Abstract: Methane is a potent greenhouse gas (GHG) that has 28 times stronger global warming potential than carbon dioxide within 100 years after its release. A significant fraction of global methane emissions can be attributed to the oil and gas (O&G) sector, where a large part is released from point sources with exceptionally high emission rates. These "super emitters" were found to be a disproportionately large contributor to total methane emissions in the United States and other major O&G producing nations. Monitoring and addressing methane super emitters could be a feasible near-term solution for reducing GHG emissions and mitigating global warming.
Great efforts have been made in recent years to advance the monitoring of methane point sources using spaceborne instruments. Here we present work using deep learning to automatically detect methane point sources from Landsats and Sentinel-2 satellite imagery. We re-investigated historical trends in atmospheric methane using observations from Landsat-5 over Turkmenistan from 1986 to 2011, one of the largest methane hotspots in the world. Our results suggest increased methane emissions from Turkmenistan's O&G sector following the collapse of the former Soviet Union in 1991, casting doubt on the hypothesis that the Soviet Union's collapse drove the observed methane slowdown. I will also cover our recent work using NOAA's GOES ABI instruments to realize near real-time monitoring of extreme methane releases, with observation intervals from 10 minutes to 7 seconds.