Post-wildfire erosion and debris-flow susceptibility mapping
Intense wildfire destroys the vegetation and soil structure that hold slopes together. Combining burn severity from satellite spectral indices with slope, soil type, and storm intensity data produces debris-flow susceptibility maps that can reach emergency managers before the first post-fire rain.
Sensors
- Sentinel-2 MSI: 10 m resolution in visible and near-infrared bands; 20 m in shortwave infrared (SWIR) bands 11 and 12, which are the spectral basis for dNBR calculation. Five-day revisit at mid-latitudes (2-3 days with both satellites). Free and open archive from 2015.
- Landsat-9 OLI-2: 30 m multispectral resolution including SWIR bands 6 and 7 used for NBR. Sixteen-day repeat cycle, offset eight days from Landsat-8 to give an effective eight-day combined revisit. Archive continuity with Landsat-5/7/8 allows pre-fire baselines going back decades.
- MODIS Terra/Aqua: 250 m to 500 m resolution; daily to twice-daily revisit. Too coarse for slope-scale susceptibility mapping but useful for rapid burn-extent triage over large fires before Sentinel-2 or Landsat-9 acquire a clear-sky pass.
- Planet SkySat: 50 cm resolution tasked optical imagery. Useful for validating debris-flow deposit boundaries and detecting early channel incision in high-priority catchments, but not a free or systematic archive; requires commercial tasking.
Why fire makes slopes dangerous in a completely different way
Wildfire removes two things a slope depends on: the root network that mechanically binds soil particles, and the organic layer that allows water to infiltrate rather than run off. A third effect is less intuitive. Combustion drives hydrophobic compounds from burning organic matter downward into the soil profile, creating a water-repellent layer a few centimetres below the surface. Rain that would normally soak in instead sheets across the surface, picking up ash and loose sediment and concentrating rapidly in channels. The result is debris flows that can travel at several metres per second, carrying boulders and charred timber, and arriving with almost no warning.
The USGS Emergency Assessment of Post-Fire Debris-Flow Hazards methodology, developed after destructive events in southern California and Colorado, formalises the physical logic. It combines burn severity, watershed slope, soil erodibility, and a rainfall intensity threshold (typically the 15-minute peak intensity of a design storm) to estimate both the probability and volume of debris flows at the catchment outlet. Satellite data enters directly at the burn-severity step.
What dNBR actually measures, and where it misleads
The Normalised Burn Ratio uses the ratio of near-infrared to shortwave infrared reflectance. Healthy vegetation reflects strongly in the near-infrared and absorbs in the SWIR; fire reverses that relationship, leaving char, exposed soil, and ash with low NIR and elevated SWIR. The differenced NBR (dNBR) subtracts the post-fire image from the pre-fire baseline. USGS and the US Forest Service have published thresholds classifying dNBR values into unburned, low, moderate, and high severity classes, with high severity generally above a dNBR of roughly 0.66, though those thresholds vary by vegetation type and should be validated against field data.
Sentinel-2 at 20 m SWIR resolution and Landsat-9 at 30 m are both adequate for catchment-scale severity mapping. The honest limit is cloud cover: smoke and post-fire cloud formation can delay a usable acquisition by days, which matters when the first thunderstorm may arrive within a week of containment. MODIS can fill the gap for extent triage, but at 500 m it cannot resolve the fine-grained severity patterns that determine which sub-catchments are most dangerous.
A subtler problem arises weeks to months after the fire. Certain fast-growing pioneer species, particularly annual grasses, can green up enough to push NBR values back toward pre-fire levels within a single growing season. A repeat-pass image acquired two months post-fire may show a recovering spectral signal while the hydrophobic soil layer is still intact and root cohesion has not returned. Treating that image as evidence of reduced hazard would be wrong. Susceptibility assessments should anchor to the dNBR computed from the first clear-sky post-fire acquisition, not from imagery acquired weeks later.
Building the susceptibility layer: what goes in beyond the burn map
Burn severity is one input. The USGS methodology also requires slope gradient derived from a digital elevation model (the agency uses 10 m National Elevation Dataset data for US assessments), soil texture and erodibility from databases such as STATSGO or SSURGO, and a rainfall intensity estimate for a specified return period, typically the two-year, 15-minute peak intensity from NOAA Atlas 14 in US contexts. Each catchment upstream of a road crossing, structure, or community is treated as a unit of analysis.
The probability and volume estimates are empirical, calibrated against historical debris-flow inventories. They carry real uncertainty, particularly for soil types not well represented in the calibration dataset or for fire behaviour unusual enough to produce severity patterns outside the training range. Outputs should be presented as susceptibility classes rather than precise flow volumes. Emergency managers need to know which drainages to watch; they do not need a false sense of numerical precision.
Temporal urgency: the window between fire and rain
Post-fire debris-flow hazard is highest in the first one to two years after a fire, and acutely highest in the first wet season. In Mediterranean climates, that window can be very short. The 2018 Montecito, California debris flows that killed twenty-three people occurred less than five weeks after the Thomas Fire was contained, triggered by a storm delivering roughly 0.5 inches of rain in five minutes over the burn scar.
That timeline sets a hard operational requirement. A susceptibility map that arrives after the first significant rain event is useful for the next storm but useless for the one that just caused casualties. Satellite data pipelines need to be running before the fire is fully contained, using near-real-time acquisition scheduling and automated dNBR computation so that a draft severity layer is ready within 24 to 48 hours of the first cloud-free post-fire pass. Latency in data ordering or processing is not an abstract technical problem; it is a life-safety issue.
Limits to be honest about before presenting results
Several sources of error deserve explicit acknowledgement in any operational product. First, satellite-derived burn severity does not directly measure the depth or continuity of the hydrophobic soil layer; it is a proxy. Two areas with identical dNBR values may have very different infiltration behaviour depending on fire residence time and soil moisture at the time of burning. Second, DEM quality matters enormously for slope-based routing. A 10 m DEM is adequate for catchment delineation; a 30 m DEM can misroute flow paths in complex terrain. Third, the USGS empirical models were calibrated primarily on fires in the western United States. Applying them to other fire regimes, say, eucalyptus-dominated slopes in southern Europe or Australia, requires local validation.
Finally, vegetation recovery imagery can mislead if used uncritically, as noted above. Any programme that uses repeat-pass satellite monitoring to track post-fire landscape recovery should maintain a separate susceptibility layer anchored to the initial post-fire severity assessment and update it only when field evidence or direct soil measurements confirm that hydrophobicity has genuinely diminished. Satellize applies this discipline in its analytics workflows, keeping the initial dNBR-derived severity layer as a fixed reference separate from any subsequent greenness monitoring.
From map to decision: what emergency managers actually need
A susceptibility map is not an evacuation order. It is a prioritisation tool. The most useful format is a catchment-level table ranked by combined probability and potential volume, cross-referenced with the infrastructure or population at the outlet. That allows a road authority to decide which culverts to pre-clear, or a county emergency manager to decide which communities need early-warning rain gauges deployed before the next storm.
Rainfall thresholds derived from the USGS methodology can also be converted into operational alert triggers. When a weather forecast or gauge network indicates that 15-minute rainfall intensity is approaching the threshold for a high-susceptibility catchment, an automated alert to the relevant authority can compress the warning time. The satellite data is the foundation; the operational value comes from connecting it to the forecast and the decision.
Typical figures
| Burn severity spatial resolution | 20 m (Sentinel-2 SWIR); 30 m (Landsat-9 OLI-2); 500 m (MODIS, triage only) |
| Revisit interval (clear-sky) | 2–3 days at mid-latitudes (Sentinel-2A+B combined); 8 days (Landsat-8+9 combined); 1–2 days (MODIS) |
| Typical post-fire acquisition latency | 24–72 hours for first usable Sentinel-2 or Landsat pass, subject to cloud cover |
| Key spectral bands for dNBR | NIR (~865 nm) and SWIR (~2200 nm); Sentinel-2 bands 8A and 12; Landsat-9 bands 5 and 7 |
| DEM resolution for slope input | 10–30 m recommended; coarser than 30 m degrades catchment delineation accuracy |
| Minimum mappable burn patch | Approximately 0.04 ha at 20 m resolution (Sentinel-2); 0.09 ha at 30 m (Landsat-9) |
| Susceptibility model calibration basis | USGS empirical debris-flow probability and volume equations, calibrated on western US fire inventory |
| Archive depth | Sentinel-2 from 2015; Landsat continuous from 1972 (Landsat-1) enabling multi-decade pre-fire baselines |
| Delivery formats | GeoTIFF severity raster, vector catchment susceptibility polygons, ranked CSV table, PDF hazard summary |
Analytics Satellize can run
| dNBR burn severity raster | Pre/post-fire NBR differencing using Sentinel-2 MSI or Landsat-9 OLI-2; USGS/USFS severity classification thresholds | GeoTIFF severity layer (unburned / low / moderate / high) with confidence notes on cloud-affected pixels |
| Catchment-level debris-flow susceptibility map | USGS Emergency Assessment methodology: burn severity + slope + soil erodibility + design-storm rainfall intensity | Vector polygon layer with per-catchment probability class and indicative volume range, ranked priority table |
| Rainfall intensity alert thresholds | USGS empirical threshold equations applied to susceptibility classes; linked to forecast or gauge inputs | Per-catchment threshold table (mm/15 min) for integration with weather warning systems or gauge networks |
| Vegetation recovery masking layer | Time-series NDVI and NBR from repeat Sentinel-2 passes; flagging of pixels where spectral recovery precedes confirmed soil recovery | Monthly raster update with recovery-mask overlay, plus advisory note distinguishing spectral from physical recovery |
| Infrastructure exposure summary | Spatial intersection of susceptibility catchment outlets with road network, building footprints, and critical facility datasets | PDF report and GIS layer showing assets within high-susceptibility flow paths, for emergency manager briefing |
| Rapid triage severity estimate (pre-Sentinel/Landsat) | MODIS Terra/Aqua burn area product (MCD64A1 or MOD14) for same-day extent approximation | Preliminary burn perimeter and rough severity zones within 24 hours of fire containment, superseded by higher-resolution product when available |
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.