VIIRS and MODIS active fire detection and radiative power
VIIRS and MODIS mid-infrared channels detect active burning at continental scale, twice daily, and convert radiance excess into fire radiative power, a direct proxy for fuel consumption rate. Resolution floors and saturation set real limits that every operational user must understand.
Sensors
- VIIRS I-band (Suomi NPP, NOAA-20, NOAA-21): 375-metre resolution at nadir; the primary operational active-fire sensor since 2012. The I4 band centred near 3.74 micrometres is the fire-detection channel; I5 at 11.45 micrometres provides the background thermal reference. Near-daily global revisit per satellite, roughly twice daily with two satellites combined.
- VIIRS M-band (Suomi NPP, NOAA-20, NOAA-21): 750-metre resolution; M13 band at 4.05 micrometres used for fire radiative power retrieval. Lower resolution than I-band but better calibrated for quantitative radiance work at high fire temperatures.
- MODIS (Terra and Aqua, heritage): 1-kilometre resolution fire products (MOD14/MYD14). Band 21 and Band 22 straddle 3.96 micrometres. Two overpasses per day combined across both platforms. Archive runs from 2000, making it the longest consistent global fire record available. Terra and Aqua are in orbit decline; MODIS is a heritage system.
- Sentinel-3 SLSTR: 1-kilometre resolution at nadir; S6 channel at 1.61 micrometres and F1/F2 fire channels at 3.74 and 10.85 micrometres. ESA's operational continuation of thermal fire monitoring for Europe and globally. Dual-view geometry adds some cloud-gap capability. Two satellites give roughly daily revisit.
- Landsat 8/9 OLI and TIRS: 30-metre multispectral and 100-metre thermal resolution. Not a fire-detection system in the operational sense, but the Band 6 shortwave infrared (2.2 micrometres) and TIRS Band 10 (10.9 micrometres) are used to validate fire perimeters and to detect small, smouldering fires that VIIRS and MODIS miss entirely. 16-day revisit per satellite.
Why 3.7 micrometres? The physics of a burning pixel
Planck's law dictates that a blackbody near 800 Kelvin, a reasonable estimate for a flaming fire front, emits orders of magnitude more energy in the mid-infrared around 3.7 micrometres than the surrounding land surface at 300 Kelvin. The contrast ratio between a fire pixel and its neighbours in this band can exceed 1,000 to 1. In the thermal infrared at 11 micrometres, the same fire raises the apparent brightness temperature by only a few degrees above background. That asymmetry is why the 3.7-micrometre channel is the detection channel and the 11-micrometre channel is the reference.
The contextual fire detection algorithm, first formalised for MODIS by Giglio and colleagues and now applied to VIIRS, exploits exactly this contrast. A candidate pixel is flagged as an active fire if its 3.7-micrometre brightness temperature exceeds both an absolute threshold (typically around 310 K for MODIS, adjusted for VIIRS) and a relative threshold above the mean of its valid, non-cloudy, non-water neighbours. The neighbourhood comparison suppresses false alarms from sun glint, hot desert surfaces and industrial thermal sources, though it does not eliminate them entirely.
Fire radiative power: what the radiance excess actually measures
Fire radiative power (FRP) is derived from the radiance excess in the 3.9-micrometre band above the estimated background. The relationship between FRP and the rate at which dry biomass is consumed is approximately linear, established empirically and grounded in combustion physics: roughly 0.368 megawatts of FRP per kilogram of dry matter consumed per second, a figure from published work using the MODIS fire products. This makes FRP operationally useful for emissions estimation, not just fire mapping.
FRP is expressed in megawatts per pixel. Integrating it over time gives fire radiative energy (FRE), which can be converted to total biomass consumed and, with fuel-type assumptions, to carbon and trace-gas emissions. Atmospheric chemistry models and global fire emissions databases such as GFED use exactly this chain. The conversion is not clean: fuel moisture, combustion efficiency and flame geometry all introduce uncertainty, and the published uncertainty on individual FRP retrievals from MODIS is typically 20 to 40 percent for moderate-intensity fires.
Detection limits: what VIIRS and MODIS will and will not find
The minimum detectable fire size is not a fixed number; it depends on fire temperature, time of day and the background thermal environment. As a practical guide, MODIS at 1 kilometre can detect fires occupying roughly 50 hectares or more under daytime conditions. VIIRS I-band at 375 metres pushes that floor down considerably, with published studies suggesting detection of fires as small as 0.1 hectares under ideal night-time conditions, though 1 to 5 hectares is a more honest operational expectation during the day when solar reflection raises background noise in the 3.7-micrometre channel.
Saturation is the opposite problem. When a fire is intense enough, the 3.7-micrometre detector reaches its upper radiance limit and the pixel is flagged as saturated rather than quantified. Crown fires in dense boreal or tropical forest can saturate VIIRS I4 and MODIS Band 22 simultaneously. The MODIS Band 21 channel has a higher saturation threshold and is used as a fallback for FRP retrieval in those cases, but very intense fires still produce underestimated FRP values. Smoke is a separate issue: thick pyroconvective plumes can obscure the fire front from the sensor entirely, causing omission errors that are essentially unrecoverable from passive optical data alone.
Latency and the operational fire monitoring chain
NASA's Fire Information for Resource Management System (FIRMS) distributes VIIRS and MODIS active fire detections with latency as low as three hours from observation to public availability, depending on ground station contact and processing queue. The Near Real-Time product carries a caveat: it has not undergone the quality screening applied to the standard product, so false-alarm rates are higher. For emergency dispatch or aviation hazard warnings, three-hour latency is still too slow in fast-moving fire conditions.
Sentinel-3 SLSTR fire products from EUMETSAT follow a similar near-real-time pipeline for European coverage. None of these systems approaches the sub-hour latency available from geostationary instruments such as SEVIRI on Meteosat or ABI on GOES, which sacrifice spatial resolution (3 to 4 kilometres) for continuous temporal coverage. The practical answer for operational fire management is to combine geostationary detection for timing with VIIRS or Sentinel-3 for spatial detail and FRP quantification.
Validation, cross-sensor consistency and honest caveats
Landsat 8 and 9 are the standard validation reference for fire perimeter mapping because 30-metre shortwave infrared imagery can resolve individual burn scars and active fire lines that VIIRS pixels average over. The comparison consistently shows that VIIRS I-band captures the gross fire extent well but misses small spot fires and underestimates perimeter complexity in fragmented landscapes. In tropical deforestation contexts, small agricultural burns under a few hectares, which are ecologically and legally significant, fall below the reliable VIIRS detection floor.
Cross-sensor consistency between MODIS and VIIRS FRP is good but not perfect. Calibration differences and the different spatial footprints mean that a direct pixel-to-pixel comparison shows scatter, and trend analyses spanning the MODIS-to-VIIRS transition require careful intercalibration. The long MODIS archive from 2000 is irreplaceable for trend work, but users should not treat MODIS and VIIRS FRP values as interchangeable without applying published correction factors.
Satellize runs VIIRS and MODIS fire analytics on open data streams for clients who need fire monitoring integrated with land-use or agricultural context. The Tonga crop-estimation programme is one example of how fire and vegetation signals are combined in a single analytical workflow. For fire specifically, the value is usually in connecting the detection layer to what was there before burning and what the emissions implications are, rather than in the detection itself.
Typical figures
| Spatial resolution (VIIRS I-band) | 375 m at nadir, degrading to approximately 750 m at swath edge |
| Spatial resolution (MODIS) | 1 km at nadir (Bands 21, 22, 31) |
| Spatial resolution (Sentinel-3 SLSTR fire channels) | 1 km at nadir |
| Revisit (VIIRS, two satellites combined) | Approximately twice daily global; up to 4 passes per day at high latitudes |
| Revisit (MODIS Terra + Aqua combined) | Approximately twice daily global; heritage system, orbit declining |
| Fire detection latency (FIRMS near-real-time) | Approximately 3 hours from observation; standard product within 24 hours |
| Primary fire detection band | 3.7 to 4.05 micrometres (mid-infrared); background reference at 11 micrometres |
| Minimum detectable fire size (VIIRS I-band, practical) | 1 to 5 hectares daytime; potentially below 1 hectare night-time under ideal conditions |
| Minimum detectable fire size (MODIS, practical) | Approximately 50 hectares daytime; smaller at night |
| MODIS archive depth | February 2000 to present (Terra); July 2002 to present (Aqua) |
Analytics Satellize can run
| Active fire alert layer | Contextual fire detection algorithm applied to VIIRS I-band brightness temperatures; neighbourhood comparison against non-fire, non-cloud pixels | GeoJSON or GIS point layer updated every overpass, with confidence class (low, nominal, high) and FRP value per detection |
| Fire radiative power time series | Radiance excess retrieval from VIIRS M13 or MODIS Band 22/21 above estimated background; converted to megawatts per pixel | CSV or API feed of FRP per fire cluster per overpass, aggregated to daily and weekly totals by region of interest |
| Cumulative burn area and perimeter mapping | Temporal compositing of VIIRS active fire detections combined with Sentinel-2 or Landsat burned-area index (dNBR) for perimeter delineation | Polygon shapefile of burn scar extent with date-of-burn attribution and pre-fire land cover class |
| Biomass consumption and emissions estimate | Integration of FRP over time to derive fire radiative energy; conversion to dry matter consumed using published emission factors (GFED methodology) | Tabular report of estimated carbon, CO and PM2.5 emissions by fire event, with uncertainty range stated |
| Fire seasonality and interannual trend analysis | MODIS archive from 2000 combined with VIIRS from 2012; intercalibrated FRP time series with breakpoint detection for trend and anomaly identification | Annual summary report with charts showing FRP anomaly relative to 20-year baseline, by land cover type and administrative unit |
| Small-fire gap analysis using Landsat | Comparison of VIIRS detections against Landsat 8/9 shortwave infrared imagery to identify fires below the VIIRS detection floor; statistical characterisation of omission rate by landscape type | Assessment report quantifying the fraction of burning area missed by VIIRS in the client's region of interest, with recommended supplementary data strategy |
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.