On September 9, 2026, Google and NASA's Jet Propulsion Laboratory published a study in the Proceedings of the National Academy of Sciences. The paper introduces MAPL-EMIT, a deep-learning model that detects and quantifies methane plumes worldwide using NASA's EMIT instrument on the International Space Station.

The model was trained on 3.6 million physics-simulated methane plumes. It identifies 50% more plumes than human experts. It adds more than 23,000 previously undetected plumes to the global record.

MAPL-EMIT achieves 84% recall on expert-annotated plumes. It also delivers a higher signal-to-noise ratio than traditional matched-filter methods, according to the Google Research blog. The system detects 24 of the world's 25 largest-emitting landfills.

Visualizations show plumes across sites in California, Turkmenistan, Delhi, São Paulo, Katowice, and Shanxi. A color-coded scale measures plume enhancement in parts per million-meter (ppm-m).

Confirmed

  • Model name: Methane Analysis and Plume Localization with EMIT (MAPL-EMIT)
  • Training data: 3.6 million physics-simulated methane plumes
  • Detection performance: 84% recall on expert-annotated plumes; 50% more plumes detected than human experts
  • Global plume count: 23,000+ additional plumes identified; 24 of 25 largest-emitting landfills detected
  • Instrument: NASA EMIT hyperspectral imager on the International Space Station (originally designed for mineral mapping)
  • Publication: "Global monitoring of methane point sources using deep learning on hyperspectral radiance measurements from EMIT" in PNAS
  • Open releases: Global plume database on Google Earth Engine; trained model and synthetic plume dataset on Kaggle; inference library on GitHub
  • Authors: Vishal Batchu (Research Engineer) and Michelangelo Conserva (Research Scientist), Google Research

Unknown

  • Precision and false-positive rates for the 23,000+ newly detected plumes (the study reports recall but not precision on the global set)
  • Quantitative flux estimates (tonnes/year) for individual sources — the model predicts enhancement in ppm-m; conversion to emission rates requires additional atmospheric modeling
  • Operational latency: time from EMIT overpass to plume availability in Earth Engine
  • Independent replication of the 84% recall figure on held-out expert annotations outside the Google/NASA team
  • Coverage gaps: EMIT's orbit and swath width leave temporal and spatial sampling gaps compared to daily global mappers like TROPOMI

Our take

MAPL-EMIT shifts methane monitoring from expert-intensive analysis to a scalable, automated pipeline — but the 23,000 new plumes are model inferences, not ground-truthed measurements. The real test is whether operators and regulators can act on these detections without independent verification.

Sources