On the Improvement of the Indirect Quantification of Methane Emissions: A Stationary Single Sensor Approach

Abstract

There is a need for improved indirect quantification of methane emissions, particularly from the natural gas industry. Natural gas production continues to increase along with the number of wells and production sites. These production sites have the potential to contribute significant levels of methane emissions. Methane is a potent greenhouse gas and the primary component of natural gas. A more complete understanding of emissions through improved quantification will lead to enhanced mitigation. Emissions have historically been quantified with both direct and indirect methods, with varying results. Emissions have been shown to be both temporally and spatially variable. These facts have led government agencies to fund programs that are focused on improving the quantification methods used. The National Science Foundation (NSF) recently awarded a grant to West Virginia University (WVU) that focused on the “inter-comparison and advancement” of direct and indirect methane emissions measurements. The primary objectives were to directly quantify emissions, develop and deploy a data acquisition system to record variables required for indirect measurements, and employ machine learning to combine current methods with the goal of reducing indirect quantification uncertainty. To accomplish these objectives, two previously utilized techniques for quantifying methane fluxes were analyzed. OTM 33A (OTM) and eddy covariance (EC) are two unique measurement techniques typically used for different applications but requiring similar instrumentation. However, both techniques have high uncertainties (30-70%) associated with their estimates. Methods were used to estimate minimal attainable uncertainty, which was determined to be ±17% for OTM and ±42% for EC. This lends credence to the hypothesis that indirect measurements from stationary single sensors will never obtain the accuracy of direct quantification estimates (~10%). Here, methods of point source estimation were enhanced through modification and combination with machine learning. The OTM method was enhanced by optimizing its governing parameters utilizing a Taguchi design array. This resulted in an average reduction of 22% in the root mean squared error (RMSE) and 30% in the standard deviation of estimates across a series of controlled release experiments. EC footprint functions previously used to determine point source emissions were evaluated but produced results that tended to significantly underestimate emissions. Combining traditional method results with machine learning could improve indirect quantification. To test this hypothesis, a mobile eddy covariance tower (MECT) was developed with the ability to measure variables necessary for both techniques. The MECT was deployed to measure a series of controlled releases for testing and verification. Using