Climate & Research / AI news for Malaysia
Google opens a free global AI methane map Malaysians can query
Google has put a global, AI-built methane plume map online under an open licence. Its own documentation says detection degrades under heavy cloud and dark surfaces, and that half the medium confidence finds may be wrong.
In brief
- Google and NASA's Jet Propulsion Laboratory published MAPL-EMIT, a deep learning model that finds methane plumes in EMIT satellite measurements, plus a global plume database anyone can open.[1][2]
- Google says the model detects 50 percent more plumes than human experts, adds more than 23,000 plumes globally and maps plumes at 24 of the world's 25 largest-emitting landfills.[1]
- The Earth Engine dataset is licensed CC-BY 4.0 and runs from 10 August 2022 to 9 June 2026, but Google records a 50 to 55 percent false positive rate for medium confidence plumes.[3]
Google and NASA JPL put a global AI methane plume map in the open
Google and NASA's Jet Propulsion Laboratory have released a deep learning model that reads satellite light measurements and marks where methane is leaking. Google described MAPL-EMIT on its main blog on 9 September 2026, a week after its research team published the method.[1][2]
The output is the part that matters outside a research lab. The global plume database sits in the Earth Engine data catalogue under a CC-BY 4.0 licence, the trained model is published on Kaggle and the inference code is on GitHub.[1][3]
That makes it reachable from Malaysia. It does not make it reliable here, and Google's own documentation says why.[3]
What the model does that an analyst could not
MAPL-EMIT stands for Methane Analysis and Plume Localization with EMIT. Google Research describes it as a vision transformer that reads the full radiance spectrum of every pixel together with the pixels around it, instead of relying on manual analyst review or matched-filter techniques that miss smaller emissions.[2][3]
It was trained on 3.6 million physics-simulated methane plumes injected into real EMIT scenes. Google says the result captures 84 percent of expert-annotated plumes and surfaces around 50 percent more plausible plumes across roughly 1,100 EMIT granules. Run across the full catalogue, that adds more than 23,000 plumes worldwide.[2][1]
The instrument was never built to watch a rainforest
EMIT is not a methane satellite. NASA launched it to the International Space Station on 14 July 2022 to map ten surface minerals, and NASA's own descriptions put its target area in arid regions between 50 degrees south and 50 degrees north, over dry and sparsely vegetated ground. Methane detection is a by-product.[4][7]
That heritage sets the limits. The instrument sees an 80 kilometre swath at 60 metre resolution from a station that passes where it passes. Google states plainly that coverage is constrained by EMIT's orbital sampling and field of view, so a location appears in the dataset only when the instrument actually measured it.[2][3]
Google publishes the failure modes as clearly as the wins
The Earth Engine entry carries a limitations section most product pages would bury. Plumes seen three or more times across separate EMIT passes are tagged high confidence, with a reviewed false positive rate of about 3 to 5 percent. Plumes seen once are tagged medium confidence, and there the reviewed false positive rate is about 50 to 55 percent.[3]
Two further constraints matter for anyone acting on a detection. Roughly 16 percent of expert-annotated EMIT plumes are missed altogether, and Google warns that performance degrades under heavy cloud cover, shadows or low-albedo dark surfaces, because those weaken the radiance signal the model reads.[3]
Why Malaysia should care
Malaysia is named nowhere in this release. What it gives a Malaysian regulator, council or university is a free, openly licensed way to check a site, plus Google's own warning that cloudy, dark-canopy ground is the hardest case.
Environmental regulators and local councils
A free, openly licensed plume layer now exists, but a single detection over a Malaysian landfill is a lead rather than proof.[3][1]
Practical move: Check whether EMIT observed the site at all, then work only from high confidence plumes with repeat observations.
What Malaysians can do now
- Treat a medium confidence plume as a candidate, not a finding. Google's own review puts its false positive rate near 50 to 55 percent.
- Check EMIT coverage for your coordinates first. The dataset contains only the places and times the instrument actually measured.
- Do not read the pixel values as emission rates. They are column enhancements in parts per million-metre and need wind data to become kilograms per hour.
What we still do not know
An open dataset is not a Malaysian measurement
- How often, if at all, EMIT has observed specific Malaysian landfills, oil and gas facilities or plantation areas, since no per-country coverage figure is published.
- How much the heavy cloud, shadow and dark-surface degradation Google describes reduces detection over Malaysia in practice.
- Whether any Malaysian agency, university or operator is using the dataset, because none is named.
Sources
- 1.A new deep learning model maps global methane emissions from space. Google, 9 September 2026
- 2.Mapping global methane emissions from space with deep learning Google Research, 1 September 2026
- 3.MAPL-EMIT: Modeled Methane Plumes Google Earth Engine Data Catalog
- 4.EMIT (Earth Surface Mineral Dust Source Investigation) NASA Jet Propulsion Laboratory
- 5.iss067e191368 NASA, 24 July 2022
- 6.EMIT Put to the Test at JPL NASA/JPL-Caltech, 1 June 2022
- 7.EMIT's Components Come Together at JPL NASA/JPL-Caltech, 2 May 2022
