Tai-Long He: Detection of methane emissions from super emitters using probabilistic deep learning

Methane leakages during oil and gas (O&G) production and transmission represent a major source of atmospheric methane. Monitoring these emissions is critical for mitigating greenhouse gas emissions and curbing climate change.  Building on previous work using land surface imaging satellites, we developed a deep-learning based approach to automate detection of large methane point sources using images from Landsat (5-9) and Sentinel-2 (A/B). The deep learning model was trained using human-labelled methane plumes reported by previous work, a notable distinction from previous work using land surface imagers. We applied the methane plume detection system to analyze nearly three decades of images from Landsat 5. Our algorithm has detected hundreds of large historical methane plumes between 1984 and 2011 over major O&G production basins in the former Soviet Union, including a number of plumes in previously unidentified locations. This forms the basis for our ongoing work estimating emission rates of the detected methane point sources and assessing the role of the collapse of the former Soviet Union on variations in the methane growth rate. 

Bio

Tai-Long He is a postdoctoral researcher at the University of Washington. His research focuses on the application of machine learning and data assimilation in the carbon cycle and air quality.