Wildfire data methods · Worked example
A burn-severity map can show a high-severity percentage without describing every hectare inside the study boundary. If clouds, shadows or other gaps remove part of that boundary from classification, the denominator becomes part of the result. This guide shows how to report the observed percentage, the missing share and transparent bounds without pretending the missing area was measured.
Documentation reviewed 30 October 2026. All numbers below are synthetic teaching examples, not measurements of an actual fire. This is an accounting guide, not a flood-risk or ecological-condition assessment.
Missing data is not an unburned class
MTBS distinguishes background outside the mapped boundary from a non-processing mask inside it. Its documentation lists sensor gaps and obscuration among reasons for missing classifications. Its “unburned to low” category is also not a pure inventory of untouched land: it can include light effects or rapid vegetation response. Keep these categories separate when counting area. [1]
The product description lists thematic severity classes separately from a mask for problematic features. Inspect the definitions delivered with the particular file rather than assuming that every raster uses the same codes. [2]
MTBS also describes scene selection and the masking of clouds, shadows, snow and other anomalies during mapping. A mapped gap therefore needs interpretation in the context of the imagery and processing workflow. [3]
Tecnobosque interpretation: a missing value does not tell you whether the ground experienced high severity, low severity or no burning. Replacing it with zero changes an unknown observation into an asserted outcome.
Worked example: 10,000 hectares with 20% unclassified
Imagine one fixed study boundary and one classified product. All areas have already been calculated on a consistent grid, using the same boundary rule. There are no overlapping categories.
| Accounting category | Area | Share of boundary |
|---|---|---|
| Mapped high severity | 2,400 ha | 24% |
| Other valid mapped classes | 5,600 ha | 56% |
| Unclassified inside the boundary | 2,000 ha | 20% |
| Total study boundary | 10,000 ha | 100% |
The classified area is 2,400 + 5,600 = 8,000 hectares. High severity occupies 2,400 ÷ 8,000 × 100 = 30% of the classified area. The mapped high-severity area occupies 2,400 ÷ 10,000 × 100 = 24% of the whole boundary.
Both calculations are correct. They answer different questions. The first describes the composition of the classified subset. The second describes the known mapped contribution to the entire boundary. Neither establishes the severity distribution within the missing 2,000 hectares.
For the preceding pixel-to-area step, use the separate guide to burn-severity area calculation, boundary rules and resampling. Do not mix a raster class total with an unrelated perimeter area simply because both are labelled hectares.
What range is possible without filling the gap?
Let H be mapped high-severity area, U be unclassified area inside the boundary and T be total boundary area. Holding the existing classifications fixed, the possible high-severity share after assigning the missing area lies between:
Lower bound = H ÷ T × 100
Upper bound = (H + U) ÷ T × 100
In the example, that gives 24% to 44%. The lower endpoint assigns none of the missing area to high severity. The upper endpoint assigns all of it to high severity.
These are accounting bounds, not a confidence interval. They do not assign probabilities to the endpoints and do not include classification errors in the observed area. They also assume that the study boundary, class definitions and area measurements remain fixed.
If you instead assume the missing area has the same 30% high-severity share as the classified subset, you would assign 600 additional hectares to high severity. The resulting estimate would be 3,000 ÷ 10,000 = 30%. Label this as an assumption-based scenario. The arithmetic cannot validate the assumption.
Why a higher reported percentage may not mean a more severely affected fire
Consider a second synthetic fire with a 10,000-hectare boundary, complete classification and 2,800 hectares mapped high severity. Its share is 28%. The first fire’s classified-only figure is 30%, but its whole-boundary accounting range is 24–44%.
A headline comparison of 30% against 28% hides the first fire’s missing area. With these inputs alone, its whole-boundary share could be either lower or higher than the second fire’s. Even complete coverage would not establish ecological comparability without checking observation dates, classification methods and landscape context.
A useful comparison table therefore includes the mapped high-severity area, classified area, total boundary area and missing share for every fire. State any excluded categories explicitly. Compare like denominators before interpreting differences.
A reporting template readers can reproduce
Within [boundary name and version], [product and observation dates] maps [H] hectares as high severity. This is [percentage] of [C] classified hectares and [percentage] of the [T]-hectare study boundary. [U] hectares ([percentage] of the boundary) are unclassified. No severity class was assigned to that missing area. Holding mapped classes fixed, the corresponding whole-boundary bounds are [lower]–[upper]%. These bounds do not account for classification error.
Preserve the original file, class legend, boundary, area ledger and processing settings with the report. If you later obtain a revised map, retain both versions and explain whether the change came from new observations, new classifications or a different boundary.
Related evidence questions
Combining several incidents? Read simple versus area-weighted burn-severity averages before averaging their percentages. Consistent missing-data treatment is necessary, but the weighting rule also changes the question being answered.
- How to read burn-severity maps: understand the product before calculating percentages.
- Why a fire may be missing from MTBS: distinguish a missing incident from a gap inside an included incident.
- Sierra Bermeja 2021: see why source-specific area figures should retain their dates and scope.
Sources and method
- MTBS: Frequently Asked Questions — class definitions and non-processing masks.
- MTBS: Product Descriptions — thematic layers and associated products.
- MTBS: Mapping Methods — image selection, boundary delineation and masks.
Method: review of official documentation and an original synthetic area-accounting example. No incident raster was processed and no missing pixels were estimated from field observations. The numerical bounds follow directly from assigning between zero and all unclassified hectares to the specified class.
Cite this guide
Tecnobosque. Missing Data in Burn Severity Maps: Percentages, Denominators and Bounds. https://tecnobosque.com/burn-severity-missing-data-percentages/. Accessed [your access date]. Cite the original product documentation for its class definitions and preserve the inputs for any calculation you reuse.