Fire radiative power and combustion rate estimation from geostationary imagers

Geostationary imagers sample active fires every 10–15 minutes, enabling fire radiative power integration over entire burn events and near-real-time smoke emission estimates for atmospheric chemistry models.

Sensors

SEVIRI on Meteosat Second Generation (MSG): 3 km pixel at nadir (4.8 km at mid-latitudes), 15-minute full-disc repeat. The 3.9 µm channel is the primary FRP retrieval band; operational FRP products are distributed by EUMETSAT in near-real-time. Saturates over very intense fire cores above roughly 10 GW per pixel.
ABI on GOES-16 and GOES-18: 2 km pixel in Band 7 (3.9 µm), full-disc every 10 minutes, CONUS sector every 5 minutes. NOAA distributes the operational Fire Detection and Characterization (FDC) product with per-pixel FRP in megawatts. Covers the Americas.
AHI on Himawari-9: 2 km pixel at 3.9 µm, full-disc every 10 minutes. Japan Meteorological Agency provides near-real-time fire products covering East Asia, Southeast Asia and Australasia, closing the geostationary coverage gap left by MSG and GOES.
VIIRS on NOAA-20 and Suomi-NPP: 375 m active-fire pixels with per-pixel FRP in the I4 band (3.74 µm), twice-daily polar overpass. Used as a gap-fill and cross-calibration reference for geostationary FRP, and to resolve sub-pixel fire structure invisible at 2–3 km. Distributed via NASA FIRMS.

What the 3.9 µm channel is actually measuring

At ambient temperatures, terrestrial surfaces emit almost nothing at 3.9 µm. A fire at 600–1,200 K shifts enough of its Planck curve into that window to dominate the signal entirely. The radiance excess above a background estimate is therefore a direct proxy for the rate at which chemical energy is released as heat, which is what fire radiative power quantifies.

The retrieval follows from the Stefan-Boltzmann relation applied to the excess mid-infrared radiance. Wooster et al. (2003, published in Geophysical Research Letters) established the linear relationship between FRP in megawatts and the rate of dry matter combustion in kilograms per second, with a coefficient of roughly 0.368 kg MJ⁻¹ derived from laboratory and field measurements. That coefficient is the link between a satellite radiance and a fuel-consumption rate, and it underpins every operational smoke-emission estimate that follows.

Why temporal sampling matters more than spatial resolution here

A single MODIS or VIIRS overpass captures a snapshot of fire intensity. Geostationary imagers capture the arc: ignition, peak spread, suppression or natural extinction. Integrating FRP over time gives fire radiative energy (FRE) in megajoules, which is proportional to total fuel consumed across the event. For a fire that burns for 18 hours, two polar snapshots miss most of the combustion curve; a 10-minute geostationary time series does not.

This matters practically because smoke-emission inventories fed into atmospheric chemistry models such as GEOS-Chem or the Copernicus Atmosphere Monitoring Service (CAMS) fire module depend on total fuel consumption, not instantaneous intensity. Errors in combustion timing propagate into errors in plume height, aerosol loading and downwind air-quality forecasts. The geostationary FRP time series is the only satellite-based way to constrain those errors without relying on post-hoc burned-area products, which typically have a 24–48 hour lag.

The saturation ceiling and the cloud floor

Two systematic biases run in opposite directions. At the high end, the 3.9 µm channel on current geostationary imagers saturates when FRP per pixel exceeds roughly 10 GW. This happens in the most intense fire cores, particularly in boreal peatland fires and large Australian eucalypt fires. When saturation occurs, FRP is underestimated and must be flagged or corrected using VIIRS sub-pixel data or adjacent unsaturated pixels. No current operational product fully solves this; users should treat geostationary FRP as a lower bound during extreme events.

At the low end, partial cloud cover within a pixel suppresses the retrieved FRP proportionally to the obscured fraction. Because cloud fraction is rarely known at sub-pixel scale, this bias is systematic and difficult to correct. Studies using SEVIRI have shown that FRP underestimation from cloud can reach 30–50% in regions with frequent convective cloud development during fire season, such as the Congo Basin. Algorithms that screen cloudy pixels entirely are conservative but miss real fire; those that attempt partial-cloud correction introduce their own uncertainty. Honest use of geostationary FRP products requires tracking the cloud-flagged fraction of the time series alongside the FRP values themselves.

From FRP to smoke: the emission-factor chain

Converting FRP to smoke-constituent emissions requires two further steps. First, FRP is converted to a fuel-consumption rate using the Wooster coefficient. Second, fuel-consumption rate is multiplied by emission factors, expressed in grams of pollutant per kilogram of dry matter burned, for species including PM2.5, CO, CO₂, NOₓ and volatile organic compounds. Emission factors vary substantially by vegetation type: savanna fires produce different ratios of trace gases than boreal forest or tropical peat. The Global Fire Emissions Database (GFED), maintained by researchers at the University of Maryland and VU Amsterdam, provides the standard emission-factor tables used in most operational systems.

The resulting near-real-time emission estimates are ingested by CAMS, which assimilates them into global atmospheric transport simulations updated twice daily. Latency from satellite observation to CAMS fire-emission analysis is typically three to six hours for the geostationary component. That is fast enough to influence 24-hour air-quality forecasts but not fast enough to guide active suppression decisions, where ground-based intelligence still dominates.

What the archive can and cannot tell a buyer

SEVIRI FRP data extend back to 2004, ABI to 2017 and AHI to 2015. That depth supports multi-year trend analysis, seasonal climatologies and fire-regime change detection across the regions each instrument covers. Comparing fire seasons year-on-year using integrated FRE is more physically meaningful than comparing burned-area estimates, which depend on the sensitivity of the optical change-detection algorithm used.

The honest limit of the archive is geographic: geostationary coverage has fixed footprints. High-latitude fires above roughly 70°N, including the Siberian and Alaskan events that have grown substantially in recent decades, fall outside the useful viewing angle of current geostationary imagers. VIIRS fills that gap but without the temporal resolution needed for FRE integration. Buyers with interests in boreal or Arctic fire should not expect geostationary FRP to be their primary data source.

Satellize runs FRP time-series analytics on open geostationary products and VIIRS gap-fill data, producing per-event combustion budgets and emission estimates as structured feeds or GIS layers. The same analytical pipeline that supports the Tonga crop-estimation programme applies here: open-constellation data, processed to a client specification, without dependence on proprietary sensor access.

Practical thresholds for procurement decisions

A government or enterprise buyer evaluating geostationary FRP data should ask three questions. What is the minimum fire size the application requires to detect? At 2–3 km pixel size, a fire must typically exceed 10–50 MW of FRP to register above background noise; sub-megawatt ignitions are invisible. What latency is acceptable? Operational products from EUMETSAT and NOAA are available within 15–30 minutes of observation; research-grade reprocessed products with better calibration have longer delays. And what geographic region matters? The answer determines which instrument is primary and whether VIIRS gap-fill is necessary.

For air-quality agencies, the 10-minute ABI cadence over the Americas or the 10-minute AHI cadence over Asia-Pacific is likely sufficient for model forcing. For insurance or carbon-market applications requiring accurate total fuel consumption, integrating the full geostationary FRP time series with VIIRS cross-calibration gives a more defensible combustion budget than any single-overpass method.

Typical figures

Spatial resolution (geostationary, nadir)	2 km (ABI, AHI) / 3 km (SEVIRI)
Temporal sampling	10 min (ABI full-disc, AHI) / 15 min (SEVIRI full-disc)
Primary retrieval band	3.9 µm mid-infrared (MIR)
FRP detection floor (typical)	~10–50 MW per pixel depending on background temperature and sensor noise
FRP saturation ceiling (current sensors)	~10 GW per pixel; flagged but not corrected in standard products
Polar gap-fill revisit (VIIRS)	~12-hour repeat at mid-latitudes; 375 m active-fire pixel resolution
Product latency (operational)	15–30 minutes from observation (EUMETSAT, NOAA GOES-R series)
Archive depth	SEVIRI from 2004; ABI from 2017; AHI from 2015
Geographic coverage	Africa/Europe (SEVIRI), Americas (ABI), Asia-Pacific (AHI); high-latitude gaps above ~70°N
Fuel-consumption conversion coefficient	~0.368 kg dry matter MJ⁻¹ (Wooster et al., GRL 2003)

Analytics Satellize can run

Per-event fire radiative energy (FRE) budget	Temporal integration of geostationary FRP time series with cloud-flag tracking and VIIRS cross-calibration	Structured CSV or GeoJSON report per fire event: total FRE in terajoules, duration, peak FRP, cloud-flagged fraction
Near-real-time combustion rate feed	Wooster FRP-to-fuel-consumption conversion applied to 10/15-minute geostationary retrievals	JSON feed updated every 15 minutes with per-pixel fuel consumption rate in kg s⁻¹, suitable for model ingestion
Smoke emission estimate by species	Fuel consumption multiplied by GFED vegetation-type emission factors for PM2.5, CO, CO₂, NOₓ	Gridded emission inventory (0.1° resolution) per fire event or per day, in NetCDF or GeoTIFF
Saturation-flagged FRP anomaly alert	Pixel-level saturation detection with VIIRS sub-pixel FRP substitution where available	Alert message with event location, time, estimated lower-bound FRP and saturation flag, delivered via webhook or email
Multi-year FRP climatology and trend layer	Archive aggregation of SEVIRI/ABI/AHI FRP by season and vegetation class, normalised by area burned	GIS layer showing mean and trend in annual FRE by 0.5° grid cell, with uncertainty bounds
Cloud-bias-corrected FRP time series	Cloud-flagged pixel exclusion with temporal interpolation across gaps; bias range reported explicitly	Annotated time-series plot and underlying data table per fire event, with cloud-gap fraction quantified

Sources

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.