Google Research and NASA’s Jet Propulsion Laboratory have developed MAPL-EMIT, a deep-learning model capable of identifying methane emissions from space at substantially greater scale than conventional manual analysis.
The system analyzes observations collected by NASA’s EMIT instrument and was trained using 3.6 million physics-simulated methane plumes.
According to Google, MAPL-EMIT detects 50% more methane plumes than human experts and identified more than 23,000 additional plumes globally.
The system also detected methane emissions associated with 24 of the world’s 25 highest-emitting landfills.
Methane is a particularly important target for climate monitoring because its warming potential over a 100-year period is approximately 30 times greater than carbon dioxide.
Detecting leaks and other major emissions quickly can allow operators and governments to prioritize mitigation efforts at the facilities responsible for the largest releases.
Satellite-based methane monitoring can be difficult because emissions must be distinguished from complex surface conditions and other noise across an enormous geographic area.
MAPL-EMIT is designed to automate much more of that analysis, allowing scientists to search satellite observations at greater scale.
Google has released the global methane plume database through Earth Engine alongside an application for visualizing the data.
The company has also made open-source models available through Kaggle and released inference tools through GitHub, allowing researchers, policymakers and industrial operators to build on the work.
