The estimates of global and national fossil CO₂ emissions (EFOS) include the oxidation of fossil fuels through both combustion (e.g., transport, heating) and chemical oxidation (e.g., carbon anode decomposition in aluminium refining) activities, and the decomposition of carbonates in industrial processes (e.g. the production of cement). We also include CO₂ uptake from the cement carbonation process. Several emissions sources are not estimated or not fully covered: coverage of emissions from lime production are not global, and decomposition of carbonates in glass and ceramic production are included only for the “Annex 1” countries of the United Nations Framework Convention on Climate Change (UNFCCC) for lack of activity data. These omissions are considered to be minor. Short-cycle carbon emissions – for example from combustion of biomass – are not included here but are accounted for in the CO₂ emissions from land use (see Sect. 2.2).

Our estimates of fossil CO₂ emissions rely on data collection by many other parties. Our goal is to produce the best estimate of this flux, and we therefore use a prioritisation framework to combine data from different sources that have used different methods, while being careful to avoid double counting and undercounting of emissions sources. The CDIAC-FF emissions dataset, derived largely from UN energy data, forms the foundation, and we extend emissions to 2024 using energy growth rates reported by the Energy Institute (a dataset formerly produced by BP). We then proceed to replace estimates using data from what we consider to be superior sources, for example Annex 1 countries' official submissions to the UNFCCC. All data points are potentially subject to revision, not just the latest year. For full details see Andrew and Peters (2025).

Other estimates of global fossil CO₂ emissions exist, and these are compared by Andrew (2020a). The most common reason for differences in estimates of global fossil CO₂ emissions is a difference in which emissions sources are included in the datasets. Datasets such as those published by the Energy Institute, the US Energy Information Administration, and the International Energy Agency's “CO₂ emissions from fuel combustion” are all generally limited to emissions from combustion of fossil fuels. In contrast, datasets such as PRIMAP-hist, CEDS, EDGAR, and GCP's dataset aim to include all sources of fossil CO₂ emissions. See Andrew (2020a) for detailed comparisons and discussion.

Cement products absorb CO₂ from the atmosphere over their lifetimes, a process known as “cement carbonation”. We estimate this CO₂ sink, from 1931 onwards, as the average of two studies in the literature (Cao et al., 2020; Guo et al., 2021). Both studies use the same model, developed by Xi et al. (2016), with different parameterisations and input data, with the estimate of Guo and colleagues being a revision of Xi et al. (2016). The trends of the two studies are very similar. Since carbonation is a function of both current and previous cement production, we extend these estimates to 2024 by using the growth rate derived from the smoothed cement emissions (10-year smoothing) fitted to the carbonation data. In the present budget, we always include the cement carbonation carbon sink in the fossil CO₂ emission component (EFOS) unless explicitly stated otherwise.

We use the Kaya Identity for a simple decomposition of CO₂ emissions into the key drivers (Raupach et al., 2007). While there are variants (Peters et al., 2017b), we focus here on a decomposition of CO₂ emissions into population, GDP per person, energy use per GDP, and CO₂ emissions per energy. Multiplying these individual components together returns the CO₂ emissions. Using the decomposition, it is possible to attribute the change in CO₂ emissions to the change in each of the drivers. This method gives a first order understanding of what causes CO₂ emissions to change each year.

We provide a projection of global fossil CO₂ emissions in 2025 by combining separate projections for China, USA, EU, India, Japan, and for all other countries combined. The methods are different for each of these. For China we use growth rates from the 2026 edition of the National Bureau of Statistics' Statistical Communique along with cement clinker trade data and estimated changes in lime production. For the USA our projection is taken directly from the Energy Information Administration's (EIA) Short-Term Energy Outlook (EIA, 2023), combined with the year-to-date growth rate of cement clinker production. For the EU we use monthly energy data from Eurostat to derive estimates of monthly CO₂ emissions through December (Andrew, 2021). EU cement emissions are based on available production data through December from three of the largest producers, Germany, Poland, and Spain as well as production indices for Italy and France. India's projected emissions are derived from monthly estimates through December using the methods of Andrew (2020b). Japan's emissions are based on monthly estimates through December. Emissions from international transportation (bunkers) are estimated separately for aviation and shipping. Changes in aviation emissions are derived primarily from OECD monthly estimates, extrapolated using the growth rates of global flight miles from Airportia. Shipping emissions are derived from estimates of sales of bunker fuels provided by Ship and Bunker (Ship and Bunker, 2026). Emissions for the rest of the world are derived for coal and cement using projected growth in economic production from the IMF (2025) combined with extrapolated changes in emissions intensity of economic production; for oil using a global constraint from EIA; and for natural gas using a global constraint from IEA. More details on the EFOS methodology and its 2025 projection can be found in Supplement Sect. S1.

We compare our 2025 projection with the Carbon Monitor (2025). Carbon Monitor is a dataset of daily emissions constructed using hourly to daily proxy data (e.g., electricity consumption, travel patterns, etc) instead of energy use data (Liu et al., 2020a, b). Emissions estimates from January to December are combined to give a full-year 2025 estimate.

The net CO₂ flux from land-use, land-use change and forestry (ELUC, called land-use change emissions in the rest of the text) includes CO₂ fluxes from deforestation, afforestation, logging, and forest degradation (including harvest activity), shifting cultivation (cycle of cutting forest for agriculture, then abandoning), regrowth of forests (following wood harvest or agriculture abandonment), peat burning, and peat drainage.

Updated estimates from three bookkeeping models are used to quantify gross emissions, gross removals, and the resulting net ELUC: BLUE (Hansis et al., 2015), LUCE (Qin et al., 2024), and OSCAR (Gasser et al., 2020). An important improvement compared to previous GCBs is the use of transient carbon densities by all three bookkeeping models, i.e., they consider the effects of environmental changes, such as atmospheric CO₂ increase, on vegetation and soil carbon densities, to estimate ELUC (Gasser et al., 2020; Dorgeist et al., 2024). The GCB assessments have for a long time also (and initially exclusively) used estimates from the bookkeeping model HandC2023 (Houghton and Castanho, 2023), its predecessor HandN2017 (Houghton and Nassikas, 2017), and earlier versions. However, HandC2023 does not consider transient carbon densities and only provides data up to 2020, which implies the need to extrapolate its data to the time after 2020. As the model does thus not incorporate the most recent developments in bookkeeping modeling, it is not used anymore in GCB2025.

Emissions from peat burning and peat drainage are added from external datasets (see Supplement Sect. 2.2): peat fire emissions from the Global Fire Emission Database (GFED4s; van der Werf et al., 2017) and peat drainage emissions averaged from estimates of the Food Agriculture Organization (Conchedda and Tubiello, 2020; FAO, 2025a) and from simulations with the DGVM ORCHIDEE-PEAT (Qiu et al., 2021) and the DGVM LPX-Bern (Lienert and Joos, 2018; Müller and Joos, 2021).

Uncertainty estimates were derived from the Dynamic Global Vegetation Models (DGVMs) ensemble for the time period prior to 1960, and using for the recent decades an uncertainty range of ±0.7 GtC yr⁻¹, which is a semi-quantitative measure for annual and decadal emissions and reflects our best value judgement that there is at least 68 % chance (±1σ) that the true land-use change emission lies within the given range, for the range of processes considered here.

The GCB ELUC estimates follow the CO₂ flux definition of global carbon cycle models and differ from IPCC definitions adopted in National GHG Inventories (NGHGI) for reporting under the UNFCCC. The latter typically include terrestrial fluxes occurring on all land that countries define as managed, following the IPCC managed land proxy approach (Grassi et al., 2018). This partly includes fluxes due to environmental change (e.g., atmospheric CO₂ increase), which are part of SLAND in our definition. As a result, global emission estimates are smaller for NGHGI than for the global carbon budget definition (Grassi et al., 2023). The same is the case for the FAO estimates of carbon fluxes on forest land, which include both anthropogenic and natural fluxes on managed land (Tubiello et al., 2025; FAO, 2025b). Using the NGHGI data collected and processed in the LULUCF data hub V3.1 (Melo et al., 2025), we translate the GCB and NGHGI definitions to each other, to provide a comparison of the anthropogenic carbon budget as reported in GCB to the official country reporting to the UNFCCC convention. We further compare these estimates with the net atmosphere-to-land flux from atmospheric inversion systems (see Sect. 2.7), averaged over managed land only.

ELUC contains a range of fluxes that are related to Carbon Dioxide Removal (CDR). CDR is defined as the set of anthropogenic activities that remove CO₂ from the atmosphere, in addition to the Earth's natural processes (such as carbon uptake in response to atmospheric CO₂ increase), and store it in durable form, such as in forest biomass, soils, long-lived products, oceans or geological reservoirs. Here, we quantify vegetation-based CDR that is implicitly or explicitly captured by land-use fluxes (CDR not based on vegetation is discussed in Sect. 2.3). We quantify CDR through re/afforestation from the three bookkeeping estimates by separating permanent increases in forest cover, which counts as CDR, from forest regrowth in shifting cultivation cycles (not part of CDR; see Supplement Sect. 2.2). It should be noted that the permanence of the storage under climate risks such as fire is increasingly questioned. Other CDR activities related to land use but not fully accounted for in our ELUC estimate include the transfer of carbon to harvested wood products (HWP), bioenergy with carbon capture and storage (BECCS), and biochar production (Babiker et al., 2022; Smith et al., 2024). The different bookkeeping models all represent HWP but with varying details concerning product usage and their lifetimes. BECCS and biochar are currently only represented in bookkeeping and DGVM models regarding the CO₂ removal through photosynthesis, without accounting for the durable storage. HWP, BECCS, and biochar are typically counted as CDR once the transfer to the durable storage site occurs and not when the CO₂ is removed from the atmosphere, which complicates a direct comparison to the GCB approach to quantify annual fluxes to and from the atmosphere. We provide estimates for CDR through HWP, BECCS, and biochar based on independent studies in Sect. 3.2.2. HWP and BECCS estimates reflect updated 2024 data of the State of CDR report (Smith et al., 2024), while biochar estimates correspond to 2023 due to unavailability of newer data. We do not add them to our ELUC estimate to avoid potential double-counting that arises from the partial consideration of HWP, BECCS, and biochar in the bookkeeping and DGVM models and to avoid inconsistencies from the temporal discrepancy between transfer to storage and removal from the atmosphere. More details on the ELUC methodology can be found in Supplement Sect. S2.

We project the 2025 land-use emissions for BLUE, OSCAR, and LUCE based on their ELUC estimates for 2023 and adding the anomalies in carbon emissions from peat fires in equatorial Asia and tropical deforestation and degradation fires (2025 emissions relative to 2023 emissions) from GFED4s (van der Werf et al., 2017) estimated using active fire data (MCD14ML; Giglio et al., 2016). 2023 is used as base year since the ELUC estimate for the year 2024 is informed by extrapolated land-use data (Supplement Sect. S2.1). Peat drainage is assumed to be unaltered as it has low interannual variability.

The rate of growth of the atmospheric CO₂ concentration is provided for years 1959–2024 by the US National Oceanic and Atmospheric Administration Global Monitoring Laboratory (NOAA/GML; Lan et al., 2025), which includes recent revisions to the calibration scale of atmospheric CO₂ measurements (WMO-CO₂-X2019; Hall et al., 2021). For the 1959–1979 period, the global growth rate is based on measurements of atmospheric CO₂ concentration averaged from the Mauna Loa and South Pole stations, as observed by the CO₂ Program at Scripps Institution of Oceanography (SIO, Keeling et al., 1976). For the 1980–2024 period, the global growth rate is based on the average of multiple stations selected from the marine boundary layer sites with well-mixed background air (Lan et al., 2024), after fitting a smooth curve through the data for each station as a function of time, and averaging by latitude band (Masarie and Tans, 1995). The annual growth rate is estimated by Lan et al. (2025) from atmospheric CO₂ concentration by taking the average of the most recent December–January months corrected for the average seasonal cycle and subtracting this same average one year earlier. To obtain GATM, the observation-based growth rate in units of ppm yr⁻¹ is converted to units of GtC yr⁻¹ by multiplying by a factor of 2.124 GtC ppm⁻¹, assuming instantaneous mixing of CO₂ throughout the atmosphere (Ballantyne et al., 2012; Table 1). There is high confidence in the observations because they are based on direct measurements from stations distributed around the world (Lan et al., 2024) with all CO₂ measurements consistently measured against the same CO₂ standard scale (WMO X2019) defined by a suite of gas standards (Hall et al., 2021). However, the conversion to estimates of GATM in GtC yr⁻¹ incurs large uncertainty on annual time scale as discussed next.

The uncertainty around GATM is due to three main factors. First, the network composition of the marine boundary layer sites with some sites being introduced or removed over time, gaps in sites time series, etc. This uncertainty was estimated with a bootstrap method by constructing 100 “alternative” networks (Steele et al., 1992; Masarie and Tans, 1995; Lan et al., 2025). Second, the analytical uncertainty that describes the short- and long-term uncertainties associated with the CO₂ analyzers. A Monte Carlo method was used to estimate the total analytical uncertainty by randomly selecting errors to add to each observation from a normal distribution of combined short- and long-term uncertainties. Prior to the 1980s when analyzers were less precise and CO₂ measurement scale was slightly less well defined, larger analytical errors were assigned to account for these factors. However, the network uncertainty remains the larger term of uncertainty. The first and second uncertainties are reported as 1σ standard deviations (i.e., 68 % confidence interval), and summed in quadrature to determine the global surface growth rate uncertainty, which averaged to 0.085 ppm (Lan et al., 2024). Third, the uncertainty associated with using the average CO₂ concentration from a surface network to approximate the true atmospheric average CO₂ concentration (mass-weighted, in 3 dimensions) as needed to assess the total atmospheric CO₂ burden. In reality, CO₂ variations measured at the stations will not exactly track changes in total atmospheric burden, with offsets in magnitude and phasing due to vertical and horizontal mixing (Pandey et al., 2025). This effect must be very small on decadal and longer time scales, when the atmosphere can be considered well mixed. The long-term CO₂ increase in the stratosphere lags the increase (meaning lower concentrations) that we observe in the marine boundary layer, while the continental boundary layer (where most of the emissions take place) leads the marine boundary layer with higher concentrations. These effects nearly cancel each other on decadal time scales, when the growth rate is nearly the same everywhere (Ballantyne et al., 2012). We therefore maintain an uncertainty around the annual growth rate based on the multiple stations dataset ranges between 0.11 and 0.72 GtC yr⁻¹, with a mean of 0.61 GtC yr⁻¹ for 1959–1979 and 0.17 GtC yr⁻¹ for 1980–2024, when more measurement sites were available (Lan et al., 2025). We estimate the uncertainty of the decadal averaged growth rate after 1980 at 0.02 GtC yr⁻¹ based on the annual growth rate uncertainty but stretched over a 10-year interval. For years prior to 1980, we estimate the decadal averaged uncertainty to be 0.07 GtC yr⁻¹ based on a factor proportional to the annual uncertainty prior and after 1980 (0.02×[0.61/0.17] GtC yr⁻¹).

To estimate the total carbon accumulated in the atmosphere since 1750 or 1850, we use an atmospheric CO₂ concentration of 278.0 ± 3 ppm or 287.7 ± 3 ppm, respectively (Gulev et al., 2021). For the construction of the historical budget shown in Fig. 3, we use the fitted estimates of CO₂ concentration from Joos and Spahni (2008) to estimate the annual atmospheric growth rate (see Supplement Sect. S7). The uncertainty of ±3 pm (converted to ±1σ) is taken directly from the IPCC's AR5 assessment (Ciais et al., 2013). Typical uncertainties in the growth rate in atmospheric CO₂ concentration from ice core data are equivalent to ±0.1–0.15 GtC yr⁻¹ as evaluated from the Law Dome data (Etheridge et al., 1996) for individual 20-year intervals over the period from 1850 to 1960 (Bruno and Joos, 1997).

Further opportunities to estimate global CO₂ growth rates are offered by space-based platforms such as GOSAT (since 2010) and OCO-2 (since 2015). Their recorded short-wave infrared spectra allow retrieval of column CO₂ abundances (XCO₂) over cloud-free scenes over land and ocean with footprints of 80 (GOSAT) and 3 (OCO-2) km² respectively. The columns are not full integrals; the sensitivity to surface and lower-tropospheric CO₂ mole fractions is much higher than to upper tropospheric and stratospheric CO₂. However, for the purpose of calculating atmospheric growth rates, the sensitivity is not limiting, as the pressure-weighted sensitivity is still above 0.5 in the stratosphere (Pandey et al., 2024) and priors used in OCO-2 account for the slow stratospheric-tropospheric exchanges using the age of air (Laughner et al., 2023). Here, we use OCO-2 based whole atmosphere growth rates using the Growth Rate from Satellite Observations (GRESO) approach (Pandey et al., 2024). We specifically note that their retrievals are evaluated against surface-based remote sensing (Total Column Carbon Observing Network) data, which in turn are tied to atmospheric observations (from aircraft and using aircore) (Wunch et al., 2017). Furthermore, OCO-2 retrievals use a priori CO₂ mole fraction profiles tied to NOAA GML in-situ flask records at Mauna Loa (MLO) and American Samoa (SMO) measurement stations (Laughner et al., 2023). For details on evaluation, bias-correction, and spatiotemporal coverage of OCO-2 we refer to O'Dell et al. (2018).

Relative to surface observations, the GRESO product typically reflects tropical growth rate anomalies earlier and sees whole atmosphere carbon stock changes with a lower latency. The GRESO growth rates presented here use both land and ocean observations of OCO-2 providing global sampling of the atmosphere. We used GRESO post 2015 to present an alternative quantification of the atmospheric carbon stock changes in GATM, using the reported 1σ uncertainty of 0.08 ppm yr⁻¹. The mean differences between GRESO and surface-data derived GATM is 0.05 ± 0.26 ppm yr⁻¹ over the 2015–2024 (n=10 years) period.

We provide an assessment of GATM for 2025 as the average of two methods. First, the GCB regression method models monthly global-average atmospheric CO₂ concentrations and derives the increment and annual average from these. The model uses lagged observations of concentration (Lan et al., 2025): both a 12-month lag, and the lowest lag that will allow model prediction to produce an estimate for the following January, recalling that the GATM increment is derived from December/January pairs. The largest driver of interannual changes is the ENSO signal (Betts et al., 2016), so the monthly ENSO 3.4 index (Huang et al., 2017) is included in the model. Given the natural lag between sea-surface temperatures and effects on the biosphere, and in turn effects on globally mixed atmospheric CO₂ concentration, a lagged ENSO index is used, and we use both a 5-month and a 6-month lag. The combination of the two lagged ENSO values helps reduce possible effects of noise in a single month. To help characterise the seasonal variation, we add month as a categorical variable. Finally, we flag the period affected by the Pinatubo eruption (August 1991–November 1993) as a categorical variable.

The second method uses the multi-model mean and uncertainty of the 2025 GATM estimated by the ESMs prediction system (see Sect. 2.9). We then take the average of the GCB regression and ESMs GATM estimates, with their respective uncertainty combined quadratically.

Similarly, the projection of the 2025 global average CO₂ concentration (in ppm), is calculated as the average of the estimates from the two methods. For the GCB regression method, it is the annual average of global concentration over the 12 months of 2025; for the ESMs, it is the observed global average CO₂ concentration for 2024 plus the annual increase in 2025 of the global average CO₂ concentration, which is an average of NOAA/GML measurements from January to June (Lan et al., 2025) and predictions of the ESMs multi-model mean from July to December (see Sect. 2.9).

The reported estimate of the global ocean anthropogenic CO₂ sink SOCEAN is derived as the average of two estimates. The first estimate is derived as the mean over an ensemble of ten global ocean biogeochemistry models (GOBMs, Tables 4 and S2). The second estimate is obtained as the mean over an ensemble of nine surface ocean fCO₂-observation-based data-products (Tables 4 and S3).

The GOBMs simulate both the natural and anthropogenic CO₂ cycles in the ocean. They constrain the anthropogenic air-sea CO₂ flux (the dominant component of SOCEAN) by the transport of carbon into the ocean interior, which is also the controlling factor of present-day ocean carbon uptake in the real world. They cover the full globe and all seasons and were evaluated against surface ocean carbon observations, suggesting they are suitable to estimate the annual ocean carbon sink (Hauck et al., 2020). We derive SOCEAN from GOBMs by using a simulation (sim A) with historical forcing of climate and atmospheric CO₂ from GCB (Sect. 2.4), accounting for model biases and drift from a control simulation (sim B) with constant atmospheric CO₂ and normal year climate forcing. A third simulation (sim C) with historical atmospheric CO₂ increase and normal year climate forcing is used to attribute the ocean sink to CO₂ (sim C minus sim B) and climate (sim A minus sim C) effects. A fourth simulation (sim D; historical climate forcing and constant atmospheric CO₂) is used to compare the change in anthropogenic carbon inventory in the interior ocean (sim A minus sim D) to the observation-based estimates of Gruber et al. (2019) and Müller et al. (2023) with the same flux components (steady state and non-steady state anthropogenic carbon flux).

As there is accumulating evidence for a 10 %–20 % underestimation of SOCEAN by the GOBMs based on ocean interior data (Sect. 3.6.5), atmospheric oxygen (Sect. 3.6.2), atmospheric inversions (Sect. 3.8) and supported by eddy-covariance measurements (Dong et al., 2024a), we scale up the global GOBM multi-model mean estimates by 10 % when estimating SOCEAN (Friedlingstein et al., 2025b) (see Supplement Sect. S3.3). Analysis of Earth System Models and GOBMs indicate that an underestimation by about 10 % may be due to biases in ocean carbon transport and mixing from the surface mixed layer to the ocean interior (Goris et al., 2018; Terhaar et al., 2021; Bourgeois et al., 2022; Terhaar et al., 2022), biases in the chemical buffer capacity (Revelle factor) of the ocean (Vaittinada Ayar et al., 2022; Terhaar et al., 2022) and partly due to a late starting date of the simulations (mirrored in atmospheric CO₂ chosen for the pre-industrial control simulation, Table S2, Bronselaer et al., 2017, Terhaar et al., 2022, 2024). GOBMs are evaluated against key metrics and their capability to reproduce various statistical measures of physical and biogeochemical fields using the International Ocean Model Benchmarking (IOMB) Scheme (Fu et al., 2022, Sect. S3.3).

The fCO₂-products are tightly linked to observations of fCO₂ (fugacity of CO₂, which equals pCO₂ corrected for the non-ideal behaviour of the gas; Pfeil et al., 2013), which carry imprints of temporal and spatial variability, but are also sensitive to uncertainties in gas-exchange parameterizations and data-sparsity (Fay et al., 2021, Gloege et al., 2021, Hauck et al., 2023a). Their asset is the assessment of the mean spatial pattern of variability and its seasonality (Hauck et al., 2020; Gloege et al., 2021; Hauck et al., 2023a). To benchmark biases and trends derived from the fCO₂-products, we update a model subsampling exercise following Hauck et al. (2023a, see Sect. S3) and use independent measurements from the SOCAT (Surface Ocean CO₂ Atlas) flag E dataset (uncertainty less than 10 µatm) (Bakker et al., 2016) and calculated fCO₂ from the Global Ocean Data Analysis Project GLODAP (Lauvset et al., 2024). New in GCB2025, we now also include the UExP-FNN-U product in the ensemble mean, which was previously shown but not included. We include it since new evidence emerged from field and modelling studies, recommending adopting a temperature correction to the fCO₂ from measurement depth to the surface skin layer where the gas exchange takes place (Dong et al., 2022; Ford et al., 2024a; Bellenger et al., 2023). Furthermore, a second fCO₂-product submitted temperature corrected fCO₂ (JMA-MLR) this year, so we now include these corrected estimates in the ensemble mean. Additionally, following Friedlingstein et al. (2025b), we now also add 0.18 GtC yr⁻¹ (multiplied by 7/9 as two products already include a temperature correction) to the multi fCO₂-product average when calculating SOCEAN to account for the warm layer and cool skin effect (See Supplement Sect. S3.1). This correction is based on a recent field study (Ford et al., 2024a) and broadly consistent in magnitude with a GOBM study (Bellenger et al., 2023).

The global fCO₂-based flux estimates were adjusted to remove the pre-industrial natural ocean source of CO₂ to the atmosphere of 0.65 ± 0.3 GtC yr⁻¹, arising from the transfer of carbon from land to ocean via rivers (Regnier et al., 2022), to satisfy our definition of SOCEAN (Hauck et al., 2020). This CO₂ outgassing adjustment was distributed over the latitudinal bands using the regional distribution of Lacroix et al. (2020; North: 0.14 GtC yr⁻¹, Tropics: 0.42 GtC yr⁻¹, South: 0.09 GtC yr⁻¹). Acknowledging that this distribution is based on only one model, the advantage is that a gridded field is available, and the adjustment can be calculated for the three latitudinal bands and the RECCAP regions (REgional Carbon Cycle Assessment and Processes (RECCAP2; Ciais et al., 2022; Poulter et al., 2022; DeVries et al., 2023). This dataset suggests that more of the river-induced outgassing occurs in the tropics than in the Southern Ocean and is thus opposed to the previously used dataset of Aumont et al. (2001). Accordingly, the regional distribution is associated with an additional uncertainty in addition to the large uncertainty around the global estimate (Crisp et al., 2022; Gruber et al., 2023). Anthropogenic perturbations of river carbon and nutrient transport to the ocean are not represented in the process models used to quantify SOCEAN, but an a-posteriori correction is applied to the global SOCEAN estimate (see Sect. 2.10 and Supplement Sect. S8.3). We calculate SOCEAN as the average of the GOBM ensemble mean and the fCO₂-product ensemble mean from 1990 onwards, including the corrections on the GOBM and fCO₂-product ensemble means as described above. For the 1959–1989 period, it is calculated as the GOBM ensemble mean plus half of the offset between GOBMs and fCO₂-products ensemble means over 1990–2001, also including the above corrections. In addition, two diagnostic ocean models are used to estimate SOCEAN over the industrial era (1781–1958, Khatiwala et al., 2013 and DeVries, 2014).

SOCEAN is evaluated against the change in the total dissolved inorganic carbon inventory in the interior ocean from the observation-based estimates of Keppler et al. (2023) with the same flux components.

We assign an uncertainty of ±0.4 GtC yr⁻¹ to the ocean sink based on a combination of random (ensemble standard deviation) and systematic uncertainties (GOBMs bias in anthropogenic carbon accumulation, previously reported uncertainties in fCO₂-products; see Supplement Sect. S3.4). While this approach is consistent within the GCB, an independent uncertainty assessment of the fCO₂-products alone suggests a somewhat larger uncertainty of up to 0.7 GtC yr⁻¹ (Ford et al., 2024a). We assess a medium confidence level to the annual ocean CO₂ sink and its uncertainty because it is based on multiple lines of evidence, it is consistent with ocean interior carbon estimates (see Sect. 3.6.5) and the interannual variability in the GOBMs and data-based estimates is largely consistent and can be explained by climate variability. We refrain from assigning a high confidence because of the deviation between the GOBM and fCO₂-product trends between around 2002 and 2020 and the higher SOCEAN estimate from O2:N2 (Sect. 2.8). More details on the SOCEAN methodology can be found in Supplement Sect. S3.

The SOCEAN forecast for the year 2025 is based on (a) the historical (Lan et al., 2025) and our 2025 estimate of atmospheric CO₂ concentration, (b) the historical and our 2025 estimate of global fossil emissions, and (c) the boreal spring (March, April, May) Oceanic Niño Index (ONI) (NCEP, 2025). Using a non-linear regression approach, i.e., a feed-forward neural network, atmospheric CO₂, ONI, and the fossil emissions are used as training data to best match the corrected SOCEAN from 1959 through 2024 from this year's carbon budget. Using this relationship, the 2025 SOCEAN can then be estimated from the projected 2025 input data using the non-linear relationship established during the network training. To avoid overfitting, the neural network training was done using a Monte Carlo approach, with a variable number of artificial neurons (varying between 2–5) and 20 % of the randomly selected training data were withheld for independent internal testing.

Based on the best output performance (tested using the 20 % withheld input data), the best performing number of neurons was selected. In a second step, we trained the network 10 times using the best number of neurons identified in step 1 and different sets of randomly selected training data. The mean of the 10 trainings is considered our best forecast, whereas the standard deviation of the 10 ensembles provides a first order estimate of the forecast uncertainty. This uncertainty is then combined with the SOCEAN uncertainty (0.4 GtC yr⁻¹) to estimate the overall uncertainty of the 2025 projection. As an additional line of evidence, we also assess the 2025 atmosphere-ocean carbon flux from the ESM prediction system (see Sect. 2.9).

The terrestrial land sink (SLAND) is thought to be due to the combined effects of rising atmospheric CO₂, increasing N inputs, and climate change, on plant growth and terrestrial carbon storage. SLAND does not include land sinks directly resulting from land-use and land-use change (e.g., regrowth of vegetation) as these are part of the land-use change emissions (ELUC), although system boundaries make it difficult to attribute exactly CO₂ fluxes on land between SLAND and ELUC (Erb et al., 2013).

SLAND is derived from the multi-model mean of 22 DGVMs (Tables 4 and S1 in the Supplement). DGVMs simulations include all climate variability and CO₂ effects over land. In addition to the carbon cycle represented in all DGVMs, 15 models also account for the nitrogen cycle and hence can include the effect of N inputs on SLAND. The DGVMs estimates of SLAND do not explicitly include the export of carbon to aquatic systems or its historical perturbation, which is discussed in Supplement Sect. S8.3. DGVMs need to meet several criteria to be included in this assessment (see Supplement Sect. S4.2). In addition, we use the International Land Model Benchmarking system (ILAMB; Collier et al., 2018) for the DGVMs evaluation (see Supplement Sect. S4.2), with an additional comparison of DGVMs with a data-informed, Bayesian model-data fusion framework (CARDAMOM) (Bloom and Williams, 2015; Bloom et al., 2016). The uncertainty on SLAND is taken from the DGVMs standard deviation.

New to GCB2025 is a correction applied to the SLAND estimate to account for its overestimation resulting from the assumption of pre-industrial land-use in the DGVM simulations, when in reality a large portion of the land surface has been converted to pasture and cropland, with a lower sink capacity. This bias is termed the Replaced Sinks and Sources (RSS) (Gitz and Ciais, 2003; Sitch et al., 2005; Pongratz et al., 2009; Gasser et al., 2020; Obermeier et al., 2021; Dorgeist et al., 2024). The correction, which is only applied when estimating the global SLAND, utilises results from a subset of DGVMs that were able to supply Net Biome Productivity estimates at a Plant Functional Type basis combined with time-varying PFT area fractions from the simulation with land use and land use change considered (O'Sullivan et al., 2025; Friedlingstein et al., 2025b), see Sect. S4.1 for details on methodology. The corrected SLAND is reduced by 19 % globally when accounting for the RSS. More details on the SLAND methodology can be found in Supplement Sect. S4.

In previous versions of the GCB, the land sink projection was, like the ocean sink forecast, based on a non-linear regression approach with a feed-forward neural network. This approach, however, resulted in a large uncertainty and underestimated the land sink variability in the past. This year, we update the projection and simply calculate the land sink for 2025 as the residual of the projection of the other components of the carbon cycle (SLAND=EFOS+ELUC-GATM-SOCEAN). Hence, by construction the 2025 BIM is set to zero.

Cumulative fossil CO₂ emissions for 1850–2024 were 495 ± 25 GtC, including the cement carbonation sink (Table 8, with all cumulative numbers rounded to the nearest 5 GtC). In this period, 46 % of global fossil CO₂ emissions came from coal, 35 % from oil, 15 % from natural gas, 3 % from decomposition of carbonates for cement production, and 1 % from flaring. In 1850, the UK stood for 62 % of global fossil CO₂ emissions. In 1893 the combined cumulative emissions of the current members of the European Union reached and subsequently surpassed the level of the UK. Since 1917 US cumulative emissions have been the largest. Over the entire period 1850–2024, US cumulative emissions amounted to 120 GtC (24 % of world total), the EU's to 80 GtC (16 %), China's to 80 GtC (15 %), and India's to 18 GtC (4 %).

In addition to the estimates of fossil CO₂ emissions that we provide here (see Sect. 2.1), there are three global datasets with long time series that include all sources of fossil CO₂ emissions: CDIAC-FF (Erb and Marland, 2025), CEDS version 2024_07_08 (Hoesly et al., 2024) and PRIMAP-hist version 2.6 (Gütschow et al., 2016, 2023), although these datasets are not entirely independent from each other (Andrew, 2020a). CEDS has cumulative emissions over 1750–2022 at 480 GtC, CDIAC-FF has 481 GtC, GCP 484 GtC, PRIMAP-hist CR 490 GtC, and PRIMAP-hist TR 492 GtC. CDIAC-FF excludes emissions from lime production. CEDS estimates higher emissions from international shipping in recent years, while PRIMAP-hist has higher fugitive emissions than the other datasets. However, in general these four datasets are in relative agreement as to total historical global emissions of fossil CO₂.

Global fossil CO₂ emissions, EFOS (including the cement carbonation sink), have increased every decade from an average of 3.0 ± 0.2 GtC yr⁻¹ for the decade of the 1960s to an average of 9.8 ± 0.5 GtC yr⁻¹ during 2015–2024 (Table 7, Figs. 2 and 4). The growth rate in these emissions decreased between the 1960s and the 1990s, from 4.3 % yr⁻¹ in the 1960s (1960–1969), 3.1 % yr⁻¹ in the 1970s (1970–1979), 1.5 % yr⁻¹ in the 1980s (1980–1989), to 1.0 % yr⁻¹ in the 1990s (1990–1999). After this period, the growth rate began increasing again in the 2000s at an average growth rate of 2.8 % yr⁻¹, decreasing to 1.2 % yr⁻¹ in the 2010s, and 0.8 % yr⁻¹ for the last decade (2015–2024). China's emissions increased by 2.5 % yr⁻¹ on average over the last 10 years dominating the global trend, and India's emissions increased by 3.6 % yr⁻¹, while emissions decreased in EU27 (-2.5 % yr⁻¹), and in the USA (-1.2 % yr⁻¹) (Fig. 5a). Figure 6 illustrates the spatial distribution of fossil fuel emissions for the 2015–2024 period.

EFOS reported here includes the uptake of CO₂ by cement via carbonation which has increased with increasing stocks of cement products, from an average of 20 MtC yr⁻¹ (0.02 GtC yr⁻¹) in the 1960s to an average of 210 MtC yr⁻¹ (0.21 GtC yr⁻¹) during 2015–2024 (Fig. 5b).

Global fossil CO₂ emissions were slightly higher, 1.1 %, in 2024 than in 2023, with an increase of 0.11 GtC to reach 10.3 ± 0.5 GtC (including the 0.22 GtC cement carbonation sink) in 2024 (Fig. 4), distributed among coal (41 %), oil (32 %), natural gas (21 %), cement (4 %), flaring (1 %), and others (1 %). Compared to 2023, the 2024 emissions from coal, oil, and gas increased by 0.8 %, 1.3 %, and 2.4 % respectively, while emissions from cement decreased by 4.6 %. All annual growth rates presented are adjusted for the leap year, unless stated otherwise.

In 2024, the largest absolute contributions to global fossil CO₂ emissions were from China (32 %), the USA (13 %), India (8 %), and the EU27 (6 %). These four regions account for 59 % of global fossil CO₂ emissions, while the rest of the world contributed 41 %, including international aviation and marine bunker fuels (3 % of the total). Growth rates for these countries from 2023 to 2024 were 0.7 % (China), -0.6 % (USA), -2.6 % (EU27), and 4.0 % (India), with +1.3 % for the rest of the world, including international aviation and marine bunker fuels (9.8 %). The per-capita fossil CO₂ emissions in 2024 were 1.3 tC person⁻¹ yr⁻¹ for the globe, and were 3.9 (USA), 2.4 (China), 1.5 (EU27) and 0.6 (India) tC person⁻¹ yr⁻¹ for the four highest emitters (Fig. 5c).

Globally, we estimate that global fossil CO₂ emissions (including cement carbonation, -0.21 GtC) grew by 1.0 % in 2025 (0.2 % to 1.7 %) to 10.4 GtC (38.1 GtCO₂), an historical record high

Growth rates in this section use a leap year adjustment that corrects for the extra day in 2024.

The uncertainty presented for projections relates only to the estimated growth rate, and excludes the uncertainty of the baseline (i.e., 2024) level. Given that our projection is effectively of what the data will show when observations are first available, the uncertainty drops to zero for any country where we already have data reported for all of 2025. This is the case below for both India and Japan, and for Carbon Monitor's estimates of growth in 2025, since they report data through December 2025.

For China, projected fossil emissions in 2025 are expected to have increased slightly by 0.4 % (range -0.1 % to 0.9 %) compared with 2024 emissions, bringing 2024 emissions for China around 3.4 GtC yr⁻¹ (12.3 GtCO₂ yr⁻¹). Our projected changes by fuel for China are 0.3 % for coal, 3.8 % for oil, 2.3 % for natural gas, and -7.3 % for cement.

For the USA, using the Energy Information Administration (EIA) emissions estimate for 2025 combined with cement clinker data from USGS, we estimate an increase of 2.5 % (range 2.2 % to 2.8 %) compared to 2024, bringing USA 2025 emissions to around 1.4 GtC yr⁻¹ (5.0 GtCO₂ yr⁻¹). Our estimated changes by fuel are 10.4 % for coal, 1.1 % for oil, 1.4 % for natural gas, and -3.1 % for cement.

For India, our estimate for 2025 is an increase of 1.1 % over 2024, with 2025 emissions around 0.9 GtC yr⁻¹ (3.2 GtCO₂ yr⁻¹). Our projected changes by fuel are 1.3 % for coal, -0.5 % for oil, -5.7 % for natural gas, and 10.5 % for cement.

For the European Union, our projection for 2025 is for a decrease of -0.1 % (range -2.6 % to 2.3 %) relative to 2024, with 2025 emissions around 0.7 GtC yr⁻¹ (2.4 GtCO₂ yr⁻¹). Our projected changes by fuel are -2.6 % for coal, -0.7 % for oil, 2.8 % for natural gas, and -0.8 % for cement.

New this year, we provide a projection for Japan, with a 2025 decrease of 0.9 %, with 2025 emissions around 0.3 GtC yr⁻¹ (0.9 GtCO₂ yr⁻¹). Our projected changes by fuel are -0.7 % for coal, -0.8 % for oil, -1.2 % for natural gas, and -2.9 % for cement.

International aviation and shipping are expected to have increased by 4.3 % in 2025, reaching 0.3 GtC yr⁻¹ (1.2 GtCO₂ yr⁻¹), with international aviation and international shipping projected to be respectively 6.7 % and 2.0 % over 2024. For the rest of the world, the expected change for 2025 is an increase of 0.9 % (range -1.1 % to 3.0 %) with 2025 emissions around 3.7 GtC yr⁻¹ (13.7 GtCO₂ yr⁻¹). The fuel-specific projected 2025 growth rates for the rest of the world are: 1.3 % for coal, 0.3 % for oil, 1.2 % for natural gas, 2.6 % for cement.

Compared to the GCB estimate, Carbon Monitor estimates an increase in global fossil CO₂ emissions of 0.7 % for 2025, lower than GCB but with overlapping uncertainties. In contrast to GCB, Carbon Monitor projects the 2025 emissions from China to decline by -0.6 %. Conversely, Carbon Monitor projects the 2025 emissions to increase by 1.4 % for the USA, by 0.7 % for India, by 0.8 % for the EU27, but to decrease by -0.3 % for Japan. The Carbon Monitor estimates that international aviation and shipping increased by 3.4 % in 2025. For the rest of the world, the estimated change is an increase of 1.3 %.

For traceability, Table S6 provides a comparison of annual projections from GCB since 2015 with the actual emissions assessed in the subsequent GCB annual report.

Cumulative net CO₂ emissions from land-use change (ELUC) for 1850–2024 are 250 ± 60 GtC (Table 8) with a large spread among individual bookkeeping estimates of 210 GtC (OSCAR), 240 GtC (LUCE) and 290 GtC (BLUE). DGVMs show a lower estimate of 160 ± 60 GtC. Vegetation biomass observations provide independent constraints on the ELUC estimates over the 1901–2012 period (Li et al., 2017). Over that period, the GCB bookkeeping models' cumulative ELUC amounts to 180 GtC [155 to 210 GtC] and the DGVM cumulative ELUC to 121 ± 45 GtC, both not significantly different from the observation-based estimate of 155 ± 50 GtC (Li et al., 2017). The substantially lower cumulative ELUC estimates from DGVMs compared to GCB2024 are due to the correction for the RSS bias, which is implemented in this assessment (see Sect. 2.6.1; O'Sullivan et al., 2025).

In contrast to growing fossil emissions, net CO₂ emissions from land use, land-use change, and forestry remained relatively constant (around 1.8 GtC yr⁻¹) over the 1959–1999 period. Since then, they have shown a statistically significant decrease of about 0.2 GtC per decade (p<0.001), reaching 1.4 ± 0.7 GtC yr⁻¹ for the 2015–2024 period (Table 7), with a spread from 1.3 to 1.6 GtC yr⁻¹ across the three bookkeeping estimates (Table 5, Fig. 4b). Different from the bookkeeping average, the DGVM average grows slightly larger over the 1980–2010 period, but as bookkeeping models, DGVMs show decreasing emissions in the most recent decade, 2015–2024 (Table 5).

Compared to GCB2024, the overall trends in ELUC remained similar but the ELUC estimates increased. The main reason for the larger estimates in this assessment is that the HandC2023 model, included up to GCB2024, is not included anymore (see Sect. 2.2). HandC2023 had the lowest ELUC estimates among all bookkeeping models (Fig. S3). Larger ELUC estimates in the last few decades (compared to GCB2024) are also due to the consideration of transient, instead of constant, carbon densities by all three bookkeeping models (Sect. 2.2.1). The inclusion of transient carbon densities increases gross fluxes for all ELUC components in the last few decades (Fig. S3). Net ELUC increases because emissions from deforestation are still the dominating term.

We separate ELUC into five component fluxes to gain further insight into the drivers of net emissions: deforestation, forest (re-)growth, wood harvest and other forest management, peat drainage and peat fires, and all other transitions, with CO₂ emissions to the atmosphere being positive and CO₂ removals from the atmosphere being negative (Fig. 7b; Supplement Sect. S2.2). We further decompose the deforestation and the forest (re-)growth term into contributions from shifting cultivation vs permanent forest cover changes (Fig. 7c). Averaged over the 2015–2024 period and over the three bookkeeping estimates, fluxes from deforestation amount to an emission of 1.9 [1.5 to 2.3] GtC yr⁻¹ (Table 5), of which 1.1 [0.9, 1.2] GtC yr⁻¹ are from permanent deforestation. Fluxes from forest (re-)growth amount to a removal of -1.3 [-1.5, -1.0] GtC yr⁻¹ (Table 5), of which -0.6 [-0.7, -0.5] GtC yr⁻¹ is from re/afforestation and the remainder from forest regrowth in shifting cultivation cycles. Wood harvest and other forest management causes net emissions of 0.4 [0.1, 0.7] GtC yr⁻¹ (with substantial gross fluxes largely compensating each other; see Fig. S2). Emissions from peat drainage and peat fires (0.2 [0.2, 0.3] GtC yr⁻¹) and the net flux from other transitions (0.1 [0.1, 0.1] GtC yr⁻¹) are of smaller magnitude globally (Table 5).

The split into component fluxes clarifies the potentials for emission reduction and carbon dioxide removal: emissions from permanent deforestation – the largest of our component fluxes – could be halted (largely) without compromising carbon uptake by forests, contributing substantially to emissions reduction. By contrast, reducing wood harvesting would have limited potential to reduce net emissions as it would be associated with less forest regrowth; removals and emissions cannot be decoupled here on long timescales. A similar conclusion applies to removals and emissions from shifting cultivation, which we have therefore separated out. Carbon Dioxide Removal (CDR) in forests could instead be increased by permanently increasing the forest cover through re/afforestation. Currently, re/afforestation creates a removal of -0.6 GtC yr⁻¹ from the atmosphere averaged over 2015–2024. This value is similar to independent estimates derived from NGHGIs for CDR in managed forests (through re/afforestation plus forest management) for 2013–2022 (-0.5 GtC yr⁻¹, Pongratz et al., 2024). In contrast to NGHGIs, bookkeeping estimates do not consider the impacts from natural disturbances (like fires, windthrow, insect outbreaks), which may lead to an overestimation of removals from re/afforestation. Re/afforestation constitutes most of all current CDR (Pongratz et al., 2024). Though they cannot be compared directly to annual fluxes from the atmosphere and are thus not included in our estimate of ELUC, CDR through transfers between non-atmospheric reservoirs such as in durable HWPs, biochar, or BECCS comprise much smaller amounts of carbon. 218 MtC yr⁻¹ have been estimated to be transferred to HWPs, averaged over 2013–2022 (Pongratz et al., 2024). The net flux of HWPs, considering the re-release of CO₂ through their decay, amounts to 91 MtC yr⁻¹ over that period (Pongratz et al., 2024). Note that some double accounting between the CDR through HWPs and the CDR through re/afforestation exists if the HWPs are derived from newly forested areas. CDR from Biochar and BECCS in 2024 amount to 0.31 and 0.18 MtC yr⁻¹ respectively. “Blue carbon”, i.e., coastal wetland management such as restoration of mangrove forests, saltmarshes, and seagrass meadows, though at the interface of land and ocean carbon fluxes, are counted towards the land-use sector as well. Currently, bookkeeping models do not include blue carbon; however, current CDR deployment in coastal wetlands is small globally, less than 0.003 MtC yr⁻¹ (Powis et al., 2023).

The statistically significant decrease in ELUC since the late-1990s, including the larger drop within the most recent decade, is due to the combination of decreasing emissions from deforestation (in particular permanent deforestation) and increasing removals from forest regrowth (with those from re/afforestation stagnating globally in the last decade). Net emissions in 2015–2024 are 23 % lower than in the late-1990s (1995–2004) and 19 % lower than in 2005–2014. The steep drop in ELUC after 2015 is due to the combined effect from a peak in peat fire emissions in 2015 and a long-term decline in deforestation emissions in many countries over the 2010–2020 period. The processes behind gross removals, foremost forest regrowth and soil recovery, are all slow, while gross emissions include a large instantaneous component. Short-term changes in gross emissions dynamics, such as a temporary decrease in deforestation, influences ELUC faster than a change in gross removals, which rather act on longer-term timescales. Component fluxes often differ more across the bookkeeping estimates than the net flux (Fig. 7b), which is expected due to different process representation; in particular, the treatment of shifting cultivation, which increases both gross emissions and gross removals, differs across models, but also net and gross wood harvest fluxes show high model spread (Fig. S2). By contrast, models agree relatively well for emissions from permanent deforestation.

Overall, highest net land-use emissions occur in the tropical regions of all three continents. The top three emitters over 2015–2024 are Brazil (in particular the Amazon Arc of Deforestation), Indonesia, and the Democratic Republic of the Congo, with these 3 countries contributing 0.8 GtC yr⁻¹ or 57 % of the global net land-use emissions (average over 2015–2024; Figs. 6b, 7a). This is related to massive expansion of cropland (FAO, 2025c), particularly in the last few decades in Latin America, Southeast Asia, and sub-Saharan Africa (Hong et al., 2021), to a substantial part for export of agricultural products (Pendrill et al., 2019). Emission intensity is high in many tropical countries, particularly of Southeast Asia, due to high rates of land conversion in regions of carbon-dense and often still pristine, undegraded natural forests (Hong et al., 2021). Emissions are further increased by peat fires in equatorial Asia (GFED4s, van der Werf et al., 2017), contributing substantially to emissions in individual years, typically related to dry conditions during El Niño (0.2 GtC in 2015; 0.03 GtC yr⁻¹ averaged over 2015–2024). Uptake due to land-use change occurs in several regions of the world (Fig. 6b) particularly due to re/afforestation. Highest nature-based CDR in the last decade is seen in China, the USA, and the EU27, partly related to expanding forest area as a consequence of the forest transition in the 19th and 20th century and subsequent regrowth of forest (Mather, 2001; McGrath et al., 2015). Substantial uptake through re/afforestation also exists in other regions such as Brazil, Russia, or Indonesia, where, however, emissions from deforestation and other land-use changes dominate the net land-use flux. While the mentioned patterns are robust and supported by independent literature, we acknowledge that model spread is substantially larger on regional than global levels, as has been shown for bookkeeping models (Bastos et al., 2021) as well as DGVMs (Obermeier et al., 2021). Independent assessments exist for selected regions (e.g., for Europe by Petrescu et al., 2020; for Brazil by Rosan et al., 2021; for China by Zhu et al., 2025; and for 8 selected countries/regions in comparison to inventory data by Schwingshackl et al., 2022).

The NGHGI data under the land-use, land-use change, and forestry (LULUCF) sector (Melo et al., 2025) and the LULUCF estimates from FAOSTAT (FAO, 2025d) differ from the global models' definition of ELUC (see Sect. 2.2.1). In the NGHGI reporting, natural land fluxes (SLAND) are counted towards ELUC when they occur on managed land (Grassi et al., 2018) whereas FAOSTAT LULUCF estimates generally include natural fluxes in managed and unmanaged forests (Tubiello et al., 2021). To compare our results to the NGHGI approach, we perform a translation of our ELUC estimates by adding SLAND in managed forest from the DGVMs simulations (1.9 GtC yr⁻¹ in 2015–2024) to the bookkeeping ELUC estimate (following the methodology described in Grassi et al., 2023; see Supplement Sect. S2.4). Adding this sink changes ELUC from being a source of 1.4 GtC yr⁻¹ to a sink of 0.5 GtC yr⁻¹, much closer to the NGHGI estimate reporting a sink of 1.0 GtC yr⁻¹ (Fig. 8a, Table S11). Remaining differences in the bookkeeping and NGHGI estimates are mainly stemming from fluxes due to other transitions and permanent deforestation, whereas peat emissions and the net flux in managed forests agree well (Fig. 8b, c). We further apply a mask of managed land to the net atmosphere-to-land flux estimate from atmospheric inversions to obtain inverse estimates that are comparable to the NGHGI estimates and to the translated ELUC estimates from bookkeeping models (see Supplement Sect. S2.4). The inversion-based net flux in managed land indicates a sink of 0.7 GtC yr⁻¹ for 2015–2024, which is in broad agreement with the NGHGI and the translated ELUC estimates (Fig. 8, Table S11). Additionally, the interannual variability of the inversion estimates and the translated ELUC estimates show a remarkable agreement (Pearson correlation of 0.71 in 2000–2024), which supports the suggested translation approach. Though estimates of NGHGI, FAOSTAT, and atmospheric inversions and the translated bookkeeping estimates still differ in value and need further analysis, the approach suggested by Grassi et al. (2023), which we adopt here, provides a feasible way to relate the global models' and NGHGI approach to each other and thus link the anthropogenic carbon budget estimates of land-use CO₂ fluxes directly to the Global Stocktake, as part of the UNFCCC Paris Agreement. The translation approach has been shown to be generally applicable also at the country-level (Grassi et al., 2023; Schwingshackl et al., 2022).

The global CO₂ emissions from land-use change are estimated as 1.3 ± 0.7 GtC in 2024, similar to the 2023 estimate. However, confidence in the annual change remains low, as the land-use forcing underlying the ELUC estimates was informed directly by data only up to and including 2023, with a trend extrapolation to 2024 (see Supplement Sect. S2.1). The impacts of the 2023/2024 droughts linked to El Niño on the 2024 ELUC estimate are thus not captured – while natural fires are not counted towards ELUC, deforestation fires spreading further or drained peatlands burning more vastly because of dry conditions would be part of ELUC. The extent of degradation was anomalously high in the Amazon in 2024 and partly masked relatively low deforestation rates (https://terrabrasilis.dpi.inpe.br, last access: 23 October 2025). However, this is not captured by our approach that is based on forest cover changes. For degradation it is currently impossible to be separated into effects of land-use activity versus effects of climate variability.

In equatorial Asia, peat fire emissions remained low (1 MtC in 2025 through 15 October 2025, after 2 MtC in 2024; GFED4.1s, van der Werf et al., 2017), as did deforestation and degradation fires (4 MtC in 2025; 8 MtC in 2024). South America saw a big reduction in emissions from deforestation and degradation fires, from 334 MtC in 2024 to 35 MtC until 15 October 2025. Pantropical 2025 fire emission estimates from deforestation and degradation through 15 October are 105 MtC, which is not just a massive drop compared to the anomalously high 2024 emissions, but also only about one third of the long-term average (1997–2024). The reduction in emissions is mainly attributable to anomalously low deforestation and degradation fire emissions in South America, likely connected to the ceasing of the El Niño conditions and their impacts on ecosystems.

Since the ELUC estimate for the year 2024 is not informed directly by data on land-use dynamics (Sect. 3.2.3), we estimate the 2025 projection based on fire anomalies over the year 2023. We expect ELUC emissions of around 1.1 GtC (4.1 GtCO₂) in 2025, 0.1 GtC below the 2024 level (0.3 GtC below the 2015–2024 average). Note that the confidence in the 2025 ELUC projection remains low, as it is based on deforestation, degradation and peat fire emissions, which are only a proxy for land-use change. Projections in past GCB assessments are only partially confirmed by updated ELUC estimates based on bookkeeping models in the following GCB assessments (Fig. S4). Although our extrapolation includes tropical deforestation and degradation fires, the degradation attributable to selective logging, edge-effects or fragmentation is not captured. Further, deforestation and fires in deforestation zones may become more disconnected, partly due to changes in legislation in some regions. For example, van Wees et al. (2021) found that the contribution from fires to forest loss decreased in the Amazon and in Indonesia over the period of 2003–2018.

Atmospheric CO₂ concentration was approximately 278 parts per million (ppm) in 1750, reaching 300 ppm in the late 1900s, 350 ppm in the late 1980s, and reaching 422.80 ± 0.1 ppm in 2024 (Lan et al., 2025; Fig. 1). The mass of carbon in the atmosphere increased by 52 % from 590 GtC in 1750 to 898 GtC in 2024. Current CO₂ concentrations in the atmosphere are unprecedented in the last 2 million years and the current rate of atmospheric CO₂ increase is at least 10 times faster than at any other time during the last 800 000 years (Canadell et al., 2021).

The growth rate in atmospheric CO₂ level increased from 1.7 ± 0.07 GtC yr⁻¹ in the 1960s to 5.6 ± 0.02 GtC yr⁻¹ during 2015–2024 with important decadal variations (Table 7, Figs. 3 and 4c). During the last decade (2015–2024), the growth rate in atmospheric CO₂ concentration continued to increase, albeit with large interannual variability (Fig. 9). This interannual variability is highly consistent between the surface-based CO₂ growth rate and GRESO, the XCO₂-based growth rate (Pearson correlation of 0.84), but substantial differences are seen in 2019 and 2022–2024. These reflect differences in the total atmospheric carbon stock seen within the set boundaries of 1 January and 31 December used to create an annual growth rate, as discussed in Sect. 2.4.2. Especially with large anomalies in tropical regions and close to 1 January, the calculation becomes sensitive to when the change is detected. This affected the 2023 and 2024 growth rates most strongly as they originated from tropical regions, and late in the calendar year (see next section). Figure 9 also shows the total annual fluxes (in GtC yr⁻¹) from the inverse models, which are derived to match the observed atmospheric increase while numerically accounting for the atmospheric transport-related delay in detection. The inversely modeled values generally match the GRESO growth rate better than the surface observations-based growth rate, indicating that the latter is more strongly influenced by the delay between the fluxes occurring and the actual observation of these signals. Note that this effect is mainly important on the annual timescale, and not over longer periods.

The growth rate in the surface based atmospheric CO₂ concentration was 7.9 ± 0.2 GtC (3.73 ± 0.08 ppm) in 2024 (Fig. 4c; Lan et al., 2025), well above the 2023 growth rate (5.7 ± 0.2 GtC, 2.7 ± 0.08 ppm) or the 2015–2024 average (5.6 ± 0.02 GtC, 2.6 ± 0.008 ppm), as to be expected during an El Niño year. The 2024 atmospheric CO₂ growth rate was the largest over the 1959–2024 atmospheric observational record, more than 1.5 GtC above the growth rates of 1998, 2015, and 2016 and 1998, all strong El Niño years.

In contrast, the satellite-based GRESO growth rate was 6.8 ± 0.2 GtC yr⁻¹ (3.20 ± 0.09 ppm yr⁻¹) for 2024 which was also above its 2023 growth rate (6.5 ± 0.2 GtC yr⁻¹, 3.06 ± 0.07 ppm yr⁻¹) and the highest on its 10-year record. But it shows a more equal split of the anomaly between both years (2023–2024). This likely reflects more accurately that a substantial fraction of the flux anomaly occurred in 2023, but this was only detected by the surface network in 2024. For the interpretation of the annual global budget this has strong implications, as 2024 surface fluxes would need to add up to a smaller anomaly (by 1.1 GtC yr⁻¹) than when using the surface-based annual growth rate. Hence the budget imbalance in 2024 (Table 7) would be reduced from -1.7 to -0.6 GtC if the GRESO growth rate was used for the budget estimate.

As the satellite-based GRESO growth rate is more evenly split between 2023 and 2024, so would be the BIM, -0.3 GtC in 2023 and -0.6 GtC in 2024 with GRESO, compared to +0.4 GtC in 2023 and -1.7 GtC in 2024 with the GATM from the surface based network.

The 2025 atmospheric CO₂ concentration, averaged over the year reached the level of 425.64 ppm, 53 % over the pre-industrial level (Lan et al., 2025). We estimate the annual growth in atmospheric CO₂ (GATM) to be about 4.4 GtC (equivalent to a 2.1 ppm increase in the global mean concentration), in line with a neutral ENSO year (ENSO 3.4 Index between -0.5 and 0.5). The projected growth rate estimated by the ESMs multi-model mean is slightly larger (5.5 GtC, 2.6 ppm).

For traceability, Table S7 provides a comparison of annual projections of the GATM, SOCEAN and SLAND from GCB since 2021 with the actual estimate assessed in the subsequent GCB annual report.

Cumulated since 1850, the ocean sink SOCEAN adds up to 200 ± 35 GtC, with more than 70 % of this amount (145 ± 25 GtC) being taken up by the global ocean since 1959. Over the historical period, the ocean sink increased in pace with the anthropogenic emissions exponential increase (Fig. 3). Since 1850, the ocean has removed 27 % of total anthropogenic emissions.

SOCEAN increased from 1.3 ± 0.4 GtC yr⁻¹ in the 1960s to 3.2 ± 0.4 GtC yr⁻¹ during 2015–2024 (Table 7), with interannual variations of the order of a few tenths of GtC yr⁻¹ (Figs. 4d, 10b). As described in Sect. 2.5.1, SOCEAN is now corrected for identified biases in the GOBMs and fCO₂ products estimates. Compared to the uncorrected estimate, SOCEAN is increased by 0.2 GtC yr⁻¹ for the 2015–2024 period.

The ocean-borne fraction (SOCEAN/(EFOS+ELUC)) has been remarkably constant around 27 % on average, with variations around this mean illustrating the decadal variability of the ocean carbon sink. So far, there is no evidence of a sustained decrease in the ocean-borne fraction from 1959 to 2024 (Fig. S16).

The increase of the ocean sink is primarily driven by the increased atmospheric CO₂ concentration, with the strongest CO₂ induced signal in the North Atlantic and the Southern Ocean (Figs. 12a, S10). The effect of climate change is much weaker, reducing the ocean sink globally by 0.20 ± 0.05 GtC yr⁻¹ (-7.1 % relative to the simulation that only accounts for the effect of atmospheric CO₂ increase) during 2015–2024 (all models simulate a weakening of the ocean sink by climate change, range -4.2 % to -10.7 %). The climate change effect leading to a reduced ocean sink is evident in all large-scale latitudinal bands (north, tropics, south, Figs. 12b, S10). This is the combined effect of change and variability in all atmospheric forcing fields, previously attributed to wind and temperature changes (Le Quéré et al., 2007; Bunsen et al., 2024). The effect of warming is smaller than expected from offline calculation due to stabilising feedback from limited exchange between surface and deep waters (Bunsen et al., 2024; Müller et al., 2025).

The global net air-sea CO₂ flux is a residual of large natural and anthropogenic CO₂ fluxes into and out of the ocean with distinct regional and seasonal variations (Figs. 6c and S5). Natural fluxes dominate on regional scales, but largely cancel out when integrated globally (Gruber et al., 2019; DeVries et al., 2023). Mid-latitudes in all basins and the high-latitude North Atlantic dominate the ocean CO₂ uptake where low temperatures and high wind speeds facilitate CO₂ uptake at the surface (Takahashi et al., 2009). In these regions, formation of mode, intermediate and deep-water masses transport anthropogenic carbon into the ocean interior, thus allowing for continued CO₂ uptake at the surface. Outgassing of natural CO₂ occurs mostly in the tropics, especially in the equatorial upwelling region, and to a lesser extent in the North Pacific and polar Southern Ocean, mirroring a well-established understanding of regional patterns of air-sea CO₂ exchange (e.g., Takahashi et al., 2009; Gruber et al., 2019; DeVries et al., 2023). These patterns are also noticeable in the Surface Ocean CO₂ Atlas dataset, where an ocean fCO₂ value above the atmospheric level indicates outgassing (Fig. S5). This map further illustrates the data-sparsity in the Indian Ocean and the Southern Hemisphere in general.

The largest variability in the ocean sink occurs on decadal timescales (Fig. 10b). The ensemble means of GOBMs and fCO₂-products show the same patterns of decadal variability, although with a larger amplitude of variability in the fCO₂-products than in the GOBMs. The ocean sink stagnated in the 1990s and strengthened between the early 2000s and the mid-2010s (Fig. 10b; Le Quéré et al., 2007; Landschützer et al., 2015, 2016; DeVries et al., 2017; Hauck et al., 2020; McKinley et al., 2020, Gruber et al., 2023). Different explanations have been proposed for the decadal variability in the 1990s and 2000s, ranging from the ocean's response to changes in atmospheric wind systems (e.g., Le Quéré et al., 2007; Keppler and Landschützer, 2019), including variations in upper ocean overturning circulation (DeVries et al., 2017) to the eruption of Mount Pinatubo in the 1990s (McKinley et al., 2020, Fay et al., 2023). The main origin of the decadal variability is a matter of debate with several studies initially pointing to the Southern Ocean (see review in Canadell et al., 2021 and Gruber et al., 2023), but also contributions from the North Atlantic and North Pacific (Landschützer et al., 2016; DeVries et al., 2019), or a global signal (McKinley et al., 2020) were proposed. The GOBM decomposition into climate and CO₂ effects emphasizes the role of the climate effect (that would include cooling signals from volcanic eruptions) for the stagnation in the 1990s until 2002, which are seemingly more pronounced in the tropics and north than in the south (Fig. S10). The atmospheric pCO₂ growth rate is a spatially uniform driver that amplifies with large-scale averaging, while climate exerts spatially heterogeneous effects that cancel with averaging. Thus, climate dominates small-scale flux variability, but is matched on the global mean by the impact of the pCO₂ growth rate (Fay et al., 2024).

More recently, the sink seems to have entered a phase of stagnation 2016–2022, largely in response to large inter-annual climate variability. The first-order effect of interannual variability stems from a stronger ocean sink during large El Niño events leading to a reduction in CO₂ outgassing from the Tropical Pacific (e.g., 1997/98, 2015/16) and a weaker sink in neutral years following El Niño (2017) and during La Niña events (2020–2022, Fig. 10b; Rödenbeck et al., 2014; Hauck et al., 2020; McKinley et al., 2017). After the triple La Niña event 2020–2022, the ocean sink rebound in 2023 linked to the onset of an El Niño event. Warming in the extratropics, in particular in the Northern Hemisphere, however, weakened the overall increase in the ocean carbon sink (Müller et al., 2025). In the GOBMs, the ocean sink stagnation 2016–2022 is associated with a period of a stronger climate effect reducing the ocean sink, mostly stemming from the tropics, in line with the expected patterns from El Niño and La Niña (Fig. S10). The CO₂ effect has also shown a lower growth rate since around 2016.

The ensemble means of GOBMs, and fCO₂-products (adjusted for the riverine flux) show a mean offset increasing from 0.28 GtC yr⁻¹ in the 1990s (first decade available) to 0.59 GtC yr⁻¹ in the decade 2015–2024. The offset is larger than in previous GCB versions, because two fCO₂-products (UExP-FNN-U, JMA-MLR), that use an adjusted version of the Surface Ocean CO₂ Atlas data to represent the sea surface fCO₂ within the surface skin layer where gas exchange takes place (Ford et al., 2025), are now included in the fCO₂-product mean. Additionally, an update in the SOCAT dataset with regards to Southern Ocean data led to an increase in the sink strength in three of the fCO₂-products (VLIZ-SOMFFN, LDEO-HPD, CMEMS-LSCE-FFNN – see Sect. S3.1, Fay et al., 2025). In this version of the GCB, the positive trends in the ocean sink estimated by the GOBMs and the fCO₂-products diverges over time by a factor of 1.6 since 2002 (GOBMs: 0.27 ± 0.05 GtC yr⁻¹ per decade, fCO₂-products: 0.46 GtC yr⁻¹ per decade [0.22 to 0.81 GtC yr⁻¹ per decade], corrected SOCEAN: 0.38 GtC yr⁻¹ per decade), but the uncertainty ranges overlap. A hybrid approach recently constrained the trend for 2000 to 2022 to 0.42 ± 0.06 GtC yr⁻¹ per decade (Mayot et al., 2024), which aligns with the updated trend of SOCEAN (0.46 GtC yr⁻¹ per decade), while the fCO₂-products exhibit a larger (0.54 [0.29, 0.85] GtC yr⁻¹ per decade) and the GOBMs a lower trend (0.34 ± 0.05 GtC yr⁻¹ per decade) over the same period.

In the current budget, the discrepancy between the two types of estimates stems from a persistently larger ocean sink in the fCO₂-products in the northern and southern extra-tropics since around 2002 (Fig. 14). Note that the discrepancy in the mean flux, which was located in the Southern Ocean in GCB2022 and earlier, was reduced due to the choice of the regional river flux adjustment (Lacroix et al., 2020 instead of Aumont et al., 2001). This comes at the expense of a discrepancy in the mean SOCEAN of about 0.2–0.3 GtC yr⁻¹ in the tropics. Likely explanations for the discrepancy in the trends and decadal variability in the high latitudes are data sparsity and uneven data distribution (Bushinsky et al., 2019; Gloege et al., 2021; Hauck et al., 2023a; Mayot et al., 2024). In particular, two fCO₂-products were shown to overestimate the Southern Ocean CO₂ flux trend by 50 % and 130 % based on current sampling in a model subsampling experiment (Hauck et al., 2023a) and the largest trends in the fCO₂-products occurred in a data devoid region in the North Pacific (Mayot et al., 2024). In Supplement Sect. S3 we show that the strength of the trends in fCO₂ products may be linked to reconstruction biases of the true trend signal. In this respect it is highly worrisome that the coverage of fCO₂ observations has declined since 2016/17 (Dong et al., 2022, 2024b; Bakker et al., 2025b) and is now down to that of the mid-2000s (Fig. 10b). The temperature correction applied in two products further worsens the agreement in trends between GOBMs and fCO₂-products. Similarly, model biases likely contribute to the discrepancy between GOBMs and fCO₂-products (as indicated by the comparison with Mayot et al. (2024), and by the large model spread in the South, Fig. 14).

The SOCEAN estimate is 2.9 ± 0.4 GtC yr⁻¹ over the period 2004 to 2019, which agrees within the ranges of uncertainty with the ocean interior estimate of 3.2 ± 0.7 GtC yr⁻¹ obtained with the MOBO-DIC approach (Keppler et al., 2023). The carbon flux components of the observation-based ocean interior estimate match the definition of SOCEAN used here (Hauck et al., 2020). Furthermore, the decadal SOCEAN estimates agree well with the corresponding ocean interior estimates for all decades from the 1960s to the 2010s, which represent a composite estimate of three observation-based products (Table 6).

The SOCEAN estimate agrees within uncertainties with other lines of evidence from atmospheric oxygen-based estimates, atmospheric inversions, and ocean interior observation-based constraints (Table 6). The atmospheric oxygen-based estimate shows a lower estimate than SOCEAN in the 1990s, but a larger estimate over the 2015–2024 period, indicating a strong growth of > 1.5 GtC yr⁻¹ over that period. ESMs with data assimilation result in lower estimates of the ocean sink than all other estimates (Table 6). Also, the atmospheric inversion estimates of the ocean sink are generally lower than SOCEAN on average but are within the uncertainties of the GOBMs and fCO₂ products.

The estimated ocean CO₂ sink is 3.4 ± 0.4 GtC for 2024. This is an increase of 0.1 GtC compared to 2023 (Fig. 4d). The sink strengthening was expected from the atmospheric CO₂ growth and El Niño conditions in the beginning of the year (January to April). However, the continuation of the anomalous warm conditions in the extra tropics, especially of the Atlantic Ocean, led to a weaker than expected increase in the sink (Fig. 11c). The GOBMs suggest that the increase is driven by the anomalously high atmospheric CO₂ growth rate and dampened by climate effects, thus suggesting that warming overcompensated the effect from El Niño (Fig. S10). GOBMs and fCO₂-products largely agree on patterns of ocean sink anomalies with a reduced sink in parts of the subtropical and subpolar North Atlantic and North Pacific that coincides with regions of anomalously warm sea surface temperatures (Fig. S9). GOBM and fCO₂-product ensemble mean estimates consistently result in an ocean sink increase in 2024 (GOBMs: 0.13 ± 0.14 GtC, fCO₂-products: 0.08 [-0.34,0.43] GtC). Eight GOBMs and six fCO₂-products show an increase in SOCEAN, while only two GOBMs and three fCO₂-products show a decrease in SOCEAN. The fCO₂-products have a larger uncertainty at the end of the reconstructed time series, potentially linked to uncertainties related to fewer available observations in the final year (see e.g. Watson et al., 2020; Pérez et al., 2024). Specifically, the fCO₂-products' estimate of the last year is regularly adjusted in the following release owing to the tail effect and an incrementally increasing data availability. While the monthly grid cells covered may have a lag of only about a year (Fig. 10b inset), the values within grid cells may change with 1–5 years lag (see absolute number of observations plotted in previous GCB releases), potentially resulting in annual changes in the flux magnitude from fCO₂-products.

Using a feed-forward neural network method (see Sect. 2.5.2) we project an ocean sink of 3.3 ± 0.4 GtC for 2025, which is a declining sink of -0.1 GtC compared to 2024 and can be explained by the generally lower ocean CO₂ uptake following the El Niño event ending in 2024, consistent with the projected recovery of the atmospheric CO₂ growth rate after a record high in 2024. The set of ESMs predictions support this estimate with a 2025 ocean sink of around 3.1 [3.0, 3.2] GtC. Taking the average of both estimates, we project an ocean sink of 3.2 ± 0.4 GtC for 2025.

The process-based model evaluation draws a generally positive picture with GOBMs scattered around the observational values for Southern Ocean sea-surface salinity, Southern Ocean stratification index, albeit outliers exist (Sect. S3.3 and Table S12). The Revelle factor is high in most GOBMs when compared to GLODAP but appears less biased when compared to OceanSODA. However, the Atlantic Meridional Overturning Circulation at 26° N is underestimated by 8 out of 10 GOBMs and overestimated by one GOBM. IOMB summarizes the GOBMs' performance across physical and biogeochemical data sets and various statistics relative to the GOBM ensemble (Fig. S6). All GOBMs perform better in some variables compared to the other models, and worse in other variables. Despite the indication that GOBMs underestimate the ocean sink, low sink models do not perform generally worse than the other models; and similarly, high sink models do not perform better than the others.

The model simulations allow to separate the anthropogenic carbon component and to compare the GOBMs DIC inventory change directly to the interior ocean estimates of Gruber et al. (2019), Müller et al. (2023) and the GOBM anthropogenic surface fluxes to DeVries (2022) without further assumptions (Table S12). The GOBMs ensemble average of anthropogenic carbon inventory changes 1994–2007 amounts to 2.3 GtC yr⁻¹ and is thus 10.5 % lower than the 2.6 ± 0.3 GtC yr⁻¹ estimated by Gruber et al. (2019) although within the uncertainty. Five models fall within the range reported by Gruber et al. (2019). Comparison to the decadal estimates of anthropogenic carbon accumulation (Müller et al., 2023) are close to the interior ocean data based estimate for the decade 2004–2014 (GOBMs sim D minus sim A, 2.6 ± 0.4 GtC yr⁻¹, Müller et al. (2023) 2.7 ± 0.3 GtC yr⁻¹), but do not reproduce the supposedly higher anthropogenic carbon accumulation in the earlier period 1994–2004 (GOBMs sim D minus sim A, 2.2 ± 0.3 GtC yr⁻¹, Müller et al. (2023) 2.9 ± 0.3 GtC yr⁻¹). Finally, the mean anthropogenic carbon uptake in GOBMs from 1985 to 2018 amounts to 2.3 GtC yr⁻¹, slightly lower than the corresponding observation-based uptake rate of 2.4 ± 0.2 Gt yr⁻¹ based on OCIM (DeVries, 2022; DeVries et al., 2023). The underestimation of anthropogenic carbon accumulation by 10 % in the period 1994 to 2007, in the 1990s and 2000s (Table 6) and by 15 % in the 2010s (Table 6) justifies the correction of GOBMs by 10 % (Friedlingstein et al., 2025a, Sect. 2.5.1). Interestingly, and in contrast to the uncertainties in the surface CO₂ flux, we find the largest mismatch in interior ocean carbon accumulation in the tropics, with smaller contributions from the north and the south (Table S12). The large discrepancy in accumulation in the tropics highlights the role of interior ocean carbon redistribution for those inventories (Khatiwala et al., 2009; DeVries et al., 2023).

Additional benchmarking of the fCO₂-products with independent data generally shows low and consistent biases and RMSEs, with the exception of the comparison with SOCAT flag E data in the tropics where the biases range from -6.6 µatm in JMA-MLR to -10.29 µatm in OceanSODA_ETHv2, (Fig. S7) and the Northern Hemisphere, where biases range from -2.68 µatm in VLIZ-SOMFFN to -40.47 µatm in UExP-FNN-U (Fig. S7), although with few measurements covering the latter area. Furthermore, we evaluate the trends derived from a subset of fCO₂-products by subsampling five GOBMs used in Friedlingstein et al. (2023; covering the period up to the year 2022) following the approach of Hauck et al. (2023a) and evaluating the air-sea CO₂ flux trend for the 2001–2021 period, i.e. the period of strong divergence in the air-sea CO₂ exchange excluding the tail effect, against trend biases identified by the GOBM reconstruction. The results indicate a relationship between reconstruction bias and strength of the decadal trends (Fig. S8), indicating a tendency of the fCO₂-products ensemble to overestimate the air-sea CO₂ flux trends in agreement with Mayot et al. (2024). This relationship, however, remains uncertain and its sensitivity to the GOBM used needs further investigation.

Cumulated since 1850, the terrestrial carbon sink SLAND amounts to 175 ± 50 GtC, 24 % of total anthropogenic emissions, with more than two thirds of this amount (120 ± 40 GtC) being taken up by the terrestrial ecosystems since 1959. Over the historical period, the land sink increased in pace with the anthropogenic emissions exponential increase (Fig. 3). As described in Sect. 2, SLAND estimate now includes the RSS correction.

SLAND increased from 0.9 ± 0.3 GtC yr⁻¹ in the 1960s to 2.4 ± 0.8 GtC yr⁻¹ during 2015–2024, with important interannual variations of up to 2 GtC yr⁻¹ generally showing a decreased land sink during El Niño events (Fig. 10a), responsible for the corresponding enhanced growth rate in atmospheric CO₂ concentration. The larger land CO₂ sink during 2015–2024 compared to the 1960s is reproduced by all the DGVMs in response to the increase in both atmospheric CO₂, nitrogen deposition, and the changes in climate, and is broadly consistent with the residual estimated from the other budget terms, that is EFOS+ELUC-GATM-SOCEAN which amounts to 2.4 GtC for the last decade (See Table 5). As described in Sect. 2.6.1, SLAND is now corrected for the Replaced Sinks and Sources (RSS) bias. Compared to the uncorrected estimate, SLAND is reduced by 0.6 GtC yr⁻¹ for the 2015–2024 period.

Over the historical period, the increase in the global terrestrial CO₂ sink is largely attributed to the CO₂ fertilisation effect (Prentice et al., 2001; Piao et al., 2009; Schimel et al., 2015) and increased nitrogen deposition (Huntzinger et al., 2017; O'Sullivan et al., 2019), directly stimulating plant photosynthesis and increased plant water use in water limited systems, with a smaller negative contribution of climate change (Fig. 12). There is a range of evidence to support a positive terrestrial carbon sink in response to increasing atmospheric CO₂ (Walker et al., 2021), including a new synthesis of free-air CO₂-enrichment experiments across forests and ages, which concluded that Net Primary Productivity increased by 22 % for a common 41 % CO₂ enrichment (Norby, 2025). As expected from theory, the greatest CO₂ effect is simulated in the tropical forest regions, associated with warm temperatures and long growing seasons (Hickler et al., 2008) (Fig. 12a). However, evidence from tropical intact forest plots indicate an overall decline in the land sink across Amazonia (1985–2011), attributed to enhanced mortality offsetting productivity gains (Brienen et al., 2015; Hubau et al., 2020). During 2015–2024 the land sink is positive in all regions (Fig. 6d) with the exception of eastern Brazil, Bolivia, northern Venezuela, Southwest USA, central Europe and Central Asia, North and South Africa, and eastern Australia, where the negative effects of climate variability and change (i.e. reduced rainfall and/or increased temperature) counterbalance CO₂ effects. This is clearly visible in Fig. 12 where the effects of CO₂ (Fig. 12a) and climate (Fig. 12b) as simulated by the DGVMs are isolated (see also Fig. S12). The negative effect of climate can be seen across the globe, and is particularly strong in most of South America, Central America, Southwest US, Central Europe, western Sahel, southern Africa, Southeast Asia and southern China, and eastern Australia (Figs. 12b, S12). Globally, over the 2015–2024 period, climate change reduces the land sink by 0.8 ± 0.7 GtC yr⁻¹ (25 % of the CO₂ effect, which is 3.2 ± 1.1 GtC yr⁻¹ for the corresponding period, see Supplement Sect. S4.1).

Most DGVMs have similar SLAND averaged over 2015–2024, and 14/22 models fall within the 1σ range of the residual land sink [1.4 to 3.3 GtC yr⁻¹] (see Table 5), and all models but one are within the 2σ range [0.5 to 4.3 GtC yr⁻¹]. The ED model is an outlier, with a land sink estimate of 4.9 GtC yr⁻¹ for the 2015–2024 period (accounting for the RSS correction), driven by a strong CO₂ fertilisation effect (6.4 GtC yr⁻¹ in the CO₂ only (S1) simulation). There are no direct global observations of the land sink (SLAND), or the CO₂ fertilisation effect, and so we are not yet able to rule out models based on component fluxes if their net land sink (SLAND-ELUC) is within the observational uncertainty provided by O₂ measurements. The important role of non-living carbon pools in carbon budgets has also been recently highlighted from an Earth-observation (EO) perspective, where contrary to DGVMs, EO products do not show an increase in biomass, thus inferring a larger role of dead carbon in land carbon cycle dynamics (Bar-On et al., 2025).

Since 2020 the globe has experienced La Niña conditions which would be expected to lead to an increased land carbon sink. This 3-year long period of La Niña conditions came to an end by the second half of 2023 and transitioned to an El Niño which lasted until mid-2024. A clear transition from maximum to a minimum in the global land sink is evident in SLAND, from 2022 to 2023 and we find that an El Niño-driven decrease in tropical land sink is offset by a smaller increase in the high-latitude land sink. In the past years several regions experienced record-setting fire events (see also Sect. 3.8.3). While global burned area has declined over the past decades mostly due to declining fire activity in savannas (Andela et al., 2017), forest fire emissions are rising and have the potential to counter the negative fire trend in savannas (Zheng et al., 2021). Noteworthy extreme fire events include the 2019–2020 Black Summer event in Australia (emissions of roughly 0.2 GtC; van der Velde et al., 2021), Siberia in 2021, where emissions approached 0.4 GtC or three times the 1997–2020 average according to GFED4s, Canada in 2023 and 2024 (Byrne et al., 2024), and unprecedented wildfires in Brazil and Bolivia in 2024 (partly related to land-use activity, see Sect. 3.2.3 and 3.2.4) (Bourgoin et al., 2025), partially offset by a negative trend in wildfire over Northern Hemisphere Africa. While other regions, including Western US and Mediterranean Europe, also experienced intense fire seasons in 2021 their emissions are substantially lower.

Despite these regional negative effects of climate change on SLAND, the efficiency of land to remove anthropogenic CO₂ emissions has remained broadly constant over the last six decades at around 23 % (including the RSS correction) (Fig. S16).

The terrestrial CO₂ sink from the DGVMs ensemble SLAND was 1.9 ± 0.9 GtC in 2024, 37 % below the 2022 La Niña induced strong sink of 3.1 ± 1.0 GtC, and below the 2015–2024 average of 2.4 ± 0.8 GtC yr⁻¹ (Fig. 4e, Table 7). We estimate that the 2024 land sink was the lowest since 2015. The severe reduction in the land sink in 2024 is likely driven by the El Niño conditions, leading to a 66 % reduction in SLAND in the tropics (30° N–30° S) from 2.8 GtC in 2022 to 1.0 GtC in 2024. This is combined with intense wildfires in Canada, Bolivia and Brazil that led to a significant CO₂ source (see also Sect. 3.8.3). We note that the SLAND estimate for 2024 of 1.9 ± 0.9 GtC is much larger than the 0.3 ± 1.0 GtC yr⁻¹ estimate from the residual sink from the global budget (EFOS+ELUC-GATM-SOCEAN, Table 5), although the residual sink would be substantially larger (at around 1.4 GtC yr⁻¹) if using the satellite-based GRESO in this equation. A large budget imbalance is often associated with El Niño years (e.g. 1986/87, 1997/98, 2005/06, 2023/24), and the possible underestimation in DGVMs of the tropical land carbon losses in response to drought, high temperature extremes, and fires. Few DGVMs have explicit representation of drought-mortality and only 8 from 22 include some form of forest demography, although recent efforts are underway to fill these critical research gaps (Eckes-Shephard et al., 2025; Yao et al., 2022, 2023).

An overestimate in SLAND in 2024 can in part be attributed to the inability in DGVMs to reproduce extreme fire emissions in 2024. Globally fire emissions as calculated by GFED were 0.43 GtC yr⁻¹ higher in 2024 compared to the decadal average (2015–2024), similar to the anomaly in 2023 (0.44 GtC yr⁻¹). This was mainly due to an increase in fire over forests in Boreal North America (0.16 GtC yr⁻¹) and forests, savannahs and grasslands over Southern Hemisphere South America (0.35 GtC yr⁻¹) in 2024, partly offset by a negative trend in fires over Northern Hemisphere Africa (-0.05 GtC yr⁻¹ anomaly). In fact, across the Amazon basin emissions from degradation fires (two thirds Brazil, one third Bolivia) surpassed those from deforestation fires in 2024 (Bourgoin et al., 2025). In contrast DGVMs simulate an anomaly of 0.2 GtC yr⁻¹ in 2024, i.e., an underestimate of 0.23 GtC yr⁻¹ in fire emissions for 2024. Fire enabled DGVMs simulate a larger reduction in SLAND in 2024 compared to non-fire models (0.6 vs 0.3 GtC), indicating the importance of representing fire extremes when explaining reductions in the land sink.

Finally, a large uncertainty relates to climate forcing datasets, particularly over the critical carbon-rich tropical forest regions with poor coverage of meteorological stations. For example, there are large differences in climate forcing (e.g., CRUJRA-3Q and ERA5) and downstream DGVM carbon simulations over the Congo basin and tropical central and northern Africa in 2024 (Ke et al., 2024).

Calculating the land sink as the residual of the other projections for 2025, we project a land sink of 3.1 GtC for 2025, 1.1 GtC larger than the 2024 estimate, consistent with an expected recovery of the land sink after an El Niño event. The ESMs do not provide an additional estimate of SLAND as they only simulate the net atmosphere-land carbon flux (SLAND-ELUC).

The evaluation of the DGVMs shows generally higher agreement across models for runoff, and to a lesser extent for GPP, and ecosystem respiration. These conclusions are supported by a more comprehensive analysis of DGVM performance in comparison with benchmark data (Sitch et al., 2024). A relative comparison of DGVM performance (Fig. S11) suggests several DGVMs (CABLE-POP, CLASSIC, OCN, ORCHIDEE) may outperform others at multiple carbon and water cycle benchmarks. However, results from Seiler et al. (2022) also show how DGVM differences are often of similar magnitude compared with the range across observational datasets. All models score high enough over the metrics tests to support their use here. There are a few anomalously low scores for individual metrics from a single model, and these can direct the effort to improve models for use in future budgets (See also Supplement Sect. S4.2).

In the period 2015–2024, the bottom-up view of global net ocean and land carbon sinks provided by the GCB, SOCEAN for the ocean and SLAND– ELUC for the land, agrees closely with the top-down global carbon sinks delivered by the atmospheric inversions. This is shown in Fig. 13, which visualises the individual decadal mean atmosphere-land and atmosphere-ocean fluxes from each, along with the constraints on their sum offered by the global fossil CO₂ emissions flux minus the atmospheric growth rate (EFOS-GATM, 4.2 ± 0.5 GtC yr⁻¹, Table 7, shown as diagonal line in Fig. 13). The GCB estimate for net atmosphere-to-surface flux (SOCEAN+SLAND-ELUC) during 2015–2024 is 4.2 ± 1.1 GtC yr⁻¹ (Table 7), implying a zero budget imbalance (BIM) (see Sect. 3.9). The atmospheric inversions estimate of the net atmosphere-to-surface flux during 2015–2024 is 4.3 GtC yr⁻¹, with a < 0.1 GtC yr⁻¹ imbalance, and thus scatter across the diagonal, with inverse models trading land for ocean fluxes in their solution. The independent constraint on the net atmosphere-to-surface flux based on atmospheric O₂ by design also closes the balance and is 4.2 ± 0.9 GtC yr⁻¹ over the 2015–2024 period (orange symbol on Fig. 13), while the ESMs estimate for the net atmosphere-to-surface flux over that period is 4.8 [2.4, 6.0] GtC yr⁻¹ (Tables 5 and 6).

The distributions based on the individual models and fCO₂-products reveal substantial spread but converge near the decadal means quoted in Tables 5 to 7. Sink estimates for SOCEAN are mostly non-Gaussian, while the ensemble of DGVMs and inverse models appears more normally distributed justifying the use of a multi-model mean and standard deviation for their errors in the budget. Noteworthy is that the tails of the distributions provided by the land and ocean bottom-up estimates would not agree with the global constraint provided by the fossil fuel emissions and the observed atmospheric CO₂ growth rate. This illustrates the power of the atmospheric joint constraint from the global CO₂ observation capacity.

The GCB estimate of the net atmosphere-to-land flux (SLAND-ELUC), calculated as the difference between SLAND from the DGVMs and ELUC from the bookkeeping models, amounts to a 1.0 ± 1.0 GtC yr⁻¹ sink during 2015–2024 (Table 5). The estimate of net atmosphere-to-land flux (SLAND-ELUC) from the DGVMs alone (1.4 ± 0.7 GtC yr⁻¹, Table 5, green symbols on Fig. 13) is slightly larger, although within the uncertainty of the GCB estimate and within uncertainty of the global carbon budget constraint (EFOS-GATM-SOCEAN, 1.0 ± 0.6 GtC yr⁻¹; Table 7). Also, for 2015–2024, the inversions estimate the net atmosphere-to-land flux is a 1.3 ± 0.3 GtC yr⁻¹ sink, similar to the mean of the DGVMs estimates (purple versus grey symbols on Fig. 13). The independent constraint based on atmospheric O₂ is slightly lower, 0.7 ± 0.8 GtC yr⁻¹ (orange symbol in Fig. 13), although its uncertainty overlaps with the uncertainty range from other approaches. Last, the ESMs estimate for the net atmosphere-to-land flux during 2014–2023 is a 2.3 [-0.1, 3.6] GtC yr⁻¹ sink, larger than all other estimates (Table 5).

As discussed in Sect. 3.5.3, the atmospheric growth rate of CO₂ derived from the NOAA surface stations was very high in 2024, 7.9 GtC (3.73 ppm) the largest on the 65 years long observational record. Both DGVMs and inversions assign this large CO₂ growth rate to a continued reduction of the net atmosphere to land flux since 2023, in particular in the tropics (Figs. 11 and 14), especially pronounced in the inversions. DGVMs simulate a 2024 global net atmosphere-to-land flux of 1.1 GtC yr⁻¹, a 50 % decline relative to the 2.2 GtC yr⁻¹ sink in 2022, primarily driven by the severe reduction in SLAND (-37 %, see Sect. 3.7.3). The tropics (30° N–30° S) are recording a dramatic decrease in the net atmosphere-to-land flux from a 1.3 GtC yr⁻¹ sink in 2022 to a 0.2 GtC yr⁻¹ source in 2024. The atmospheric inversions show a continued reduction with the global net atmosphere-to-land flux declining from 2.9 GtC yr⁻¹ in 2022 to 0.8 GtC yr⁻¹ in 2023 to 0.2 GtC yr⁻¹ in 2024 (-94 % from 2022 to 2024), with the tropics turning from a 1.3 GtC yr⁻¹ sink in 2022 to a 1.2 GtC yr⁻¹ source in 2024. This discrepancy between the DGVMs and inversions estimates of the tropical and hence global atmosphere-land flux largely explains the negative BIM in 2024, (see also Sect. “Tropics” below).

For the 2015–2024 period, the GOBMs (2.7 ± 0.4 GtC yr⁻¹) produce a lower estimate of the ocean sink than the fCO₂-products with 3.3 [2.9, 3.8] GtC yr⁻¹, which shows up in Fig. 13 as separate peaks in the distribution from the GOBMs (dark blue symbols) and from the fCO₂-products (light blue symbols). Atmospheric inversions (3.0 ± 0.3 GtC yr⁻¹) suggest an ocean uptake more in line with the average of the GOBMs and fCO₂-products for the recent decade (Table 6). The inversions are not fully independent as 7 out of 14 inversions covering the last decade use fCO₂-products as ocean priors and one uses a GOBM (Table S4). The independent constraint based on atmospheric O₂ (3.5 ± 0.5 GtC yr⁻¹) is at the high end of the distribution of the other methods. However, as mentioned in Sect. 2.8, the O₂ method requires a correction for global air-sea O₂ flux, which induces a non-negligible uncertainty on the decadal estimates (about 0.5 GtC yr⁻¹). The large growth in the ocean carbon sink from O₂ is compatible with the GOBMs and fCO₂-products estimates when accounting for their uncertainty ranges. Lastly, the ESMs estimate, 2.5 [2.2, 2.8] GtC yr⁻¹, suggest a lower average ocean carbon sink than the other estimates. We caution that the riverine transport of carbon taken up on land and outgassing from the ocean, accounted for here, is a substantial (0.65 ± 0.3 GtC yr⁻¹) and uncertain term (Crisp et al., 2022; Gruber et al., 2023; DeVries et al., 2023) that separates the GOBMs, ESMs and oxygen-based estimates on the one hand from the fCO₂-products and atmospheric inversions on the other hand.

Figure 14 shows the latitudinal partitioning of the global atmosphere-to-ocean (SOCEAN), atmosphere-to-land (SLAND-ELUC), and their sum (SOCEAN+SLAND-ELUC) according to the estimates from GOBMs and ocean fCO₂-products (SOCEAN), DGVMs (SLAND-ELUC), and from atmospheric inversions (SOCEAN and SLAND-ELUC). SOCEAN estimates were not corrected for the regional analysis.

Despite being one of the most densely observed and studied regions of our globe, annual mean carbon sink estimates in the northern extra-tropics (north of 30° N) continue to differ. The atmospheric inversions suggest an atmosphere-to-surface sink (SOCEAN+SLAND-ELUC) for 2015–2024 of 2.7 ± 0.4 GtC yr⁻¹, which is higher than the process models' estimate of 2.0 ± 0.4 GtC yr⁻¹ (Fig. 14). The GOBMs (1.1 ± 0.2 GtC yr⁻¹), fCO₂-products (1.4 [1.2–1.6] GtC yr⁻¹), and inversion systems (1.2 ± 0.1 GtC yr⁻¹) produce largely consistent estimates of the ocean sink. However, the larger flux in the fCO₂-products may be related to data sparsity (Mayot et al., 2024). Thus, the difference mainly arises from the net land flux (SLAND-ELUC) estimate, which is 0.9 ± 0.3 GtC yr⁻¹ in the DGVMs compared to 1.5 ± 0.4 GtC yr⁻¹ in the atmospheric inversions (Fig. 14, second row).

Discrepancies in the northern land fluxes conform with persistent issues surrounding the quantification of the drivers of the global net land CO₂ flux (Arneth et al., 2017; Huntzinger et al., 2017; O'Sullivan et al., 2022) and the distribution of atmosphere-to-land fluxes between the tropics and high northern latitudes (Baccini et al., 2017; Schimel et al., 2015; Stephens et al., 2007; Ciais et al., 2019; Gaubert et al., 2019; O'Sullivan et al., 2024).

In the northern extra-tropics, the process models, inversions, and fCO₂-products consistently suggest that most of the interannual variability stems from the land (Fig. 14). Inversions generally agree on the magnitude of interannual variations (IAV) over land, more so than DGVMs (0.27–0.38 vs 0.07–0.55 GtC yr⁻¹, averaged over 1990–2024).

In the tropics (30° S–30° N), both the atmospheric inversions and process models estimate a net carbon balance (SOCEAN+SLAND-ELUC) that is relatively close to neutral over the past decade (inversions: 0.05 ± 0.5 GtC yr⁻¹, process models: 0.4 ± 0.6 GtC yr⁻¹). The GOBMs (-0.003 ± 0.3 GtC yr⁻¹), fCO₂-products (0.3 [0.1, 0.6] GtC yr⁻¹), and inversion systems (0.3 ± 0.1 GtC yr⁻¹) indicate a neutral to positive tropical ocean flux (see Fig. S5 for spatial patterns). DGVMs indicate a net land sink (SLAND-ELUC) of 0.4 ± 0.5 GtC yr⁻¹, whereas the inversion systems indicate a small source in the net land flux although with larger model spread (-0.2 ± 0.6 GtC yr⁻¹, Fig. 14, third row).

A continuing conundrum is the partitioning of the global atmosphere-land flux between the Northern Hemisphere land and the tropical land (Stephens et al., 2007; Pan et al., 2011; Gaubert et al., 2019). It is of importance because each region has its own history of land-use change, climate drivers, and impact of increasing atmospheric CO₂ and nitrogen deposition. Quantifying the magnitude of each sink is a prerequisite to understanding how each individual driver impacts the tropical and mid/high-latitude carbon balance. We define the North–South (N–S) difference as net atmosphere-land flux north of 30° N minus the net atmosphere-land flux south of 30° N. For the inversions, the N–S difference is 1.7 ± 1.0 GtC yr⁻¹ across this year's inversion ensemble. In the ensemble of DGVMs the N–S difference is 0.6 ± 0.5 GtC yr⁻¹, a much narrower range than the one from atmospheric inversions. The smaller spread across DGVMs than across inversions is to be expected as there is no correlation between Northern and Tropical land sinks in the DGVMs as opposed to the inversions where the sum of the two regions being well-constrained by atmospheric observations leads to an anti-correlation between these two regions.

The tropical lands are the origin of most of the atmospheric CO₂ interannual variability (Ahlström et al., 2015), consistently among the process models and inversions (Fig. 14). The interannual variability in the tropics is similar among the ocean fCO₂-products (0.06–0.18 GtC yr⁻¹) and the GOBMs (0.07–0.16 GtC yr⁻¹). The DGVMs and inversions indicate that atmosphere-to-land CO₂ fluxes are more variable than atmosphere-to-ocean CO₂ fluxes in the tropics, with interannual variability of 0.3 to 1.13 and 0.96 GtC yr⁻¹ for DGVMs and inversions, respectively for 1990–2024. The year 2024 saw the largest growth rate in the atmosphere so far, largely caused by reductions in the tropical land sink, similar to earlier extreme years, such as during the 2015–2016 El Niño period. The inversions seem to capture the response of the tropical land sink to a larger degree than the DGVMs, with a 2024 tropical land source of 1.2 ± 0.9 GtC yr⁻¹ for the inversions versus a 0.2 ± 0.7 sink for the DGVMs.

In the southern extra-tropics (south of 30° S), the atmospheric inversions suggest a net atmosphere-to-surface sink (SOCEAN+SLAND-ELUC) for 2015–2024 of 1.6 ± 0.2 GtC yr⁻¹, similar to the process models' estimate of 1.5 ± 0.4 GtC yr⁻¹ (Fig. 14). An approximately neutral net land flux (SLAND-ELUC) for the southern extra-tropics is estimated by both the DGVMs (0.05 ± 0.1 GtC yr⁻¹) and the inversion systems (0.01 ± 0.1 GtC yr⁻¹). This means nearly all carbon uptake is due to oceanic sinks south of 30° S. The Southern Ocean flux in the fCO₂-products (1.6 [1.5, 1.9 GtC] yr⁻¹) and inversion estimates (1.6 ± 0.2 GtC yr⁻¹) is marginally higher than in the GOBMs (1.5 ± 0.4 GtC yr⁻¹) (Fig. 14, bottom row). This agreement is subject to the choice of the river flux adjustment (Lacroix et al., 2020; Hauck et al., 2023b). Nevertheless, the time-series of atmospheric inversions and fCO₂-products diverge from the GOBMs. A substantial overestimation of the trends in the fCO₂-products could be explained by sparse and unevenly distributed observations, especially in wintertime (Fig. S5; Hauck et al., 2023a; Gloege et al., 2021). Model biases likely contribute as well, with biases in mode water formation, stratification, and the chemical buffer capacity known to play a role in Earth System Models (Terhaar et al., 2021; Bourgeois et al., 2022; Terhaar et al., 2022).

The interannual variability in the southern extra-tropics is low because of the dominance of ocean areas with low variability compared to land areas. The split between land (SLAND-ELUC) and ocean (SOCEAN) shows a substantial contribution to variability in the south coming from the land, with no consistency between the DGVMs and the inversions or among inversions. This is expected due to the difficulty of separating exactly the land and oceanic fluxes when viewed from atmospheric observations alone. The SOCEAN interannual variability was found to be higher in the fCO₂-products (0.04–0.18 GtC yr⁻¹) compared to GOBMs (0.03 to 0.06 GtC yr⁻¹) in 1990-2024. Inversions give an interannual variability of 0.13 to 0.15 GtC yr⁻¹. Model subsampling experiments recently illustrated that fCO₂-products may overestimate decadal variability in the Southern Ocean carbon sink by 30 % and the trend since 2000 by 50 %–130 % due to data sparsity, based on one and two fCO₂-products with strong variability (Gloege et al., 2021; Hauck et al., 2023a). The trend benchmark test using the method of Hauck et al. (2023a) and a subset of 6 fCO₂-products confirms the sensitivity of the decadal trends in fCO₂-products to reconstruction biases, particularly in the Southern Ocean, indicating an overestimation of the ensemble mean trend (See Supplement Sect. S3.4). However, we also find compensating positive biases in the ensemble so that the ensemble mean bias is smaller than the bias from some individual fCO₂-products.

Aligning with the RECCAP-2 initiative (Ciais et al., 2022; Poulter et al., 2022; DeVries et al., 2023), we provide a breakdown of this GCB paper estimate of the ELUC, SLAND, Net land (SLAND-ELUC), and SOCEAN fluxes over the 10 land, and 5 ocean RECCAP-2 regions, averaged over the period 2015–2024 (Fig. 15). The DGVMs and inversions suggest a positive net land sink in all regions, except for South America, Africa, and Southeast Asia, where the inversions indicate a small net source of respectively -0.1 ± 0.5, -0.3 ± 0.3, and -0.1 ± 0.3 GtC yr⁻¹ compared to a small sink of 0.1 ± 0.3, 0.1 ± 0.3, and 0.03 ± 0.1 GtC yr⁻¹ for the DGVMs. The uncertainty in the sign of net tropical carbon fluxes is driven by opposing gross fluxes and relatively large uncertainty in the gross fluxes themselves. For South America, DGVMs estimate SLAND of 0.4 ± 0.4 GtC yr⁻¹ and ELUC of 0.3 ± 0.2 GtC yr⁻¹. Bookkeeping models suggest a larger ELUC source of around 0.5 GtC yr⁻¹. Similarly, in Southeast Asia, the DGVMs estimate an ELUC of 0.2 ± 0.1 GtC yr⁻¹, compared with the bookkeeping model estimate of 0.4 ± 0.02 GtC yr⁻¹. Therefore, DGVMs may underestimate ELUC in these regions, which could explain the net sink discrepancy with inversions. In Africa, ELUC is similar for DGVMs and bookkeeping models (∼ 0.4 GtC yr⁻¹), and DGVMs estimate an SLAND of 0.5 ± 0.2 GtC yr⁻¹.

The inversions suggest the largest net land sinks are in North America (0.4 ± 0.3 GtC yr⁻¹), Russia (0.6 ± 0.2 GtC yr⁻¹), Europe (0.3 ± 0.2 GtC yr⁻¹), and East Asia (0.2 ± 0.3 GtC yr⁻¹). This agrees well with the DGVMs in North America (0.4 ± 0.1GtC yr⁻¹), which indicate a large natural land sink (SLAND) of 0.4 ± 0.2 GtC yr⁻¹, and near-zero net land-use related carbon losses (0.02 ± 0.1 GtC yr⁻¹). The DGVMs suggest a smaller net land sink in Russia compared to inversions (0.3 ± 0.1 GtC yr⁻¹), and a similar net sink in East Asia (0.2 ± 0.1 GtC yr⁻¹).

There is generally a higher level of agreement in the estimates of regional SOCEAN between the different data streams (GOBMs, fCO₂-products and atmospheric inversions) on decadal scale, compared to the agreement between the different land flux estimates. All data streams agree that the largest contribution to SOCEAN stems from the Southern Ocean due to a combination of high flux density and large surface area, but with important contributions also from the Atlantic (high flux density) and Pacific (large area) basins. In the Southern Ocean, GOBMs suggest a sink of 1.1 ± 0.4 GtC yr⁻¹, in line with the fCO₂-products (1.1 [1.0, 1.2] GtC yr⁻¹) and atmospheric inversions (1.1 ± 0.2 GtC yr⁻¹). There is similar agreement in the Pacific Ocean, with GOBMs, fCO₂-products, and atmospheric inversions indicating a sink of 0.6 ± 0.1 GtC yr⁻¹, 0.7 [0.6, 0.9] GtC yr⁻¹, and 0.6 ± 0.2 GtC yr⁻¹, respectively. However, in the Atlantic Ocean, GOBMs simulate a sink of 0.5 ± 0.1 GtC yr⁻¹, noticeably lower than both the fCO₂-products (0.8 [0.7, 0.9] GtC yr⁻¹) and atmospheric inversions (0.8 ± 0.1 GtC yr⁻¹). It is important to note the fCO₂-products and atmospheric inversions have a substantial and uncertain river flux adjustment in the Atlantic Ocean (0.3 GtC yr⁻¹) that also leads to a mean offset between GOBMs and fCO₂-products/inversions in the latitude band of the tropics (Fig. 14). The Indian Ocean due its smaller size and the Arctic Ocean due to its size and sea-ice cover that prevents air-sea gas-exchange are responsible for smaller but non negligible SOCEAN fluxes (Indian Ocean: (0.3 ± 0.1 GtC yr⁻¹, 0.3 [0.3, 0.4] GtC yr⁻¹, and 0.3 ± 0.1 GtC yr⁻¹ for GOBMs, fCO₂-products, and atmospheric inversions, respectively, and Arctic Ocean: (0.1 ± 0.03 GtC yr⁻¹, 0.2 [0.1, 0.3] GtC yr⁻¹, and 0.1 ± 0.04 GtC yr⁻¹ for GOBMs, fCO₂-products, and atmospheric inversions, respectively). Note that the SOCEAN numbers presented here deviate from numbers reported in RECCAP-2 where the net air-sea CO₂ flux is reported (i.e. without river flux adjustment for fCO₂-products and inversions, and with river flux adjustment subtracted from GOBMs in most chapters, or comparing unadjusted datasets with discussion of uncertain regional riverine fluxes as major uncertainty, e.g. Sarma et al., 2023; DeVries et al., 2023).

Fire emissions so far in 2025 have been below the average of recent decades, chiefly due to the lowest emissions on record since 2003 across the tropics. Figure S14 shows global and regional emissions estimates for the period 1 January–30 September in each year 2003–2025. Estimates derive from two global fire emissions products: the global fire emissions database (GFED, version 4.1s; van der Werf et al., 2017), and the global fire assimilation system (GFAS, operated by the Copernicus Atmosphere Service; Kaiser et al., 2012). The two products estimate that global emissions from fires were 1.2–1.4 GtC (the range reflecting the difference between the two fire emission datasets, actual uncertainties are larger, and estimates are likely conservative, see Chen et al., 2023) during January–September 2025. These estimates are 19 %–20 % below the 2015–2024 average for the same months (1.5–1.7 GtC). Both the GFED4.1s and GFAS products show the second-lowest January-September fire emissions since 2003 (2022 is the lowest year in both products).

Notably, the year 2025 follows two years with above-average global fire emissions (Fig. S14), with global emissions through September totalling 1.7–2.1 GtC in 2023 and 1.6–2.2 GtC in 2024. These high emissions totals were caused by high emissions in North America in 2023 principally in Canada; Jones et al., 2024b), and in both North America and South America in 2024 (principally in Canada and Brazil; Kelley et al., 2025). For example, in January–September 2024, GFED4.1s suggests that global emissions totalled 2.2 GtC, the highest total on record for these months and 32 % above the average for the decade prior (2014–2023; GFAS also suggests that emissions were 11 % above the average for January–September).

Despite low global fire emissions so far in 2025, a pattern of extremely high fire emissions from Canada has now persisted for three consecutive years commencing with the record-breaking year in 2023 (Jones et al., 2024b; Byrne et al., 2024; Kelley et al., 2025). In January–September 2025, fire emissions from Canada (0.3–0.4 GtC) were slightly greater than in the same months of 2024 (0.2–0.3 GtC yr⁻¹) and around half as large as those in the same months of 2023 (0.5–0.8 GtC). The emissions totals for Canada in January–September 2025 were 2.2 times the average of January–September periods during the prior decade 2015–2024 and 4–6 times greater than the average of those months in 2003–2022 [excluding the record-breaking year in 2023]; Fig. S14). According to GFED4.1s, January–September fire emissions from Canada in just the past three years (1.5 GtC) exceeded the country's total January–September fire emissions throughout the 20 years prior (1.2 GtC during 2003–2022). The continued anomaly in Canada propagated to the Northern Hemisphere, where January–September 2025 emissions of 0.4–0.5 GtC were 13 %–21 % above the average of 2015–2024.

Fire emissions anomalies in Africa strongly influence global fire emissions totals because the continent typically contributed near 50 % of global fire emissions during 2015–2024 (average of January–September periods). For 2025, fire emissions in Africa through September were 0.4–0.6 GtC, 19 %–25 % below the average of 2015–2024 (0.6–0.8 GtC). Synchronously, emissions through September from South America were around 0.1 GtC in both GFED4.1s and GFAS systems, less than half of the average for 2015–2024 (0.2–0.3 GtC), and emissions from Southeast Asia were around 0.05 GtC, also less than half of the average for 2015–2024 (0.1 GtC in both GFED4.1s and GFAS). Low emissions from the tropical parts of Africa, South America, and Southeast Asia contribute to low emissions across the global tropics so far in 2025 (0.7–0.9 GtC), which were only around two-thirds of the average for 2015–2024 (1.1–1.3 GtC) in both GFED4.1s and GFAS.

We caution that the fire emissions fluxes presented here should not be compared directly with other fluxes of the budget (e.g., SLAND or ELUC) due to incompatibilities between the observable fire emission fluxes and what is quantified in the SLAND and ELUC components of the budget. The fire emission estimates from global fire products relate to all fire types that can be observed in Earth Observations (Randerson et al., 2012; Kaiser et al., 2012), including (i) fires occurring as part of natural disturbance-recovery cycles that would also have occurred in the pre-industrial period (Yue et al., 2016; Keeley and Pausas, 2019; Zou et al., 2019), (ii) fires occurring above and beyond natural disturbance-recovery cycle due to changes in climate, CO₂ and N fertilisation and to an increased frequency of extreme drought and heatwave events (Abatzoglou et al., 2019; Jones et al., 2022; Zheng et al., 2021; Burton et al., 2024), and (iii) fires occurring in relation to land use and land use change, such as deforestation fires and agricultural fires (van der Werf et al., 2010; Magi et al., 2012). In the context of the global carbon budget, fire emissions associated with (ii) should be included in the SLAND component, fire emissions associated with (iii) should already be accounted for in the ELUC component, while fire emissions associated with (i) should not be included in the global carbon budget as part of the natural carbon cycle. However, it is not currently possible to derive specific estimates for fluxes (i), (ii), and (iii) using global fire emission products such as GFED or GFAS. In addition, the fire emissions estimates from global fire emissions products represent a gross flux of carbon to the atmosphere, whereas the SLAND component of the budget is a net flux that should also include post-fire recovery fluxes. Even if emissions from fires of type (ii) could be separated from those of type (i), these fluxes may be partially or wholly offset in subsequent years by post-fire fluxes as vegetation recovers, sequestering carbon from the atmosphere to the terrestrial biosphere (Yue et al., 2016; Jones et al., 2024c). Increases in forest fire emissions and severity (emissions per unit area) globally during the past two decades have highlighted the increasing potential for fire emissions fluxes to outweigh post-fire recovery fluxes, though long-term monitoring of vegetation recovery is required to quantify the net effect on terrestrial C storage (Jones et al., 2024c).

Emissions during the period 1850–2024 amounted to 745 ± 65 GtC and were partitioned among the atmosphere (290 ± 5 GtC; 39 %), ocean (200 ± 40 GtC; 27 %), and land (175 ± 50 GtC; 24 %). The cumulative land sink is lower than the cumulative land-use emissions (250 ± 65 GtC), making the global land a source of 75 ± 80 GtC over the whole 1850–2024 period (Table 8).

The use of nearly independent estimates for the individual terms of the global carbon budget shows a cumulative budget imbalance of 80 GtC (11 % of total emissions) during 1850–2024 (Table 8), which suggests that emissions could be slightly too high by the same proportion or that the combined land and ocean sinks are underestimated (by up to 20 %). A large part of the BIM occurs over the first half of the 20th century (Fig. 3, red dashed line) and could originate from the estimation of significant increase in EFOS and ELUC between the mid 1920s and the mid 1960s which is unmatched by a similar growth in atmospheric CO₂ concentration as recorded in ice cores (Fig. 3). Also, we now correct the SLAND estimate for historical reduction in forest cover (RSS, see Sect. 2.6). This reduces the SLAND estimate by about 40 GtC over the 1850–2024 period, contributing to the increase of the historical budget imbalance in comparison to GCB2024 (Friedlingstein et al., 2025a).

For the more recent 1959–2024 period where direct atmospheric CO₂ measurements are available, total emissions (EFOS+ELUC) amounted to 530 ± 50 GtC, of which 420 ± 20 GtC (79 %) were caused by fossil CO₂ emissions, and 110 ± 45 GtC (21 %) by land-use change (Table 8). The total emissions were partitioned among the atmosphere (230 ± 5 GtC; 44 %), ocean (145 ± 30 GtC; 27 %), and the land (120 ± 30 GtC; 23 %), with a budget imbalance of 35 GtC (7 % of total emissions). All components except land-use change emissions have significantly grown since the 1960s, with important interannual variability in the growth rate in atmospheric CO₂ concentration primarily mirrored in the land CO₂ sink (Fig. 4), and some decadal variability in all terms (Table 7). Differences with previous budget releases are documented in Fig. S15.

The global carbon budget averaged over the last decade (2015–2024) is shown in Fig. 2 and Table 7. For this period, 88 % of the total emissions (EFOS+ELUC) were from fossil CO₂ emissions (EFOS), and 12 % from land-use change (ELUC). The total emissions were partitioned among the atmosphere (50 %), ocean (29 %) and land (21 %), with a near zero budget imbalance (0.1 %, < 0.1 GtC yr⁻¹). We note that, compared to GCB2024, the corrections on SOCEAN and SLAND largely reduced the BIM over the 2015–2024 period, by about 0.4 GtC, but at the cost of increasing the cumulated BIM over the longer 1959–2024 period, from 17 GtC (uncorrected estimate) to 37 GtC (corrected estimate).

For single years, the budget imbalance can be larger (Fig. 4f). For 2024, the combination of our estimated anthropogenic sources (11.6 ± 0.9 GtC yr⁻¹) and partitioning in atmosphere, land and ocean (13.3 ± 0.9 GtC yr⁻¹) leads to a large negative BIM of -1.7 GtC (Table 7), indicating that, despite the lower than average SLAND in 2024, the land sink is still too large to explain the record-high atmospheric CO₂ growth rate of 2024. We note that using the 2024 GRESO atmospheric growth rate of 6.8 ± 0.2 GtC yr⁻¹ (Sect. 3.5.3) would reduce the 2024 BIM to about -0.6 GtC.

The carbon budget imbalance (BIM; Eq. 1, Fig. 4f) quantifies the mismatch between the estimated total emissions and the estimated changes in the atmosphere, land, and ocean reservoirs. The budget imbalance from 1959 to 2024 is small (35 GtC cumulated over the period, i.e. 0.5 GtC yr⁻¹ on average) and shows no significant trend over the 1959–2024 period (Fig. 4). The process models (GOBMs and DGVMs) and fCO₂-products have been selected to match observational constraints in the 1990s, but no further constraints have been applied to their representation of trend and variability. Therefore, the small mean and trend in the budget imbalance can be seen as evidence of a coherent community understanding of the emissions and their partitioning on those time scales (Fig. 4). However, the budget imbalance shows substantial variability of the order of ± 1 GtC yr⁻¹, particularly over semi-decadal time scales, although most of the variability is within the uncertainty of the estimates.

We cannot attribute the cause of the budget imbalance with our analysis, we only note that the budget imbalance is unlikely to be explained by errors in the emissions alone because of its large semi-decadal variability component, a variability that is atypical of emissions, and also because the budget imbalance has not increased in the past 60 years despite a near tripling in anthropogenic emissions (Fig. 4). Errors in SLAND and SOCEAN are more likely to be the main cause for the budget imbalance, especially on interannual to semi-decadal timescales. For example, underestimation of the SLAND by DGVMs has been reported following the eruption of Mount Pinatubo in 1991 possibly due to missing responses to changes in diffuse radiation (Mercado et al., 2009). Although we account for aerosol effects on solar radiation quantity and quality (diffuse vs direct), most DGVMs only used the former as input (i.e., total solar radiation) (Table S1). Thus, the ensemble mean may not capture the full effects of volcanic eruptions, i.e., associated with high light scattering sulphate aerosols, on the land carbon sink (O'Sullivan et al., 2021), potentially explaining the large positive BIM in 1991–1992. DGVMs are suspected to underestimate the land sink reduction in response to El Niño events (see Sect. 3.7.3), which could explain the large negative BIM in 1986–1987, 1997–1998 or 2023–2024). Quasi-decadal variability in the ocean sink has also been reported, with all methods agreeing on smaller than expected ocean CO₂ sink in the 1990s and a larger than expected sink in the 2000s (Fig. 10b; Landschützer et al., 2016; DeVries et al., 2019; Hauck et al., 2020, McKinley et al., 2020; Gruber et al., 2023) and the climate-driven variability could be substantial but is not well constrained (DeVries et al., 2023; Müller et al., 2023). Errors in sink estimates could also be partly driven by errors in the climatic forcing data, particularly precipitation for SLAND and wind for SOCEAN.

Although the budget imbalance is near zero for the most recent decade, it could be due to a compensation of errors. We cannot exclude an overestimation of CO₂ emissions, particularly from land-use change, given their large uncertainty, as has been suggested elsewhere (Piao et al., 2018), and/or an underestimate of the land or ocean sinks. A larger estimate of the atmosphere-land CO₂ flux (SLAND-ELUC) over the extra-tropics would reconcile bottom-up model results with inversion estimates (Fig. 14). Likewise, a larger SOCEAN is also possible given the higher estimates from the fCO₂-products and the oxygen-based estimates (see Table 6 and Fig. 10b), the underestimation of interior ocean anthropogenic carbon accumulation in the GOBMs (Sect. 3.6.5, Müller et al., 2023), or known biases of ocean models (e.g., Terhaar et al., 2022, 2024). More integrated use of observations in the global carbon budget, either on their own or for further constraining model results, should help resolve some of the budget imbalance.