TECNOBOSQUE RESEARCH METHODS · Published 5 November 2026 · Version 1.0

Wildfire Perimeter Comparison: How to Reconcile Official Boundaries, Satellite Burn Scars and Evidence Timelines

Research question: Two reputable datasets assign different areas to the same wildfire. Does that establish an error, new growth or a contradiction? Not without matching the incident identity, measurement definition, observation time and geometry.

This practical research guide brings together four methods: what boundaries measure, a reproducible QGIS comparison, evidence timestamps and versions, and burn-severity interpretation. It complements the Tecnobosque Fire Archive, the Wildfire Black Box and the technical calculators. No results below are presented as measurements of a real wildfire.

1. Why two wildfire perimeters may disagree

Area estimates are often presented as if they answer one question. In practice, datasets may describe different things at different moments. Before comparing acreage, record the source institution, product definition, observation date, download date, incident identifier, spatial reference system and known limitations.

NASA FIRMS cautions against interpreting a thermal detection as the total burned area. MTBS also documents reasons why mapped burn boundaries and incident perimeters can disagree. Copernicus EFFIS may revise mapped extent when new imagery is available. These are differences in observation and mapping, not automatic evidence of error.

A proposed discrepancy taxonomy

Tag each unresolved difference with one or more explanatory hypotheses. These are Tecnobosque editorial labels, not an international standard: T (observation time), D (product definition), R (sensor resolution or visibility), G (geometry and projection), V (dataset revision), I (incident identity or complex), and U (unresolved). Do not assign a confirmed cause without incident-specific evidence.

2. Reproducible QGIS perimeter-comparison protocol

A reproducible comparison goes beyond subtracting headline acreage. The polygons must be associated with the same incident and equivalent observation periods when possible. Keep source copies immutable, document dataset licenses and prepare an analysis layer in a suitable projected, equal-area coordinate system.

1. Identify the incident: preserve NIFC and MTBS IDs, event name, administrative area and whether the record is a complex or individual fire.
2. Preserve provenance: original source URL, publication/observation timestamps, retrieved date, product version, filename and optional SHA-256 checksum.
3. Inspect quality: geometry validity, multipart features, holes, duplicates, overlaps, positional uncertainty and source CRS.
4. Reproject: choose and record an appropriate equal-area CRS in metres for the study area; never calculate hectares in latitude/longitude degrees.
5. Compare geometry: use QGIS Intersection, Union and Symmetrical Difference; inspect unique areas with Difference A−B and B−A.
6. Calculate indicators: A, B, overlap, union, symmetric difference and intersection-over-union (IoU).
7. Investigate causes: compare historical imagery and source notes; do not equate a cartographic change with newly burned land unless supported by dated observations.
8. Publish limitations: include QA decisions, temporal mismatches, assumptions and unresolved questions.

Worked example — synthetic polygons, not a real fire

Suppose polygon A has an area of 1,200 ha, polygon B 1,130 ha and their intersection 1,060 ha. The union is 1,200 + 1,130 − 1,060 = 1,270 ha. Their intersection-over-union is 1,060 / 1,270 = 0.835 (83.5%). The symmetric difference is 1,200 + 1,130 − 2 × 1,060 = 210 ha. A-only area is 140 ha; B-only area is 70 ha. Their net area difference is only 70 ha, but the spatial disagreement is 210 ha. Net size alone is not a measure of boundary agreement.

3. Evidence timestamps, revisions and “what was known when?”

Historical wildfire reconstruction requires a distinction between an event and the record of that event. A report published on Wednesday describing conditions observed on Monday was not necessarily available to decision-makers on Monday.

For every evidence item, record at least observed_at (measurement or event time), published_at (public availability), retrieved_at (when the investigator downloaded it), recorded_at (when added to this archive), and effective_at where an official decision has a specified effective time. Store timezone and confidence. A missing timestamp must remain missing, not silently inferred.

Maintain source-specific versions and store corrections without deleting earlier observations. To test a historical known-at claim, establish that the relevant document was accessible by the cutoff time, not just that the underlying phenomenon had already occurred. Keep a link, cited excerpt, document version, hash where appropriate, and the reasoning behind every claim. This approach underpins the Wildfire Black Box.

Example: If a public agency first says 1,200 hectares on Monday and later revises it to 1,235 on Wednesday, the change may reflect new spread, better mapping, or incident-boundary adjustments. Preserve both statements. Do not label the additional 35 hectares as newly burned without progression evidence.

4. NBR, dNBR and why burn severity is not fire intensity

The Normalized Burn Ratio compares near-infrared (NIR) and short-wave infrared (SWIR) reflectance: NBR = (NIR − SWIR) / (NIR + SWIR). The commonly used differenced index is dNBR = NBR before fire − NBR after fire. These indices describe spectral change. They do not directly measure temperature, flame length or energy release.

A higher positive dNBR can indicate stronger post-fire change, but interpretation depends on ecosystem, acquisition season, vegetation and soil, cloud/shadow masking, sensor processing and the chosen classification thresholds. The same numeric threshold should not be transferred blindly between landscapes. Unburned areas, regrowth and pre-existing disturbance also matter.

Synthetic numerical example: if NBR before fire is 0.64 and after fire is 0.08, dNBR = 0.56. This is an arithmetic illustration only; it is not a severity-class assignment for a real fire. Any real severity map should cite its product and scene dates, classification scheme and validation limits.

5. How this helps interpret the Tecnobosque archive

A case study should not be assessed solely by its date, country or displayed acreage. A robust record links its official incident identifier, source datasets, historical versions, perimeter definitions, computations, confidence and explanatory narrative. Cross-links between cases can then identify recurring discrepancies, map-treatment differences and evidence gaps rather than merely similar fire names.

Explore the historical wildfire archive for case records, Wildfire Black Box for versioned evidence, and Wildfire Tools & Calculators for related technical utilities. This page presents a research method, not an emergency prediction or tactical fire-safety instruction.

Primary sources and methods references

• NASA FIRMS — thermal anomaly data and usage guidance
• Monitoring Trends in Burn Severity (MTBS) — frequently asked questions
• Copernicus EFFIS — rapid damage assessment methods
• National Interagency Fire Center — open geospatial datasets
• QGIS — vector overlay algorithms
• USGS — Landsat normalized burn ratio

Editorial transparency: The discrepancy labels and timestamp schema are proposed Tecnobosque research conventions, not official incident findings. Numerical examples are synthetic. Source data must be inspected and licensed independently before a real comparison is published. Updated 5 November 2026.