A smaller pixel does not automatically mean a more accurate wildfire map. Pixel size, positional accuracy, classification accuracy and observation time answer different questions. A map can look detailed while still being unsuitable for a property-level conclusion.
This guide explains how to check those differences and includes an original worked example of resampling a 30-metre grid to 10 metres. The numerical examples are illustrative, not measurements from an incident.
Four questions to ask before using a wildfire raster
| Property | Question it answers | What it does not establish |
|---|---|---|
| Pixel or cell size | How large is each grid cell in the delivered dataset? | The correctness of its location or assigned class |
| Positional accuracy | How well do mapped locations agree with reference locations? | Whether a burn-severity label is correct |
| Classification accuracy | How well do mapped categories agree with suitable reference observations? | Whether the map represents today’s conditions |
| Observation time | When was the underlying observation acquired? | When the file was downloaded or the webpage refreshed |
Do not interpret a 30-metre cell as a claim of ±30 metres positional accuracy. That would require a separate accuracy statement and its measurement definition. Likewise, a file downloaded today may contain observations acquired weeks earlier.
Worked example: 30 metres becomes 10 metres
Imagine a square classified raster covering exactly 300 × 300 metres. Its original cells are 30 × 30 metres. Assume a projected metre-based grid, exact alignment, no missing cells and no change of extent.
- Original grid: 300 ÷ 30 = 10 cells along each side, or 100 cells.
- Area per original cell: 30 × 30 = 900 m² = 0.09 hectares.
- Total area: 100 × 900 = 90,000 m² = 9 hectares.
Now resample to a 10-metre grid with nearest-neighbor assignment and boundaries aligned so that each original cell becomes nine smaller cells.
- Output grid: 30 × 30 = 900 cells.
- Area per output cell: 10 × 10 = 100 m² = 0.01 hectares.
- Total area: 900 × 100 = 90,000 m² = 9 hectares.
The file now contains nine times as many cells. It does not contain nine times as many independent observations. In this deliberately aligned example, the nine output cells simply repeat each original class.
If 20 original cells carry a selected class, their area is 20 × 0.09 = 1.8 hectares. They become 180 output cells: 180 × 0.01 = 1.8 hectares. Multiplying the new cell count by the old cell area would instead produce 16.2 hectares—an impossible result for this 9-hectare example.
This exact area preservation belongs to the stated example. Reprojection, shifted grids, clipping and different resampling choices can change class counts. Record those operations when comparing outputs.
Why smoothing can corrupt categorical interpretation
Esri’s Resample documentation distinguishes nearest-neighbor assignment from bilinear and cubic interpolation. It cautions against bilinear and cubic methods for categorical data because they can alter class values.
For an invented legend where 1 means “unburned” and 3 means “high severity,” averaging the codes to obtain 2 does not prove that the ground belongs to a middle severity class. Class codes are labels. Use a method appropriate to the data and document it; a smoother-looking boundary is not validation.
A hotspot square is not a burned-area measurement
NASA FIRMS describes its map symbols as approximating 1 km × 1 km MODIS pixels and 375 m × 375 m VIIRS pixels, and explicitly explains that a whole displayed square need not be on fire. Do not calculate burned area by counting those squares and multiplying by their nominal area.
Read hotspot versus fire perimeter before comparing detections with an incident boundary. The FIRMS interpretation guide covers acquisition times and evidence records.
A practical checklist before reporting a result
- Identify the product: keep the provider, product name, version and download link.
- Read the units: establish whether cell dimensions are in metres, degrees or another unit. Do not treat degrees as metres.
- Separate source and output grids: record the original observation resolution and any delivered or resampled cell size.
- Record processing: retain projection, resampling method, alignment, extent and NoData handling.
- Check validation: look for positional and thematic accuracy information relevant to the intended use. If absent, say it was unavailable.
- Match the question: decide whether the data support regional comparison, area accounting or the finer-scale claim you intend to make.
For example, a regional severity layer alone does not establish what happened to an individual building. A narrow mapped strip may cross several land-cover types within one source cell. Use evidence at a suitable scale and keep unresolved detail explicit.
How to describe a resampled map honestly
Illustrative reporting statement: The source classification used 30-metre cells. We resampled it to an aligned 10-metre grid using nearest-neighbor assignment for compatibility with the analysis grid. The output cell size does not represent new 10-metre observations.
Continue with map scale and distance for screenshot measurements, or reading burn-severity maps for legends and missing-data denominators.
Sources and scope
- Esri: Cell size and resampling in analysis — resampling and the absence of new observations.
- Esri: Resample tool — interpolation methods and categorical-data cautions.
- NASA FIRMS map information — nominal detection-pixel symbols and interpretation limits.
Published 21 October 2026. Original educational calculations by Tecnobosque; no incident-specific accuracy assessment or field validation is claimed. Use official local information for current emergency decisions.