When Soil Moisture Is a Fire Early-Warning Signal, and When It Is Not
When Soil Moisture Is a Fire Early-Warning Signal, and When It Is Not
1. The question
Soil moisture is the most physically obvious leading indicator of fire: fuel has to dry before it burns. If dryness precedes fire by a predictable interval, that interval is an early-warning horizon a system can act on. The founding question of this thread was whether that horizon exists, how long it is, and, crucially, whether it is the same everywhere. An early-warning system built on a five-week moisture lead is only as good as the assumption that the lead is present where the system is deployed. So the real test was not "does soil moisture predict fire" in one place, but "where does it, where does it not, and why".
2. Method: a comparative lead-lag
For each region we built a weekly time series of soil moisture and of fire activity, then cross-correlated fire against soil moisture at leads of zero to twelve weeks. The lead that maximizes the correlation is the early-warning horizon; the strength of that correlation is how usable it is. We ran the identical procedure across five tropical zones (the southern Amazon basin as a whole, Bolivia, Guaviare, the Xingu, and Bogotá) and, for the temperate comparison, across the US West, Canada boreal and Mediterranean Europe using annual burned area against concurrent and accumulated drought. Two deliberate controls guard against fooling ourselves: we repeated the tropical test with a different satellite (VIIRS, which sees far more small fire than MODIS), and in Bogotá we repeated it against the Bomberos ground incident record rather than any satellite at all.
3. The five-week lead is real, in the moisture-limited south
In the southern Amazon the signal is unambiguous. Fire activity is most strongly anti-correlated with soil moisture measured about five weeks earlier: for the southern basin the peak lead is five weeks at r = −0.85, and for Bolivia five weeks at r = −0.82. The correlation curve has a clear minimum at that lag rather than at zero, which is the signature of a genuine lead rather than mere concurrency. This is the case that motivated the whole EWS: a month or more of warning, driven by the fact that in this system moisture is what stands between available fuel and fire.
4. But it is not universal: elsewhere the lead is zero
Run the same test a few hundred kilometres away and the horizon vanishes. In Guaviare, in the Xingu, and in Bogotá, the correlation is weak and peaks at zero lag: there is no interval at which past moisture forecasts future fire. The contrast is not subtle. Where the southern basin bottoms out at −0.85 five weeks ahead, Bogotá drifts around −0.22 with its strongest value at zero weeks. The seasonal phase loops tell the same story from the other side: the moisture-limited zones trace a wide loop from dry to burning to wet, while the no-lead zones collapse toward a point. Soil moisture is simply not the variable that gates fire in those places.
| Zone | Peak lead | r at peak | Reading |
|---|---|---|---|
| Southern basin | 5 weeks | −0.85 | strong moisture lead (usable EWS) |
| Bolivia | 5 weeks | −0.82 | strong moisture lead (usable EWS) |
| Guaviare | 0 weeks | −0.42 | no lead |
| Xingu | 0 weeks | −0.24 | no lead |
| Bogotá | 0 weeks | −0.22 | no lead (ignition-limited) |
5. Why: three fire regimes, one binding constraint each
The pattern makes sense once fire is read through its locally binding constraint. Fire needs fuel, dryness and an ignition all at once; whichever is scarcest is what governs, and that is what a leading indicator must track. In the moisture-limited south, fuel is abundant and ignitions are frequent, so drying is the last domino and moisture leads fire by weeks. In fuel-limited systems (deserts, and to a degree boreal forest), there is always enough dry weather, so fire instead follows antecedent wetness that grew the fuel, and concurrent dryness rather than a moisture lead is what tracks burned area. In ignition-limited systems like Bogotá, a dense human-wildland interface where people start the fires, dryness is neither scarce nor predictive on its own, and no moisture horizon appears.
The temperate retrospective confirms the split. Concurrent fire-season dryness tracks annual burned area robustly in the US West (ρ = −0.85) and Canada boreal (ρ = −0.66), so concurrent dryness generalizes widely. But the multi-year accumulated-drought effect that matters in the Amazon does not carry over: it is weak and non-significant in the US West, Canada and the Mediterranean. In the fire-versus-drought phase portraits this appears as hysteresis. Canada boreal traces a large memory loop, the fingerprint of a fuel-limited system where wet years build fuel and a later dry year ignites it, while the Amazon tracks a near-single line, the fingerprint of a moisture-limited system responding directly to current dryness.
6. It is a regime, not a measurement artifact
A zero lead could in principle mean the instrument is too coarse to see the fire, rather than that no lead exists. Two checks rule that out. First, the tropical result is invariant to sensor: VIIRS detects three to five times more fire than MODIS, especially the small fires, yet returns the identical lead structure, strong in the south and absent in the no-lead zones. If the zero lead were a detection gap it would move when we changed detectors; it does not. Second, and most directly, Bogotá's zero lead holds against the Bomberos ground incident record, not just against satellites. Even though MODIS caught only about 27 percent of real Bogotá fire-weeks, the ground record itself shows no soil-moisture lead. The absence of a horizon there is a fact about the fire regime, not about what we could see.
7. What this means for early-warning design
The soil-moisture EWS is a precision tool, not a general one. It should be deployed where moisture is the binding constraint, the moisture-limited tropics like the southern Amazon and Bolivia, where it buys a real multi-week horizon. It should not be the basis of warning in ignition-limited systems. Bogotá is the clear case: there is no moisture horizon to forecast, so the leverage is spatial rather than temporal, knowing where fires will become large from the fuel structure, which is exactly what the companion vegetation-heterogeneity and receptivity work provides. A useful operational rule falls out of this: before standing up a moisture-based EWS in a new region, run this one-page lead-lag test first; a flat, zero-lag correlation curve means the moisture EWS will not work there and a different indicator is needed.
One important nuance remains open. A zero concurrent lead does not by itself distinguish a fuel-limited regime from an ignition-limited one; both return "no lead". The disambiguator is antecedent-rainfall coupling, the positive correlation between prior-season wetness and fire that cleanly separated desert from Amazon elsewhere in the project. Running that test on the no-lead zooms (Guaviare, the Xingu) would turn each "no lead" into a positive regime label, and is the natural next step for this thread.
title: When soil moisture is a fire early-warning signal, and when it is not
creator: Alejandro Feged-Rivadeneira, Tecnologico de Monterrey
date: 2026-07-20
sources: weekly soil-moisture and MODIS/VIIRS active-fire series (southern Amazon basin, Bolivia, Guaviare, Xingu, Bogota); UAECOB Bomberos forest-fire record 2009-2025; regional annual burned area and drought (US West, Canada boreal, Mediterranean Europe)
methods: weekly fire-vs-soil-moisture cross-correlation lead-lag (0-12 wk); seasonal phase loops; fire-drought phase portraits and hysteresis; concurrent vs accumulated drought retrospective; sensor-invariance and ground-record controls
data-sha256: b2ae83ac8b79f62749a821672819507c73cd324146d07f6a2b8621fb0b988c3d
license: CC-BY 4.0
Note: Only a member of this blog may post a comment.