Mapping the Jasper burn scar to within 6.5% of the national reference

Lightning lit three fires in the Athabasca valley on the evening of 22 July 2024. The evacuation alert went out at eight o'clock and the order at ten, moving more than 20,000 residents and summer visitors out of Jasper on a single highway in the dark.

Two days later the south fire crested Whistlers Mountain with flames thirty to fifty metres high, spotting five hundred metres ahead of itself in winds gusting over a hundred kilometres an hour. It entered the town at 18:40 on 24 July and structures were burning by eight. When it was done, 358 of 1,113 buildings were gone, about 800 homes, roughly a third of the townsite. Firefighters held the water treatment plant, the school and the hospital.

Insured losses reached just under $1.3 billion, the second costliest fire in Canadian history. The fire was held on 17 August and under control on 7 September, with 278 kilometres of perimeter. A 24 year old Alberta Wildfire crew member was killed by a falling tree on 3 August. Park visitation halved the following year, from 2.41 million to 1.14 million, and visitor spending came in $278 million under forecast.

The analysis covers the park and the town on a box drawn from the national burn perimeter with about ten kilometres of margin, on four archive passes: 24 June and 10 July before the fire, then 19 August and 4 September after it.

Two conditions govern whether an answer of this kind is worth anything, and both are fixed before scoring. The area sits at the corner of three acquisition paths and only one covers the whole burn, so every scene is drawn from that path and each reaches full coverage of the box. Cloud, shadow and snow are masked from the mission's own quality band, which matters more here than on most ground: melting snow is bright in the near infrared and dark in the shortwave, precisely the signature a burn leaves.

The test itself is published, not ours. Ground counts as burned where the normalised burn ratio falls by 0.27 or more, the Key and Benson reference break the platform already carries. It is not fitted to this fire. The comparison runs only over ground that carried vegetation beforehand.

The same test also runs across the two pre-fire scenes. Nothing happened between 24 June and 10 July, so whatever that pair returns is what the method produces from ordinary variation, and every number after it is read against that.

The cleanest scene of the series was in the archive twenty six days after ignition, thirteen months before the definitive map was published. For an insurer setting reserves, or a park authority scoping salvage and closure, that is the difference between a decision made on evidence and a decision made on an incident command estimate.

Two things about the comparison are worth stating plainly to anyone who will rely on the number. The national reference is derived from the same family of optical instruments, so this is an independent agency, an independent team and an independent method, but not an independent sensor family: a systematic error common to that family would move both figures the same way. And a fixed threshold catches a fire's low severity edge differently from an analyst's hand drawn perimeter, which is why total change across the box is 45,599 hectares against a 29,597 hectare connected scar. The connected component is the quantity that corresponds to what the reference measures.

This is burn extent, and only extent. The platform's severity classes were scored against the United States federal reference across thirty fires with the sample fixed before scoring. Extent passed its gate. The severity classes failed theirs, are published as failed, and are not offered. The Forest Service's own report notes the same weakness from the other side: burn ratio methods do not separate severity well in stands already disturbed by mountain pine beetle, which describes a good deal of this park.

Nothing here speaks to structures. The 358 buildings lost were counted on the ground.