Methane, carbon dioxide, hydrogen sulfide, and isotopic ratios of methane observations from the Permian Basin tower network

. We describe the instrumentation, calibration, and uncertainty of the network of ground-based, in-situ, cavity ring down spectroscopy (CRDS) greenhouse gas (GHG) measurements deployed in the Permian basin. The primary goal of the 10 network is to be used in conjunction with atmospheric transport modeling to determine methane emissions of the Delaware sub-basin of the Permian Basin oil and natural gas extraction area in Texas and New Mexico. Four of the measurements are based on tall communications towers, while one is on a building on a mountain ridge, with the recent addition of a small tower at that site. Although methane (CH 4 ) is the primary species of interest, carbon dioxide (CO 2 ), hydrogen sulfide (H 2 S), and the isotopic ratio of methane ( δ 13 CH 4 ) are also reported for a subset of the sites. Measurements were reported following 15 WMO X2004A scale for CH 4 , and the WMO X2019 scale for CO 2 . CRDS instruments were calibrated for CH 4 and CO 2 in laboratory prior to deployment. For H 2 S, data was offset-corrected using the minimum 40-min running mean value of the day, and for δ 13 CH 4 , calibrations were based on laboratory data. We describe the characteristics of the data set with a set of illustrative analyses. Methane and carbon dioxide showed strong seasonality, with a well-defined diurnal cycle during the summer, which was opposed to the winter, when a diurnal cycle was absent. CH 4 enhancements to the background, during 20 the winter, are up to twice the summer values, which is attributed to the changes in boundary layer depth and wind speed. The largest CH 4 enhancements occurred when winds


Introduction
Emissions of methane (CH 4 ), such as from oil and natural gas producing regions, are an environmental concern since CH 4 is 30 a greenhouse gas with a global warming potential 28-36 times larger than that of carbon dioxide (CO 2 ) over a 100-year period, and 80 times larger than CO 2 over a 20-year period (IPCC, 2021). This large difference in radiative forcing is a result of the relatively short atmospheric lifetime of CH 4 (~ 10 years) compared to CO 2 (lifetime of ~ hundreds or thousands of years, IPCC, 2021). As a result of its strong short-term impact, reductions in CH 4 emissions are an efficient way to quickly reduce radiative forcing. According to Ocko et al (2021), full implementation of all methane abatement technologies that are 35 already technically feasible could cut anticipated global methane emissions in 2030 by 57%, and global-mean average methane warming rates between 2030 and 2050 would consequently be reduced by 26%.
About 60 % of the global CH4 budget arises from anthropogenic emissions (Saunois et al., 2020). Some of these emissions are fairly well-known since they are large point sources amenable to direct emissions monitoring (e.g., coal mines) 40 (Kirchgessner et al., 2000). Some are relatively diffuse, large-area, low-intensity sources such as agricultural activities (Carlson et al., 2017;Moraes et al., 2013) and leaks inside homes and buildings (Wennberg et al., 2012). A large component of anthropogenic emissions, however, comes from relatively compact, high-intensity, regional sources such as oil and gas (O&G) production basins (Alvarez et al., 2018;Maasakkers et al., 2019;Pandey et al., 2019;Schwietzke et al., 2016). O&G producing regions include numerous point sources covering a wide range of expected emissions (e.g., well pads and 45 processing plants) and more diffuse sources as gathering pipelines, sometimes intermingled with other methane sources such as livestock. Thus, O&G emissions are large in magnitude, and often highly uncertain. In addition, O&G production in the United States has increased dramatically since around 2005, driven primarily by hydraulic fracturing and horizontal drilling (Alvarez et al., 2012). This expansion of O&G production has prompted an increasing interest in monitoring of methane emissions from O&G basins on regional scales for the purpose of possible regulations and commercial incentives for 50 operators to prove low emissions.
Atmospheric CH4 measurements can reduce uncertainty in CH 4 emissions by providing "top-down" assessments of emissions. Top-down estimates are based on empirical data and atmospheric scientific methods, opposed to "bottom-up", which uses an inventory approach and extrapolate regional emissions from smaller spatial scale measurement data as a 55 component. The top-down emissions estimates can be compared with and used to improve more traditional accountingbased, or inventory methods (e.g., Maasakkers et al., 2016;U.S. EPA, 2019). Emissions of CH 4 in the U.S. derived from atmospheric data have differed significantly from inventory assessments (Alvarez et al., 2018;Turner et al., 2015;Barkley et al., 2019b;Barkley et al., 2021;Zhang et al., 2020), showing the importance of such independent data. Many past atmospheric studies of CH 4 have used aircraft data to quantify emissions from O&G production basins (Baray et al., 2018;60 Barkley et al., 2017;2019a;2019b;Karion et al., 2015;Schwietzke et al., 2017) and cities (Cambaliza et al., 2014;Conley et al., 2016;Heimburger et al., 2017;Plant et al., 2019). Automobile-based measurements (e.g., Caulton et al., 2019;Omara et al., 2016;Robertson et al., 2020) have also been used to great advantage to characterize emissions from individual O&G production sites. Aircraft and automobile measurements, which are spatially rich in information content, are typically shortterm in nature. Even extended airborne campaigns (e.g., Heimburger et al., 2017;Barkley et al., 2021) have limited 65 availability to date to capture temporal trends in basin-or city-scale emissions.
Observations of atmospheric methane are sparse, with limited in-situ sites located mostly in North America and Europe (e.g., Andrews et al., 2014;Karion et al., 2015). Satellite-based measurements such as the Greenhouse Gas Observing Satellite (GOSAT) and the TROPOspheric Monitoring Instrument (TROPOMI) data provide global, remote sensing of column 70 atmospheric methane (XCH4; e.g., Qu et al., 2021), and will strengthen our understanding of the global methane budget, but quantification of the level of bias and uncertainty in methane emissions estimates as a function of conditions known to affect satellite retrievals is critical. Tower-and building-based in-situ networks measuring CH 4 dry mole fractions within the boundary layer have been used to quantify urban emissions (e.g., Lamb et al., 2016;McKain et al., 2015;Yadav et al., 2019;Sargent et al., 2021). Chan et al. (2020) used in-situ monitoring to quantify methane emissions for the Western Canadian 75 Sedimentary O&G Basin and Lin et al. (2021) quantified methane emissions from the Uinta O&G Basin. Here, we present observations from a tower-based atmospheric monitoring network designed to track CH 4 emissions from the Delaware subbasin of the Permian Basin.
The Permian Basin is the largest-producing region of oil in the United States, and the second largest-producing region of 80 natural gas, accounting for 40% of the U.S. oil production and 15 % of the U.S. natural gas production (Enverus Drillinginfo, 2021). The Permian Basin is also a large emitting U.S. oil and natural gas producing region according to satellite observations (Zhang et al., 2020), with a natural gas productions normalized loss rate of 3.7 %. Emissions of methane are associated with midstream processing (e.g., Mitchell et al., 2015), flares (e.g., Allen et al., 2016), and with low-producing marginal wells (e.g., Deighton et al., 2020). 85 This tower-based network was deployed in late February 2020, just prior to the COVID-19 pandemic lockdown in the United States. The data reported here stops on 9 November 2021, but the network is still in operation. Using these tower observations alongside aircraft measurements, satellite observations, and model estimates, Lyon et al. (2021) found a correlation between decreasing oil prices and the CH4 emissions during the COVID-19 lockdown (March -April 2020), and 90 hypothesized that under normal conditions, production exceeds the midstream capacity, resulting in more venting/flaring, and consequently higher methane emissions.
The primary purpose of this paper is to describe the high-accuracy mole fraction measurements of CH 4 including the network and site characteristics (Section 2), instrumentation used for data collection (Section 3), as well as the associated 95 calibration, processing, uncertainties, and data coverage (Section 3). We also describe opportunistic measurements of CO 2 , hydrogen sulfide (H 2 S), and methane isotope ratio (δ 13 CH 4 ) which are of interest but are not the focus of this network.
Section 4 presents summary data for all gases (CH 4 , CO 2 , H 2 S, and δ 13 CH 4 ) to date, and an example analysis of each gas.
The example analyses include diurnal cycles and enhancements for CH 4 , CO 2 , and H 2 S, and the determination of source isotopic signature using δ 13 CH 4 and CH 4 measurements. We have included brief methods for these analyses in each 100 subsection, rather than a separate section for methods.

Network and sites characterization
The Permian in-situ tower observation network includes five monitoring stations in the Delaware sub-basin (Fig. 1a, Table   1 The towers encompass an area of approximately 160 km x 220 km. Of the monitoring stations, four are communications towers with gas inlets installed at 91 -134 m height above ground level (AGL). Due to the lack of availability of a tower, the instrumentation at the Carlsbad location was initially deployed on a rooftop (4 m AGL). On 13 May 2021, the instrumentation was moved to a 9 m tower approximately 250 m to the east of the original location. The Carlsbad site is at 1349 m above sea level on a mountain ridge, significantly higher than the surroundings' elevation (e.g., the elevation in 110 White City, 6 km to the east of Carlsbad tower, is 1112 m ASL). The location is within Carlsbad Caverns National Park, hence is buffered from oil and gas infrastructure immediately adjacent to the measurement site. At Hobbs, additional inlets were installed at 2 m AGL and inside the building at the base of the tower.  The prevailing wind direction in the region varies seasonally (Fig. 1b). During the winter (i.e., December, January, February 135 (DJF)), the wind most often comes from SW -W directions, while during the summer (i.e., June, July, August (JJA)), the prevailing direction is from the South. This behavior makes the measurements obtained from Carlsbad (during winter) and Fort Stockton (during summer) the most likely background sites since most of the time they are not directly impacted by CH 4 emissions originating from the oil and gas fields located within the Delaware sub-basin to the east and north of these towers, respectively. 140 New Mexico and Texas are within a region of extensive production of oil and natural gas, and the landscape surrounding the towers are mostly shrub/scrub, and grassland (Fig. 1c). Hobbs is the only site with significant agricultural and urban landcover. Within a 10 km radius of the Hobbs tower, the landscape is ~ 40 % shrub, scrub, and grassland, ~ 40 % cultivated crops (to the east of the tower), and ~ 20 % urban area (to the west of the tower). These relatively simple 145 surroundings in terms of methane emissions simplifies the task of isolating emissions from the basin's O&G infrastructure, which accounts for > 90 % of methane emissions in the Permian Basin (Maasakkers et al., 2016).

Instrumentation, sampling details, and calibration
Mole fraction measurements were made with wavelength-scanned cavity ring down spectroscopic (CRDS) instruments 150 (Picarro, Inc., models G2301, G2401, G2204, and G2132-i). The primary species of interest for this network was CH 4 .
Most of the instruments also reported CO 2 , one reported H 2 S and, at various locations and time periods, δ 13 CH 4 was reported.
Instrument failures necessitated multiple exchanges of instrumentation for repairs.
The in-situ sampling method was similar to the procedures described in Richardson et al. (2017) and the schematic for the 155 systems as deployed in the field is shown in Fig. 2. Collocated at the top sampling level of each tower were two 1/4 in (0.64 cm) OD Synflex 1300 (Eaton Corp.) tubes with rain shields to prevent liquid water from entering the sampling line. Air was drawn down from the inlet on the tower, through the Nafion dryer (MD Series, Permapure LLC, 24 in (61 cm) to 96 in (244 cm) lengths, depending on availability), into the CRDS instrument for analysis, and then used as the purge gas in the Nafion dryer (i.e., re-flux method). Field calibration tank gas was introduced upstream of the dryer, humidifying the calibration gas 160 via 1/8 in (0.32 cm) OD stainless steel tubing. Scott Specialty Gas (now Air Liquide) two-stage regulators (part number 51-14 A-590, similar to Air Liquide, part number Q1-14B-590) were used for sampling the field calibration tanks. 3-way solenoid valves (part number 091-0094-900, Parker Hannifin Corp.) were used to switch between sample and field calibration gas. For the Hobbs site, the top level was sampled for 40 min of each hour and the lower levels were each sampled for 10 min. We ignored 4 min after every transition for equilibration. Correction factors, based on factory default 165 H2O factors, were applied to adjust the CH 4 and CO 2 values for the effects of the remaining water vapor (Rella et al., 2013). The measurements are reported on the World Meteorological Organization (WMO) X2004A scale for CH 4 , and the WMO X2019 scale for CO 2 . H 2 S and δ 13 CH 4 (tied to the Vienna Pee Dee Belemnite (VPDB) scale) are reported, but with limited calibration. The G2301 and G2401 CRDS instruments (Picarro, Inc.) were calibrated for CH 4 and CO 2 in the laboratory prior to deployment using four NOAA (National Oceanic and Atmospheric Administration) tertiary standards, ranging between 1790 and 2350 ppb CH 4 , and 360 and 450 ppm CO 2, sampling each tank at least five minutes after equilibration, and 175 repeating at least once. We characterized the G2132-i and G2204 instruments using 2 -4 tanks and applied the calibration in the post-processing stage.
To calibrate the field tanks prior to deployment, we first calibrated a dedicated laboratory CRDS instrument as above. We then used the CRDS instrument to sample each field calibration tank and assigned the mean to each tank. Field calibration 180 tanks (size N80, expected to last 12 -18 months, and listed in Table 2) were sampled nominally every 23 hours for 6 min for G2301, G2401 and G2204 instruments, and every 6 hours for 20 min for G2132-i instruments, but adjustments were occasionally made to limit field calibration gas usage. After each transition between calibration gas and atmospheric sample, 4 min of data are ignored. An offset correction was applied daily. One disadvantage of this procedure is the potential introduction of tank drift to the data, but tank drift for CH4 has not been observed (Andrews et al., 2014). Allan deviations 185 were < 0.2 ppb CH 4 and 0.02 ppm CO 2 for 2 min samples for the G2301 and G2401 instruments (Yver Kwok et al., 2015), and 0.1 ppb CH 4 and 0.1 ppm CO 2 for ~ one hour sample for the G2132-i instruments (Miles et al., 2018) and thus the noise for these instruments for the calibration cycles is insignificant. Ideally, more than one tank would be sampled at each site, but this was not practical for this network. Data from the first four minutes after a transition between gases was discarded.
Flow rates for G2301, G2401 and G2204 instruments were about 240 cc/min whereas the flow rates for the G2132-i 190 instruments were about 30 cc/min. Residence time from the top of the tower to measurement was 6 to 9 min for the G2301, G2401, and G2204 instruments and 45 to 70 min for the G2132-i. For the Carlsbad site sampling from a rooftop, then a 9 m tower, residence times were < 1 min. The data were adjusted to report the sampling time, rather than the measurement time, taking into account the residence times. We offset-corrected the H 2 S data using the minimum 40-min running mean value of the day (assumed to be zero) instead of 200 a field calibration tank because a tank was not available for an extended period of the deployment. For the period for which a field calibration was available, the mean was within 0.05 ppb of the lowest H 2 S throughout the day, which was small compared to the observed signals. The standard deviation of the instrument drift based on the field calibration tank was ±0.43 ppb throughout the period June -August 2021. A calibration tank with known non-zero H 2 S mole fraction was not available, so we were not able to assess the calibration slope. We did not apply any correction to the H 2 S for water vapor. 205

195
We applied δ 13 CH 4 calibrations based on laboratory data (Miles et al., 2018) and tested with a tank characterized by the Institute of Arctic and Alpine Research (INSTAAR) prior to deployment, but field calibration tanks with varying δ 13 CH 4 values were not available for this network. A daily offset-correction was applied to the δ 13 CH 4 , using a field-calibrated tank.
The uncertainty is estimated to be 1 ‰, compared with 0.15 ‰ uncertainty reported by Miles et al. (2018) when utilizing 210 multiple field calibration tanks.

Uncertainty
The uncertainty of the reported hourly values for CH 4 and CO 2 include contributions from measurement uncertainty, extrapolation, and water vapor (Andrews et al., 2014;Verhulst et al., 2017;Karion et al., 2020). The measurement uncertainty is composed of uncertainties attributable to short-term precision, calibration baseline and scale. We assessed the 215 effects of instrument short-term precision and drift between calibration cycles (i.e., calibration baseline) using a 31-day running standard deviation of the daily tank residuals, reporting the 2-σ value. Typical values vary with instrument type (e.g., 0.6 ppb CH 4 and 0.06 ppm CO 2 for G2301, 1.0 ppb CH 4 and 0.18 ppm CO 2 for G2132-i and 6.8 ppb CH 4 for G2204). Scale (i.e., tank assignment) uncertainty is set to 1.0 ppb CH 4 and 0.06 ppm CO 2 , following Andrews et al. (2014). Ideally, a target tank is used to independently assess the measurement uncertainty, but for this network only one tank was deployed at 220 each site.
Because we performed a full calibration of the instruments using four NOAA-calibrated tanks prior to deployment (and upon any factory repairs), extrapolation error is expected to be small and is not specifically reported here. Round-robin style tests have indicated that if full calibrations are performed at least every 2 years, in addition to a daily single-point adjustment, 225 differences from known values are within 1 ppb CH4 and 0.1 ppm CO 2 .
We assessed the uncertainty due to water vapor based on the difference in water vapor mole fraction between the dried sample and the humidified field calibration gas. The difference varied due to length of Nafion dryer, building temperature, and instrument flow rate, but was typically between 0.05 and 0.7 % H 2 O. No drying was employed at the Hobbs site for 230 March 2020 -May 2021, during which time the water vapor was up to 2 %. Errors in the coefficients used to determine the water vapor correction can vary by instrument and are the largest contributor for cases with moderate or no drying. Rella et al. (2013) showed that errors associated with the water vapor correction, even with no drying, are less than WMO compatibility goals for CO 2 and CH 4 (0.1 ppm CO 2 and 2 ppb CH 4 ) if instrument specific correction factors are determined periodically. The error associated with relying on general correction factors (as used here) is up to 0.25 ppm CO 2 and 2.0 ppb 235 CH 4 for 3 % water vapor. We have therefore assumed the uncertainty due to the water vapor correction to be a linear function between these values and no error at 0 % water vapor. The uncertainty due to the water vapor correction when drying was 0.00 -0.06 ppm CO 2 and 0.0 -0.6 ppb CH 4 . For the period without drying at the Hobbs site, the uncertainty was up to 0.17 ppm CO 2 and 1.7 ppb CH 4 .

240
The initial instrument at the Notrees site did not report water vapor for 1 March -27 July 2020 due to a laser problem. For that period, we applied a water correction and uncertainty based on the subsequent mean and standard deviation of the water vapor (0.53 % ± 0.19 %). The uncertainty in the water vapor value led to uncertainty in the CO 2 of ~ 1.1 ppm CO 2 and 5 ppb CH 4 for this period. For 28 July -31 December 2020, the instrument did report H 2 O but CH 4 and CO 2 uncertainty due to noise continued to be higher (2.3 ppb CH 4 and 0.58 ppm CO 2 ) than subsequent values for this instrument (0.4 ppb CH 4 and 245 0.04 ppm CO 2 ). The intra-network CH 4 differences across the Delaware Basin were 135 ppb (winter) and 51 ppb (summer), whereas the CO 2 differences were 0.8 ppm (winter) and 1.2 ppm (summer) (Section 4.3). Thus, the CH 4 uncertainty during this period was small compared to intra-network differences, but nearly the same magnitude for CO 2 . Consequently, we flagged CO 2 for 1 March -31 December 2020 as unsuitable for use and replaced the mole fractions with a placeholder value (NaN). The instrument was replaced on 15 May 2021. 250 The results for these contributions to the uncertainty, summed in quadrature, are shown in Fig. 3. The uncertainty (2-sigma) of the G2301 and G2132-i instruments when operating normally is better than 1.5 ppb CH 4 . For CO 2 , the uncertainty is better than 0.07 ppm for the G2301 instruments and 0.30 ppm for the G-2132i instruments. Note that the Hobbs instrument does not measure CO 2 and that the uncertainty for CH 4 is larger due to the instrument type (G2204). Manufacturer precision 255 specifications of the instrument model at this site indicate CH 4 precision for a 5-s sample of 2 ppb compared to the other instrument models used in this network (CH 4 precision for a 5-s sample of < 0.5 ppb).

265
H 2 S and δ 13 CH 4 are reported in this dataset but are not the focus of this research. Since a field calibration tank with non-zero H 2 S was not available, we reported the manufacturer-specified precision of 1 ppb + 0.4% H 2 S as an estimate of instrument uncertainty. The standard deviation of two months of quasi-daily field calibration data for the isotopic ratio of methane at the may be additional slope errors not captured by the single field calibration tank, and we have reported the uncertainty on the 270 isotope ratio as 1 ‰.

Data coverage
The data coverage of the hourly-averaged observations for each species for the Permian Basin tower network through the beginning of November 2021 is indicated in Fig. 4. For the period 9 June 2020 -16 May 2021, a leak near the inlet to the 275 instrument was identified at the Hobbs site. The field calibration tank emptied unusually quickly, and flow rates measured near the instrument inlet on a site visit indicated a leak. Although the leak is not apparent upon inspection of the data in isolation or via comparison with the other network sites, we have flagged this data as unsuitable for use and replaced with a placeholder value (NaN) in an abundance of caution. The original data is, however, available online.

Diurnal cycle and seasonality
Composited means of hourly CH 4 and CO 2 mole fractions (averaged over summer and winter seasons) indicate clear 290 seasonality (Fig. 5). The atmospheric boundary layer is typically deeper in the summer, which is consistent with the observed lower mole fractions of CH 4 and CO 2 , compared to the wintertime. Methane and carbon dioxide have a distinct diurnal cycle during the summer months, with the highest mole fractions between 10 and 15 UTC (night). In Texas the local time is UTC-5 and in New Mexico, UTC-6. The observed CH 4 mole fraction in the summer is lowest at Fort Stockton, which is consistent with predominantly southerly winds (Fig. 1b) and the majority of the oil and natural gas facilities being to the 295 northwest of this tower. During the winter months, when winds are predominantly from the SW -W direction, the lowest CH 4 is measured at Maljamar on the northern edge of the network for most hours of the day. The diurnal amplitude in CO 2 is 1 ppm during the winter, compared to 5 ppm during the summer.

305
Due to data availability, the seasonality of H 2 S cannot be assessed. Hourly composites of H 2 S mole fraction during the summer months show a clear diurnal cycle (Fig. 6). The concentrations at 91 m AGL and inside the building have similar diurnal cycle signature and magnitude. On the other hand, 2 m AGL presented the highest concentration, and is significantly different from the 91 m AGL observations. Filtration by the air conditioning system appears to have reduced the H 2 S mole 310 fractions inside the building. While background levels of H 2 S are essentially zero, the mean H 2 S mole fraction at direct exposure level for humans (2 m AGL) is on the order of 8 ppb around 12 UTC, which is 4 times larger than the levels observed inside the building and at 91 m AGL. These levels of H 2 S are not a health concern (OSHA, 2022), but do indicate the existence of an H 2 S source near the tower.

Methane, carbon dioxide, and hydrogen sulfide enhancements
Atmospheric inversion techniques, used to estimate gas emissions, rely on accurate quantification of enhancements, which 320 are defined as the difference between the tower network background which may be defined in a variety of ways (e.g., Karion et al., 2021;Sargent et al., 2018) and the mole fraction observed at each tower. Typically, enhancements are calculated for afternoon hours, and here we defined afternoon hours from 20 to 23 UTC. To determine the enhancements, we averaged the afternoon mole fractions at each tower, and obtained the background for CH 4 and CO 2 from the minimum averaged afternoon mole fraction of the entire network. Thus, each day has one value for the background, and each tower has 325 one enhancement value per day. For H 2 S, we assumed that the background is zero since this gas is not expected to be found naturally in this region. We used the wind data from Lea International Airport in Hobbs and excluded directions from which the wind originated for five or fewer days during the analysis period (e.g., summer and winter months). Calm winds (< 1.6 m/s) were also excluded from the analysis.

330
Enhancements of methane have strong seasonality, with smaller enhancements during the summer months (Fig. 7a) when compared to the enhancement during the winter months (Fig. 7b). During the summer, there is intense surface heating in the region, and deep boundary layer depths, compared to the winter months when more stable atmospheric conditions and lower atmospheric boundary layer depths occur. The largest enhancements of CH4 occur when the wind blows from the center of the Delaware sub-basin (Fig. 7), which is coincident with the high estimated emission rates of methane (Zhang et al., 2020). 335 The seasonality of methane observations is also apparent in the daily afternoon differences between the largest and smallest 345 CH 4 mole fraction measured from the tower network from 1 March 2020 to 1 August 2021 (Fig. 8). While in summer months the afternoon differences do not exceed 150 ppb, in the winter months the differences reached more than 900 ppb.
For the 30-day mean, the summer differences ranged from 50 to 100 ppb, and winter differences were twice the summer values, and are usually between 150 and 200 ppb. Note that Figure 5 is averaging across a 3-month period while Figures 7 and 8 are daily maximum and minimum differences. In the summer, FORT is the site with minimum afternoon CH 4 50% of 350 the days, but during the winter, the site exhibiting the minimum CH 4 is much less consistent, with CARL at 29%, FORT at 27%, and MALJ at 20% of the days. Thus, while the daily afternoon intra-network differences are large in the winter (Fig.   8), the mean difference between sites when averaged over 3 months is 40 ppb (Fig. 5b).

360
Even though the composite diurnal cycle of CO 2 mole fractions presented some seasonality ( Fig. 5c and 5d), the magnitude of CO 2 enhancements did not (Fig. 9). There were, however, changes in enhancements related to the seasonality of the wind direction (Fig. 9). We did not expect to observe significant enhancements of CO 2 coming from the O&G basin, but some interesting patterns emerged. The observations revealed that at Notrees tower the enhancements were larger than 3 ppm with winds coming predominantly from South (Fig. 9b) and South-Southwest directions (Fig. 9a). The Notrees enhancements also 365 corroborate the enhancements observed at Fort Stockton, which has the largest enhancements to SSW during the summer; and during the winter, Fort Stockton has enhancements coming from the North, pointing to a possible source that is between Notrees and Fort Stockton. As for Carlsbad and Maljamar, during the summer, the enhancements are more isotropic, from NE to SSW, and the largest enhancements from both towers come from the directions of the city of Carlsbad. During the winter, Carlsbad, particularly, has a strong enhancement coming from the east. 370 Only one of the network towers has measurements of H 2 S during the summer, as stated above, and thus we cannot verify the 380 seasonality of this dataset. However, the H 2 S enhancements obtained from the Hobbs tower at 91 m AGL, in the order of 2 ppb, indicate a potential source NE of the tower (Fig. 10). There is a patch of oil and gas wells 0.5 -1.5 km to the east and northeast of the tower that may be the source of H 2 S. The enhancements computed from the 2 m AGL inlet, not shown here, presented the same pattern as the 91 m AGL, and are, on average, 0.05 ppb larger than the enhancements at the top level.
During the summer, the enhancements obtained from the inlet temporarily installed inside the building, on average, did not 385 exceed 0.2 ppb, coming from the southeast.

Isotopic ratio source signature
We used the Keeling plot approach (Keeling, 1961;Röckmann et al., 2016;Miles et al., 2018), determining the intercept of 395 the best-fit line of the isotopic ratio as a function of the inverse methane mole fraction, to estimate the isotopic ratio of the methane source at the Maljamar tower. The intercepts of the best-fit lines for the peaks (Fig. 11) indicate that the sources contributing to the peaks have a mean isotopic ratio of −40.8 ± 0.5 ‰. Oil and natural gas extraction is the only significant source of local methane in this region (Maasakkers et al., 2016). The methane is lighter than that observed in the Marcellus region (−31.2 ‰; Miles et al., 2018), and similar to that observed in the Barnett region (−41.8 ‰;Milkov et al., 2020). The 400 correlation coefficients were lower than that observed via a similar tower-based method in the Marcellus (Miles et al., 2018).

Data availability
The data is available at The Pennsylvania State University Data Commons under DOI https://doi.org/10.26208/98y5-t941 (Monteiro et al., 2021). We plan to update the data repository annually.

410
The dataset is organized by yearly data files and named by the host institution identification (PSU), the project identification (PERMIAN), type of measurements (INSITU), tower identification (e.g., CARLSBAD, HOBBS, FORTSTOCKTON, MALJAMAR, NOTREES), and year. The earliest files start on 01 March 2021.
In the datasets, the columns include location code, instrument serial number, inlet height (m AGL), minimum time included 415 in the hourly average (decimal day of year, UTC), maximum time included in the hourly average (decimal day of year, UTC), year, day of year, hour (UTC), calibrated CO2 (dry mole fraction, ppm), standard deviation of the raw (2-3 s) CO 2 data within the hour (ppm), estimated CO 2 uncertainty for that hour (ppm), calibrated CH 4 (dry mole fraction, ppb), standard deviation of the raw (2-3 s) CH 4 data within the hour (ppb), estimated CH 4 uncertainty for that hour (ppb), H 2 S (ppb) or δ 13 CH 4 (‰) (depending on instrument type), standard deviation of the raw (2 -3 s) H 2 S or δ 13 CH 4 data within the hour, 420 estimated uncertainty for that hour, and a user flag (1 = good, 0 = not recommended for use or not available).
Another sub-product of this dataset is the Permian Map website, developed by EDF (PermianMAP, 2021), providing access to intermediate data products and a map of Permian Basin emissions, updated periodically.

Conclusions 425
The data presented show that regional tower networks can be operated to monitor methane emissions from O&G basins.
Data quality and continuity successfully document regional methane enhancements associated with O&G production in the Delaware sub-basin of the Permian basin. The location of upwind/downwind sites both change significantly as a function of season, illustrating the need to surround a basin with measurements. The magnitude of the enhancements also changes significantly vs. season, illustrating that accurate descriptions of boundary layer depths and winds are needed to interpret the 430 data. A greater density of sites, more readily available instrument spares or more reliable GHG measurement instruments could increase the data density, but the existing network performed sufficiently to document the basic characteristics of enhancements associated with this production basin. Basins with more complex methane background conditions and/or smaller emission rates may prove more challenging to characterize.

Author contributions 435
SR and NM collected data. BJH performed data ingest and initial screening. NM performed quality control, data processing, and uncertainty analysis. VM performed data analysis and produced the majority of the figures. VM and NM wrote the manuscript, with contributions from KD, SR, and ZB. NM, KD, SR, ZB, and DL designed the study.