Google and NASA JPL release AI model to map global methane plumes

Google Research and NASA Jet Propulsion Laboratory (JPL) have released a deep learning model designed to map and quantify global methane plumes from space. Utilizing raw spectral data from the EMIT instrument on the International Space Station, the vision transformer model operates at a resolution of 60 metres per pixel.

The new AI model identifies approximately 50% more plumes than human analysts and has detected 23,000 additional plumes compared to existing products. Notably, the model identified plumes at 24 of the world’s 25 largest-emitting landfills, whereas the previous NASA product identified 17.

This initiative is part of Google Earth AI, a collection of geospatial models aimed at converting planetary data into actionable intelligence. While existing tools like the Copernicus atmosphere service provide broad regional coverage, this high-resolution approach allows for the identification of specific point sources, such as individual landfill cells, which are often missed by lower-resolution satellite monitoring.