Comparing Burn Severity Percentages: Simple vs Area-Weighted Averages

Wildfire data methods · Comparing incidents

To calculate the high-severity share across a set of fires, add the high-severity areas and divide by the sum of the corresponding analysis areas. Taking the simple mean of each fire’s percentage answers a different question: it gives every incident equal weight, regardless of size. Both summaries can be useful, but they should never share an ambiguous label such as “average damage.”

Reviewed 30 October 2026. All numerical examples are invented and independently checked. They demonstrate arithmetic, not observed regional trends or treatment effectiveness.

Two fires, two valid answers: 50% or 26%?

Assume two non-overlapping incidents, fully classified within their respective analysis boundaries. Use the same class interpretation and area method for this teaching example.

The mean of the two incident percentages is (80 + 20) ÷ 2 = 50%. That is the mean high-severity share per incident in this two-fire set.

The pooled area share is (80 + 180) ÷ (100 + 900) × 100 = 26%. That describes the proportion of the combined analysis area mapped in the high-severity class.

Fire B contributes nine times as much area as Fire A, so it contributes nine times as much weight to the pooled calculation. The 24-percentage-point difference between the answers is not a data error: it comes from changing the question.

Choose the question before choosing the formula

For each incident i, let Hᵢ be its high-severity area, Aᵢ its chosen analysis area and pᵢ = 100 × Hᵢ/Aᵢ. For n incidents with positive denominators:

• Mean incident percentage: (p₁ + p₂ + … + pₙ) ÷ n.
• Pooled area percentage: 100 × (H₁ + H₂ + … + Hₙ) ÷ (A₁ + A₂ + … + Aₙ).

The second expression equals a weighted mean of pᵢ using Aᵢ as the weights. Use unrounded areas for the calculation and round the final result. Do not combine whole-boundary areas for some fires with classified-only areas for others.

The first measure treats each included fire as one observation. The second treats each unit of included area as contributing equally. Neither represents fires omitted from the dataset without additional evidence about coverage.

A seasonal percentage can fall while high-severity area rises

Imagine Year 1 contains only Fire A: 80 high-severity hectares out of 100 hectares, or 80%. Imagine Year 2 contains a separate pair of incidents with the same dimensions as A and B: 260 high-severity hectares out of 1,000 hectares, or 26%.

The pooled percentage falls from 80% to 26%, while high-severity area rises from 80 to 260 hectares. A statement that “high severity decreased” would therefore be incomplete. Specify whether you mean area, percentage, incident frequency or another outcome.

This constructed example does not establish a climate trend or show that management improved. It illustrates how the composition and size of a dataset can change a summary. Report absolute area alongside proportions before discussing causes.

Four checks before combining real maps

1. Comparable observations. MTBS documents initial and extended assessments and selection of prefire imagery to match seasonal conditions. Preserve observation dates and assessment types; a shared fire year alone does not establish matching observation windows. [1]
2. Comparable class meaning. MTBS notes that threshold interpretation involves uncertainty and that ecological meaning can vary across landscapes. Matching class names do not make every ecological comparison valid. [2]
3. Consistent missing-area treatment. Report excluded or unclassified area separately. See the worked example on missing data and severity-percentage bounds. A pooled classified-area share is still conditional on what was classified.
4. A stated rule for repeat burning. Summing incident areas can count the same ground more than once across events. That may suit an event-area total, but it is not the unique land footprint. A unique-area statistic requires an explicit spatial overlap rule and a decision about which observation describes each location.

For pixel dimensions, clipping and resampling, use Burn Severity Area Calculation. Correct weighting cannot repair incompatible input areas.

A concise reporting template

Across [n] included incidents, [H] hectares were mapped high severity within [A] hectares of [defined analysis area], giving a pooled share of [P]%. The unweighted mean incident share was [M]%. Results use [product, dates and class definitions]. Missing areas were treated as [rule], and overlapping incident footprints as [rule]. These statistics describe the selected dataset, not a causal estimate of management effectiveness.

Keep one row per incident with its identifier, numerator, denominator, missing area and source version. That lets a reader reproduce both averages and identify whether one large fire dominates the pooled result.

Sources and method

1. MTBS Product Descriptions — imagery, assessment types and thematic products.
2. MTBS Mapping Methods — interpretation, thresholds and ecological limits.

The equations and hypothetical examples are Tecnobosque’s mathematical explanation. No regional dataset was analysed and no new incident measurements were produced. For map interpretation, continue to How to Read Burn Severity Maps.

Cite this guide

Tecnobosque. Comparing Burn Severity Percentages: Simple vs Area-Weighted Averages. https://tecnobosque.com/comparing-burn-severity-percentages/. Accessed [your access date]. Preserve the input table when reusing a calculation.