Evaluation, Prediction, and Monitoring of Methane Emission from Oil and Gas Development

Abstract

Human beings are experiencing man-made climate change due to the emission of greenhouse gases, among which methane is a highly potent one, with a 20-year global warming potential 80 times higher than that of carbon dioxide. A systematic approach is needed to evaluate, monitor, and mitigate methane emissions such as those from the oil and gas (O&G) industry (22% of total anthropogenic emissions). In this thesis, I addressed three critical challenges to control such emissions, namely, (i) the uncertainty in a potential O&G methane emission pathway via the groundwater system, (ii) the large population of potential leaking infrastructural elements making routine inspection inefficient and expensive, and (iii) the intermittent emissions that cannot be captured via periodic surveys. To address (i), groundwater samples were collected from more than 300 sites in O&G-producing Northern Appalachia. Dissolved methane concentration was negatively correlated with the distance to O&G well in one of our study regions, but such correlation was confounded by topographic variation. Furthermore, dissolved sulfate concentration was negatively correlated with methane concentration and with distance to coal mine, and these correlations were robust even when considering topographic confounding. In conclusion, groundwater methane could be attributed to natural geological sources and sulfate-mediated biogeochemical processes, rather than O&G development. To investigate (ii), Machine Learning (ML) models were used to predict O&G well integrity issues related to methane leakage to guide prioritized sensor allocation. Different ML models (e.g., Random Forrest, XGBoost, and Logistic Regression) were compared on a dataset consisting of 1,250 O&G wells, and a test F-1 score above 65% was achieved. Furthermore, the most important physical parameters for the prediction were identified, and the geospatial clustering of integrity issues was observed and analyzed. These findings could enable prioritized sensor allocation near O&G facilities with high emission risk and inform better design of future O&G wells. To study (iii), inexpensive continuous methane sensors are needed, but such sensors can suffer from signal interferences. Given such, an ML signal deconvolution strategy was proposed and an experimental apparatus, consisting of mass flow controllers, a gas chamber, and a data logging system, had been built to collect data for ML model training and testing. In addition, preliminary tests were conducted to study the influence of humidity and gas flow rate on the performance of the sensors. Lastly, the apparatus is being upgraded to integrate commercially available methane sensors and temperature control system. Overall, the research of this thesis deepens our understanding of O&G methane emissions and enhances our capability to monitor and mitigate those contributions.