Abnormal tank emissions in the Permian Basin identified using ethane to methane ratios

There has been considerable increase in oil and natural gas (ONG) production in the United States in the past decade that creates the possibility of an increase in associated methane (CH₄) emissions, which numerous studies have noted (Schneising et al., 2014; Franco et al., 2016; Hausmann et al., 2016; Helmig et al., 2016; Alvarez et al., 2018; Nisbet et al., 2019; Raimi, 2019). The Permian Basin in Texas and New Mexico covers more than 75,000 square miles (Energy Information Administration [EIA], 2020). It is the largest oil producing shale formation in the United States with 5.208 million barrels per day of oil and 20.280 billion cubic feet per day of natural gas as of April 2022 (EIA, 2022). Hence, there has been interest to quantify and mitigate the CH₄ emissions from this region. A recent ground-based study reported well-pad CH₄ emissions in the Permian 5–9 times higher than Environmental Protection Agency (EPA) inventory estimates (Robertson et al., 2020). Airborne and satellite analysis has also produced CH₄ emission rates that are higher than inventory estimates (Schneising et al., 2020; Zhang et al., 2020; Irakulis-Loitxate et al., 2021; Chen et al., 2022). Recent studies to constrain total CH₄ emissions from the Permian Basin have reported emissions from the production sector contributing approximately 50% of the total basin emissions (Cusworth et al., 2021; Chen et al., 2022). These studies also suggest that the largest emissions are well above the emission range seen from ground campaigns but could not distinguish the onsite source of emissions in most cases, though intermittent flares were identified as contributing 12% of emissions (Cusworth et al., 2021). Ground-based sampling require large sample sizes to catch these “superemitters,” which are infrequent and/or short-lived and have a low probability of being randomly sampled (Wang et al., 2022). Additionally, sources with lofted plumes (such as flares) may be impossible to quantify via ground-based methods if the plume remains above the measurement height.

The production sector includes well pads and tank batteries, where a typical ONG well pad may consist of oil derricks or wellheads, compressors, crude or condensate tanks, produced water tanks, pneumatic controllers, and flaring units (EIA, 2021). Some of the routine activities like venting, use of pneumatic controllers, unintentional leakages, malfunctioning flaring units, and storage tanks contribute to the overall emissions from the production sector (Allen et al., 2015a; Allen et al., 2015b; Tyner and Johnson, 2021; Allen et al., 2022; Zimmerle et al., 2022). One method to identify a specific CH₄ source type is by measuring a tracer gas emitted along with CH₄ such as ethane (C₂H₆). C₂H₆ is primarily emitted from ONG sources and thus has been used as a suitable tracer to distinguish ONG emissions from other CH₄ sources such as livestock (Smith et al., 2015; Peischl et al., 2018; Pollack et al., 2022).

Previous work has provided limited differentiated ethane to methane ratios (EMRs) for specific sources of various types of fossil fuel extraction and refining (Yacovitch et al., 2014; Yacovitch et al., 2017; Yacovitch et al., 2020). More commonly, EMRs are reported for large areas. Kort et al. (2016) determined the C₂H₆ emissions and the EMR from the Bakken Shale region in North Dakota using aircraft measurements. Similarly, using airborne CH₄ and C₂H₆ measurements, Smith et al. (2015) determined EMRs for the microbial, low C₂H₆ fossil, and high C₂H₆ fossil sources in the Barnett Shale region in Texas. Peischl et al. (2018) characterized CH₄ and C₂H₆ fluxes for several ONG regions around the United States. Both Peischl et al. (2018) and Smith et al. (2015) were able to quantify ONG CH₄ emissions in the regions of mixed sources and demonstrate the use of these EMRs in constraining their results. More recently, estimates of EMRs for different oil-bearing and dry gas regions were used to identify the importance of oil reservoirs (like the Permian Basin) as dominant sources of CH₄ among ONG activities (Tribby et al., 2022).

EMRs for specific ONG processes may be expected to change with geology, which affects the initial gas composition and can be quite variable (Tzompa-Sosa et al., 2017). Downstream of the production sector, the EMR of gas is lowered as C₂H₆ and other natural gas liquids are separated and the processed gas (>95% CH₄) is sent via the transmission sector to customers (American Petroleum Institute [API], 2021). Flaring may lower the EMR from the source gas as C₂H₆ is expected to combust more efficiently than CH₄, but this will depend on meteorology, gas exit velocity, and flame stability (Leahey et al., 2001; API, 2021). At many sites, produced water, condensate, and oil containing dissolved gases are stored onsite in tanks at near-atmospheric pressure after being passed through a high-pressure separator that separates natural gas from liquids. The tanks periodically vent as pressure exceeds a set point, causing a quick release of the dissolved hydrocarbons that have partitioned out of the oil/water and into the gas phase. These emissions are known as tank “flashing” and the EMR will be a function of the dissolved hydrocarbon concentrations and each species’ solubility, which is affected by temperature and pressure (API, 2021). Crude and condensate tank flashing typically has higher EMRs than the associated produced gas (Cardoso-Saldana et al., 2021). This study focuses on the use of C₂H₆ as a tracer to differentiate sources within the production sector in the Permian Basin. We measured C₂H₆ concentrations simultaneously with CH₄ and calculated site specific EMRs that were then assigned to the identified emitting source(s) onsite.

Using the University of Wyoming mobile lab (Robertson et al., 2017), 2 sampling campaigns were completed in January 2020 and October–November 2020. ONG well pads and tank batteries in the Permian Basin in Texas and New Mexico were sampled randomly based on optimal wind direction. The region sampled primarily covered the Delaware Basin, which is the western portion of the Permian Basin. A map of the sampled locations is provided in Figure 1.

The University of Wyoming mobile lab included a 2D weather station, 3D sonic anemometer, and an inlet mounted 4 m above the ground connected to a gas sampling manifold. Inside the van, a 2-Hz Picarro Cavity Ring-Down Spectrometer (CRDS, Model G2204) was used to measure CH₄ by sampling from the manifold. C₂H₆ measurements were also collected from the manifold using an Aerodyne Ethane-Mini spectrometer, a tunable infrared laser direct absorption spectroscopy instrument (QC-TILDAS), which has a frequency of 1 Hz (Yacovitch et al., 2014). The Picarro CRDS was calibrated using a high-precision standard CH₄/C₂H₆ air mixture created by the World Meteorological Organization/Global Atmospheric Watch Central Calibration Laboratories at the National Oceanic and Atmospheric Administration’s Global Monitoring Division with 1,936.3 ± 0.2 ppb CH₄ and 2.09 ± 0.01 ppb C₂H₆ (prepared October 30, 2019). This was carried out twice throughout the sampling campaign. The reported precision for the CH₄ measurements was 2 ppb in 5s, and the reading was always within 2.5 ppb of the standard. Similarly, for the calibration of Aerodyne Ethane-Mini spectrometer, the CH₄/C₂H₆ air mixture was used, and the instrument was zeroed every 30 min using ultra high-purity zero air. The calculated precision for the C₂H₆ measurements was 80 ppt in 1s, and the reading was always within 0.3 ppb of the standard. In addition, the C₂H₆ analyzer was calibrated using dilution of a commercial 50 ppm C₂H₆ standard by ultra high purity zero air. The analyzer showed excellent linearity (R² > .99) for readings up to at least 5,000 ppbv C₂H₆.

As part of the Environmental Defense Fund’s Permian Methane Analysis Project, this campaign was designed to capture data suitable for CH₄ emission calculations using the Other Test Method (OTM) 33A (EPA, 2013; Brantley et al., 2014). OTM 33A is an emission quantification technique that uses an inverse Gaussian flux method to calculate fluxes at the well pad level. High-frequency CH₄ and meteorological data collected at a stationary point downwind of a source along with source distance are the required inputs to produce an emission rate (in kg CH₄ hr⁻¹). Accordingly, data suitable for OTM 33A emission calculations were collected, while the van was stationary and downwind of a source for at least 20 min. Emissions were calculated for sites that had CH₄ enhancements above approximately 50 ppbv at a distance of less than 150-m downwind, which is the minimum enhancement suitable to calculate emissions by OTM 33A. Occasionally, during this campaign, transects were driven downwind of sources suitable for emission calculation by transect method (Caulton et al., 2018). The OTM 33A and transect data were used to calculate EMRs, which is the focus of this work.

Ratios were calculated by total least squares regression between the CH₄ and C₂H₆ volumetric mixing ratios, where the slope of the fit represents the molar EMR. The total least squares approach allows for errors on both the x and y data. The reported 95% confidence interval (CI) for each ratio is calculated from the uncertainty of the slope. These ratios are expressed as a percentage of the CH₄ mixing ratio (ppbv/ppbv × 100). Ratios were screened to remove sites that showed low correlation (R² value) between C₂H₆ and CH₄. The R² value used to screen out sites was .65 (Yacovitch et al., 2014).

Optical gas imaging using an FLIR camera (model GF300) was taken during sampling, and whenever possible, the source(s) of emissions was noted. Sites were never sampled when onsite operator activity (e.g., maintenance) was observed; thus, no sources should represent maintenance activity or manual liquid unloadings. Bell et al. (2017) observed that OTM 33A underestimated emissions from sites that had manual liquid unloadings, which contributed a significant fraction of the total emission in their data set. The high emission rates of the manual liquid unloadings may result in lofted plumes, which can be transported above the downwind sampling location. Although Bell et al. (2017) also noted general underestimation of emissions from OTM 33A, more recent work using controlled releases from a variety of well pad infrastructure generally showed good agreement. However, their release rates did not span above 2.15 kg hr⁻¹ (Edie et al., 2020). Due to our sampling criteria, we assume that the emissions estimates for these sites are robust and do not primarily contain sources that would have lofted plumes the ground sampling technique could not measure accurately. Error estimates for OTM 33A derived CH₄ emission rates are reported to be +54%/−26% (Edie et al., 2020) and +170%/−50% for transect-derived emission rates (Caulton et al., 2018).

Sites were grouped by common emission source type for the analysis of differences in EMRs. As sites can have multiple sources, this analysis is not without some subjectivity, and it is possible that the identified source was not the only or primary source at a site. However, this procedure is consistent with the general way leaks are detected via optical gas imaging (e.g., Bell et al., 2017) albeit without onsite access. The team made observations by FLIR camera as close to the sources as possible from publicly accessible land (typically at the edge of the well pad or road). The source categories defined for this analysis included “compressor,” “pneumatics,” “separator,” and “tank.” A source category contained any emission relating to that source type (e.g., tanks include any type of valve, fitting, vent, pipe, or thief hatch on a tank and any type of tank: oil, condensate, produced water, and saltwater). Any site with more than one source noted was put into a “mixed signal” category. The pneumatics category contains emissions from wellheads and any pipe infrastructure on the site. Additionally, some sites had no obvious source, or no information recorded at the time of sampling, and were grouped together as “none/not identifiable.” Full details of the source observations from each site are reported in Table S1.

A few sites sampled (n = 12) displayed 2 distinct EMR signals (Figure 2). Many of these sites were initially screened out due to low correlation coefficients stemming from the fact that a single fit could not represent the data. In order to separate the EMR signals, principal component analysis (PCA) was performed. PCA was used to identify distinct groups, which were subsequently analyzed separately. The results from an example of such an analysis are shown in Figure 2. The analysis of the wind direction data was used to identify the probable sources from site notes and photographs when possible. Figure 2 shows an example of a site with distinct wind directions associated with each signal. The EMR signatures can also be used to parse the total CH₄ flux from the site into the contributions from individual signals, as detailed in Text S1. This calculation requires the total site CH₄ and C₂H₆ emissions calculated either via OTM 33A or transect method and individual EMRs. Not all dual-signal sites produced 2 signals that passed the R² screening threshold, and thus, not all of these sites could be parsed. Results for this analysis are presented in Tables S2 and S3. In general, the few sites with multiple EMR signatures where one signal could be attributed to a tank showed that nontank sources typically were the largest CH₄ source. Further discussion of tanks with respect to EMRs and CH₄ emissions overall is presented in Section 3.2.

As an example of this process, we discuss Sites S02 and S03 measured on January 22, 2020, in more detail. These were the repeated measurements of the same site and present a unique case study. Prior to sampling, FLIR videos identified that the emissions were coming from a separator and a tank, which were about 65 m apart on the site. Premeasurement transects showed consistent distinct peaks for these sources (Figure S1). The initial sampling (123-m downwind) was oriented in the centerline of the separator plume, and it was observed that the emission from the tank onsite would not be fully captured. The initial OTM 33A measurement (S02) was completed, and afterward, the team moved position to the centerline of the tank plume and completed another measurement (S03). An average wind speed of 4.4 m s⁻¹ was observed (range: 1.2–9.0 m s⁻¹). The low EMR is remarkably consistent between these sites (3.3% for S02 and 3.5% for S03). However, in S03, there is an additional signal observable in the data that return an EMR of 21% coming from a tank. Additional analysis to corroborate the source signals and contributions is provided in Text S2, which included using the transect plumes to calculate component emissions. Using the parsing method, the contribution to the total S03 CH₄ emission from the tank is small (7% of the total emission) and reasonably consistent with source specific emission estimates calculated from the transects (13% of the total emission).

This analysis of separate signals increased our sample size of screened EMRs from 88 to 100 and is used for the remainder of the analysis. These EMRs correspond to a unique site or to a unique component on a site. There are at most 2 EMRs per site. The range of ratios calculated varied from 3.3% to 157%. Statistics of this data set are reported using bootstrapping of 1,000 samples with replacement (meaning observations can be randomly selected multiple times). The arithmetic mean and median ratios with 95% CIs were 27 (±6)% and 14 (±1)%, respectively. In addition, the geometric mean was calculated as 18 (±2)%. More than 50% of the observations had EMRs between 10% and 20%, which likely represents produced gas; however, we observed several ratios over 100% indicative of oil or condensate tank flashing. The distribution of EMRs displays right hand skewness (skew = 3.0). Figure 3 shows the distribution of EMRs on normal and lognormal axes.

EMR statistics were calculated for the identified source categories. The results of this analysis are shown in Figure 4 with statistics reported in Table 1. There is a clear increase in average EMR for sites that have only tank emissions. This is consistent with observations that have reported high EMRs from tanks and processing equipment of wet-gas regions (Yacovitch et al., 2014; Goetz et al., 2015; Goetz et al., 2017). In addition, data have shown the enhancement of alkane emissions specifically from tank venting and flashing, through modeling and measurements (Pétron et al., 2012; Cardoso-Saldana et al., 2021). Mixed signals also show an elevated mean EMR. Most of the sites with mixed signals included at least one tank source, so it is reasonable to assume that the mixed signal EMR is enhanced from the presence of tanks. These source categories also showed a large range in EMRs as seen from their large standard errors (Table 1). Comparatively, compressors, separators, and pneumatics had relatively consistent and lower EMRs. The sites with no identified source information also showed elevated EMR signals and higher variability, like the mixed signal category. Although there is a broad range of EMRs in the sample, most (>70%) of the sites had ratios <21%. For the 28 observations with EMRs ≥21%, 16 were directly attributed to tanks, 7 to sites with mixed signals where tanks were present, 2 to pneumatics, and 3 were not identifiable. All 6 sites with EMRs over 100% came from tanks or mixed signals where tanks were present. We also sampled a ground/pipeline leak (EMR = 13.7%) and an isolated flare emission (EMR = 16.3%).

Generally, sites with sources identified as tanks or mixed signals also had higher CH₄ emissions, but the range of CH₄ emission observations is larger than the range of EMRs. Also shown in Figure 4 and reported in Table 1 are emission weighted mean EMRs. This is calculated by multiplying the EMR in a source category by that site’s CH₄ emission rate (in kg hr⁻¹) divided by the total CH₄ emission rate (in kg hr⁻¹) from that category and summing the individual contributions, as shown in Equation 1:

Some source categories show consistency between the mean and weighted mean EMR including compressors, pneumatics, separators, and even the category, where sources were not identifiable. Mixed signals and tanks, however, show the largest difference between the mean and weighted mean. Tanks in particular have a very high mean (47%) and comparatively low weighted mean (17%). This suggests that while high EMRs indicate the presence of a tank, the tanks that cause high emissions do not have high EMRs. For the purpose of this analysis, we have defined “high-emitters” (HEs) as sites with a CH₄ emission 10× the geometric mean of the CH₄ emissions of this data set (HE > 17 kg hr⁻¹). This value is not meant to be a universal standard (i.e., we are not suggesting this as a threshold for other data sets), and all but one of the sites measured had CH₄ emission rates <100 kg hr⁻¹. This procedure identified 7 observations or approximately 10% of the data set of CH₄ emissions as HEs. For tanks with EMR values ≥21%, none of the emissions can be classified as HEs. All the tanks associated with HEs in this data set had EMRs between 10% and 21%. The mean CH₄ emission rate for these low EMR (<21%) tanks (n = 8) was over 30 times higher and statistically different (using a 2-sample t test) from than the mean CH₄ emission rate for tanks with high EMRs (n = 12). Statistics for the tanks broken up by EMR are reported in Table 2.

To explain these observations, we hypothesize that the high EMRs represent normal tank operations (e.g., flashing, working, and standing losses) that do not appear to be primarily associated with high CH₄ emissions. Rather, the high CH₄ emissions may occur during abnormal conditions, where separator or other issues pass unprocessed gas directly to the tank where it can leak to the atmosphere. This hypothesis is supported by other work that has suggested emissions from ONG primarily arise from abnormal conditions (Zavala-Araiza et al., 2017; Alvarez et al., 2018; Luck et al., 2019). The precise source of the abnormal condition may be similar or related to a known emission point from separator dump valves, which are used to release accumulated liquids and can become stuck open due to debris or other issues (API, 2021). During oil extraction, the emulsion of fluids and gas must pass through a separator. Dump valves control the transfer of liquids (e.g., oil and water) to onsite storage tanks. Dump valves are known to occasionally become stuck open, which can allow raw gas to leak into (and out of) the tanks they are connected to. Tanks are typically kept near ambient pressure and have set points on their relief valves at just above ambient pressure. In this region, oil production dominates, and coproduced gas may be combusted or transported for sale via pipeline. We primarily measured at sites that did not have active flaring operations.

There is little direct support for this hypothesis without onsite reports of equipment status. In the absence of such data, we have looked at site characteristics for further evidence. Not all of the low EMR tanks produced large CH₄ emissions. If these low EMR tanks primarily represent abnormal conditions, one factor limiting the amount of CH₄ that can be emitted is the amount of produced gas. Generally, there has been little evidence for significant relationships between gas production and site-level CH₄ emissions. We assume that most of the ONG infrastructure in these studies were operating normally; thus, for normally operating sites, we expect a moderate to weak relationship between these parameters (Brantley et al., 2014; Lyon et al., 2016; Omara et al., 2016; Zavala-Araiza et al., 2018). We separated the tanks by low and high EMR and regressed them against the gas production corresponding to the month of measurement to investigate the significance of these relationships. The results of this analysis are shown in Figure 5. Low EMR tanks showed a positive correlation (r = .74) between the natural log of monthly gas production and CH₄ emissions. However, the slope of this fit is not statistically different from 0 at the 95% confidence level. On the other hand, the regression between the natural log of monthly gas production and CH₄ emissions for high EMR tanks shows weak negative correlation (r = −.35). For reference, the entire data set showed weak correlation between there parameters (r = .11). Following this analysis, the identification of HEs from tanks from this data set is consistently predicted by the presence of a low EMR and high gas production value. More observations are needed to corroborate this relationship, particularly with sites with even higher CH₄ emission rates (>100 kg hr⁻¹).

We caution that as subcategories are further divided, the number of observations in any category becomes increasingly small and prone to spurious relationships. It should also be noted that there is no evidence for a direct correlation between EMR and CH₄ emission. Using a regression of the calculated ratios versus the calculated OTM 33A or transect CH₄ emissions (n = 65), we found a Pearson correlation coefficient of −0.1. The correlation is only statistically significant (P < 0.05) for tanks with a correlation coefficient of −0.53 (Figure 5). However, there appears to be 2 distinct regions to the tank EMR versus CH₄ correlation corresponding to the 21% EMR threshold previously identified that suggests the linear relationship is not meaningful. Complicating this analysis is the fact that the CH₄ distribution of this data set is more positively skewed (skew = 5.1) than the EMR distribution (skew = 3.0). This is consistent with observations of CH₄ emissions from ONG operations that show extreme right skew behavior, which has been observed in the Permian as well (Brandt et al., 2016; Robertson et al., 2020). The presence of HEs that occur at low frequency has the effect of substantially altering the mean of any data set. Therefore, it is appropriate to use caution when interpreting trends associated with these extremes. The conclusion that tanks have statistically higher (using a 2-sample t test) mean EMRs than any other identified source is robust and consistent with previous observations (Goetz et al., 2015; Goetz et al., 2017; Cardoso-Saldana et al., 2021). The observation that tanks with low EMRs have on average higher CH₄ emissions than tanks with high EMRs is also statistically robust and a novel finding of this work. The interpretation that there is a direct relationship between gas production and CH₄ emission for low EMR tanks and an inverse relationship between EMR and CH₄ emission from tanks requires more observations and access to onsite operational detail to corroborate because HEs occur infrequently and can dramatically alter the regressions.

As mentioned, the range of EMRs observed in this data set produced a skewed distribution. This distribution yielded a range of statistics with different values, as stated earlier, with an arithmetic mean, median, and geometric mean of 27 (±6)%, 14 (±1)%, and 18 (±2)%, respectively. This gives rise to the question of which statistic is most appropriate to represent the region or useful for other analysis. Because the uncertainty on most of the statistics was relatively high, we also calculated a regional EMR through total least squares regression analysis of the background concentration data collected when transiting between sites. For this analysis, background data were calculated as a running mean of the lowest 30% of the data in 30-s bins, which removed sharp peaks but preserved large scale variations in the background. Some examples of this procedure are shown in Figure S2. The background data were then regressed to produce an EMR that should be representative of the weighted EMR for the region including sectors other than production. However, because this area was dominated by production sites, the ratio is expected to be similar to the production sector. We also separated the data based on the season of measurement to observe spatiotemporal trends in the EMR. The EMR ranged from 16.78 (±0.02)% in winter to 19.14 (±0.03)% the following fall with a combined ratio of 17.4 (±0.2)% (Figure 6). There is some overlap in the sampling area between the winter and fall campaigns though the area is not exactly the same (Figure 1). The geometric mean of the production sector EMRs (18%) compares best with the EMR calculated by regression for the region and may be the best statistic to represent skewed EMR distributions.

An additional vector of comparison can be made using available gas composition data, which has been previously explored in other studies. Kort et al. (2016) found that their EMR was consistent with the composition of NG production data from 710 sites in the Bakken Shale which had an EMR of 42%. Peischl et al. (2015; 2018) also reported C₂H₆ and CH₄ fluxes for several regions and compared to available gas composition data in those regions and generally found good agreement. This previous work suggests that gas composition may be used as a proxy for expected EMRs from production sites. For this study, we compared our results to the gas composition statistics from 19 wells in the Permian Basin (Eastern Research Group, 2012; Fairhurst and Hanson, 2012; Howard et al., 2015). The bootstrapped statistics for the gas composition mean and median EMR are 13 (±3)% and 15 (±6)%, respectively. The gas composition mean EMR is statistically lower than the mean ratio calculated from this study of 26%. It is also slightly lower (and statistically different) than the regional EMR (17.4%) and geometric mean (18%). The median gas composition EMR is more uncertain but compares better with the site EMR statistics and regional EMR calculated in this study. The gas composition data used in this analysis primarily came from wells in Texas in both the Delaware and Midland basins, which span a wide geographical area and include areas outside of our study region (Figure 1). Large data sets of gas composition are not always readily available (as in this case) and composition varies from well to well, thus comparison to a few wells is not very meaningful. The gas composition EMR from the available data varied from <1% to 24% and was not normally distributed. In addition, the results presented here show that the surface source types do not have uniform EMRs. These results suggest it is more appropriate to actually measure EMRs at the surface than assume gas composition is an equivalent metric.

This work reports the largest sample of production site specific EMRs in the Permian Basin to date. Tanks displayed distinct behavior with respect to CH₄ emissions for sites that were close to the regional EMR (10%–21%) and sites that had elevated EMRs (21%–157%). The highest CH₄ emissions from tanks in this data set had lower EMRs and high gas production values. Of the 5 highest emitting sites in this study, which contributed 75% of emissions, 4 sites were categorized as tanks and one as a mixed signal. However, none of these sites had EMRs over 21%. The observation that tanks are a primary source of elevated CH₄ emission rates in this data set is consistent with recent observations that also identify tanks as a major source of CH₄ emissions (Tyner and Johnson, 2021).

We have put forth a hypothesis for these observations, which implies that the elevated CH₄ emissions from tanks are mainly from produced gas escaping through the tank rather than tank flashing. This indicates that these high tank CH₄ emissions are driven by abnormal conditions and perhaps caused by or related to separator issues such as stuck dump valves. Because this emission is due to a component specific leak process, we would expect such observations to hold for ONG fields other than the Permian Basin. For example, as we have noted, there are other observations of elevated EMRs from tanks in the Marcellus Shale (Goetz et al., 2015; Goetz et al., 2017) and modeling efforts return high EMRs for tank flashing from multiple regions (Cardoso-Saldana et al., 2021). Therefore, EMRs could have use for determining when detected tank emissions are normal versus abnormal in many regions. Such observations could be used in support of achieving CH₄ reduction from ONG in accordance with the regulations the EPA is currently preparing, which include sections specific to tanks (EPA, 2021).

EMRs are computationally easy as they can be calculated directly from mixing ratio measurements and do not rely on meteorology. We used approximately 20 min of data for these calculations, but we were also able to calculate EMRs from aborted OTM 33A measurements and transects that lasted only a few minutes. It may be possible to quickly quantify the EMR from tanks to determine whether they are behaving abnormally and implement remediation, regardless of gas production value. Gas production values may have use for optimizing such a strategy to target high CH₄ emission sites.

In addition to distinguishing production sources, EMRs have use in defining the contribution of sector emissions at large scales (Tribby et al., 2022). Although previous work has shown that the mean EMR is generally close to the gas composition of the region (Peischl et al., 2015; Kort et al., 2016; Peischl et al., 2018), this work showed that EMRs varied by source, suggesting raw gas composition data are not an accurate representation of the surface emission EMRs, especially in areas that have equipment such as tanks (i.e., wet gas/associated gas regions). Other recent work has made an argument against using gas composition in regions, where transmission sector equipment produces lower EMRs than expected by gas composition data (Zimmerle et al., 2022). In addition, the measurements of processing equipment have shown a large range of EMRs and even scenarios where C₂H₆ is released without CH₄ (Yacovitch et al., 2014; Roscioli et al., 2015; Yacovitch et al., 2015). This illustrates the likely variability of EMR signatures across sectors in addition to regional differences. Cusworth et al. (2021) estimated the distribution of emissions based on sector in the Permian Basin in 2019 (Cusworth et al., 2021). It may be possible to corroborate such contributions from different sectors using C₂H₆ observations if they are associated with different mean EMRs. As Smith et al. (2015) showed, there is considerable uncertainty when attempting to use only C₂H₆ and CH₄ to partition signals in a region with multiple EMRs (Smith et al., 2015). Observations of other tracers, like CO₂ or H₂S, may be necessary to fully implement such analysis.

Finally, the results presented here provide some implications for low-cost sensors, which have been gaining increasing attention as a cheap and large-scale monitoring solution (Zhou et al., 2021; Riddick et al., 2022). Most low-cost sensors like metal oxide sensors (MOS) and photoionization detectors (PIDs) are not very selective, meaning they respond to hydrocarbons other than CH₄ (for MOS) or a suite of hydrocarbons (for PIDs which do not detect CH₄). These sensors may be used to infer CH₄ emissions or total VOC emissions if the makeup of the emitted gases remains constant (e.g., the EMR doesn’t change). However, when used to infer emission rates, they may be prone to producing false high emission rates when they are in the plume of a tank due to the elevated presence hydrocarbons. For example, an MOS sampling a normally operating wellhead and tank emitting the same magnitude CH₄ emission should show higher readings from the tank. This could lead to consistent “false positive” readings for tank emissions and limit the efficiency of leak detection from MOS or PIDs.

The data containing ethane to methane ratios, methane emissions, and supporting information are available at https://data.permianmap.org/datasets. This data repository is maintained by the Environmental Defense Fund and is free and open to the public with agreement to abide by the terms of use, which are available at the website.

The supplemental files for this article can be found as follows:

Text S1 to S2.

Figures S1 to S2.

Tables S1 to S3.

Available as a merged Docx file.

The authors would like to thank Megan McCabe for help in data collection during the January ‘20 campaign, Matt Burkhart and Zane Little for technical support and maintenance of the mobile lab, and Jack Warren for help in obtaining and aggregating the gas production data from Enverus.com.

This work was funded by the Environmental Defense Fund as part of the Permian Methane Analysis Project (PermianMAP) campaign. PermianMAP, which includes aerial, tower, and flare survey data, is grateful for the support of Bloomberg Philanthropies, Grantham Foundation for the Protection of the Environment, High Tide Foundation, the John D. and Catherine T. MacArthur Foundation, Quadrivium, and the Zegar Family Foundation. The School of Energy Resources at the University of Wyoming also provided financial support for the mobile lab, instrumentation, and students.

The authors declare no competing financial interests.

Contributed to conception and design: AMR, DRC, SMM.

Contributed to acquisition of data: AMR, KP.

Contributed to analysis and interpretation of data: AMR, DRC, DRL, PDG.

Drafted and/or revised the article: AMR, DRC, DRL, KP, PDG, SMM.

Approved the submitted version for publication: AMR, DRC, DRL, KP, PDG, SMM.

How to cite this article: Caulton, DR, Gurav, PD, Robertson, AM, Pozsonyi, K, Murphy, SM, Lyon, DR. 2023. Abnormal tank emissions in the Permian Basin identified using ethane to methane ratios. Elementa: Science of the Anthropocene 11(1). DOI: https://doi.org/10.1525/elementa.2022.00121

Domain Editor-in-Chief: Detlev Helmig, Boulder AIR LLC, Boulder, CO, USA

Associate Editor: Gunnar W. Schade, Texas A&M University, College Station, TX, USA

Knowledge Domain: Atmospheric Science