Advancing Automated Methane Peak Detection for Mobile Surveys: Accuracy, Robustness and Implications for Scalable Deployment

Methane emission measurements are crucial in emerging reporting frameworks such as UNEP’s Oil and Gas Methane Partnership (OGMP) 2.0 Standards and the European Union Methane Regulation. Whilst aerial platforms increasingly provide site-level quantification for upstream operations, advanced mobile leak detection (AMLD) remains the dominant methodology for municipal natural gas distribution networks. A growing number of service providers commercialize this methodology, but open-source academic models remain essential to promote transparency and harmonize quantification across regions. Colorado State University introduced an algorithm that correlates leak rates with the methane mole fraction peak maxima measured when driving downwind methane plumes; Utrecht University improved this method by focusing on the peak-integrated area to reduce instrument-specific bias. However, the area quantification is sensitive to the errors in the detection of the peak bases and currently requires substantial human-based (HB) quality control; thus, limiting scalability of this algorithm and opening up to bias introduction by the individual operator’s HB actions.

This study refines the original algorithm by revising detection logic to reduce the need for HB intervention. Unlike the previous single-step approach, the revised version leverages the benefits of signal smoothing to improve peak detection while mitigating the delays introduced by the high-frequency component filtering. Performances have been evaluated on two replication datasets from the original study (November 2022 and June 2024), observing recall ranging from 93.0% - 95.7%, enabling a clear one-to-one matching of algorithm-detected and HB-validated peaks. For 83.6% of the peaks, the algorithm-integrated area was within 20% from the HB-validated counterpart, with precision losses being attributed to the faulted detection of the peak bases at small peaks close to the validation threshold of the method. Finally, the revised algorithm is used on public AMLD data collected in several municipalities across Europe to benchmark similarities and differences across regions and assess usability potential and challenges of integrating AMLD data to support robust methane emission reporting within city networks.

Our findings suggest that the revised algorithm can evolve into a practical proxy for HB area quantification, reducing HB effort by focusing only on peaks characterized by target features such as anomalous duration. This would preserve the overall transparency and reproducibility of the algorithm across different data sources, enabling scalability and benchmarking across different operators and regions, and promote harmonization in city network methane emission reporting initiatives.