The Emissions Database for Global Atmospheric Research (EDGAR) v6.0 is a bottom-up inventory providing detailed information about global anthropogenic emissions of various air pollutants and greenhouse gases. The yearly emission data have a spatial resolution of 0.1°×0.1° and are available from 1970 to 2018. The emissions of a specific gas are calculated using international activity data and emission factors using the IPCC methodology. Activity data describe the activities producing emissions, such as the amount of fossil fuel that is exploited or the number of animals on a farm. Emission factors are coefficients that relate the emitted amount of a specific gas to a certain activity or process. The data required to calculate the emissions are collected from a variety of sources, including international organizations such as the International Energy Agency (IEA), national emission inventories and industry reports. EDGAR is well-suited to determine the anthropogenic source types of the detected potential since this inventory provides sector-specific emissions, which enables the differentiation between individual source types within the source regions. For methane, EDGAR v6.0 provides sector-specific anthropogenic emissions from, for example, enteric fermentation; landfills; rice cultivation; and fossil fuel exploitation, which is further separated into coal, oil, and gas emissions. We use the EDGAR v6.0 methane data for 2018.

The Global Fuel Emission Inventory (GFEI) v2.0 is a methane emission database providing global anthropogenic emissions for the fossil fuel sectors: coal, oil and gas. The emission data are gridded to yearly maps (2010–2019) with a resolution of 0.1°×0.1°. GFEI v2.0 uses fossil-fuel-related emission data reported by countries to the United Nations Framework Convention on Climate Change (UNFCCC); separated into the sectors of coal, oil, and gas; and assigned to the appropriate infrastructure locations like coal mines or oil and gas wells. The infrastructure data are taken from several databases. For countries that do not report their emissions to the UNFCCC, the emissions are calculated using the IPCC methods and activity data from the US Energy Information Administration (EIA). Due to the different methods and data used for emission quantification in EDGAR v6.0 and GFEI v2.0, both databases show differences in their fossil fuel emissions, especially on a regional scale. Therefore, GFEI v2.0 can be used as a useful supplementary database to assign the appropriate fossil fuel source type to the detected source regions. We use GFEI v2.0 data for 2019.

WetCHARTs v1.3.1 is a global wetland methane emission ensemble that provides monthly emissions with a resolution of 0.5°×0.5° for the time period of 2001–2019 . The ensemble is based on different wetland extent scenarios, multiple terrestrial biosphere models and various temperature dependence parameterizations, resulting in 18 different model configurations. We use WetCHARTs to also include wetlands as a potential dominant source type of a source region. To compare the wetland emissions from WetCHARTs with the other emission databases, we create a yearly averaged wetland emission map for 2019, with a resolution of 0.1°×0.1°, by averaging the emissions of all configurations and months.

A PPSR is characterized by the appearance of enhanced anomalies at a certain frequency over a certain time period. Therefore, to define a PPSR, we have to specify the term enhanced anomaly and to introduce variables to quantify how often the region shows enhanced anomalies. We define an anomaly as enhanced if 2ΔXCH4≥Nσ⋅σ. We set Nσ=2. The σ is the standard deviation of the XCH4 in the HPFA(n) around the analyzed grid cell (Sect. ).

To characterize the persistent enhancement of a certain region, e.g., consisting of several grid cells, we first define the number of months in which the region contains at least one anomaly (measurement) as Nmeas. In addition, the number of months in which the region contains at least one enhanced anomaly is defined as Nenh. As a measure for the persistence of enhancements, we define the fraction Fenh=Nenh/Nmeas, which characterizes in how many of the months with measurements at least one of the anomalies is enhanced. Figure a–c illustrates the calculation of these variables for a region of 3×3 grid cells. We define a region as a PPSR if 3Fenh≥Fenh,min,Nmeas≥Nmeas,min. The parameters Fenh,min and Nmeas,min define the lower limits of Fenh and Nmeas. We set Fenh,min=0.5 and Nmeas,min=16. This means that a region is defined as a PPSR if it contains data in at least 16 of the 48 months and also contains an enhanced anomaly in at least half of the months in which an anomaly is in the region.

We have chosen Fenh,min=0.5 for the following reasons. Persistent methane sources do not always show enhanced methane anomalies in all months. For example, some sources such as wetlands or rice paddies show seasonal variations in emissions. Emissions from coal mines can also vary over time, as they depend on mining activity. In addition, we also want to take into account persistent sources in the detection process that started emitting during 2018–2021 and therefore do not show emissions over the entire period. With Nmeas,min=16, we also take into account regions that do not contain data in all 48 months.

To detect PPSRs in an anomaly dataset, we define small areas around every grid cell of the dataset and calculate for each of those areas the number of months with at least one anomaly, Nmeas; the number of months with enhanced anomalies, Nenh; and the fraction of months with enhanced anomalies, Fenh, by analyzing the monthly XCH4 and σ maps from 2018 to 2021. In detail, for each grid cell, we apply the following steps, which are illustrated in Fig. . We first define an area of 3×3 grid cells consisting of the grid cell itself and the directly adjacent grid cells (black-outlined area in Fig. a). We are using a small 3×3 area for the calculation of Nmeas, Nenh and Fenh rather than only analyzing a single grid cell for the following reason. The ΔXCH4 enhancements within a persistent source region depend on the source itself and on the meteorological conditions. Therefore, enhancements show temporal and spatial variability. Consequently, the ΔXCH4 enhancements can occur at different grid cells in different months of the persistent source region. To account for this in the detection process, we analyze the ΔXCH4 and σ maps of multiple grid cells simultaneously rather than considering each grid cell independently. We use an area of 3×3 to take into account the fact that the varying meteorological situations in the monthly XCH4 maps are not as strong as in the daily XCH4 data. In the monthly maps, the daily plumes, which vary with wind strength and direction, typically average out and result in a XCH4 enhancement over the source region, which shows only slight monthly variability.

After defining the 3×3 area, we analyze all monthly ΔXCH4 and σ maps from 2018 to 2021 for the 3×3 area to calculate Nmeas, Nenh and Fenh (Fig. b and c). We also count the number of months, Nenhgc, in which the anomaly in the grid cell is enhanced for each grid cell in the 3×3 area. If the 3×3 area fulfills the persistence conditions from Eq. () and if the center grid cell shows an enhanced anomaly in at least 1 month, we mask the area as PPSR (yellow area in Fig. d). To label a 3×3 area as PPSR, we mark all grid cells within the area that show an enhanced anomaly in at least 1 month (Nenhgc≥1). Thus, grid cells with Fenh<0.5 can also be part of a PPSR if their enhancements contribute to the 3×3 area being marked as a PPSR. We only consider 3×3 areas that have no complicated topography (median of surface roughness <80m and standard deviation of the surface elevation <150m) as PPSRs. As can be seen in Fig. c, the analysis of an area rather than a single grid cell enables the detection of source regions in which the individual grid cells show no persistent enhancement, but the area does. This means that the enhanced anomalies need not occur at the same grid cell every month but can vary monthly within the area.

PPSRs that are directly adjacent or overlapping are merged into one PPSR (Fig. e and f). For this, we apply a labeling algorithm in which each individual PPSR is assigned its own number, with directly adjacent or overlapping PPSRs getting the same number. In the end, we get a global map containing the separated and labeled PPSRs of the anomaly dataset considered.

We apply the detection process to the five anomaly datasets and obtain five global maps with the detected PPSRs.

The detected PPSR masks describe the locations and shapes of the corresponding source regions. However, some of the masks do not cover the entire spatial extent of the source regions. Therefore, in the next step, we refine the PPSR masks. One example is shown in Fig. a. It can be seen that the two PPSR masks do not contain all the grid cells that would be identified by eye as part of the source regions because their fractions Fenh do not exceed the threshold Fenh,min=0.5 required for detection (Eq. ). These grid cells are nevertheless part of the source region since they have a high fraction of Fenh and are located in the immediate surroundings of the source regions. To add them to the source regions, we could lower the Fenh,min parameter. But this would imply a change in the persistence condition. To determine the total spatial extent of the source regions without changing the persistence condition, we choose the following approach. We add grid cells to the PPSR masks that are in the immediate vicinity and whose fractions Fenh indicate that they are part of the source. For this, we identify all grid cells with Fenh≥0.33 that also fulfill all other conditions from Eq. (). We refer to these grid cells as “toseeds” (green grid cells in Fig. b). The grid cells detected with Fenh,min=0.5 are called seeds (yellow grid cells in Fig. b). We chose 0.33 as the lower threshold, since Fenh≥0.33 indicates that the grid cells show enhanced anomalies in a certain number of months and are therefore still strongly influenced by the sources within the PPSR, although its Fenh is smaller than 0.5. Grid cells with Fenh<0.33 indicate a weaker influence of the sources on the grid cells, which is why we did not include them in the refining process. Next, we apply a random walker algorithm to assign the toseeds to the seeds. A random walker algorithm is an image segmentation algorithm, which can divide an image into several sections based on threshold values. A first threshold is used to define the pixels of the image that represent the foreground of the image and are called seeds (the grid cells detected with Fenh,min). The seeds can have different labels so that the foreground can be divided into different areas. With a second threshold, which is below the first one, the pixels of the background that are not to be considered further are defined. The pixels between the first and second threshold are the so-called undefined pixels that the random walker algorithm assigns to the corresponding seeds using a diffusion equation (the grid cells with 0.33≤Fenh≤0.5). Based on the gradient between an undefined pixel, the different seeds and the distance between them, the probability of which seed the respective undefined pixel is assigned is calculated. The lower the gradient, i.e., the more similar the values of the undefined pixel and a seed are, the higher the probability that this pixel will be assigned to this seed. Undefined pixels that do not have a contiguous path to at least one seed are discarded. As the basis on which the grid cells detected with Fenh,min are assigned to the PPSRs, we use the multi-year (2018–2021) ΔXCH4 of the analyzed anomaly dataset. Figure c shows the mask created by assigning the toseeds to the seeds. It can be seen that the spatial extent of the source regions is now better described by the masks and that grid cells connecting the separate source regions are added. But some of the toseeds have a low multi-year ΔXCH4 mean compared to the seeds. Here, we only want to consider toseeds that have comparable high multi-year ΔXCH4 as part of the source region and remove added toseeds with ΔXCH4 smaller than 25% of the maximum ΔXCH4 of the seeds. In the end, we obtain the refined PPSR masks, which now describe the spatial extent of the source regions better (Fig. d). We emphasize that the example shown in Fig. , in which two PPSRs are first merged and then separated, does not appear often. We only used it to illustrate all the steps of the refinement process for one region. Due to the refinement, the number of final PPSRs can differ from the number of PPSRs detected in Sect. . On the one hand, multiple PPSRs can be combined into one PPSR by adding new grid cells to the masks. On the other hand, a PPSR can be split into multiple PPSRs by removing grid cells with 2018–2021 ΔXCH4 means that are too low. We apply the refinement to each of the five global maps containing the detected PPSRs (Sect. ).

Much effort was made to minimize systematic biases when generating the WFMD v1.8 XCH4 data product . However, it is not guaranteed that the WFMD v1.8 product is entirely unbiased. This means that despite the good quality of the product, it is not certain that every individual XCH4 enhancement has its origin in a real methane source. For example, localized XCH4 enhancements could be caused by scenes with inhomogeneous albedo (e.g., coastal regions, lakes and rivers) and complex topography. To take this into account, the PPSRs are filtered for surface features that could potentially lead to a false positive detection. We use a conservative approach and prefer to accept false negatives rather than false positives. We decide whether a PPSR has challenging surface features based on the following properties: the correlation between SWIR surface albedo and XCH4, the standard deviation of the surface elevation within the PPSR mask, the frequency of months in which the largest XCH4 enhancements occur in or adjacent to grid cells with high surface roughness, the fraction of coastal grid cells in the PPSR mask, and the frequency of months in which the largest XCH4 enhancements occur over or next to water grid cells. If a PPSR is identified by one of these criteria, then it is filtered and not considered further. Excluded from this are PPSRs in which very strong XCH4 enhancements occur. By this, we ensure that important source regions are not excluded due to their surface features. As we focus in this study on source regions that contribute significantly to the global methane budget, we filter out PPSRs with weak XCH4 enhancements. Additionally, we filter out PPSRs that occur in the Bodélé Depression in Chad. This is a region where strong dust storms occur on average 100 d yr-1, always directed towards the southwest and with a plume-like structure. Analyses of the WFMD data product have shown that these special conditions, which only occur in this region, can lead to false-positive detections. We apply the filtering to each detected PPSR of each anomaly dataset to obtain five global maps comprising the refined and filtered PPSR masks.

The PPSR with the highest emission estimate for 2018–2021, called PPSR 1, is detected in the Sudd in central South Sudan, one of the world's largest wetlands. South Sudan, and in particular its wetland region, is a well-known methane source region that has been subject of several studies . By comparing the emission databases within PPSR 1 as described in Sect. , we determine its dominant source type as wetland, which corresponds to its location in the Sudd. In Fig.  we show an overview of the PPSR 1 results. Figure a shows the 2018–2021 XCH4 of the South Sudan region, including the detected PPSR 1 mask, as well as one other identified PPSR in eastern South Sudan. It can be seen that the XCH4 within PPSR 1 is strongly enhanced compared to its surroundings. The area outlined in black in Fig. a indicates the surrounding region, which is used to calculate the XCH4 enhancements δXCH4 of PPSR 1 (see Sect. ). The corresponding time series of the δXCH4 for 2018–2021 is shown in Fig. c. The mean for the entire time period is 12.9±1.3ppb, and the standard deviation is 10.3ppb. The δXCH4 shows a seasonal cycle with its peak enhancement at the end of each year, as well as a strong increase since the end of 2020. Due to the frequent occurrence of clouds during the wet season from April to November, only a few days with data are available for this period of the year. In Fig. b we show the emission estimates of PPSR 1 for 2018–2021, which we calculated as described in Sect. . The mean of the emission estimates is 4.5±0.9Mtyr-1, where ± indicates the long-term emission estimate uncertainty calculated via Eq. (). By comparing the time series in Fig. b–d, it can be seen that due to the small variations in the mean wind speed V, the δXCH4 variations determine the temporal variations in the emission estimates, including the strong increase since the end of 2020. This strong increase is in good agreement with the finding that tropical wetlands are a major contributor to the strong methane growth rate in 2020 and 2021 .

estimated the methane emissions of the entire wetland region in South Sudan, including the Sudd and other wetlands, to be 8.0±3.2Mtyr-1 for 2018–2019. In a study from , emissions of the Sudd region were estimated using GOSAT XCH4 data, resulting in 5.2–6.9 Mt yr-1 for 2016. Our estimate is lower compared to the two results, which can be explained by the smaller source region of this study. By combining PPSR 1 with the PPSR that we detected in the east of South Sudan (0.8±0.4Mtyr-1 for 2018–2021), we get a total emission estimate of 5.3±1.3Mtyr-1, which is in agreement within the uncertainties of the emissions calculated in and .

The emissions from the databases WetCHARTs v1.3.1, EDGAR v6.0 and GFEI v2.0 for the South Sudan region are shown in Fig. e–g. We compute the emissions of the databases in a PPSR by adding all emissions within the extended mask of the PPSR (see Sect. ), which include the directly adjacent outer grid cells of the PPSR, to consider possible source location variations in the databases. WetCHARTs emissions for 2019 in PPSR 1 are 0.88Mtyr-1. EDGAR emissions for 2018 for PPSR 1, which are mostly from the agriculture sector, combine to 0.17Mtyr-1 and the emissions from GFEI for 2019 are 0.01Mtyr-1. It can be seen that the emissions from the databases show a large difference from the emission estimates of this study and from those of and .

The PPSR with the second-highest emission estimate for 2018–2021, called PPSR 2, is detected on the west coast of Turkmenistan, in the Balkan province that borders the Caspian Sea. The dominant source type is determined as oil and gas. The west coast of Turkmenistan is a methane source region with oil and gas infrastructure over almost the entire coastal belt, including oil and gas power plants, compressor stations, and pipelines . An overview of the results for PPSR 2, as well as the mask that defines the PPSR, can be seen in Fig. . The mean emission estimate for 2018–2021 is 3.5Mtyr-1 with an uncertainty of 0.9Mtyr-1 and a standard deviation of 0.6Mtyr-1. All months except January and February 2018 contribute to the emission estimate. The mean of the δXCH4 for the time period is 17.5±1.4ppb and the mean wind speed 4.3±1.0ms-1, where ± indicates the corresponding uncertainties.

Methane emissions on the west coast of Turkmenistan have been detected in recent studies . In , areas on the west coast were identified as hotspot regions using TROPOMI, where hyperspectral (ZY1 and PRISMA) and multispectral (Sentinel-2) satellites detected several localized emission events in the range of kilotons per year from January 2017 to November 2020. In , a methane source was detected at a compressor station in Korpezhe, in the middle of the west coast of Turkmenistan. Using TROPOMI data, the total emissions within a 12×12km2 region around this source were calculated to be 0.45Mtyr-1 (0.19–0.75) for December 2017 to January 2019. The emissions calculated in these studies refer to individual events or to smaller regions of the west coast and therefore cannot be directly used for comparison with the emission estimates calculated in this study but provide an overview of the magnitude of the emissions.

The spatial distribution of methane emissions from EDGAR v6.0 for 2018 and GFEI v2.0 for 2019 for the region considered are shown in Fig. e–g. The emissions from EDGAR of 0.64Mtyr-1 and GFEI of 0.62Mtyr-1 for the entirety of PPSR 3 are significantly lower than our estimate of 3.5±1.8Mtyr-1. Several studies suggested that the inventories may underestimate Turkmenistan's emissions . For example, calculated emissions of 3.6±1.3Mtyr-1 related to oil and gas in Turkmenistan using TROPOMI, which are higher than the emissions reported by GFEI of 1.5Mtyr-1. If we add the mean emission estimates of all oil- and gas-related PPSRs in Turkmenistan, we get a total emission estimate of 5.0±1.4Mtyr-1, which is in agreement within the uncertainties in .

The PPSR with the third-highest emission estimate for 2018–2021, called PPSR 3, is detected in the region of the border between northeastern Argentina and southern Paraguay and is assigned to the wetland type. PPSR 3 is located in the northern part of the Paraná region, a well-known methane source region, which extends from the Iberá Wetland in the north, the second-largest wetland in the world, to the area where the Paraná River flows into the Atlantic Ocean . In Fig.  we show an overview of the results of PPSR 3. The mean emission estimate for 2018–2021 is 3.3±1.1Mtyr-1 with a standard deviation of 1.3Mtyr-1, and the mean of the corresponding δXCH4 is 8.9±1.9ppb. The emissions show a seasonal cycle, which also can be seen in the δXCH4 time series and which is in good agreement with the wet season . Furthermore, the emission estimates show a slight decrease from 2020 onward, which agrees with the results in , where methane emission changes between 2019 and 2021 are analyzed, including the emission changes in the Paraná region.

WetCHARTs v1.3.1 shows enhanced methane emissions for the entire Paraná region, especially for Iberá Wetland, whereas the anthropogenic databases indicate only low emissions (Fig. e–g). WetCHARTs emissions for PPSR 3 are 0.64Mtyr-1, which is below our emission estimate. Although the Paraná region is a known methane source region, until now, no studies have calculated the absolute values of the emissions from this region that we can use to further assess our emission estimates. For example, in , XCH4 retrieved from GOSAT observations is used to analyze how well the methane interannual variability is described by model simulations for several regions, including the Paraná, without reporting explicit emission estimates.

The PPSR with the fourth-highest emission estimate for 2018–2021, called PPSR 4, is detected in Liaoning province in northeast China and is assigned to type other anthropogenic. Liaoning is known for its high agricultural production (e.g., rice cultivation and livestock) as well as for its large heavy industry, including strong coal mining activities. The results of PPSR 4 are shown in Fig. . The PPSR mask covers the region of Liaoning province where most of the rice production takes place and where a majority of the coal mines are located . The mean emission estimate is 2.9Mtyr-1 with an uncertainty of 0.9Mtyr-1 and a standard deviation of 1.0Mtyr-1. The δXCH4 has a mean of 8.1±1.6ppb and shows strong variability over the years with a standard deviation of 2.5ppb, with the minimum usually in spring. In all months from 2018 to 2021, the PPSR as well as the background region are filled with sufficient XCH4 values to calculate the δXCH4 and the emission estimates.

So far, there are only a few studies that have analyzed or identified methane emissions in the region considered. For example, two plumes were detected in 2021 by , which are located in the PPSR, with one plume assigned to the coal source type and one plume assigned to the landfill type. In , coal-related emissions in 2011 for China, including the Liaoning region, were estimated by analyzing reports from over 10 000 coal mines in China. For Liaoning, the coal-related emissions were calculated to be 1.04Mtyr-1. The different time periods, as well as the larger region considered in , make it difficult to compare the results with the results of this study. In emissions of urban areas were estimated using TROPOMI data and compared with EDGAR, including the Shenyang region in Liaoning, where the emissions were estimated at 1.6Mtyr-1. If we take into account the fact that the Shenyang region is smaller than PPSR 5 and thus some emissions from the surrounding area are not included in the estimate, our result is in good agreement with that of .

It can be seen from Fig. e–g that anthropogenic emissions are the dominant source type in this region. Emissions from EDGAR for PPSR 4 are 1.3Mt in total for 2018, with large emissions seen in Shenyang, the capital of Liaoning. Of the 1.3Mt, 52% are from the category of other anthropogenic sources, which is composed of emissions from several sectors, such as rice cultivation or landfills. The remaining emissions from EDGAR are related to the fossil fuel sector, mainly to coal production, which is in the range of the fossil-fuel-related emissions from GFEI in 2019 for the PPSR of 0.49Mtyr-1. The emissions from the databases are significantly lower than the emissions calculated in this study of 2.9±0.9Mtyr-1, which is also reported in for their emission estimate of the Shenyang region.

The PPSRs with the fifth-, sixth- and seventh-highest emission estimates for 2018–2021, called PPSRs 5, 6 and 7, are detected in the Shanxi province in north China. The Shanxi province is a known methane source region with emissions resulting primarily from heavy coal mining activity (Peng et al., 2023). This corresponds to the dominant source type of the three PPSRs that was determined, which is coal. An overview of the results of the individual PPSRs is shown in Fig. . Figure a shows the 2018–2021 XCH4 for Shanxi and the surroundings, including the detected PPSR masks, as well as the corresponding background regions for PPSRs 5, 6 and 7. It can be seen that the XCH4 in the PPSRs is enhanced compared to the XCH4 in the surrounding regions. The time series of the δXCH4 for the PPSRs are shown in Fig. c. PPSR 5 has a mean δXCH4 for 2018–2021 of 25.1±2.5ppb, PPSR 6 of 20.6±1.8ppb and PPSR 7 of 22.3±2.2ppb, which are the highest mean δXCH4 values of all detected PPSRs. The δXCH4 shows a strong variability in all three PPSRs with standard deviations of 10.4ppb for PPSR 5, 6.5ppb for PPSR 6 and 4.8ppb for PPSR 7. This variability can also be seen in the emission estimates of the PPSRs shown in Fig. b. The mean emission estimates are 2.6±0.8Mtyr-1 for PPSR 5, 2.6±0.7Mtyr-1 for PPSR 6 and 2.4±0.7Mtyr-1 for PPSR 7, and in all three PPSRs almost all months contribute to the corresponding mean emission estimate.

Methane emissions in Shanxi have already been detected in several studies. The main focus was on the detection of individual plumes, which were identified, for example, by analyzing TROPOMI data as in and ; by data from the Worldview 3 satellite as in ; or by data from the PRISMA satellite mission as in . The detected transient plumes in these studies are not suitable for comparison with our emission estimates, which were evaluated for persistent hotspot regions for several years. But this is the case for the study by in which the coal-related methane emissions for the entire Shanxi region for the years 2019 and 2020 were calculated by inversion of TROPOMI data. estimated emissions for 2019 of 8.5±0.6Mtyr-1 and for 2020 of 8.6±0.6Mtyr-1. To compare, we computed the sum of the emissions of all the detected PPSRs in Shanxi (PPSRs 5, 6, 7 and one other PPSR with a mean emission estimate of 1.1±0.3Mtyr-1 for 2018–2021; see Fig. a) and obtained an emission estimate of 8.8±2.4Mtyr-1 for the period 2018–2021, which is in agreement within the uncertainties in the results from . Moreover, by considering the emission estimates for 2019 and 2020, we obtained 8.5±2.1Mtyr-1 for 2019 and 8.7±1.8Mtyr-1 for 2020 for the combined PPSRs in Shanxi. In , the entire Shanxi region is considered, while we only focused on parts of the region. However, if we assume that our identified hotspots in the Shanxi region contain the majority of methane emissions, the comparison of the two results is reasonable.

Figure e–g shows the methane emissions of WetCHARTs v1.3.1, EDGAR v6.0 and GFEI v2.0 for Shanxi and the surrounding area. It can be seen that the region is dominated by anthropogenic emissions. The emissions for 2018 from EDGAR are mainly related to coal production and are 1.2Mtyr-1 for PPSR 5, 2.8Mtyr-1 for PPSR 6 and 1.2Mtyr-1 for PPSR 7 in the corresponding extended PPSR masks. In total, the EDGAR emissions of all PPSRs in Shanxi combine to 5.2Mtyr-1, which is below our emission estimate of 8.8±2.4Mtyr-1 for 2018–2021. The emissions from GFEI for 2019 are mostly related to the coal sector and are concentrated in a few hotspots, which correlate with the locations of the detected PPSRs. For PPSR 5, the GFEI emissions are 1.5Mtyr-1, 2.5Mtyr-1 for PPSR 6 and 1.9Mtyr-1 for PPSR 7. The total GFEI emissions of the PPSRs considered are 5.9Mtyr-1, which is slightly higher than the emissions reported by EDGAR but lower than the emission estimates of this study and the study by .

The PPSR with the eighth-highest emission estimate for 2018–2021, called PPSR 8, is detected in a region encompassing Dhaka, the capital of Bangladesh, which is one of the most populated cities in the world. The dominant source type is determined to be other anthropogenic sources. Dhaka and the surrounding area are a known methane source region with the main sources being agricultural production (rice, livestock) and waste management (wastewater, landfills) but also with contributions from wetlands . The results for PPSR 8 are shown in Fig. . The 2018–2021 XCH4 shows a strong enhancement in the PPSR, especially in and around Dhaka, compared to the XCH4 of the surrounding area (see Fig. a). The δXCH4 values for 2018–2021 are shown in Fig. c, averaging to a mean of 21.4±2.0ppb, which is in the range of the enhancements of the PPSRs in the Shanxi region. For the years considered, no XCH4 is present for the period from March/April to October/November due to the monsoon season and the resulting frequent high cloud coverage. Figure b shows the emission estimates for 2018–2021 with a mean of 2.4±0.5Mtyr-1 and increasing values from October/November until April/May of the following year. This period is also one of two phases in which the rice is cultivated in Bangladesh. The first phase is in summer, which starts around June and ends in October with the harvest. The second phase is during the winter from November to April, when the fields are artificially irrigated .

Methane emissions in Dhaka have already been detected and quantified in several studies . used TROPOMI data to detect plumes worldwide and detected as many plumes as in any other urban area in Dhaka. The emissions from Dhaka are calculated in using TROPOMI data and a two-dimensional plume model, resulting in emissions of 1.3Mtyr-1, which is lower than our estimate of 2.4±0.5Mtyr-1. It must be taken into account that our region is larger than that of and can therefore include emissions from other cities in the surrounding area, as well as wetland emissions from the Ganges delta.

Figure e–g shows the emissions from WetCHARTs v1.3.1, Edgar v6.0 and GFEI v2.0 for the Dhaka region. For WetCHARTs, the emissions in the PPSR amount to 0.13Mt for 2019; for EDGAR, 0.92Mt for 2018; and for GFEI, 0.02Mt for 2019. The emissions from EDGAR are mainly from the agricultural sector, with 0.38Mtyr-1 from rice production and 0.15Mtyr-1 from enteric fermentation, and are lower than our calculated emission estimate. In , the calculated emissions were also higher compared to EDGAR. They concluded that part of the difference between EDGAR and their emission estimate is due to the fact that untreated wastewater is not taken into account, which can be a major factor, especially in very densely populated cities such as Dhaka.

The PPSR with the ninth-highest emission estimate for 2018–2021, called PPSR 9, is detected in the Kuznetsk Basin (also called Kuzbass) in southwestern Siberia, Russia. Its dominant source type is determined to be coal, which coincides with the fact that Kuzbass is one of the largest coal production areas worldwide . Figure  shows an overview of the results for PPSR 9. In the 2018–2021 XCH4 map shown in Fig. a, a strong enhancement can be seen in the entire PPSR mask compared to the XCH4 of the surrounding area. To quantify the XCH4 enhancements within the PPSR, we computed the monthly δXCH4 for the time period 2018–2021, which is on average 17.3±0.6ppb with a standard deviation of 6.6ppb. The mean emission estimate is 2.4Mtyr-1 with an uncertainty of 0.5Mtyr-1, which is computed from emission estimates of 30 months (Fig. b).

Even though the Kuzbass is one of the largest coal production areas worldwide, there is still a need for studies reporting methane emissions from this region. In , methane plumes are detected in this region but not discussed in more detail. Due to the limited number of studies, we only compare our emission estimate with the emissions from the databases, which are shown in Fig. e–g for the region considered. It can be seen that the emissions from the databases are dominated by anthropogenic activity and that the emission hotspots reported by EDGAR and GFEI show a high spatial correlation. EDGAR reports emissions of 1.6Mt for 2018 and GFEI of 1.4Mt for 2019, in which the emissions from both databases are mainly related to the coal sector. Compared to the emission estimate of this study, the emissions from EDGAR and GFEI are lower but still within the uncertainty range.

The PPSR with the 10th-highest emission estimate for 2018–2021, called PPSR 10, is detected in the Permian Basin in the USA and is assigned to the source type oil and gas. The Permian Basin is the most prolific oil field in the USA and is also a high-producing natural gas region, which is located in western Texas and eastern New Mexico. The Permian Basin consists of several sub-basins, including the Delaware Basin in the west and the Midland Basin in the east, where mostly non-conventional exploitation techniques, such as hydraulic fracturing, are used. An overview of the results for PPSR 10 are shown in Fig. . It can be seen that we detect two regions in the Permian Basin: PPSR 10 in the Delaware Basin and a PPSR in the Midland Basin, which shows the 13th-strongest emission estimate. Since the literature often refers to the emissions of the entire Permian Basin, we analyze these two PPSRs together. The monthly emission estimates for 2018–2021 are shown in Fig. b. The mean emission estimate for PPSR 10 is 2.2±0.6Mtyr-1 and 2.0±0.5Mtyr-1 for PPSR 13, which leads to a combined mean emission estimate of 4.1±1.1Mtyr-1 for 2018–2021 (taking into account the second decimal place). The δXCH4 time series for 2018–2021 for PPSRs 10 and 13 can be seen in Fig. c. The mean δXCH4 enhancement for PPSR 10 is 7.5±0.6ppb with a standard deviation of 3.3ppb and 7.2±0.6ppb for PPSR 13 with a standard deviation of 1.7ppb.

Methane emissions from the Permian basins have already been quantified in several studies . In the studies by and , emissions were calculated based on the TROPOMI/WFMD XCH4 data product. used a Gaussian integral method and estimated emissions of 3.2±1.1Mtyr-1 for the period 2018-2019, whereas calculated emissions of 3.0±0.7Mtyr-1 for 2019 using a divergence method. The emissions reported in the studies by , and are based on the operational TROPOMI data product and different inversion frameworks. calculated emissions of 2.9±0.5Mtyr-1 for the period from May 2018 to March 2019, whereas estimated emissions of 2.9±0.4Mtyr-1 for the period from May 2018 to February 2020 and of 3.7±0.5Mtyr-1 for the same period but with an adjusted prior. In , the period from May 2018 to October 2020 is considered and mean emissions of 3.7±0.9Mtyr-1 are reported, which is higher than the previous emission estimates. The emission estimate of 4.1±1.1Mtyr-1 for 2018–2021 calculated in this study is slightly higher than the emissions of the other studies presented but agrees within the uncertainties.

The emissions from EDGAR v6.0, GFEI v2.0 and WetCHARTs v1.3 are shown in Fig. e–g. For EDGAR, the emissions within the extended PPSR mask (see Sect. 3.6) are 1.2Mtyr-1 and 0.2Mtyr-1 for GFEI and relate to the oil and gas sector. The emissions of both databases are significantly lower than the emission estimates of this study and the other studies mentioned above. The emissions of these two databases also differ from one another.