Every satellite heat detection over Britain & Ireland, archived every 30 minutes

Data starts 1 January 2026 — live collection every 30 minutes since 9 August 2026, extended back through NASA’s FIRMS archive (reprocessed VIIRS detections, fetched once with an API key). Fires before that date never enter the archive.

Fire events, last 7 days

Detections within 3 km and 48 h of each other are grouped into one event, named after the ward or district nearest the fire’s centre — the place column says where the fire is, not where you are (moorland fires far from any postcode show as “open country”). Events with satellite heat in the last 24 h are listed first, then by peak power. Peak power is fire radiative power (FRP) — the strongest single pixel the satellites measured, in megawatts. Recurring clusters at industrial sites are heat the satellite genuinely measured; we label nothing automatically. Click any row to see that event on the map, or a column heading to sort. Showing the 40 most significant of 77 events; the full set is in the CSV.

Every archived detection, with timestamps, satellites, confidence and event assignment: download the 7-day CSV. For a single fire, click its event on the map — the popup links a CSV of that event’s complete detection history.

The ledger runs forward from here — because backwards failed our checks

Tracking is half the programme. Each substantial fire this tracker catches gets its burn scar measured from Sentinel-2 imagery two to four weeks after the heat stops: hectares from surface-reflectance change (dNBR), severity, and what kind of land burned — derived figures, never quoted estimates. We restrict ourselves to fresh scars because we tested the method to destruction first, and we print the results:

1. Where the method works — verified: measured promptly after summer fires, dNBR agrees with ground reports well. Tintwistle (Peak District, June 2026): we measured 784 ha against the fire service’s “over 600”. Rhinogydd (Eryri, July 2026): 927 ha measured vs 951 reported — a ratio of 0.97.
2. Control we ran and failed — the retrospective ledger: we tried to measure all 13 major named UK fires of 2026 season-to-date against three pre-registered pass/fail criteria, and failed all three. Spring fires (Marsden Moor, Elan Valley) could not be recovered months later — moorland regrows and the scar signal fades to near nothing (Elan Valley measured 1% of its reported 8,000 ha). And on Kinder Scout, a moor with no catalogued 2026 fire, the detector found ~730 ha of apparent “scar” — drought-browned vegetation crossing the burn threshold in a record-dry summer. A season-long retrospective would have been quietly wrong in both directions, so it does not ship.
3. Control we ran and failed — the tinder map: an earlier attempt to predict fire risk from satellite-measured vegetation dryness performed at chance level against where 2026 fires actually happened, and was killed before publication.

Fresh-scar measurements of events on this page will appear here once their post-fire imagery windows open — the first candidates from the current fires are measurable from late August 2026.

Method

1. Source: NASA FIRMS active-fire detections from the VIIRS instrument on Suomi NPP, NOAA-20 and NOAA-21 — 375 m pixels, day and night passes, ~1–3 h from overpass to publication. FIRMS publishes a rolling 24-hour window; our cron archives it every 30 minutes, so nothing is lost (live collection began 9 August 2026; a one-off backfill through the FIRMS archive API takes the data back to 1 January 2026).
2. Region: Britain and Ireland, all of both — fires ignore the border.
3. Events: detections within 3 km and 48 h of an event’s footprint join it; otherwise a new event opens. Event statistics (first/last seen, counts, peak FRP, centroid) are recomputed from the underlying detections every poll — derived, never hand-maintained. Names come from the nearest postcode’s ward and district (postcodes.io).
4. Known limits, stated plainly: detections are moments, not perimeters; small, short-lived or smouldering fires between passes are missed; cloud blinds the sensor; industrial heat sources recur as false wildfire lookalikes; and satellite “last heat” is not an all-clear — a fire can smoulder below detection.

Data: NASA FIRMS (VIIRS, LANCE) · postcodes.io. For live official information: your local fire and rescue service, the Met Office, and in an emergency 999. We did not produce the detections; we produced the archive and the join.

Is this a live fire alert service? Should I rely on it in an emergency?

No, twice over. Detections reach us ~1–3 hours after the satellite passes, satellites pass roughly six times a day (with gaps), and cloud hides fires entirely. If you can see fire or smoke, call 999. For official information use your fire and rescue service and the Met Office. This page is an archive and a map of what the satellites measured — useful for understanding fires, useless for escaping one.

Is each dot a fire? Is the dot the size of the fire?

Each dot is one satellite pixel (375 m across) in which the VIIRS instrument measured a strong heat anomaly at that moment. It is not a perimeter and not an area: a large moorland fire shows as dozens of dots across several passes, a small bonfire-scale heat source can be one dot, and persistent industrial heat — steelworks, refineries, flares — shows up as a cluster that recurs in the same spot week after week. We archive every detection and label nothing automatically; the recurring industrial clusters are visible on the map for exactly that reason.

Why is a fire I read about not shown here?

Four honest reasons: our data only starts on 1 January 2026, so earlier fires never enter it; the map shows the last 7 days by default (its window selector runs from the last 24 hours to the whole archive); cloud blocks the sensor completely; and a fire that burns low and smoulders between overpasses can be missed. Absence from this map is not evidence a fire did not happen.

What happens to the fires this tracker catches?

Each substantial event gets its burn scar measured from Sentinel-2 imagery two to four weeks after the heat stops — measured hectares from surface-reflectance change (dNBR), not quoted estimates. That is the one regime where our scar method verifiably works, and we know because we tested the alternative to destruction: a season-long retrospective ledger failed its pre-registered checks, and the failure is printed on this page rather than hidden.

More measurement from the same discipline: Sinking Britain · the hottest block in every city · the clay forecast · the water debt — or explore all our layers on the interactive map.