New AI Model Increases Methane Plume Detection by 50%

The research-stage model uses satellite data to identify greenhouse gas emissions more effectively than manual review.

Updated on Oct. 26, 2026 in Artificial Intelligence

Isometric editorial illustration of an industrial gas plume rising from a site, representing automated atmospheric emission monitoring. — Google and NASA researchers developed the MAPL-EMIT AI model, which improved methane plume detection by 50% using satellite imagery analysis. AI Illustration. Upload story photo >

Live Poll

Should government agencies rely more on AI to monitor and regulate industrial methane leaks?

Google and NASA researchers have published a study detailing MAPL-EMIT, an AI model designed to detect methane plumes in satellite data from the International Space Station. The research-stage system identified 50% more plumes than human analysts in recent tests.

Why it matters

Methane traps heat 80 times more effectively than carbon dioxide over a 20-year period, making accurate detection critical for climate mitigation. This model automates the identification process, overcoming the limitations of manual labeling which previously bottlenecked satellite data analysis.

The system analyzed 1,084 observation scenes at a 60-meter spatial resolution, achieving a false-positive rate of 3% to 5% for high-confidence detections. The model was trained on 235,000 real images augmented with 3.6 million synthetic plumes.

The players

Google

A multinational technology conglomerate focused on search, cloud computing, and advanced artificial intelligence research.

NASA

The United States government agency responsible for the civilian space program and aeronautics research, operating the EMIT instrument on the International Space Station.

The details

MAPL-EMIT utilizes a Swin-v2 vision transformer, a type of neural network architecture that processes image data by capturing both fine-grained spectral information and the relationships between neighboring pixels. Because manual annotation of satellite imagery is labor-intensive, the researchers used a Lagrangian puff model—a simulation technique that calculates the trajectory of atmospheric pollutants—to generate millions of synthetic plumes for training. The model uses this data to infer spatial context, allowing it to identify plumes that often escape human detection in isolated image tiles.

Sept 1, 2026: The study was published in the Proceedings of the National Academy of Sciences.

The Tech Race

Automated methane detection represents a shift from manual expert review toward machine learning-based monitoring of atmospheric data. This development places the model as a core processing layer for the NASA EMIT mission, moving beyond static imagery toward active, scalable emission reporting.

This research provides a mechanism to improve the accuracy of atmospheric tracking, though the system remains in the research phase. Policymakers and environmental scientists are the primary users likely to integrate this model into existing emission monitoring workflows.

The takeaway

The research highlights how synthetic data generation can overcome training data bottlenecks for complex environmental modeling. Interested readers should watch for future implementations of this model in public climate reporting datasets to confirm the gains in detection reliability.