Why wildfire perimeters disagree: incident boundaries, burn scars and satellite hotspots

Tecnobosque Research Methods · Research note · Prepared 5 November 2026 · Reviewed 6 November 2026 · Methodological synthesis (not an assessment of a particular incident)

Research question. Two reputable sources assign different areas to the same wildfire. Does that prove an error, a larger fire, or a later spread? No. The answer first requires establishing what each source measured, when it observed the fire, and which geometry rules it applied.

Four maps that answer four different questions

Data class	What it principally represents	Appropriate use	Common misuse
Incident/operational perimeter	A boundary recorded or compiled by incident-management authorities	Operational history and official reporting at a stated time	Treating it as a uniformly verified burn scar
Satellite burn-scar boundary	Spectral evidence of burned vegetation/surface, delineated from imagery	Post-fire extent and spatial comparison	Assuming it has exactly the same definition as an operational perimeter
Thermal anomaly or hotspot	A sensor detection of anomalous heat in a footprint at overpass time	Indication of active-fire location and development	Converting pixels or detection symbols directly into hectares burned
Burn-severity raster	Classification or index of spectral/biophysical change	Spatial patterns of post-fire change	Equating a severity class with flame intensity or an incident’s reported acreage

NASA FIRMS explicitly cautions that a VIIRS or MODIS detection does not mean that its entire displayed pixel is burning. Hotspots can be missed by cloud and overpass timing; some anomalies may have other causes. The use of hotspot footprints as a final burned-area perimeter is therefore inappropriate without further delineation. NASA FIRMS

MTBS states that burn-area boundaries derived from Landsat/Sentinel imagery and incident perimeters need not coincide. Reasons include under-canopy fire that spectral mapping does not show clearly, the treatment of unburned islands and operational boundaries that may extend toward containment lines. MTBS FAQ

Copernicus EFFIS maps burn scars from optical imagery and refines boundaries when better imagery becomes available; it also documents sensor resolution, small-fire thresholds and the fact that deliberate prescribed fires can appear in its mapped products. Therefore EFFIS mapped area is not necessarily equivalent to a country’s incident statistics. EFFIS methods

The Tecnobosque discrepancy taxonomy (proposed editorial method)

Assign one or more explanations to each discrepancy. Do not force a single cause before evaluating alternatives.

T — Time mismatch. Boundaries correspond to different observation dates or incident stages.
D — Definition mismatch. One product measures operational perimeter, another burn scar, another thermal detections.
R — Resolution and visibility. Different image resolution, smoke, cloud, terrain, canopy or small islands alter what is detectable.
G — Geometry treatment. Slivers, multiparts, holes, topology repair or projection change computed area.
V — Version revision. A supplier reprocesses a prior observation or replaces a provisional polygon without a new burning episode.
I — Incident identity. Complexes, merged incidents, duplicate names or split records lead to a false one-to-one comparison.
U — Unresolved. Metadata are insufficient: preserve ambiguity and do not invent an explanation.

These codes are Tecnobosque’s proposed analytical labels, not an external standard. A single comparison can carry T+D+R, for example. The classification is a hypothesis until supported by incident-specific evidence.

Area difference is not spatial agreement

Suppose two synthetic sample perimeters have areas of 1,200 ha and 1,130 ha. Their net area difference is 70 ha, or about 5.8% of the first perimeter. That does not reveal whether they overlap. If the overlap is 1,060 ha, the union is 1,270 ha and the intersection-over-union (IoU) equals 1,060 / 1,270 = 0.835. Their nonoverlapping area totals 210 ha: 140 ha unique to the first perimeter and 70 ha unique to the second. These numbers are illustrative, not measurements of a real wildfire.

Useful indicators include:

Net difference: Area(A) − Area(B); useful for reporting, but it hides shifts in shape.
Intersection: Area(A ∩ B); actual common footprint.
Symmetric difference: Area(A △ B); total area belonging to only one geometry.
Intersection-over-union: Area(A ∩ B) / Area(A ∪ B); 1 means identical footprint, 0 means no overlap.
Asymmetric differences: Area(A \ B) and Area(B \ A); reveal which side contributes the disagreement.

Never calculate areas directly from unprojected degrees. Reproject both layers into a suitable metre-based equal-area CRS and document the CRS, original dataset IDs and download times. Fix invalid geometries cautiously and retain the source originals.

A source-conscious decision path

Identify the incident ID and distinguish incidents from complexes or merged records.
Record source owner, product name, download URL, acquisition date, published version and license.
Confirm both geometries answer the same measurement question before calculating a comparison.
Compare at matched time points whenever possible. Otherwise tag the time mismatch.
Calculate the geometry indicators and inspect disagreement zones against contemporaneous imagery, not only a headline area value.
Classify the discrepancy using T/D/R/G/V/I/U and describe evidence supporting each code.
Publish a conclusion at the evidence’s actual confidence level. Preserve both source statements even if they remain incompatible.

Worked interpretation: a revision is not necessarily renewed burning

An agency posts 1,200 ha on Monday and 1,235 ha on Wednesday. Three competing explanations are possible: new spread; a revised boundary derived from better mapping; or a change in which sectors were attributed to the incident. Without a progression map or dated image observations, the 35 ha change cannot be labelled new spread as a fact. Record the two publications, their observation periods and the interpretation as provisional.

Limits and safeguards

This framework is suitable for historical research and evidence auditing, not evacuation decisions or tactical deployment. Different layers can both be valid within their definitions. Map agreement metrics can be sensitive to the choice of reference scale, data gaps and the amount of boundary complexity. A low IoU alone is not evidence of misconduct, error or misreporting.