Low NO_X and high organic compound emissions from oilfield pumpjack engines

The United States is the world’s largest producer of oil and natural gas (Energy Information Administration, 2022). Many studies over the past decade have sought to understand how U.S. oil and gas production impacts local air quality (Ahmadov et al., 2015; Prenni et al., 2016; Evans and Helmig, 2017; Schade and Roest, 2018) and global climate change (Zavala-Araiza et al., 2015; Robertson et al., 2017; Alvarez et al., 2018; Riddick et al., 2019). These studies have included measurements and analyses of emissions across all scales, from individual components (Brantley et al., 2015; Subramanian et al., 2015; Thoma et al., 2017; Vaughn et al., 2018) to entire basins (Robertson et al., 2017; Omara et al., 2018; Foster et al., 2019; Pétron et al., 2020).

Stationary internal combustion engines, usually fueled by natural gas, are routinely used in the oil and gas industry to compress gas and actuate pumpjacks that pull oil from the ground (these are usually called artificial lift engines or pumpjack engines) (Temizel et al., 2020). Bottom-up emissions estimates consistently predict stationary engines to be among the largest sources of oxides of nitrogen (NO_X) in oil and gas basins (Bar-Ilan et al., 2006; Gorchov Negron et al., 2018; Dix et al., 2020), though we have not found any peer-reviewed studies that include direct measurements of oilfield engine NO_X emissions in field conditions. Studies of methane (CH₄) emissions from natural gas-fueled 4-stroke compressor engines show that CH₄ passing through engines uncombusted (a.k.a., fuel slip) is a common problem (Johnson et al., 2015; Zimmerle et al., 2015; Vaughn et al., 2018; Vaughn et al., 2021). Fuel slip is a challenge for natural gas-fueled engines outside of the oil and gas industry, as well (Kuppa et al., 2019; Speirs et al., 2020). Catalysts can alleviate this problem (Hutter et al., 2018), but they are not used uniformly in the oil and gas industry (Gorchov Negron et al., 2018).

Natural gas-fueled pumpjack engines are common in basins without electrical infrastructure (Gorchov Negron et al., 2018; Pétron et al., 2020; Lyman et al., 2021). To our knowledge, no peer-reviewed study had directly measured emissions from pumpjack engines, but 2 recent papers report enhancements of organic compounds in ambient air influenced by pumpjack engine exhaust. Pétron et al. (2020) encountered emission plumes from compressor and pumpjack engines during a mobile survey of ambient air in the San Juan Basin in Colorado and New Mexico, USA. The plumes had ratios of CH₄ and CO to CO₂ that were several times higher than those predicted by regulatory emissions inventories, leading Pétron et al. to call for more comprehensive work to characterize engine CH₄, CO, and CO₂ emissions. Lyman et al. (2021) found elevated alkenes and carbonyls in areas of the Uinta Basin in Utah, USA, that had a higher prevalence of 2-stroke pumpjack engines.

Most pumpjack engines in the United States are 2-stroke, 1- or 2-cylinder engines. These engines are used because they are mechanically simple, require little maintenance, and can operate in challenging environmental conditions (Brown, 2017). Two-stroke engines are particularly prone to fuel slip, however, because they take fuel into the combustion chamber at the same time that exhaust is expelled, allowing for fuel to be expelled with the exhaust (Nuti and Martorano, 1985; Meeks, 1992). Most of these engines run lean (i.e., they take in more air than is needed for combustion), which also leads to uncombusted or incompletely combusted fuel exiting with the exhaust (Harrington and Shishu, 1973; Kim and Bae, 2000).

We directly measured emissions of a comprehensive suite of compounds from 58 two-stroke, natural gas-fueled pumpjack engines in the Uinta Basin to better understand how these engines contribute to air pollution from the oil and gas industry. This article describes our measurement results and compares those results to emissions inventories and other available studies.

We calculated bootstrapped 95% confidence limits around means using the scikits.bootstrap tool in Python. We report values in this document as mean (lower confidence limit, upper confidence limit). We calculated calibration recoveries as the value reported by the instrument divided by the expected value.

Among regulatory agencies, the term volatile organic compounds (VOCs) is used to mean organic compounds that are more reactive than ethane (Environmental Protection Agency [EPA], 2022). Usually, this means it includes all organic compounds except methane and ethane. We avoid use of the acronym VOC in this work and included methane and ethane in our analyses, except when comparing our results to a regulatory emissions inventory. In that case, we used the regulatory definition of VOC.

We measured emissions from 58 engines belonging to 3 companies in Uintah and Duchesne Counties, Utah, between January and May 2021. We resampled 5 of these engines in January 2022. Table 1 provides a summary of the engines measured. All engines except the GM Vortecs powered pumpjacks. The GM Vortec engines powered hydraulic lift systems.

During each engine measurement, we measured ambient temperature and relative humidity (NM150WX; New Mountain Innovations, Marathon, FL, USA), wind speed and direction (WindSonic; Gill, Lymington, Hampshire, UK), and barometric pressure (CS100; Campbell, Logan, UT, USA) on a tower extended from the measurement trailer to a height of 6 m. We checked these measurements against NIST-traceable standards annually.

The pumpjack engines we encountered did not have reliable gauges to indicate rotations per minute (RPM), engine temperature, or other engine properties. We estimated RPMs by tapping out the number of audible rotations over a period (usually 30 s) on a smartphone app (Tap Counter, Svimph) and dividing the number of rotations by time. Our RPM measurements were within the range of manufacturer specifications, but we did not use these data for further analysis, except for Section 3 of the Supplemental Material.

We wrote a physical description of each engine’s exhaust apparatus and estimated the height above ground of each exhaust stack to within 0.3 m. We used a digital caliper to measure the internal diameter of each exhaust stack at its exit.

We used an Ecom (Gainesville, GA, USA) J2KN Pro Industrial analyzer to measure the physical and inorganic chemical properties of exhaust, with data collected every 2 s over 10 min periods in most cases. The analyzer included a cooled condensation trap to remove excess water vapor. It utilized standard electrochemical sensors for oxygen (O₂; p/n 7529701H), carbon monoxide (CO; p/n 7529281H for 0–4,000 ppm, p/n 7529301H for >4,000 ppm), nitric oxide (NO; p/n 7529801H), and nitrogen dioxide (NO₂; p/n 7529601H), and an infrared sensor for carbon dioxide (CO₂; 7533601H). Dewpoint was estimated by the instrument from O₂, temperature, and barometric pressure. We inserted the instrument’s standard 0.3 m sampling probe with pistol grip into exhaust stacks a few minutes before recording measurements and began recording after measurements stabilized. We attached one of several 13 mm outer diameter stainless steel tubes to the instrument’s sampling probe when it was possible to insert the probe deeper than 0.3 m into the stack, in accordance with manufacturer specifications. We used these tubes to collect chemical measurements at a depth of 1 m or more when possible. Depths less than 1 m resulted in a low bias in pollutant measurements due to entrainment of ambient air into the engine exhaust. We corrected for this bias (see Figure S1 and associated discussion in the Supporting Information). The Ecom analyzer’s probe incorporated a thermocouple to measure exhaust temperature (at 0.3 m depth), and the analyzer also measured exhaust pressure.

The Ecom analyzer was calibrated by the manufacturer before and at the end of the study. We also calibrated measurements of NO, NO₂, CO, CO₂, and O₂ weekly in our laboratory with compressed gas standards by inserting the instrument’s sampling inlet into a 3 cm outer diameter tube and flooding the tube with calibration gas. We used a 79 ppm calibration gas for NO (Airgas), a 100 ppm gas for NO₂ (GASCO, Oldsmar, FL, USA), 200 (Airgas, Vernal, UT, USA), 2,000 (GASCO), and 50,000 (GASCO) ppm gases for CO, 10% gas for CO₂ (Airgas), and ambient air for O₂. Ecom calibrated instruments with NIST-traceable gas standards at 30 and 100 ppm for NO, 10 and 100 ppm for NO₂, 1,000 and 10,000 ppm for CO, 12% for CO₂, and 12% for O₂. Calibration checks in air free of the measured gases produced responses of zero in all cases. Average calibration recoveries are shown in Table 2. Ecom found calibration recovery for NO, NO₂, CO (low sensor), CO (high sensor), and CO₂ at the end of the study to be 105%, 97%, 118%, 100%, and 97%, respectively.

We measured exhaust flow with the Ecom analyzer’s pitot tube-based flow measurement probe. In most cases, we affixed the probe at the exhaust stack outlet during the entire period of gas sampling. In some cases, we held the probe for at least 1 min at the stack outflow to collect the measurement.

We compared the Ecom pitot tube against a recently factory-calibrated Pacer (Keene, NH, USA) DA400 rotating vane anemometer. Because of the bidirectional nature of engine exhaust flow, rotating vane anemometers are not useful for exhaust measurements. We performed the comparison using a 20 cm duct and a blower to generate unidirectional airflow, and we partially blocked the blower intake to vary the flow between 2 and 6 m s⁻¹. The difference between the two instruments was 5.3 (3.9, 6.9)%. More information about flow measurement quality is available in the Supporting Information (including Figure S2).

We sampled organic compounds via a custom-built sampling line that was heated to 55°C. The tip was a 1.3 cm outer diameter stainless steel tube. We used tubes of different lengths, depending on the stack depth, and sampled at a depth of 1 m or more when possible. The maximum length used was 1.5 m. We inserted the probe into the end of the exhaust stack to collect exhaust gas samples. We corrected measurements for the depth of the probe in the stack as described above.

Downstream of the stainless tube was a stainless steel sintered filter (7 µm pore size) and a PFA Teflon filter pack that held a 47 mm diameter PTFE filter with a 5 µm pore size. After the filter pack was a 1 cm PFA Teflon tube of 15 m length. The final 0.5 m of the stainless tube, the filters, and all the Teflon tubing were heated to 55°C, which was hotter than the dewpoint of the exhaust samples. The sample line led to a generator-powered trailer, and a pump flushed sample gas through the line at 4 L min⁻¹.

We used a Los Gatos Research (San Jose, CA, USA) Fast Greenhouse Gas Analyzer to measure methane mixing ratios in the exhaust gas. The analyzer pulled gas from the sample line in the trailer. We calibrated the methane analyzer at a minimum of 2 points along its measurement range, including one zero point, on each field measurement day. We performed calibrations with 4 nonzero points weekly. We used a custom-built scrubber system to generate methane and carbon dioxide-free air, and we used calibrated mass flow controllers to dilute NIST-traceable compressed gas standards with this air for span calibrations. Calibrations standards were from Airgas (Vernal, UT, USA). We calibrated at 0, 35, 700, 25,000, and 80,000 ppm. The instrument’s response when introduced to air scrubbed of methane was 1 (0, 2) ppb. The analyzer has two lasers for methane, one for 0–2,000 ppm and one for mixing ratios greater than 2,000 ppm. Calibration recovery was 105 (104, 107) and 98 (96, 99)% for the low and high lasers, respectively.

Within the trailer, evacuated (<10 mtorr) silonite-coated 6 L stainless steel canisters with flow regulated by Alicat (Tucson, AZ, USA) mass flow controllers passively collected gas samples from the sample line. The line to the canisters, and the canisters themselves, were unheated (the trailer itself was heated, but its temperature fluctuated with the ambient temperature and with the opening and closing of doors). This may have caused condensation within the canisters, which could have inhibited recovery of alcohols. Also, alcohols could have reacted with NO₂ in canisters, which could also have inhibited recovery (Koda et al., 1985).

After sampling, we diluted fuel gas and exhaust gas sample canisters in the laboratory with ultrapure nitrogen gas to bring the mixing ratios of organic compounds within them into the range of our gas chromatography–mass spectrometry (GC-MS) system. We waited 24 h between each dilution step and prior to analysis to allow for mixing of gases within the canisters. We diluted most samples by pressurizing the samples to 3,800 mbar with ultrapure nitrogen, attaching a PFA Teflon line to one end of the sample canister, slightly opening the canister valve to allow a flow of sample gas, waiting 30 s, withdrawing a quantity of gas from the PFA line with a syringe, and injecting the gas into a nitrogen stream generated by an Entech (Simi Valley, CA, USA) 4600 dynamic diluter. The flow from the diluter passed into a clean evacuated silonite-coated 6 L stainless steel canister. The diluter tracked the volume of nitrogen added to the canister, and we determined the dilution ratio from the volume of sample injected and the volume of nitrogen added. We diluted some canister samples by first pressurizing the samples to 3,800 mbar with ultrapure nitrogen and then diluting them with ultrapure nitrogen by attaching the samples to the Entech 4600 dynamic diluter in the place of calibration gas standards. Exhaust gas samples were diluted to an average of 0.030% of their original mixing ratios.

We analyzed diluted canisters for a suite of 55 hydrocarbons and 3 alcohols (see the list of compounds in Table S1) within 7 days and analyzed diluted canisters within 45 days of collection. We have shown previously that the organic compounds analyzed in the canisters are stable for at least 45 days (Lyman et al., 2018). We used an Entech 7200 preconcentrator (in cold trap dehydration mode) and 7016D autosampler to concentrate samples and introduce them to a gas chromatograph (GC) system for analysis. The GC system consisted of 2 Shimadzu (Somerset, NJ, USA) GC-2010 GCs, one with a flame ionization detector (FID) and another with a Shimadzu QP2010 Mass Spectrometer (MS). The FID detected C2 and C3 compounds, and the MS detected all other compounds. All analytes first passed through a Restek (Bellefonte, PA, USA) rtx1 ms column (60 m, 0.32 mm I.D., 1.0 µm film thickness). A heated VICI 4-port valve with a Valcon E rotor then directed the C2 and C3 compounds to a Restek AluminaBOND/Na₂SO₄ column (50 m, 0.32 mm I.D., 5.0 µm film thickness) and into the FID. After these compounds passed (6.5 min), the valve switched to direct the remaining compounds into a Restek rtx1 ms column (30 m, 0.25 mm I.D., 1.0 µm film thickness), and then into the MS.

We used helium as the carrier gas. Both ovens started each analysis at 45°C. The oven that led to the MS contained both rtx1 ms columns. It maintained the start temperature for 10 min, then increased to 170°C at a rate of 7.5°C min⁻¹. Upon reaching 170°C, the oven increased to 250°C at a rate of 15°C min⁻¹, and then held at that temperature for 8 min. The oven that contained the AluminaBOND/Na₂SO₄ column and the FID maintained the start temperature for 15 min and then increased to 170°C at a rate of 5°C min⁻¹.

Duplicate analyses of canister samples were 0 (−2, 2)% different. Blanks contained 0.12 (0.11, 0.14) ppb of individual organic compounds. All results were blank-corrected. We used calibration standards from Airgas (Vernal, UT, USA) that contained about 1 ppm of each compound, and we diluted them to 100 ppb and collected different volumes and the 7,200 to achieve calibration points of 0, 50, 100, 150, and 200 ppb. Calibration curves always had r² values >0.99. Recovery of calibration standard checks was 99 (99, 100)%.

An independent pump pulled exhaust gas from the sample line through Waters (Milford, MA, USA) SepPak 2,4 dinitrophenylhydrazine (DNPH)-coated amorphous silica bead cartridges (350 mg), which collected carbonyl compounds. We used two cartridges in series, which allowed the second cartridge to capture any breakthrough, and we used the total of both cartridges in our analyses. The line to the cartridges and the cartridges themselves were heated to 55°C to avoid water condensation. A calibrated mass flow controller regulated flow through the cartridges. We kept cartridges refrigerated to the manufacturer-recommended temperature of <4°C before and after sampling. Gloves and clean surfaces were used to minimize contamination of the cartridge and sample container during handling and sample collection.

We eluted and analyzed DNPH cartridges by modifications of the methods of Uchiyama et al. (2009), Anneken et al. (2015), Shimadzu method LAAN-J-LC-E090 (Shimadzu, 2011), and Restek Lit. Cat.# EVSS2393A-UNV (Restek, 2018). These techniques are somewhat different from U.S. EPA Method TO-11A (EPA, 1999), which has become outdated due to improved instrumentation capabilities. We eluted cartridges within 14 days of sampling and analyzed the eluent within 30 days. To elute DNPH cartridge samples, we flushed cartridges with 5 mL of a solution of 75% acetonitrile and 25% dimethyl sulfoxide (percent by volume). We collected the solution into 5 mL volumetric flasks and brought the flasks to a volume of 5 mL using 0.5–1 mL of the acetonitrile/dimethyl sulfoxide solution. Finally, we pipetted a 1.6 mL aliquot from the 5 mL flask into two 2 mL autosampler vials for analysis by high performance liquid chromatography (HPLC). The second vial was kept as a spare in case of contamination or equipment failure.

We used a commercial standard mixture (M-1004; AccuStandard, New Haven, CT, USA) of derivatized aldehydes in acetonitrile for calibration. We analyzed samples with a Shimadzu (Somerset, NJ, USA) Nexera-i LC-2040C 3d Plus HPLC and a Shimadzu Shim-Pack Velox C18 column. We used a mixture of acetonitrile, tetrahydrofuran, and water as the eluent. We calibrated the instrument on each analysis day with a 5-point calibration curve and ran at least 1 additional calibration standard at the beginning and end of each analysis batch to check for retention time drift or other errors. Duplicate analyses of DNPH samples were 14 (10, 23)% different (average of all compounds). Blanks contained 0.02 (0.01, 0.0.02) ppb of individual carbonyl compounds. All results were blank-corrected. Recovery of calibration standards was 100 (99, 102)%.

Williams et al. (2019) cautioned that measurement of carbonyls in engine exhaust by DNPH cartridge methods presents a challenge because NO₂ and CO react with and consume DNPH, lowering the capacity for cartridges to capture carbonyl compounds. Also, NO₂ and CO can create peaks that interfere with correct identification of carbonyl compounds. Lange and Eckhoff (1996) found that these effects were minimized if less than 200 µg NO₂ were sampled through the cartridge. Tang et al. (2004) reported successful DNPH cartridge measurements of carbonyls in engine exhaust and even quantified NO₂-DNPH peaks in their HPLC analysis. Only 7% of our DNPH cartridges sampled greater than 200 µg NO₂. We did, however, identify some cartridges (14% of the total) that had no DNPH peak in the chromatogram (i.e., all the DNPH had been used). These samples also had an equal concentration of carbonyl compounds on both DNPH cartridges that were placed in the flow path in series (i.e., the first and second cartridges had both become completely loaded). Samples with completely loaded DNPH cartridges are indicated in the final data set and were not used for analyses of organic compound composition.

We collected fuel gas samples from 17 of the engine measurement locations in inert-coated stainless steel 500 mL sampling canisters with stainless steel valves on both ends. We connected the canisters to existing sampling ports on fuel gas lines at well pads (i.e., lines that connected to engines to supply gas used by the engines as fuel), opened both canister valves, and allowed fuel gas to flow through the canisters for about 60 s. We then closed the downstream canister valve, followed by the upstream valve, to collect each sample. We diluted these samples the same way as engine exhaust canister samples, except that we did not pressurize them with ultrapure nitrogen gas prior to dilution. We analyzed the diluted fuel gas samples for nonmethane hydrocarbons and alcohols by the same method as exhaust gas canister samples. Fuel gas samples were diluted to an average of 0.0025% of their original mixing ratios. We did not analyze fuel gas samples for carbonyls.

After analyzing canister samples for nonmethane hydrocarbons and alcohols with the GC-MS system, we analyzed them with an LGR (San Jose, CA, USA) UGGA methane analyzer. We connected a PTFE Teflon-lined pump to each canister, and the pump pulled air from the canister and pushed it into a tee. One outlet of the tee vented to the atmosphere and the other was connected to the analyzer. In this way, the analyzer could experience canister air at near-atmospheric pressure, even as the pressure in the canister itself varied during sampling.

The methane mixing ratio in each diluted canister divided by the mixing ratio measured in exhaust gas during field measurements constituted an independent estimate of the dilution amount for each canister sample. We used this value to determine the original mixing ratio of nonmethane organic compounds in the exhaust gas and to correct calculated dilution rates based on this value if needed. We used this same method to determine methane mixing ratios in fuel gas samples.

We calibrated the methane analyzer on each day of use using ultrapure nitrogen gas and NIST-traceable compressed gas standards (Airgas, Vernal, UT, USA). The instrument’s response when exposed to air scrubbed of methane was 1 (0, 1) ppb. We calibrated at 1, 4, 7, 15, 30, and 130 ppm. Recovery of methane calibration gas was 100 (99, 101)%.

To calculate the emission rate for each compound, we first multiplied the measured exhaust velocity (m s⁻¹) by the cross-sectional area of the stack (m²) at its exit to determine the flow rate of the exhausted gas (m³ s⁻¹). We then multiplied this flow rate by the concentration of each measured compound (g m⁻³) to determine the emission rate of the compound (g s⁻¹). All compounds except carbonyls were originally measured as mixing ratios (units of parts-per-billion by volume, or ppb). We converted mixing ratios to density units (g m⁻³) for emission rate calculations at standard conditions of 25°C and one atmosphere. Before calculating emission rates, we converted measured volumetric flow rates to standard flow using these same conditions.

We calculated the air–fuel equivalence ratio of each engine using the Brettschneider equation (Brettschneider, 1979):

where λ is the air–fuel equivalence ratio, [XX] is the gas concentration in percent by volume, H_CV is the atomic ratio of hydrogen to carbon in the fuel, O_CV is the atomic ratio of oxygen to carbon in the fuel, and n is the number of carbon atoms per molecule of fuel. We used 3.9, 0, and 1.5 for H_CV, O_CV, and n, respectively.

We performed several additional quality assurance tests, including matrix spikes (i.e., additions of known amounts of the gases analyzed) in laboratory and field settings and a comparison against an existing engine emissions data set. These are available in the Supporting Information (Figures S3 and S4; Table S2).

Many characteristics of engine emissions in the study were correlated with the engine air–fuel equivalence ratio (Figure 1). The air–fuel equivalence ratio is the ratio of air taken in by the engine to the amount of air needed for stoichiometric combustion of the fuel. Engines with equivalence ratios less than 1 tended to have the most CO in their exhaust (Harrington and Shishu, 1973). NO_X in exhaust was highest at an equivalence ratio just greater than 1, with declines below and above 1 (Harrington and Shishu, 1973; Wang et al., 2019) (r² = 0.46 for the power relationship of NO_X with equivalence ratios greater than 1). Many of the engines in the study operated at very high equivalence ratios, with 1 engine at an equivalence ratio of 10.6, meaning that the engine was taking in 10.6 times more air than was needed for combustion. Engines with equivalence ratios greater than 3 had median NO_X and CO of only 3 and 191 ppm, respectively. Low NO_X occurs at high equivalence ratios because the combustion temperature is too cool for abundant NO_X production (Yu et al., 2022).

Figure S5 shows that the ratio of NO to NO₂ increased in exhaust gases with higher total NO_X (r² = 0.73 for power regression), as was found by Varde et al. (1995). This means that a somewhat consistent NO₂ mixing ratio existed in all exhausts sampled, while the NO mixing ratio was higher at higher total NO_X. Engines with a high equivalence ratio, which tended to have low NO_X, thus tended to have more NO₂ as a percentage of total NO_X.

For this work, we calculated fuel slip as the ratio of carbon atoms in exhaust gas that were associated with organic compounds to the total carbon atoms in the gas. Figure 2 shows that the percent fuel slip of measured engines depended on the equivalence ratio, with fuel slip increasing at higher equivalence ratios (r² = 0.68 for the logarithmic relationship with equivalence ratios greater than 1), as others have shown (Varde et al., 1995; Kim and Bae, 2000; Liu et al., 2021). For engines with equivalence ratios greater than 3, the average fuel slip was 55 (51, 60)% (median of 56%). In other words, 55% of the carbon in the fuel passed through the engine and into the atmosphere in the organic form.

While engines with higher equivalence ratios tended to exhaust more fuel, engines with lower ratios tended to emit a larger percentage of total organics as secondary organic compounds formed in the combustion process (including alkenes, acetylene, and carbonyls; Figure 3). Exhaust from engines with air–fuel equivalence ratios greater than 3 had the lowest percentage of these compounds and the lowest mixing ratios (carbonyls and alkenes + acetylene were 83% and 72% lower, respectively, for equivalence ratios greater than 3). Alkenes, acetylene (Klimstra, 1990; Nine et al., 1997), and carbonyls (Mitchell and Olsen, 2000) are products of incomplete combustion. CO, another product of incomplete combustion, showed the same pattern (Figure 1). Alkenes + acetylene were correlated with CO (r² = 0.40), but carbonyls were not (r² = 0.01).

No measurements of engine or combustion temperature were available for this study. Instead, we used stack temperatures to derive an indirect estimate of combustion temperature (see Figure S6 and additional discussion in the Supporting Information). NO_X was correlated (r² = 0.47 for exponential relationship), while fuel slip (r² = 0.36 for exponential relationship) and equivalence ratio (r² = 0.36 for exponential relationship) were anticorrelated, with estimated combustion temperature. Higher combustion temperatures are associated with low equivalence ratios (Kim and Bae, 2000). We measured emissions from 5 of the same engines during summer and winter to determine whether ambient temperatures impacted emissions, but we did not find a statistically significant difference (Figure S7).

Figure 4 shows the average organic compound composition, by compound type, of fuel gas, exhaust gas, and a 2-stroke, natural gas-fueled engine exhaust gas composition profile in the EPA SPECIATE database (Oliver and Peoples, 1985; EPA, 2020). Alkenes and acetylene comprised a very small percentage of fuel gas but made up an average of 2% of the organic compounds in exhaust gas (also see Table S3). We did not measure carbonyls in fuel samples in this study. Instead, we used measurements of carbonyls in Uinta Basin raw natural gas collected by Wilson et al. (2020) to estimate carbonyls in fuel gas. Using this estimate, carbonyls were several orders of magnitude more abundant in exhaust gas than in fuel gas (Figure 4; Table S3). The profile from the SPECIATE database has a similar amount of alkenes and acetylene to the engines in this study that had relatively low equivalence ratios, but carbonyls in the SPECIATE profile are similar to the engines in this study with high equivalence ratios.

Many alkenes and carbonyls are very reactive with respect to ozone production (Carter, 2009; Carter and Seinfeld, 2012; Lyman et al., 2021), which means that emissions of organics from pumpjack engines can be expected to produce more ozone than an equivalent mass of emitted uncombusted natural gas. While emissions from engines with low equivalence ratios are much richer in alkenes, carbonyls (Figure 4), and NO_X (Figure 1), engines running under higher equivalence ratios emit more uncombusted fuel hydrocarbons overall (Figure 2). Photochemical modeling would be needed to definitively determine the optimal engine conditions for a given airshed.

Exhaust gas was somewhat richer in hydrocarbons with 2–4 carbon atoms (i.e., C2–C4 compounds) than fuel gas, but exhaust gas was poorer in C7–C10 compounds (Figure S8). The EPA SPECIATE profile was poorer in C3–C10 hydrocarbons than exhaust gas in this study, probably because of a difference in the composition of the fuel gas used by the engines measured for the SPECIATE profile. Formaldehyde comprised 89% of carbonyls in exhaust gas in this study (by weight), and acetaldehyde comprised another 6% (Figure S9). Methanol comprised 97% of alcohols in both fuel gas and exhaust gas in this study (Figure S10).

The different types of engines had different magnitudes and compositions of pollutant emissions. In general, Ajax engines tended to have higher total emissions and higher variability than Arrow engines (Figure 5). Ajax engines also tended to have more organic compound emissions as a percentage of the total, which was probably because they tended to have higher equivalence ratios (Figure S11). We only measured 2 GM Vortec engines, which are 4-stroke. They had very low CO and organic compound emissions and the highest emissions of NO_X. Arrow 795 engines are the most commonly used engine in the Uinta Basin (Table 1), and they had relatively low NO_X and relatively low fuel slip (Figure S11). They had the highest emissions of carbonyls (Figure 5) and were often observed to emit visible smoke (we noted the presence of smoke during field sampling but did not quantify particulate emissions).

Figures 6 and 7 show the distribution of NO_X and VOC emissions, respectively, from our measurements and those of natural gas-fueled pumpjack engines in the latest version of the Utah Division of Air Quality’s 2017 oil and gas emissions inventory (version 1.89, which was released in April 2021) (UDAQ, 2021). On average, measured NO_X emissions were only 9% of values listed in the inventory (median of 2%), and measured VOC emissions were 15 times higher than values listed in the inventory (median of 10 times). Measured formaldehyde and CO emissions from pumpjack engines were also higher than inventoried values (formaldehyde emissions are listed explicitly in the inventory). Average measured formaldehyde emissions were 6 times the inventory average (Figure S12; median measured was two times the median inventory), and average measured CO emissions were 9 times the average of all inventory values (Figure S13; median measured was 2 times the median inventory).

Emissions from engines account for 58% of total NO_X and 2% of total VOC emissions in the Utah 2017 inventory. To determine how the results of this study would impact these numbers, we changed emission values for all engines designated as pumping units in the inventory. We did not change emission values for other types of engines. We reduced NO_X emissions from pumping unit engines to 9% of their original values, and we increased VOC emissions to 15 times their original values. After these changes, engines accounted for 37% of NO_X and 15% of VOC emissions. These changes reduced total basin-wide emissions of NO_X from oil and gas sources by 33% and increased total VOC emissions by 16%. Table S4 shows emissions of NO_X and VOC in the inventory before and after these changes.

Estimates of NO_X emissions by Gorchov Negron et al. (2018) were also higher than the results of this study. They used a method based on fuel use to estimate that 42% of NO_X in the Uinta Basin is emitted from pumpjack engines and that more than 65% of NO_X is from all engine sources. Their estimate of pumpjack engine emissions is from the EPA Oil and Gas Tool (EPA, 2015), which assumes an average of 21 g of NO_X are emitted per kg of CO₂ (see figure S3 in Gorchov Negron et al.). In this study, only 1.0 (0.6, 2.1) g of NO_X were emitted per kg of CO₂ (excluding 4-stroke engines).

Figure S14 compares ratios of CH₄ to CO₂ and CO to CO₂ measured in exhaust gas samples in this study with regression slopes from ambient air measurements collected by Pétron et al. (2020) downwind of well pads with pumpjack engines in the San Juan Basin. The sample size from Pétron et al. is limited (n = 6), and their ambient air measurements can be expected to include emissions from nonengine sources at well pads, but results from the two studies are within the same range, providing evidence that the findings in this study are applicable beyond the Uinta Basin. Pétron et al. showed that the CH₄ to CO₂ and CO to CO₂ ratios they measured in ambient air downwind of engines were much higher than those assumed in an official emissions inventory for the San Juan Basin.

Ahmadov et al. (2015), Matichuk et al. (2017), and Tran et al. (2018) found that 3-dimensional photochemical models of wintertime ozone in the Uinta Basin failed to produce realistic levels of ozone unless organic compound emissions were increased above inventoried levels. More recently, several emission source types have been found to emit more organics than was estimated in the inventories used in those studies (Robertson et al., 2017; Lyman and Mansfield, 2018; Lyman et al., 2018; Mansfield et al., 2018; Wilson et al., 2020), and this study shows that pumpjack engines are another underestimated source of organics. Ahmadov et al., but not Matichuk et al., found NO_X emissions in the Uinta Basin to be overestimated in inventories. This study shows that actual NO_X emissions from pumpjack engines are much lower than inventoried values, but quantification of other stationary engines and oilfield combustion sources is needed.

Engine emissions values in inventories typically come from manufacturer specifications or emission factors, such as those provided in EPA’s AP-42 emission factor compilation (EPA, 1995; Alpha-Gamma, 2000), rather than from measurements of engines operating in real field conditions. This study is inadequate to definitively determine why such large differences exist between field measurements and inventory emission factors, though several possibilities exist. Organics emissions were highest, and NO_X emissions were lowest, for engines with high air–fuel equivalence ratios. Emission factors are typically derived from new engines in laboratory conditions, and it is probable that new engines run at lower equivalence ratios than engines that have been in continuous use for many years. The 2-stroke engines we encountered during the study had only a few analog controls, and they generally had no sensors to indicate engine performance, so it would be difficult for a mechanic to quantitatively assess whether an engine is operating within manufacturer specifications. Additionally, worn seals, suboptimal spark timing, or other mechanical problems in older engines could increase equivalence ratios and fuel slip and decrease NO_X.

Another possible explanation for the discrepancy between the measurements in this study and available emissions inventories is that inventories assume pumpjack engines operate at full load, while actual loads experienced by engines in field conditions are lower. At lower loads, emissions of methane (Varde et al., 1995), total hydrocarbons (Jensen et al., 2021), formaldehyde (Olsen and Willson, 2002), and CO (Brown, 2017) increase, while NO_X emissions decrease (Brown, 2017). Further, the relationship between NO_X emissions and load is nonlinear for 2-stroke natural gas-fueled engines (Varde et al., 1995; Griffin and Jacobs, 2015; Brown, 2017), such that NO_X emissions at 80% of the manufacturer-rated load are only about 10% of NO_X emissions at 100% load (Griffin, 2015). One of the companies that participated in the study provided us with load data they calculated from well-site pump controller output for some engines in the study. Those engines were found to be operating at 73 (63, 86)% (median of 62%) of the manufacturer-rated load. We did not find any statistically significant relationships between calculated load and any other measured parameter for this subset of engines. This could be because differences due to engine type, equivalence ratio, or other properties obscured the influence of engine load. Also, the calculated load provided by the company did not take into account the counterweights used to aid in lifting liquids from the well, so actual load experienced by engines was likely lower than calculated load.

Also, we measured emissions from 5 engines that were not being used to power pumpjacks. At these low-producing wells, the operator periodically disengaged the pumpjack for a period of days or weeks to allow pressure to build in the reservoir, after which the pumpjack was reengaged to continue producing oil. The operator kept the engine running to power other operations at the well even when the pumpjack was disengaged. NO_X emissions were lower from these partially idled engines than from equivalent engines that were powering pumpjacks (1.8 [0.7, 2.7] vs. 19.1 [2.6, 62.5] g h⁻¹; equivalent engines had the same distribution of engine types and were operated by the same company). Total organics emissions from the two groups were within the same range, however (1,153 [431, 2,399] vs. 2,904 [323, 10,264]).

We also investigated changes in emissions as the load experienced by the engine changed during each pumping cycle (Figure 8). We did not measure pumping rod position for this study, but we observed during field measurements that NO_X increased as the rod rose (which would lead to an increase in engine load) and decreased as the rod fell (which would lead to a decrease in engine load). O₂ and NO_X were anticorrelated over the pumping cycle (r² = 0.67 for the data in Figure 8, with a 2-s lag). NO_X and CO both increased at the same time in the pumping cycle, though CO was poorly correlated with O₂ (r² = 0.11, with a 2-s lag). CH₄ was positively correlated with O₂ (r² = 0.52, with a 4-s lag). Thus, the peak of engine load during each pumping cycle was associated with higher NO_X and CO, lower CH₄, and lower equivalence ratios (as evidenced by lower O₂), as expected (Varde et al., 1995; Kim and Bae, 2000).

Two-stroke pumpjack engines in this study emitted less NO_X and more organics than has been assumed in emissions inventories. The reason for this discrepancy is not known with certainty. NO_X emissions were lower and organics emissions were higher for higher air–fuel equivalence ratios, and higher equivalence ratios are associated with lower engine load. Relatively low load, possibly combined with engine deterioration over time, could be causing engines operating in real field conditions to perform much differently than those in the ideal conditions used for emission factor development. Accounting for actual engine emissions would change established inventory values substantially.

The organic compound composition of engine emissions in this study was also substantially different from the composition in a natural gas-fueled 2-stroke engine profile from the EPA SPECIATE database. Application of region-specific composition profiles may have a significant impact on photochemical model performance.

Fuel slip was very high for many of the engines measured in this study. Reducing equivalence ratios would decrease fuel slip, but it would also increase emissions of NO_X, CO, alkenes, and carbonyls. Photochemical modeling is needed to determine whether lower equivalence ratios would be beneficial for air quality in the Uinta Basin or elsewhere. Switching to 4-stroke engines or switching from Ajax to Arrow engines would have the same general effect as reducing equivalence ratios in existing engines.

This study adds to previous work showing that emissions of organic compounds from a variety of oil and gas equipment and processes have been underestimated and that NO_X emissions from oil and gas sources have been overestimated. This and future work to characterize component-level and process-level emissions of speciated organics and NO_X will improve the ability of industry and regulators to effectively and efficiently reduce the air quality impacts of oil and gas development and will also improve the ability of photochemical models to simulate those impacts (Ahmadov et al., 2015; Matichuk et al., 2017; Tran et al., 2018). In particular, measurements of emissions from other combustion sources, including compressor engines, heaters, combustors, and flares, are needed.

An anonymized data set containing engine properties, emission rates, exhaust mixing ratios, and fuel gas composition is publicly available at the DOI listed.

Lyman, S. 2022. Air pollutant emissions from natural gas-fueled pumpjack engines in the Uinta Basin [dataset]. Utah State University. DOI: http://dx.doi.org/10.26078/H62D-3F89.

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

Additional information, Figures S1–S14, and Tables S1–S4 are available as a supplemental PDF file.

Three energy companies allowed our team to access their facilities and provided information used in the study. Student Makenzie Holmes processed and analyzed most of the whole-air canister samples. Students Brant Holmes and Tyler Elgiar participated in field measurements.

This work was funded by the Utah Division of Air Quality, the Utah Legislature, and the Uintah Impact Mitigation Special Service District. Chevron Corporation provided funding for a liquid chromatograph to analyze samples for carbonyl mixing ratios in 2020, and we made extensive use of that instrument for the measurements described here. An endowment from Anadarko Petroleum Corporation funded the wages of students who participated in the project.

Direct funding for this work came from state and local government entities. An analytical instrument used in this work and an endowment for student wages were provided by oil and gas companies (see Funding). Over the past decade, 4% of research funding at the Bingham Research Center has been from private companies in the oil and gas, electric power, and mining sectors.

Substantial contributions to conception and design: SNL, HNQT, TLO, MLM.

Acquisition of data: SNL, HNQT, TLO.

Analysis and interpretation of data: SNL, MLM.

Drafting the article or revising it critically for important intellectual content: SNL, HNQT, TLO, MLM.

Final approval of the version to be published: SNL, HNQT, TLO, MLM.

How to cite this article: Lyman, SN, Tran, HNQ, O’Neil, TL, Mansfield, ML. 2022. Low NO_X and high organic compound emissions from oilfield pumpjack engines. Elementa: Science of the Anthropocene 10(1). DOI: https://doi.org/10.1525/elementa.2022.00064

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

Knowledge Domain: Atmospheric Science

Part of an Elementa Special Forum: Oil and Natural Gas Development: Air Quality, Climate Science, and Policy