Methane leak or false positive? An automated probabilistic treatment of detected emissions from oil and gas facilities in multi-spectral satellite imagery at continental scale.

Satellite observations are an important tool in large emission event detection and global monitoring of methane emissions from oil and gas facilities, but current satellite-based methods face significant uncertainty. One of the major sources of uncertainty is a high rate of false positives. Current methods to mitigate false positive rates typically involve manual inspection of plume imagery, a time-consuming process which introduces human error. The Sentinel-2 multispectral satellite is widely used in global methane observation, as methane enhancements can be identified by a signal in shortwave infrared bands 11 and 12.  However, physical, biological and other anthropogenic processes can have a similar spectral signature, leading to a high rate of false positives. We present an empirical False Discovery Rate approach for quantifying the false positive probability for a given candidate plume. Imagery data is divided into a near-to-well set (within 200m of oil and gas infrastructure) and a far-from-well control set (between 400m-1200m away from oil and gas infrastructure), which is conservatively assumed to consist entirely of false positives. With these datasets, we define the probability of a false positive given proximity to oil and gas infrastructure as a function of plume quality and distance to infrastructure.  The results from this approach are shown for a case study over the contiguous United States, where we found a strong relationship between the selected plume quality metrics, distance to infrastructure and the false positive probability. We also identified significant differences in plume characteristics between the near-to-well and far-from-well datasets.  This work presents a more efficient and data-driven false positive algorithm, which can significantly reduce the manual step in false positive identification, resulting in larger scale deployment and data processing of satellite-based methane emission monitoring.