How the 2015 Southeast Asian fires exploded: tracing fire origins reveals multi-point ignition

On the second week of September 2015, satellite sensors over Sumatra and Kalimantan returned an unbroken ribbon of red pixels stretching from the southern tip of Sumatra to the northern coast of Borneo. By the end of that month, Indonesia's National Disaster Management Agency had logged more than 100,000 active fires across the archipelago, and Singapore's three-hour Pollutant Standards Index had crossed 400, well past the threshold at which the city issues a public health alert. The 2015 equatorial fire season, the most severe on record at the time, became the reference event against which every subsequent haze crisis in the region is measured. A study published in late July 2026 argues that the standard reference frame is wrong: the disaster was not a single fire, but a synchronised set of fires that ignited at multiple, geographically separate points.

The paper, summarised by Phys.org on 17 July 2026, used a back-tracing method on the earliest detectable ignition footprints of the largest fires recorded during the 2015 season. The headline finding is statistical rather than visual: most of the catastrophic fires had multiple origin points, not one. The implication is uncomfortable for anyone designing an early-warning system around a single ignition source. If a fire's footprint can be assembled from a dozen distributed starts, then a system that watches for one spark will, by construction, miss the conditions that allow those sparks to combine into a single conflagration.

A different way of counting where a fire begins

Conventional fire mapping tends to assign each burned patch to a single ignition point. Where the source data are coarse, that single point is often a centroid, a derived pixel that has no physical meaning on the ground. The new analysis took the opposite path. Working from the earliest detectable thermal signatures, the researchers walked each fire backward, asking: how many distinct origins can be resolved before the signal collapses into noise? In most of the large fires, the answer was more than one, and in many cases, several points were active within the same day, separated by tens of kilometres.

This is not a small methodological wrinkle. Fire management in the region has historically concentrated on peatland drainage, land-clearing burns, and concession mapping. Indonesian and Singaporean agencies have invested heavily in hotspot alerts and concession-permit enforcement. A multi-origin finding does not invalidate any of that work. It does suggest that the fire load is more distributed than the hotspot map suggests, and that weather windows, rather than individual operators, may be doing most of the assembly work. A dry spell sets the stage; dozens of legally and illegally lit fires then act in concert.

The weather window did most of the amplification

The 2015 season was conditioned by a strong El Niño, which suppressed rainfall across Sumatra and Kalimantan and dried out peat domes that had accumulated carbon over decades. When the dry window opened, the ignition events that would otherwise have smouldered locally spread into sustained burns. Multi-origin ignition means that no single operator, no single concession, no single village, can be cleanly assigned responsibility for the eventual smoke plume that shut Southeast Asian airports for days. The system, not the spark, was the unit of harm.

This finding matters for how responsibility gets assigned, and to whom. The regional haze agreements that emerged after 2015, including the ASEAN Agreement on Transboundary Haze Pollution and various bilateral arrangements between Indonesia, Singapore and Malaysia, rely on being able to trace plumes back to specific concession holders. Multi-origin ignition complicates that picture without relieving anyone of obligation. It does, however, push the policy conversation upstream, toward land-use planning, peat-dome rehydration, and the timing of burn windows, rather than only toward enforcement at the moment of ignition.

What an effective early-warning system would actually watch for

If the fires that matter most are assemblies rather than single events, then the warning system that matters is the one that flags the conditions under which assemblies form. The study points toward three operational changes. First, hotspot density over time, not hotspot presence at a moment, is the leading indicator worth alerting on. Second, weather windows defined by soil-moisture collapse, wind patterns, and peat dryness need to be published on the same dashboard as the hotspots, so that concession holders and fire managers are reading the same forecast. Third, the back-tracing method itself can be run in near-real time during fire seasons, allowing agencies to recognise a multi-origin event in its first hours rather than in its post-mortem.

None of this is a technical barrier. The satellite infrastructure is in place: NASA's MODIS and VIIRS instruments, the European Sentinel fleet, and regional assets such as Indonesia's LAPAN platforms provide continuous coverage. What is missing is operational integration, the unglamorous work of running the back-trace algorithm inside the agencies that issue the public alerts.

The case that the multi-origin finding is overstated

It is fair to ask whether the back-trace method is over-reading its own signal. Multi-origin ignition is sensitive to how the algorithm defines a distinct origin. Set the threshold too tight, and every fire looks like many fires; set it too loose, and the most consequential conflagrations collapse back to a single point. The paper acknowledges this in its discussion of sensitivity, but does not yet settle it.

There is also a regional-knowledge argument against the headline. Veteran Indonesian fire managers, working the same 2015 season from helicopters and ground teams, tended to describe the crisis in terms of a small number of industrial land-clearing operations and the wind direction on a given day. Their intuition was that a few large actors drove most of the smoke. The multi-origin finding does not contradict that, but it does distribute moral and operational weight in a way that the practitioners' narrative does not. Either the satellite back-trace is resolving structure that ground teams could not see at the time, or it is over-resolving structure that does not exist on the ground. Future work, ideally combining isotopic smoke-source fingerprinting with the satellite back-trace, will settle it. For now, the result is strong enough to change what the early-warning dashboards should display, and not strong enough to retire the operator-centric story.

Desk note: this piece leans on the Phys.org summary of the underlying paper rather than the paper itself; readers seeking the full method should go to the source link in the references.