Thermal Anomalies

Every hot spot a weather satellite actually sees — a wildfire front, a gas flare, a volcano’s heat — is organized into the thermal_anomaly Eventdex. Each slot is one event, assembled from the individual satellite fire-pixel detections that make it up: where it burned, how long, and how much energy it radiated. These are measured thermal anomalies the sensor observed, not a model of where fire might be.

The dex

One Eventdex holds the record. A satellite doesn’t see “a fire” — it sees a scatter of hot pixels, hundreds or thousands of them across the life of a single blaze. This dex clusters those detections in space and time: detections close together (within about a kilometre) and close in time (within a day) belong to the same event. A slot is therefore a whole event — its footprint, its duration, and its full set of underlying detections — not a single pixel.

Coverage is honest about its age: this dex is newly launched. It opens with a seed from the August 2020 Northern California fire siege — including the SCU and LNU Lightning Complexes, recovered as coherent multi-day events — and a live global feed that grows it forward every day. The full historical backfill, reaching back to the start of the satellite fire record (MODIS in 2000, VIIRS in 2012), is in progress.

Not just wildfire

FIRMS detects heat, and heat has more than one source. Roughly 95% of detections are vegetation fires, but the rest are real too, and each event is tagged with what it most likely is:

• Vegetation fire — wildfires, agricultural and land-clearing burns
• Gas flare / static source — persistent industrial burn-off at oil and gas fields
• Volcano — active volcanic thermal output
• Offshore — heat over water, typically offshore flaring

Persistent sources — a flare that burns for years — are split into one slot per calendar year, so a permanent hot spot doesn’t masquerade as one endless event.

Provenance

Every event traces to NASA FIRMS — the Fire Information for Resource Management System, which distributes active-fire detections from the VIIRS (375 m) and MODIS (1 km) instruments aboard NASA and NOAA satellites. The data is public domain (U.S. Government), and FIRMS is the direct upstream source that most third-party fire feeds simply redistribute. It is registered against the thermal_anomaly dex, so every detection carries a direct line back to the satellite overpass that recorded it.

What’s in a record

Each event slot carries:

• Footprint — centroid, bounding box, and spatial extent in kilometres
• Time — first and last detection (UTC) and total duration
• Intensity — peak, total, and mean fire radiative power, in megawatts
• Subtype — vegetation fire, gas flare, volcano, or offshore, where classified
• Sensors — which satellites and instruments contributed
• Detections — every underlying fire-pixel: position, time, radiative power, brightness temperature, and detection confidence

How to read it honestly

FIRMS is a fire-detection-class thermal sensor, tuned to catch combustion-scale hot spots — it is not a general thermometer, and it does not see heat below its fire threshold. Two more honest notes:

• The live edge is unclassified. FIRMS only assigns a detection’s type (fire / flare / volcano) in its standard-processing archive, which lags the near-real-time feed by two to three months. Very recent events are held as unclassified thermal anomalies until that classification catches up.
• Event boundaries are a choice. The clustering distance and time gap decide where one event ends and the next begins; a cluster of fire fronts that grew together — a “complex” — is held as a single event. Those thresholds are recorded on every slot.