source: Google Research: Mapping global methane emissions from space with deep learning
level: technical
google research and nasa jpl released mapl-emit, a deep learning model that finds methane plumes in hyperspectral images from the emit instrument on the international space station. the model uses a vision transformer to process full spectral and spatial context, detecting plumes, measuring methane amounts, and locating emission sources. it was trained on 3.6 million synthetic plumes injected into real emit scenes because labeled real plumes are scarce.
on a benchmark of expert-annotated plumes from nasa's l2b dataset, mapl-emit recalls 84 percent of known plumes and finds about 50 percent more plausible plumes across roughly 1100 emit granules. it detected plumes at 24 of the world's 25 top-emitting landfills. false positives remain a challenge in complex terrain, so each detection includes a confidence score based on spectral fit and repeated observations, letting users filter results.
the team released the trained model, synthetic plume dataset, inference library, and a global plume database on google earth engine. methane has 30 times the warming potential of carbon dioxide over 100 years and causes about 25 percent of human-induced warming. cutting methane emissions is a fast way to slow near-term warming, and this tool helps regulators and companies find leaks from oil and gas, agriculture, and waste sites.
why it matters: automated methane detection from satellite data can speed up leak repairs and emissions reporting, directly supporting climate targets and reducing a major source of near-term warming.
source: Google Research: Mapping global methane emissions from space with deep learning