chip counts)Current— Epoch AI Data Centers · geocoded via Nominatim (OSM)CC BY 4.0 (Epoch AI) · ODbL for the geocoded coordinatesIngested 2026-07-23 (upstream is edited continuously, so this is the vintage) · 44 rows shippedEpoch AI's maintained register of large AI training clusters: current power draw in MW, owner, H100-equivalent chip counts and capital cost, sourced per site. The dataset carries a postal address and no coordinates, so each site is geocoded at ingest against Nominatim under its 1 req/s policy. **Precision is recorded, not assumed** — roughly half resolve to the plot and are marked `site`, the rest to a town centroid and are marked `area`; anything that resolved no better than a county, or to a place unrelated to the address asked for, is dropped rather than pinned. Sites already covered by the curated frontier campuses are skipped, since those carry full-build capacity and a lifecycle status that a current-power snapshot does not replace. Re-ingest with npm run ingest:ai-datacentres.
- Transmission lines & substationsCurrent— Curated seed (OpenStreetMap power=line / power=substation ingest available)Seed: original work · OSM: ODbL — attribution + share-alikeIngested 2026-06-22 (upstream is edited continuously, so this is the vintage) · 94,064 rows shippedCurated seeds of ~35 major interconnectors + ~16 converter stations; npm run ingest:transmission / :substation add OSM geometry (≥ MIN_KV).
- Grid-scale energy storage — pumped hydro (965 schemes; 206 GW operating)Current— Global Energy Monitor — Global Integrated Power, March 2026CC BY 4.0 — attribution required · data modifiedSource release 2026-03 · 5 months old · 965 rows shippedPumped-storage units filtered out of the integrated power tracker and aggregated to sites by GEM location ID, cancelled and shelved schemes dropped. Modified by OverWatts. Batteries are not yet in this layer: v5 named Sandia's GESDB and the licence audit blocked it (no redistribution grant), so BESS awaits a GEM storage-tracker release in data-sources/. Every site here also appears in the power-plant layer as hydro — a pumped-storage station generates and stores, so the two counts must never be summed. Re-ingest with npm run ingest:grid-storage.
- Cross-border interconnectors (704 pathways; 999 GW existing transfer capacity)Reference— Global Transmission Database v0.1 — Brinkerink et al., Zenodo, 2023CC BY 4.0 — attribution + dataset citation requiredSource release 2023-11 · 2.7 years old · 704 rows shippedCountry-to-country transfer capacity worldwide plus sub-national pathways for twelve large countries (AU, BR, CA, CN, IN, ID, JP, PH, RU, TH, US, VN), land-based and subsea, existing and planned. Cite as: Brinkerink, M., Sherman, G., Osei-Owusu, S., Mohanty, R., Majid, A., Barnes, T., Niet, T., Shivakumar, A. & Mayfield, E. (2023). Global Transmission Database (v0.1) [Data set]. Zenodo. https://doi.org/10.5281/zenodo.10063446 — GEOMETRY IS SCHEMATIC: GTD publishes capacity between nodes, not cable routes, so each pathway renders as a straight dashed line between the two nodes' reference points and is labelled as such. Modified by OverWatts: zero-capacity pathways dropped, multi-circuit voltage reduced to its maximum, named GB cables annotated from Elexon BMRS. Re-ingest with npm run ingest:interconnectors.
- Per-country grid statistics (mapped line-km by voltage, substations, coverage grade)Current— OverWatts — derived from the OpenStreetMap transmission + substation ingestsODbL (derived from OSM) — attribution + share-alikeIngested 2026-07-26 (upstream is edited continuously, so this is the vintage) · 153 rows shippednpm run derive:grid-stats measures every ingested line segment by haversine and attributes it to the country containing its midpoint (OSM ways carry no country tag). Figures are *mapped* line-km, not installed line-km, and the coverage grade ranks how completely OSM has drawn a country's network — never how much network the country has. Backs /country/[iso]/grid.
- Public domain
Live feeds
- Free key; verify redistribution ToSTerrestrial/coastal coverage only — vessels go dark mid-ocean.
- National oil stocks (67 countries, monthly)Current— Joint Organisations Data Initiative (JODI-Oil World Database)Free — attribution requiredIngested 2026-07-22 (upstream is edited continuously, so this is the vintage) · 67 rows shippedSelf-reported closing stock levels held on national territory — commercial plus government/strategic reserves (US crude includes the SPR). Reported, not observed: these are national submissions, not satellite tank readings. Coverage is uneven and gaps are left blank rather than interpolated — China reports no stocks at all. Republished monthly, roughly a month in arrears. Re-ingest with npm run ingest:jodi.
- Free with a registered API key — attribution requiredDaily percent-full of underground gas storage against working gas volume, plus injection/withdrawal rates, for the EU aggregate and 21 reporting countries. Fetched server-side with AGSI_API_KEY so no visitor IP reaches GIE. One request per country — AGSI+ has no bulk form and silently ignores a comma-separated country filter. Percent full is seasonal by design (storage fills all summer and drains all winter), so the level is always shown beside its day-on-day trend.
- EU LNG terminal send-out and inventory (31 live terminals)Live— GIE ALSI (Gas Infrastructure Europe)Free with a registered API key — attribution requiredDaily send-out, tank inventory, declared regasification capacity and contracted share for every live European LNG import terminal, matched onto the Global Energy Monitor asset it belongs to. Same API family and the same AGSI_API_KEY as gas storage, fetched server-side so no visitor IP reaches GIE. Coverage is EU-only: Isle of Grain and South Hook stopped reporting to ALSI in early 2022 and their series are closed, so Great Britain appears on the platform as history rather than live data. Superseded rows ("Prior to …") and Spain's virtual balancing tank are excluded — counting them would double-count Spain and Italy. Utilisation is send-out against declared capacity and is not a demand signal on its own: Europe holds spare regas capacity by design, so the booked share is always shown beside it.
- Free with a registered token — attribution + platform termsEvery generation unit ≥100 MW currently out or derated, as filed by the TSO under Regulation 543/2013 — the operators' own regulatory record, not a news report. Fetched server-side with ENTSOE_TOKEN so no visitor IP reaches ENTSO-E. Unavailable MW is nominal power minus the capacity still available, so a derating counts only for what it lost. Cancelled and withdrawn filings are returned alongside live ones and are excluded here. Planned maintenance is reported apart from unplanned outages: a European summer runs tens of GW of scheduled nuclear work, and a combined total reads as a crisis that is not happening.
- Free, keyless — attribution + GIE termsStorage, LNG and transmission operators' own REMIT filings (Reg. 1227/2011) of capacity out of service — the gas-side twin of the ENTSO-E generation-outage feed, joined to the atlas asset each one concerns via its EIC. As a regulatory record rather than a news report, a filing bypasses the corroboration pass. Capacity is normalised to GWh/d from the kWh/h and GWh/d the feed mixes, and revisions of the same filing (republished with the same series and left marked Active) are collapsed to the latest, or a single extended outage would be counted several times over. Planned maintenance is reported apart from unplanned. GIE serves only the 50 most-recently-submitted filings — its pagination parameter is not honoured and the endpoint behind it needs credentials GIE does not issue publicly — so this is a rolling window over recent activity, not the complete active set. Only LNG-terminal filings appear on the map (the atlas has no underground-storage layer to pin a storage notice to); the rest render in the card.
- CC-BY-4.0 — attribution required, commercial use permittedIngested 2026-07-23 (upstream is edited continuously, so this is the vintage) · 214 rows shippedElectricity actually generated by fuel (TWh), total generation, demand, net imports, clean share and power-sector carbon intensity, on the country pages, the dossier export and the country_energy_profile MCP tool. This is the answer to 'what did a country produce', which installed-capacity data (GEM/WRI) cannot give — generated TWh and nameplate GW are different quantities and are kept in separate sections so neither is read as the other. Ingested from Ember's yearly bulk release by npm run ingest:ember (the JODI pattern — a compact committed JSON, not a runtime dependency on data that changes about twice a month). Each country is pinned to its own most recent reported year (latest 2025 for ~91 major economies, 2024 for ~105 more), carried with every figure. Refreshed quarterly.
- Free, keyless — attribution + platform termsNet physical gas flow between European countries — Norwegian exports, TurkStream into the Balkans, TAP into Italy, Baltic Pipe into Poland — the numbers a gas desk watches each morning. Aggregated from ENTSOG's per-point daily Physical Flow, joined to the interconnection catalogue so each point resolves to a country pair, and netted for reverse flow (a border carrying gas both ways reports the net transfer, not the gross). The point→country join is bidirectional: a physical flow reported only by the downstream TSO (much of the Norwegian supply) is captured, not just the upstream side. A daily figure that settles over ~2 days, so the covered gas day travels with the numbers. No map layer: ENTSOG's point coordinates are schematic canvas positions, not geographic (a Portuguese point sits in the Pacific), so a border-point map is a separate curation task rather than a fake-coordinate render.
- Free, keyless — per-provider terms
- Grid mix — US / Canada / GBLive— EIA · IESO · National Grid ESOFree / public — per-provider terms
- Free, keyless — commercial use permitted with the BMRS attribution stringDirectional power flow over each named GB electricity interconnector (IFA, IFA2, ElecLink, BritNed, Nemo Link, North Sea Link, Viking Link, East-West, Greenlink, Moyle) from the BMRS FUELINST dataset at 5-minute resolution — the flow per individual cable, which the grid-mix card's single net figure cannot show. Positive is import into Great Britain, negative is export, and a cable can reverse within the hour. Elexon grants a worldwide, royalty-free, perpetual licence including commercial use, conditional on the attribution "Contains BMRS data © Elexon Limited copyright and database right", which is carried in every API response and on the card.
- Free personal API tier — non-commercial
- Free, public
- Free with an API token — attribution requiredHuman-reviewed annotations of national internet outages, each with a cause (power outage, cable cut, government-directed) and the countries affected. Used as the country-scoped second observation behind an event's `corroborated` label: a nationwide traffic collapse attributed to loss of power is often the fastest public confirmation that a grid went down. Corroborates grid outages only — it says nothing about a refinery fire — and matches are labelled country-scope, never given a false distance.
- Free, keylessGeocoded CAMEO conflict events over the last two hours, binned onto a 0.5° grid; hover/click any hotspot for its place + activity. Filtered hard, because GDELT is machine-coded from news and codes crime reports, court rulings and politics as conflict: city-level geocodes only (country and state centroids are dropped), each source story counted once, ordinary-crime codes excluded, military-specific codes weighted above merely martial-sounding ones, and two independent news outlets required. Expect ~10–20 hotspots — a quiet theatre drops out until it is corroborated. One artifact remains by design: foreign-policy coverage is geocoded to the capital it was announced from, so a strike decision can light up the seat of government rather than the target. Replaced the curated theatre-centroid set in v0.53.0.
- Free, keyless — links to publishers
- Public domain (U.S. Government work) — redistributableDaily spot prices, fetched server-side with an EIA key. **Published ~weekly with a multi-day lag**, so the readout carries its age and turns amber past four days — it is not a live market price and is not represented as one. Yahoo Finance front-month futures led this feed from 0.59.10 and were reverted: near-live, but an undocumented endpoint whose ToS bars commercial redistribution, which this site cannot rely on now that it carries advertising. **TTF is no longer shown** — EIA publishes no European gas series, and no free, redistributable TTF benchmark has been found.
- IMF Copyright and Usage terms — attribution required, commercial reuse terms check pendingDaily AIS-derived transit calls for 28 world chokepoints, published by the IMF as an open ArcGIS feature service (no key). We fetch it server-side, so a visitor's browser never contacts the IMF. The dataset's own licence field points at imf.org/external/terms.htm, which permits attributed reuse but reserves commercial redistribution — flagged pending a read of those terms before any paid export includes this series. PortWatch's chokepoint geographies are close to but not identical with ours (the Øresund is not the whole Danish Straits; the Bosporus is not the whole Turkish Straits), and each mismatch is stated on the readout rather than silently averaged away.
- Chokepoint maritime security risk levelsCurrent— Derived from Joint Maritime Information Centre (JMIC) advisory information · via UKMTOCrown copyright (UK MoD) / coalition product. No licence granted or sought; JMIC neither authorises nor endorses OverWatts. Derived facts only (region, level, date).Source release 2026-07-26 · 5 days oldWe publish only the derived fact — region, assessed threat level, assessment date — with attribution and a link to the publisher. The note's PDFs are not mirrored and its regional narrative prose is not reproduced. Levels are transcribed by hand from the latest note (npm run match:jmic proposes a diff; a human promotes it), and the five chokepoints outside JMIC's advisory area show no level rather than a default. Levels older than 14 days are withdrawn from display entirely and replaced with a pointer to UKMTO, rather than shown as current. Confirmed with JMIC (July 2026): they raise no objection to this use but grant no permission and endorse nothing here. An assessment, not ground truth — and not for navigation or operational decisions.
- Sanctioned-vessel listsCurrent— OFAC SDN · EU FSF · UK OFSIPublic government dataSource release 2026-06 · 2 months oldVessel entries matched to live AIS by IMO/MMSI/name.
Map & weather
- Free, no key — attribution required
- Place labels (raster)Undated— CARTO · OpenStreetMapFree tier — attribution required
- Public domainSource release 2022-05 · 4.2 years old1:50m land for the ocean-overlay coastline mask; 1:110m admin-0 countries as an invisible click hit-target that resolves a tap to the country containing it.
- CC BY 4.0 — attribution required · data modified147 rows shippedWorld ECS, retrieved from the Marine Regions GeoServer WFS (MarineRegions:ecs). Modifications: geometry Douglas–Peucker simplified to ~1.1 km and coordinates rounded to 3 dp; the `pol_type` legal status folded into three classes (CLCS-recommended/deposited, submitted, overlapping/non-UNCLOS). Depicts claims as submitted — not an adjudication of any boundary. Re-ingest with npm run ingest:ecs.
- ODbL — attribution requiredThe basemap's own `boundary` layer, filtered to features tagged maritime=1. Coverage is OSM-quality: good on agreed limits, incomplete on contested ones.
- Free — personal / educational use only
- CC BY 4.0 — free tier non-commercial
- Marine (waves, sea-surface temp)LiveOff by default— Open-Meteo MarineCC BY 4.0 — free tier non-commercial