U-Plume: Automated Algorithm for Methane Plume Detection and Source Quantification by Satellite Point Source Imagers

Methane is a potent greenhouse gas. Identification and quantification of methane point sources is a crucial step in understanding primary contributors to regional emissions and identifying targets for high impact mitigation. Point source imaging satellites like GHGSat-C1 are powerful tools for detection of point sources globally. Current methods for plume identification and source rate estimation are often error prone and labor intensive. They are not scalable to the quantity of high spatial resolution methane data that is presently becoming available from point source imagers. We have developed a hybridized method, UPlume, combining machine learning with a U-Net architecture for plume detection with physics-based source rate estimation methods. The UPlume method is successful in producing accurate plume masks in GHGSat-C1 imagery and separating out artifacts. Source rate estimations from the UPlume method support the viability of this method for automated detection and quantification over large datasets. The separation of masking and source rate estimation within UPlume allows visualization of the plume mask, identification of the point source, and quality control on the accuracy of image segmentation.