Skip to content

Fires follow different developmental pathways

Finding 01 · Developmental morphology

The primary morphospace reveals broad, reproducible gradients in when observed growth is allocated—not a catalog of discrete fire types.

Question

Can observed wildfire histories be compared on common developmental coordinates without defining those coordinates by final size, duration, or weather?

Answer

Yes. A 20-bin shape-only representation places 10,246 consecutively observed fires in a shared space. Its leading gradients contrast earlier with later allocation and concentrated with more distributed growth. Five axes summarize 89.4% of standardized variance.

Shape-only developmental morphospace, axis loadings, observation support, and example VASEs.

What to notice. The occupied space is broad and continuous. PC1 (34.1%) separates earlier from later allocation; PC2 (28.2%) contrasts middle-concentrated with more endpoint-weighted allocation. The example VASEs are landmarks along gradients, not named natural classes. The lower-left census shows why the strict cohort must not be presented as all fires. Technical caption

How to read this figure
- **Panels:** A is the shape space; B shows which relative-time bins define its first two axes; C audits observation support; D shows real example VASEs. - **Axes:** Moving along PC1 shifts allocation from earlier toward later development. PC2 contrasts middle-concentrated with more endpoint-weighted histories. - **Look here:** The points occupy a continuum rather than separate islands. Use the loadings and example VASEs together to interpret that continuum. - **Supported conclusion:** The broad gradients describe reproducible developmental variation. They are not established natural fire classes.

Challenge this result. If the gradients vanished under stricter observation requirements or another valid compositional geometry, they could be artifacts of short records or an analysis choice. See what those tests found →

Evidence trail

ClaimBroad developmental gradients, not discrete types
FigureFigure 2 · primary N = 10,246
Analysisanalysis/v2/ PCA scores, loadings, variance, and sensitivity tables

What we measured

Each consecutive history was transformed into 20 nonnegative relative-time allocation bins summing to one. The PCA used those bins only. Final area, duration, observation count, absolute peak, mean growth, slenderness, and weather were projected later rather than used to define the axes.

10,246primary consecutive histories
34.1%variance on PC1
89.4%variance on the first five axes

How well does it hold up?

  • At ≥7 consecutive observations, 1,171 fires remain. Distance ranks on shared anchors correlate 0.969 with the primary space, although 15-neighbor overlap falls to 71.8% and five-axis coverage to 74.2%.
  • Five-dimensional bootstrap subspace overlap has median 0.999 with a 2.5–97.5% range of 0.998–1.000.
  • Hellinger compositional geometry preserves the broad structure but changes some local neighborhoods and extreme exemplars.
  • Null allocations also compress, so low dimensionality alone is not evidence of a uniquely biological constraint.

What this does not show

This analysis does not establish natural fire classes, independence from endpoint attributes, transfer to long or intermittently observed fires, or a universal restricted wedge. PC1 remains moderately associated with final area, duration/count, and observed daily peak even though those quantities were excluded from fitting.

Technical details

See growth and shape coordinates, second-pass validation, and the complete figure gallery.