Decoding methane concentration in Alberta oil sands: A machine learning exploration

作者：Liubov Sysoeva, Ilhem Bouderbala, Miles H. Kent, Esha Saha, B. A. Zambrano-Luna, Russell Milne, Hao Wang · 发表于：Ecological Indicators · 年份：2024 · DOI：10.1016/j.ecolind.2024.112835 · 被引用次数：11 · 研究领域：Hydrocarbon exploration and reservoir analysis、Atmospheric and Environmental Gas Dynamics、Hydraulic Fracturing and Reservoir Analysis

Most activities associated with Alberta’s oil sands industry are widely recognized as a serious threat to the environment, particularly the emission of greenhouse gases; the industrial residue that accumulates in oil sands tailings ponds (OSTPs) has the potential to emit large quantities of methane. Mathematical modeling of these emissions, and hence deducing where and why high methane concentrations can be found, is often infeasible due to complex interactions between different sources of methane and lack of availability of appropriate data. Additionally, stationing advanced monitoring devices either inside or in the vicinity of methane emitting sources can be expensive, and may require permits that are hard to obtain. Interpretable machine learning techniques, coupled with existing data from weather monitoring stations, offer a cost-effective alternative approach for modeling and understanding methane emissions sources. We introduce a multi-step framework for finding the primary factors associated with higher methane concentrations, powered by machine learning models (such as random forest) trained on high dimensional datasets sourced from multiple weather monitoring stations located in the Lower Athabasca region. The proposed framework can predict methane concentration levels, illustrate the dependence between the important features and their impact on these levels, and (via the incorporation of wind data) uncover locations of methane sources. We use it to locate such sour...