Capability 02 · Probability Assessment — A Field Guide

How to think about wildfire probability.

This page is a methods guide, not a scoreboard. It walks through how a defensible wildfire-probability assessment is actually conducted — what evidence to start from, which environmental signals to layer on, and how to combine and communicate them honestly. There are no live scores here: probability work is legacy work, built on the historical record.

Read this first

A probability is not a prediction — and it is not live. Probability assessment asks how strongly a place resembles the places where fire has historically occurred. It is an archival exercise: dated, versioned, and revisited — never a tracker of active fires, and never the sole basis for emergency, insurance, or land-use decisions. This platform deliberately does not track live fire response.

Step one · the evidence

Start from the historical record

The honest anchor for any probability statement is what has already happened. Before touching a model, assemble the record and interrogate it.

Fire perimeterswhere & how often
Map every recorded burn. Recurrence is the strongest single signal — fire-prone ground tends to stay fire-prone, because the conditions that made it burn persist.
Ignition recordswhere fires start
Starts cluster along human corridors — roads, communities, infrastructure. Ignition density tells you where the sparks are, independent of whether they grew.
Frequency as base ratethe anchor
Fires per unit area per decade, from the record itself, is your base rate. Every adjustment you make afterward should be explainable as a departure from it.
Vintage & coveragehonesty check
State the record's time span and its gaps. A quiet decade in the data can mean quiet ground — or missing reports. The assessment inherits every weakness of its record.
Step two · the modifiers

Layer on the environment

History says where fire has been welcome; the landscape says whether it still is. For each factor, the question is the same: does this raise or lower the resemblance to past fire ground?

Fuelsthe carrier
In Hawaiʻi the dominant carrier is dry invasive grass. Ask what fraction of the ground is fine fuel, and how cured it is right now versus the record's typical state.
Drynessthe slow variable
Drought loads the landscape over weeks and months. Treat it as the slow dial that moves everything else — the same fuels behave differently at different dryness.
Windthe spread driver
Wind rarely starts fires; it decides what they become. Consider prevailing patterns and terrain alignment — wind moving upslope with the fuel is the dangerous geometry.
Terrainthe amplifier
Slope preheats the fuel above the flame; aspect decides which faces cure first. Terrain doesn't create risk — it multiplies whatever the fuels and weather supply.
Exposurethe stakes
Where the wildland meets people, the same probability matters more. Keep likelihood and consequence distinct in the analysis — but present them together.
What to leave outdiscipline
Live detections, active incidents, response status — operational feeds belong to monitoring, not probability. Mixing them in turns an archive into a false forecast.
Step three · the combination

Combine transparently, or not at all

However you weigh your factors, the method — not the number — is the product. A few practices separate a defensible assessment from a black box:

Few, named factors
A handful of drivers you can defend beats dozens you can't. Every factor should have a stated source and a stated reason for being there.
Explicit weights, versioned
Write the weights down, date them, and version every change. A score that can't say which weighting produced it is not evidence — it's an opinion with decimals.
Carry the breakdown
Every assessed value should travel with its component contributions, so a reader can see why — and argue with the parts rather than the whole.
Handle gaps honestly
When a factor has no data somewhere, say so and renormalize across what remains — never let silence quietly drag the result toward zero, and never invent a value.
Invite expert review
Weightings are hypotheses about fire behavior. Publish them for the people who fight and study these fires, and expect to revise.
Bands, not decimals
Communicate in a few labeled bands. Two decimal places imply a precision the underlying record cannot support — banding is the honest resolution.
Step four · the artifact

Treat every assessment as legacy

A probability assessment is a dated document, not a dashboard. Archive each run with its inputs, weights, and vintage; revisit on a schedule, and when the record grows. The value compounds in the archive — the comparison between last year's assessment and this year's is often more informative than either alone. When someone asks "what's the risk right now," the answer is the monitoring page; when they ask "where should mitigation go," the answer is the archive.

Working here

The evidence base on this platform

Everything above can be conducted with the historical layers already in the atlas: fire perimeters with year filtering, ignition density, fuels and landcover, terrain, and land exposure — with area reports that assemble the record for any drawn boundary, and the open API for anyone building their own assessment on top of the same evidence.