Multisatellite Imaging of a Gas Well Blowout Enables Quantification of Total Methane Emissions

Incidents involving loss of control of oil/gas wells can result in large but variable emissions whose impact on the global methane budget is currently unknown. On November 1, 2019, a gas well blowout was reported in the Eagle Ford Shale. By combining satellite observations at different spatial and temporal scales, we quantified emissions 10 times during the 20‐day event. Our multisatellite synthesis captures both the short‐term dynamics and total integrated emissions of the blowout. Such detailed event characterization was previously not possible from space and difficult to do with surface measurements. We present 30‐m methane and carbon dioxide plumes from the PRISMA satellite, which let us estimate flare combustion efficiency (87%). Integrating emissions across all satellites, we estimate 4,800 ± 980 metric tons lost methane. Blowouts occur across the globe and multisatellite observations can help to determine their pervasiveness, enable corrective action, and quantify their contribution to global methane budgets.

of emissions from blowouts was possible, studies have shown that these events may be among the largest point source emissions on a national scale over the event duration (Conley et al., 2016;Pandey et al., 2019). This class of well control events are not uncommon occurrences, with many reported each year in Texas (RRC, 2020a). However, the same reporting protocols in Texas do not consistently exist in other jurisdictions nationally and internationally, which creates a large underlying uncertainty.
Independent detection and quantification of emissions from these events using satellite remote sensing is an avenue to close the uncertainty gap. Global tiered remote sensing systems for CH 4 event detection and quantification are now possible with recent launches of satellites with sufficient spatial resolution and coverage, temporal coverage, and targeting capabilities. For instance, from February to March 2018, a significant gas well blowout (∼120 t h −1 ) in Ohio was detected and quantified using the TROPOspheric Monitoring Instrument (TROPOMI; Pandey et al., 2019) onboard the Sentinel-5P satellite (Hu et al., 2018;Veefkind et al., 2012). TROPOMI provides atmospheric dry air column mixing ratio (XCH 4 ) maps with 7 × 5.5 km 2 nadir spatial resolution globally and daily in cloud-free conditions (Hu et al., 2018). Similarly, the GHG-Sat-D satellite, with its finer 50 m spatial resolution, detected prolonged venting (10-43 t h −1 ) operations at a gas compressor station in Turkmenistan from June 2018 to January 2019 , results which were consistent with TROPOMI estimates. Airborne remote sensing surveys imaged and quantified emissions from the Aliso Canyon blowout January-February 2016 in California (20 t h −1 ; Thorpe et al., 2020) with results consistent with airborne in situ methods for the same period (Conley et al., 2016). The Hyperion EO-1 satellite imaging spectrometer (2000Folkman et al., 2001) also imaged plumes three times during the event at 30 m spatial resolution and detected significant CH 4 emissions .
Here, we focus on a discrete gas well blowout and show that with a combination of satellite observations, we are able to detect, quantify, and track emissions from this event. These instruments include TROPOMI, GHGSat-D, the Visible Infrared Imaging Radiometer Suite (VIIRS; Cao et al., 2013) instrument, and the PRecursore IperSpettrale della Missione Applicativa (PRISMA) satellite imaging spectrometer (launched March 2019; Loizzo et al., 2018). PRISMA has 30 m spatial resolution and measures backscattered solar radiances from 400 to 2500 nm at <12 nm spectral resolution, a range which includes both CO 2 and CH 4 absorption features. Previous theoretical work showed the potential for retrieving large CH 4 point sources with PRISMA (Cusworth et al., 2019). In this study, we provide evidence of this capability and show we can simultaneously retrieve column averaged CO 2 and CH 4 concentrations with high spatial resolution, derive emissions, and thus quantify the combustion efficiency of the flared blowout. This new capability has not been available with any single satellite instrument previously. When we combine data collection and analysis from PRISMA with other satellite instruments, we demonstrate an enhanced capability to resolve the evolution of the blowout, which has major implications for global monitoring of heavy-tailed emitters.

Event Timeline
At 02:40 (local time) on November 1, 2019, a blowout was reported at a gas well in the Eagle Ford Shale near Victoria, Texas (28.9°N, 97.6°W). The event occurred at a surface site where four horizontally drilled wells are colocated. Evacuation orders were issued soon after for residences within a 2-mile radius (Chapa, 2019). On 14 November, a cap was placed on the well and gas was diverted to an open pit where it was flared. On 20 November at 11:00, dense fluids were injected into the well, which effectively shut it in. From 2 to 12 November, the Devon Energy Corporation contracted CTEH, LLC to measure in situ volatile organic compounds (VOCs) within 5 km of the blowout at several sampling sites. Instantaneous benzene concentrations peaked at 236 ppb downwind of the blowout on 2 November. The first CTEH 24-h averaged benzene concentration was reported between 3 and 4 November as 16.5 ppb ( Figure S1). The TCEQ Karnes County air quality station (located 30 km west of the blowout) recorded a much lower maximum benzene concentration of 7.9 ppb on 2 November, with levels diminishing afterward, but spiking to 4.6 ppb briefly between 14 and 15 November ( Figure S1). No direct CH 4 measurements were made during in situ monitoring, but we infer CH 4 emissions based on chemical transport modeling and the gas composition of the well (Section S1). The Texas Railroad Commission (RRC) reports blowouts and well-control problems publicly and includes information regarding causes, injuries, and operators (RRC, 2020a). TCEQ maintains the Air Emission Event report system, which publishes endorsed company reports of emission events. However, neither RRC nor TCEQ have posted official reports at the time of this study's submission.
From space, we retrieve XCH 4 from TROPOMI, GHGSat, and PRISMA, and XCO 2 from PRISMA, and use this information to quantify total emissions from the event. We also infer CH 4 emission rates using satellite proxies, specifically flaring radiant heat from VIIRS. Availability of satellite observations depends on favorable weather conditions (in particular clear skies), and on the revisit and sampling strategy of the instrument. In this study, we use two sampling approaches jointly to constrain the blowout. The first approach uses coarse resolution global mappers with daily revisit, which here include TROPOMI for direct XCH 4 retrieval, and VIIRS, which is used to measure nighttime gas flaring radiant heat at 750 m resolution and derive methane emission rates (Elvidge et al., 2016). The second approach uses high-resolution target instruments that must be tasked to image a particular region. For this study, we requested tasking of GHG-Sat-D, PRISMA, and Planet Lab's SkySat, a 70 cm four-channel red-blue-green (RGB) and near infrared (740-900 nm; Murthy et al., 2014) instrument to image the blowout. When we coordinate tasking and combine information from all instruments, we gain much more insight into the evolution of and the integral emission from the blowout. Figure 1 shows the CH 4 emission rates derived from each measurement platform. The details of deriving emission rate estimates from each instrument are provided in the supporting information. Starting with TROPOMI, we have observations available on 2, 3, 15, and 18 November ( Figure S2). The spatial resolution of TROPOMI generally limits its ability to identify individual CH 4 point sources (Cusworth et al., 2018), but recent studies have shown that for very large events like blowouts or large venting, plumes are visible at this spatial resolution Varon et al., 2019). We estimate emissions from a TROPOMI scene by simulating atmospheric transport using the Weather Research and Forecasting Chemistry (WRF-Chem) model (Powers et al., 2017). We drive an initial WRF-Chem simulation using a somewhat arbitrary emission rate and then sample the resulting simulated concentration fields at the TROPOMI spatial resolution (Section S1). We then scale the blowout enhancement from this simulation to fit the observed TROPOMI concentrations and apply this scaling to the initial WRF-Chem emission rate to estimate blowout emissions. For 2 November ( Figure S2), we estimate a possible mean emission range of 18-163 t h −1 . This wide range is due to sparse pixel sampling from quality filtering ( Figure S2) and variable meteorology on that day ( Figure S3). We compare this emission estimate against another derived using the Integrated Methane Enhancement (IME) algorithm (Cusworth et al., 2019;Frankenberg et al., 2016;Varon et al., 2018). This method does not invoke a transport model but instead estimates emissions by multiplying the excess methane mass generated within a retrieved methane plume by the inverse lifetime of the plume, which is calculated from the boundary layer wind speed and plume length (Section S2). Using this method, for 2 November, we CUSWORTH ET AL. estimate an emission rate of 27.6 ± 8.8 t h −1 (1σ). This IME-based estimate is on the lower end of the WRF-Chem range, but consistent with bottom-up methods described later. IME methods rely on wind speed and not wind direction, which may explain the lower uncertainty for 2 November when compared to the WRF-Chem simulation. By 3 November, the IME-based emission rate dropped to 15.8 ± 5.1 t h −1 . Cloudy conditions persisted at the 7 km scale over the next several days, limiting additional TROPOMI scenes, and by 15 November, a discernible plume signal was lost, indicating that any remaining emissions following flaring were below the instrument's detection threshold. For subsequent calculations that integrate total CH 4 emissions over the event (see Section 4), we use the IME-based estimates for TROPOMI, given their lower uncertainty. We also set the emission rate between the blowout start and the first TROPOMI overpass to the 2 November TROPOMI estimate.

Emission Estimates During the Blowout
GHGSat-D imaged the blowout on 10 November at a much finer 50 × 50 m 2 effective spatial resolution. GHG-Sat-D has previously quantified CH 4 point sources as low as 3-4 t h −1 from a single overpass . In Figure 2, a CH 4 plume is clearly visible emanating from the blowout source and extending along the southerly winds. For emission quantification, we apply specific IME and cross-sectional flux methods tailored to the GHGSat-D instrument (Section S2; Varon et al., 2019). We estimate an emission rate of 10.5 ± 5.7 t h −1 , a 67% drop in emission rate from the TROPOMI IME estimate for 2 November. The reduction in emissions throughout the duration of the blowout was previously been observed in the case of the widely documented Aliso Canyon blowout (Conley et al., 2016) and here may be indicative of decreasing well pressure.
We corroborate the trend between TROPOMI and GHGSat-D emission estimates by inferring CH 4 emissions from in situ VOC measurements. Although CH 4 was not directly monitored from the ground, if we assume a fixed gas composition of the well (Table S1), we can estimate near surface CH 4 concentrations from several measured proxy VOCs (i.e., pentanes and butanes). Then, using a similar WRF-Chem setup as for comparisons to TROPOMI data, we simulate atmospheric transport to derive emission rates (Section S1). On 2 November, we estimate a CH 4 emission rate of 61 ± 32 t h −1 , which is higher but within the uncertainty of the TROPOMI IME-based emission estimate, and within the TROPOMI WRF-Chem emission estimate range. Wind conditions did not allow for the in situ network to adequately sample the plume on 3 and 4 November. Between 5 and 8 November, mean CH 4 emissions derived from the in situ proxies fluctuated between 6.8 and 12.6 t h −1 . The GHGSat-D emission rate is consistent with the range of in situ estimates, and together, they show a decreasing trend from the initial emission rate, albeit with significant variability. Though reduced, the range of 6. On 15 November, both PRISMA and SkySat imaged the blowout within a 2-h period, after flaring had been initiated to diverted well gas. Figure 3 shows the RGB images from both PRISMA and SkySat. The large flare is visible in both images, covering four PRISMA pixels. Also visible is a large aerosol, condensate, and/or condensation plume, which extends approximately 1.5 km from the well pad, along the northerly winds. We retrieve column XCO 2 and XCH 4 by applying the Iterative Maximum A Posteriori-Differential Optical Absorption Spectroscopy (IMAP-DOAS) algorithm (Cusworth et al., 2019;Frankenberg et al., 2005;Thorpe et al., 2017) to each pixel within the scene (Section S3). We infer emission rates by applying the IME method to retrieved CH 4 and CO 2 plumes (Section S2). Retrievals near the plume source are discarded due to strong sensor saturation from scattered radiance . Despite missing enhancements over the source location, we still estimate significant CO 2 and CH 4 emissions of 172 ± 71 and 5.0 ± 1.4 t h −1 , respectively. Assuming that the CH 4 emission came primarily from incomplete combustion, we follow the method of Caulton et al. (2014) and Gvakharia et al. (2017) and use the ratio of CH 4 to CO 2 emissions to derive an estimate for combustion efficiency (CE): where E represents the emission rates of CH 4 and CO 2 , A is the carbon composition of methane in the well gas (based on  represents the 1σ spread of CE when taking into account the uncertainties in CH4 E , CO 2 E , and A. This combustion efficiency estimate is lower than what is reported as typical flare operations (98%; US Environmental Protection Agency, 2015). However, field studies have shown that although combustion efficiency for typical gas operations in some regions may center on 98%, efficiency also follows a right-skewed distribution, meaning many conventional flares operate below 98% (Gvakharia et al., 2017).
We can go a step further and quantify flare temperature using PRISMA flare pixels, which serves as a qualitative check on flare combustion efficiency derived in Equation 1. Temperature retrievals from imaging spectrometers can be performed using mixture models that separate contributions from reflected solar and flare-emitted radiance (Dennison et al., 2006). We do this through a clustering of nonflared pixel spectra into representative background endmembers and combining with MODTRAN emitted radiance simulations over a range of plausible flare temperatures (Section S4). Applying these methods to PRISMA pixels where flaring is visibly apparent (Figure 3), we retrieve temperatures from 1650 to 1800 K ( Figure S7). The PRISMA retrieved temperatures are within the normal oil/gas operational range of flare temperatures (Elvidge et al., 2016), although as previously mentioned, our 87% combustion efficiency estimate is below typical flaring operations (98%). Operator efforts to control the well allowed for large amounts of gas to be flared, which can explain the high retrieved PRISMA temperatures. However, gas diversion to an open pit, though ultimately successful in helping regain well control, may have led to some gas to escape unflared, leading to lower than normal combustion efficiency for typical flare operations.
VIIRS can be used to retrieve radiant heat, which is related to the volumetric quantity of flared gas (Elvidge et al., 2016). If flare combustion efficiency is known, this relationship between VIIRS radiant heat and the flared gas quantity can be used to estimate noncombusted CH 4 emissions (Section S5). VIIRS retrieved a flare temperature of 1,585-1,615 K during its nighttime overpass on 15 November, which is slightly lower than the 1,650-1,800 K temperature range retrieved during the day from PRISMA. However, PRISMA temperature falls broadly within the VIIRS temperature range over the entirety of flaring from 15 to 20 November (1,470-2,170 K). Therefore, we take the 87% ± 4.0% combustion efficiency derived from PRISMA and apply it to VIIRS radiant heat retrievals to estimate CH 4 emission rates. On 15 November, we estimate CH 4 emissions of 6.8 ± 1.3 t h −1 from VIIRS, which is higher, but consistent within the uncertainties of the PRISMA estimate. The flare and subsequently the well were shut-in on 20 November. The VIIRS emission rate for 20 November prior to the shut-in was 1.6 ± 0.3 t h −1 . Though this is still significant, it represents a small fraction of peak emissions from the blowout. We have no additional emission constraints after the twentieth, so are unable to determine if any residual emissions were present after the blowout. We assume that emissions associated with the event ended after 20 November.

Discussion
Combining multiple satellite and in situ emission estimates for individual days allows us to estimate the total gas lost during the blowout event. Integrating satellite emission rates, we estimate 4,830 ± 980 metric tons of CH 4 were lost to the atmosphere. Therefore, with effective combination of multisensor satellites aided by favorable meteorological conditions, this work shows a clear path forward for quantifying CH 4 emissions from blowouts entirely from space. In the present analysis, no one satellite was able to capture all the dynamics of the blowout, due to observational factors including spatial resolution, spatial coverage, and target revisit frequency, as well as region-specific issues including cloud cover and the presence of flaring. Only after coordinated targeting and multisensor analysis were we able to more accurately constrain the emissions.
We compare our top-down emission estimates with a bottom-up lifecycle analysis from the Oil-Climate Index (OCI) model (Brandt et al., 2018;Gordon et al., 2015). The OCI model estimates upstream GHG emissions, gas at the well, and fugitive methane loss rate at the well level using field depth, well dimension, well pressure, gas-to-oil ratio, and liquids gravity, among other inputs (Table S2). To model a blowout with OCI, we use reported reservoir actuation pressures, as these represent pressures that the hydraulic fracturing process must overcome to produce gas (see Section S6). Though an official report of this event is not yet available (RRC, 2020a), RRC reports four horizontally drilled gas wells on the Devon Migura Lease that were directionally drilled from the same surface location where the blowout occurred (RRC, 2020c). These wells had successive spud dates dated between 30 April 2019 and 9 May 2019, several months before the blowout occurred (RRC, 2020b). The existence of pressure communication between adjacent wells and reservoirs has been shown to exist in the Eagle Ford Shale under fracture treatment (Sukumar et al., 2019). Given that the four horizontal gas wells (1H, 2H, 5H, and 6H) have the same surface location where the blowout occurred ( Figure S8), we assume well communication and model CH 4 emissions using actuation pressure assumptions (Table S2). With OCI, we estimate 22.4 t h −1 maximum emission rate associated with the blowout, consistent with the TROPOMI 2 November overpass.
We simulate daily bottom-up emissions using OCI during the blowout (Table S4). Daily emission changes are driven by applying operational information (e.g., specific time blowout started and flaring was applied) and ancillary atmospheric observations (e.g., trends in VOCs). Table S4 compares the top-down emission time series with OCI daily estimates and shows close correspondence. OCI total methane over the 20-day course of the blowout event is estimated at 4,630 metric tons, consistent with the satellite estimate. Finally, OCI estimates that under normal operating conditions, the four wells involved in this blowout together would have emitted 66 metric tons of fugitive methane in total over the course of 20 days, using an average leakage rate of 0.012 kg CH 4 per kg CH 4 at the wellbore (Table S3). Therefore, the blowout is estimated by OCI to have increased CH 4 emissions from the wells involved by a factor of 75 over routine operations.
Blowouts represent a potentially large global GHG source, but their emissions are poorly understood. In Texas, blowouts are required to be reported to the RRC, and all such events from 1960 to present are logged ( Figure S9; RRC, 2020c). Since the 1980s, flared and unflared blowouts in Texas have decreased, but they are still frequent today. Globally, there is much less open-access information about blowout statistics, so we lack bottom-up constraints to quantify the contributions to the anthropogenic CH 4 budget. Satellites will be critical to closing this uncertainty gap. Similarly, satellites could be key to identify and quantify methane emissions from other unexpected events (e.g., unlit flares in remote regions, underground pipelines, and failed controls on elevated point sources). The GeoCARB mission will extend TROPOMI's capability by providing daily CH 4 maps across North and South American land masses at 5-10 km resolution (Moore et al., 2018), and the MethaneSAT mission is planned to provide ∼1 km CH 4 maps focusing on major oil/gas basins globally at least weekly (Wofsy & Hamburg, 2019). GHGSat is planning on expanding its observing capacity with two additional instruments, GHGSat-C1 (launched 2020) and GHGSat-C2 (expected launch 2021; Jervis et al., 2020). Many imaging spectrometers with targeting capabilities similar to PRISMA with similar or better spectral resolution and signal-to-noise are set to launch in the next few years (EnMAP, Guanter et al., 2015;EMIT, Green et al., 2018). Globally mapping imaging spectrometers with 10-15 days revisit cycles are also planned to launch in the next decade (SBG, Hochberg et al., 2015;CHIME, Nieke, & Rast, 2018). Even with these new observing platforms, constraining all the dynamics of large emission events will require coordinated tasking and combining of all satellite information. As we showed with PRISMA, these instruments will have the ability to map both CH 4 and CO 2 plumes, and comparison of the corresponding emission rate estimates allowed us to assess CH 4 combustion efficiency. This confirms previous theoretical work that has indicated the possibility of doing CH 4 monitoring with spaceborne imaging spectrometers (Ayasse et al., 2019;Cusworth et al., 2019). Using the visible and near infrared PRISMA channels, we can also retrieve flare temperature, which can be an ancillary check on combustion efficiency. The ability to quantify CH 4 emissions, CO 2 emissions, flare temperature, and combustion efficiency from one satellite instrument is a new opportunity in satellite remote sensing that will greatly improve total carbon footprint accounting.

Data Availability Statement
TROPOMI data are available at https://s5phub.copernicus.eu/dhus/#/home. WRF-CHEM model code is available at https://ruc.noaa.gov/wrf/wrf-chem/. PRISMA data are publicly available to registered users at https://prisma.asi.it/. Registration is free and can be obtained at https://prismauserregistration.asi.it/. VIIRS data are available at https://doi.org/10.5067/VIIRS/VJ103IMG.002. ing operator information and insights regarding this blowout event. Portions of this work were undertaken at the Jet Propulsion Laboratory, California Institute of Technology, under contract with NASA. Some of the work was supported by NASA's Carbon Monitoring System program. This research contains Copernicus Sentinel data 2019. S. Pandey is supported through the GALES (Gas LEaks from Space) project (Grant 15597) by the Dutch Technology Foundation, which is part of the Netherlands Organisation for Scientific Research (NWO). WRF simulations were carried out on the Dutch National supercomputer Cartesius maintained by SurfSara (www.surfsara.nl). The authors thank colleagues at Planet Labs for tasking SkySat in response to this event.