Assessment and mitigation of errors associated with a large-scale field investigation of methane emissions from the Marcellus Shale

Abstract:

Recent work suggests the distribution of methane emissions from fracking operations is a skewed distributed with a small percentage of emitters contributing a large proportion of the total emissions. In order to provide a statistically robust distributions of emitters and determine the presence of super-emitters, errors in current techniques need to be constrained and mitigated. The Marcellus shale, the most productive natural gas shale field in the United States, has received less intense focus for well-level emissions and is here investigated to provide the distribution of methane emissions.

In July of 2015 approximately 250 unique well pads were sampled using the Princeton Atmospheric Chemistry Mobile Acquisition Node (PAC-MAN). This mobile lab includes a Garmin GPS unit, Vaisala weather station (WTX520), LICOR 7700 CH₄ open path sensor and LICOR 7500 CO₂/H₂O open path sensor. Sampling sites were preselected based on wind direction, sampling distance and elevation grade. All sites were sampled during low boundary layer conditions (600-1000 and 1800-2200 local time). The majority of sites were sampled 1-3 times while selected test sites were sampled multiple times or resampled several times during the day. For selected sites a sampling tower was constructed consisting of a Metek uSonic-3 Class A sonic anemometer, and an additional LICOR 7700 and 7500. Data were recorded for at least one hour at these sites.

A robust study and inter-comparison of different methodologies will be presented. The Gaussian plume model will be used to calculate fluxes for all sites and compare results from test sites with multiple passes. Tower data is used to provide constraints on the Gaussian plume model. Additionally, Large Eddy Simulation (LES) modeling will be used to calculate emissions from the tower sites. Alternative techniques will also be discussed. Results from these techniques will be compared to identify best practices and provide robust error estimates.