A Bayesian technique for quantifying methane emissions using vehicle-mounted sensors with a Gaussian plume model

Understanding the uncertainties associated with methane emission estimates is crucial for prioritizing leak repair interventions, enforcing environmental regulations, and modeling climate change. This paper presents a model-based Bayesian approach for describing the uncertainties associated with methane emissions estimates derived from vehicle-based concentration measurements, combined with the Gaussian plume dispersion model (GPM) and anemometry data. The approach begins by deriving a probability density function (pdf) that defines the likelihood of measuring a given release rate conditional on the true release rate. The width of the likelihood pdf is dominated by the GPM model error, which is explored using computational fluid dynamics simulations. The likelihood pdf is combined with a prior pdf that encodes what is known about the emission before the measurement to yield the posterior pdf, which comprehensively defines what is known about the release rate based on measurements and prior information. The technique is assessed by comparing releases inferred from single plume transects with ground truth emission rates, and it is found that the 90% creditability interval contains the true release rate approximately 90% of the time. The Bayesian approach can also be used to optimize measurement paths and/or consider the limitations of these technologies with respect to atmospheric conditions. • Methane emission estimates must be considered in the context of uncertainty. • Uncertainty estimates may be derived empirically or through a model-based approach. • Bayesian inference can be used to incorporate prior information and model error. • Model error for the Gaussian plume model is derived from CFD. • Technique demonstrated on controlled releases measured with a truck-mounted TDLAS sensor.