Daily Satellite Observations of Methane from Oil and Gas Production Regions in the United States

Production of oil and natural gas in North America is at an all-time high due to the development and use of horizontal drilling and hydraulic fracturing. Methane emissions associated with this industrial activity are a concern because of the contribution to climate radiative forcing. We present new measurements from the space-based TROPOspheric Monitoring Instrument (TROPOMI) launched in 2017 that show methane enhancements over production regions in the United States. In the Uintah Basin in Utah, TROPOMI methane columns correlated with in-situ measurements, and the highest columns were observed over the deepest parts of the basin, consistent with the accumulation of emissions underneath inversions. In the Permian Basin in Texas and New Mexico, methane columns showed maxima over regions with the highest natural gas production and were correlated with nitrogen-dioxide columns at a ratio that is consistent with results from in-situ airborne measurements. The improved detail provided by TROPOMI will likely enable the timely monitoring from space of methane emissions associated with oil and natural gas production.

Methane is the second most important anthropogenic greenhouse gas in the atmosphere after carbon dioxide 1 . Methane emissions are related to (i) the production of fossil fuels, i.e. crude oil, natural gas and coal, and can also be released from natural seeps, (ii) microbial processes in wetlands, rice cultivation, livestock, landfills and termites, and (iii) biomass burning 2 . Because of the imbalance between methane sources and sinks, the atmospheric mixing ratio has increased from a pre-industrial value of 722 ppbv 1 to the current global average of approximately 1850 ppbv 3 . As the atmospheric lifetime of methane is relatively short at 9.1 years 1 , a reduction in methane emissions would lower the combined radiative forcing from greenhouse gases on a timescale of years and is therefore a relatively efficient option to mitigate climate change.
Methane emissions associated with the production of fossil fuels (coal, oil and natural gas) constitute about 24% of the global emissions of methane and a larger fraction of the anthropogenic emissions. The global emissions from this industry have stayed relatively constant since 1985 even as the global production of fossil fuels increased over this same time period 2 . Methane emissions from the U.S. oil and natural gas supply chain have received much recent attention and are estimated to be 2.3% of gross U.S. gas production 4 and 41% of the anthropogenic U.S. emissions 5 . The long-term trends in U.S. emissions of methane are the subject of some debate, with one study finding a 30% increase between 2002-2014 6 , and another finding no significant trends between 2006-2015 7 .
Global measurements of methane are made from a network of ground-based and airborne in-situ measurements 8 . These measurements are mostly made away from sources to reflect background concentrations. Measurements of methane emissions in oil and gas production areas have been made using surface monitors 9,10 and from research aircraft [11][12][13] . The latter are labor intensive and only provide snapshots of emissions, which may be an issue as episodic, large emissions can represent a relatively large fraction of the total 14 . This calls for more frequent, basin-wide monitoring of emissions, which explains the interest in methane measurements using space-based sensors. However, observing methane from oil and gas operations from space has been challenging. Schneising et al. derived methane emissions from two oil producing regions, the Bakken and Eagle Ford, using data from the SCIAMACHY instrument 15 . Kort et al. described the observation in SCIAMACHY data of a hotspot in methane over the Four Corners region, a region with natural gas and coal-bed methane production 16 . Turner et al. used data from the GOSAT instrument to describe trends in U.S. methane emissions but could not attribute the trends to a specific sector 6 . All of these studies presented satellite measurements of methane that were averaged over a year or more. One study that used observations from single satellite overpasses was reported by Thompson et al. following the Aliso Canyon blow-out 17 , which was one of the largest accidental releases of methane in recent history 18 . Similarly large emissions of methane in Turkmenistan were recently reported using data from GHGSat and TROPOMI 19 .
The TROPOspheric Monitoring Instrument (TROPOMI) is the sole instrument on the EU Copernicus Sentinel 5 Precursor satellite that was launched in October 2017. In comparison with its predecessor, the Ozone Monitoring Instrument (OMI) 20 , TROPOMI has an improved spatial resolution of up to 7 × 3.5 km 2 and also added a short-wave infrared radiation (SWIR) band that allows the detection of methane and carbon monoxide 21 . The TROPOMI measurements of methane have been compared with those from the GOSAT instrument and were found to be in good agreement 22 . Here, we study the enhancements in methane observed over two oil and gas production regions in the U.S., the Uintah basin in Utah and the Permian Basin on the border of Texas and New Mexico. Using corroborating evidence from ground-based and airborne measurements, we show that the observed methane enhancements in single overpasses can be attributed to emissions into the boundary layer from oil and natural gas production. These results suggest that the future determination of regional methane emissions with a high time resolution and soon after the time of emission will be feasible.

Results
tRopoMi observations. Average methane columns over the contiguous U.S. between 1 December 2018 and 31 March 2019 are shown in Fig. 1. The number of TROPOMI retrievals per grid box during this period is shown in Fig. S1 and was limited over some regions due to cloud cover, low surface albedo, topography and other factors that affect the data quality. Enhancements in methane columns are for example observed: • Over the Central Valley of California, a region with strong methane emissions from agriculture (both livestock and rice cultivation) and from oil and natural gas production. The enhancements were particularly clear in December 2018 and January 2019, as a result of wintertime inversions that trap emitted methane in the boundary layer. • In the Uintah basin in Utah in the northeastern corner of the state. This is a region where strong inversions can trap emissions from oil and gas production in the winter. Accordingly, the enhancements were particularly clear in December 2018. • Over Texas, which has several oil and gas producing regions. Both oil and gas are produced in the Permian Basin in the Southeast corner of New Mexico and the adjacent parts of Texas, and in the Eagle Ford south of San Antonio. Natural gas production also takes place in the Barnett Shale west of Dallas -Fort Worth, and in the Haynesville on the border of Texas and Louisiana. • Over several states with intense agricultural production of corn and soybean 23 ranging from North Dakota through Minnesota, Illinois and Ohio as well as the Mississippi Delta region. Rather than an actual emission of methane, we suspect that retrieval biases due to low surface albedo in the winter, when the agricultural soils are bare, may play a role and need to be further investigated. • Over Florida, which may be related to emissions from wetlands.
The San Juan basin in the northwestern part of New Mexico does not show up consistently in the TROPOMI data so far, although enhancements at the northern end of the basin were observed on two days in January 2019 (Fig. S2). A previous study identified a hot spot in methane in this basin using data from SCIAMACHY 16 .
Average tropospheric NO 2 columns between 1 December 2018 and 31 March 2019 are also shown in Fig. 1. In this case, the strongest enhancements are observed over metropolitan areas such as New York City, Chicago, Denver, Houston, Los Angeles and many others. Enhanced NO 2 columns are also observed over oil and gas producing regions, most clearly over the Permian Basin and Eagle Ford in Texas, and the Bakken in North Dakota. Enhanced NO 2 over these regions was previously identified in the OMI data 24 .
The methane columns shown in Fig. 1 have not been validated in detail. A comparison with GOSAT data was previously published 22 and a comparison with the measurements by the Total Carbon Column Observing Network (TCCON) is currently in progress. We include here a comparison between TROPOMI measurements that are co-located with the Lamont TCCON site in Oklahoma within a 300-km radius (Fig. S3). The dynamic range in methane columns over the comparison period was limited, but the two data sets both show periods with enhanced columns in August 2018 and in October-December 2018, and were correlated with a linear correlation coefficient, r, of 0.6. The average bias between the TROPOMI and Lamont TCCON data sets was −7.5 ± 9.9 ppbv, where the error represents the 1-σ variability in the bias from individual days.
The accuracy of the TROPOMI methane column product for low surface albedo needs further investigation and so data included in this paper are all limited to a surface albedo higher than 0.05, except for Fig. S2 where an albedo threshold of 0.03 was applied. Methane column mixing ratios over the Great Lakes region and into Canada are occasionally lower than the surface values and possibly affected by low tropopause heights and stratospheric air masses with reduced methane. Nevertheless, the column enhancements over Utah and Texas appear to be www.nature.com/scientificreports www.nature.com/scientificreports/ readily attributable to emissions associated with oil and gas production, and this hypothesis is evaluated in the next two sections.
Uintah basin. The Uintah Basin in Utah is a smaller natural gas and, to a lesser extent, oil producing region in the northeastern part of Utah. Drilling activity has declined since 2008 and natural gas production is currently decreasing (Fig. S4). The basin is surrounded by mountains and atmospheric studies in the region were sparked by the occurrence of winter ozone formation 25,26 . This unusual phenomenon occurs when emissions from oil and gas production accumulate under cold-pool conditions strengthened by a snow cover, and react photochemically over the course of days to build up ozone to levels that far exceed other polluted regions of the U.S.
The TROPOMI methane measurements show a clear hotspot over the Uintah basin ( Fig. 2A). Enhanced methane columns were observed on the south side and deeper part of the basin, where most of the oil and natural gas wells are situated 11 . Data gaps in the TROPOMI data are caused by clouds, topographic features and low surface albedo. The methane hotspot in the Uintah Basin became less obvious from December 2018 through March 2019 as the cold-pool conditions that trap emissions in the basin became less prevalent (Fig. S5). NO 2 columns over the Uintah Basin are strongly affected by emissions from the Bonanza power plant 26 and are not included in this analysis.
Methane has been measured in-situ in the basin since 2015 at three different locations indicated in Fig. 2A. The Horsepool and Roosevelt monitors are situated at lower elevation in the basin and are frequently exposed to high methane in the winter when emissions are trapped under cold-pool conditions. The Fruitland monitor is situated at a higher elevation and does not see similarly high methane enhancements 27 . Figure 2B   www.nature.com/scientificreports www.nature.com/scientificreports/ The in-situ measurements were averaged over a 60-min interval centered on the overpass time of the TROPOMI instrument. TROPOMI data were limited to those pixels that were within 30 km of the monitor, with the size of the symbol representing the proximity. Data were also selected for surface albedos >0.05. The column average methane mixing ratios from TROPOMI correlate with the in-situ measurements with a correlation coefficient of 0.71. The effect of selecting for surface albedo is shown in Fig. S6. A less strict criterion results in more outliers, but the correlation between the column and in-situ measurements remains. Figure 2B shows a linear regression slope between the column-averaged and surface methane mixing ratios of 0.053. This value is much lower than one, because methane is only enhanced in a shallow boundary layer. Assuming a tropopause altitude of ~10 km and a surface altitude of 1.5 km, this implies that methane accumulates in a layer of ~450 m thick (0.053 × 8500 m). This layer height is in the range of previous observations in the basin. During cold-pool conditions, the height of the polluted layer is lower initially (200-300 m) 26 but can grow as cold-pool conditions persist and emissions gradually mix to higher elevations. Deeper boundary layers of more than ~1500 m were also observed in the winter in the absence of cold-pool conditions 11 . More insight into the altitude distribution of the methane is obtained by plotting the column average mixing ratios of methane observed on a single day, 9 December 2018, as a function of the surface pressure, i.e. elevation (Fig. 2C), with the location of www.nature.com/scientificreports www.nature.com/scientificreports/ the columns shown in Fig. 2D. In the absence of local emissions, the column-averaged methane mixing ratios do depend slightly on surface elevation, which may explain the observations outside the basin, but the much stronger altitude dependence inside the basin can only be attributed to the build-up of methane in a shallow layer. This analysis also suggests that the depth of the layer with enhanced methane is ~50 hPa or ~500 m. It should be noted that the Uintah basin was snow-free on 9 December 2018 (Fig. 2D), so the inversion may not have been strong and the mixed layer deeper than during cold-pool conditions 28 .
We conclude that the TROPOMI hotspot observed over the Uintah Basin is caused by enhanced methane in the boundary layer in agreement with surface monitors. Continued surface and TROPOMI measurements over the Uintah Basin will be useful to further develop and evaluate emissions estimates under cold-pool conditions 27 , which can then be applied with more confidence for other, similar basins. permian basin. The Permian Basin in Texas and New Mexico has been a very large production region for both oil and gas in the United States for many years (Fig. S4). The region has seen increases in extraction of both oil and natural gas by unconventional means since 2010, i.e. horizontal drilling and hydraulic fracturing that allow extraction from shale and tight sands formations. Unlike the Bakken in North Dakota and the Eagle Ford in Texas, the Permian Basin did not see a decrease in oil production after the 2015 downturn in drilling and is currently developing very rapidly. Figure 3 shows the methane and tropospheric NO 2 columns over the Permian basin averaged over the 1 December 2018 through 31 August 2019 period. Natural gas and oil production volumes are shown in Fig. 3C,D. Both methane and NO 2 show enhancements over the Delaware basin in the west, and the Midland basin in the eastern part of the Permian Basin. NO 2 is also enhanced over Midland and Odessa, the two main cities in the region, but NO x emissions in these cities are clearly not the only source. Figure 4 illustrates the relationship between methane and oil and gas production in more detail. For each month between December 2018 and August 2019, we regressed gridded methane columns vs. gridded natural gas and oil production, and the best fits for each month are shown in Fig. 4. Examples of the individual fits are shown for August 2019 in Fig. S7. It is seen that pixels www.nature.com/scientificreports www.nature.com/scientificreports/ with higher production are associated with significantly higher methane columns, and that the slopes are similar between different months.
Using airborne measurements, it has been shown that methane and NO x emissions in oil and gas production regions are correlated 29 . While these two trace gases are not from the exact same sources, the emissions are collocated thereby causing their correlation in the atmosphere. For example, methane is released from the wellhead, separator and compressor on each well site, while NO x is released from internal combustion engines that are used to run this equipment 30 . All these parts of the infrastructure are sufficiently close to each other that their emissions appear to come from the same location when sampling at a distance. Also, while the emissions from each well site can be different, an average ΔNO x /ΔCH 4 ratio is observed when averaging over 100 s of sites. Here, ΔNO x and ΔCH 4 refer to their enhancements over the background and are determined from the linear regression between their mixing ratios, with the background in methane signifying the background mixing ratio. During airborne measurements in 2015 in the Permian Basin, this ratio was observed to be (9.0 ± 0.2) × 10 −3 29 .
The TROPOMI methane and NO 2 columns in the Permian Basin were often found to be correlated on individual days. Examples are shown for 31 January 2019 in Fig. 5 and for 25 January 2019 in Fig. S8. Within a box that encompasses the Permian Basin indicated with the thick line, NO 2 and methane correlated with a correlation coefficient of 0.81 on 31 January (Fig. 4C) and of 0.86 on 25 January (Fig. S8). No concurrent enhancements were seen for the aerosol optical thickness derived from the NPP VIIRS measurements, which were obtained within minutes from the TROPOMI overpass, ruling out an aerosol interference. In the December 2018-March 2019 period, NO 2 and methane were correlated with r > 0.4 on 45 out of 55 days with available data.
The relation between the NO 2 and CH 4 in the boundary layer is assessed for each day for which more than 10 NO 2 -CH 4 data pairs are available. First, we subtract the background for both NO 2 and CH 4 . The background is estimated as the 10th percentile of the values. Next, we convert the NO 2 column and CH 4 mixing ratios into boundary layer mixing ratios, by assuming a boundary layer thickness of 100 hPa. For NO 2 this is done by converting the number concentration into a mixing ratio. For CH 4 we rescale the column mixing ratio by the surface pressure divided by 100 hPa. We fit a linear relationship through the NO 2 -CH 4 data points, using an orthogonal least-squares fit. Before applying this fit, we apply a scaling such that both NO 2 and CH 4 are in the range 0-1. On 31 January, the ΔNO 2 /ΔCH 4 ratio was found to be (4.9 ± 0.2) × 10 −3 . The distribution in the ΔNO 2 /ΔCH 4 ratios from all overpasses between 1 December 2018 and 31 March 2019 with r > 0.4 is shown in Fig. 5D. Also added to this graph is the ΔNO x /ΔCH 4 ratio of (9.0 ± 0.2) × 10 −3 reported from airborne measurements in the Permian Basin 29 . This ratio is at the high end of the ΔNO 2 /ΔCH 4 ratios from TROPOMI, which is entirely reasonable given that NO x includes NO in addition to NO 2 , and because of photochemical removal, which can remove a fraction of NO x on the timescale of atmospheric transport across the basin.
We conclude that the TROPOMI methane enhancements observed over the Permian Basin are caused by emissions in the boundary layer that include those from oil and gas production. The TROPOMI data show separate enhancements over the eastern (Midland) and western (Delaware) parts of the Permian Basin, which can be explored in more detail when more data are available.

Discussion
Using methane and NO 2 column measurements from the new TROPOMI instrument, we have shown that emissions from oil and gas production in the Uintah and Permian Basins can be observed in the data from individual overpasses. This is a vast improvement over measurements from previous satellite instruments, which typically needed to be averaged over a year or more to quantify trends and regional enhancements in methane emissions.
In the future, TROPOMI will likely allow the study of trends in methane and NO x emissions from oil and gas production. It will be of interest to look for the correlations between, on one hand, methane and NO 2 columns with, on the other hand, proxies for industrial activity such as active drill rig counts, production volumes of oil, gas and condensate, and natural gas flare data. Such analyses may allow the relative importance of different www.nature.com/scientificreports www.nature.com/scientificreports/ industrial activities to be separated. The current analysis also needs to be extended to other oil and production regions in North America once more data are available, and to production regions outside of North America.
The TROPOMI data will likely also allow the quantification of methane and NO 2 emissions through inverse modeling. For methane, this can possibly be done directly, although the accurate modeling of tropopause height, stratospheric removal and boundary conditions of the modeling domain are all significant challenges 31 . An indirect way of determining methane emissions may be through the inverse modeling of NO x emissions 32,33 and the scaling to methane through observed NO 2 /CH 4 ratios.
Finally, in-situ measurements in oil and gas production regions and coordinated airborne measurements would allow further verification of the results obtained in this analysis and extension to other regions.

Methods tRopoMi satellite instrument. TROPOMI is the single payload on board of the EU Copernicus
Sentinel-5 Precursor satellite, which was launched in October 2017 into a Sun-synchronous orbit with an overpass time of 13:30 local solar time 21 . TROPOMI is a push-broom imaging spectrometer with a swath width of 2,600 km and a ground pixel of 3.5-7 × 7 km 2 at the sub-satellite point. TROPOMI has two spectrometer modules, the first covering bands in the ultraviolet, visible, and near-infrared (NIR) and the second dedicated to the shortwave infrared (SWIR) band centered around 2.3 μm.
tRopoMi methane retrieval. We use the TROPOMI operational CH 4, data product (Hu 2016) version 1.2 (https://doi.org/10.5270/S5P-3p6lnwd), which implements the RemoTec retrieval algorithm 34,35 . The operational algorithm uses a two-band retrieval approach using the NIR and SWIR bands. The SWIR band contains the CH 4 information and the NIR band is used for correction for aerosols. From the operational data we use bias-corrected www.nature.com/scientificreports www.nature.com/scientificreports/ column average mixing ratios, with a quality value larger than 0.5, which filters the data for low data quality, high solar and viewing zenith angles and high aerosol loading. We estimate the uncertainty in the column average CH 4 mixing ratios by multiplying the reported retrieval precision with a factor of 2, as recommended by the data providers.
For the maps shown in Fig. 1, and for the comparisons between NO 2 and CH 4 we constructed daily data on a 0.1° × 0.125° latitude-longitude grid. For the CH 4 data, only retrievals with a surface albedo larger than 0.05 are included in the gridded data. tRopoMi no 2 retrieval. We use the TROPOMI operational NO 2 data product version 1.2 (https:// doi.org/10.5270/S5P-s4ljg54). The NO 2 processing system uses a DOAS (Differential Optical Absorption Spectroscopy) fit on the visible spectra. The tropospheric NO 2 column is determined by subtracting the stratospheric NO 2 column amount and corrections for the effects of surface reflectance and clouds 36 . We use the NO 2 retrievals with a quality value larger than 0.75.
In-situ measurements in the uintah basin. In-situ measurements of CH 4 (along with CO 2 and H 2 O) at three sites in the Uintah Basin have been carried out by the University of Utah since 2015 using Los Gatos Ultra-Portable Greenhouse Gas Analyzers (Los Gatos Research Inc., San Jose, CA). The analyzers take data every 10 seconds and are calibrated every three hours with reference gas cylinders traceable to the World Meteorological Organization scale. For more details regarding the measurement setup and instrument uncertainties, we refer the reader to Foster et al. 10 and Bares et al. 37 .
tccon measurements at lamont. Measurements from the Total Carbon Column Observing Network (TCCON) at Lamont, Oklahoma (36.604 N, 97.486 W) are used to evaluate TROPOMI data for a location that is closest to the region of interest (Fig. S3). TCCON consists of a network of ground-based Fourier-Transform Infrared Spectrometers that record spectra of the sun and allow accurate and precise retrievals of column-averaged abundances of atmospheric constituents including CH 4 . Data are available at https://doi.org/10.14291/tccon. ggg2014.lamont01.R1/1255070 38 .
Permian basin production data. Oil and natural gas production data were obtained at the well level using the Drillinginfo tool provided by Enverus and were averaged over the same 0.1° × 0.125° latitude-longitude grid boxes as used for the methane columns. Total production volumes of oil and natural gas for the U.S. and for the Permian Basin were also obtained from the U.S. Energy Information Administration (www.eia.gov) and for Utah from the Utah Division of Oil, Gas and Mining (oilgas.ogm.utah.gov) summing the data for the Uintah and Duchesne counties. Drill rig counts are from Baker Hughes (bakerhughes.com) with only statewide data available for Utah.