Wildfire Burn Severity Mapping with Landsat — How dNBR Works
When a wildfire happens, the standard way to quickly and objectively answer "how much burned, and how badly" is satellite-based dNBR analysis. Without deploying a helicopter or walking the entire perimeter, two satellite images — one from before the fire, one from after — are enough to grade the burned area by severity. This guide explains the underlying principle and how to run it yourself in the browser, with no API key.
NBR: putting a number on burned ground
The core idea is simple. Healthy vegetation strongly reflects near-infrared (NIR) light and reflects relatively little shortwave infrared (SWIR). Burned, charred ground does the opposite — NIR reflectance drops sharply while SWIR reflectance increases. Turning that contrast into a single index gives you NBR (Normalized Burn Ratio).
Landsat uses Band 5 (NIR) and Band 7 (SWIR2). Higher NBR values indicate healthy vegetation; low or negative values suggest the ground has burned.
dNBR: the difference between before and after
A single NBR snapshot can't tell you whether an area is "naturally like this" or "burned and became like this." So you subtract the post-fire NBR from the pre-fire NBR. That difference is dNBR (differenced NBR).
The larger the change (the higher the dNBR), the more severe the damage at that location. This approach filters out seasonal variation and areas that were already bare ground, isolating only what was newly burned by this specific fire.
Burn severity classes (USGS)
Splitting dNBR values using the classification scale compiled by the US Geological Survey (USGS) turns the raw number into a severity class. The scale is conventionally expressed as the value × 1000.
| dNBR (×1000) | Class |
|---|---|
| < −100 | High post-fire regrowth |
| −100 to 99 | Unburned / unchanged |
| 100 to 269 | Low severity |
| 270 to 439 | Moderate-low severity |
| 440 to 659 | Moderate-high severity |
| ≥ 660 | High severity (near-total loss) |
Image selection makes or breaks the result
In practice, what determines result quality isn't the formula — it's which images you choose. A few principles:
- Pre-fire image: as close as possible to right before the fire, and cloud-free. If the season is very different, it introduces noise into the dNBR.
- Post-fire image: right after the fire is contained. If it's too late, vegetation will have started regrowing and severity will be underestimated.
- Clouds: mountainous terrain often has frequent cloud cover. Cloud and cloud-shadow pixels need to be masked out (via a QA_PIXEL mask) to reduce false positives.
- Resolution: Landsat's 30m resolution comes with an archive going back to the 1980s, which is useful for historical events.
Running it in the browser
- Define the area you want to analyze on the map.
- Enter the wildfire date (or the month it occurred).
- Before/after satellite imagery is selected automatically and dNBR is computed.
- Damage is shown color-coded by severity class, with area (in hectares) tallied per class. Results can be exported as GeoJSON or Shapefile.
Limitations and caveats
Satellite dNBR is optimized for broad-area screening. It won't tell you whether an individual tree survived, or reveal damage below the surface. And during periods of heavy cloud or haze, there may simply be no usable imagery at all. Treat the results as a first-pass tool for prioritizing field verification and detailed assessment — final damage figures should still be confirmed on the ground.