This season
—days
Checking this week’s air in Kuala Lumpur and Singapore…
This week
When will the haze end?
Running the daily-hazard simulation…
The season model did not answer.
The fires are unaffected by this outage.
—more days
until the Kalimantan fire season ends, by the model’s median. That is not a date for clear air in Kuala Lumpur or Singapore: Sumatra burns too, and the wind decides whose sky the smoke reaches. Haze can clear before that date or linger after it.
The sky above is drawn from the model's reading once it arrives.
Until the median end date
—days
—hours
—min
—sec
This week, where you live
The last seven days of air in Singapore and the Klang Valley, and the fires behind it. Each chart uses one index only. Singapore’s PSI and Malaysia’s API (IPU) are different scales, so they never share an axis.
Fetching this week’s readings…
This week’s readings are not available.
For official readings, see haze.gov.sg (Singapore) and APIMS (Malaysia).
The fires behind it
Where these numbers come from
- Singapore PSI and PM2.5: National Environment Agency, via data.gov.sg, under the Singapore Open Data Licence v1.0. Measured.
- Malaysia API (IPU): Department of Environment Malaysia. Where this page has no official DOE reading it says so; the official figures are on APIMS.
- Modelled US AQI and PM2.5: Copernicus Atmosphere Monitoring Service (CAMS), via Open-Meteo, CC BY 4.0. A forecast model on a grid of about 40 km, not a measurement.
- Fire detections: NASA FIRMS, VIIRS near-real-time (LANCE NRT). firms.modaps.eosdis.nasa.gov
Licence notices
And then there is the other ending.
The fuel-exhaustion model did not answer.
Kalimantan runs out of things to burn in —
The odds, as the model sees them
The same simulation as the answer above, asked as yes-or-no questions. Each percentage is a count, not an opinion: the share of simulated seasons in which it happened.
Model odds. Not a market — no money changes hands. Each percentage is the share of N simulated seasons.
Reading the odds off the simulation…
Odds are not available yet.
Which week does it end?
Share of simulated seasons ending in each week.
How the odds have moved
Each line is one question, re-asked every day with only the weather known on that day.
Where it burns
Provinces shaded by peat area — peat is what makes Kalimantan smoke rather than simply burn. Move through the seasons to replay each year's fire clusters, lit in the order they started, sized by area and coloured by what was burning.
Loading provinces and fire clusters…
The map data did not load.
loading
Bars above the slider: burned area each year, coloured by ENSO phase.
Every season on record
The pattern is not subtle: El Niño years burn wide and burn long, La Niña years barely register. That is the whole forecast, and it has been the whole forecast for decades.
Loading the season record…
The history endpoint did not answer.
Burned area in Kalimantan, with the ONI index
Bars: burned area (Mha), coloured by ENSO phase; the wide pale bar behind is all of Indonesia. Line: peak Oceanic Niño Index, right axis. Select a bar to load that year on the map.
When each season ran
Every season's burn window laid across the calendar, most recent at the top. Late June to early November, year after year.
Change the inputs
The model is not an oracle; it is a set of assumptions with a sigmoid attached. Move the assumptions and watch the two answers move. Nothing here is a policy recommendation.
Fuel exhaustion
—
| Pool | Gone by (p50) | Left in 2050 |
|---|
What is holding the season open
Each driver's portion of the termination hazard's log-odds magnitude. Together they make one — they describe how the model is weighting things, not how the atmosphere is.
No drivers to show.
Methods, caveats & provenance
This section is not collapsed, and it is not an appendix. Everything above is a model output. Here is what the model does, what it cannot do, and where its numbers come from.
How the season number is produced
A discrete-time daily survival model. For each day after today the model computes a termination hazard from recent rainfall, the Keetch–Byram Drought Index, the ONI index and the progress of the northeast monsoon, then Monte-Carlos thousands of trajectories. The headline is the median of the simulated end dates; the interval is the 10th to 90th percentile.
How the exhaustion year is produced
A stock-and-flow model over five fuel pools, with ENSO evolving as a three-state Markov chain. "Exhausted" means burnable stock has fallen below 5% of its 2025 value. It is arithmetic on assumptions, carried out to a horizon far beyond anything anyone can forecast. It is here to show a direction, not a date.
How the picture at the top is drawn
The haze over the answer is not decoration. Its density is set from the season model's Keetch–Byram Drought Index and its probability that the season is still burning two weeks from now. A wet season leaves the sky clear; a bad one hides the far treeline and turns the sun into an orange disc you can look at directly, which is what the real thing does. Where the browser can run it, a 3D model of the island's smoke fades in over that sky once the answer has loaded; it never replaces the answer.
How the odds are counted
Each percentage in the odds section is the share of the season model's simulated trajectories in which the event happened. There is no market behind it, nobody stakes anything, and the numbers move only when the model's inputs do.
Who built it
This site is built and run by ALPHV Technologies, an AI-native systems integrator in Kuala Lumpur. The model, its caveats and its error rate are published here in full either way.
What this is not
It is not an air-quality service. If you need to know whether it is safe to go outside today, use your national agency's measured readings, not a model on a website with a joke for a domain name.
The haze is not a natural disaster with a season attached. It is drained peat, cleared land and a market for both. The worst seasons on record are associated with tens of thousands of excess deaths across Indonesia, Malaysia and Singapore, and with billions of dollars of economic loss. Where published studies disagree on those figures, the page shows the disagreement rather than the most dramatic number. They are not a punchline.
Caveats, as returned by each endpoint
Data provenance
| Dataset | Source as reported | Status |
|---|