Most big Southeast Asian fires don't start where fire crews look first, study finds

On a single working day in late September 2015, satellite sensors catalogued more than 1,200 active fire detections across Sumatra and Kalimantan as smoke closed airports in Singapore, Kuala Lumpur and parts of Thailand. By the time the haze lifted weeks later, the crisis had cost the region tens of billions of dollars in economic damage and pushed daily pollution readings in several Indonesian provinces more than thirty times the World Health Organization's guideline for fine particulates. The fires are the worst the equatorial belt has recorded in the satellite era.

A new study of that exact episode, surfaced by Physical Sciences on 17 July 2026, argues that the response playbook designed around those sensors may be looking in the wrong place. By retracing the earliest detectable origins of the largest blazes, the authors find that almost none of the catastrophic fires began as a single point on a map. They grew.

The one-point fire is a satellite artefact

Modern fire response treats each hot-spot detection as a discrete ignition. Logistics, dispatch and suppression budgets are then built around that assumption. The new analysis keeps the satellite detections but goes back further, asking where each cluster first appears at sub-kilometre resolution and how those clusters accumulate in the days before the plume is large enough to dominate imagery. The headline result is uncomfortable: the biggest fires almost all had multiple origin points spread over tens of kilometres, with new ignitions joining existing complexes at a near-daily cadence. Single-point starts were the exception, not the rule, and when they did occur they rarely produced the largest blazes.

The practical upshot is straightforward. A response model that treats each new hot spot as a fresh start has to chase the fire; a response model that expects coalescence can pre-position crews, machinery and peat-water assets along the corridors where new starts are statistically most likely. Singapore's 2015 air-quality bill, Indonesia's fire-fighting budget, and Malaysia's haze agreements have all been written assuming the first model. The paper says the second is closer to the ground truth.

Where the counter-narrative holds

The dominant regional framing treats the 2015 fires as a peatland story: drained, degraded carbon soils on the coasts of Sumatra and Kalimantan, dried by El Niño, ignited by the cheapest available hand, and impossible to extinguish without monsoon rains or massive water bombing. The new findings do not undercut that account so much as they re-rank it. Peat remains the fuel that turns ordinary agricultural burns into a multi-week atmospheric event. But the ignition pattern is not what most of the published literature has assumed. A peatland that is dry enough to burn is dry enough to host dozens of starts a day, several kilometres apart, and any of them can hook the same carbon layer.

This is also where the paper speaks most directly to corporate accountability. The widely publicised palm-oil concession map for 2015 puts the largest of those concession blocks inside the very provinces where multiple-origin growth was most pronounced. That doesn't by itself prove a specific corporate actor deliberately set the fires. What it does is remove the plausible cover that a given large fire 'originated' somewhere outside a concession and 'drifted' inward. The growth pattern the authors document makes that single-point story much less credible for the largest events.

A structural read of the response lag

Southeast Asia's fire crisis is now in its third decade of escalating policy attention without a clear downward trend in bad-air years. The same paper indirectly explains why: the metric being optimised is hot-spot count and area-burned, both of which look very different after a multi-origin growth phase than they do after a single ignition. By the time daily hot-spot counts peak, the fire has effectively been burning for a week or more and any chance of a cheap suppression outcome has already gone. The reason the metric hasn't moved much is partly that the metric is being measured on a single-point model of a multi-point phenomenon.

The same structural problem shows up in carbon accounting. The method used to attribute emissions to concession holders relies, in practice, on fire-perimeter mapping at the date of satellite overpass. A fire that grew by accretion across the boundary of two concessions will get split between them on the day of overpass and re-attributed later, often after political negotiation. A method that tracked daily origin-point clusters would, in principle, identify who was responsible for the ignition more cleanly. Indonesia's national peat agency and several provincial governors have the data; what has been missing is a published method that handles multi-origin growth explicitly.

What to watch before the next El Niño

The 2026 dry season is mid-cycle across equatorial Asia. The strongest El Niño signature in the historical record sits on the 2015 episode the study examines, and forecasters at regional meteorological agencies have already pointed to a weak-to-moderate signal in mid-2026. If the paper's findings hold under independent replication, three things should be observable in the next bad-air year: a higher fraction of large events traceable to multi-origin growth than the current single-point model predicts; measurable cost reductions in provinces that pre-position suppression assets along empirically identified ignition corridors; and at least one jurisdiction shifting its concession-level fire attribution method toward origin-cluster tracking rather than single-overpass perimeter mapping.

The open questions are honest ones. The dataset covers one watershed year; the multi-origin pattern needs to be re-tested for the 2019 and 2023 episodes and for non-El Niño years before it can be treated as a default rather than a finding. The paper does not specify how its detection thresholds generalise to cloudier fire seasons, nor does it resolve the long-running academic dispute over peatland ignition versus peatland combustion as the limiting step. Those are the points where replication and argument will land.

Monexus framed this study for its operational consequences rather than its theoretical novelty: the next bad-air year arrives on a calendar, not a publication schedule.

Wire provenance

This editorial synthesis draws on the following public wire/social posts:

• https://en.wikipedia.org/wiki/2015_Southeast_Asian_haze
• https://en.wikipedia.org/wiki/Peatland_fire