The atmosphere is now being asked to file its own methane audit
the atmosphere is now being asked to file its own methane audit: a deep learning model built with nasa jpl maps global methane plumes from satellite spectra published in pnas, a physics guided neural operator detects emissions onboard hyperspectral satellites with a false positive rate three times lower than the strongest neural baseline, and machine learning now turns tropomi observations into emission rates for global methane super emitters.
the sky now files its own leak list.
Context
The PNAS model is the Methane Analysis and Plume Localization with EMIT model (MAPLE-EMIT is not used as a name in the excerpt read). Google Research's page describes it as a deep-learning framework that automates detection, enhancement quantification and source estimation of methane plumes globally from data of NASA's EMIT instrument on the International Space Station, developed with scientists at NASA's Jet Propulsion Laboratory. The paper is titled Global monitoring of methane point sources using deep learning on hyperspectral radiance measurements and appeared in PNAS in 2026.
The onboard detector is FLAME, a physics-guided neural operator that builds methane absorption physics into its architecture. Its arXiv abstract says it reduces the pixel-level false positive rate by nearly 3 times over the strongest neural baseline. The super-emitter work is ML-SPERE in Atmospheric Measurement Techniques (2026): a convolutional network trained on simulated TROPOMI methane observations and meteorological data to estimate emission rates for super-emitters.
The note says three times lower, and the FLAME abstract says nearly 3 times, so the note rounds up slightly. The comparison is against the strongest neural baseline on the paper's own methane detection benchmark, so it is a benchmark result from the authors, not an in-orbit measurement. The abstract excerpt read here also states FLAME has the highest detection accuracy among the evaluated methods. ML-SPERE is described as outperforming the Integrated Mass Enhancement method in the excerpt read, but the size of that gain was not read. The note's headline that a model maps global plumes from satellite spectra is supported for that model, with the caveat that EMIT covers selected regions as the ISS passes, which the sources read did not quantify.
Related work
- the lake is now being asked to file its own bloom forecast ↗A companion note on satellite AI for environmental monitoring.
Watch next
- Detection limits of the EMIT plume model and coverage numbers. ML-SPERE accuracy figures against the baseline.
Sources
- Global monitoring of methane point sources using deep learning on hyperspectral radiance measurements (PNAS)pnas.org
- Mapping global methane emissions from space with deep learning (Google Research)research.google
- Global monitoring of methane point sources using deep learning (arXiv 2604.10094)arxiv.org
- FLAME: Physics-Guided Neural Operators for Onboard Satellite Methane Detection in Hyperspectral Imagery (arXiv 2606.01577)arxiv.org
- Machine learning-based emission rate estimates of global methane super-emissions (Atmospheric Measurement Techniques)amt.copernicus.org
Provenance
The note above is reproduced unedited from the original post, first published on Threads on 4 October 2026 at 13:10 IST. Sources are the papers and datasets the note draws on.
View the original post ↗Embed this note
More notes
The air is now being asked to keep its own ledger
the air is now being asked to keep its own ledger: ecmwf’s aifs compo becomes the first ai model to forecast atmospheric composition globally every three hours, cleanair simulates 365 days of pm2.5 over china in ten seconds, and a unified framework maps six pollutants at one kilometer across the whole country. the air now files its own composition report.
read the note →The current is now being asked to draw its own map
the current is now being asked to draw its own map: china’s langya 2.0 predicts six ocean phenomena including internal waves and mesoscale eddies, a deep net called wenhai resolves eddies globally with air sea flux formulas built in, and scripps infers surface currents from the way temperature patterns deform in satellite images. the ocean now files its own circulation report.
read the note →The soil is now being asked to report its own carbon
the soil is now being asked to report its own carbon: a nix color sensor paired with generative data augmentation predicts soil organic carbon without a lab, random forest drives 74 percent of soil health mapping studies, and sentinel 2 tracks five year carbon change across france and italy from 922 samples. the dirt now files its own carbon account.
read the note →