For comparison with the G2L2 product, we used the GOSAT TANSO-FTS SWIR Level 2 CH₄ column-amount product (version V03.05; hereinafter G1L2). The V03 series provides XCO₂ and XCH₄ retrieved from GOSAT Level 1B spectra using the full-physics method . Version V03.05 corresponds to the bias-corrected product validated against TCCON measurements, and only data meeting the product quality criteria were used in this study.

Following the launch of GOSAT-2 in October 2018 and completion of its initial calibration phase, the G2L2 product has been available since March 2019. Several short interruptions in TANSO-FTS-2 operations occurred during the period, with the full list provided by the EORC-JAXA archive.

Monthly variations in the number of quality-screened XCH₄ observations are shown in Fig. . To ensure comparability between TANSO-FTS and TANSO-FTS-2, repeated soundings over the same location within short time intervals were counted only once per day. Such repeated soundings mainly arise from targeted observation operations, in which the same location may be observed multiple times to increase the likelihood of obtaining cloud-free retrievals. After this adjustment, TANSO-FTS-2 still yielded roughly 2–3 times more valid observations than TANSO-FTS over both land and ocean, reflecting improvements in sensor performance, pointing strategy, and observational coverage.

Figure shows the spatial distribution of G2L2 observations. TANSO-FTS-2 achieved substantially enhanced global coverage, particularly over the oceans, where nearly all 2.5° grid cells contain valid retrievals. Over land, the Intelligent Pointing system increased the sampling of cloud-free scenes relative to TANSO-FTS, although persistent cloudiness in the deep tropics remains a limiting factor. These improvements reduce spatial sampling gaps and provide a more uniform global dataset, forming a stronger observational basis for the CH₄ flux estimation presented in later sections.

The retrieval counts shown here represent only successful observations included in the GOSAT and GOSAT-2 Level 2 products and therefore reflect the combined effects of orbital sampling, observing strategy, cloud screening, and retrieval performance.

Uncertainties in GOSAT-2 XCH₄ retrievals arise from surface reflectance, aerosol and cirrus properties, instrument characteristics, and limitations in the forward-model representation. Validation of the G2L2 retrievals has been reported previously. evaluated version V02.00 through comparisons with TCCON and the G1L2 product, while the provides validation results for the version used in this study, V02.10.

According to the , validation of the G2L2 V02.10 product against TCCON shows small negative biases over both land and ocean, with larger biases over ocean sites. Typical mean differences fall within a few ppb for land and around -5 to -8 ppb for ocean, with standard deviations of approximately 10–15 ppb depending on the site and collocation criteria. These results indicate that G2L2 V02.10 generally underestimates XCH₄ relative to TCCON, particularly over ocean scenes. No bias correction is applied in this study; therefore, the flux estimates presented below reflect the original characteristics and remaining retrieval biases of the G2L2 V02.10 product.

The comparison of G1L2 and G2L2 products is subject to inherent limitations. First, the observation locations of TANSO-FTS and TANSO-FTS-2 do not perfectly coincide, making a direct one-to-one comparisons challenging. Second, differences in their retrieval algorithms can produce systematic variations in retrieved XCH₄ even when the measurements are taken at nearly the same time and location.

Figure presents two-dimensional histograms comparing XCH₄ retrieved from TANSO-FTS (x-axis) and TANSO-FTS-2 (y-axis). Matchups were selected when the fields of view were within 20 km and the observation times were within 30 min. Repeated soundings at the same location within consecutive time intervals were averaged, and only the closest matchup pair was retained.

Over land, the mean difference (G2L2–G1L2) is -2.66 ppb with a standard deviation of 11.53 ppb (N=12799), whereas over ocean the mean difference is -6.84 ppb with a standard deviation of 10.74 ppb (N=3614). Correlation coefficients exceed 0.93 for both surface types, indicating that the two products show high consistency in XCH₄ variability. Combined with the TCCON validation results in Sect. 3.2, these comparisons confirm that G2L2 XCH₄ is generally slightly lower than G1L2 and TCCON, with the negative bias being more pronounced over the ocean.

Figure shows the spatial distributions of G1L2 and G2L2 XCH₄ on a 2.5° grid, averaged over four consecutive three-month periods from December 2021 to November 2022. G2L2 provides significantly improved coverage, particularly over oceans, high latitudes, and tropical regions where cloud contamination often limits the number of valid GOSAT retrievals. A notable exception is the central-eastern region of South America, where the spatial coverage of G2L2 is degraded. Although the cause has not yet been fully identified, reduced radiance levels in Band 1 have been observed in this region, and a possible influence of the South Atlantic Anomaly has been suggested.

The seasonal distributions are broadly consistent between G1L2 and G2L2, showing higher XCH₄ values during boreal fall and lower values from spring to early summer, which is likely influenced by the seasonality of Northern Hemisphere wetland emissions . Regions of enhanced XCH₄ include South and East Asia, the Arabian Peninsula, central Africa, and parts of the Americas, and these patterns are largely reproduced across both products.

Mean spatial differences (G2L2–G1L2) averaged over three-month periods range from -2.6 to -5.4 ppb over land and from -5.2 to -10.8 ppb over ocean, with standard deviations of 12–14 ppb across both surface types. A systematic latitudinal pattern is also evident: negative differences are larger in low-latitude regions and diminish toward higher latitudes, implying a reduced north-south contrast within each hemisphere in G2L2 relative to G1L2. To investigate the influence of bias correction, the same analysis was repeated using G1L2 without bias correction (V03.00). In this case, the latitudinal trend largely disappears, suggesting that the G1L2 bias-correction scheme is a major contributor. Therefore, part of the spatial differences between G1L2 and G2L2 may reflect differences in the bias-correction strategy rather than differences in the underlying satellite observations themselves. Accordingly, caution is required when interpreting the spatial differences between the two products. Because assessing this correction is beyond the scope of this study, no further analysis is conducted, but users should be aware that the presence or absence of bias correction can introduce substantial spatial differences in XCH₄.

The global CH₄ budget synthesized by the Global Carbon Project GCP; provides the most authoritative benchmark for evaluating top-down CH₄ flux estimates. Against this benchmark, our satellite-only inversion using G2L2 retrievals reproduces the global budget with remarkable accuracy. The posterior global totals are 582.9 and 589.5 Tg CH₄ yr⁻¹ for 2020 and 2021, respectively. The 2020 estimate lies within the range reported by the Global Methane Budget top-down assessment 608 Tg CH₄ yr⁻¹, range 581–627 Tg CH₄ yr⁻¹;. This high level of agreement suggests that GOSAT-2 SWIR observations are capable of reproducing the global-scale CH₄ budget, despite inherent retrieval uncertainties in satellite-based measurements e.g.,. However, agreement at the global scale does not necessarily imply uniformly strong observational constraints at regional scales, where the influence of prior assumptions and observational coverage may vary substantially.

Figure a shows the spatial distribution of posterior CH₄ fluxes for 2021. Large emissions appear across regions where elevated XCH₄ was identified in Fig. , including South and East Asia, the Arabian Peninsula, central Africa, and the Americas. These hotspots reflect contributions from both natural wetlands and anthropogenic activities such as fossil-fuel extraction and waste management. The G2L4 flux distributions broadly align with established global CH₄ source patterns.

Figure b displays posterior-prior differences, highlighting regions where the G2L2 product induces substantial adjustments. Positive corrections occur mainly in tropical and subtropical regions – South Asia, central Africa, and northern to central South America – where prior estimates are typically uncertain and where ground-based observations are sparse. Negative adjustments, in contrast, appear over parts of China, the Persian Gulf region, Europe, and central North America. It is important to note that large posterior adjustments do not necessarily indicate strong observational constraints. Substantial corrections may arise either from (i) strong satellite constraints correcting prior biases, or (ii) weak observational constraints in regions with persistent cloud cover or limited viewing opportunities, where the inversion becomes more sensitive to residual retrieval biases or prior uncertainties. The degraded G2L2 coverage over central-eastern South America is one such example.

Overall, regions with large adjustments generally coincide with areas where prior uncertainties are large and observational coverage is limited, indicating where GOSAT-2 observations have the potential to provide valuable additional information beyond the prior estimates. These results emphasize both the value of G2L2 for improving flux estimates and the need for continued satellite observations to strengthen constraints in regions where current observational coverage remains limited.

Figure presents monthly CH₄ flux variations for six latitude bands between 70° N and 50° S for the period May 2019–October 2022. Prior fluxes show only weak seasonal variability across all latitude bands, whereas posterior fluxes exhibit pronounced seasonal cycles, particularly north of 10° N. In the high northern latitudes (70°–50° N), posterior fluxes increase from around 2 Tg CH₄ month⁻¹ during winter to more than 10 Tg CH₄ month⁻¹ in July–August, reflecting well-known wetland-driven seasonality. Similar enhancements in seasonal amplitude are found in the 50°–30° N band and in the tropics (10° N–10° S), where prior fluxes show only limited variability.

A notable feature in Fig. is the clear phase shift in the 30° N–10° N latitude band, where the posterior flux peak moves from May–June in the prior to July–August. This change is consistent with independent studies showing that CH₄ emissions from wetlands and rice paddies in South and East Asia tend to peak during the boreal summer monsoon e.g.,. Although satellite sampling in this region decreases during the monsoon season due to persistent cloud cover, the available G2L2 retrievals still exhibit enhanced XCH₄ in July–August, suggesting that the inversion adjusts the seasonal phase to align more closely with observational signals and to compensate for known biases in the prior. At the same time, the limited sampling also implies that part of the adjustment may reflect prior uncertainties rather than strong observational constraints. The posterior phase shift therefore likely results from a combination of satellite-derived information and structural uncertainties in the prior representation of wetland and rice-paddy emissions.

Overall, posterior fluxes peak in tropical and subtropical regions (30° N–10° S), whereas prior fluxes show maximum emissions in the mid-latitude to tropical Northern Hemisphere. Averaged over 2020 and 2021, posterior fluxes increase by roughly 20–40 Tg CH₄ yr⁻¹ in the tropics and subtropics relative to the prior, while high-latitude Northern Hemisphere fluxes remain broadly similar. The stronger seasonality and spatial redistribution evident in the posterior estimates suggest that GOSAT-2 observations contribute useful information for adjusting flux patterns in regions where prior constraints are relatively weak, particularly in tropical regions dominated by wetlands and agricultural activity.

To evaluate the impact of the G2L2 product, we compared CH₄ flux estimates derived from inversions using the same model system as G2L4 but based on different observational datasets. Two datasets were employed: (1) fluxes estimated using only the G1L2 V03.05 satellite product, as in the Results section, and (2) fluxes from , which were derived from ground-based and aircraft (SURF+AIR) observations. The ground-based data include ObsPack GLOBALVIEWplus version 6.0 and measurements from NIES and collaborative networks . Aircraft data consist of flask samples collected through the CONTRAIL program and Tohoku University campaigns . Although used the same inversion framework as this study, the prior biomass burning emissions differ between the two approaches. The dataset provides a benchmark based on dense in situ measurements, offering a complementary perspective to satellite-driven inversions and enabling assessment of consistency and potential biases in regional and seasonal CH₄ flux estimates.

We first compared the global annual CH₄ budgets among the three inversions. The G1L2-based inversion yielded 594.3 and 599.2 Tg CH₄ yr⁻¹, and SURF+AIR-based inversion produced 593.2 and 595.4 Tg CH₄ yr⁻¹ for 2020 and 2021, respectively. Differences across datasets remain within ∼10 Tg CH₄ yr⁻¹, indicating that the choice of observational constraint does not materially change the global CH₄ budget in this inversion framework. This cross-dataset convergence provides confidence that the regional analyses below primarily reflect the characteristics of the observational constraints rather than structural instabilities of the inversion system itself.

Seasonal patterns of CH₄ fluxes are broadly consistent across the three inversions (Fig. ), as noted in previous studies e.g.,. Comparisons between G2L4 and G1L2-based inversions show small mean differences across latitude bands (-0.12 to 0.58 Tg CH₄ month⁻¹) and high correlations (r= 0.83–0.99), indicating similar seasonal cycles. Variance ratios (posterior-to-reference standard deviations; values >1 indicate stronger seasonal amplitude) range from 1.05 to 1.99 – largest in 10° N–10° S – reflecting the combined influence of (i) differences in retrieval algorithms and product post-processing (including G1L2 bias-correction), (ii) differences in sampling and viewing geometry between missions (targeting strategies, off-nadir angles, cloud screening), and (iii) mission-dependent L1B radiance characteristics. In contrast, differences relative to the SURF+AIR-based inversion are more pronounced: the mean difference reaches -2.03 Tg CH₄ month⁻¹ in the tropics, root-mean-square difference exceeds 2.38 Tg CH₄ month⁻¹, and variance ratios rise to 4.36, indicating changes in both amplitude and phase. Correlations drop south of 10° S (r= 0.55–0.31), consistent with sparse in-situ coverage there.

Next, we extend our analysis to the regional scale using the 42 land regions defined in the GOSAT Level 4 product (Fig. ). This framework enables a systematic comparison of CH₄ flux estimates derived from different observational datasets and helps identify where observational constraints most strongly affect inversion outcomes. At the regional scale (Fig. ), the three inversions show broadly consistent spatial patterns, with all methods identifying similar high-emission regions and capturing the major continental-scale features of the CH₄ budget. This overall agreement indicates that G2L4, G1L2-based, and SURF+AIR-based inversions share a common large-scale structure and that the differences among them reflect regional observational characteristics rather than fundamental inconsistencies. Within this general consistency, regional differences become apparent. For example, in Southern Central Asia (CAS, RGN 30), the annual mean flux from G2L4 (50.5 Tg CH₄ yr⁻¹) is ∼20 % lower than the SURF+AIR-based estimate (60.2 Tg CH₄ yr⁻¹). Similar gaps (4–11 Tg CH₄ yr⁻¹) appear in Northeastern Siberia (SIB-NE, RGN 28) and the Arabian Peninsula-Western Middle East (WME, RGN 29). While these differences broadly correspond to sparse in-situ observational coverage, we cannot rule out contributions from region-dependent retrieval challenges-such as high aerosol loading, low surface albedo, or large solar zenith angles-that may influence satellite column retrievals. Thus, both observational distribution and potential retrieval systematics likely contribute to the regional spread.

Beyond differences in regional observational coverage, mission-specific observation strategies and retrieval characteristics also contribute to divergences among inversions. Both the GOSAT and GOSAT-2 missions allocate a considerable fraction of their observations to developed regions in order to target large emission sources such as megacities and power plants . While this targeting strategy enhances the ability to capture urban greenhouse-gas enhancements , it can introduce a sampling bias toward high-concentration areas and may reduce the representativeness of regional fluxes. In addition, the two missions do not necessarily observe identical footprints, leading to differences in spatial sampling even within the same region. The case of Eastern Asia (RGN 32) illustrates this complexity. Although multiple TCCON sites are available in the region and are used in the bias correction of G1L2, its flux estimates still differ substantially from those derived from the SURF+AIR-based inversion. This divergence arises despite both inversions being strongly constrained by extensive observational data. The differences likely reflect the fundamentally distinct nature of the observational datasets-satellite column retrievals versus in-situ and aircraft measurements-as well as differences in coverage, bias-correction approaches, data-selection criteria, and the spatial representativeness of observation sites. These factors can lead to meaningful discrepancies in regional flux estimates even when the overall inversion frameworks are robust. Rather than indicating inconsistencies among the inversions, these results underscore the importance of considering observation-system characteristics when interpreting regional CH₄ budgets. They also highlight the need for a balanced combination of satellite, ground, and aircraft observations to improve representativeness and reduce structural uncertainties in future CH₄ monitoring systems.

Figure also highlights where the three inversion systems show noticeable differences in their regional estimates. While G1L2-based inversions generally show amplitudes and correlations closer to those of G2L4, the SURF+AIR-based inversions display larger differences in regions where observational coverage is limited or where surface networks do not sufficiently capture the dominant emission regimes. This pattern is consistent with previous studies reporting that the performance of ground-based-driven inversions is strongly dependent on the spatial density and placement of in-situ sites e.g.,. Our results reflect this behavior across several regions, including Pampas (PAM, RGN 14), Central South America (SAM-, RGN 15), Northwestern Northern Africa (AFR-NNW, RGN 19), Maritime Southeast Asia (SEA-MAR, RGN 33), and Southeastern Australia (AUS-SE, RGN 36), where correlations with G2L4 are comparatively low. Amplitude differences also emerge in high-latitude regions such as Southwestern Siberia (SIB-SW, RGN 25) and Northeastern Siberia (SIB-NE, RGN 28), as well as in well-monitored areas like the Nordic Region (NOR, RGN 41) and mid-latitude North America (NAM-SW and NAM-NE; RGNs 5 and 8). These discrepancies do not undermine the overall consistency among the inversions; rather, they show how regional flux estimates respond differently to the distinct observational constraints available in each system. These regions illustrate where satellite observations offer broader spatial sampling that complements limited in-situ coverage. Looking ahead, integrating GOSAT-2 with complementary satellite missions-together with targeted in-situ deployments where feasible-will help reduce regional uncertainty and support the development of a more balanced, globally consistent CH₄-monitoring framework.

Recent studies have demonstrated the capability of satellite observations to quantify national-scale CH₄ emissions using atmospheric inversion frameworks and to compare these estimates with national inventories e.g.,. Inversion-based estimates, however, are generally sensitive to the total atmospheric CH₄ field, which integrates contributions from multiple source sectors, CH₄ sinks, and atmospheric transport processes. This property is not unique to satellite measurements; it is inherent to all top-down inversion systems. As a result, satellite-based inversions primarily optimize the net surface flux and cannot directly isolate sector-specific emissions. To gain insight into sectoral contributions despite this limitation, previous studies have adopted an indirect approach: in regions where the net prior flux is strongly dominated by a limited number of anthropogenic sectors, the direction and magnitude of posterior adjustments can help infer whether emissions from those dominant sectors are likely over- or underestimated. While this approach does not fully resolve sector-specific budgets, it can provide useful qualitative information on sectoral patterns. Building on this concept, we assess whether the G2L4 product offers meaningful constraints on sector-dominated regions. We focus on regions where aggregated anthropogenic emissions account for more than 80 % of prior total net surface flux, allowing us to examine whether posterior adjustments inferred from satellite observations align with expected sectoral contributions. Here, anthropogenic emissions refer to the sum of coal mining, oil and gas, landfill and waste, biofuel, enteric fermentation, manure management, and rice paddy fluxes.

Figure illustrates that the G2L4 product provides useful observational information for anthropogenic-dominated regions, offering new insight into regional CH₄ flux patterns. In East Asia (EAS), where anthropogenic emissions constitute the majority of the prior total flux, the G2L4 posterior suggests a reduction of 12.3 Tg CH₄ yr⁻¹ relative to the prior. Although this posterior estimate (34.4 Tg CH₄ yr⁻¹) is lower than values reported in previous satellite-based studies such as and , those studies also emphasize that East Asia exhibits large variability across inversion systems due to differences in observational datasets, retrieval configurations, and bias correction methods. The spread among inversion estimates therefore likely reflects this region's well-known sensitivity to differences in observational data sources, rather than any inconsistency among the inversion systems themselves. In this context, the G2L4 result contributes an independent satellite-based constraint that is useful for assessing the plausible range of anthropogenic emissions in East Asia.

In the Western Middle East (WME), where emissions are dominated by oil and gas production, the G2L4 posterior (18.2 Tg CH₄ yr⁻¹) falls within the range of previous estimates, including the Persian Gulf fossil emissions inferred by . This consistency-combined with the region's frequent clear-sky conditions, which enable dense satellite sampling-demonstrates that GOSAT-2 observations provide useful observational information for fossil-fuel-dominated regions. At the same time, the strong influence of dust aerosols on SWIR retrievals e.g., highlights the importance of continued refinement of aerosol correction methods to fully leverage the strengths of satellite data. Overall, the G2L4 results in EAS and WME illustrate how satellite-based inversions can provide valuable and region-specific insight into sectoral emissions, while also identifying observational and algorithmic factors that should be prioritized to further enhance the robustness of future CH₄ flux assessments.

Beyond East Asia and the Western Middle East, several other regions-such as Northern Africa and Central Asia-show only modest differences between prior and posterior estimates. Importantly, small posterior adjustments in these regions should not be interpreted as evidence that prior inventories are highly accurate. Rather, these areas are characterized by sparse ground-based observations and limited socioeconomic activity data, making the underlying inventories inherently uncertain e.g.,. In such contexts, even small satellite-driven adjustments provide independent and valuable constraints that would otherwise be difficult to obtain. These regions also underscore a broader challenge: observational gaps persist in many parts of the world where establishing and maintaining surface networks is logistically difficult and financially costly e.g.,. Satellite observations offer spatial coverage that complements limited in-situ measurements and helps identify where additional observations would most improve regional constraints. Looking ahead, expanding the use of multi-satellite datasets-along with targeted deployments of ground-based sensors where feasible-offers a promising pathway for improving the sectoral attribution of CH₄ emissions in regions that have historically been under-constrained.

The G2L4 product provides monthly global CH₄ flux estimates derived from a satellite-only atmospheric inversion system. While the results presented in this study suggests that GOSAT-2 observations can improve the spatial and temporal distribution of CH₄ flux estimates, several limitations should be considered when interpreting the data product.

First, the current G2L4 release does not include formal posterior uncertainty estimates, posterior covariance matrices, inversion averaging kernels, or degrees-of-freedom-for-signal (DFS) diagnostics . The inversion is performed using a global high-dimensional 4D-Var framework based on NISMON-CH₄ , and explicit estimation of posterior covariance matrices would require additional covariance calculations or large ensemble-based experiments that are not included in the operational G2L4 processing system. Consequently, quantitative estimates of uncertainty reduction and observation impact are not presently provided as part of the G2L4 product.

Second, it is important to distinguish between retrieval-level and inversion-level diagnostics. Averaging kernels and a priori profiles distributed with the G2L2 product are incorporated into the observation operator in order to account for retrieval smoothing effects and to ensure consistency between modeled and retrieved XCH₄ . However, these retrieval averaging kernels should not be interpreted as inversion averaging kernels, nor do they provide information on posterior uncertainty reduction within the flux inversion framework.

Therefore, posterior-minus-prior flux differences presented in this study should primarily be interpreted as inversion adjustments rather than direct quantitative measures of observational constraint or uncertainty reduction. Interpretation of regional flux estimates should also take into account variations in observational coverage, prior uncertainty, retrieval characteristics, and the prescribed inversion error statistics.

Despite these limitations, comparisons with independent inversion systems indicate that the G2L4 product reproduces the large-scale characteristics of the global CH₄ budget and provides useful observational information in many regions where conventional surface observations remain sparse. Future developments of the G2L4 framework will investigate practical approaches for characterizing posterior uncertainty and information content while maintaining computational feasibility for global high-resolution inversions.