Methane emissions from US low production oil and natural gas well sites

Eighty percent of US oil and natural gas (O&G) production sites are low production well sites, with average site-level production ≤15 barrels of oil equivalent per day and producing only 6% of the nation’s O&G output in 2019. Here, we integrate national site-level O&G production data and previously reported site-level CH4 measurement data (n = 240) and find that low production well sites are a disproportionately large source of US O&G well site CH4 emissions, emitting more than 4 (95% confidence interval: 3—6) teragrams, 50% more than the total CH4 emissions from the Permian Basin, one of the world’s largest O&G producing regions. We estimate low production well sites represent roughly half (37—75%) of all O&G well site CH4 emissions, and a production-normalized CH4 loss rate of more than 10%—a factor of 6—12 times higher than the mean CH4 loss rate of 1.5% for all O&G well sites in the US. Our work suggests that achieving significant reductions in O&G CH4 emissions will require mitigation of emissions from low production well sites.


1
We use the monthly O&G well-level and production data available from Enverus Prism 1 , a 2 commercial platform which collects and aggregates public and proprietary O&G data for wells in 3 the US For the year 2019, we aggregated the production data to annual production metrics for 4 2019 based on each well's unique well ID. The summary statistics for this initial data 5 aggregation are shown in Supplementary Table 1  Ninety four percent of all the wells in the database had data on the operator-reported total 10 number of production days. Among the wells with non-zero production days in 2019, 666,712 11 wells reported total production days of over 300 days in 2019 ( Supplementary Fig. 1). The Next, we filtered the aggregated dataset for specific well types for which we did not have methan 25 e emissions data. These are based on Enverus Prism's-derived well types and included cola-bed 26 methane wells, water wells, brine wells, CO2 injection wells, and service, storage and disposal w 27

ells. 28 29
This resulted into a total of 849,611 wells. We then filtered the dataset for wells located on only 30 onshore locations, using the shapefile data for US states available from the US Census Bureau 3 . 31 As part of this assessment, we excluded wells that were identified in Enverus Prism to be in 32 offshore counties or basins while located within the state boundaries. 33 The summary statistics for the final dataset are shown in Supplementary Table 1.

58
The clustering of spatial well-level data to site-level information requires an appropriate choice 59 of buffer radius. To test for sensitivity to buffer radii, we selected 10,000-20,000 wells in each for O&G production and performed geospatial clustering analysis on the well-level locations at 62 various buffer radii ranging from 10 m to 100 m, treating vertically-and directionally-drilled 63 wells (including those with unknown drilling trajectory) separately from horizontally-drilled 64 wells. As buffer radius is increased, more wells are aggregated to form well sites, resulting into 65 fewer well sites. An optimal buffer radius can be assessed based on the successive percentage 66 changes in the number of well sites as buffer radii is increased: at optimal buffer radius, a 67 minimum percentage change, or a leveling off in the percentage change in the number of well 68 sites, is observed ( Supplementary Fig. 3).  horizontally-drilled well sites, respectively. For each well with known location (lat/lon), we 90 applied an appropriate coordinate reference system based on the European Petroleum Survey 91 Group's (EPSG) codes for the specific UTM Zone for each location. A 25-m or 50-m buffer was 92 then generated around each well location, depending on well drilling trajectory, and overlapping 93 buffers were spatially merged. For multi-well sites, each well's production attributes were 94 summed to yield site-level attributes. Supplementary Fig. 4 shows examples of single-well and 95 multi-well sites for low production well sites as seen in satellite imagery (imagery basemap: 96 Google Maps). 97 Using this approach, the total number of active onshore well sites, defined as a well site with 98 total combined O&G production > 0 boed/site, was 700,000 (3 significant figures) well sites in

104
The top row shows single-well sites; the bottom row shows multi-well sites. In all cases, the red circles 105 represent a 25-m radius buffer around each wellhead's location. For multi-well sites, the overlapping 106 buffers are merged and site-level O&G production obtained by summing the production data for all the 107 wells on the site. Imagery basemap from Google Maps.

128
There are uncertainties in the estimates for the total number of well sites and their production 129 that are difficult to quantify based on available data. These include (i) uncertainty in location 130 data for wells and (ii) inaccuracies in the operator-reported data. In addition, some states, e.g., 131 Texas, require operators to report production data at the lease-level, not at the well-level. Data    Fig. 7 shows the distribution of low production well site count for the major US 174 O&G producing states, indicating that Texas and the Appalachian states of Pennsylvania, West 175 Virginia and Ohio dominate the total number of low production well sites nationally. 176 Additionally, the distribution for site age indicates a bimodal distribution ( Supplementary Fig.  177 7b), with peaks around 12 years old and 30 years old. Finally, as shown in Supplementary Fig.  178 7c, the majority of low production well sites (57%) produce <2 boed/site.

188
Analysis of data from Enverus Prism 1 .

189
To simplify our analysis, we focused on those wells that we classified as single-well low 190 production well sites that were operational in 2019, tracing their monthly production history 191 forward from the first reported production date. This resulted into a total of 44,649 low 192 production well sites that were actively producing in 2019 and had their first reported 193 production date between 2012 and 2019. 194 If the first production date is assigned as month zero for each well, it is possible to track their 195 monthly production in subsequent months and generate a production decline curve based on 196 reported data. Grouping the data by month from first month of production to last reported 197 month (in 2019), we compute the average boed for each month for all sites using Supplementary 198 Eq. 1. Supplementary Fig. 9 shows the average production decline curve for the 44,649 single-199 well sites (84% vertically-drilled, 16% horizontally-drilled) that were low production sites in 200 2019 and had their first reported production date between 2012 and 2019. For these sites, site-201 level O&G production fell to below 15 boed/site generally within the first one to two years (for 202 vertically-drilled wells) and two to five years (for horizontally-drilled wells) following initial 203 production ( Supplementary Fig. 9). We reviewed data from previously reported site-level measurement studies and selected study 220 data based on the following criteria (see Main text): 221 (i) The measurements were focused on quantifying total site-level CH 4

emissions, 222
(ii) Measurements captured both low and high-emitting sites, and 223 (iii) Both oil and gas production data were reported for each site where they could be 224 obtained (e.g., based on proprietary data, state-level reports or other reported attributes 225 such as location of the measured site and date of measurement). We were not able to use 226 data from site-level studies that did not report, or for which it was difficult to obtain, 227 both oil and gas production data, as these production data were needed to assess the 228 production category of sampled sites. 229 The summary statistics for O&G CH 4

239
(2020). 10 The new data for the 10 low production well sites are included in Supplementary Data 1 ). The study's method provides sitelevel quantification with a detection limit of 0.01 g/s (or 0.036 kg/h). A total of 80 sites were low-producing during the month of measurement, with combined O&G production per site ranging from 0.6 to 14.7 boed. For these sites, the reported site-level methane emissions ranged from 0.05 to 28 kg CH4/h. The study reported that "Methane emissions were positively correlated with gas production, but only approximately 10% of the variation in emission rates was explained by variation in production levels. The weak correlation between emission and production rates may indicate that maintenancerelated stochastic variables and design of production and control equipment are factors determining emissions." Low production well site data used in this study? Yes Omara et al. used the tracer flux measurement approach to quantify site-level methane emissions at 18 low production well sites in SW Pennsylvania and N West Virginia. The sites had gas production ranging from 0.68 to 44 Mcfd and combined O&G production in the range of 0.12 to 7.8 boed. The reported site-level methane emissions ranged from 0.02 to 4.5 kg/h/site. This study also quantified methane emissions at 17 unconventional gas well pads in the Marcellus and found significant differences between the emissions from low production and non-low production well sites. The authors observed that "these differences were attributed, in part, to the large size (based on number of wells and ancillary NG production equipment) and the significantly higher production rate of UNG sites. However, C v NG sites generally had much higher production-normalized CH 4   Measurements in the UGR, DJB and Uinta were performed on public roads while measurements in Fayetteville were performed either on public roads or on site access roads because the authors had site access granted by an anonymous operator. The authors reported that for well sites in the Fayetteville, Uinta and DJB, 20% of sites accounted for 72-83% of cumulative methane emissions, indicating the presence of high emitting sites. The authors also reported that average methane mass emission rates per well pad were similar among different basins despite large differences in average gas production, with the exception of Fayetteville which was dominated by dry gas production. The authors excluded any sites where an operator was visibly present and maintaining equipment, but reported that "episodic events (e.g., flash emissions, automated liquid unloadings) and failed components (e.g., thief hatch stuck open, malfunctioning pressure relief valves) may have been captured during a measurement." From this study, 39 sites were low production (i.e. <15 boed) and combined O&G production in the range of 1.2 to 14.6 boed. The measured methane emission rates for low production well sites ranged from 0.02 to 47 kg CH 4 /h/site.

Omara et al. (2018) 7
This study included measurements at 92 O&G well pads in Uinta, Denver-Julesburg and Marcellus (NE PA) regions. Methane measurements were performed using a combination of measurement techniques, including the dual tracer flux, OTM-33A and mobile transects followed by Gaussian dispersion modeling (GDM). 17 sites measured in the Uinta and DJB were low production sites. Their combined O&G production ranged from 0.2 to 14.6 boed. The measured CH 4 emission rate ranged from 0.06 to 9.3 kg CH4/h. The study used mobile transects with Gaussian dispersion modeling techniques for quantifying methane emissions. They found that the distributions were -extremely‖ skewed, with the top 10% of emitters contributing 77% of total methane emissions. For all sites, the authors reported a methane loss rate of 0.53% with a 95% CI of 0.46 to 0.64%. We identified 61 sites that were low production sites in this dataset, filtering for only actively producing sites at the time of measurement. Using the reported site location, we also reviewed available data from Enverus Prism for the selected low production well sites to confirm that they were indeed low-producing at the time of measurement. Their reported gas production ranged from 6 to 90 Mcfd and combined O&G production ranged from 1 to 15 boed. The reported methane emissions ranged from 0 to 136 kg CH 4 /h. The study also reported that a comparison of methane emissions across production classes, operator sizes, well status and regions produced "almost no significant" differences.

Low production well site data used in this study? Yes
Riddick et al. (2019) 10 Riddick et al. measured CH 4 emissions from abandoned and active conventional wells in West Virginia using the dynamic flux chamber and Gaussian dispersion modeling approaches. The study's method focused on quantification of wellhead methane emissions. The authors reported wellhead CH 4 leakage rates for 49 low production well sites, with gas production in the range of 0.02 to 24 Mcfd and combined O&G production in the range of ~0 to 4 boed. The reported CH 4 emissions ranged from ~0 to 3.2 kg/h. The authors reported that their measured methane emission factor for active conventional wells in this state (0.138 kg/h) was a factor of 7.5 times higher than EPA's EF for these wells. Overall, they found that wellhead CH 4 emissions from active conventional wells in this state represent loss rates of 8.8% of their CH 4 production, on average.
Low production well site data used in this study? Cited in the discussion, not used in the roll-up for national estimates because it focused only on wellhead emissions.

Deighton et al. (2020) 11
This study collected methane emissions from 43 low production well sites that produce <1 boed, focusing only on wellhead CH 4 emissions. The authors reported that the average wellhead CH 4 emission rate was 0.128 kg/h (median: 0.018 kg/h; range: 0 -0.91 kg/h). In addition, they found that the emissions were not episodic and that some wells were emitting all or more of the gas they produced. The study did not measure emissions from tanks and the authors acknowledge that their results may be conservative. The authors attribute their observations to maintenance issues, which were prevalent at most of the sites. For example, "some sites visited were in a state of disrepair (e.g. rusty well shafts, broken valves, fallen trees)". The authors proposed that "the main driver of emissions from the wells visited is neglect. The state of maintenance at these wells was poor. Often, pumpjacks, tanks, and other infrastructure were rusty and sometimes appeared to have temporary fixes to just keep the well mechanically operational." The authors calculate an average CH 4 loss rate of 21%. The authors estimate that oil and gas wells in this lowest production category emit approximately 11% of total annual CH 4 from oil and gas production in the EPA greenhouse gas inventory, although they produce about 0.2% of oil and 0.4% of gas in the US per year.  We follow the procedure in Zavala-Araiza et al. 12 and provide an alternative assessment of the 315 site-level CH 4 emission factors for low production well sites. We split the CH 4 emission rate data 316 into two cohorts representing the sites producing 2 boed (n = 41) and those producing 2 boed 317 (n = 199). Binning the data into these two cohorts was necessary as the empirical data suggest 318 lower CH 4 emissions for the 2 boed cohort (Supplementary Fig. 15). In both cases, we assume 319 the underlying distributions are lognormal, and verify this assumption using both the Lilliefors 320 and Shapiro-Wilk statistical tests for normality, performed on the log-transformed data for each 321 cohort of sites. The null hypothesis for both tests is that the values are drawn from a normal 322 distribution (unknown parameters μ and ), with critical p-value established at 0.05. 323 The results of the normality tests are shown in Supplementary Table 3, indicating that one 324 cannot reject the null hypothesis that the site-level data for both cohort of sites arise from a 325 lognormal population distribution. We therefore use the lognormal distribution as a reasonable 326 assumption for the site-level CH 4 data. 327 We fit the site-level CH 4 emissions data for each cohort of sites to a lognormal distribution, 328 deriving the mean (μ) and standard deviation ( parameters using the maximum likelihood 329 estimation approach, 12 treating data for the below-detection-limit sites (as zeros or <0.036 kg 330 The CH 4 emission factor for each cohort of sites is then computed as: . Combining 334 these results with activity data for low production well sites results into a mean CH 4 estimate of 335 3.2 (95% CI: 0.8-18) kg CH 4 /h/site for CH 4 emissions from low production well sites, reflecting 336 uncertainty due to extrapolation of a small sample size to a large population of well sites. The 337 overall emission factors are higher, but have confidence intervals that overlap with the results 338 from our primary model, which more comprehensively assesses the distribution of emissions 339 relative to the emitter characteristics of the high-emitting sites (top 5% of sites), the bottom 95% 340 of sites with detectable emissions and the below-detection-limit sites.  methane emission rates at low production well sites. Permian Basin for large CH 4 emission sources. In total they found 3,067 plumes of methane 357 above an ~10-20 kg/hr detection limit and by use of repeated overflights aggregated their 358 detections into 1,756 unique large emission sources by clustering all plumes within a 150m 359 spatial buffer. Well-level production data 1 was spatially aggregated to ‗well sites' as previously than 15 barrels-of-oil equivalent per day (boed). All 62 source locations were manually reviewed 364 using satellite imagery to ensure the well site was isolated from emission sources on neighboring 365 well sites or other O&G facilities by ~200m and linkable to the production data with high 366 confidence. 367 Using the polygon outline of coverage area by the aircraft swath for the entire campaign, we 368 estimate the Cusworth et al. dataset observed >95,600 well sites, with >62,600 of them meeting 369 the ≤ 15 boed criterion to be defined as low production sites. When accounting for repeat 370 overflights conducted on different flight days, this amounts to >194,800 observations of 371 Permian well sites, >127,800 of which are low production. The massive extent of this campaign 372 with respect to well site coverage suggests the rarity of observing emission plumes from low 373 production well sites in the Permian Basin above the ~10-20 kg/hr detection limit is on the 374 order of 0.05%, or one out of every 2,000 site visits. 375 The measured CH 4 emission rates in this sample are anomalously high ( Supplementary Fig. 17), 376 and may be related to site-specific safety control measures, for example, venting from pressure 377 relief valves due to upstream overpressure conditions at tanks or separators. We note that these 378 extreme emissions were highly impersistent: of the sites that were flown over on more than one The 240 site-level methane emissions data are based on previous measurements conducted in 405 seven major US oil and gas basins ( Supplementary Fig. 18). These basins are diverse in their 406 characteristics, including gas-dominant basins (e.g., Appalachian) and oil-dominant basins (e.g., 407 Delaware and Denver-Julesburg), with average gas-to-oil ratios that range from 4 Mcf/barrel to 408 88 Mcf/barrel. Low production well sites in basins or regions where site-level measurement data 409 were collected (see Supplementary Fig. 18) account for 42% of all US low production well sites 410 and ~20% of total US O&G production from low production well sites. Related to the broad spatial coverage, the production characteristics of the sampled sites are 422 broadly representative of the production distribution for all low production well sites within the 423 measured basins or regions and for all low production well sites nationally ( Supplementary Fig.  424   19).    While our model does not directly resolve operator-specific CH 4 emissions-because the input 517 variables are exclusive of operator information due to limited data-it is nevertheless possible to 518 derive further insights based solely on the distribution of site count, production characteristics, 519 and related marginal well site CH 4 emission profile. More than three-quarters of low production 520 well sites and four-fifths of their O&G production are owned by a small fraction of operators 521 with 100 well sites each. Treating these operators' low production well sites as a single cohort 522 of sites and applying the same emissions modeling scheme as before, we estimate that, in 523 aggregate, sites owned by midsize to large operators (>100 well sites each) dominate CH 4 524 emissions, accounting for 80% (95% CI: 62-100%) of the total.  Texas regulations apply to new sources, relative to 2000, 2011, or 2012 depending on location and type of permit. Texas requires a leak detection and repair ("LDAR") program for certain mid-sized to large oil and gas facilities. The specific requirements vary depending on the facility's location and potential to emit uncontrolled volatile organic compounds ("VOC"). Most well sites are not subject to LDAR due to the high emissions threshold uncontrolled VOC emissions (>10 or 25 tpy) and distance from a sensitive receptor, such as a home or school, that triggers the application of LDAR. 20