Satellites unveil easily-fixable super-emissions in one of the 1 world ' s largest methane hotspot regions 2 3

Affiliations: 7 1 Research Institute of Water and Environmental Engineering (IIAMA), Universitat Politècnica 8 de València (UPV), 46022, Valencia, Spain. 9 2 SRON Netherlands Institute for Space Research, 3584 CA, Utrecht, The Netherlands. 10 3 Environmental Defense Fund, Reguliersgracht 79, 1017 LN Amsterdam, The Netherlands. 11 4 Institute for Marine and Atmospheric Research Utrecht, Utrecht University, 3584 CC Utrecht, 12 The Netherlands. 13 14


79
from space. Since 2017, the TROPOMI sensor onboard Sentinel-5P provides daily global CH4 80 concentration data with a 7x5.5 km 2 pixel resolution (19). This allows detection of CH4 concentration 81 enhancements at the regional scale (e.g., 17-21), but in general, does not enable the determination 82 of single point sources. On the other hand, the GHGSat instruments and so-called hyperspectral 83 satellite missions like PRISMA, ZY1 AHSI and Gaofen-5 AHSI are able to map CH4 plumes from 84 single emitters at high spatial resolution (25-50 m GHGSat and 30m the rest) with a detection limit 85 roughly between 100 and 1000 kg/h, suitable to detect medium to strong point emitters worldwide 86 (13,25,26). The systematic application of these measurements, however, is limited by their sparse 87 spatio-temporal coverage (see Materials and Methods). The recent realisation of the CH4 mapping 88 potential of so-called multispectral missions with frequent global coverage holds promise to 89 alleviate this gap (27). Missions like Sentinel-2 (S2) and Landsat 8 (L8) cover the entire world with 90 a relatively high spatial and temporal resolution (20 m and less than 5 days revisit time for S2, and 91 30 m and less than 15 days revisit time for L8), so they are able to continuously monitor CH4 plumes 92 under favorable conditions (typically, strong emissions over spatially homogeneous areas). In 93 particular, S2 provides a very high spatio-temporal sampling and data volume, which makes it to 94 be the best mission for systematic monitoring of CH4 sources in those locations where the site 95 characteristics enable CH4 retrievals with multispectral missions. L8 and its precursors in the 96 Landsat series do not provide such a high density of observations but allow to extend the time 97 series to years and even decades before the S2 era. This recently-developed satellite-based CH4 98 monitoring scenario allows to detect single point emissions of the largest CH4 hotspot regions in 99 the world, which are identified with TROPOMI's moderate resolution observations (28).

100
One example of those CH4 hotspot regions is the west coast of Turkmenistan, located in the Balkan 101 province on the shores of the Caspian Sea, within the South Caspian Basin (SCB). This is a desert 102 area where the main human activity is the production of O&G and derived products, with a residual 103 presence of other possible anthropogenic CH4 sources such as livestock, rice fields or landfills (29, 104 30) and an abundant presence of mud volcanoes (more than twenty), some of which are associated emissions over the west coast of Turkmenistan based on the hotspot locations provided by the 125 TROPOMI observations. This survey covers an area of approximately 21500 km 2 and the time 126 period between January 2017 and November 2020. Our analysis relies on three different types of 127 space-based CH4 measurements, which are used synergistically: TROPOMI data facilitate the 128 delimitation of the study area and the identification of the most active regions; the hyperspectral 129 images from PRISMA and ZY1 AHSI allow the identification of medium-to-strong emitters and the 130 accurate quantification of emission rates for those regions in a limited set of days; finally, the 131 multispectral data from S2 and L8 enable the constant monitoring of the emissions from the 132 emission points unveiled by the hyperspectral data (see Materials and Methods). We choose the 133 west coast of Turkmenistan for this study because it offers an ideal combination of extreme CH4 134 emissions with a bright and relatively homogeneous surface. This allows us to best evaluate this 135 unprecedented combination of CH4 data streams as well as to extract its full potential.

141
Combining the hyperspectral and multispectral high spatial resolution satellite data, we have 142 detected 29 emission points with activity between January 2017 and November 2020 (Fig. 2). The 143 areas with the highest density of point sources in our high-resolution survey coincide with the 144 strongest CH4 enhancements over the west coast of Turkmenistan, as seen in the regional-scale 145 maps generated from TROPOMI moderate resolution data ( Fig. 1)

146
The 20-30 m sampling of the hyperspectral and multispectral satellites in combination with very 147 high-resolution imagery from Google Earth, Bing Maps and Esri (<2.5m/pix) provide sufficient 148 information to determine the coordinates of emission sources with high precision, especially for 149 those emitters with many detected plumes (see Materials and Methods). Combining these data, we 150 have identified the sources of 26 of the 29 points. We find that the vast majority of the emitters (24 151 of them) are inactive flares that vent gas. Several of them have flaring activity before 2017 152 according to the historical record of the S2, Landsat 5, 7, and 8 satellites, and Google Earth, Bing 153 and Esri images, and three of them had an active flare at the beginning of the study period (Fig. 154 S1), followed by CH4 emissions as soon as the flare disappeared. The flaring activity is discussed 155 in more detail in the following sections.

156
The 24 emitting flares are distributed across different onshore fields of the SCB with a higher 157 density in the Goturdepe, Barsa-Gelmez and Korpeje fields (Fig. S2, and labeled with emitters A.X,

158
B.X and D.X, respectively in Fig. 2). These three fields have the highest production (Table 1) and 159 are also three of the oldest ones in the basin. This coincides with the 2013 Carbon Limits report, 160 which indicates that most of the flares are concentrated in fields built before 1990 (37). Most of the 161 emitters are in fields where the predominant activity is crude oil and condensate production, except 162 for the Korpeje field that extracts mainly gas (see Table 1). Two of the emitting flares are around

168
Regarding the two other emitters with a known origin, the plumes from points A.10 and E.2 (see 169 Fig. 2) are due to pipeline leaks that persist over several months. In the case of A.10, the leak is 170 active for more than a year between 2019 and 2020, while at E.2, we observe emissions from April 5 to October 2018. It has been possible to confirm that these two emissions are due to leaks because above pipes, that the facilities are old in these fields and that, according to the 2013 Carbon Limits 178 report, the pipeline network (controlled by the national gas company Turkmengas) "is characterised 179 by its old and inefficient equipment" (37). However, we do not have access to records of incidents 180 or leaks recorded by the operators and cannot confirm the source of the emissions because the 181 very high-resolution imagery available is not sufficiently up to date to support this hypothesis, and 182 the resolution of S2 and Landsat imagery is not sufficient in these cases to distinguish a clear

197
The coincident overpass time of S2, PRISMA and ZY1 (2 -5 minutes difference) has enabled us 198 to capture emissions concurrently with S2 and the hyperspectral systems (see Fig. S5). Using the 199 accurate CH4 concentration enhancement maps from the hyperspectral systems as a reference,

200
we can assess the detection limits of the substantially lower signal-to-noise ratio S2 observations.

201
This exercise shows that S2 can detect emissions of at least 1800 ± 200 kg/h for the Turkmenistan 202 desert scenes, as this is the smallest emission for which we have a coincident detection with the 203 hyperspectral data. This is the minimum flux rate that we set for the plumes detected by S2 (944 204 plumes in total) between January 2017 and November 2020 (Fig. S4).

205
We have estimated the approximate annual flux emitted from the 29 emitters identified in the study 206 area, i.e., the total CH4 flux emitted from the sources that we sample in our study. This calculation  and Gogerendag fields (labelled in Fig. 2 with emitters A.X, D.X and C.X, respectively) as 220 representative cases of different temporal evolution patterns. Goturdepe is one of the fields with 221 the highest number of identified emitters, and its temporal evolution clearly shows a decrease in Immediately after the article submission (May 2019) emissions stopped from that source, but both 226 our analysis and the one by Varon et al. (2021) (27)

238
Analysing the emitters individually, we also see that there is wide variability in their emitting 239 frequency. Of the 29 points, 6 show emissions on only between 1 and 3% of the observed clear-240 sky days, i.e., they rarely present emissions above our 1800 kg/h detection limit. On the opposite 241 side, 5 points show emissions in more than 38% of the observed days. For example, Figure 3 242 shows a S2 detection series from A.3 (29% emission frequency) whose emissions persist during

251
We also look at the emissions of the region before our 2017-2020 core study period. First, the 252 longer time series of L8 satellite data reveal that at least 15 of the 29 emitters identified in the study 253 period were already emitting large amounts of CH4 before January 2017, as shown in Figure   high-resolution data for these dates, nor detailed information about the infrastructure, we have not 279 attributed these emissions to any specific infrastructure.

280
All these data demonstrate that this type of emission has been occurring for many years and that 281 the origin of these long-term CH4 enhancements is in the venting of gas, mainly from oil and 282 condensate fields.

295
The effect of the use of flaring can also be noticed in the TROPOMI data where, for example, we

304
In this study, we have used a combination of satellites to produce a large-scale survey of individual 305 CH4 emitters active between 2017 and 2020 on the west coast of Turkmenistan, one of the world's 306 largest CH4 hotspot regions as shown by TROPOMI observations. First, areas of interest within the 307 region have been identified using medium-resolution data from TROPOMI. Two types of high-308 resolution data (multi-and hyperspectral) have then been used to detect, quantify, and monitor the

317
of flaring as a method to treat excess gas. Secondly, the emissions not related to venting are linked would be preferable (40); in the case of pipeline leaks, it is necessary to improve maintenance and 324 surveillance. Identifying these high emitting sources is fundamental for any mitigation strategy, as 325 their elimination would result in an important reduction of CH4 emissions. In particular, we estimate 326 that the emissions identified in this study amount to 0.28 Tg a -1 (0.25-0.31 Tg a -1 95% confidence 327 interval), which could be easily avoided. It is unknown how these numbers would scale to the global 328 scale, but we can already speculate that a massive amount of CH4 emissions could indeed be 329 avoided if greater control actions were taken on oil and gas extraction operations.

331
The emitting sources found in the study only represent emitters above the detection limit of the 332 satellites used in this work. In these cases, synergy with a regional mapper (and inverse modelling)

346
Our results also point to the risks of penalizing flaring without effective measures to control venting.

347
The possibility of flaring cessation at the expense of venting is a problem that has been discussed 348 in the past (40) since monitoring flaring is easy to carry out by satellites, but venting was easy to 349 hide until now. Furthermore, the methods we use here can also be applied to track the progress of 350 flare reduction strategies in other areas of the world.

358
(19) provides daily global coverage of CH4 data with 7 km x 7 km (since August 2019 5.5 km x 7 359 km) pixel resolution in nadir that allows finding areas with high CH4 concentration enhancements.

360
The approximate location of the strongest sources in the study area has been identified using the

371
the detection and monitoring of CH4 emissions. Hyperspectral instruments offer a relatively high 375 sensitivity to CH4 thanks to tens of spectral channels located around the strong CH4 absorption observations over any region on Earth, but with very limited sensitivity to CH4.

380
Use of hyperspectral data for CH4 detection and quantification

382
For this study, we have collected data from the ZY1 AHSI and PRISMA missions, which are the

389
All hyperspectral data acquisitions took place during 2020 (the last year covered by this study).

390
Acquisition requests were first made with a focus on the key points identified by TROPOMI, and 391 then those were extended to other possible key areas (see the following subsection). Due to the 392 difficulty to obtain data from these sensors in the short term, we could not cover some areas in that 393 time range. Many PRISMA images have been acquired from the catalogue, while others have been 394 obtained based on requests for targeted locations. In total, we have obtained 12 images from 395 PRISMA and one from ZY1 (see Fig. S11).

397
The hyperspectral images have allowed us to observe CH4 emissions with 30m spatial resolution 398 and quantify the emissions using the matched filter method (13). The quantification has been done 399 with the integrated mass enhancement (IME) method (41)

404
Use of multispectral data for CH4 monitoring

406
For the temporal monitoring of emissions, we have used the Sentinel-2 Level 2A (L2A) product 407 from both S2-A and B satellites of ESA's Copernicus program, whose data are openly available on 408 the Copernicus Open Access Hub official portal.

410
The S2 CH4 detection limit and the estimation of the emissions detected in S2 monitoring has been 411 defined using the quantified plumes coincident with S2 detections, as the three satellites have 412 approximately the same overpass time with a few minutes difference (between 2 and 5) in the 413 observations used. We have identified nine simultaneous plumes indicating that the detection limit 414 of S2 is close to 1800 kg/h (see Fig. S5). This relationship holds if the plume maintains  Figure S4, where hyperspectral sensors detect plumes on 2020-07-31 and 2020-09-11 that 419 S2 missed, i.e., S2 has not detected emissions with fluxes lower than 1800 kg/h that PRISMA and 420 ZY1 have with a few minutes difference. This detection limit value is slightly lower than Varon et al.

421
(2021) (27) indicated (~3000 kg/h) for the most optimal surfaces, as is the case in most of 422 Turkmenistan.

424
pinpoints (see Fig. S10), emission points detected in the ZY1 and PRISMA hyperspectral images 428 (see Fig. S4), O&G extraction fields in the SCB according to (35,36), pipeline crossings, flares that 429 in the past had shown an active flame, and mud volcanoes.

431
To detect CH4 emissions with S2, we have selected bands B11 and B12, with 20 m pixel resolution.

432
The B11 band extends over a set of weak CH4 absorption lines near 1650 nm, and the B12 band

442
The S2 detection figures shown in this paper ( Fig. 3 and Fig. S5) have been obtained applying the 443 B12 and B11 bands ratio of two contiguous days from the same satellite and with the same orbit 444 whenever possible, i.e., the equation described below but ensuring that the detection is taken by 445 the same satellite, S2A or S2B from the same viewing, on both days. In this way, we try to avoid 446 the increase of noise in the result due to miss-registration and viewing differences (45).

466
The 20m pixel resolution of S2 and multiple observations of plumes from the same source have 467 provided sufficient accuracy to identify the emission source.

469
We have obtained the results for L5 and L8 in the same way as S2, but in this case, the bands

481
We have estimated an integrated annual emission rate (Qa) from all 29 sources detected in this 482 study with S2. For this estimation, we rely on the Q values estimated for the single plumes obtained 483 from the hyperspectral data (Fig. S4) in order to obtain an average hourly flux rate ( ̅ ) 484 characterizing the emissions in the area. This average flux rate is scaled in time using an average 485 emission frequency number ( ̅ ) which is obtained from the S2 plume detections (O. E. % in Table   486 S1). The total annual emission rate is then given by: where N is the number of emitters, i.e., 29 emission sources.

494
This estimate is based on statistics from emission intensity and frequency data sampling the four

507
The identification of the sources was carried out by inspection of high-resolution visual images from 508 Google Earth, Bing Maps and Esri, depending on the acquisition date available for each area on 509 each platform.

510
In the initial approach of the study, we also considered mud volcanoes as possible sources of CH4 511 emission. However, after observing the different potential areas, it has not been possible to link any 512 of the observed plumes to a mud volcano.

513
In three cases, we were not able to identify the origin of the emissions due to lack of up-to-date

526
In Turkmenistan, we have detected emissions from all three types of flares. Throughout the study, 527 they have all been referred to as the same "flare" emitter type, although in Table S1, there is a 528 more precise classification separating them into the three groups.

12
The identification of the emitters, mainly flares, has been verified by the Carbon Limits group, which 530 has experience in field measurements in Turkmenistan.

534
Flaring can be detected by satellites with bands in the SWIR, due to the flame's strong signal in 535 that spectral region, with the emission peak at 1.6 µm (47).

537
In

551
The authors thank the team that realized the TROPOMI instrument and its data products, consisting 552 of the partnership between Airbus Defense and Space Netherlands, KNMI, SRON, and TNO, 553 commissioned by NSO and ESA. Sentinel-5 Precursor is part of the EU Copernicus program,

554
Copernicus (modified) Sentinel-5P data (2018-2020) have been used. We thank the Sentinel Hub 555 service for providing the EO Browser service, which was key to the development of the study.

556
Thanks to the Environmental Defense Fund (EDF) for providing data about the O&G fields of the 557 study area, and the Carbon Limits group for contributing to the verification of the emission sources.

558
We thank the Italian Space Agency for the PRISMA data used in this work. Dr. Yongguang Zhang 559 from the University of Nanjing is also thanked for his support to get access to ZY1 AHSI data, and

806
S2, and this data is used throughout the document to refer to the emission frequency. "Field" field 810 where it is located.