Active fire detection and burned-area mapping
Thermal anomaly detection from VIIRS and MODIS locates active fires within hours of ignition, while post-fire NBR differencing and SAR coherence map what burned after the smoke clears. Each method has hard limits that operational users must understand.
Sensors
- VIIRS (Suomi-NPP / NOAA-20): 375 m active fire pixels in the I-band (3.74 µm and 11.45 µm); global daily coverage from each satellite, giving two overpasses per day per location when both platforms are used together. Detects fires roughly four times smaller in area than MODIS can resolve.
- MODIS (Terra / Aqua): 1 km thermal anomaly detection at 3.96 µm and 11 µm. Long archive from 2000 (Terra) and 2002 (Aqua) makes it the backbone of multi-decadal fire climatology. Minimum detectable flaming area is roughly 50–100 m² under ideal conditions, but at 1 km pixel scale small fires are easily diluted.
- Sentinel-2 MSI: 10–20 m multispectral imagery including SWIR bands at 1610 nm (Band 11) and 2190 nm (Band 12), which are the basis for the Normalised Burn Ratio. 5-day revisit at the equator (2–3 days at mid-latitudes with both satellites). Cloud and persistent smoke block acquisition entirely.
- Landsat 8 / 9 OLI-TIRS: 30 m SWIR bands (Band 6 at 1570 nm, Band 7 at 2110 nm) for NBR computation; 16-day single-satellite revisit, 8-day combined. Archive back to 1972 (Landsat 1) for long-term burn scar history. Thermal band (Band 10) at 100 m resampled to 30 m can flag residual heat.
- Sentinel-1 SAR (C-band, 5.6 cm): Cloud- and smoke-penetrating synthetic aperture radar. Interferometric coherence between pre- and post-fire image pairs drops sharply over burned areas where canopy structure is destroyed, enabling burned-area delineation when optical sensors are obscured. 6-day revisit in IW mode at 10 m resolution.
What the thermal bands actually measure
Active fire detection rests on a straightforward piece of physics. The mid-infrared window around 3.7–4 µm sits on the peak of the Planck curve for temperatures between roughly 500 K and 1000 K, the range of flaming combustion. A fire occupying a tiny fraction of a pixel can still raise the pixel's radiance in that band by orders of magnitude above the background, even when the surrounding land surface is cool. This is why VIIRS and MODIS can detect fires far smaller than their nominal pixel size.
The practical difference between VIIRS at 375 m and MODIS at 1 km matters most for small fires. A MODIS pixel covers roughly seven times the area of a VIIRS I-band pixel. A smouldering agricultural burn of a few hectares may produce a clear VIIRS anomaly while contributing only a marginal, easily missed signal to the much larger MODIS pixel. NASA's Fire Information for Resource Management System publishes both datasets with near-real-time latency, typically under three hours from satellite overpass.
The Normalised Burn Ratio: what a charred canopy gives away
Once a fire is extinguished, the active thermal signal disappears. Mapping what actually burned requires a different approach. The Normalised Burn Ratio compares near-infrared reflectance (high in healthy vegetation) with shortwave-infrared reflectance (elevated in bare, charred or ash-covered soil). The formula is (NIR − SWIR) / (NIR + SWIR). Healthy forest returns high NBR values, typically above 0.4. Severely burned ground drops to negative values. The differenced NBR (dNBR), computed between a pre-fire and post-fire image pair, quantifies burn severity across a continuous scale.
Sentinel-2 at 20 m SWIR resolution gives substantially finer burn-scar delineation than Landsat at 30 m, though both are operationally useful. The choice often comes down to archive availability and cloud cover at the time of the relevant acquisition. Neither sensor can see through smoke during an active fire event, which is precisely when perimeter mapping is most urgently needed.
When smoke wins: SAR coherence as a fallback
Persistent smoke columns can obscure an active fire perimeter for days or weeks, particularly in tropical peatland fires where smouldering can continue for months. Sentinel-1's C-band radar penetrates smoke without attenuation. The relevant technique is interferometric coherence: when two SAR images acquired before and after a fire are compared, burned areas show a sharp drop in coherence because the canopy structure that produced a stable radar return has been destroyed or fundamentally altered. Intact forest maintains coherence; burned forest does not.
The method has its own limits. Coherence also degrades over dense, wind-disturbed canopy, over agricultural areas that change rapidly between acquisitions, and in areas of heavy rainfall between passes. It works best as a complement to optical NBR mapping, filling in the gaps where cloud or smoke has prevented a usable Sentinel-2 or Landsat acquisition, rather than as a standalone product.
Latency, revisit and the gap that matters most
For emergency response, the useful window is the first 24 to 72 hours after ignition. VIIRS and MODIS deliver active fire point data within two to three hours of overpass via FIRMS. That is fast enough to direct ground crews, but the data gives a point location and a fire radiative power estimate, not a perimeter. Perimeter data from optical NBR requires a cloud-free post-fire acquisition, which may not arrive for days if the fire is in a persistently cloudy region.
Sentinel-1 revisit of six days in standard IW mode is the binding constraint for SAR coherence mapping. In practice, the European Space Agency's emergency mapping activations can sometimes access ascending and descending orbit passes to shorten this, but six days is the baseline expectation. For fast-moving fires in dry savanna or Mediterranean shrubland, a perimeter can expand dramatically in that window. Buyers of fire-mapping products should ask explicitly about latency guarantees, not just sensor specifications.
What the numbers cannot tell you
Fire radiative power from VIIRS or MODIS correlates with fuel consumption rate and, by extension, with emissions estimates for carbon and aerosols. Published studies have used FRP to estimate CO2 and particulate emissions at regional scale. The relationship is not precise: fuel moisture, fuel type and combustion phase all introduce uncertainty that satellite radiance alone cannot resolve. An FRP figure from a single overpass is a snapshot, not an integrated measure of total energy released.
Burn severity from dNBR is similarly a spectral proxy. It captures the surface signature of fire, not necessarily the ecological impact on soil, seed bank or root systems. A high-severity dNBR score in a conifer stand and in a grassland carry very different ecological meanings. Ground-truthing remains necessary for any application that requires ecological interpretation rather than spatial extent. Satellize's analytics pipeline for fire products, like those used in the Tonga crop-estimation programme, flags these proxy uncertainties explicitly in delivered outputs rather than presenting a single authoritative number.
Building an operational fire-monitoring product
A practical fire-monitoring system for a national forestry agency or a large concession holder typically layers three data streams. VIIRS and MODIS active fire alerts provide near-real-time ignition detection and rough location. Sentinel-2 or Landsat NBR difference maps, generated as soon as a cloud-free post-fire image is available, provide the burned perimeter and severity classification. Sentinel-1 coherence maps fill the optical gap and provide an independent check on perimeter extent.
The archive depth of these systems is considerable. MODIS Terra goes back to February 2000, giving over two decades of fire history at 1 km. Landsat extends the burned-area record further still, with usable SWIR data from Landsat 5 TM onwards (1984). Long archives matter for understanding fire return intervals, which in turn inform fuel accumulation models and risk assessments. That historical dimension is often underused in operational programmes that focus only on current-season events.
Typical figures
| Active fire detection resolution | VIIRS: 375 m (I-band); MODIS: 1 km |
| Active fire detection latency | Typically 2–3 hours post-overpass via NASA FIRMS |
| Thermal anomaly revisit (combined VIIRS SNPP + NOAA-20) | Up to 4 overpasses per day at mid-latitudes; 2 per day at equator per satellite |
| Burned-area mapping resolution (optical NBR) | Sentinel-2: 20 m SWIR; Landsat 8/9: 30 m SWIR |
| Optical sensor revisit for NBR | Sentinel-2: 5 days (equator), 2–3 days (mid-latitudes); Landsat 8+9 combined: ~8 days |
| SAR burned-area mapping resolution | Sentinel-1 IW mode: 10 m (range × azimuth ~5 × 20 m before multi-looking) |
| SAR revisit for coherence mapping | Sentinel-1: 6 days (single orbit direction); 12-day interferometric baseline standard |
| Key spectral bands | MIR 3.7–4.0 µm (active fire); NIR ~865 nm + SWIR 1610/2190 nm (NBR); C-band 5.6 cm (SAR coherence) |
| Archive depth | MODIS from 2000; Landsat SWIR from 1984 (Landsat 5 TM); Sentinel-1/2 from 2014/2015 |
| Minimum detectable flaming area | VIIRS: on the order of a few hundred m² under ideal conditions; MODIS: ~50–100 m² theoretically but practically limited by 1 km pixel dilution |
Analytics Satellize can run
| Near-real-time active fire alert feed | Contextual thermal anomaly detection on VIIRS 375 m I-band and MODIS 1 km data, following the NASA FIRMS algorithm family | GeoJSON point feed with fire radiative power, confidence class and timestamp, updated per overpass |
| Burned-area perimeter and severity map | Differenced Normalised Burn Ratio (dNBR) from pre- and post-fire Sentinel-2 or Landsat SWIR image pairs | GeoTIFF severity raster and vector perimeter shapefile, with severity classified into low / moderate / high / very high burn severity classes |
| Smoke-obscured perimeter estimate | Sentinel-1 C-band interferometric coherence change detection between pre- and post-fire image pairs | GeoTIFF coherence-loss layer and derived burned-area polygon, flagged as SAR-derived with associated uncertainty notes |
| Fire radiative power time series | Aggregation of VIIRS and MODIS FRP point data over a defined area of interest across an event duration | CSV and chart report of cumulative FRP by day, usable as a proxy input for emissions estimation workflows |
| Multi-year fire return interval map | Stack analysis of annual burned-area products (MODIS MCD64A1 or equivalent) over the available archive | Raster map of fire return interval in years, with summary statistics per land-cover class |
| Burn severity validation report | Cross-comparison of dNBR-derived severity classes against SAR coherence loss and, where available, field-survey reference points | PDF report with accuracy assessment table and flagged areas of disagreement between optical and SAR-derived extents |
Who does the work
We can get this done for you. Satellize runs its own analyst desk and a strong science team. You do not buy a data feed and work out what it means; our people source the imagery, run the analysis described on this page, and hand you the answer with its confidence limits stated. Discuss this requirement.