Identifying and quantifying source contributions of air quality contaminants during unconventional shale gas extraction

Abstract. The United States experienced a sharp increase in unconventional natural gas (UNG) development due to the technological development of hydraulic fracturing ("fracking"). The objective of this study is to investigate the effect of unconventional natural gas development activities on local air quality as observed at an active Marcellus Shale well pad at the Marcellus Shale Energy and Environment Laboratory (MSEEL) in Morgantown, Western Virginia, USA. Using an ambient air monitoring laboratory, continuous sampling started in September 2015 during horizontal drilling and ended in February 2016 when wells were in production. High resolution data were collected for the following air quality contaminants: volatile organic compounds (VOCs), ozone (O3), methane (CH4), nitrogen oxides (NO and NO2), carbon dioxide, (CO2), as well as typical meteorological parameters (wind speed/direction, temperature, relative humidity, and barometric pressure). Positive Matrix Factorization (PMF), a multivariate factor analysis tool, was used to identify possible sources of these pollutants (factor profiles) and determine the contribution of those sources to the air quality at the site. The results of the PMF analysis for well pad development phases indicate that there are three potential factor profiles impacting air quality at the site: natural gas, regional transport/photochemistry, and engine emissions. There is a significant contribution of pollutants during horizontal drilling stage to natural gas factor. The model outcomes show that there is an increasing contribution to engine emission factor over different well pad drilling through production phases. Moreover, model results suggest that the major contributions to the regional transport/photochemistry factor occurred during horizontal drilling and drillout stages.


BS-DISP % cases with swaps 0 0 3 The base run was automatically selected by the program based on the lowest Qrobust.Since finding a rotationally unique solution is rare, it is acceptable to observe an increasing Qvalue due to the Fpeak rotation with a less than 5% change in Q (dQ).For both time period results, Q values did not vary significantly with Fpeak values of -1.0, -0.5, 0.5, and 1.0); therefore, we can consider all four model results for evaluation.Factor profiles and contributions were examined to determine the impact of the rotation by comparing to the base run results.As a result, for baseline conditions some optimization is gained using an Fpeak of 1.0.

S9
There is a small deviation in species for the three factors.Furthermore, Fpeak-rotated factor fingerprints were compared with the base model outputs.The optimized distribution of pollutants in the three factors provides more interpretable source profiles with respect to marker species.
Also, there is small improvement with the source profiles for the well pad drilling through production phases.
Figure S9.Error estimation summary plot of range of concentration by pollutants in each factor, active phase.Factor profiles for natural gas, regional transport/photochemistry, and engine emissions factors.

Methods/rationale for uncertainty calculations
All parameters have instrument error (2 x detection limit) as the base uncertainty.
Uncertainty is added based on the number of measurements included in the average (as standard error).For the TEOM and VOCs data, where only two measurements are used and the standard error becomes "range/2", an additional factor is calculated to account for the time-weighted averages.
In special cases, where there is missing data before or after a given measurement (as a result of instrument malfunction, power failure, etc), an additional uncertainty is added based on the number of minutes of available data relative to the total number of minutes possible.where x is the number of minutes for the measurement within the target hour (for example: for an original data point at "02:41", x = 41) and a, b, c are constants that were calculated based on the following criteria:

TEOM
-Maximum added uncertainty for measurements at x=30 ("02:30" "03:30" etc.) -No added uncertainty for measurements at x=0/60 ("02:00" "03:00" etc.) Factor related to the number of minutes used in the average: where x is the number of minutes for the measurement within the target hour (for example: for an original data point at "02:41", x = 41) and a,b,c are constants that were calculated based on the following criteria: -Maximum added uncertainty for measurements at x=30 ("02:30" "03:30" etc) -No added uncertainty for measurements at x=0/60 ("02:00" "03:00" etc) Factor related to the number of minutes used in the average:

Figure S3 .
FigureS3.The time series of the total gas production for the four wells (mseel.org).

Figure
Figure S2.Unconventional natural gas production process activity diagram.

Figure S6 .
Figure S6.The PMF factor contribution roses for Engine Emissions factor, Regional Transport/ Photochemistry factor, and Natural Gas factor.

Figure
Figure S8.O3, NOx, CO2, and CH4 concentration time series at well pad development site.

Figure
Figure S8 (continue).O3, NOx, CO2, and CH4 concentration time series at well pad development site.
60 minute sample = no added uncertaintyLargest increase in uncertainty for samples where the adjacent sample is missing (before or after)=((60/minutes used)-1)*(hourly averaged concentration 0.5 ) VOCs IE: Instrument error (2*detection limit) = 0.2 SE: Standard error (for n=2) IE+SE = range/2 multiplied by a factor to account for the averages being weighted based on the number of minutes at each concentration (quadratic equation)=(range/2)*(ax 2 +bx+c) 40-minute sample (sample collection starts and is injected within the same hour) = no added uncertainty 35-minute sample (start and injection in consecutive hours) -slight increase in uncertainty Largest increase in uncertainty for samples where the adjacent sample is missing (before or after): ((40/minutes used)-1)*(hourly averaged concentration 0.5 )

Table S2 .
Average concentrations (ppb) of the most significant volatile organic compounds in different operational phases

Table S3 .
Evaluation of PMF solutions for drilling through production phases.