Domain Editor-in-Chief: Detlev Helmig; Institute of Alpine and Arctic Research, University of Colorado Boulder, Boulder, Colorado, United States
Guest Editor: Michael Griffin; Carnegie Mellon University, United States

1. Introduction

The confluence of several technologies, namely horizontal drilling, slickwater mixtures, and hydraulic fracturing, have led to a rapid increase of domestic oil and gas production from shale rock resources in the United States in the last decade (Energy Information Administration, 2015). Collectively referred to as “fracking”, the industrial exploration of shale resources for oil and gas extraction consist of a series of activities: well pad and drill rig establishment, drilling, well bore preparation, hydraulic fracturing, well completion, and production. One pad site typically serves production from multiple bores. To distinguish the process, we follow a previously used naming convention, calling it Unconventional Oil and Gas (UOG) exploration. Unlike in conventional oil and gas exploration, UOG production occurs at hundreds to thousands of smaller sites throughout a shale area, creating a landscape dotted with well pads (Jones et al., 2015). Produced oil and/or gas is transported from these sites either via pipelines, or temporarily stored onsite until trucked to its destination.

Environmental impacts of UOG exploration can be manifold (Jackson et al., 2014), and concerns have focused on water contamination (Jackson et al., 2013; Vengosh et al., 2014; Alawattegama et al., 2015) and air emissions (Field et al., 2014; Moore et al., 2014), which can come from various steps in the UOG exploration process, such as during the hydraulic fracturing backflow period. Atmospheric emissions are dominated by emissions of methane, the dominant component of natural gas and a strong greenhouse gas (Brandt et al., 2014). While there are an increasing number of field studies addressing methane leakage (Allen et al., 2013; Karion et al., 2013; Miller et al., 2013; Phillips et al., 2013; Caulton et al., 2014b; Karion et al., 2015; Lan et al., 2015; Lavoie et al., 2015; Peischl et al., 2015; Yacovitch et al., 2015; Zimmerle et al., 2015), other field studies have addressed simultaneously emitted higher hydrocarbon emissions as a result of UOG exploration (Petron et al., 2012; Gilman et al., 2013; LaFranchi et al., 2013; Peischl et al., 2013; Swarthout et al., 2013; Edwards et al., 2014; Helmig et al., 2014; Macey et al., 2014; Petron et al., 2014; Thompson et al., 2014; Zavala-Araiza et al., 2014; Smith et al., 2015; Townsend-Small et al., 2015; Yuan et al., 2015). Ethane has been used as an indicator of the UOG exploration methane source (Simpson et al., 2012; Smith et al., 2015; Townsend-Small et al., 2015; Vinciguerra et al., 2015), and the large variety of volatile, non-methane hydrocarbons (NMHCs) associated with oil and gas and their potential impacts on atmospheric ozone formation have been investigated in detail (Kemball-Cook et al., 2010; Carter and Seinfeld, 2012; Edwards et al., 2014; Helmig et al., 2014; Ahmadov et al., 2015; Field et al., 2015; Koss et al., 2015).

A complicating factor in shale areas such as the Eagle Ford in Texas is the amount of gas flaring at oil wells (Schade and Roest, 2015), which can act as an additional source of NMHCs and partially oxidized volatile organic compounds (VOCs), such as formaldehyde and acetaldehyde (Strosher, 2000; Al-Fadhli et al., 2012; Knighton et al., 2012; Torres et al., 2012b; Wood et al., 2012; Pikelnaya et al., 2013). Flaring serves as a replacement for venting, with the intention of converting all methane and associated NMHCs into CO2. While it is mandatory in Texas, the operated flares are diffusion flares (Strosher, 2000), for which less information on emissions exists as compared to controlled industrial flares. Although investigated diffusion flares in the Marcellus shale area have been found to combust methane highly efficiently (Caulton et al., 2014a), no other field measurements on diffusion flare VOC emissions in US shale areas are available (Buzcu-Guven and Harriss, 2012), and even a small percentage of incomplete combustion and NOx formation (Duncan et al., 2016) could affect ozone formation in flare plumes significantly (Al-Fadhli et al., 2012; Olaguer, 2012).

The Eagle Ford shale (EFS) is a geographically large region in south central Texas (Figure 1), whose oil and gas production has been rapidly increasing since 2008 (, when the first UOG wells were completed in the area. Total natural gas production increased from approximately 4000 to 5000 million cubic feet (MMcf) per day between 2013 and 2014, while raw oil and condensate production increased from approximately 966,000 to 1,340,000 barrels per day during the same period. Due to a continuous drop in depth of the shale layer from north northwest to south southeast (Energy Information Administration, 2014), gas production (gas–to-oil ratio, GOR, >6000) is prevalent south of Karnes City, while oil production (GOR<2000) is prevalent to its north, with a “wet gas” window in between. Associated NMHC emissions in the EFS have been interrogated due to their possible influence on ozone formation in south Texas (AACOG Natural Resources Department, 2013, 2015; Pacsi et al., 2015) and various emissions inventory estimates in the form of reports to the Texas Commission on Environmental Quality (TCEQ) exist that are used in air quality modeling (Pring, 2012; AACOG Natural Resources Department, 2014; Lange et al., 2014). Due to increasing concerns about local and regional air quality as affected by the industry’s NMHC emissions (e.g. ICN, 2014), TCEQ installed a permanent air quality monitoring site north of the shale area in the city of Floresville, population 6500, in summer 2013 (Fig. 1). In addition, due to the vast area covered by UOG exploration in the Eagle Ford, the TCEQ commissioned a limited mobile air sampling study to assess whether the Floresville monitor’s measurements downwind of the shale area are representative for a larger area (Sullivan, 2014). The study, carried out in May and June 2014, concluded that, overall, the monitor shows qualitatively representative NHMC abundances as compared to other sampling sites downwind of the shale area. The main difference encountered was for the highly reactive alkene isoprene, which comes nearly exclusively from inhomogeneously distributed isoprene-emitting trees in the region (Wiedinmyer et al., 2001; Yu et al., 2015).

doi: 10.12952/journal.elementa.000096.f001.
Figure 1.  

South Texas County map highlighting site location and shale area.

A south Texas county map of the Eagle Ford shale area (light brown), the monitoring site’s location in Floresville indicated by a red dot. The inset shows the local road network surrounding the site. Major roads are depicted as thick gray lines. Thin black lines mark county borders, and thin grey lines in the inset show residential roads.

Our objective for this study was to provide an initial assessment of the first year of measurements at the newly established Floresville monitoring site, which occurred during the continuously expanding UOG exploration in the EFS area (Figure 1). Karnes County, which lies to the south and southeast of the monitor features some of the highest well densities in the shale area, and has been estimated to have contributed 12.9 metric tons of NMHCs per ozone season day (period between 1 April and 31 October) (AACOG Natural Resources Department, 2014). We thus hypothesized that the NMHCs monitored would stem from a combination of sources including UOG exploration, the emissions of which would accumulate into the boundary layer during transport over the shale area.

2. Methods

We used publicly available monitoring station data from TCEQ’s Floresville Hospital location (Figure 1), EPA Site Number 484931038, Continuous Air Monitoring System (CAMS) station 1038, activated in July 2013. The site is located in Wilson County, Texas, 29.1307 degrees N and 98.1481 degrees W, 122.0 m above sea level. It consists of an air conditioned trailer housing an automated gas chromatograph for hydrocarbon measurements, a NO/NOx analyzer, and standard meteorological instrumentation. Data were obtained in August 2014 through TCEQ but are now also available online at

TCEQ’s Standard Operating Procedure for VOC precursor analysis (TCEQ, 2005) is available through their Field Operations Division. Briefly, ambient hydrocarbon concentrations are measured hourly using an automated sampler connected to a dual column, dual GC-FID system provided by Perkin-Elmer. C2–C10 Hydrocarbons in air are collected through a Nafion™ dryer onto a cooled (-30 °C) adsorbent trap for 40 minutes at 15 mL min-1. The automated thermal desorber then heats the trap to 325 °C and flushes the volatiles directly on-column. A two column system (a 50-m, 0.22 mm ID BP-1 and a 50 m, 0.32 mm ID PLOT) is used for NMHC compound separation. Calibration is based on regular injections of known concentrations of propane (PLOT column) and benzene (BP-1) using the FID’s carbon response characteristics to calibrate all other identified compounds. Quality control includes daily blanks and detector response verification, weekly precision checks and annual calibration curves. The method detection limit is typically less than 0.4 ppbC (i.e., e.g., 0.2 ppb ethane or 0.05 ppb octane). TCEQ assures data quality when system accuracy tests yield results less than 30%, and precision tests less than 20% deviation, both assessed using propane and benzene gas standards.

An important caveat with this widely established sampling and analysis technique are the use of a Nafion™ dryer, and the lack of ozone removal from the air sample upon collection. Nafion dryers can cause positive alkene artifacts, particularly for isobutene (Plass-Dülmer et al., 2002), and all data for this compound were reported “NA” by TCEQ. Potential alkene losses due to ozone reactions during cryogenic or solid sorbent sampling and subsequent thermal desorption are a known bias (Helmig, 1997). While the TCEQ is not aware of significant alkene losses during their sampling and desorption technique, we cannot exclude that significant fractions of alkenes, particularly those highly reactive toward ozone, were lost from the sample before analysis. However, since ethene and propene losses are typically small (Plass-Dülmer et al., 2002), we decided to retain all alkene data, but will present only results for the least reactive alkenes ethene and propene.

Upon receipt of the quality-assured data, we converted all hydrocarbon data into volume mixing ratios (VMRs), given in ppb, by dividing the given unit ppbC by carbon number in the molecule. Data were arranged into a continuous timeline, and combined with the hourly NOx and meteorological averages, replacing missing data or periods of instrument issues or calibration times with “NA”. R language software (R Core Team, 2015) was used for data analysis. Furthermore, NOAA’s online HYSPLIT (Hybrid Single Particle Lagrangian Integrated Trajectory) model (Draxler and Rolph, 2015) was used in selected case study periods for further analysis of air mass origins and/or plume dispersion.

3. Results and discussion

We first analyzed the abundances and the time series of the dominant hydrocarbons, then compared hydrocarbon ratios to published data from other shale areas, and to raw oil and gas composition data as available from non-peer-reviewed sources (section 3.1). We next carried out a factorial analysis to determine the dominant sources contributing to the observed composition at the site, then relate the obtained factors to air mass origin (section 3.2). This information was then used together with calculated atmospheric OH radical reactivities of the observed average composition at the site for different air mass origins and contrasted with past studies in the Houston ship channel, considered one of the most polluted air sheds in Texas (section 3.3). Lastly, we present two case studies of high NMHC concentrations at the monitor (section 3.4).

3.1 VOC abundances and ratios

Boxplots of observed VMRs of selected hydrocarbons and NOx are shown in Figure 2, compared with typical values observed in US cities (Baker et al., 2008). Ethane, a compound associated with oil and gas exploration, showed a median VMR of 9 ppb and high variability, with extreme mixing ratios over 100 ppb. Higher alkanes were strongly correlated with ethane but lower in abundance (Table S1), probably due to both lower emissions and shorter atmospheric lifetimes. BTEX compounds such as benzene and xylenes had significantly lower abundances not exceeding any human health-based comparison values as specified by TCEQ (2015). Figure 2 demonstrates that the data are skewed toward higher values, with all mean values exceeding medians. Skewness is highest for the photochemically short-lived species isoprene and NOx, but also very large for the alkanes and xylenes, while much lower for benzene. As discussed below, this is dominantly due to recurring high VMRs, particularly at night, under wind directions from the EFS in combination with the EFS’s source emission profile. In contrast, the very short-lived biogenic hydrocarbon isoprene, only emitted during daytime by vegetation such as oak trees, was elevated dominantly when ambient temperatures exceeded 20 °C, but did not display a strong preference for wind direction (data not shown). Figure 2 also includes some typical northern hemisphere annual background values (progressed to 2013) for the longer lived hydrocarbons and the xylenes as reported by Helmig et al. (2009). The comparison demonstrates that very clean air masses are occasionally observed at this monitoring site, typically in summer when tropical air masses are rapidly advected to the site from the Gulf of Mexico. On the other hand, the observed median and mean alkane abundances are similar than those observed in Houston’s ship channel in 2006 (Table S2), a region with a high density of petrochemical industries (Gilman et al., 2009). While no long-term history of air quality measurements exists for the EFS, the large scale grab sample measurements in spring 2002 presented by Katzenstein et al. (2003) already showed elevated ethane and n-butane abundances of 2–6 ppb and 0.2–1 ppb, respectively, and were attributed by the authors to then existing oil and gas industry emissions in the region. At that time (2002) Texas crude oil production was 52% and natural gas production was 72% of 2013 values; however, production numbers in Karnes County in 2002 were less than 1% of oil and less than 5% of natural gas compared to its 2013 rates.

doi: 10.12952/journal.elementa.000096.f002.
Figure 2.  

Selected hydrocarbon and NOx VMR boxplots.

Boxplot of annual hydrocarbon and NOx VMRs measured at the Floresville monitor site during July 2013 to July 2014. Horizontal bars are medians, crosses are averages, and open squares represent typical background VMRs (Helmig et al., 2009). For comparison, the range of mean VMRs in 28 cities presented by Baker et al. (2008) is given (brown bars), although 2013/14 VMRs in urban areas are likely lower (Warneke et al., 2012). All compounds showed right-skewed distributions and skewness is indicated across the top of the graph.

The alkanes, which dominate the observed hydrocarbons, showed a typical diurnal cycle of higher nighttime than daytime abundances, depicted for propane in Figure 3. It is indicative of a diurnally consistent source that is diluted into a diurnally varying boundary layer height. The observed median early morning to late afternoon ratio of 3.2 for propane under southerly winds (winds blowing from the south) was approximately 20% higher than the 3-hr composite boundary layer heights ratios in this area as retrieved from the North American Regional Reanalysis data (Mesinger et al., 2006). Since propane has a chemical lifetime of ten days for typical 12-h daytime OH radical abundances of 2×106 molecules cm-3 (Atkinson, 2000), the lower daytime values may indicate small daytime losses due to photochemistry and entrainment of air with lower propane mixing ratios from the free troposphere. Seasonal changes were also observed; higher abundances were generally observed during the colder seasons of the year most likely due to (i) lower boundary layer depths and (ii) longer photochemical lifetimes.

Correlations between the dominant hydrocarbons in the dataset are summarized in Table S1 in the supplementary materials. There are very strong relationships between the alkanes, but weaker correlations between the alkanes and the BTEX species or ethene, all of which are known to be emissions in car traffic exhaust (Kawashima et al., 2006). Alkanes prominent in traffic exhaust and gasoline evaporation are dominated by butanes, pentanes and hexanes (Liu et al., 2014; Pang et al., 2014). Typically observed car traffic dominated urban ratios are approximately 1:1 for isopentane to n-butane, 2.5:1 for isopentane to isobutane, 2:1 for isopentane to n-pentane, 5:1 for isopentane to n-hexane, and 12.5:1 for isopentane to n-heptane (Parrish et al., 1998; Kawashima et al., 2006; Baker et al., 2008; Parrish et al., 2009). The respective ratios at the Floresville site were approximately 1:3, 2:3, 1:1, 5:1, and 15:1, similar to findings in the Denver-Julesburg Basin, an oil and gas development region in Colorado (Gilman et al., 2013; Swarthout et al., 2013), and signifying a strong contribution to abundances by other short-chain alkane emitting sources. The ratios of ethane to propane, 5:4, propane to butanes, 3:2, and butanes to pentanes, 2.5:1, (approximated from Table S1), strongly suggest emissions from oil and gas exploration as that source (Todd, 2011; Pring, 2012; Gilman et al., 2013; Swarthout et al., 2013; Lange et al., 2014).

doi: 10.12952/journal.elementa.000096.f003.
Figure 3:  

Diurnal variation of propane.

Boxplot of the diurnal variation of propane at the Floresville monitor under southerly wind directions (plotting data outside two standard deviations–dashed line error bars–was omitted for visibility reasons). Blue, horizontal dashed bars show the median early morning (4-8 am) and early afternoon (1–5 pm) VMRs.

3.2 Factor analysis

To further elucidate the emission sources contributing to the observed ambient VMRs at the Floresville monitoring station, we carried out a factor analysis in R (R Core Team, 2015). A factor analysis was preferred over positive matrix factorization (McCarthy et al., 2013) because of the rural nature of the monitoring site, removed from many major emission sources. Data were arranged into 32 hydrocarbon species or groups and NOx. Sets of measurements that were missing one or more of ethane, propane, or n-butane data were discarded. Missing data or zero measured abundances were removed and replaced by randomly generated data at approximately one tenth the instrument’s detection limit. The factanal function in R was run with increasing factor numbers from 2 to 8, Bartlett’s least squares regression scores, and varimax rotation. In Table 1, we show the result of the factor analysis with 5 prescribed factors as an example. In each analysis the first two factors remained virtually unchanged, their relative contributions varying between 47% and 50% for factor one, and between 29% and 32% for factor two. Factor one dominates variability, and based on the dominance of alkanes in its loadings composition, was assigned to oil & gas exploration related emissions. Factor two is dominated by alkenes and acetylene, with large contributions from NOx and aromatic compounds, and was thus assigned to combustion emissions (Liu et al., 2014). Although these assignments appear straightforward, we note that neither factor presents a unique loadings distribution that could be assigned to a known source. Since the monitoring site is relatively far downwind of much of the shale area, various oil or gas exploration related emissions have likely blended, thus creating covariance due to co-advection in addition to co-emission. However, since the oil exploration section of the shale is closer than the gas exploration section, emissions from gas flaring at oil wells might be a reason for significant loadings of ethene, propene and acetylene on the oil & gas factor. Similarly, the combustion factor is not necessarily dominated by car traffic exhaust emissions in the area, but may be a combination of various combustion sources including compressor engines or flares at well sites.

Compound Factor 1 Factor 2 Factor 3 Factor 4 Factor 5
NOx * 0.687      
ethane 0.911 0.286 *   *
propane 0.930 0.271 *   *
n-butane 0.951 0.244 *   *
isobutane 0.947 0.226     *
n-pentane 0.958 0.238 *    
isopentane 0.937 0.298      
n-hexane 0.946 0.246      
n-heptane 0.868 0.292   0.353 *
n-octane 0.879 0.309   * 0.250
n-nonane 0.819 0.387     0.205
n-decane 0.741 0.400 *   *
methylhexanes 0.734 0.376 * 0.506  
dimethylbutane 0.844 0.334      
dimethylpentanes 0.615 0.548 * 0.392  
cyclopentane 0.895 0.315 *   *
cyclohexane 0.929 0.252 *   *
methylcyclopentane 0.885 0.355 * * *
methylcyclohexane 0.923 0.277   * *
ethene 0.362 0.849 0.244    
propene 0.611 0.672 0.220    
1-butene * 0.891 *    
1-pentene * 0.824 *    
acetylene 0.276 0.763 0.260    
benzene 0.684 0.565 0.202 *  
toluene 0.545 0.734     *
m/p-xylene 0.543 0.759 -0.239   *
o-xylene 0.448 0.834 -0.204   *
ethylbenzene 0.453 0.777 *   0.233
styrene 0.227 0.590 *    
trimethylbenzenes 0.477 0.754 -0.212   *
isoprene *   -0.204    
butadiene   0.876 *    
sum square loadings 16.261 9.999 0.723 0.649 0.570
prop. variability 0.493 0.303 0.022 0.020 0.017
cum. variability 0.493 0.796 0.818 0.837 0.855
doi: 10.12952/journal.elementa.000096.t001.

Table 1.

Example factor tablea

a Only absolute loadings larger than 0.2 are shown; absolute loadings less than 0.1 were discarded, while those between 0.1 and 0.2 are indicated by an asterisk.

The two leading factors explained 80% of the data set variability, and no additional factor was contributing more than at most another 3%. The compounds with the highest factor one over factor two loadings are butanes, pentanes and cyclohexanes, all highly volatile compounds present in natural gas, condensate, and oil produced in the EFS (Hendler et al., 2009; Pring, 2012; Lange et al., 2014). The factor one weighted ratios were very similar to ratios listed in section 3.1, most likely due to the dominance of factor one and only small loadings differences between these compounds. However, the same ratios for the combustion factor were closer to the typical urban ratios, such as 1.5:1 for the isopentane to pentane ratio, or 2:1 instead of 3:1 for n-butane to isopentane.

The two factors showed markedly different features as a function of wind direction: Figure 4 shows two representative compounds, n-butane for the oil & gas factor and ethene for the combustion factor. High n-butane mixing ratios are associated with transport from the ESE and SW sectors, both of which are rural areas with low traffic density, allowing air masses to spend a long period of time over the shale area (Figure 1) and can thus accumulate higher VMRs from ongoing emissions relative to a trajectory perpendicular to the shale, such as for southerly winds. Ethene mixing ratios are elevated with winds from the NW sector, which is explained by the location of the monitoring station in the SE extent of the town of Floresville, and the fact that the San Antonio metro area lies further to the NW. Emissions from dense car traffic in these areas are probably the dominant sources of the measured hydrocarbons. Note, however, that very few outliers for either n-butane or ethene occurred for winds from the NW quadrant, while numerous outliers occurred for both species from all other quadrants, and especially for southerly wind directions (150–210 degrees). From the perspective of the monitoring site, UOG activities in the EFS occur in the SE and SW quadrant, and partially also the NE quadrant (Figure 1), and this is appears to be reflected in the outliers in addition to the elevated abundances. Ethene, and especially propene, also showed elevated abundances for easterly winds. This, in addition to numerous outliers for the SE and SW quadrants, and significant loadings on factor one, supports the hypothesis that they may also be a UOG activity emission, possibly from the large amount of flaring in the shale area. This is further supported by an additional factor and the finding that most outliers occurred at nighttime when photochemical loss rates are minimal, further discussed below.

doi: 10.12952/journal.elementa.000096.f004.
Figure 4.  

Variation of butane and ethene with wind direction.

Boxplots of n-butane, representative of factor 1, and ethene, representative of factor 2, VMRs as a factor of wind direction (see Figure S1 for a windrose). Elevated VMRs for factor 1 occur with air mass trajectories spending longer periods over the shale area for E-SE and SW to W directions. Factor 2 is also slightly elevated for those directions, but dominantly for NW directions due to emissions from the urban area of Floresville and beyond (some very high VMRs were omitted from the graphs for visibility reasons). All shale directions from E to S to SW show a higher amount of data outside two standard deviations (open circles), dominantly occurring at night.

Although no additional factor contributed significantly to the data set’s variability, additional factors mostly reduced factor two loadings, and two of these factors showed noteworthy loadings compositions. Factor three in Table 1 shows a composition that might be related to the flaring in the shale area. Its scores showed a similar variation with wind direction as the oil & gas factor, maximizing for E and SW wind directions, directions in which very large flaring sources are located (Tedesco and Hiller, 2014). Factor four shows high loadings of heptanes, possibly related to shale oil exploration. The composition of heptane isomers in the data set (∼50% n-heptane, ∼15% dimethyl-pentanes, and ∼35% methyl-cyclohexane, not including iso-heptanes) suggests that the source of the heptanes is an oil-associated gas (Zecheng et al., 2013), likely of sapropelic origin, which aligns with what is known about the Eagle Ford’s geology (Hunt and McNichol, 1984; Robison, 1997). These two factors, albeit contributing very little to the total data set variability and showing varying compound loadings, consistently appeared in all factor analyses with four or more factors. Therefore, we speculate them to be indicative of two minor, but possibly significant aspects of the air quality at this location. No other factor could be related to other potential sources or source variability in the area.

Presuming the two leading factors are sufficient to predict NMHC air quality at the monitoring site, we calculated selected abundances using isobutane and ethene as the predictors in a multilinear model similar to Gilman et al. (2013). Figure 5 shows the mean diurnal cycle and measured versus predicted benzene VMRs for winter 2013/14 (winter was chosen since ethene VMRs did not drop below the detection limit). This analysis allows quantification of individual source contributions, and shows, in this case, that winter benzene levels at the Floresville site could be 11–32% lower (IQR, median: 19%) if there were no UOG activity emissions coming from the Eagle Ford shale area. The respective median values for toluene and xylenes were 28% and 37% lower, respectively, suggesting significant, but not dominant contributions of UOG activity to these aromatic compounds in this rural area.

doi: 10.12952/journal.elementa.000096.f005.
Figure 5.  

Modeled mean benzene diurnal variation.

The mean diurnal cycle of measured (dashed line) and predicted (colored sections) benzene VMRs during December 2013 to February 2014 using a multilinear model based on isobutane and ethene VMRs (cf. Gilman et al., 2013). The inset shows the goodness of the fit using hourly data.

3.3 Hydrocarbon reactivity

The high abundances of hydrocarbons at the Floresville monitor identified above raise concerns about possible impacts on regional ozone and public health within and downwind of the shale area. Thus, we considered the overall OH radical reactivity of monitored NMHCs and NO2 at the site and further contrasted it to a similar data set from Houston, obtained during the Texas Air Quality Study II in July to September 2006 in the Houston/Galveston Bay (HGB) area (Gilman et al., 2009).

The HGB data are comprised of a larger set of VOCs, mostly due to a lack of measurement capability of oxygenated VOCs at the Floresville site. Thus, the comparison is also performed using a subset of the HGB data from which these compounds have been removed. The results are summarized in Table 2, which lists estimated (methane and carbon monoxide) and calculated reactivities for the Floresville and HGB data sets. Note, that two of the largest single contributions missing from the shale area measurements are formaldehyde and acetaldehyde, which are expected to be (i) formed from local and regional hydrocarbon photochemistry, and (ii) directly emitted from flaring in the shale area. Without the inclusion of these compounds, median OH reactivity at the Floresville site during 2013/14 was similar to that observed at HGB in 2006. Meaning, most of Houston’s higher OH reactivity was attributable to oxygenated VOCs and to higher CO and NOx concentrations, as would be expected for a large metropolitan area and its associated diverse emissions from stationary and mobile sources. Furthermore, since Houston’s NMHC abundances have stood out in the past (Kleinman et al., 2002), extending this comparison to other cities illustrates that OH reactivities at the Floresville monitor resemble those in much larger cities (Baker et al., 2008).

Reactivity contribution All median Shale medianc Shale summerd Shale percent HGB medianb HGB percent
CH4e 0.315 0.315 0.315 10 0.27 6
COe 0.40 0.41 0.33 11 0.62 14.5
sumVOCs 1.23 2.01 1.82 57 2.1 (1.25)f 49 (29)f
NO2 0.62 0.81 0.57 20 1.28 30
Rtota 2.75 3.77 3.07 100 4.28 100 (80)f
doi: 10.12952/journal.elementa.000096.t002.

Table 2.

Comparisons of OH radical reactivitya expressed in units of [s-1]

a Rtot = RCH4 + RCO + RNO2 + RsumVOC = kOH+CH4 [CH4] + kOH+CO [CO] + kOH+NO2 [NO2] + ∑ (kOH+VOC [VOC]); may not match sum of individual contributions due to skewness.

b Data from Table 1 in Gilman et al. (2009).

c Wind directions limited to 60–140 degrees, all times of year.

d Only summer (DOY 120–270, May through September) in addition to wind direction limitation.

e Methane was assumed to be constant at 2000 ppb, and carbon monoxide was made to vary smoothly between 100 (winter maximum) and 60 ppb (summer minimum), at the Floresville site.

f Number in parentheses reflects reactivity excluding oxygenated VOCs.

Similar results to those presented in Table 2 were obtained at the Boulder Atmospheric Observatory in late winter (Gilman et al., 2013; Swarthout et al., 2013). Compared to urban areas, however, the OH reactivity downwind of both the Colorado and Texas shale areas is largely associated with saturated hydrocarbons, not unsaturated ones such as in Houston. This is further illustrated as a bar chart in Figure 6, which includes a reasonable estimate of missing relative reactivity associated with oxygenates: while alkanes contributed only 20% of OH reactivity in the HGB, they contributed 70% downwind of the shale area, while alkenes contributed 26% and 13%, respectively. The composition did not change much between summer and winter likely because the Floresville site is not near significant sources of biogenic hydrocarbons (Yu et al., 2015). However, since there are several regions with higher densities of oak trees on the eastern side of the shale, their isoprene emissions may contribute more significantly to OH reactivity in those regions of the shale area (Sullivan, 2014).

doi: 10.12952/journal.elementa.000096.f006.
Figure 6.  

Comparative median contributions to OH radical reactivity.

Relative, median contributions of different VOC groups to OH radical reactivity in the HGB (in 2006) vs. the Floresville monitor (2013/14). The amount that oxygenated VOCs may contribute to the latter site was estimated to be 10% of the total, equivalent to approximate median VMRs of 0.6 ppb formaldehyde and 0.3 ppb acetaldehyde.

Whether the observed high OH reactivities downwind of the shale area have a significant impact on regional ozone depends on local and regional NOx abundances. While UOG exploration is a source of NOx from local, well-site fuel combustion, flares, and increased regional truck traffic, current ozone modelling using the established emission inventory (AACOG Natural Resources Department, 2015; Pacsi et al., 2015) suggests that the shale area’s ozone formation is largely NOx-limited. Nevertheless, the emission inventory may underestimate NOx emissions from flares and the current modelling strategy does not resolve flare plumes. Olaguer (2012) showed emissions of reactive hydrocarbons and NOx from flares can first reduce, then produce ozone in the plume, which, depending on radical abundances versus NOx, may reduce or enhance regional ozone similar to observed ozone in power plant plumes (Ryerson et al., 2001). Given the high OH reactivity observed at the Floresville site, shale area flare plumes may become ozone sources at short distances from the emission point due to the large pool of radical precursors in background air, and this should be explored further in support of area-wide surface ozone modelling.

3.4 Case studies

Here, we discuss two case studies that illustrate the impacts that UOG exploration in the Eagle Ford can have on air quality. To put the two case studies into context, we show the complete time series of hourly average ethane VMRs in supplementary Figure S2. Ethane mixing ratios regularly peaked throughout the year-long data set. Mixing ratios exceeded 50 ppb (roughly equivalent to the 95% quantile of all ethane data) for at least two hours on 23% of all measurement days. Such high values at such high frequency appear unusual.

3.4.1 March 2014 hydrocarbon plumes

Our first case study focusses on the maximum ethane VMRs exceeding 200 ppb observed in March 2014. This case was chosen because it coincides with a local resident’s air quality complaint found in TCEQ’s complaint data base (, #195126, on 6 March 2014. At the time, south central Texas was under the influence of high pressure and weak winds, with nighttime temperatures between 5 and 10 °C, and clear skies (see supplementary materials Text S1). These conditions were conducive of shallow nighttime boundary layer depths and most likely contributed to the very high hydrocarbon VMRs observed during those days. Figure 7 shows two individual pollutant plumes observed at the Floresville monitor on 5 and 7 March. Both plumes contained mostly saturated hydrocarbons but also unsaturated ones. The latter can be formed as the result of incomplete combustion in flares. The first plume arrived in Floresville, located at the north edge of the shale (Figure 1), just before midnight on 4 March. Concentrations rose slowly, dipped mid-plume, and then dropped again in the late morning hours. The second plume two days later showed significantly steeper edges, meaning the plume passing the monitor was more well-defined on that day. Both plumes contained NMHCs from very recent emissions as demonstrated by looking at changing hydrocarbon ratios during passage, supplementary Figure S3 (e.g. Helmig et al., 2008).We performed a HYSPLIT (Draxler and Rolph, 2015) plume dispersion analysis (supplementary Figures S4S6), placing the presumed emission source at a large midstream (gathering) facility west of Karnes City for which the aforementioned complaint was made. The analysis showed that a slowly west-northwest moving, laterally dispersing plume approached Floresville the night of 4–5 March, before being displaced south by northerly turning winds in the evening of 5 March. On 6 March, only the plume edges reached Floresville before it was again displaced southwards. However, on 7 March, a stronger southerly breeze and much less lateral dispersion brought a more well-defined plume to Floresville, with its edges passing over the monitoring site from west to east with changing wind directions from SE to S. The first plume scenario (5 March) appears to miss the receptor site until much later in the day but the second (7 March) matches the onsite observations in their timing and location well, which suggests that the chosen emitter vicinity was likely accurate, and that the emissions causing this extreme event must have lasted for at least three days. Additional support for the hypothesis that these plumes were dominantly coming from the presumed emitter comes from a rare FLIR video of opportunity (Video S1) obtained at the site, which confirmed not only the emissions, but also that shallow boundary layer depths existed at the time. While this is relatively strong evidence, we cannot exclude that the observed hydrocarbon plumes may in part have been due to sources along a trajectory including the presumed dominant emitter and the receptor site. A good match between the observed plume and the plume model arrival timing was obtained for the 7 March but not the 5 March plume, suggesting that additional uncertainty comes from the resolution and spatial coarseness of the HYSPLIT model (12 km grid input data).

doi: 10.12952/journal.elementa.000096.f007.
Figure 7.  

March 2014 hydrocarbon plumes.

Selected hydrocarbon VMRs observed at the Floresville monitor during two plume passages in March 2014. Wind directions were southerly during plume observations and northerly in between (daytimes). See the supplementary materials for the plumes scenarios.

If the emitter site allocation is accurate, and if the emitter was the far dominating emission source contributing to this plume, then plume dispersion model results (supplementary materials Text S1) suggest that local hydrocarbon VMRs at the emitter site could have been on the order of 100 to 1000 times higher than observed at the Floresville monitoring site (Stein et al., 2015). At such high VMRs, numerous hydrocarbons would have locally exceeded their odor threshold values, which explains the odor complaints made to TCEQ. As a result, local values at the emitter site might have exceeded several hundred ppb, and thus several of TCEQ’s short term air monitoring comparison values (AMCV), for example for benzene and hexane (TCEQ, 2015).

We note that both plumes contained significant amounts of ethene and propene. The 7 March nighttime plume also showed acetylene and NOx. Since traffic typically minimizes at these nighttime hours, those compounds were likely co-emitted and co-advected from combustion sources such as flaring, then preserved in the plume due to the virtual absence of OH radical chemistry at night. However, the much larger alkane abundances in the plumes suggest that flaring was likely not the main source of emissions in this case. Otherwise, the plume composition suggests an extremely inefficient flare, mostly emitting raw gas.

3.4.2 September 2013 air mass origin changes

Our second case study looks at a typical period of measurements during September 2013 that illustrates the hydrocarbon sources in the area, identifies a “clean sector”, and highlights the limitations of an air monitor as far removed from the shale area’s sources as the Floresville monitor is (Figure 1). Figure 8 shows a two-week period of measurements beginning mid-September 2013. Two hydrocarbons, one each to represent the two dominant hydrocarbon sources, are shown alongside wind direction and speed. The period began with typical daytime southeasterly flows that showed elevated n-butane and propene VMRs during nighttime and a correlation dominated by butane over propene of approximately 50:1, meaning stronger oil and gas than combustion source impacts. VMRs were lower on day of year (doy) 259 (16 September) when wind direction did not turn SE during daytime, and this effect becomes clearer when wind direction turned to NE on doy 263 as a result of a midday cold front passage. The associated change to a continental air mass under northeasterly wind directions first increased then decreased wind speeds. As a result of this air advection, n-butane and propene VMRs both dropped to much lower levels, and changed the butane-to-propene ratio to as low as 5:1, indicating much lower oil and gas source influences in this continental air mass. As winds turned northerly under high pressure another two days later, wind speed decreased further and propene VMRs increased, most likely due to local emissions from car traffic as evident from its peaks during rush hours. A high degree of photochemical processing in this air mass was obvious from benzene-to-toluene ratios exceeding 2 during daytime (not shown). Butane increased during this period only when wind directions turned temporarily easterly under low wind speeds at night. A day after a weaker, second cold front had passed on doy 267 (24 September), winds rapidly turned back to southeasterly directions, causing n-butane again to increase and propene to decrease, such that by doy 270 (27 September), an air mass origin and composition similar to the beginning of this period was reestablished.

doi: 10.12952/journal.elementa.000096.f008.
Figure 8.  

Typical butane and propene variations in differing air masses.

Time series of n-butane (open circles) and propene (filled squares) (top panel) alongside wind direction and speed (bottom panel) for a typical 2-week summer period in September 2013 (see text for discussion).

This period showed that the NE sector is a “clean” sector for this air quality monitor. It also exhibited a correlation of propene with n-butane under southeasterly wind directions that was strongest during the hours associated with higher vehicle traffic. This wind direction can bring emissions from sections of highway 181 between Floresville and Karnes City to the monitor (Figure 1), causing co-advection of the two main source factors (section 3.2), and illustrating the difficulty of assigning well-defined emissions sources in data from a non-central monitoring site. Thus, in this case, we cannot assign a relative emission to the potential flaring emission compounds ethene and propene similar to case study 1.

4. Conclusions

The first year of measurements at a recently established air quality monitor in Floresville, TX, just north of the Eagle Ford shale, shows high VMRs of UOG related hydrocarbons, clearly related to air mass trajectories traversing the shale. VMRs are regularly higher at night due to shallower boundary layer depths, and show characteristics of both “routine” emissions, elevating median ethane to 4–5 times its background VMR, and upset emissions elevating ethane to more than 100 times its background downwind of the source. Similar to ethane, nearly all compounds associated with UOG emissions through factor analysis showed contiguous elevated abundances above their 95% quantiles on average during one out of five to one out of four days. Meaning, episodes with VMRs belonging to the highest 5% of all data occurred regularly throughout the first year of measurements from summer 2013 to summer 2014. At the high abundance levels observed, UOG related NMHCs and their typically not monitored oxidation products, such as formaldehyde and acetaldehyde, may significantly contribute to ozone formation on a regional to continental scale (Ahmadov et al., 2015; Field et al., 2015). This is supported by an evaluation of local OH radical reactivities revealing conditions similar to large urban areas, especially when winds advect high NMHC mixing ratios over the shale area. Conditions are thus conducive to efficient ozone production given adequate NOx abundances.

Our factor analysis reveals that NMHC emissions from the shale area dominate pollution variability downwind of the shale, while traffic emissions dominate under the reverse wind direction over the city of Floresville. The siting and operation of the monitor does not allow an unequivocal quantification of alkene emissions from UOG related combustion sources such as flaring due to possible alkene losses and co-advection of UOG and car traffic related emissions. However, despite potential biases due to sampling methodology, short-chain alkene data at times correlate with saturated hydrocarbons suggesting co-emission from UOG activities. This possible co-location of emission sources of reactive VOCs and NOx, either due to flaring (Torres et al., 2012c; Pikelnaya et al., 2013; Duncan et al., 2016) or well-site diesel engines (Warneke et al., 2014) and truck traffic (Texas Transportation Institute, 2015), is of importance for estimating the emissions’ impacts on ozone formation, which may require smaller scale plume modeling (Olaguer, 2012). While overall NOx levels may be lower in the shale area compared to urban areas such as Houston (Sullivan, 2014), likely due to lower traffic density, co-located emissions can lead to plumes of higher concentrations directly downwind of well sites (Warneke et al., 2014); and this situation is not accessible via standard 4 km resolution ozone air quality grid modelling (AACOG Natural Resources Department, 2015). We note that flaring distinguishes the EFS from other shale areas (Duncan et al., 2016). Typical flares associated with oil wells in the EFS are neither controlled for optimal fuel-to-air stoichiometry, nor air- or steam-assisted. Thus, their emission amounts and composition may not be well represented by recent flare studies (Torres et al., 2012a, 2012b) and/or the commonly used EPA AP-42 document (EPA, 2015) emission factor estimates. We therefore recommend that measurements be located closer to various flaring sites in the Eagle Ford shale to assess both flaring emissions and near-source concentration impacts. In addition, testing the quality of the alkene data with respect to possible ozone and Nafion drier artifacts is necessary to further assess the alkene observations. We plan to further analyze TCEQ’s data at this and the newer monitoring site in Karnes City, both operating in 2015, to compare NHMC emission inventories to observations in the Eagle Ford shale area.

Data accessibility statement

A dataset was generated from raw data provided by TCEQ. R scripts generating the subsets of data analyzed and creating the Figures in this manuscript were assembled. Both are available from the authors upon request.


© 2016 Schade and Roest. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.

Supplemental materials

The supplementary materials provided with this manuscript include a wind-rose for the monitoring site, additional Figures and Tables, three plume scenarios developed using NOAA’s online HYSPLIT program, a picture of the area and two FLIR camera videos of an emissions event in Karnes County taken by Sharon Wilson on 6 and 7 March 2014.

Text S1.
Supplementary materials summary
File Type: DOC
File Size: 0.01 MB

Figure S1.
File Type: JPG
File Size: 0.13 MB

Figure S2.
Time series of ethane VMRs
File Type: JPG
File Size: 0.26 MB

Figure S3.
Temporal change in propane-to-ethane ratios
File Type: JPG
File Size: 0.46 MB

Figure S4.
Plume dispersion model for 4–5 March 2014
File Type: GIF
File Size: 3.90 MB

Figure S5.
Plume dispersion model for 5–6 March 2014
File Type: GIF
File Size: 4.66 MB

Figure S6.
Plume dispersion model for 6–7 March 2014
File Type: GIF
File Size: 4.77 MB

Figure S7.
Photograph of area west of Karnes City taken on 6 March 2014
File Type: JPG
File Size: 3.43 MB

Video S1.
FLIR footage from 6/7 March 2014
File Type: MP4
File Size: 13.5 MB

Table S1.
Hydrocarbon correlations
File Type: DOC
File Size: 0.01 MB

Table S2.
Comparative VOC and NO2 VMRs
File Type: DOC
File Size: 0.02 MB


Conception and design: Schade and Roest

Formatting and exploration of data: Roest

Analysis and interpretation of data: Schade and Roest

Drafted and revised the article: Schade

Proofread the article for publication: Roest

Competing interests

The authors declare no competing interests.

Funding information

This work was conducted with internal support and did not receive any funding from an outside agency.