REMOTE SENSING GUIDE · Reviewed 8 October 2026
A burn severity map describes mapped fire effects. Its colours are not a recording of flame temperature, a building-loss inventory or a forecast of what happens next. Reading the product name, observation dates and legend before comparing colours prevents several common mistakes.
NBR, dNBR and the coloured map are different products
The Normalized Burn Ratio (NBR) uses near-infrared and short-wave-infrared observations to contrast vegetation and fire-affected surfaces. MTBS defines its burn severity primarily around effects on above-ground biomass. It does not claim to measure every ecological consequence of fire. [1]
In the MTBS workflow, pre-fire and post-fire NBR images are differenced to produce dNBR. Analysts interpret the continuous indices with imagery and ecological context, then select thresholds to create severity classes. MTBS explicitly identifies uncertainty from interpretation and limited field data; one threshold set does not serve every landscape or purpose equally well. [2]
| Product | What the reader receives | First check |
|---|---|---|
| NBR | A spectral index from one image | Which observation date does it represent? |
| dNBR | A continuous difference between two NBR images | Which before-and-after observations were selected? |
| Thematic severity | Classes produced by thresholding an index | What do the legend and metadata say each class means? |
MTBS distributes both index images and thematic layers. Its thematic products distinguish unburned-to-low, low, moderate, high and increased greenness; a mask can flag clouds, shadows, water or other problematic features. A masked pixel must not be silently counted as low severity. [3]
A practical reading workflow
Tecnobosque editorial workflow: use the following checks to document your interpretation. They are not a substitute for the product’s technical documentation.
- Name the actual dataset. Save its provider, incident identifier, product type and version. A screenshot without those details is difficult to audit.
- Record the image dates. Keep the fire date, pre-fire image date, post-fire image date and publication date in separate fields.
- Read the legend literally. Preserve the original class names and identify masked or unclassified areas before calculating percentages.
- Define the area being summarised. A fire boundary, treatment polygon and watershed can produce different percentages from the same raster.
- Write a bounded conclusion. Describe the mapped result first. Any explanation involving management, weather or suppression needs additional evidence.
Worked example: the denominator changes the answer
Illustrative numbers only—not an observed wildfire: suppose a selected boundary covers 1,000 hectares. Within it, 180 hectares are classified as high severity and 100 hectares are masked. High severity represents 18% of the whole boundary, but 20% of the 900 hectares with valid classification.
Both calculations can be arithmetically correct. They answer different questions. A reproducible statement would say: “High severity occupied 180 hectares, or 20% of classified area; 100 hectares were excluded as masked.” Retain the boundary-wide figure separately if you also report it.
This example is an accounting exercise. It does not recommend a universal denominator, infer the condition of masked land or establish that classification is error-free.
How much could the masked area change the percentage?
Worked extension added 19 October 2026. All areas below are hypothetical.
Return to the 1,000-hectare example: 180 hectares have a high-severity label, 720 have other valid labels and 100 are masked. Reporting only “20% high severity” hides the fact that one tenth of the selected boundary remains unclassified.
For a simple sensitivity check, hold every existing classification fixed and vary only how the 100 masked hectares might eventually be labelled. This produces the following arithmetic scenarios:
| Hypothetical assignment of masked area | Total high-severity-labelled area | Share of the full 1,000 ha boundary |
|---|---|---|
| None assigned high severity | 180 ha | 18% |
| Half assigned high severity | 230 ha | 23% |
| All assigned high severity | 280 ha | 28% |
The 23% row is an illustration, not a best estimate. There is no evidence here that half the masked area belongs to the high class. The 18–28% interval is also not a statistical confidence interval: it only describes extreme assignments of missing labels under the stated assumptions. Errors in existing classes, boundary choice or area measurement are outside this calculation.
A transparent report can keep three quantities together: 180 ha labelled high severity; 20% of validly classified area; 10% of the boundary masked. Include the boundary-wide 18% separately and identify its denominator. Do not silently fill masked cells with zero, redistribute them across classes, or treat them as observed low severity.
Keep three types of absence separate
- Masked cells in an available raster: preserve their area and mask status in the summary.
- A fire missing from the archive: investigate coverage, eligibility, release and identity using Why Is a Fire Missing from MTBS? This is an incident-selection question, not a pixel-classification result.
- No satellite hotspot in a selected period: use the FIRMS interpretation guide to check observation timing and detection limits. A thermal detection layer is a different product from a post-fire severity map.
Before comparing two fires, write down the same fields for each: boundary area, valid area, masked area, high-class area, denominator, product version and image dates. Matching the arithmetic makes a comparison reproducible; it does not by itself establish that the ecological meaning of the classes is identical.
Why low severity after a reburn does not mean full recovery
MTBS warns that a recently reburned landscape can show low mapped severity after herbaceous regrowth even where an earlier fire removed the forest canopy. The map reflects change relative to the selected baseline, not necessarily accumulated change across both fires or ecological integrity. Local knowledge and other datasets are needed for those questions. [1]
Calculate area from the raster
For pixel counts, hectares, boundary-cell rules and category-preserving resampling, read Burn Severity Area Calculation: Pixels, Hectares and Resampling. Its worked examples show how the analysis grid and clipping rule affect the reported area.
Choose the next calculation guide
- Part of the map is unclassified: use Missing Data in Burn Severity Maps for a complete area ledger, whole-boundary bounds and wording that keeps observed values separate from assumptions.
- You are combining several incidents: use Simple vs Area-Weighted Burn Severity Averages to decide whether every fire or every hectare should carry equal weight. The guide also explains why a seasonal percentage can fall while high-severity hectares increase.
Complete the area accounting before combining results. These guides address reproducible summaries; they do not turn a severity class into a property-level hazard forecast or a causal estimate of treatment performance.
Use the map with a case study
Read Warm Lake 2007 for a comparison of operational and satellite evidence, and Biscuit 2002 for the importance of prior fire and management history. These records provide context; this guide does not reprocess their imagery.
For a wider selection of documented events, explore the wildfire archive. If you are comparing a burn-scar boundary with active-fire detections, first read Wildfire hotspot vs fire perimeter: the products describe different observations and should not be treated as interchangeable measurements.
For terminology, see fire intensity versus burn severity. For interpreting intervention claims, use the fuel-treatment evidence guide.
Official documentation
- MTBS: Fire Mapping FAQ — scope, NBR and reburn interpretation.
- MTBS: Mapping Methods — image comparison and classification uncertainty.
- MTBS: Product Descriptions — index images, thematic classes and masks.
Method: explanatory synthesis of official MTBS documentation, with a separately labelled editorial workflow and invented arithmetic example. No new map, field validation, hazard assessment or incident measurement was produced.