Mapping 75 days of fires in Indonesia and the spread of haze across Southeast Asia.
On 15 September, all five regions of Singapore reported unhealthy air for the first time since 2019. Schools have closed in parts of Indonesia, Malaysia, Brunei and the Philippines. TIME and the ABC report Indonesia’s worst fire season in eleven years.
I’ve mapped 75 days of satellite fire detections and modelled smoke to see where the fires are and how the haze spreads. Press play to follow the season, or select a date.
Orange dots show NASA MODIS fire detections. The brown layer shows smoke estimated by the CAMS atmospheric model. By default it shows smoke through the full air column; the toggle switches to estimated pollution at ground level.
Fire activity increased sharply around 5 August. The weekly detection footprint tripled from July’s 7,000–8,000 km², reaching 59,000 km² in the week of 26 August. It stayed around 50,000 km² a week through most of September. These figures measure the area of pixels with fire detections, not the area burned.
Since 1 August, Borneo has recorded almost four times Sumatra’s fire footprint. Three-quarters of Borneo’s total comes from West and Central Kalimantan. West Kalimantan peaked in August, while Central Kalimantan’s September footprint is nearly double its August total.
On Sumatra, most activity is in the south: South Sumatra, Bangka Belitung, Lampung and Jambi. Riau has been relatively quiet, consistent with the Ministry of Forestry’s hotspot reports.
The monsoon carries Kalimantan’s smoke towards Kuching and Brunei, then across the South China Sea towards the Philippines. Select 31 August to see the plume over the central Philippines. Smoke from southern Sumatra travels up the Strait of Malacca towards Singapore and the Klang Valley. Singapore’s surface pollution rises in mid-September as fires increase in South Sumatra.
Each row is a city and each cell is a day. Colours show modelled daily average PM2.5, grouped by air-quality index thresholds. The WHO daily guideline is 15 µg/m³.
Since 1 August, Palangka Raya has spent 36 of 58 days above 150 µg/m³ in the model, with a peak of 1,023 µg/m³ on 21 September. Kuching peaks on 30 August, Brunei on 31 August, Singapore on 14 September and Kuala Lumpur on 15 September.
Switch to smoke aloft to see the difference over the Philippines. Cebu’s smoke signal on 1 September is nine times its July average; Manila’s peaks at four times its average on 31 August. The model shows little of this reaching the ground, despite reports of very unhealthy air in Manila. I’d be cautious about its surface estimates this far from the fires: a roughly 40 km grid can miss local differences in how smoke mixes downwards.
Surface PM2.5 also includes local pollution. Jakarta, upwind of the fires, stays around 85 µg/m³ in the model. Near the fires, the model may underestimate concentrations.
The MODIS record starts in 2001, giving us 26 years to compare.
Borneo’s fire footprint through 28 September ranks sixth against the full-year totals since 2001, just behind 2015’s. August was its largest since 2004. No earlier year had a higher January–September total, and 2026’s still has two days to run.
Sumatra’s year-to-date footprint ranks fifteenth. It is about 5% behind 2019 at the same point in the year, about 23% below 2015 and less than half of 2006. Activity is increasing: September’s footprint is already a third larger than August’s, largely because South Sumatra’s has more than doubled.
These comparisons measure detected fire activity. They don’t tell us how much land burned or how much smoke was produced. MODIS can miss smouldering peat fires beneath thick smoke, which helps explain why 2015 looks less severe here than its haze would suggest. YKAN estimates that around 6,000 km² actually burned in August, about a tenth of the detection footprint.
Everything here is built from public data through Google Earth Engine; the scripts are in github.com/tuttinator/satellite-images.
| Layer | Source | What was done |
|---|---|---|
| Active fires | NASA FIRMS, MODIS Collection 6.1 NRT, 1 km | Daily presence of a detection, aggregated to the fraction of each 0.1° cell, × cell area |
| Smoke aloft | ECMWF CAMS NRT, organic-matter AOD at 550 nm, 0.4° | Mean of the forecast steps under 12 hours’ lead for each UTC day |
| Surface PM2.5 | ECMWF CAMS NRT, 0.4° | Same averaging; nearest cell for the city series |
| Fire seasons | FIRMS monthly footprint, 2001–2026 | Summed over FAO GAUL 2015 provinces for Sumatra and Borneo (incl. Sarawak, Sabah, Brunei) |
| Coastlines | Natural Earth 1:50m | Clipped and simplified |
MODIS passes four times a day and cannot see through thick cloud or smoke. Daily totals are noisy, and the last two or three days are provisional. Only Terra was operating before mid-2002, so the early record has fewer detections. CAMS combines satellite measurements and a fire-emissions inventory with an atmospheric model; its surface concentrations are estimates. All daily averages use UTC.
Updated 29 September. The fire record runs to 28 September and the smoke data to 27 September.
For current local air quality, use your national monitoring service: NEA in Singapore, APIMS in Malaysia or BMKG in Indonesia.