California's Airborne AI Methane-Leak Detection — From AVIRIS-NG Plumes to Statewide Enforcement Data

The California Air Resources Board, with NASA's Jet Propulsion Laboratory, UC Riverside and Lawrence Berkeley National Laboratory, flew the AVIRIS-NG imaging spectrometer across roughly 22,000 square miles of California in 2016 and 2017, examining more than 272,000 facilities and components statewide.

A trained neural network automatically detected methane plumes in the imagery, and the Geospatial Source Attribution Automated Model (GSAAM), a hierarchical algorithm, then attributed each plume to a specific facility, with peer-reviewed results showing attribution accuracy rising from 51.3% under a simple nearest-distance method to 99.6% under GSAAM.

CARB found that fewer than 0.2% of surveyed infrastructure was responsible for 34-46% of the state's total methane emissions, with landfills alone accounting for 41% of detected point-source emissions, and the resulting Vista-CA dataset identified 15-18% more emitting facilities than existing EPA or CARB inventories; a named leak at the Sunshine Canyon Landfill detected in an October 2016 flight was confirmed remediated on a subsequent 2017 flight.

The programme has since evolved into a satellite-based near-real-time detection pipeline feeding CARB's oil-and-gas and landfill methane regulations, though CARB itself cautions that coverage remains prioritised rather than comprehensive and that formal enforcement mechanisms tied directly to satellite detections were still being developed for landfill rules as of its most recent reporting.