Assessing the Relative Importance of Satellite-Detected Methane Superemitters in Quantifying Total Emissions for Oil and Gas Production Areas in Algeria Authors: S Naus, J D Maasakkers, R Gautam, M Omara, R Stikker, A K Veenstra, B Nathan, I Irakulis-Loitxate, L Guanter, S Pandey, M Girard, A Lorente, T Borsdorff, I Aben Published: 2023 ABSTRACT AbstractMethane emissions from oil and gas production provide an important contribution to global warming. We investigate 2020 emissions from the largest gas field in Algeria, Hassi R’Mel, and the oil-production-dominated area Hassi Messaoud. We use methane data from the high-resolution (20 m) Sentinel-2 instruments to identify and estimate emission time series for 11 superemitters (including 10 unlit flares). We integrate this information in a transport model inversion that uses methane data from the coarser (7 km × 5.5 km) but higher-precision TROPOMI instrument to estimate emissions from both the 11 superemitters (>1 t/h individually) and the remaining diffuse area source (not detected as point sources with Sentinel-2). Compared to a bottom-up inventory for 2019 that is aligned with UNFCCC-reported emissions, we find that 2020 emissions in Hassi R’Mel (0.16 [0.11–0.22] Tg/yr) are lower by 53 [24–73]%, and emissions in Hassi Messaoud (0.22 [0.13–0.28] Tg/yr) are higher by 79 [4–188]%. Our analysis indicates that a larger fraction of Algeria’s methane emissions (∼75%) come from oil production than national reporting suggests (5%). Although in both regions the diffuse area source constitutes the majority of emissions, relatively few satellite-detected superemitters provide a significant contribution (24 [12–40]% in Hassi R’Mel; 49 [27–71]% in Hassi Messaoud), indicating that mitigation efforts should address both. Our synergistic use of Sentinel-2 and TROPOMI can produce a unique and detailed emission characterization of oil and gas production areas.Keywords: methane, superemitter, diffuse, oil, gas, emissions, Algeria, satellite Introduction Atmospheric methane has contributed at least a quarter to anthropogenic global warming since 1750.1 The oil and gas (O/G) sector is responsible for around a quarter of anthropogenic methane emissions.2 Various studies3−7 have shown that O/G methane emissions have a heavy-tailed source distribution, with a few large point sources contributing a relatively large fraction of total emissions from O/G facilities. As addressing such large point sources can be cost-beneficial, they are an attractive target for climate mitigation.8 In terms of regional emissions, smaller sources also contribute significantly as an aggregate, and achieving optimal reductions in O/G methane emissions will require mitigation of these emissions as well.9−11 However, the relative contribution of superemitters versus other sources to total O/G emissions remains uncertain and will likely vary between regions. Here, we present a novel atmospheric emission inversion approach to estimate total area emissions along with the contributions from superemitters, which we apply to two important O/G production areas in Algeria. Satellite instruments that measure atmospheric methane can provide valuable information in remote areas on O/G emissions with regular revisit times that ground- or aircraft-based campaigns cannot match. This is especially important for O/G sources, since these are often intermittent7 and can be related to production activities in remote areas that are challenging to monitor. In recent years, the number of satellites that measure atmospheric methane has quickly expanded.12 In particular, the launch of the TROPOspheric Monitoring Instrument (TROPOMI) onboard Sentinel-5P has been a major step forward, as it is the first instrument with daily, global coverage that can detect individual superemissions above 8 t/h [e.g., ref (13)] due to its high precision and spatial resolution of up to 7.5 km × 5.5 km at nadir.14,15 However, in most cases it is not possible to link methane plumes detected by TROPOMI to individual facilities due to its spatial resolution. Satellite instruments with a higher spatial resolution can serve a complementary role.16,17 For example, Sentinel-2 provides global coverage every 2–5 days at 20 m resolution and, as such, can help to trace large plumes to their sources.18−20 However, due to its lower spectral resolution than TROPOMI (100–200 nm versus 0.25 nm), Sentinel-2 can only detect the largest point sources (>1–2 t/h; hereafter adopted as our definition of superemitters) over relatively uniform surfaces.20 Targeting, hyperspectral instruments like the GHGSat constellation21 or PRISMA22,23 provide additional information with better detection limits, but have limited coverage. Previous work has explored the synergy between these satellite instruments measuring at different spatial scales [e.g., ref (24)], but has not yet integrated the different data streams to estimate total area emissions for longer time periods. Estimating area emissions by combining these satellite data showing both individual superemitters and smaller sources observed as an aggregate (hereafter referred to as the diffuse area source, as in e.g., Duren et al.6) is not straightforward. To estimate area sources at regional scale, inversions of atmospheric transport models are often used [e.g., refs (25−27)]. However, it is challenging to capture emission plumes in these relatively coarse region transport model simulations due to limited model resolution, transport errors, and the large volume of satellite data that makes it time-intensive to verify transport model performance on individual emission events. On the other hand, studies that focus on emission events [e.g., refs (17,28−30)] or satellite-observed plumes in general8,13 do not integrate these estimates in satellite-constrained area estimates. Therefore, the long-term area emission estimates do not fully exploit the increased data density of new satellites, while event-focused work only informs on emissions that result in large, concentrated plumes. An accurate assessment of total area emissions for longer time periods, combined with the characterization of relative contributions from superemitting point sources and the diffuse area source, is essential for efficient emission mitigation of the overall emissions from the O/G sector. In this study, we estimate the total area emissions, including the contribution from superemitting point sources (>1–2 t/h; the Sentinel-2 detection limit) and the diffuse area source (individually <1–2 t/h; detectable with TROPOMI as an aggregate) for two Algerian O/G production areas in 2020: Hassi R’Mel and Hassi Messaoud. In 2020, Algeria was the 10th largest gas producer and the 16th largest oil producer globally,31 and provided 10% of the gas imports of the European Union (EU).32 Accurate characterization and mitigation of the methane emissions associated with these imports fits the intention of the EU to reduce their greenhouse gas emissions and to make all imports of fossil fuels into the EU subject to compliance with EU rules on emission monitoring and leak detection.33 Moreover, previous work has suggested relatively high emission intensities from oil production in Algeria (Masnadi et al).34 The two areas investigated in this work represent distinctly different production activities, so their combination provides insight into how source distributions might vary. Hassi R’Mel contains the largest gas field of Algeria (43% of national gas production in 202035), oil production facilities (16% of national production), and acts as a processing hub for other Algerian O/G fields. Hassi Messaoud is mainly an oil production area (36% of national oil production), where frequent detections of methane emission events have been reported.16,23,36 In this study, we incorporate methane concentrations from TROPOMI and facility-level plumes detected by Sentinel-2 in a transport model inversion to provide a detailed characterization of the area emissions, separating the contribution from individual superemitters (>1–2 t/h) and the diffuse area source. In this way, we produce emission estimates that provide information on the differences between an oil- and a gas-production-dominated area, as well as on the relative roles of the diffuse area source and superemitters in both. Materials and Methods To estimate methane emissions in the two Algerian production fields, we first use Sentinel-2 methane plume detections to identify individual superemitters. As such, we define superemitters as sources detectable with Sentinel-2 (>1–2 t/h). Based on Sentinel-2 methane enhancements, we use a mass-balance method to derive the emissions for unique detected plumes. We then perform a transport model (WRF-Chem) inversion based on TROPOMI data that includes the Sentinel-2 emission estimates as prior information on the superemitting point sources. Additionally, our simulations include the O/G methane emissions from the bottom-up Scarpelli et al.37 emission inventory, as a first estimate for the diffuse area source. The latter represent sources that are too small to be detected as individual sources in Sentinel-2, but which can be constrained as an aggregate in TROPOMI methane data, since TROPOMI has higher precision and integrates enhancements over a larger area. In the inversion, superemitters and the diffuse area source are optimized independently. TROPOMI and Sentinel-2 Satellite Data TROPOMI is an atmospheric monitoring instrument that provides daily global coverage of column-averaged, dry air mixing ratios of methane (XCH4; in ppb) at a resolution of 7.5 km × 5.5 km in nadir, based on measurements in the shortwave infrared (SWIR) spectrum at 2.3 μm.14,15 TROPOMI can detect large (>8 t/h) concentrated methane sources in a single overpass,8,13 and it frequently detects plumes in both of our study areas. We use version 2.04 of the TROPOMI-CH4 product.38 We correct for along-track stripes following the FMD approach of Borsdorff et al.39 and apply a custom data quality filter (qa >0.1; precision <10 ppb; cloud fraction <0.1; SWIR aerosol optical thickness <0.1; see Figure 1 panel A). Using these filters, we find 224 and 238 days in 2020 with sufficient coverage for emission estimation (>20 pixels per day) over Hassi R’Mel and Hassi Messaoud, respectively. The 2020-averaged TROPOMI methane field is shown in Figure 1A. We note that some of the large-scale methane gradients are ”artifacts” driven mostly by albedo gradients (e.g., in the northeast extending to the middle of Tunisia) and/or low coverage (near the coast). However, such effects in the two regions investigated here are small. Overview of Algeria and our study domains. Panel A: Average 2020 TROPOMI XCH4 data over Algeria. Panel B: Bottom-up O/G methane emissions in Algeria.37 Panels A and B show inner (solid) and outer (dashed) zoom domains used in the WRF-Chem simulations for Hassi R’Mel (blue) and Hassi Messaoud (red). Panels C and D: Zoom-in of bottom-up emissions37 in the two inner zoom domains: Hassi R’Mel (C) and Hassi Messaoud (D). Sentinel-2-detected superemitter locations are marked with colored circles. The Sentinel-2 mission is a constellation of two satellites, with the Multispectral Imager as their main instrument, which is a land and ocean surface imager with lower spectral (100–200 nm) but higher spatial resolution (20 m) than TROPOMI.18 Sentinel-2 provides global coverage with a revisit time of 2–5 days. Sentinel-2 has previously been used to study large methane point sources.19,24,29,40 Here, we retrieve XCH4 enhancements using the multiband-multipass (MBMP) method as described in ref (19), where enhancements are calculated using the ratios between two SWIR spectral bands (one with stronger methane absorption than the other) on the day of interest normalized with a reference day. For each scene, we take the median over the MBMP with the 10 best reference days rather than use a single reference day. As shown by Gorroñno et al.,20 use of multiple reference days can make resultant plume detections less prone to surface reflectance variations than using a single day. We select the 10 reference days as those with the most similar surface reflectance as the day of interest, based on the correlation with the day of interest in the red spectral band. Due to the use of only two spectral bands, Sentinel-2 can only detect methane enhancements (as opposed to absolute methane column concentrations) under favorable observation conditions as it is sensitive to surface albedo (e.g., buildings) and aerosol (e.g., smoke) features.20 For high-albedo and relatively uniform surfaces, as those investigated in this study, the Sentinel-2 methane plume detection limit has been estimated at 1 to 2 t/h depending on the wind conditions.20 To better understand the (limitations of the) Sentinel-2 data, we also used five observations from the methane-designated GHGSat-C1 satellite instrument over Hassi Messaoud. GHGSat observes methane at 25 m resolution over ∼10 km × 15 km spatial domains with a reported detection limit of ∼100 kg h–1.41 More details are provided in Supporting S3. While Hassi R’Mel and most prominently Hassi Messaoud are areas with the strongest and most frequent plume signals in TROPOMI and Sentinel-2, Algeria does have additional areas with O/G production (Figure 1B), where we observe methane plumes with TROPOMI and Sentinel-2. The largest of these are the fields south of the inner Hassi Messaoud domain (red square in Figure 1A,B). However, near this field we find frequent interference from aerosols that affects the TROPOMI methane product and would make an inversion more complicated than in the other two regions. Therefore, we focus on the aforementioned two areas in our current analysis. Sentinel-2 Plume Detection and Emission Estimates We use Sentinel-2 to detect emissions from individual superemitters and provide a first quantification of their emission rates using a mass-balance approach. For the detection of Sentinel-2 sources, both the Hassi R’Mel and Hassi Messaoud areas were explored, partly guided by plume detections in TROPOMI data as in Schuit et al.13 We note that detection of individual sources is limited by the Sentinel-2 spatial resolution (20 m) and that we cannot separate sources located closer than that. Once a superemitter source is detected with Sentinel-2, we construct a 2020 time series for that source by manually labeling all scenes as “with plume”, “no plume”, or “bad data” (e.g., cloudy days). For all scenes with a plume, we calculate an emission source rate using the integrated mass enhancement (IME) approach.4,42 We follow the approach described by Varon et al.,16 including their effective wind speed calibration with ERA-5 10-m wind speeds,43 but we apply a different plume masking. For our plume masking, we only consider pixels in a 1075 m (1000 m downwind and 75 m upwind of the source) × 500 m box downwind of the source. Since we find that the wind direction in the ERA-5 10-m wind data often does not align with the plume direction, we determine the plume direction based on the observed Sentinel-2 methane plume. To do so, we take the pixel with the highest methane enhancement within a 10 pixel (∼200 m) square around the source, and then we iteratively include adjacent pixels with methane enhancements higher than 1.8 standard deviations above the scene average. We then calculated the plume direction as an average of the angle between the source and these selected pixels, weighted by the pixel distance from the source. In this way, our IME calculation depends only on the ERA-5 10-m wind speed and not on the wind direction. Finally, we select all pixels in the resulting box that are 1.8 SD above the scene average as part of our own plume mask. The background enhancement is calculated as the average over a 500 m × 500 m box placed 150 m upwind of the plume. Figure S1 shows the resulting boxes and plume masks for one plume detection per source location. In the inversion, we use the Sentinel-2-derived emission estimates as a prior for the superemitting point sources, where we assume a fixed prior error of 50% (1-σ Gaussian) of the Sentinel-2 emission estimates. In reality, the uncertainty can vary per plume detection, with uncertainty increasing at lower wind speeds, and 50% is conservatively high.19 Therefore, we vary this error between 30 and 70% in the sensitivity ensemble (Section 2.4). WRF-Chem Transport Model We use the Weather Research and Forecast model coupled to Chemistry (WRF-Chem) version 4.144 to simulate the atmospheric transport of methane emissions in the Hassi R’Mel and Hassi Messaoud areas separately. For both areas, we include three nested domains centered on the respective O/G field, with each domain consisting of 100 by 100 grid cells with a resolution of 3, 9, and 27 km, respectively (Figure 1). We use methane fields from the Copernicus Atmospheric Monitoring Service (CAMS) at 0.25° × 0.25° resolution for initialization and as 6 hly boundary conditions.45 Meteorology in the WRF-Chem simulations is driven at the boundaries by meteorological fields provided by the National Centers for Environmental Prediction (NCEP).46 We sample the 3-D CH4 simulation output at the TROPOMI pixel locations and the vertical levels of the TROPOMI averaging kernel, after which we apply the averaging kernels to retrieve simulated XCH4 values that can be compared to the TROPOMI data in the inversion. For the quality of our inversion estimate, it is important to obtain a good match between simulated and observed plumes. Therefore, following Maasakkers et al.,30 we perform simulations for each domain with five different planetary boundary layer (PBL) schemes (Table S1; Figure 2). For all other settings, we use the CONUS physics suite. Additionally, we sample the simulations at 11, 12, and 13 UTC, which roughly covers the variation in the TROPOMI overpass times over the areas. This approach produces an ensemble of simulated scenes (15 per day) from which we select the optimal plume match on each day for the inversion (Section 2.4 and Figure 2). Large methane plume as observed by TROPOMI on January 4, 2020 (panel A), regridded to 0.2° resolution (panel B). Panels C–G show the corresponding five WRF-Chem simulations with different PBL schemes (sampled at 12 UTC). To select the best-performing PBL scheme, we select the scheme with the lowest posterior cost for that day: in this case, PBL scheme 2 (panel C; the Mellor–Yamada–Janjic TKE scheme). Prior emissions for the WRF simulations are based on Sentinel-2 for superemitting point sources (>1–2 t/h; the Sentinel-2 detection limit) and on Scarpelli et al.37 for diffuse area emissions, even though the latter is supposed to encapsulate total emissions. Scarpelli et al.37 provide a global oil, gas, and coal methane emission inventory for 2019 at 0.1° × 0.1° resolution. National total methane emissions are consistent with those reported to the United Nations Framework Convention on Climate Change (UNFCCC), extrapolating to 2019 when only historic reports are available, and are spatially distributed using a variety of infrastructure data sets. For Algeria, 95% of O/G methane emissions are reported to come from gas exploitation and distribution, and only 5% from oil activity.47 Inversion Framework We obtain our optimized emission estimate from an inversion where prior emissions (individual superemitters and the diffuse area source) and the CAMS background are scaled to best reproduce TROPOMI data. For this, we solve the analytical Bayesian problem48 of minimizing a cost function J to retrieve an optimal state vector x̂ as where x̂ is the optimized state vector, xA is the prior state vector, SA is the prior error covariance matrix, K is the Jacobian, y is the observational vector, and SO is the observational error covariance matrix. The state vector x includes per-orbit scaling factors for the CAMS background, daily emissions for each point source, and monthly diffuse emissions. We subdivide the Scarpelli et al.37 inventory in the inner domain in a 6 × 6 grid (36 elements; ∼ 50 km × 50 km each) and in the middle and outer domain in a 4 × 4 grid, optimizing each grid element independently. We optimize the CAMS background per orbit because on some days the domain is covered by two TROPOMI overpasses, and the gradients between these overpasses are not always captured in the CAMS boundary conditions. We assume a 50% error on the Sentinel-2-based emission estimates (see Section 2.2), as well as on the prior diffuse emissions. Here and in previous work,30 we find that the CAMS background needs to be adjusted upward by 3–5% relative to the prior, and we therefore assume a 10% error on the CAMS background. We assume a fixed observational error of 15 ppb on individual TROPOMI XCH4 columns, which is based on the standard deviation of the difference between the prior model and the observations. We aggregate TROPOMI data to a 0.2° × 0.2° grid for the model observation comparison to reduce the impact of small differences between simulated and observed plumes (Figure 2A,B). We assume that prior errors (SA) and observational errors (SO) are uncorrelated. Both prior errors and observational errors reported here are 1-σ Gaussian. To ensure the best match between simulated and observed plumes, we perform WRF simulations with 5 PBL schemes (Section 2.3). To find the best PBL scheme per day, we performed two sets of preliminary inversions. First, we perform an inversion for 2020 based on one PBL scheme to get optimized monthly diffuse emissions. These are then used in daily inversions for each PBL scheme and each sample time step (for a total of 15 inversions per day), to select the PBL scheme producing the best match with TROPOMI for each day, based on the lowest posterior cost (Figure 2). Additionally, in selecting the optimal PBL scheme, preference was given to the daily inversions with positive posterior point source emissions. We then perform the final inversion in which we optimize the full state vector with the Jacobian K constructed using one optimal PBL scheme and sampling time for each day. We remove days in which the final inversion produces negative point source emissions from the analysis (caused by a poor fit to the observations), which has only a small effect (<5%) on the annual average source rates. To estimate uncertainties, we report the range of a sensitivity ensemble of different inversion configurations (Table 1). We find that the optimized emissions are especially sensitive to scaling the priors. Therefore, we independently vary the Sentinel-2-derived and the Scarpelli et al.37 priors in our ensemble with ±50% and we also vary the prior error between 30 and 70%. Additionally, in some months (e.g., May 2020 in Hassi R’Mel), monthly emissions are disproportionately influenced by observations from a single day, and removing this day from the inversion results in significantly different monthly emissions. We therefore include a sensitivity inversion, in which we exclude the day with the largest contribution to the cost function J. Finally, we add a sensitivity inversion where we use the second-best-performing PBL for each day. Reported posterior uncertainties cover the full range of the sensitivity ensemble. Table 1Ensemble of Inversions That Is Used to Quantify Posterior Uncertaintiesaparameterreferencelowhighscale superemitters1.00.51.5scale diffuse area source1.00.51.5prior error emissions50%30%70%aggregation resolution0.2°0.1°0.3° Other Ensemble MembersExclude for each month the day with the most observational weight.Use the second-best PBL scheme for each day in the optimization.aPrior emissions and their errors, as well as the aggregation resolution, are perturbed discretely to their low, reference, and high value (Table 1), i.e., three variations for each. For the sensitivity ensemble, we include all permutations of the different components. Reported posterior uncertainties cover the full range of the sensitivity ensemble. Ensemble of Inversions That Is Used to Quantify Posterior Uncertaintiesa Prior emissions and their errors, as well as the aggregation resolution, are perturbed discretely to their low, reference, and high value (Table 1), i.e., three variations for each. For the sensitivity ensemble, we include all permutations of the different components. Reported posterior uncertainties cover the full range of the sensitivity ensemble. To assess the performance of the inverse system and the representativeness of the sensitivity ensemble, we performed an Observation System Simulation Experiment (OSSE), where we used one PBL simulation as truth and estimate emissions based on the other four. The OSSE is especially useful to assess the influence of transport errors on the final emission estimate. For example, a mismatch between simulated and observed plumes can lead to a large error in estimated emissions but does not necessarily result in a large spread in the sensitivity ensemble. In the OSSE, we find that daily estimated (point source) emissions have large errors (50%), but these errors mostly average out on monthly and annual scales of 25 and 15%, respectively. Superemitters are generally underestimated (on average by 20%), while the bias for the diffuse area source is much smaller. This is because point source plumes are very concentrated and therefore difficult to capture in the simulation, while diffuse plumes are more likely to have at least some overlap with the observed plumes. This result implies that in our approach we give a conservatively low estimate of the contribution from superemitters. At the same time, the uncertainty in estimated total and diffuse area emissions is limited by TROPOMI’s precision as well as the inversion model. A full description of the OSSE analysis is given in Supporting S1. Results and Discussion Sentinel-2 Detections and Emission Estimates In our Sentinel-2 analysis, we detect one superemitter source in the Hassi R’Mel field and 10 in the Hassi Messaoud field (Figures 3 and S1), with a combined total of 347 plume detections in 2020. Since two Hassi Messaoud sources are close together (1.1 km), we group them as one cluster (M03) in our analysis. The two sources are still spaced apart sufficiently to quantify their emissions independently using Sentinel-2. The IME-based emission estimates (Figure 3) show that during 2020 some sources emit on nearly every day with Sentinel-2 coverage (e.g., at M03a, we detect plumes on 90% of days with coverage; see also Table S3), while others emit above the Sentinel-2 detection limit only on a few days (e.g., 1 day at M07). We note that where we quantify emissions as zero, it is still possible that a source emits below the Sentinel-2 detection limit. By far the largest emissions from a single source are detected at M06 on January 4 (29.8 ± 14.9 t/h) and January 7 (68.4 ± 34.2 t/h), while the January 9 Sentinel-2 overpass reveals flaring at this location (also reported in Pandey et al.24). TROPOMI shows large plumes from January 3 to January 8. This event demonstrates the complementarity of the Sentinel-2 high-spatial resolution with daily TROPOMI coverage. Source M08 has been described in Varon et al.19 as emitting methane consistently over a 10 month period, ending in August 2020, which is confirmed in our analysis, and we produce emission estimates consistent with theirs (Figure S1). Daily source rate estimates are based on an IME analysis of the Sentinel-2 methane data. The top and bottom panels show methane plumes as observed by TROPOMI (left, wind vectors showing 10-m ERA-5 winds43), Sentinel-2 (middle), and Bing imagery (right, ⒸVexcel Imaging, 2023) for 1 day in Hassi R’Mel (top) and Hassi Messaoud (bottom) that have both Sentinel-2 and TROPOMI coverage. For the emission time series, we linearly interpolate between days with Sentinel-2 coverage. Emission time series are shown as the sum of all sources in each respective domain, where colors differentiate between the contributions of the different clusters of superemitting point sources (also shown in Figure 1). On days for which we quantify emissions as zero, it is possible that a source still emits below the Sentinel-2 detection limit. Sentinel-2 can only be used to detect the largest superemitters (>1–2 t/h20). Inspection of targeted GHGSat data, which has a better detection limit than Sentinel-2, reveals an additional source location near M06 (see Supporting S3). The emission rates calculated from the GHGSat data are below or close to the Sentinel-2 detection limit (0.4–1.2 t/h). Sentinel-2 does show plume signals at this location on 6 days in 2020 (no overlap with GHGSat detections), but since this source was identified after performing the WRF simulations, it is not included in the rest of our analysis. This result shows both the potential importance of large point sources emitting below the Sentinel-2 detection limit, included in this work in the diffuse area source, and the challenge in comprehensively identifying all emission signals in Sentinel-2 over a large area. Based on visual imagery (mostly from the ESRI World Imagery49), we identify 10 locations as flares and one as a production well (M06; see also ref (24)) (full list in Table S3). Since most sources are found in the oil-production-dominated Hassi Messaoud area, our results indicate that most emissions from detected superemitting point sources come from unlit flares related to oil activity. When we compare emissions from Scarpelli et al.37 to the superemitter emissions (in Supporting S4 and Figure S4), it is clear that while the two source categories largely cover the same areas, the superemitters are extremely localized. Their annual-averaged source rates are often larger than the diffuse area source aggregated over 0.1° × 0.1° grid cells, and the temporal variations shown in Figure 3 would increase the differences as Scarpelli et al.37 only report annual average emissions. Therefore, we consider their inclusion in our inverse system extremely valuable. This is expected, as superemitters tend to not be included in bottom-up inventories. Hassi Messaoud and Hassi R’Mel Emission Estimates Our inversion finds a large correction to the CAMS boundary conditions (3.4% on average) but also significantly improves the agreement between model and TROPOMI through corrections to the emissions, both in terms of mean absolute bias (reduced from 59.13 to 0.05 ppb) and correlation (r2 = 0.85 when only the optimized variations in CAMS are included; to r2 = 0.89 when emission-optimized variations are also included). Figure 4 and Table S2 show the emission estimates for Hassi R’Mel and Hassi Messaoud, where we distinguish between superemitting point sources and the diffuse area source. The two regions show different adjustments from the inversion: in Hassi R’Mel (gas-production-dominated), prior emissions are significantly too high with and without point sources, while in Hassi Messaoud (oil-production-dominated), they are approximately correct. The prior (mainly diffuse) emissions in Hassi R’Mel are high (0.39 Tg/yr), which would produce large methane signals in TROPOMI. Since we find only a single point source (5.3 t/h; see Table S3) in Sentinel-2 and small methane enhancements in TROPOMI, we consider the strong downward adjustment to 0.16 [0.11–0.22] Tg/yr highly plausible. Our posterior emission estimate for the diffuse area source in Hassi Messaoud (0.11 [0.06–0.15] Tg/yr) aligns well with the bottom-up prior emissions from Scarpelli et al.37 However, the Scarpelli et al.37 inventory is supposed to represent all O/G emissions, and if we include superemitters (0.11 [0.06–0.17] Tg/yr), we find 79 [4–188]% higher emissions. Prior (dashed lines) and posterior (solid lines) methane emission estimates for superemitters (purple, referred to as point) and the diffuse area source (orange, referred to as diffuse) in Hassi R’Mel and Hassi Messaoud. Superemitters represent those sources that were individually detectable in Sentinel-2 (>1–2 t/h individually), and the diffuse area source represents the aggregate of smaller sources (<1–2 t/h individually). Left panels show the emission time series over 2020, with daily emissions from superemitters and monthly emissions from the diffuse area source. The right panels show the annual mean estimates for each category, as well as their combined total. Uncertainty margins indicate the full range of the sensitivity ensemble, while the solid lines indicate the result from the reference inversions. Following the NGSI Methane Emissions Intensity Protocol (NGSI;50 as in e.g., Schneising et al.51), we calculate a methane intensity based on the combined oil and gas production in each area. We convert gas production and methane emissions to barrels of oil equivalent (boe) using a conversion factor of 6000 cubic feet per boe.52 Using this approach, we calculate methane intensities of 0.5 [0.4–0.7]% and 1.1 [0.7–1.4]% for Hassi R’Mel and Hassi Messaoud, respectively. Johnson et al.53 report a methane intensity of 0.38% for upstream emissions from a gas field in British Columbia, Canada, which is similar to what we find for Hassi R’Mel. They additionally recalibrated emission intensities reported across a range of studies using a consistent set of assumptions. This recalibration shows that other work report higher intensities: e.g., 1.45–1.94% (2015 U.S. average; Alvarez et al.54), 1.08%–1.52% (2019 U.S. average; Shen et al.55), and 0.55–5.59% for five different U.S. basins.51 The lower values in Johnson et al.53 and Hassi R’Mel could indicate a difference between areas with mainly gas exploitation and fields with mixed oil and gas activity. Generally speaking, the intensities we find for Algeria are on the lower end of those found for other countries, although still significantly higher than international intensity targets (e.g., 0.2%56). Previous work has also combined methane emission estimates with marketable gas production data to compare different areas (e.g., Alvarez et al.54). Using this approach, we find methane intensities of 0.9 [0.7–1.1]% and 116.6 [70.0–148.8]% for Hassi R’Mel and Hassi Messaoud, respectively (assuming 80% methane content for natural gas). We note that these methane intensities are based on marketable production:35 gross gas production might be higher, which would result in lower intensities. This result emphasizes that the first approach is more robust for comparison between areas with combined oil and gas activities, with the above-100% leakage of Hassi Messaoud, which is likely largely related to oil activity, as a good example. When we assume that O/G methane emissions in Algeria scale linearly with production, and we combine national total O/G production with the O/G production in Hassi Messaoud and Hassi R’Mel,35 we find national 2020 emissions of 0.53 [0.39–0.66] Tg/yr (oil) and 0.17 [0.09–0.28] Tg/yr (gas) for a total of 0.70 [0.47–94] Tg/yr, where we distinguish between oil and gas based on the O/G production ratio in the two domains. To test the sensitivity of the upscaling procedure, we alternatively use the ratio between the sum of prior O/G emissions in the two regions and national total O/G emissions.37 Using this ratio, we estimate Algerian emissions at 0.89 [0.56–1.18] Tg/yr. For the upscaling, we only included the upstream (exploration, production, and processing) and midstream (transmission and storage) sectors of prior emissions: Scarpelli et al.37 estimate downstream emissions at 0.18 Tg/yr, which should be added to our estimates to get total oil/gas emissions. Both our estimates are consistent with Algeria’s last UNFCCC-reported O/G methane emissions of 1.0 Tg/yr for 2000.47 However, our results suggest that substantially more methane emissions come from oil production (∼75% of total emissions) than suggested in this report (5%). We note that national reporting updated for more recent years would make this comparison more insightful. If we upscale using prior emissions based on only Hassi R’Mel or Hassi Messaoud, the best estimate changes from 0.89 to 0.61 or 1.60 Tg/yr, respectively, which indicates the significant uncertainty of estimating Algerian emissions based on a subset of fields. While it is difficult to indicate how representative these two fields are for the whole of Algeria, they do cover approximately 50% of both oil and gas production,35 which lends credibility to the upscaling. Previous satellite-based transport model inversions have estimated Algerian methane emissions at 1.82 [1.68–1.96] Tg/yr for 201757 and at 3.7 ± 0.3 Tg/yr for 2019,58 based on GOSAT, and at 3.5 [2.4–4.4] Tg/yr for 2019, based on TROPOMI.59 Attribution of these national total emissions to different source sectors differs significantly: Western et al.57 attribute 80% of total Algerian emissions to O/G activity in the prior, but do not revise this percentage based on the inversion; Worden et al.58 estimate O/G emissions at 3.2 Tg/yr; and Chen et al.59 estimate lower O/G emissions at 2.09 Tg/yr and separately estimate O/G upstream and midstream emissions at 1.60 and 0.29 Tg/year, respectively. In general, these emission estimates are somewhat higher than what we report here, but as indicated above, our national upscaling based on only two fields introduces additional uncertainty. Moreover, Algerian oil and gas production in 2020 were lower than in previous years (e.g., by 4% (gas) and 8% (oil) compared to 201935). We also use the latest TROPOMI version (2.04), which has improved accounting for albedo gradients such as those seen near Hassi R’Mel,38 which can introduce differences between studies. While we do consider our emission estimate conservatively low (e.g., based on the OSSE; see Supporting S1), we find an overall good posterior match between the annual-averaged simulated methane concentration fields and TROPOMI (annual mean bias of 0.05 ppb), which provides confidence in our emission estimate. To better compare our emission estimate to more conventional area inversions that only optimize a traditional emission inventory at a coarser resolution, we additionally performed an inversion in which we coarsen TROPOMI to 1.0° × 1.0° resolution, include only Scarpelli et al.37 emissions, and optimize the background monthly. When we do this, we find posterior emissions comparable to our reference inversion (∼10% difference; see Table S3), which confirms the robustness of our emission estimates. Part of the novelty in our approach comes from the distinction between superemitters (>1–2 t/h individually) and the diffuse area source (representing sources <1–2 t/h individually). We find that in Hassi Messaoud superemitters contribute more to the total emissions (49 [27–71]%) than in Hassi R’Mel (24 [12–40]%) and the diffuse area source contributes significantly to the total regional emissions in both regions. Some of the Sentinel-2-detected sources emit frequently, and addressing these sources is a logical target for emission mitigation. The posterior differentiation between superemitters and the diffuse area source is in large part informed by the Sentinel-2 analysis, since distinguishing between them in the TROPOMI-based inversion is difficult. For example, the prior uncertainty in the superemitter contribution to total emissions is only reduced by ∼20%. This is in part because we coarsen the TROPOMI data to 0.2° × 0.2° resolution, which reduces the difference between super and diffuse signals in the TROPOMI data. In addition, the differences between signals from superemitters and the diffuse area source are often smaller than the model transport errors. This might be different in other areas where production is more spread out (e.g., the Permian Basin11,60). The advantage of our approach, however, is that we integrate the information from Sentinel-2 and TROPOMI in one consistent statistical framework, where Sentinel-2 provides information on the importance of superemitters and TROPOMI on the total area emissions. This is an improvement on previous work, where either only (gridded) area emissions are estimated61,62 or point source emission estimates are compared to national inventories to estimate their relative contribution.8 The resulting emission estimates from our methodology provide valuable and unique information on total area emissions as well as the contributions of the diffuse area source and superemitters, and their temporal variability, where we acknowledge that this variability as derived in our approach still includes significant uncertainties. The developed system can be applied to other emission regions, realizing that each area will have its own challenges with respect to the satellite data that need to be considered, and our approach can be aided by automated plume detection.8,13 The availability of global data from both TROPOMI and Sentinel-2 makes such an approach feasible in many areas. Especially regions with low cloud cover and uniform surface albedo are well-suited to satellite-based analyses (e.g., Libya, Iran), although satellite detection of emissions is also possible in, e.g., more northern regions (e.g., Schuit et al.13). For Algeria, we find that emissions in Hassi R’Mel, the largest Algerian gas field, are overestimated in Scarpelli et al.37 by 53 [24–73]% and emissions in Hassi Messaoud (oil-dominated) are underestimated by 79 [4–188]%. Our Sentinel-2 analysis shows that most superemitters in the two Algerian fields are unlit flares (10 out of 11). The difference between the relative contribution from superemitting sources (>1–2 t/h) for the two fields highlights the importance of assessing this distribution for individual areas around the world. Since the majority of emissions in both areas are below the Sentinel-2 detection limit (1–2 t/h) individually, other approaches are needed to characterize these emissions. Designated methane satellites such as GHGSat with a better detection limit than Sentinel-2 can play a role here (see Supporting S3), while satellites with an intermediate resolution (100–500 m), such as Sentinel-3 and the upcoming MethaneSAT mission, can bridge the gap between TROPOMI and individual source detections [e.g., ref (24)]. A full emission characterization of each area requires a ground-based campaign using vehicle or aircraft measurements. Importantly, our results indicate that a large part of the emissions is persistent in time, so that a snapshot provided by a measurement campaign is likely to capture the majority of sources. A satellite-based approach with daily global coverage can identify areas that will benefit most from ground-based work and provide context to the more detailed snapshots provided by such campaigns [e.g., ref (63)]. Information about the emission distribution per O/G production area is highly relevant for effective emission mitigation. For Algeria, we find that addressing the largest superemitters can significantly reduce O/G emissions. However, the majority of emissions in both areas comes from sources that are individually below the Sentinel-2 detection limit (1–2 t/h), which means that to optimally reduce region-wide emissions, mitigation of these sources will also be needed and will likely require general infrastructure improvements beyond addressing the largest superemitters. Acknowledgments The authors thank the team that realized the TROPOMI instrument and its data products, consisting of the partnership between Airbus Defense and Space Netherlands, KNMI, SRON, and TNO, commissioned by NSO and ESA. Sentinel-5 Precursor is part of the EU Copernicus program, Copernicus (modified) Sentinel-2 and Sentinel-5P data (2020) have been used. The authors thank SURF (www.surf.nl) for the support in using the National Supercomputer Snellius. S.N. and B.N. acknowledge funding through the UNEP CCAC Methane Studies and the Environmental Defense Fund (EDF). A.L. and T.B. acknowledge funding through the TROPOMI national program through NSO. Supporting Information Available The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acs.est.3c04746. Results from the Observation System Simulation Experiment, more details on the Sentinel-2 analysis and GHGsat detections, and a grid-level comparison between the Scarpelli et al. (2022) emission inventory and the satellite-detected superemitters (PDF) Results from the Observation System Simulation Experiment, more details on the Sentinel-2 analysis and GHGsat detections, and a grid-level comparison between the Scarpelli et al. (2022) emission inventory and the satellite-detected superemitters (PDF) Author Contributions S.N. and J.D.M. developed the conceptual ideas for the study. S.N. performed the TROPOMI/WRF analysis with input from B.N. and J.D.M. R.G. and M.O. provided the production analysis and supported the contextualization of the project. R.S. and A.V. performed the Sentinel-2 analysis with input from SP, I.I.-L., and L.G. A.L. and T.B. provided the TROPOMI data and associated support. S.N. wrote the manuscript with inputs from all co-authors. The authors declare no competing financial interest. Supplementary Material es3c04746_si_001.pdf References Associated Data Supplementary Materials es3c04746_si_001.pdf