GeoAnalitikBack-trajectories say the haze-day air passed over both Sumatra and Kalimantan, Sumatra more often, and no area reaches the 70% rule.
The study follows the air reaching Johor Bahru and Singapore back five days and asks which detections it passed low over. Trajectories show where the air passed, not how much smoke it carried, and a detection is not a fire. The winds, the timing and the modelled smoke below sit beside that test; none of them can tell one island’s smoke from the other’s.
Of the agreed haze-day arrivals, 52% at Johor Bahru and 52% at Singapore had passed low over Sumatran detections in the five days before, against 39% and 40% over Kalimantan’s
A haze day is one on which Johor’s station-average daily mean API was 101 or more, on station-days with 18 or more hours; there were 26. The study traced the air arriving every 3 hours at 500, 1,000 and 1,500 m above Johor Bahru and Singapore back 120 hours, as a 27-member ensemble. An arrival passes over a detection when its path, inside the highest boundary layer of the previous 24 hours, comes within 25 km of an S-NPP vegetation-fire detection dated that day or the day either side.
The 70% rule was set before the run, and no area reaches it. An arrival counts only when 18 of the 27 ensemble members agree; 39–40% of arrivals do. Trajectories show where the air passed, not how much smoke it carried. 16 Oct MYT is partial (02:00 and 05:00 only).
Sumatran detections passedKalimantan’sElsewhereOne arrival’s path
Johor Bahru 256 agreed arrivals
Sumatra 52% Kalimantan 39% Mixed under 1% No fire contact 9% Not agreed 350
256 of 606 haze-day arrivals agreed (solid, shares of these); 350 did not (hollow).
Singapore 250 agreed arrivals
Sumatra 52% Kalimantan 40% Mixed 1% No fire contact 7% Not agreed 356
250 of 606 haze-day arrivals agreed (solid, shares of these); 356 did not (hollow).
Figure 1. Where the agreed haze-day air passed. Each ink hairline is one arrival’s centre run, 120 h back, drawn every 6 h, the three start heights pooled. Cells of 0.25° hold the S-NPP detections those runs passed: lime for Sumatra (with the Riau Islands and Bangka-Belitung), grey for Kalimantan, pale for anywhere else. A few paths run on beyond the frame to the south and east. The bar under each map holds all 606 haze-day arrivals: solid where 18 of the 27 members agreed, split by what they passed, and hollow where they did not. Trajectories: the study’s calculation on ERA5, not an ECMWF product. “Contains modified Copernicus Climate Change Service information 2026. Neither the European Commission nor ECMWF is responsible for any use that may be made of the Copernicus information or data it contains.” Detections: NASA FIRMS. Land: geoBoundaries outlines, simplified by the study. Map data from OpenStreetMap (https://www.openstreetmap.org/copyright), via geoBoundaries, available under the Open Database License (ODbL).
Sumatra is the top label under the main rule and each of the eight sensitivities, at 46–72% of agreed haze-day arrivals at Johor Bahru and 47–70% at Singapore
Each setting changes which arrivals the ensemble agrees on (from 236 to 371 at Johor Bahru, 237 to 384 at Singapore); none changes the top label. Only the shortest run, 72 h back, comes near the 70% rule (72% at Johor Bahru, 69.8% at Singapore); the rule is judged on the main setting, where Sumatra holds 52%.
| Main rule | 52% | |
|---|---|---|
| 72 h back instead of 120 | 72% | |
| 168 h back | 48% | |
| Hourly boundary layer | 46% | |
| Fixed 1,500 m layer | 64% | |
| 50 km instead of 25 | 50% | |
| All three VIIRS satellites | 53% | |
| Weighted by fire power (FRP) | 54% | |
| Leaving out low paths over 10% outside the fire table | 51% | |
| 070% rule100% |
| Main rule | 52% | |
|---|---|---|
| 72 h back instead of 120 | 70% | |
| 168 h back | 47% | |
| Hourly boundary layer | 47% | |
| Fixed 1,500 m layer | 65% | |
| 50 km instead of 25 | 49% | |
| All three VIIRS satellites | 52% | |
| Weighted by fire power (FRP) | 54% | |
| Leaving out low paths over 10% outside the fire table | 51% | |
| 070% rule100% |
Figure 2. Sumatra’s share of the agreed haze-day arrivals under each setting, in a fixed order (lime dot), against the 70% rule (ink line). The main rule: 120 h back, the highest boundary layer of the previous 24 h, 25 km, S-NPP detections. The others change one thing each: 72 or 168 h back, the hourly boundary layer, a fixed 1,500 m, 50 km, all three VIIRS satellites, weighting by fire radiative power, or leaving out arrivals with more than 10% of their low path outside the fire table’s area.
The last four haze days, 13–16 October, turned to Kalimantan: none of their agreed arrivals was labelled Sumatra
Of the agreed arrivals on 13–16 Oct, 60 at Johor Bahru and 59 at Singapore passed over Kalimantan’s detections, and 3 at Johor Bahru passed none. 16 Oct is partial. Four days do not overturn the full-window finding.
| Label | Johor Bahru | Singapore |
|---|---|---|
| Kalimantan | 60 | 59 |
| Sumatra | 0 | 0 |
| No fire contact | 3 | 0 |
| Agreed arrivals | 63 | 59 |
Table 1. Agreed arrivals on 13–16 Oct MYT by label, the three start heights pooled; 16 Oct has its 02:00 and 05:00 arrivals only.
Diagnostic, not part of the rule: weighted by fire contacts the trajectories lean to Kalimantan (54–56%), and counted once per detection the lean is stronger (62–63%)
Weighting each haze-day centre run by the detections it passed, rather than counting arrivals, gives Kalimantan 54% at Johor Bahru and 56% at Singapore. Counting each detection once across all those runs (81,937 at Johor Bahru, 86,592 at Singapore) gives 62% and 63%. This measures something other than the arrival rule: a detection passed by many runs counts once, so it is a diagnostic, not part of the rule. Height changes the answer: at 500 m contacts lean Kalimantan (79% and 80%), at 1,500 m Sumatra (60% and 61%).
| Arrival rule (Figure 1) | 52% 39% | |
|---|---|---|
| Weighted by fire contacts | 42% 54% | |
| Each detection once | 29% 62% | |
| By start height | ||
| 500 m, by contacts | 13% 79% | |
| 1,000 m, by contacts | 39% 57% | |
| 1,500 m, by contacts | 60% 37% | |
| 500 m, each detection once | 9% 78% | |
| 1,000 m, each detection once | 29% 65% | |
| 1,500 m, each detection once | 42% 52% | |
| Arrival rule (Figure 1) | 52% 40% | |
|---|---|---|
| Weighted by fire contacts | 39% 56% | |
| Each detection once | 27% 63% | |
| By start height | ||
| 500 m, by contacts | 11% 80% | |
| 1,000 m, by contacts | 32% 64% | |
| 1,500 m, by contacts | 61% 36% | |
| 500 m, each detection once | 9% 74% | |
| 1,000 m, each detection once | 26% 67% | |
| 1,500 m, each detection once | 41% 53% | |
Figure 3. A diagnostic. It measures something other than the arrival rule; the 70% rule is unchanged. Rows: the arrival rule of Figure 1; the haze-day centre runs weighted by their fire contacts; each detection those runs passed counted once; then the last two by start height. Sumatra lime, Kalimantan grey, everywhere else pale, as in Figure 1.
From September the 850 hPa wind over the Malacca Strait turned to blow from the south and south-south-east, the bearing of South Sumatra and Jambi rather than Kalimantan
Over the Malacca Strait and the Riau Islands the monthly mean wind blew from 200–217° in July and August, 167–175° in September and 149–168° on 1–10 October. From Johor Bahru and the Riau Islands, South Sumatra and Jambi lie at 159–197° and Kalimantan’s fire points at 96–118°. From 1 Sep to 10 Oct the daily wind blew from 90–130° on 1 of 40 days at Johor Bahru and 2 at the Riau Islands, and from 150–195° on 21 and 24. Point winds are not trajectories and cannot tell a Sumatran source from a Kalimantan one.
| Point on the smoke path | Jul | Aug | Sep | Oct 1–10 |
|---|---|---|---|---|
| Malacca Strait, off Klang | SSW 211° | SSW 206° | SSE 167° | SSE 149° |
| Kuala Lumpur | WSW 241° | SSW 213° | SSW 192° | SSE 162° |
| Malacca Strait, off Dumai | SW 217° | SW 215° | S 171° | SSE 157° |
| Riau Islands | SSW 200° | SSW 205° | S 175° | SSE 168° |
| Johor Bahru | SSW 210° | SSW 211° | S 176° | SSE 165° |
| Karimata Strait | SSE 161° | SE 143° | SE 128° | SE 129° |
| South China Sea, off Kuching | SSW 207° | SSW 204° | S 186° | SSW 213° |
| Kuching | SSW 194° | S 177° | SSE 158° | S 182° |
Figure 4. ERA5 850 hPa wind at eight points on the smoke path, monthly vector means, 2026 (October: 1–10 Oct). The arrow points the way the wind blew; the figure under it gives the direction it blew from. Ticks on the rim at three points mark the bearings to the fire points, lime for Sumatra’s and grey for Kalimantan’s: a wind from a tick blows from that fire point. “Contains modified Copernicus Climate Change Service information 2026. Neither the European Commission nor ECMWF is responsible for any use that may be made of the Copernicus information or data it contains.”
Sumatra’s detections track Peninsular haze within about two days (r about 0.65); Kalimantan’s track East Malaysia only weakly, and the Peninsula not at all
Pearson r between NOAA-20 detections and the mean of station daily-mean API, on Malaysian days: 40 days, 2 Sep – 15 Oct. Sumatra against the Peninsula holds 0.63–0.67 at lags of zero to two days and falls to 0.39 at three; Kalimantan against East Malaysia is 0.22 the same day and 0.28–0.33 at one to four days; Kalimantan against the Peninsula is −0.14 to 0.06. The two fire days without NOAA-20’s night pass over Sumatra and Kalimantan, 8 Sep and 13 Oct, are left out; keeping them moves r by up to 0.07. These series are autocorrelated: a pattern, not a test. No fixed delay can be read from it.
Sumatra’s detections, haze over the Peninsula
0.650.670.630.390.18
01234
Kalimantan’s detections, haze over the Peninsula
0.000.06−0.02−0.04−0.14
01234
Kalimantan’s detections, haze over East Malaysia
0.220.310.310.280.33
01234
Sumatra’s detections, haze over East Malaysia
−0.28−0.25−0.18−0.040.00
01234
Lag in days: haze that many days after the detections.
Figure 5. Pearson r at lags of 0 to 4 days for each pair, ink dots joined by a hairline; no lag is singled out. Detections: NASA FIRMS, NOAA-20, by island. Haze: DOE APIMS station daily means, station-days with 18 or more hours, by side of Malaysia.
Modelled smoke over East Malaysia rose from 10–11 µg/m³ in late July to 75 in the week of 31 August; the Peninsula’s model peak was 50, in the week of 5 October
Station-average PM2.5 in Copernicus’ CAMS forecasts at the 68 DOE station points, in Monday–Sunday weeks, 1 Jul – 15 Oct. A model, used for timing and comparison, never a reading: against DOE’s daily means over 40 days, 2 Sep – 15 Oct, pooled r is 0.74 and the median station’s 0.71. At the 65 fixed stations’ points alone, East Malaysia’s peak week is 75.7 and the Peninsula’s 49.3. Seasons are not ranked: the EAC4 reanalysis reverses 2015 and 2019 on both sides. 2026 is forecast only.
East Malaysia
Peninsula
Figure 6. Weekly means of modelled PM2.5 (µg/m³) at the station points of East Malaysia (Sarawak, Sabah, Labuan) and the Peninsula, 1 Jul – 15 Oct: 2026 in lime (forecast), 2015, 2019 and 2023 in grey, each season on its own forecast model, drawn by date. The first and last weeks are cut by the window. “Contains modified Copernicus Atmosphere Monitoring Service information 2026. Contains modified Copernicus Climate Change Service information 2026. Neither the European Commission nor ECMWF is responsible for any use that may be made of the Copernicus information or data it contains.”
Regional context: ASMC raised its alert to Level 3, its highest, on 26 August, and Singapore’s 24-hour PSI reached 229 on 16 October
PSI is Singapore’s index and is not API; these readings are quoted from their sources and are not in the study’s data.
| When | Reading | What the source says | Reported by |
|---|---|---|---|
| ASMC alert level, southern ASEAN region | |||
| 10 Jul | Level 1 | start of the dry season for the southern ASEAN region | ASEAN Specialised Meteorological Centre |
| 3 Aug | Level 2 | raised as prolonged dry weather increased hotspots and smoke haze | ASEAN Specialised Meteorological Centre |
| 26 Aug | Level 3 | raised to Level 3, ASMC’s highest; no later change listed as of 16 Oct | ASEAN Specialised Meteorological Centre |
| Singapore 24-hour PSI | |||
| 9 Oct | 192 | central region, 5 pm to 8 pm, then the 2026 high, just below the 201 Very Unhealthy line | The Star |
| 14 Oct | 203 | west region at 9 pm, the first Very Unhealthy reading (201–300) of 2026; the west stayed there until 2 pm on 15 Oct | National Environment Agency |
| 15 Oct | 212 | west region at noon, its peak | National Environment Agency |
| 16 Oct | 229 | central region at 3 pm, the highest reading of 2026 as of 16 Oct | National Environment Agency |
Table 2. ASMC alert levels for the southern ASEAN region, and Singapore’s 24-hour PSI, each from one report; each outlet’s name links to its report. PSI is Singapore’s index and is not API. “Contains information from Pollutant Standards Index (PSI) (National Environment Agency) accessed on 17 Oct 2026 from data.gov.sg which is made available under the terms of the Singapore Open Data Licence version 1.0 https://data.gov.sg/open-data-licence”.
Limits
- No area reaches the 70% rule set before the run, so the study names no single source area: both, Sumatra more often.
- Trajectories show where the air passed, not how much smoke it carried. The ensemble agrees on 39–40% of arrivals, and the shares rest on those.
- The trajectories have no turbulence or convection, and sea breezes below the 1000 hPa level are not resolved; 8.5% of them leave the wind box within 120 h, all outside the area of the detections.
- Point winds are not trajectories. There is no fixed delay between burning in Indonesia and haze in Malaysia, and the study does not claim one.
- Seasons are not ranked on modelled smoke: the reanalysis does not confirm the forecasts’ order.
- Papua and West Papua are counted in every Indonesia total; their smoke is not traced.
- The study names no company or concession.
Data, method and credits
- Back-trajectory method: follows HYSPLIT’s published scheme, the two-step advection of Stein et al. (2015), https://doi.org/10.1175/BAMS-D-14-00110.1, and the 27-member offset-meteorology ensemble of the NOAA Air Resources Laboratory HYSPLIT User’s Guide (https://www.ready.noaa.gov/hysplitusersguide/S253.htm); HYSPLIT software not used.
- Winds and modelled smoke: “Contains modified Copernicus Atmosphere Monitoring Service information 2026. Contains modified Copernicus Climate Change Service information 2026. Neither the European Commission nor ECMWF is responsible for any use that may be made of the Copernicus information or data it contains.” ERA5 and CAMS, CC BY 4.0; the trajectories are the study’s calculation, not a Copernicus or ECMWF product.
- Detections: “We acknowledge the use of data and/or imagery from NASA's Fire Information for Resource Management System (FIRMS) (https://earthdata.nasa.gov/firms), part of NASA's Land, Atmosphere Near real-time Capability for Earth observations (LANCE) (https://earthdata.nasa.gov/lance) and NASA's Earth Science Data and Information System (ESDIS).”
- Air quality: Department of Environment Malaysia (Jabatan Alam Sekitar), Air Pollutant Index Management System (APIMS), https://eqms.doe.gov.my/APIMS/main. “Data dan maklumat tertakluk di bawah Terma Penggunaan Data Terbuka Kerajaan Malaysia 1.0” (https://www.doe.gov.my/wp-content/uploads/2021/07/TermaPenggunaan.pdf).
- Boundaries: geoBoundaries (Runfola et al. 2020, https://doi.org/10.1371/journal.pone.0231866), CC BY 4.0; the land outline contains information from OpenStreetMap, under the ODbL.
- Boundary files under the ODbL, free online at geoBoundaries’ commit-pinned URLs: github.com/wmgeolab/geoBoundaries/raw/9469f09/releaseData/gbOpen/IDN/ADM1/geoBoundaries-IDN-ADM1.geojson, github.com/wmgeolab/geoBoundaries/raw/9469f09/releaseData/gbOpen/MYS/ADM1/geoBoundaries-MYS-ADM1.geojson, github.com/wmgeolab/geoBoundaries/raw/9469f09/releaseData/gbOpen/TLS/ADM0/geoBoundaries-TLS-ADM0.geojson, github.com/wmgeolab/geoBoundaries/raw/9469f09/releaseData/gbOpen/PNG/ADM0/geoBoundaries-PNG-ADM0_simplified.geojson and github.com/wmgeolab/geoBoundaries/raw/9469f09/releaseData/gbOpen/THA/ADM0/geoBoundaries-THA-ADM0_simplified.geojson. We repaired and simplified them (tolerance 0.0005°) and used them to assign fire detections to provinces, states and countries.
- Regional context: ASMC alerts and Singapore PSI as reported, one row each in the study’s research file, with links.
- No credited source or agency endorses this study, and no figure here is official.
- Every licence, DOI and the method: Method and credits.