Source Attribution and Emissions Quantification for Methane Leak Detection: A Non-linear Bayesian Regression Approach.

Methane leak detection and remediation efforts are critical for combating climate change due to methane's role as a potent greenhouse gas. In this work, we consider the problem of source attribution and leak quantification: given a set of methane ground sensor readings, our goal is to determine the sources of the leaks and quantify their size in order to enable prompt remediation efforts and to assess the environmental impact of such emissions. Previous works considering a Bayesian inversion framework have focused on the over-determined (more sensors than sources) regime and a linear dependence of methane concentration on the leak rates. In this paper, we focus on the opposite, industryrelevant regime of few sources per sensor (under-determined regime) and consider a non-linear dependence on the leak rates. We find the model to be robust in determining the location of the major emission sources, and their leak rate quantification, especially when the signal strength from the source at a sensor location is high.