What the instrument is
The Visible Infrared Imaging Radiometer Suite flies on three sun-synchronous polar orbiting spacecraft: Suomi NPP, NOAA-20 and NOAA-21. Suomi NPP crosses the equator at about 13:30 and 01:30 local time, with the other two about 50 minutes ahead.
The detection product here is the 375 m imagery-band product. Its primary channel is I4 at about 3.74 microns, compared against I5 at about 11 microns for background. The 375 m product yields roughly three times more fire pixels by day and twenty-five times more at night than the older 750 m baseline, with better fire-perimeter consistency on multi-day burns.
The swath is about 3,000 km wide, which gives global coverage every twelve hours or less per spacecraft, and three to four looks a day at mid latitudes across the constellation.
One detail that affects how detections should be drawn: the pixel is not a constant size. The footprint grows from about 375 by 375 m at nadir to roughly 795 by 784 m at the swath edge, a two-fold linear growth. A detection plotted as a fixed-size square is wrong by a factor of two at the edge of a pass.
Detection is contextual, not a temperature threshold
The algorithm does not ask whether a pixel exceeds some fixed temperature. For each candidate it compares the mid-infrared brightness temperature, and the difference between the mid-infrared and thermal channels, against statistics of the surrounding fire-free background, together with absolute threshold tests and explicit screens for cloud, water, sun glint, desert and bright surfaces. Nominal confidence requires a temperature anomaly greater than 15 K relative to background.
Three framings keep this honest:
- It detects radiant heat at the instant of overpass. Not flame, not smoke, not burned area. A fire that started after the pass, or was out before it, does not appear at all.
- It detects thermal anomalies. Volcanoes, gas flares and industrial plant are not failures of the algorithm, they are in-scope detections of hot things.
- Cloud and dense smoke block it entirely. A cloud-topped fire is simply absent from the feed, which is a very different statement from "no fire".
What confidence means, and what it does not
Detections carry a confidence of low, nominal or high. Two things must be said about that, because both are counterintuitive.
Confidence is not a probability. The product documentation calls it a heuristic measure intended to help users gauge the quality of individual pixels. There is no reading under which high means ninety percent.
High confidence means hot, not important. The high-confidence class is defined by saturated pixels, meaning the detector pegged, which happens at 367 K on the primary channel. A genuine small grass fire can be nominal confidence, and a large industrial gas flare can be high. About 8 percent of daytime fire pixels and about 1 percent of nighttime ones saturate.
Low-confidence detections over water, and those associated with the South Atlantic Magnetic Anomaly, are removed from the distributed records entirely, because they are predominantly spurious. Some genuine fires are lost with them.
Fire radiative power, and why not to add it up
Fire radiative power is the pixel-integrated rate of radiative energy emission, in megawatts. It is the most useful single number attached to a detection and the easiest to misuse.
The retrieval combines resolutions: a single coarser retrieval is assigned to its coincident 375 m fire pixel, or split between several coincident sub-pixels. So two adjacent detections may each be carrying half of one physical measurement. They are not two independent fires of that power.
On top of that, some fire pixels legitimately carry zero or null power, and pixels flagged as residual bow-tie duplicates at the swath edge are redundant samples of the same ground. The product documentation is explicit that the bow-tie flag must be used when computing emissions estimates, precisely to avoid double counting.
A "total power in view" figure is therefore not a physical quantity unless duplicates are excluded and nulls handled. This layer shows power per detection, with units, and does not offer a naive sum.
Why a persistent dot may not be a fire
This is the limitation most worth stating plainly, and it is a genuine constraint of the real-time feed rather than a shortcoming of this site.
The standard-quality product carries a field inferring the source of each hotspot: presumed vegetation fire, active volcano, other static land source, or offshore detection. That field exists only in standard-quality data, not in the real-time versions. A worldwide near-real-time layer therefore cannot tell you that a persistent dot is a gas flare, a volcano or a steel mill.
Known and documented sources of detections that are not wildfires: gas flares, which the algorithm deliberately processes water pixels in order to catch; reflective rooftops on large industrial buildings under sun glint; large industrial parks surrounded by cooler background; volcanoes; energetic particle hits over the South Atlantic Magnetic Anomaly; and superheated smoke plumes displaced outside the fire perimeter by parallax at high scan angles. Measured commission error for nominal-confidence pixels is below about 1.2 percent overall, which is good, but a persistent industrial source reappearing every pass is not an error at all, it is the instrument working correctly.
Why it matters to aviation
Smoke and visibility. Smoke degrades flight visibility and is reported in aviation weather under its own qualifier. Pyroconvection on large fires produces strong vertical motion and, in extreme cases, pyrocumulonimbus with turbulence and lightning well above the fire itself.
Temporary flight restrictions. In US airspace, wildfire restrictions are issued under 14 CFR 91.137, whose purposes include protecting a surface hazard that low-flying aircraft would magnify, providing a safe environment for disaster relief aircraft engaged in suppression, and preventing unsafe congestion of sightseeing aircraft.
Firefighting aircraft operations. Air tankers and helicopters work at low altitude in reduced visibility, in congested orbits over the fire, often under non-standard procedures. This is why unauthorized drone incursions over fires routinely ground air operations entirely.
The limit is the important part: a detection here is a satellite observation up to about three hours old at a 375 m to 800 m footprint, with no restriction status, no fire perimeter, and no indication whether aircraft are operating. NOTAMs and TFRs are the authority for airspace status.
What this layer does not tell you
Every data source has an edge. These are the ones that matter for reading this layer correctly.
- It detects radiant heat at the moment of overpass, not flame, smoke or burned area. A fire between passes is unobserved.
- Cloud and dense smoke block detection completely. Absence of a detection is not absence of fire.
- The real-time feed cannot distinguish a wildfire from a gas flare, volcano or industrial heat source, because the field that does so exists only in standard-quality data.
- Confidence is a heuristic, not a probability, and high confidence is defined by sensor saturation, so it means hot rather than significant.
- Fire radiative power can be split across adjacent pixels and duplicated at swath edge, so summing it without handling duplicates and nulls produces a meaningless number.
- The pixel footprint doubles in linear size from nadir to swath edge, so detection positions near the edge of a pass are correspondingly coarse.
- Newest global detections may be up to about 3 hours old. This is not a real-time fire map.
AeroScope is an situational awareness and research tool. Nothing here is approved for air traffic control, navigation, flight planning or collision avoidance. For operational decisions, use the authoritative source named above.