Forest fire risk assessment for carbon credit permanence ratings
Fire is the leading cause of non-permanence reversals in forest carbon projects. Satellite-derived fuel moisture, fire history and drought indices can convert that risk into defensible, quantitative buffer-pool inputs for Verra and Gold Standard methodologies.
Sensors
- MODIS Terra/Aqua: 500 m burned-area product (MCD64A1) and active fire detections (MOD14/MYD14) at 1 km, with daily global revisit from the combined constellation. Archive runs from 2000, giving over two decades of fire-frequency data for baseline construction.
- VIIRS S-NPP / NOAA-20: 375 m active fire product (VNP14IMG / VJ114IMG), improving on MODIS spatial resolution for small fires. Daily global revisit. The NOAA-20 addition since 2018 provides a second daily overpass, reducing detection gaps in cloudy periods.
- Sentinel-2 MSI: 10 m visible and 20 m shortwave infrared bands (SWIR1 at 1610 nm, SWIR2 at 2190 nm) used to compute the Normalized Difference Moisture Index and Moisture Stress Index as fuel-moisture proxies. Five-day revisit at the equator, better at higher latitudes. Cloud cover is the principal operational constraint.
- Landsat 8/9 OLI/TIRS: 30 m SWIR bands for fuel-moisture indices, plus 100 m (resampled to 30 m) thermal infrared for land surface temperature anomalies used in drought-stress scoring. Sixteen-day single-satellite revisit, eight days when both Landsat 8 and 9 are combined. Archive back to 1984 for long-term fire-scar mapping.
Why buffer pools need a number, not a narrative
Verra's Verified Carbon Standard requires project proponents to contribute credits to a pooled buffer account sized according to a risk rating. Fire risk is one of the scored categories, and the methodology (VM0007, VM0015 and others) asks for evidence of historical fire occurrence and current conditions. Most project developers answer this with a qualitative paragraph. Registries are beginning to expect more.
The practical problem is that fire risk is not static. A project area that scored low risk in 2015 may look very different after three consecutive below-average rainfall years. A buffer contribution set at validation and never revisited understates true risk as climate conditions shift. Satellite time series make annual or even seasonal risk re-scoring tractable, which matters for projects seeking periodic verification.
What a two-decade fire archive actually tells you
The MODIS burned-area product MCD64A1 provides monthly global coverage at 500 m resolution from November 2000 onwards. Over a 20-plus-year window, fire frequency per pixel can be computed as a simple count or as a return-interval estimate. A pixel that has burned four times in 22 years has a very different risk profile from one with no recorded fire, and that difference is defensible in a verification report.
VIIRS adds value from 2012 onwards, particularly for detecting smaller fires that MODIS misses at 500 m. The two archives are not directly interchangeable: MODIS burned area is a post-fire mapping product, while VIIRS active fire detections are near-real-time thermal anomalies. Used together, they cross-validate. Discrepancies, where VIIRS detects active fire but MODIS shows no burned area in the subsequent monthly product, often indicate low-intensity surface fires that consume understorey fuel without killing the canopy. Those fires still consume carbon.
One honest limit: both sensors undercount fire in dense cloud cover, which is common in humid tropical forests during transitional seasons. A fire-frequency map built on optical and thermal data alone will have gaps. Projects in persistently cloudy regions should treat fire-frequency estimates as lower bounds.
Fuel moisture from shortwave infrared: the physics and the caveats
Liquid water in plant tissue absorbs strongly at 1450 nm and 1940 nm, with secondary absorption features at 1200 nm and around 970 nm. Sentinel-2's SWIR bands at 1610 nm and 2190 nm sit on the shoulders of these features, meaning reflectance in those bands rises as canopy water content falls. The Normalized Difference Moisture Index, computed as (NIR minus SWIR1) divided by (NIR plus SWIR1), is the most widely used proxy. Published studies using field-measured fuel moisture content show reasonable correlation with NDMI in open-canopy forests, though dense closed-canopy tropical forest presents a confound: the sensor sees the top of the canopy, and understorey fuel moisture, which drives surface fire spread, may diverge substantially from the canopy signal.
Sentinel-2's 20 m SWIR resolution is adequate for project-level mapping but not for sub-stand fuel heterogeneity. The five-day revisit means that a rapid drying event during a gap in clear-sky acquisitions can be missed entirely. For projects in seasonally dry forests, a compositing approach that tracks the driest clear-sky observation within a rolling 30-day window is more reliable than single-date analysis.
Land surface temperature anomalies as a drought-stress signal
Landsat 8/9 TIRS measures land surface temperature (LST) at 100 m native resolution. When a forest is water-stressed, transpiration falls and LST rises relative to the local climatological baseline. The anomaly, expressed as degrees above the multi-year mean for the same calendar period, is a practical drought-stress index that does not require rainfall station data.
The approach has published precedent in fire-danger forecasting research and is consistent with the logic behind the Keetch-Byram Drought Index, though LST anomaly is not a direct substitute for that index. Its advantage is spatial continuity across project areas that may lack ground weather stations. Its limit is that cloud cover blocks thermal retrieval at exactly the moments, during active convective weather, when conditions are changing fastest. Combining LST anomaly with vegetation indices and the VIIRS active fire record gives a multi-signal picture that is harder to dismiss than any single indicator.
Assembling a permanence risk score from satellite inputs
A quantitative fire-risk component for a buffer-pool calculation can be structured as a weighted composite of three satellite-derived sub-scores: historical fire frequency from the MODIS/VIIRS archive, current-season fuel-moisture anomaly from Sentinel-2 SWIR, and drought-stress severity from Landsat LST anomaly. Each sub-score is normalised against a reference distribution, either global forest pixels or a biome-specific peer group, so that the output is interpretable relative to comparable projects.
The composite score maps directly onto the risk-tier language used in Verra's buffer contribution tables. A project in the lowest fire-risk tier might contribute two per cent of issued credits to the buffer; a project scoring in the highest tier might contribute ten per cent or more. The difference in credit revenue over a 30-year project lifetime is substantial, which gives project developers a direct financial incentive to produce credible, annually updated risk evidence rather than a one-time qualitative assessment.
Satellize structures this kind of multi-signal composite as a reproducible analytical pipeline, with the same architecture used in the Tonga crop-estimation programme applied to forest-risk scoring: open-constellation inputs, documented processing steps, and outputs formatted for direct ingestion into verification reports. The pipeline is auditable, which matters when a third-party verifier needs to trace a risk score back to its source pixels.
What satellite data cannot do here
Fire risk has components that no satellite currently resolves well. Ignition probability depends on human activity patterns, road access, land-use pressure at project boundaries and local fire management capacity. None of those are directly observable from orbit, though road networks and settlement proximity can be mapped as proxy indicators. Wind speed and direction at the time of a fire event, which determine spread rate and severity, are not retrievable from the sensors described here.
Cloud cover remains the most operationally significant constraint. A project in the Congo Basin or the western Amazon may have fewer than 20 cloud-free Sentinel-2 acquisitions per year over parts of its area. In those conditions, fuel-moisture time series will have gaps of weeks or months, and the analyst must be explicit about which risk periods are unobserved. Presenting a fire-risk score without stating the cloud-gap fraction is misleading. Any credible permanence risk report should include an observation-density map alongside the risk output.
Typical figures
| Burned-area spatial resolution | 500 m (MODIS MCD64A1); 375 m active fire detection (VIIRS VNP14IMG) |
| Fuel-moisture index resolution | 20 m SWIR-derived (Sentinel-2 MSI); 30 m (Landsat 8/9 OLI) |
| Land surface temperature resolution | 100 m native, resampled to 30 m (Landsat 8/9 TIRS) |
| Revisit frequency | Daily global (MODIS/VIIRS combined); 5 days at equator (Sentinel-2); 8 days combined (Landsat 8+9) |
| Fire-frequency archive depth | MODIS from November 2000 (24+ years); Landsat burn-scar mapping possible from 1984 |
| Active fire detection latency | VIIRS near-real-time products available within 3 hours of overpass via NASA FIRMS |
| Cloud cover constraint | All optical and thermal bands blocked by cloud; humid tropical sites may have fewer than 20 clear acquisitions per year |
| Minimum detectable burned area | Approximately 100 ha for MODIS MCD64A1; smaller fires detectable via VIIRS active fire but not always in burned-area product |
| Delivery formats | GeoTIFF risk layers, CSV pixel-level fire-history tables, PDF scoring report with observation-density annex |
Analytics Satellize can run
| Historical fire-frequency map | Pixel-level burn count from MODIS MCD64A1 and VIIRS VJ114IMG archives, with return-interval statistics per land-cover class | GeoTIFF layer and summary table by project sub-zone, formatted for Verra risk-rating input |
| Fuel-moisture anomaly index | NDMI and Moisture Stress Index from Sentinel-2 SWIR bands, compared against pixel-level seasonal climatology; 30-day minimum-moisture composite | Seasonal GeoTIFF time series with anomaly score per reporting period, plus cloud-gap fraction map |
| Drought-stress severity score | Landsat TIRS land surface temperature anomaly relative to 10-year baseline, aggregated to project sub-zones | Annual LST anomaly raster and zonal statistics table |
| Composite permanence risk score | Weighted normalised combination of fire-frequency, fuel-moisture anomaly and drought-stress sub-scores, calibrated against biome peer-group distribution | Single numeric score per project zone with confidence interval, mapped to Verra buffer-tier language, included in verification-ready PDF report |
| Active fire alert feed | Near-real-time VIIRS FIRMS ingestion with project-boundary intersection and area estimation | Email or API alert within 6 hours of detection, with estimated affected area and severity class |
| Observation-density and data-gap audit | Cloud-mask analysis across all Sentinel-2 and Landsat acquisitions within project boundary and reporting period | Per-pixel clear-observation count map included as mandatory annex to risk report |
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.