Methodology

What our words mean

These distinctions are used consistently across the whole app. A term further down this list is a stronger claim than one above it: reaching it requires more evidence, not just more time.

Raw satellite thermal anomaly

A single pixel where a satellite instrument detected unusually high heat. On its own it is not evidence of a wildfire.

Potential incident cluster

A group of nearby, close-in-time thermal anomalies the clustering engine has linked as probably the same heat source. Still not a confirmed fire.

Fire reported by an FRS

An incident a fire and rescue service has logged as reported, before it has been confirmed as an active wildfire.

Verified active wildfire

An incident a named organisation, usually a fire and rescue service, has confirmed as an active wildfire.

Contained

The spread of a verified wildfire has been stopped, though monitoring or mop-up may continue.

Monitoring

A wildfire is being watched for renewed activity but is not currently classed as spreading.

Extinguished

A verified wildfire has been officially declared out.

Formally declared major incident

A status only ever applied when the responding service has made a formal major incident declaration. It is never inferred from fire size or duration.

Physically large fire

A fire covering a large area on the ground. This is a size description, not a formal status: a large fire is not automatically a major incident.

Forecast fire danger

A modelled forecast of how favourable conditions are for a fire to start and spread, built from weather and rainfall. It is a forecast of conditions, never evidence that a fire exists.

From pixels to incidents

A wildfire usually lights up several satellite pixels across several passes, not one. Two detections are linked into the same potential incident cluster when they fall within 1.5 km of each other and within 24 hours of each other; a nearby detection arriving more than 24 hours after a cluster's last one starts a new cluster instead. Separately, a cluster with no fresh detection for 20 hours is flagged "no recent detections" and its map marker fades. That gap is NOT evidence the fire is out: it may be extinguished, smouldering below the roughly 375 m detection threshold, hidden by cloud, or simply between satellite passes; only a fire and rescue service can confirm extinguishment. Each cluster's mapped extent is buffered by 375 m around its detections to account for pixel footprint and satellite geolocation error.

Confidence rules

Confidence answers one question: how likely are these detections to be a real ongoing fire, rather than noise or a false alarm. It never changes what status a cluster has, only how much the map and lists signal about it.

Industrial and persistent heat source checks

Flares, incinerators and similar industrial heat sources produce satellite heat signatures that can look like a wildfire on paper. Two independent checks catch most of them before they reach anyone's screen as a probable incident.

Fire danger model

Fire danger is a weighted blend of temperature, humidity, wind and rainfall deficit, plus the Copernicus EFFIS Fire Weather Index where it is available. Each factor is normalised to 0 to 1 before weighting, so no single unit dominates the score just because of its scale.

Normalisation ranges

Bands

A score at or above the very high threshold (0.84) is banded Exceptional.

When the EFFIS Fire Weather Index is unavailable for a point, its weight is not simply dropped: it is redistributed proportionally across temperature, humidity, wind and rainfall deficit, so the score stays on the same 0 to 1 scale. Fire danger is always a forecast of conditions that favour a fire starting or spreading. It is never evidence that a fire exists.

Urban Threat Index and Overall Incident Risk

The Urban Threat Index (UTI) blends how hazardous a cluster looks with how much is nearby to be affected, then applies a wind alignment factor. Overall Incident Risk instead blends hazard, exposure and the official response already under way. Both are 0 to 100 decision-support scores, not predictions of harm.

UTI blend

Hazard weights

Recency reaches full weight within 6 hours of the last detection and fades to zero by 48 hours. Persistence saturates at 4 satellite passes. Fire radiative power saturates at 60 MW.

Exposure weights

Population and buildings, by ring

When no building count is available (for example the Overpass API is disabled or unreachable), the buildings weight is redistributed to population rather than treated as zero exposure.

Downwind concern sector and wind factors

The downwind concern sector is a cone pointing the direction the wind is blowing toward, with a half angle of 35 degrees and a length of 2 km plus 0.55 km for every m/s of wind speed, capped at 10 km. It is built only from current or forecast wind direction and speed.

It is not a fire-spread perimeter prediction: the platform has no calibrated fuel load, terrain or fire-behaviour model feeding it, so it deliberately makes no claim about how far a fire would actually travel. It is a plain-English "who is downwind now, and when might that change" summary, nothing more.

Response

Duration saturates at 72 hours since first observed. Concurrency saturates at 6 other active incidents in the same fire and rescue service area. Station distance scales linearly from 3 km (near) to 20 km (far).

Overall Incident Risk weights