Aursee · method · night

Fires in the dark: how satellites find them at night

At night the ground cools and the sunlight is gone. A flame stays at a thousand kelvin. In the mid-infrared that gap is four orders of magnitude, which is why a fire a few metres across can light up a 375 m pixel seen from about 830 km up.

Planck’s law, emissivity 1, no atmosphere. Shaded: the VIIRS bands the fire algorithm reads (I4 3.55–3.93 µm, M13 3.973–4.128 µm, I5 10.5–12.4 µm). Dots mark each curve’s peak (Wien: 2,898 µm·K / T).

One band sees the flame, one sees the ground

×12,583How much brighter a 1000 K flame is than 290 K ground at 3.74 µm, the centre of the I4 band.

×30The same comparison at 11.45 µm, the centre of I5. The thermal band barely notices a small fire.

derivedA burning front emits most strongly near 3 µm; the ground at night peaks near 10 µm. The fire test looks for exactly that shape: a pixel that jumps in the mid-infrared while its thermal band stays close to its neighbours. Sunlight reflected at 3.7 µm raises and scrambles the daytime background, so by day the algorithm needs a stronger signal and extra reflectance checks. At night that noise is gone.

B(λ,T) = 2hc² / λ⁵ · 1 / (e^hc/λkT − 1) → L_pixel = p·B(λ,T_fire) + (1−p)·B(λ,T_ground)

How small a fire the night pass can see

derivedA flame front at 1000 K covering 28 m², 0.020% of one pixel, lifts a 290 K night pixel to 320 K at 3.74 µm: the fixed night threshold of the VIIRS 375 m algorithm. Its 11.45 µm temperature moves by 0.4 K.

The square on the left is that flame, to scale. The algorithm’s contextual tests, which compare each pixel with an 11×11 to 31×31 window around it, reach further down; the published figure is a fire fraction of about 10⁻⁴.

Flame temperature	Share of pixel	Area	Change at 11.45 µm
600 K	0.2603%	366 m²	+1.63 K
800 K	0.0520%	73 m²	+0.64 K
1000 K	0.0196%	28 m²	+0.37 K
1200 K	0.0101%	14 m²	+0.26 K

Error bar: these are monochromatic at the band centre, with emissivity 1 and no atmosphere, over a uniform 290 K background at nadir. Band response, water vapour, a warmer or cooler surface and the larger pixel towards the swath edge each move the numbers; read them as the order of magnitude, not a detection limit.

When the satellites look

detectedThree identical VIIRS instruments fly the same sun-synchronous orbit: Suomi NPP, NOAA-20 and NOAA-21. The orbit crosses the equator at about 13:30 local solar time going north and 01:30 going south, so every place is seen once around midday and once in the middle of the night.

derivedThe 3,060 km swath spreads that by about ±55 minutes at the equator. Towards the poles swaths overlap and passes multiply.

Instrument: VIIRS, 22 bands
Platforms: Suomi NPP · NOAA-20 · NOAA-21
Orbit: sun-synchronous, 101.5 min, about 830 km up
Equator crossing: 13:30 ↑ · 01:30 ↓ local solar
Swath: 3,060 km
Pixel: 375 m at nadir, I-bands
Fire band: I4 3.55–3.93 µm
Thermal band: I5 10.5–12.4 µm
Fire power band: M13 3.973–4.128 µm, 750 m
I4 saturation: about 358 K effective
Night: solar zenith ≥ 90°
Night fixed test: BT4 > 320 K

Band limits, saturation, swath and tests: VIIRS 375 m active fire ATBD. Orbit period: SGP4 from CelesTrak elements, all three platforms within 0.01 min of 101.50.

What “burning in darkness” means on the map

derivedA fire is counted as burning in darkness when the sun is below the horizon at its position at the moment you are looking. Your browser computes the point where the sun is overhead from the date and the UTC time, and draws the same line as the terminator on the globe, so the count moves with the night. Error bar: the solar declination is a sine approximation and the equation of time is left out, so the line can sit up to about 16 minutes of sun movement from the true terminator.

detected“Seen on night pass” on a fire is different: the share of its detections that FIRMS flags as night acquisitions, solar zenith 90° or more at the pixel when the satellite passed.

Today’s feed, split by pass

34%of 165,038 detections in the last 24 hours came from a night pass.

7,550of 28,511 fires were seen at night at least once; 4,620 only at night.

1.1 MWmedian power per detection for fires seen only at night, against 5.3 MW for fires seen only by day.

NASA FIRMS VIIRS 375 m, Suomi NPP, NOAA-20, NOAA-21, 24 hours to 2026-10-12 03:49 UTC, grouped into fires by Aursee. The power gap mixes two effects that this feed cannot separate: fires burn less intensely at night, and the night test picks up weaker ones.

What the dark does not show

Fire under thick cloud or dense smoke at pass time.
Fire that starts and goes out between two passes.
Which kind of heat it is. Gas flares, volcanoes and some industrial sites are hot at night too, and a detection alone does not tell them from a vegetation fire.
The power of the brightest pixels from I4 alone: it saturates near 358 K, so fire radiative power comes from M13 at 750 m. See fire radiative power.

Sources

Schroeder, W., Oliva, P., Giglio, L., Csiszar, I. (2014). The New VIIRS 375 m active fire detection data product: algorithm description and initial assessment. Remote Sensing of Environment 143, 85–96.
VIIRS 375 m Active Fire Algorithm Theoretical Basis Document, NASA GSFC.
NASA FIRMS, LANCE near-real-time VIIRS 375 m active fires.
CelesTrak orbital elements, propagated with SGP4.