Data assimilation method for quantifying controlled methane releases using a drone and ground-sensors

• Published in: Atmospheric Environment: X Vol. 17; p. 100210
• Main Authors: Darynova, Zhuldyz, Blanco, Benoit, Juery, Catherine, Donnat, Ludovic, Duclaux, Olivier
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.01.2023; Elsevier
• ISSN: 2590-1621; 2590-1621
• DOI: 10.1016/j.aeaoa.2023.100210

This work assesses the particle filter data-assimilation technique for estimating the methane emission rate during CH4 controlled-release experiments conducted over 3–4 days in Fall 2020 and 2021. Several controlled methane releases took place on a 40 m × 50 m platform in France, called TADI (TotalEnergies Anomaly Detection Initiative). The leaks ranged from 0.01 to 5 g CH4 s-1 over 24–71 min. A methane-detecting drone and five ground-sensors recorded the methane concentration simultaneously. The accuracy of the air contaminant dispersion estimations, based on Gaussian model, is improved by applying a data assimilation method using a particle filter. Diffusion coefficients and release rate are considered as state parameters in the data-driven modeling. A particle filter is then applied to update these parameters during each computation step. We assessed various frameworks for assimilating air data in order to monitor CH4 emissions from industrial sites and infrastructures. For most releases, the assimilations consistently give precise rate estimates, whether considering fixed or mobile data and any of the particular assimilation setups. The average relative errors in the estimated CH4 release rates typically range from approximately 35%–84% for the 2020 campaign, and from 29% to 72% for the 2021 campaign. The inversion results using the stationary measurements have an average relative error of about 72%, while the use of drone measurements yields a more accurate emission rate estimate of around 51%. The hybrid approach, which simultaneously evaluated both drone and stationary measurements using a particle filter, achieved the highest coefficient of determination and the lowest relative error between the reported and model estimated flow rates (R2 = 0.97 and 29%, respectively). •Using ground-sensors and drones in inversion simultaneously achieves better results.•Low and high wind conditions have comparable relative error result.•Inversion with particle filtering increases the accuracy of the model estimation.