Wildfire burn-scar analysis and post-fire erosion risk at heritage sites

Wildfires strip the vegetation that holds soil over buried archaeology. Mapping burn severity with dNBR and modelling post-fire erosion risk tells site managers where the ground is about to move, and how fast.

Sensors

Landsat 8/9 OLI: 30 m multispectral resolution including NIR (Band 5) and SWIR-2 (Band 7), the two bands used to compute NBR. Global 16-day revisit per satellite, combined 8-day revisit with both in orbit. Archive from 1984 (TM) gives pre-fire baseline imagery spanning decades.
Sentinel-2 MSI: 10 m visible and 20 m SWIR bands (B8A at 865 nm, B12 at 2190 nm) for NBR computation at finer spatial detail than Landsat. Five-day revisit at mid-latitudes with both Sentinel-2A and 2B. Cloud cover over post-fire smoke and Mediterranean summer haze can delay acquisition by several days.
Sentinel-1 SAR (C-band): 5.4 GHz C-band SAR at 10 m IW mode resolution. Cloud-independent, so it captures surface roughness changes and rill formation during post-fire rain seasons when optical sensors are obscured. Coherence change between pre- and post-fire acquisitions detects canopy loss and soil disturbance. Six-day revisit in Europe, 12-day elsewhere.
MODIS / VIIRS active fire products: MODIS Terra/Aqua at 1 km thermal anomaly detection, VIIRS at 375 m. Used to establish fire perimeter timing and intensity sequence rather than final burn-severity mapping. NASA FIRMS provides near-real-time alerts within hours of detection. Spatial resolution is too coarse for site-level archaeology but essential for understanding fire behaviour across a landscape.

What a burn scar reveals about buried ground

Wildfire does two things to an archaeological landscape simultaneously. It removes the vegetation mat that binds surface soils, and it bakes the upper centimetres of earth into a hydrophobic crust. Both effects are legible from orbit within days of a fire, and both are precursors to accelerated erosion of exactly the kind of stratified deposits that archaeologists depend on.

The differenced Normalised Burn Ratio (dNBR) is the standard method for quantifying this. NBR is computed from near-infrared and shortwave infrared reflectance: healthy vegetation has high NIR and low SWIR, whereas burned ground inverts that relationship. Subtracting post-fire NBR from pre-fire NBR yields dNBR, with values typically ranging from below zero (unburned or enhanced regrowth) to above 1.3 (high-severity crown fire). The USGS and NASA have published thresholds for severity classes since the early 2000s, and those thresholds have been validated against field measurements in Mediterranean shrubland, Californian chaparral and Iberian montane forests, all of which contain significant archaeological site distributions.

The erosion window that follows

Post-fire erosion is not a slow background process. The first high-intensity rainfall event after a fire, often within weeks in Mediterranean climates, can mobilise more sediment in a single storm than the preceding decade of normal erosion. For a site with shallow buried deposits, that single event can be irreversible.

Erosion risk modelling after fire typically combines dNBR-derived soil erodibility estimates with slope data from a digital elevation model and published rainfall erosivity figures. The Revised Universal Soil Loss Equation (RUSLE) provides the standard framework. The cover-management factor (C-factor) drops sharply in high-severity burn zones, driving predicted soil loss up by an order of magnitude or more on slopes above roughly 15 degrees. Sentinel-2 at 20 m SWIR resolution is fine enough to distinguish individual field terraces and archaeological earthwork scarps, which matters because even a modest terrace can redirect runoff and protect downslope deposits. Conversely, a breached terrace becomes a funnel.

Published work on Californian archaeological landscapes, including studies following the 2017 and 2018 fire seasons, has used exactly this combination: dNBR from Landsat or Sentinel-2 for severity, lidar-derived or SRTM slope data, and RUSLE to prioritise which site polygons face the greatest immediate risk. The outputs are practical: a ranked list of site locations by modelled soil-loss rate, updated after each significant rainfall event.

SAR fills the gap when smoke and cloud arrive together

Post-fire landscapes in the Mediterranean are often cloud-covered precisely when erosion is occurring, during autumn and early winter rains. Sentinel-1 C-band SAR is unaffected by cloud and operates day and night. Coherence change detection between pre-fire and post-fire SAR image pairs captures the loss of vegetative scatterers and the emergence of rougher, rilled surfaces. This is not a substitute for dNBR severity mapping, which requires optical data, but it extends monitoring through the critical wet season when optical revisit is effectively zero.

Rill formation on bare burned slopes produces a measurable increase in surface roughness at C-band wavelengths. Studies in Mediterranean Spain and Portugal have used Sentinel-1 backscatter time series to track progressive surface degradation across burned hillsides. The practical limit is that C-band cannot penetrate more than a few centimetres into dry soil, so it reads the surface, not the buried deposit. It tells you the erosion is happening; it does not tell you how much archaeology has already been lost.

Vegetation recovery as a proxy for ongoing risk

Bare ground is not a permanent state. Vegetation recovery after Mediterranean fire follows a reasonably predictable trajectory: pioneer annuals within months, shrub resprouting within one to two growing seasons, partial canopy closure within three to five years depending on fire severity and rainfall. The NDVI and NBR time series from Sentinel-2 or Landsat can track this recovery against a pre-fire baseline. Where recovery is slow or absent, erosion risk remains elevated.

For site managers, this translates into a monitoring schedule rather than a single assessment. A site on a high-severity burn patch on a 20-degree slope in a low-rainfall year needs a different intervention timetable than one on a low-severity margin with rapid grass recovery. Multitemporal spectral analysis across the archive makes that distinction quantitative rather than qualitative. Sentinel-2's five-day revisit is fast enough to catch the first flush of post-fire green-up, which is itself a useful signal: rapid early recovery correlates with lower hydrophobic crust severity and lower erosion risk.

Honest limits of the method

dNBR is a spectral index, not a direct measure of soil loss. Its accuracy depends on the quality of the pre-fire baseline image, which must be cloud-free, seasonally matched, and atmospherically corrected. A pre-fire image acquired in a drought year will already show suppressed vegetation, compressing the dNBR dynamic range and potentially underestimating severity. This is a real problem in regions where multi-year drought precedes fire, which is increasingly common in southern Europe.

RUSLE-based erosion modelling introduces further uncertainty. The C-factor and K-factor (soil erodibility) values for post-fire Mediterranean soils are published but carry wide confidence intervals. Modelled soil-loss rates should be treated as relative risk rankings across a landscape, not as precise volumetric predictions. Validation against field measurements after specific storms is the only way to calibrate the model for a particular site geology.

Finally, neither optical nor SAR satellites can see buried deposits directly at the depths relevant to most archaeological stratigraphy. The satellite assessment identifies where surface erosion is occurring and at what rate. Translating that into an estimate of archaeological impact requires ground-truth knowledge of deposit depth and fragility, which must come from existing site records or targeted field inspection.

From assessment to action

The practical output of a post-fire heritage assessment is a site-risk matrix: each known site polygon ranked by modelled erosion exposure, updated at defined intervals through the first post-fire wet season. That ranking drives field response, whether emergency stabilisation measures such as straw wattles on rilled slopes, or prioritised archaeological recording before deposits are lost.

Satellize runs this workflow on open Sentinel and Landsat archives, with commercial tasking added where sub-10 m resolution is needed for individual monument detail. The analytics approach follows the same published RUSLE and dNBR methodology used in Mediterranean and Californian heritage assessments. A useful next step is submitting a site boundary file and a recent fire perimeter to scope which archive acquisitions provide a clean pre-fire baseline.

Typical figures

Burn-severity spatial resolution	20 m (Sentinel-2 SWIR bands used for NBR); 30 m (Landsat 8/9 OLI)
Optical revisit	5 days at mid-latitudes (Sentinel-2A+B combined); 8 days (Landsat 8+9 combined)
SAR revisit (Sentinel-1)	6 days over Europe; 12 days over most other regions in IW mode
Active fire detection latency	Within 3 hours of overpass via NASA FIRMS (VIIRS 375 m, MODIS 1 km)
Key spectral bands for dNBR	NIR (~865 nm) and SWIR-2 (~2190 nm); available on both Sentinel-2 MSI and Landsat OLI
Archive depth for pre-fire baseline	Landsat from 1984; Sentinel-2 from 2015; Sentinel-1 from 2014
Minimum mappable burn patch	Approximately 0.04 ha at 20 m resolution (Sentinel-2); 0.09 ha at 30 m (Landsat)
Cloud limitation	Optical sensors blocked by post-fire smoke and seasonal cloud; SAR unaffected
Delivery formats	GeoTIFF dNBR rasters, RUSLE risk-ranked site polygon GeoJSON, PDF assessment report

Analytics Satellize can run

Pre/post-fire dNBR severity map	Differenced Normalised Burn Ratio from atmospherically corrected Sentinel-2 or Landsat imagery, using published USGS severity class thresholds	GeoTIFF raster with five severity classes, clipped to site buffer zone
Post-fire erosion risk ranking for site inventory	RUSLE C-factor derived from dNBR, combined with SRTM or Copernicus DEM slope layer and published rainfall erosivity (R-factor) for the region	Site polygon GeoJSON with modelled relative soil-loss index and priority tier (high/medium/low)
SAR coherence change map	Sentinel-1 IW SLC coherence difference between pre-fire and post-fire image pairs, detecting canopy loss and surface roughness increase	GeoTIFF coherence-change layer for wet-season monitoring periods when optical data are unavailable
Vegetation recovery time series	Multitemporal NDVI and NBR from Sentinel-2 archive, compared against pre-fire phenological baseline to track C-factor recovery trajectory	CSV time series per site polygon with quarterly recovery index values and updated erosion-risk flag
Fire perimeter and timing reconstruction	NASA FIRMS VIIRS 375 m active fire detections aggregated by day to establish fire progression sequence across the landscape	Dated fire-perimeter GeoJSON for integration with site records and insurance or heritage-authority reporting
Post-fire rill formation detection	Sentinel-1 C-band backscatter time series analysed for surface roughness increase on bare burned slopes following rainfall events	Change-detection alert layer flagging slope segments with significant backscatter increase after each major rainfall event

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.