Fire emissions quantification for forest carbon accounting

Quantifying fire emissions inside forest-carbon project areas requires active-fire detection, burned-area mapping and combustion-factor estimation. Each step carries measurable uncertainty that VCS permanence and buffer-pool rules demand you account for honestly.

Sensors

VIIRS (Suomi-NPP / NOAA-20): 375 m active-fire pixels using the I-band at 3.7–3.8 µm; daily global revisit from each satellite, giving roughly twice-daily combined coverage. The 375 m product (VNP14IMG / VJ114IMG) resolves fires missed by MODIS but still undercounts fires smaller than roughly one hectare in closed canopy.
MODIS Terra / Aqua: 1 km thermal anomaly product (MOD14/MYD14) and the MCD64A1 monthly burned-area product at 500 m. Mature archive from 2000 onward makes it the backbone of long-term emissions trend analysis, though the 1 km fire pixel misses a large fraction of sub-pixel smouldering events.
Sentinel-2 MSI: 10 m visible and 20 m shortwave-infrared bands (SWIR1 at 1.61 µm, SWIR2 at 2.19 µm) enable normalised burn ratio (NBR) differencing at a resolution that resolves individual canopy gaps. Five-day revisit at the equator under cloud-free conditions; the cloud caveat is significant in humid tropics during fire season.
Landsat 8 / 9 OLI-2 + TIRS: 30 m multispectral bands for NBR-based burned-area mapping and a 100 m thermal band for residual heat detection. Sixteen-day single-satellite revisit (eight days combined) limits temporal resolution for fast-moving fire fronts but provides a consistent archive back to 1984 for historical baseline work.

Why fire is a special problem for permanence accounting

Forest-carbon standards treat fire as a non-permanence risk. Under the Verra VCS methodology, projects must contribute to a pooled buffer account precisely because fire can reverse decades of sequestration in days. The buffer contribution percentage is partly determined by the project's assessed fire risk. When a fire actually occurs inside or adjacent to a project boundary, the methodology requires the proponent to estimate the carbon released and retire a corresponding number of buffer credits.

That estimation is not straightforward. It requires three linked quantities: the area burned, the above-ground biomass present before the fire, and the combustion completeness factor for the relevant forest type. Each carries its own uncertainty range. Satellite data can constrain the first quantity tightly and the second moderately well, but combustion completeness remains the largest unresolved source of error, with published values for tropical forests typically ranging from 0.20 to 0.45 depending on fire intensity and fuel moisture.

What mid-infrared emission at 3.9 µm actually tells you

Active fire detection exploits a physical fact: a flaming combustion front at roughly 600–1000 °C radiates orders of magnitude more energy in the mid-infrared (around 3.9 µm) than the surrounding cool canopy. VIIRS and MODIS both carry bands centred near this wavelength specifically to exploit that contrast. The result is a thermal anomaly flag, not a fire perimeter. A single VIIRS 375 m pixel flagged as active fire could represent a genuine 14-hectare blaze or a single burning log pile occupying a fraction of that area.

This ambiguity matters for carbon accounting. The active-fire product tells you when and roughly where combustion is occurring; it does not tell you how much area has burned or how completely the fuel was consumed. Relying on fire-pixel counts to estimate emissions, without a subsequent burned-area analysis, systematically overestimates emissions in some fire regimes and underestimates them in others. The two data streams are complementary, not interchangeable.

Burned-area mapping: what the normalised burn ratio resolves and what it misses

The normalised burn ratio, calculated as (NIR minus SWIR2) divided by (NIR plus SWIR2), exploits the fact that fire dramatically reduces near-infrared reflectance (healthy vegetation) while increasing shortwave-infrared reflectance (exposed char and bare soil). Differencing pre-fire and post-fire NBR images produces the differenced NBR (dNBR), which correlates with burn severity and has been used operationally in programmes such as the USGS Burned Area Emergency Response protocol.

At Sentinel-2's 20 m SWIR resolution, individual canopy patches of roughly 0.04 hectares become detectable in principle, though practical detection limits under mixed burn conditions are closer to 0.1 to 0.5 hectares depending on burn severity and background heterogeneity. Landsat at 30 m adds archive depth. The critical honest limit is cloud cover: in the Congo Basin or Borneo during fire season, persistent cloud can delay a usable post-fire Sentinel-2 image by two to six weeks, by which time secondary vegetation flush has already begun to mask shallow burn signatures. Combining VIIRS active-fire timing with the first cloud-free optical image is standard practice for constraining the burn window.

Small, low-intensity surface fires in closed-canopy tropical forest are systematically undercounted by every current spaceborne system. A fire that kills understorey vegetation without opening the canopy may be invisible in Sentinel-2 NBR imagery and generate no VIIRS thermal anomaly if it smoulders rather than flames. Published studies comparing field-mapped burn scars with satellite products in tropical forests report omission errors of 20 to 50 percent for low-severity events. This is not a reason to abandon satellite mapping; it is a reason to report emissions estimates with honest lower-bound uncertainty.

Combustion factors: the number that satellite data cannot supply directly

Once burned area and pre-fire biomass are known, the emissions calculation requires a combustion completeness factor (the fraction of available fuel actually oxidised) and emission factors for CO₂, CH₄ and N₂O. The IPCC 2006 Guidelines and the Global Fire Emissions Database (GFED) provide default values by biome, but the ranges are wide. For tropical forests, GFED4s uses a combustion completeness of around 0.36 for above-ground live biomass, but field studies report values from below 0.20 for patchy surface fires to above 0.50 for intense crown fires.

Satellite fire radiative power (FRP), measured by MODIS and VIIRS, offers a partial constraint. FRP integrates the rate of radiative energy release from the fire and correlates with fuel consumption rate in experimental settings. However, the relationship between cumulative fire radiative energy and total biomass consumed has substantial scatter across forest types and moisture conditions. FRP-based emissions estimates are more defensible than applying a single default factor, but they introduce their own calibration uncertainties and require careful handling of partially cloud-obscured fire events.

Putting a number on uncertainty for a VCS submission

VCS methodology VM0007 and its successors require that emissions estimates from unplanned deforestation and degradation, including fire, be accompanied by uncertainty analysis. A combined uncertainty of 30 to 50 percent in fire-derived carbon loss estimates is realistic for a single event in a tropical forest project, driven roughly equally by burned-area mapping error, pre-fire biomass uncertainty and combustion factor variability. Projects that have invested in high-resolution pre-fire biomass maps (from lidar or Sentinel-2 canopy structure analysis, covered on a sibling page) can reduce the biomass component substantially.

Satellize assembles the burned-area, fire-timing and FRP data streams into a structured uncertainty budget that maps directly onto the reporting tables required by standard verification bodies. The workflow runs on open Sentinel-2, Landsat and VIIRS archives, with commercial tasking added for rapid post-fire imaging when cloud persistence makes open-archive latency unacceptable.

Typical figures

Active-fire detection resolution	375 m (VIIRS I-band); 1 km (MODIS)
Burned-area mapping resolution	20–30 m (Sentinel-2 SWIR / Landsat OLI); 500 m (MCD64A1)
Active-fire revisit	~12 hours combined (Suomi-NPP + NOAA-20 VIIRS); ~12 hours combined (Terra + Aqua MODIS)
Optical burned-area revisit	5 days cloud-free (Sentinel-2, equatorial); 8 days combined (Landsat 8+9)
Near-real-time fire alert latency	Under 3 hours from overpass (NASA FIRMS VIIRS NRT)
Minimum detectable burned area (practical)	0.1–0.5 ha at 20 m resolution (severity-dependent); ~1 ha at 500 m
Key spectral bands	Mid-IR ~3.9 µm (active fire); NIR ~0.86 µm + SWIR2 ~2.2 µm (NBR)
Archive depth	MODIS from 2000; Landsat from 1984; Sentinel-2 from 2015; VIIRS from 2012
Coverage	Global daily (VIIRS/MODIS); global systematic (Sentinel-2/Landsat, cloud permitting)
Typical delivery format	GeoTIFF burn-severity raster; vector fire perimeter; CSV uncertainty budget table

Analytics Satellize can run

Active-fire event log with fire radiative power	VIIRS VNP14IMG / MODIS MOD14 contextual fire detection algorithm; FRP integration over event duration	Timestamped point dataset with FRP values (MW) per detection, clipped to project boundary buffer, delivered as GeoJSON or CSV within 4 hours of overpass
Burned-area perimeter and severity map	Pre/post dNBR differencing on Sentinel-2 SWIR2 and NIR bands; threshold classification validated against VIIRS fire-pixel timing	Classified GeoTIFF (unburned / low / moderate / high severity) and vector perimeter shapefile with area statistics in hectares
Combustion-completeness uncertainty range	GFED4s biome-specific combustion factors cross-referenced with FRP-derived fire intensity; Monte Carlo uncertainty propagation	Tabular uncertainty budget (mean estimate, 90% confidence interval) formatted for VCS VM0007 reporting tables
Carbon loss estimate per fire event	Burned area × pre-fire above-ground biomass layer × combustion factor × IPCC emission factors for CO₂, CH₄, N₂O	PDF summary report with tonne-CO₂-equivalent loss estimate, uncertainty range and methodology narrative suitable for submission to a Validation and Verification Body
Historical fire frequency map for buffer-pool risk assessment	MODIS MCD64A1 monthly burned-area archive (2000 to present) aggregated to annual fire return interval raster	GeoTIFF fire return interval map and project-area statistics table for inclusion in VCS non-permanence risk assessment
Post-fire cloud-gap-filled burn scar composite	Multi-date Sentinel-2 and Landsat median compositing to fill cloud gaps; change vector analysis against pre-fire baseline	Cloud-free burn scar raster with acquisition date metadata, flagging pixels where cloud gap exceeds 30 days and uncertainty is elevated

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.