Optimization of the Gaussian Dispersion Model Inversion for Estimating Facility Scale Methane Emissions in Canada

Abstract. This study presents improvements to the implementation of Gaussian dispersion model for estimating methane (CH₄) emissions using mobile real-time measurements at facility scale. The Gaussian plume models often rely on discretized atmospheric states, characterized by stability classes. The proposed enhancements include blending (weighted averaging) the two nearest atmospheric stability classes to create a more continuous model of the atmosphere, and then systematically adjusting the source location along the transect to improve the fit. Stability classes (and the weights assigned) were derived by comparing surface roughness and Obukhov length. The source translation was performed by examining the shapes of the observed and modeled plume. The methods were tested on a widely used, open-source Gaussian dispersion model from Polyphemus, with mobile observations and a Bayesian inversion scheme to estimate emissions. Observations included both controlled release studies that were performed in Canada, as well as observation data from Canadian refineries and waste management facilities. A wide range of methane emission rates were examined, with estimating emissions ranging from 1.7 to 14,213 kg/day. Results showed that blending the stability classes results in model performance that are roughly the weighted averages of the two initial classes, and in some instances slightly better. Source translation resulted in increased correlation and decreased root mean square error (RMSE), in many cases significantly so (e.g., from R² of 0.02 to 0.96, seen in a transect from Courtright). The algorithm was also able to locate a previously unknown source. While blending stability classes showed small improvements in some cases, it generally aligned emission estimates with observations. Our novel approach worked across various stability classes, although its sensitivity to surface roughness remains a limitation for certain situations/environments.