All UNFCCC parties shall periodically update and submit their national GHG inventories of emissions by sources and removals by sinks to the convention parties. Annex I countries submit their NIRs in Common Reporting Format (CRF) tables every year with a complete time series starting in 1990. Non-Annex I parties are required to submit their NCs roughly every 4 years after entering the convention and have submitted BURs every 2 years since 2014. Currently, there are in total 427 submissions of NCs and over 166 submissions of BURs (UNFCCC, 2021b, a) (Fig. 1).

We collected NGHGI data submitted to the UNFCCC by 28 February 2023. For Annex I countries, data collection is straightforward, as their reports are provided as Excel files under the Common Reporting Format (CRF) until the year 2020, last accessed on 28 February 2023. For non-Annex I countries, the data were directly extracted from the original reports provided in portable document format (PDF), last accessed on 28 February 2023. Data from successive reports for the same country were extracted, except when they relate to the same years, in which case only the latest version is considered. While Annex I countries are required to compile their inventory following 2006 IPCC guidelines and the subdivision between sectors established by the UNFCCC decision (dec. 24/CP.19), non-Annex I countries are increasingly adopting the 2006 IPCC guidelines, although some still utilize the older 1996 IPCC guidelines, with different approaches and sectors. Consequently, the methods used and the reported sectors may differ among NCs and BURs.

Most atmospheric inversions derive total net CH₄ emissions at the surface as it is difficult for them to disentangle overlapping emissions from different sectors at the pixel/regional scale based on atmospheric CH₄ observations only. However, five of the seven inverse systems solve for some source categories owing to different spatio-temporal distributions between the sectors. For each inversion, monthly gridded posterior flux estimates were provided at 1° × 1° grid resolution for the net flux at the surface (Enetinv); the soil uptake at the surface (Esoilinv); the total emission at the surface (Etotinv); and five emitting “super sectors” which regroup several IPCC sectors – agriculture and waste (EAgWinv), fossil fuel (EFFinv), biomass and biofuel burning (EBBinv), wetlands (EWetinv), and other natural (EOthinv) emissions. Considering the soil uptake to be a “negative source” given separately, the following equation applies: 2Enetinv=Etotinv+Esoilinv=EAgWinv+EFFinv+EBBinv+EWetinv+EOthinv+Esoilinv. For inversions solving for net emissions only, the partitioning into source sectors was created based on using a fixed ratio of sources calculated from prior flux information at the pixel scale. For inversions solving for some categories, a similar approach was used to partition the solved categories into the five aforementioned emitting sectors. Such processing can lead to significant uncertainties if not all sources increase or change at the same rate in a given region/pixel. National values have been estimated using the country land mask described in the CO₂ section (Sect. 2.3.1); thus offshore emissions are not counted as part of inversion results unless they are in a coastal grid cell.

In our previous study (Deng et al., 2022), four methods were proposed to separate CH₄ anthropogenic emissions from inversions (EAnthinv) and compare them with national inventories (EAnthni), aiming to discuss the uncertainties in anthropogenic CH₄ emissions associated with the chosen separation methods. These four methods comprise (1) summing prior estimates based on inversions for anthropogenic sectors (method 1), (2) subtracting natural emissions from total fluxes (method 2), and (3) subtracting natural emissions derived from other bottom-up assessments from the total inversion flux (methods 3.1 and 3.2, differing only in the bottom-up wetland CH₄ data used). The calculations of anthropogenic emissions by each method were performed separately for GOSAT inversions and in situ inversions. However, the uncertainty from the separation method is generally much smaller than the variability between different inversion models (see Fig. 9 in Deng et al., 2022). Therefore, we apply only one method in this study, which consists of using inversion partitioning as defined in Saunois et al. (2020): 3EAnthinv=EAgWinv+EFFinv+EBBinv-EwildfiresBU⇔EAnthni. This method has some uncertainties. First, the partitioning relies on prior fractions within each pixel, and second, emissions from wildfires are counted for in the biomass and biofuel burning (BB) inversion category, while they are not necessarily reported in NGHGIs. The BB inversion category includes methane emissions from wildfires in forests, savannahs, grasslands, peats, agricultural residues, and the burning of biofuels in the residential sector (stoves, boilers, fireplaces). Therefore, we subtracted bottom-up (BU) emissions from wildfires (EwildfiresBU) based on the GFEDv4 dataset (van Wees et al., 2022) using its reported dry matter burned and CH₄ emission factors. Because the GFEDv4 dataset also reports specific agricultural and waste fire emissions data, we assumed that those fires (on managed lands) are reported by NGHGIs, so they were not counted in EwildfiresBU. Figure S3 presents a comparison between our adjusted BB flux and the wood fuel emissions reported by Flammini et al. (2023). This comparison highlights the broader scope and definition of our adjusted BB flux, illustrating the differences in emissions estimation methodologies.

We subtracted estimates of natural N₂O sources from the N₂O emission budget (Etotinv) of each inversion to provide inversions of anthropogenic emissions (Eantinv) that can be compared with national inventories (Eantni): 4Eantinv=EMLinv-Enataq-EwildfiresGFED⇔Eantni. Here, the natural N₂O sources include natural emissions from freshwater systems (nataq) and natural emissions from wildfires (antni).

In our previous study, intact forest grid cells (assumed unmanaged) from Potapov et al. (2017) and lightly grazed grassland areas from Chang et al. (2021) were removed from the gridded N₂O emissions in proportion to their presence in each inversion grid box. Here we used the new managed-land mask defined in Sect. 2.3 to filter gridded N₂O emissions from inversions to obtain EMLinv. We verified that the inversion grid box fractions classified as unmanaged do not contain point source emissions from the industry and energy and from diffuse emissions from the waste sector to make sure that we do not inadvertently remove anthropogenic sources by masking unmanaged pixels. From the EDGAR v4.3.2 inventory (Janssens-Maenhout et al., 2019), we found that N₂O from wastewater handling covers a relatively large area that might be partly located in unmanaged land. But the corresponding emission rates are more than 1 order of magnitude smaller than those from agricultural soils. For other sectors, only very few of the unmanaged grid boxes contain point sources, and none of them have an emission rate that is comparable with agricultural soils (managed land). Thus, our assumption that emissions from these other anthropogenic sectors are primarily over managed-land pixels is solid (other sectors include the power industry; oil refineries and transformation industry; combustion for manufacturing; aviation; road transportation, no resuspension; railways, pipelines, off-road transport; shipping; energy for buildings; chemical processes; solvents and product use; solid waste incineration; wastewater handling; and solid waste landfills).

The flux Enataq is the natural emission from freshwater systems given by a gridded simulation of the DLEM (Yao et al., 2019) describing pre-industrial N₂O emissions from N leached by soils and lost to the atmosphere by rivers in the absence of anthropogenic perturbations (considered the average of 1900–1910). Natural emissions from lakes were estimated only at a global scale by Tian et al. (2020) and represent a small fraction of rivers' emissions. Therefore, they are neglected in this study. The flux EwildfiresGFED is based on the GFED4s dataset (van Wees et al., 2022) using its reported dry matter burned and N₂O emission factors. Because the GFED dataset reports specific agricultural and waste fire emissions data, we assume that those fires (on managed lands) are reported by NGHGIs, so they were not counted in EwildfiresGFED, just like for CH₄ emissions. Note that there could also be a background natural N₂O emission from natural soils over managed lands (EMLsoil) which is not necessarily reported by NGHGIs. We did not try to subtract this flux from managed-land emissions because we assumed that, after a land-use change from natural to fertilized agricultural land, background emissions decrease and become very small compared to N-fertilizer-induced anthropogenic emissions. In a future study, we could use for EMLsoil the estimate given by simulations of pre-industrial N₂O emissions from the NMIP ensemble of dynamic vegetation models with carbon–nitrogen interactions (number of models n=7), namely, their simulation S0, in which climate forcing is recycled from 1901–1920, CO₂ is at the level of 1860, and no anthropogenic nitrogen is added to terrestrial ecosystems (Tian et al., 2019).

Another important point to ensure a rigorous comparison between inversion and NGHGI data is whether anthropogenic indirect emissions (AIEs) of N₂O are reported in NGHGI reports. This is not always the case even though UNFCCC parties are required to report these in their NGHGIs according to the IPCC guidelines. For example, South Africa's BUR3 did not report indirect N₂O emissions due to the lack of activity data. AIEs arise from anthropogenic nitrogen from fertilizers leached to rivers and anthropogenic nitrogen deposited from the atmosphere to soils. AIEs typically represent 20 % of direct anthropogenic emissions and cannot be ignored in a comparison with inversions. For Annex I countries, AIEs are systematically reported, generally based on emission factors since these fluxes cannot be directly measured, and we assumed that indirect emissions only occur on managed land. For non-Annex I countries, we checked manually from the original NC and BUR documents if AIE was reported or not by each non-Annex I country. If AIEs were reported by a country, they were used as such to compare NGHGI data with inversion results and were grouped into the agricultural sector. If they were not reported or if their values were outside plausible ranges, AIEs were independently estimated by the perturbation simulation of N fertilizer leaching, CO₂, and climate applied to river and lake fluxes in the DLEM model (Yao et al., 2019) and by the perturbation simulation of atmospheric nitrogen deposition on N₂O fluxes from the NMIP model ensemble (Tian et al., 2019).

The bottom-up NGHGIs are compiled based on activity data (statistics) following the 1996/2006 IPCC guidelines (IPCC, 1997, 2006), with detailed information on subsectors. However, the top-down inversions can only distinguish between very few groups of sectors at most. Thus, in this study, we aggregated NGHGI sectors into some super sectors to make inversions and inventories comparable for each GHG (Table 4). For CO₂, the inversions are divided into two aggregated super sectors: (1) fossil fuel and cement CO₂ emissions and (2) adjusted net land flux. Inversions use a prior gridded fossil fuel dataset as summarized in Sect. 2.3.1; thus, in this study, we compare only the net land flux between inversions and inventories. To calculate the net land flux over managed lands from NGHGIs, we subtracted fossil emissions from the IPCC CRF (1) “energy” and (2) “industrial processes” (or (2) “industrial processes and product use”) sectors from the “total GHG emissions including LULUCF/LUCF” (or “total national emissions and removals”) sector. For CH₄, we compare inversions and inventories based on three super sectors, comprising “fossil”, “agriculture and waste”, and “total anthropogenic”. To compare with NGHGIs, we group the IPCC CRF sectors of (1) “energy” and (2) “industrial processes” (or (2) “industrial processes and product use”), excluding biofuel burning (reported under the (1) energy sector), into the super sector of fossil; we group sectors of (4) “agriculture” (or (3) “agriculture”) and (6) “waste” (or (5) “waste”) into the super sector of agriculture and waste; and we aggregate anthropogenic flux from “fossil”, “agriculture and waste”, and “biofuel burning” into anthropogenic. For N₂O, we group the NGHGI sectors into anthropogenic flux, being the sum of (1) energy + (2) industrial processes (or (2) industrial processes and product use) + (4) agriculture (or (3) agriculture) + (6) waste (or (5) waste) + anthropogenic indirect emissions.

For the analysis, we selected 12 countries (or groups of countries) based on specific criteria for each aggregated sector (Table 5). Firstly, each chosen country had to possess a sufficiently large land area, as the limitations of coarse-spatial-resolution inversions make it difficult to reliably estimate GHG budgets for smaller countries. Additionally, it was preferable for the selected countries to have some coverage provided by the in situ global network of monitoring stations.

For CO₂, we focus on the land CO₂ fluxes of large fossil fuel CO₂ emitters. Although inversions do not allow us to verify fossil emissions in these countries as they are used as a fixed prior map of emissions, it is crucial to compare the magnitude of national land CO₂ sinks with fossil fuel CO₂ emissions in those large emitters. It is important to note that fitting net fluxes to changes in atmospheric CO₂ and then subtracting the prior fossil fuel (FF) fluxes can result in errors in the residual values, which are typically attributed exclusively to the sum of all non-FF fluxes. Additionally, we included two large boreal forested countries (Russia – RUS and Canada – CAN), two tropical countries with large forest areas (Brazil – BRA and the Democratic Republic of Congo – COD), two large countries with ground-based stations (Mongolia – MNG and Kazakhstan – KAZ), and two large dry Southern Hemisphere countries also with high rankings in fossil fuel CO₂ emissions (South Africa – ZAF and Australia – AUS), the latter of which both possess atmospheric stations to constrain their land CO₂ flux.

For CH₄, we first ranked countries (or groups of countries) based on their total anthropogenic, fossil, and agricultural emissions. This study includes China (CHN), India (IND), the United States (USA), the European Union (EUR), Russia (RUS), Argentina (ARG), and Indonesia (IDN), all of which are among the top emitters of both fossil fuel and agricultural CH₄ and possess large areas. Criteria of large land areas and the presence of atmospheric stations are crucial for in situ inversions. The advantage of utilizing GOSAT in CH₄ atmospheric inversions is its ability to provide observations over countries where surface in situ data are sparse or absent, such as in the tropics. This allows us to consider countries with limited or few ground-based observations. Small countries were excluded due to the coarse spatial resolution. However, among the selected countries, Venezuela, with an area of 916 400 km², was chosen specifically for the analysis of CH₄ emissions. Despite being relatively small, Venezuela is a large producer of oil and gas, potentially allowing for inversions using GOSAT observations to constrain its emissions. In major oil- and gas-extracting countries that have negligible agricultural and wetland emissions like Kazakhstan (KAZ), grouped in this study with Turkmenistan (TKM) into KAZ&TKM; Iran (IRN); and Persian Gulf countries (GULF), fossil emissions should be easier to separate by inversions and thus easier to compare with NGHGIs.

For N₂O, we selected the top 12 emitters based on the NGHGI reports. Anthropogenic N₂O emissions in most of these countries are predominantly driven by the agricultural sector, which accounts for a share (including indirect emissions) ranging from 6 % in Venezuela (VEN) to 95 % in Brazil (BRA) of their total NGHGI emissions.

Together, the selected countries (or groups of countries), with a different selection for each gas, account for more than 90 % of the global land CO₂ sink, 60 % of the global anthropogenic CH₄ emissions (around 15 % of fossil fuel emissions and approximately 40 % of agriculture and waste emissions separately), and 55 % of the global anthropogenic N₂O emissions, as estimated by the NGHGIs.

Figure 4 presents the variations in anthropogenic CH₄ emissions for the 12 selected countries, where these emissions sum the sectors of agriculture and waste, fossil fuels, and biofuel burning. The distribution of emissions is highly skewed even among the top 12 emitters, with the largest and most populated countries/regions such as China (CHN), India (IND), the United States (USA), Brazil (BRA), Russia (RUS), and the European Union (EUR) which emits more than 10 Tg CH₄ yr⁻¹ annually, while other countries have smaller emissions (ranging from 3 to 10 Tg CH₄ yr⁻¹) that are more challenging to quantify through inversions. During 2010–2020, CHN has the highest total anthropogenic emissions at around 50 ± 4 Tg CH₄ yr⁻¹, followed by IND with 30 ± 1 Tg CH₄ yr⁻¹, the USA with 24 ± 1 Tg CH₄ yr⁻¹, BRA with 24 ± 1 Tg CH₄ yr⁻¹, EUR with 19 ± 1 Tg CH₄ yr⁻¹, Indonesia (IDN) with 14 ± 1 Tg CH₄ yr⁻¹, and RUS with 13 ± 1 Tg CH₄ yr⁻¹, according to the medians of the satellite-based inversion ensemble based on EDGAR v6.0 as prior. The remaining countries have emissions of approximately 5 Tg CH₄ yr⁻¹. In general, the difference between NGHGIs and inversions aligns in the same direction based on both satellite and in situ inversions. This provides some confidence in using inversions to evaluate NGHGIs as the satellite observations are independent from in situ networks. Overall, satellite-based inversions may be more robust across most countries due to better observation coverage, except in EUR and the USA, where the in situ network is more extensive.

In IND, PAK, and MEX, there is good agreement (r>0.8, p<0.01) between the in situ and satellite-based inversion ensembles (31 ± 1 and 30 ± 1 Tg CH₄ yr⁻¹ in IND, 8 ± 1 and 7 ± 1 Tg CH₄ yr⁻¹ in PAK, and 6 ± 1 and 6 ± 1 Tg CH₄ yr⁻¹ in MEX, respectively), while both of them present a significant increasing trend of anthropogenic methane emissions in these countries (Mann–Kendall p<0.05). However, when comparing to NGHGI values, the inversion results in IND and PAK indicate > 50 % larger emissions than those reported from the NGHGIs during 2010–2020. In contrast, values reported from the NGHGIs (∼ 6 Tg CH₄ yr⁻¹) by MEX also show good agreement with the inversion results.

In BRA, IDN, and Argentina (ARG), the medians for in situ and satellite-based inversion ensembles show good consistency (r=0.8, p<0.01) in these two countries, while satellite-based inversion results are generally higher than the in situ inversion results. Specifically, in BRA, the satellite-based inversions (24 ± 1 Tg CH₄ yr⁻¹) were 16 % higher than the in situ inversions (21 ± 1 Tg CH₄ yr⁻¹) and 52 % higher than the NGHGI estimation (∼ 17 Tg CH₄ yr⁻¹) during 2010–2020, possibly owing to difficulties in inversions differentiating between natural (wetlands, inland waters) and anthropogenic sources in this country and to possible flaws in the prior used for natural and anthropogenic fluxes. In IDN, NGHGIs reported a significant continuous upward trend at an annual average growth of 0.3 Tg CH₄ yr⁻¹, with a noticeable positive outlier in 2000. The medians for both in situ and satellite-based inversion ensembles also indicate an upward trend in IDN, but both of them present sudden dips in anthropogenic methane emissions in 2015 and 2019 by 15 %–23 % and 16 %–25 % compared to the previous year, respectively. It is unlikely that anthropogenic activities could contribute such large year-to-year variations except for different flooded areas used for rice paddies. In ARG, the satellite-based inversion results also indicate two sudden dips in 2016 and 2019; however, such a pattern was not found in the in situ inversion results. A cause of year-to-year variations from inversions is the lack of in situ sites and variable cloud cover affecting the density of GOSAT data.

Regarding IRN, NGHGIs only provided data for three years (1994, 2000, and 2010), making it difficult to compare them with inversion results. However, NGHGIs show a rapid growth in anthropogenic CH₄ emissions (+9.4 % yr⁻¹) during this period. There are significant differences between inversion results and for IRN, with satellite inversions generally giving lower emissions than in situ inversions and different trends. Satellite inversions suggest a declining trend between 2010 and 2015, followed by a fluctuating increase until 2020. In contrast, in situ-based inversions (by any nearby measurement stations, thus likely reflecting the prior trend) show a rapid rise in emissions after 2010, reaching a peak in 2018, followed by a decline.

NGHGIs for RUS indicate that anthropogenic CH₄ emissions have been reduced during the 1990s and remained stable since 2000 (12.0 ± 0.3 Tg CH₄ yr⁻¹ during 2000–2020), which is similar with the trend observed from satellite-based inversion results (12.7 ± 0.9 Tg CH₄ yr⁻¹ during 2000–2020). However, in 2016, there was a sudden increase in emissions in satellite inversion results (+14 % increase from 12.5 Tg CH₄ yr⁻¹ in 2015 to 14.2 Tg CH₄ yr⁻¹ in 2016), followed by a gradual decline, and then a new increase in 2020 (+11 % increase from 12.8 Tg CH₄ yr⁻¹ in 2019 to 14.3 Tg CH₄ yr⁻¹ in 2020). This recent change was not observed in the in situ inversion results or the NGHGIs.

For the USA, Australia (AUS), and EUR, NGHGIs reported a slowly declining trend (EUR: 0.4 Tg CH₄ yr⁻¹; USA: 0.2 Tg CH₄ yr⁻¹; AUS: -0.04 Tg CH₄ yr⁻¹) in anthropogenic CH₄ emissions. In the case of the USA, inversion-derived emissions are slightly lower than NGHGIs (in situ-based: 9 % lower during 2000–2020; satellite-based: 11 % lower during 2010–2020). However, both ground-based and satellite-based inversions indicate that anthropogenic CH₄ emissions have remained relatively steady since 2000, without reflecting the slow decline reported by NGHGIs. In EUR, NGHGIs indicate that anthropogenic CH₄ emissions have been decreasing rapidly since 1990 (-1.4 % yr⁻¹), consistent with the trend obtained from inversion results. However, in situ inversion emissions are on average slightly higher than NGHGIs, and this difference has been gradually increasing from 8 % in the 2000s to 15 % in the 2010s.

Figure 5 presents the fossil CH₄ emissions for the top 12 emitters from the fossil sector based on EDGAR v6.0 as the prior. The largest emitter is China (CHN), mainly from the subsector of coal extraction, followed by Russia (RUS) and the United States (USA). In CHN, the in situ (20 ± 2 Tg CH₄ yr⁻¹) and satellite inversion (17 ± 1 Tg CH₄ yr⁻¹) emissions in the 2010s are 24 % and 35 % lower, respectively, than in the NGHGIs (∼ 26 Tg CH₄ yr⁻¹). The NGHGIs in CHN suggest a decrease from 28 Tg CH₄ yr⁻¹ in 2012 to 24 Tg CH₄ yr⁻¹ in 2014. However, both in situ and satellite inversion results indicate there has been an increasing trend since 2018. In India (IND) and Indonesia (IDN), NGHGIs report a decreasing trend during the study period, while inversions suggest a rapid increase in IDN and a stable value in IND after a peak in 2012. In IND, satellite inversions suggest a peak of fossil CH₄ emissions during 2011–2012, which then dropped in 2013 and remained stable afterward. In IDN, both in situ and satellite inversions indicate a fluctuating trend, with a significant drop between 2015 and 2019. In RUS, both in situ and satellite inversion-based estimates of fossil fuel emissions are higher than those of NGHGIs and show an increasing trend, while NGHGIs report a decreasing trend. This discrepancy may be due to inversion problems for separating between wetland emissions and gas extraction industries both located in the Yamal Peninsula area or to leaks not captured in NGHGIs. In the USA, NGHGIs overall show a significant declining trend (Mann–Kendall Z=-0.8, p<0.01). In situ inversion estimates of fossil fuel emissions are 26 % lower than NGHGIs during 2000–2010 and remained consistent until around 2011. Nearly all in situ inversions show a jump in fossil fuel emissions in 2011. In the European Union (EUR), both NGHGIs and inversion results demonstrate a consistent declining trend. However, starting from 2010, both in situ and satellite inversions are higher than results in NGHGI reports.

Major oil-producing countries in the Persian Gulf are too small compared to the model resolution to be studied individually. Hence, NGHGIs from the GULF countries (Saudi Arabia, Iraq, Kuwait, Oman, the United Arab Emirates, Bahrain, and Qatar) were grouped and show much lower emissions compared to inversion results. In the 2010s, in situ and satellite inversions estimate that emissions in GULF were 9 times and 8 times higher, respectively, than the estimates reported in NGHGIs. This huge under-reporting of emissions in GULF could be partly attributed to the omission of ultra-emitters in NGHGIs. The ultra-emitters defined by Lauvaux et al. (2022) are all short-duration leaks from oil and gas facilities (e.g., wells, compressors) with an individual emission > 20 t CH₄ h⁻¹, each event lasting generally less than 1 d. Such leaks are often random occurrences and are difficult to quantify, which is why most countries do not account for these significant and episodic events in their national inventories. Indeed, recent studies by Lauvaux et al. (2022) have identified more ultra-emitters and larger emission budgets from ultra-emitters in Qatar, Kuwait, and Iraq. In KAZ&TKM, grouped together because of their rather small individual areas, both in situ (3 ± 0.2 Tg CH₄ yr⁻¹) and satellite (3 ± 0.1 Tg CH₄ yr⁻¹) inversions estimate emissions to be 2 times higher than NGHGIs (1.5 Tg CH₄ yr⁻¹) in the 2010s. Similarly, KAZ is located downwind of TKM, which has a high share of ultra-emitters. The global inversions operating at a coarse resolution may misallocate emissions from TKM to KAZ. It is worth noting that KAZ has two in situ stations for CH₄ measurements, whereas the GULF countries lack in situ station networks. On the other hand, GOSAT provides dense sampling of atmospheric column CH₄ in the Persian Gulf region due to frequent cloud-free conditions. Therefore, GOSAT inversions can be considered more accurate than in situ inversions for Iran (IRN), GULF countries, and Kazakhstan and Turkmenistan (KAZ&TKM). Additionally, it is important to note that GOSAT inversions generally give lower emissions than in situ inversions in those countries. Venezuela (VEN) is a rare case where NGHGIs report much higher CH₄ emissions than inversions. While the uncertainty in GOSAT inversions (model spread) has decreased compared to the results reported by Deng et al. (2022), the gap between inversions and NGHGIs has increased. In 2010, NGHGI reports of fossil CH₄ emissions in VEN were 298 % higher than those of GOSAT inversions and 326 % than those of in situ inversions. We do not have a clear explanation for this large difference, except that VEN has strongly decreased oil and gas extraction due to sanctions curbing its crude production from 2.7 million barrels d⁻¹ in 2015 to 0.6 million barrels d⁻¹ in 2020 (OPEC, 2023), which may not be reflected in their NGHGIs. In Nigeria (NGA) and Mexico (MEX), NGHGI estimates fall between the median of in situ and satellite inversions during 2010–2020. However, in MEX, the in situ inversion was 50 % lower than NGHGIs in the 2000s and showed a sudden large increase in 2010.

Figure 6 presents CH₄ emissions of the agriculture and waste sector for the top 12 emitters of this sector. In all countries except for the United States (USA) and Russia (RUS), the values reported by NGHGIs are systematically lower than the inversion results. The results from the previous ensemble of in situ inversions (dotted red line) are consistent with those of the inversions used in this study except in the USA, where previous inversions are 3.2 Tg CH₄ yr⁻¹ higher; in RUS, where they show a drop after 2015 although they remain in the range from the new satellite and in situ inversions; and in Mexico (MEX), where they are systematically lower by 1.6 Tg CH₄ yr⁻¹.

In China (CHN), the most recent NGHGI reports in 2012 and 2014 estimate agriculture and waste emissions at 28 Tg CH₄ yr⁻¹, which is close to the result of satellite inversions (28 ± 1 Tg CH₄ yr⁻¹) but 22.4 % lower than the median of in situ inversions (35 ± 1 Tg CH₄ yr⁻¹) and closer to their minimum value. The trend in agricultural and waste emissions is consistent between inversions and NGHGIs for CHN. In India (IND), inversions consistently show higher emissions than NGHGIs by approximately 50 % and indicate an increasing trend during 2000–2020, but with the NGHGI last communication being for 2016, we are not able to obtain a recent trend. According to the national inventory of IND, enteric fermentation is the primary source of CH₄ emissions in the agriculture and waste sector, contributing 61 % of emissions, with rice cultivation accounting for 20 % and waste contributing 16 %. A similar pattern is observed in Bangladesh (BGD), where agricultural emissions are dominated by rice production (48 % in 2012) and enteric fermentation (42 % in 2012). Satellite and in situ inversion estimate emissions in BGD are nearly double those reported by NGHGIs during 2001 and 2012, the year of the last communication. The significant discrepancies between inversions and NGHGIs in IND and BGD may be attributed to potential underestimation of livestock or waste CH₄ emissions by NGHGIs. NGHGIs utilized the Tier 1 method and associated emission factors from the 2006 IPCC Guidelines for National Greenhouse Gas Inventories (IPCC, 2006). However, a recent study (Chang et al., 2021) found that estimates using revised Tier 1 or Tier 2 methods from the 2019 Refinement to the 2006 IPCC Guidelines for National Greenhouse Gas Inventories (IPCC, 2019) give livestock emissions that are 48 %–60 % and 42 %–61 % higher for IND and BGD by 2010, respectively, compared to Tier 1 IPCC (2006) methods, which would bring bottom-up emissions closer to inversions. In Brazil (BRA), both satellite and in situ inversions consistently estimate larger emissions than the NGHGIs by 34 % and 29 %, respectively, and show a consistent increasing trend over their study periods. In the USA, the medians of satellite and in situ inversions are slightly lower than those of NGHGIs, but they exhibit a similar trend throughout the study period. The trend of inversions is comparable to the one of the NGHGIs in BRA during their period of overlap, although there is no NGHGI communication later than 2016. In Argentina (ARG), Pakistan (PAK), and Thailand (THA), the medians of in situ inversions show good consistency with satellite inversion results. Nevertheless, in situ inversion emissions in the 2010s are, on average, 47 % higher in PAK, 20 % higher in ARG, and 64 % higher in THA compared to the NGHGI reports. In the European Union (EUR), emissions from agriculture and waste were reported to have significantly decreased over time in the NGHGI data, mainly from solid waste disposal (Petrescu et al., 2021), a trend that is captured by inversions and is close to the one of the NGHGIs over the study period. In contrast, emissions from agriculture and waste in RUS are reported to have a positive trend after 2010 by the NGHGI, with in situ inversions producing a consistent trend from 2000 to 2014 but a sharp decrease thereafter, while satellite inversions produce stable emissions, although they are lower than those of the NGHGIs and in situ inversions after 2010.

In this section, we compare four different estimates of land CO₂ fluxes during the period 2010–2020 (Fig. 8), including (1) medians of in situ inversion results from our previous study (Deng et al., 2022), (2) medians of in situ and (3) satellite-based inversion results processed in this study based on the Global Carbon Budget 2022 (Friedlingstein et al., 2022), and (4) NGHGIs. This enables a comparison of the median and range of our in situ inversion results (n=5) with those from our previous study (n=6) and an assessment of the performance differences between satellite-based (n=4) and in situ inversion models. To ensure a fair comparison and avoid anomalies in the satellite-based inversion results during 2010–2015 when some of these inversions used GOSAT after 2010 and then OCO-2 after 2015, we separate the analysis into two periods: 2011–2015 and 2016–2020.

The variations in yearly land CO₂ fluxes span a comparable range between the current and previous in situ inversion ensembles, indicating consistency of the inversion results but that the uncertainty within the new in situ inversion ensemble was not improved. However, examining the median values, results from the new in situ inversion ensemble may be closer to those of NGHGIs in most countries (such as China (CHN), the United States (USA), the European Union (EUR), Canada (CAN), Kazakhstan (KAZ), India (IND)). This suggests that the new in situ inversion ensemble used in this study has partially narrowed down the gaps between inversion results and NGHGIs compared to our previous study. However, in Russia (RUS) and Brazil (BRA), the difference between the median of in situ inversion ensembles and NGHGIs has enlarged. For example, in RUS, the new in situ inversion ensemble indicates a larger carbon sink than that from Deng et al. (2022), while the difference between the median of in situ inversions and NGHGIs increased 51 % during 2011–2015 (from 208 to 314 Tg C yr⁻¹) and 49 % during 2016–2020 (from 168 to 249 Tg C yr⁻¹). Conversely, in BRA, the median of the new in situ inversion ensemble indicates a larger carbon source, while the difference increased over 100 % during 2011–2015 (from 200 to 423 Tg C yr⁻¹) and nearly 300 % during 2016–2020 (from 56 to 223 Tg C yr⁻¹).

As for the inversion ensemble used in this study, in most countries, the variations in yearly land CO₂ fluxes also span a similar range between the satellite-based inversion ensemble and in situ inversion ensemble. However, in the cases of the USA, RUS, CHN, and BRA, the spread of satellite-based inversion results is narrower than that of in situ inversion results, indicating better consistency among available satellite-based inversion models, at least when similar satellite data are assimilated. In addition, in most cases, smaller differences were found between the median of inversion results and the NGHGIs. For countries with dense surface monitoring networks such as in the USA and EUR, the satellite-based inversion results show good agreement in terms of in situ inversion results. However, for countries with sparse station coverage like Kazakhstan (KAZ) and Mongolia (MNG), satellite-based inversion results could provide more reliable estimates due to more extensive spatial sampling from satellites, although the medians of satellite-based inversion results indicate larger carbon sinks and larger differences compared with NGHGIs (than for in situ inversion results). In the USA and CAN, the difference during 2011–2015 ( GOSAT-only period) between in situ and satellite-based inversion ensembles is larger than that during 2016–2020 (OCO-2 period). This can be attributed to the use of different satellite data during these periods and different numbers of ensemble members. Before 2015, only GOSAT and only two out of four systems were available. The inversion of OCO-2 data starting in 2014 resulted in a better alignment among ACOS OCO-2 v10 inversions, indicating the in situ and satellite evaluations were similar (Byrne et al., 2023).

Following the method proposed by Grassi et al. (2023), in this study we updated the managed-land mask for Canada (CAN) and Brazil (BRA) using maps of managed land derived from NGHGIs and for Russia (RUS) by adjusting tree-cover threshold in the tree-cover map from Hansen et al. (2013) to match the average area of managed land per oblast (province) that is used for the NGHGIs. Thus, the new mask is now more consistent with the definition of managed land in the NGHGIs for these three countries, so we can further analyze the impacts of different definitions of managed-land masks to separate the managed-land CO₂ fluxes in inversions (Fig. 9). Generally, in Russia (RUS) and Canada (CAN), the managed-land CO₂ fluxes extracted from the new mask are closer to NGHGIs than those separated by the previous mask used by Deng et al. (2022). In addition, in Brazil (BRA), adjusting the national managed-land mask resulted in greater land carbon emissions, increasing the gap with NGHGIs. However, the improvement of the managed-land mask in this study is still not able to explain all the existing discrepancy between inversion estimates and NGHGIs, in which the sources of and reasons for these differences and uncertainties still need further analysis. We also observe in Fig. 9 that the impact of our new managed-land mask compared to the previous one is qualitatively similar whether it is applied to in situ inversions or satellite inversion gridded flux fields.

In our previous study, we found that satellite inversion models appear to have better agreement with NGHGIs than in situ station-based inversion models and, on the other hand, that differences between inversion models and NGHGIs in large oil- and gas-producing countries suggest an underestimation of national reports, possibly due to the omission of ultra-emitting sources by NGHGIs. With the new inversion ensemble in this study, we confirm those results (Fig. 10). In countries such as China (CHN), India (IND), and Russia (RUS), the updated inversion model set provides estimates that are closer to those of NGHGIs, but differences still exist, and the reasons for these differences are not the same. For example, differences in anthropogenic methane emissions in IND are mainly due to differences in agricultural and waste methane flux with the new inversion ensemble used in this study. In RUS, the updated inversion ensemble shows lower fossil fuel emissions, reducing the differences with NGHGIs for this sector, but higher agricultural and waste emissions than in Deng et al. (2022). Nevertheless, the updated fossil fuel emission flux is still higher than the NGHGI estimate for RUS. The remaining differences may be attributed to ultra-emitting sources or underestimated emission factors for some components of the oil and gas extraction and distribution industry in RUS. Conversely, in GULF (GULF comprises Saudi Arabia, Iraq, Kuwait, Oman, the United Arab Emirates, Bahrain, and Qatar), the new inversion model ensemble consistently reflects higher fossil fuel emission fluxes than NGHGIs, like in our previous study, and expands the difference in estimates of artificial methane flux between inversion models and NGHGIs, possibly indicating more methane leakage.

The use of different priors can also influence the inversion results of the data. Figure 11 presents the sets of inversion results using EDGAR (blue) and GAINS (purple) as priors. In most countries, the median values of the two inversion result sets are similar. However, in countries such as Russia (RUS), the United States (USA), Iran (IRN), and Mexico (MEX), significant differences are observed between the two inversion result sets, which may primarily stem from the differences in the inversion results for fossil CH₄ emissions (Fig. 12). In RUS and USA, the inversion results using GAINS as priors are consistently higher than those using EDGAR as priors. In RUS, the satellite inversion results using GAINS as priors are higher by 45 % during 2010–2020, and the ground-based inversion results are higher by 75 % during 2000–2020. In the case of the USA, the inversion results using GAINS as priors exhibit a completely different trend compared to the ones obtained using NGHGIs and EDGAR as priors. The inversion results using GAINS as priors, from both satellite and ground-based measurements, show a rapid growth trend by increasing by 24 % from 2010 to 2020. In IRN and MEX, the inversion results using GAINS as priors are lower than those using EDGAR as priors. For IRN, the differences between satellite inversion results using different priors are not significant, and the trends are similar. However, the ground-based inversion results are very close between 2000–2013, but after 2013, a steep increase is observed in the ground-based inversion results using GAINS as priors. On the other hand, in MEX, the ground-based inversion results are similar, but the satellite inversion results using GAINS as priors are relatively low, by 14 % on average. Such discrepancies may arise from differences in inventory methodologies and the resulting estimations. As shown in Supplement Fig. S1 in Tibrewal et al. (2024), similar discrepancies were found between the two inventories in these countries, which report a higher estimation from GAINS in RUS and USA compared to EDGAR during 2011–2020 and a lower estimation in IRN. As noted in Tibrewal et al. (2024), EDGAR is based on various versions of national inventory reports (NIRs) that utilize different combinations of emission factors from the IPCC, while GAINS employs an independent estimation approach. This highlights the critical role of prior data selection in determining the accuracy of CH₄ emission estimates.

The updated N₂O inversion results show systematically higher anthropogenic emissions than our previous N₂O inversion results (Deng et al., 2022), resulting in larger discrepancies between N₂O inversion results and NGHGIs in most countries in Fig. 13. Countries such as Brazil (BRA), the Democratic Republic of the Congo (COD), Indonesia (IDN), Colombia (COL), Sudan (SDN), Australia (AUS), and Venezuela (VEN) exhibit significant differences. These discrepancies may be attributed to the use of lower IPCC default emission factors in the national inventories of these tropical countries, leading to lower NGHGI results. The IPCC default emission factors are derived from measurements primarily conducted in temperate regions of the Northern Hemisphere (e.g., Europe and the United States (USA)), which explains the better alignment of inversion results with inventories in those regions. Notably, in the case of the USA, the median of the updated N₂O inversion results is very close to that of NGHGIs. The median of the N₂O inversion results from Deng et al. (2022) was 42 % lower than that of the NGHGIs between 2005 and 2015, whereas the median of the updated inversion models is only 4 % lower. This demonstrates improved consistency in the updated inversion system results for the USA. Additionally, in countries such as India (IND), IDN, COL, COD, Sudan (SDN), and VEN, our N₂O inversion results have a larger distribution compared to the previous study, indicating that the new N₂O inversion ensemble (n=4) has less consistency in these countries compared to the previous ensemble (n=3).