Leveraging AI to Augment Methane Leakage Detection

Securing quick wins renders the onerous road of restraining the adverse effects of climate change more manageable. One such win could be achieved by targeting methane, a greenhouse gas with 80 times the global warming potential (GWP) of CO2 over the 20-year timescale. In its recent Global Methane Assessment, the UN has identified abatement of anthropogenic methane emissions, primarily those resulting from extraction, processing, and distribution of hydrocarbons, as the quickest means to slow the accelerating rate of near-term warming. 

The methane crackdown gained a renewed impetus amid the energy security concerns following the invasion of Ukraine. According to the International Energy Agency, concerted efforts on tackling flaring, venting, and fugitive emissions could materialize in more than 45 bcm of annual gas savings in countries that currently export to the EU – nearly one-third of Russian imports. Many of the proposed measures could be implemented relatively quickly and in certain cases at net negative cost, given today’s elevated prices. However, methane emissions present a challenge from the standpoint of monitoring and hence are seldom considered in inventories. To improve accuracy of detection and source attribution, some companies are revolutionizing their approach by harnessing the power of AI. 

Spotting a Needle in a Haystack

Many emerging detection technologies are being deployed, with a whole spectrum of temporal, spatial, and spectral characteristics. Each of these has a unique set of attributes, and a key challenge in the space is to navigate the heterogeneity of data and identify the combination that would provide the most comprehensive picture of an area of interest. For instance, satellites could provide information about super-emitters but would fail to capture details of a local distribution system. In contrast, with ground sensors, leaks could be identified in real time with a high degree of accuracy, however, detection is restrained to a specific swath of land, which might prove to be cost-ineffective. Growing synergy between such data sources with different spatio-temporal resolutions could be utilized to develop AI models. 

Machine learning (ML) techniques could be coupled with geostatistical tools to perform analytics on satellite data such as spatial interpolation and hotspot detection at locations where high quality measurements cannot be obtained. This is often the case when the view is obstructed by aerosols and cloud cover or limited due to reflectance from the ground surface. Subsequently, multi-modal datasets comprising mobile and fixed sensor measurements can be incorporated to enhance analytics for emissions cluster detection.

Another feat that has earned ML recognition in the space is real-time anomaly detection. Temporary anomalies recorded by the fixed grid of ground sensors can be identified by models adopting an unsupervised learning approach on accrued time series data. Once a leak has been detected, the system alerts operators on its severity and reports the incident to a digital dashboard. 

ML methods could be deployed with optical gas imaging to quantify methane emissions from infrared images and video streams of methane plumes. Deep learning architectures, namely Convolutional Neural Networks and Recurrent Neural Networks, bring capabilities of automatic leak size classification, thereby substantially reducing the cost of human intervention and improving the quality of mitigation.

Future Outlook

A recent policy paper by the UK government on Atmospheric Implications of Increased Hydrogen Use adds to a growing body of research that highlights the indirect global warming effects of hydrogen leakage. There is a conjecture to a later call regarding its GWP value (20-44 on a 20-year timescale), however, the looming prospect of exacerbating the very things hydrogen is intended to ameliorate underscores the need for prudent strategic planning. Careful deliberations are required around ensuring the compatibility of cutting-edge methane leakage detection technologies with future proposals for hydrogen infrastructure.

Guidehouse is well-positioned to assist with conversion of broad methane emissions reduction targets into clear vision for mitigation projects and commercial frameworks. Guidehouse is the key delivery partner for Cadent, the largest gas distribution company in the UK, in partnership with NGGT, SGN, NGN, and WWU, on a Strategic Innovation Fund project investigating the potential of using cutting-edge technology to provide real-time and dynamic assessment of gas leakage. As part of the Digital Platform for Leakage Analytics design, Guidehouse leverages its digital expertise to develop data-driven strategies enhanced by ML techniques.