Post-wildfire burn scar extent and severity mapping

Burned area extent and fire severity are mapped from space using the differenced Normalised Burn Ratio, combining pre- and post-fire Sentinel-2 or Landsat imagery. Where smoke or cloud persists, SAR backscatter change from Sentinel-1 fills the gap.

Sensors

Sentinel-2 MSI: 10 m resolution in visible bands, 20 m in the SWIR-1 (Band 11, 1610 nm) and SWIR-2 (Band 12, 2190 nm) bands used for NBR calculation. Five-day revisit at the equator (two-satellite constellation). The primary operational sensor for dNBR mapping under clear skies.
Landsat 8/9 OLI: 30 m resolution across all bands including SWIR-1 (Band 6) and SWIR-2 (Band 7). Eight-day revisit per satellite, 16 days for a single platform. Archive extends to 1972 for Landsat 1, giving decades of pre-fire baseline. The USGS burn-severity classification scheme is calibrated to Landsat dNBR values.
Sentinel-1 IW SAR: C-band synthetic aperture radar at 10 m resolution in Interferometric Wide Swath mode. Penetrates cloud and smoke. Backscatter change between pre- and post-fire acquisitions detects structural canopy loss where optical data is obscured, though it cannot replicate the spectral precision of dNBR for severity gradation.
VIIRS SNPP / NOAA-20: 375 m active-fire detection product (VNP14IMG) provides the fire-perimeter timeline needed to select correct pre- and post-fire image pairs. Not used for burn-scar delineation itself, but essential for dating the event precisely. Daily global coverage.

What the differenced Normalised Burn Ratio actually measures

The Normalised Burn Ratio is calculated as (NIR minus SWIR-2) divided by (NIR plus SWIR-2). Healthy vegetation reflects strongly in the near-infrared and absorbs in the shortwave infrared, so its NBR is high. Burned ground does the opposite: charred material and exposed soil absorb NIR and reflect SWIR, producing a low or negative NBR. Subtract the post-fire image from the pre-fire image and you get the dNBR, a continuous variable that scales with the degree of change.

The USGS and Copernicus Emergency Management Service both use published dNBR thresholds to assign discrete severity classes: unburned (dNBR below roughly 0.1), low severity, moderate-low, moderate-high, and high severity (dNBR above approximately 0.66). These thresholds were developed from field-validated Landsat data and are widely cited, but they are not universal. Vegetation type, soil moisture at the time of the fire, and the interval between the fire and the post-event image all shift the numbers. A Mediterranean shrubland and a boreal spruce forest will not behave identically on the same dNBR scale.

Choosing the right image pair is half the work

The pre-fire image should be acquired in the same phenological season as the post-fire image, ideally within the same calendar month of a prior year. Using a summer pre-fire image against an autumn post-fire image introduces spectral confusion: natural senescence looks like burn damage. VIIRS active-fire detections, available within hours from NASA FIRMS, give the precise date range to bracket the event and guide image selection.

The post-fire image should be acquired as soon as cloud clears, ideally within two to four weeks of fire suppression. Wait too long and early re-sprouting vegetation, particularly in fire-adapted ecosystems such as Australian eucalyptus or South African fynbos, begins to raise the NBR and underestimate severity. The Extended Assessment product used by Copernicus EMS deliberately revisits at 30 and 90 days to capture this trajectory, which is useful for recovery monitoring but should not be confused with the initial severity map.

Smoke, cloud and the limits of optical retrieval

Persistent smoke haze is the most common source of error in post-fire optical mapping. Smoke attenuates NIR and scatters visible wavelengths, suppressing the NBR and causing the algorithm to underestimate burned area near active or recently extinguished fronts. Atmospheric correction algorithms such as Sen2Cor partially compensate, but dense smoke columns defeat them. The practical rule: if the aerosol optical depth over the scene exceeds roughly 0.5 at 550 nm, the optical severity map should be treated as a lower bound on affected area.

This is where Sentinel-1 becomes useful, not as a replacement but as a consistency check. C-band backscatter drops sharply when a forest canopy is removed, because the double-bounce and volume-scattering mechanisms that characterise intact vegetation are eliminated. A simple log-ratio of pre- and post-fire VV or VH backscatter highlights areas of structural loss. The limitation is real: SAR cannot distinguish high-severity burns from moderate ones with the same precision as dNBR, and it is sensitive to soil moisture changes that may have nothing to do with fire. Use it to confirm extent under cloud, not to replace spectral severity gradation.

The Copernicus EMS workflow and what it delivers

Copernicus Emergency Management Service Rapid Mapping activations produce standardised grading maps in GeoTIFF and vector format, typically within 24 to 48 hours of activation request, subject to image availability. The EMSR product set for wildfires includes a Delineation product (burned perimeter) and a Grading product (severity classes within the perimeter). Both are publicly archived and downloadable, making them a useful baseline for any subsequent deeper analysis.

Copernicus EMS uses Sentinel-2 as the primary sensor, falling back to commercial very-high-resolution imagery when cloud persists or when fine-scale urban interface mapping is needed. The service does not guarantee a specific revisit window; it depends on satellite tasking priority and atmospheric conditions. For events outside Europe or where the activation threshold is not met, analysts must run equivalent workflows independently using the same open-data sources.

What the map cannot tell you without ground truth

A dNBR map is a spectral proxy, not a direct measurement of ecological or structural damage. High-severity pixels indicate that surface reflectance changed dramatically, which correlates strongly with canopy mortality and soil heating. But correlation is not identity. Patchy burns within a high-severity zone, surviving understorey beneath a killed overstorey, and pre-existing bare soil or rock outcrops can all distort the picture. Field validation, even a modest sample of GPS-located ground plots, substantially improves the reliability of area estimates derived from the map.

Resolution also matters for reporting. At Sentinel-2's 20 m SWIR resolution, features smaller than roughly 400 square metres are sub-pixel. In wildland-urban interface fires, where the boundary between destroyed structures and surviving vegetation is sharp and spatially complex, 20 m pixels produce mixed-class errors. Landsat at 30 m is worse on this metric. For structure-level damage assessment in the interface zone, very-high-resolution optical or SAR data is necessary, and that falls under a separate analytical workflow.

Satellize runs dNBR severity mapping on open Sentinel-2 and Landsat archives, with SAR consistency checks from Sentinel-1, and delivers classified severity layers alongside uncertainty flags derived from atmospheric quality indicators. The Overhead column has covered several major fire seasons using this methodology.

Typical figures

Primary sensor spatial resolution (SWIR bands)	Sentinel-2 MSI: 20 m. Landsat 8/9 OLI: 30 m.
Revisit interval	Sentinel-2 (two-satellite): 5 days at equator, 2-3 days at mid-latitudes. Landsat 8 or 9 individually: 16 days. Combined Landsat 8+9: 8 days.
SAR supplement resolution	Sentinel-1 IW mode: 10 m range x 10 m azimuth (after multi-looking). Cloud- and smoke-penetrating.
Active-fire dating sensor	VIIRS 375 m active-fire product (VNP14IMG), daily global coverage, latency under 3 hours via NASA FIRMS.
Key spectral bands for dNBR	NIR (~865 nm) and SWIR-2 (~2190 nm) for both Sentinel-2 and Landsat OLI.
Minimum mappable burned patch	Approximately 1-4 ha at Sentinel-2 20 m SWIR resolution under clear sky. Smaller patches are sub-pixel and may be missed or blended.
Typical post-event image latency (clear sky)	First usable Sentinel-2 acquisition within 1-5 days of fire suppression, depending on overpass timing and cloud cover.
Landsat archive depth	Landsat 1 onwards from 1972; consistent OLI-class data from Landsat 8 launch in 2013.
Standard dNBR severity thresholds (USGS)	Enhanced regrowth: dNBR < -0.1. Unburned: -0.1 to 0.1. Low: 0.1-0.27. Moderate-low: 0.27-0.44. Moderate-high: 0.44-0.66. High: > 0.66.
Delivery formats	GeoTIFF (classified and continuous dNBR), GeoPackage or Shapefile vector perimeters, PDF summary report.

Analytics Satellize can run

Burned area perimeter (vector)	Thresholded dNBR from Sentinel-2 or Landsat pre/post image differencing, edge-cleaned with morphological filtering.	GeoPackage polygon layer with area in hectares and confidence flag based on atmospheric quality.
Fire severity classification raster	USGS five-class dNBR thresholding applied to atmospherically corrected imagery. Optionally adjusted for vegetation type using land-cover ancillary data.	GeoTIFF with five severity classes and accompanying class-area statistics table.
SAR-derived canopy-loss extent	Log-ratio change detection on Sentinel-1 VH backscatter between pre- and post-fire acquisitions. Threshold set at 3 dB change to reduce soil-moisture false positives.	GeoTIFF binary change layer and comparison overlay against optical dNBR perimeter.
Smoke-contamination quality mask	Scene Classification Layer (SCL) from Sen2Cor plus aerosol optical depth retrieval to flag pixels where atmospheric interference degrades dNBR reliability.	Per-pixel confidence layer included with severity GeoTIFF.
Multi-date recovery trajectory	Time-series NBR extracted from all available Sentinel-2 acquisitions at 30, 60 and 90 days post-fire to track re-vegetation onset by severity class.	CSV time-series per severity zone and annotated chart for reporting.
Burned area change report	Comparison of current-event perimeter against historical fire scars from Landsat archive, identifying re-burn areas where prior fire history may affect severity interpretation.	PDF summary with map of re-burn overlap and prior-event dates.

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.