Indicators of Global Climate Change 2023: annual update of key indicators of the state of the climate system and human influence

Historic emissions from human activity were assessed in both AR6 WGI and WGIII. Chapter 5 of WGI assessed CO2 and CH4 emissions in the context of the carbon cycle (Canadell et al., 2021). Chapter 6 of WGI assessed emissions in the context of understanding the climate and air quality impacts of short-lived climate forcers (Szopa et al., 2021). Chapter 2 of WGIII, published 1 year later (Dhakal et al., 2022), assessed the sectoral sources of emissions and gave the most up-to-date understanding of the current level of emissions. This section bases its methods and data on those employed in this WGIII chapter.

Methods of estimating greenhouse gas emissions changes

Like in AR6 WGIII, net GHG emissions in this paper refer to releases of GHGs from anthropogenic sources minus removals by anthropogenic sinks, for greenhouse gases reported under the common reporting format of the UNFCCC. This includes CO2 emissions from fossil fuels and industry (CO2-FFI); net CO2 emissions from land use, land-use change and forestry (CO2-LULUCF); CH4; N2O; and fluorinated gas (F-gas) emissions. CO2-FFI mainly comprises fossil-fuel combustion emissions, as well as emissions from industrial processes such as cement production. This excludes biomass and biofuel use. CO2-LULUCF is mainly driven by deforestation but also includes anthropogenic removals on land from afforestation and reforestation, emissions from logging and forest degradation, and emissions and removals in shifting cultivation cycles, as well as emissions and removals from other land-use change and land management activities, including peat burning and drainage. The non-CO2 GHGs – CH4, N2O and F-gas emissions – are linked to the fossil-fuel extraction, agriculture, industry and waste sectors.

Global regulatory conventions have led to a twofold categorisation of F-gas emissions (also known as halogenated gases). Under UNFCCC accounting, countries record emissions of hydrofluorocarbons (HFCs), perfluorocarbons (PFCs), sulfur hexafluoride (SF6) and nitrogen trifluoride (NF3) – hereinafter “UNFCCC F-gases”. However, national inventories tend to exclude halons, chlorofluorocarbons (CFCs) and hydrochlorofluorocarbons (HCFCs) – hereinafter “ODS (ozone-depleting substance) F-gases” – as they have been initially regulated under the Montreal Protocol and its amendments. In line with the WGIII assessment, ODS F-gases and other substances are not included in our GHG emissions reporting but are included in subsequent assessments of concentration change (including compounds formed in the atmosphere as ozone), effectivepage2628 radiative forcing, human-induced warming, carbon budgets and climate impacts in line with the WGI assessment.

There are also varying conventions used to quantify CO2-LULUCF fluxes. These include the use of bookkeeping models, dynamic global vegetation models (DGVMs) and aggregated national inventory reporting (Pongratz et al., 2021). Each differs in terms of their applied system boundaries and definitions, and they are not directly comparable. However, efforts to “translate” between bookkeeping estimates and national inventories using DGVMs have demonstrated a degree of consistency between the varying approaches (Friedlingstein et al., 2022; Grassi et al., 2023).

Each category of GHG emissions included here is covered by varying primary sources and datasets. Although many datasets cover individual categories, few extend across multiple categories, and only a minority have frequent and timely update schedules. The Global Carbon Budget (GCB; Friedlingstein et al., 2023) covers CO2-FFI and CO2-LULUCF. The Emissions Database for Global Atmospheric Research (EDGAR; Crippa et al., 2023) and the Potsdam Real-time Integrated Model for probabilistic Assessment ofpage2629 emissions Paths (PRIMAP-hist; Gütschow et al., 2016, 2024) cover CO2-FFI, CH4, N2O and UNFCCC F-gases. The Community Emissions Data System (CEDS; Hoesly et al., 2018; Hoesly and Smith, 2024) covers CO2-FFI, CH4 and N2O. The Global Fire Emissions Database (GFED; van der Werf et al., 2017) version 4.1s covers CO2, CH4 and N2O. As detailed below, for various reasons, not all these datasets were employed in this update.

In AR6 WGIII, total net GHG emissions were calculated as the sum of CO2-FFI, CH4, N2O and UNFCCC F-gases from EDGAR and net CO2-LULUCF emissions from the GCB. Net CO2-LULUCF emissions followed the GCB convention and were derived from the average of three bookkeeping models (Hansis et al., 2015; Houghton and Nassikas, 2017; Gasser et al., 2020). Version 6 of EDGAR was used (with a fast-track methodology applied for the final year of data – 2019), alongside the 2020 version of the GCB (Friedlingstein et al., 2020). CO2-equivalent (CO2e) emissions were calculated using global warming potentials with a 100-year time horizon (GWP100 henceforth) from AR6 WGI Chap. 7 (Forster et al., 2021). Uncertainty ranges were based on a comparative assessment of available data and expert judgement, corresponding to a 90% confidence interval (Minx et al., 2021): ±8% for CO2-FFI, ±70% for CO2-LULUCF, ±30% for CH4 and F-gases, and ±60% for N2O (note that the GCB assesses 1 standard deviation uncertainty for CO2-FFI as ±5% and for CO2-LULUCF as ±2.6 Gt CO2; Friedlingstein et al., 2022). The total uncertainty was summed in quadrature, assuming independence of estimates per species/source. Reflecting these uncertainties, AR6 WGIII reported emissions to two significant figures only. Uncertainties in GWP100 metrics of roughly ±10% were not applied (Minx et al., 2021).

This analysis tracks the same compilation of GHGs as in AR6 WGIII. We follow the same approach for estimating uncertainties and CO2e emissions. We also use the same type of data sources but make important changes to the specific selection of data sources to further improve the quality of the data, as suggested in the knowledge gap discussion of the WGIII report (Dhakal et al., 2022). Instead of using EDGAR data (which are now available as version 8), we use GCB data for CO2-FFI, PRIMAP-hist CR (country-reported) data for CH4 and N2O, and atmospheric concentrations with best-estimate lifetimes for UNFCCC F-gas emissions (Hodnebrog et al., 2020). As in AR6 WGIII we use GCB for net CO2-LULUCF emissions, taking the average of three bookkeeping models (BLUE by Hansis et al., 2015; H&C by Houghton and Castanho, 2023; OSCAR by Gasser et al., 2020). The GCB methodology includes CO2 emissions from deforestation and forest degradation fires but excludes wildfires, which are assumed to be natural even if climate change affects their intensity and frequency. Bunker emissions are included but military emissions excluded (e.g. Bun et al., 2024). For more completeness, this year we also include estimates of N2O and CH4 emissions from global biomass fires, sourced from GFED4.1s.

There are three reasons for these specific data choices. First, national greenhouse gas emissions inventories tend to use improved, higher-tier methods for estimating emissions fluxes than global inventories such as EDGAR (Dhakal et al., 2022; Minx et al., 2021). As GCB and PRIMAP-hist CR integrate the most recent national inventory submissions to the UNFCCC, selecting these databases makes best use of country-level improvements in data-gathering infrastructures. It is important to acknowledge, however, that national inventories differ substantially with respect to reporting intervals, applied methodologies and emissions factors (Minx et al., 2021). Notably, the PRIMAP-hist CR dataset has significantly lower total CH4 emissions relative to both the other datasets reported here and the global atmospheric inversion estimates evaluated in this paper. A substantive body of literature has evaluated national-level CH4 inversions versus inventories, finding a tendency for the former to exceed the latter (Deng et al., 2022; Tibrewal et al., 2024; Janardanan et al., 2024; Scarpelli et al., 2022). Compared to the median of reported inversion models from Deng et al. (2022), PRIMAP-Hist CR reports lower CH4 emissions for India, the EU27 + UK, Brazil, Russia and Indonesia but not in the case of China and the United States (see Fig. S1).

Second, comprehensive reporting of F-gas emissions has remained challenging in national inventories and may exclude some military applications (see Minx et al., 2021; Dhakal et al., 2022). However, F-gases are entirely anthropogenic substances, and their concentrations can be measured effectively and reliably in the atmosphere. We therefore follow the AR6 WGI approach in making use of direct atmospheric observations.

Third, the choice of GCB data for CO2-FFI means we can integrate its projection of that year's CO2 emissions at the time of publication (i.e. for 2023). No other dataset except GCB provides projections of CO2 emissions in this time frame. At this point in the publication cycle (mid-year), the other chosen sources provide data points with a 2-year time lag (i.e. for 2022). While these data choices inform our overall assessment of GHG emissions, we provide a comparison across datasets for each emissions category, as well as between our estimates and an estimate derived from AR6 WGIII-like databases (i.e. EDGAR for CO2-FFI and non-CO2 GHG emissions and GCB for CO2-LULUCF).

Updated greenhouse gas emissions

Updated GHG emission estimates are presented in Fig. 2 and Table 1. Total global GHG emissions were 55±5.3 Gt CO2e in 2022, the same as previous high levels in 2019 and 2021. Of this total, CO2-FFI contributed 36.4±2.9 Gt CO2, CO2-LULUCF contributed 4.3±3 Gt CO2, CH4 contributed 9±2.7 Gt CO2e, N2O contributed 3.1±1.9 Gt CO2e and F-gas emissions contributed 1.8±0.54 Gt CO2e. Initial projections indicate that total CO2 emissions remained similar in 2023, with emissions from fossil fuel and industry at 36.8±3 and from land-use change at 4.1±2.9 Gt CO2 (Friedlingstein et al., 2023; see also Liu et al., 2024; IEA, 2023). Note that ODS F-gases such as chlorofluorocarbons and hydrochlorofluorocarbons are excluded from national GHG emissions inventories. For consistency with AR6, they are also excluded here. Including them here would increase total global GHG emissions by 1.5 Gt CO2e in 2022.

Average annual GHG emissions for the decade 2013–2022 were 53±5.4 Gt CO2e, which is the same as the estimate from last year for 2012–2021. Average decadal GHG emissions have increased steadily since the 1970s across all major groups of GHGs, driven primarily by increasing CO2 emissions from fossil fuel and industry but also rising emissions of CH4 and N2O. Stratospheric ozone-depleting F-gases are regulated under the Montreal Protocol and its amendments, and their emissions have declined substantially since the 1990s, whereas emissions of other F-gases, regulated under the UNFCCC, have grown more rapidly than other greenhouse gas emissions but from low levels. Both the magnitude and trend of CO2 emissions from land-use change remain highly uncertain, with the latest data indicating an average net flux between 4 and 5 Gt CO2 yr-1 for the past few decades.

AR6 WGIII reported total net anthropogenic emissions of 59±6.6 Gt CO2e in 2019 and decadal average annual emissions of 56±6.0 Gt CO2e from 2010–2019. By comparison, our estimates here for the AR6 period sum to 55±5.5 Gt CO2e in 2019 and an annual average of 53±5.5 Gt CO2e for the same decade (2010–2019). The difference between these figures, including the reduced relative uncertainty range, is partly driven by the substantial revision in GCB CO2-LULUCF estimates between the 2020 version (used in AR6 WGIII) of 6.6 Gt CO2 and the 2022 version (used here) of 4.6 Gt CO2. The main reason for this downward revision comes from updated estimates of agricultural areas by the FAO, which uses multi-annual land-cover maps from satellite remote sensing, leading to lower emissions from cropland expansion, particularly in the tropical regions. It is important to note that this change is not a reflection of changed and improved methodology per se but an update of the resulting estimation due to updates in the available input data. Second, there are relatively small changes resulting from improvements in datasets since AR6, including the new addition of global biomass burning (landscape fire) emissions. Dataset impacts are largest for CH4, where the emission estimate had reduced by 1.6 Gt CO2e in 2019. This is related to the switch from EDGAR in AR6 to PRIMAP-hist CR in this study. EDGAR estimates considerably higher CH4 emissions – from fugitive fossil sources, as well as the livestock, rice cultivation and waste sectors – compared to country-reported data using higher-tier methods, as compiled in PRIMAP-hist CR (see Sect. 2.1). The estimate for CO2-FFI is also 1.6 Gt CO2e lower for 2019 in this study due to the switch from EDGAR to GCB, as the latter includes a cement carbonation sink not considered in EDGAR. Differences in the remaining gases for 2019 are relatively small in magnitude (increases in N2O (+0.42 Gt CO2e) and UNFCCC-F-gases (+0.35 Gt CO2e)).

The fossil-fuel share of global greenhouse gas emissions was approximately 70% in 2022 (GWP100-weighted), based on the EDGAR v8 dataset (Crippa et al., 2023) and net land-use CO2 emissions from the Global Carbon Budget (Friedlingstein et al., 2023). Non-fossil-fuel emissions are mostly from land-use change, agriculture, cement production, waste and F-gas emissions.

New literature not available at the time of the AR6 suggests that increases in atmospheric CH4 concentrations are also driven by methane emissions from wetland changes resulting from climate change (e.g. Basu et al., 2022; Peng et al., 2022; Nisbet et al., 2023; Zhang et al., 2023). There is also a possible effect from CO2 fertilisation (Feron et al., 2024; Hu et al., 2023). Such carbon cycle feedbacks are not considered here as they are not a direct emission from human activity, yet they will contribute to greenhouse gas concentration rise, forcing and energy budget changes discussed in the next sections. They will become more important to properly account for in future years.

Non-methane short-lived climate forcers

In addition to GHG emissions, we provide an update of anthropogenic emissions of non-methane short-lived climate forcers (SLCFs) (SO2, black carbon (BC), organic carbon (OC), NOx, volatile organic compounds (VOCs), CO and NH3). Data are presented in Table 2. HFCs are considered in Sect. 2.2.

Sectoral emissions of SLCFs are derived from two sources. For fossil fuel, industrial, waste and agricultural sectors, we use the CEDS dataset. CEDS provides global emissions totals from 1750 to 2022 in its most recent version (v_2024_04_01) (Hoesly et al., 2018; Hoesly and Smith, 2024). No CEDS emissions data are currently available for 2023. The estimate for 2023 was derived by assuming a scaled return to an underlying SSP2-4.5 emissions scenario, used for inputs to COVID-MIP (Forster et al., 2020; Lamboll et al., 2021). We find that the 2020-2022 emissions trends comparing CEDS and the COVID-MIP extrapolation are not substantially different (Fig. S2), so the COVID-MIP extension to 2023 is justifiable. In Forster et al. (2023), the CEDS dataset was only available to 2019, so the COVID-MIP extension was used to 2022. Therefore, emissions from 2020 have been revised in this year's paper with 2020–2022 data now arising from CEDS.

Overall, the net SO2 emissions were similar (within 2 Tg SO2; see Sect. S2) over the 2020–2022 period in the CEDS dataset than our estimate in Forster et al. (2023). The CEDS dataset accounts for the introduction of strict fuel sulfur controls brought in by the International Maritime Organization on 1 January 2020. Total SO2 emissions in 2019 were 84.2 Tg SO2 (Table 2). The SO2 emissions from international shipping declined by 7.4 Tg SO2 from 10.4 Tg SO2 in 2019 to 3.0 Tg SO2 in 2020, which is close to the expected 8.5 Tg SO2 reduction estimated by the International Maritime Organization, approximately -80% frompage2631 the 2019 number, accounting for a 3-month phase in period and COVID-19 changes. Non-shipping SO2 emissions were impacted slightly by COVID-19 but had rebounded to levels close to those of 2019 by 2022 in CEDS.

For biomass burning SLCF emissions, we follow AR6 WGIII (Dhakal et al., 2022) and use GFED (van der Werf et al., 2017) version 4 with small fires (GFED4.1s) for 1997 to 2023, with the dataset extended back to 1750 for CMIP6 (van Marle et al., 2017). Estimates from 2017 to 2023 are provisional. As demonstrated with the update to CEDS emissions, the potential for both sources of emissions data to be updated in future versions exists, for example, with a planned introduction of GFED5 in preparation for CMIP7.

Using our combined estimate of GFED and CEDS with a 2023 extrapolation, emissions of all SLCFs were reduced in 2022 relative to 2019 but rebounded again in 2023 (Table 2). The primary driver of the increase in 2023 is an anomalous biomass burning year, mostly related to the unprecedented 2023 Canadian fire season (Barnes et al., 2023), with a smaller contribution from a continued recovery from COVID-19. Under these assumptions, 2023 was a record year for emissions of organic carbon (driven again by a very active biomass burning season) and ammonia (driven by a steady background increase in agricultural sources and a contribution from biomass burning). Causes of the enhanced burning are not distinguished in the GFED data. Whether human-caused burning, a feedback due to the extreme heat or naturally occurring, we choose to include them in our tracking, as historical biomass burning emissions inventories have previously been consistently treated as anthropogenic (for example in CMIP6), though this assumption may need to be revisited in the future. This differs from the treatment of accounting for CO2 and CH4 emissions at present (Sect. 2.2), where we do not include natural emissions in the inventories. As described in Sect. 4, the treatment of all biomass burning emissions as a forcing has implications for several categories of anthropogenic radiative forcing. Trends in SLCF emissions are spatially heterogeneous (Szopa et al., 2021), with strong shifts in the locations of reductions and increases over the 2010–2019 decade (Hodnebrog et al., 2024).

Uncertainties associated with these emission estimates are difficult to quantify. From the non-biomass-burning sectors they are estimated to be the smallest for SO2 (±14%), largest for black carbon (BC) (a factor of 2) and intermediate for other species (Smith et al., 2011; Bond et al., 2013; Hoesly et al., 2018). Uncertainties are also likely to increase both backwards in time (Hoesly et al., 2018) and again in the most recent years. The estimates of non-biomass-burning emissions for 2023, especially SO2, are highly uncertain, owing to the use of proxy activity data used with a SSP2-4.5 scenario extension (see above). Future updates of CEDS are expected to include uncertainties (Hoesly et al., 2018). Even though trends over recent years are uncertain, the general decline in some SLCF emissions derived from inventories punctuated by temporary anomalous years with high biomass burning emissions including 2023 is supported by MODIS Terra and Aqua aerosol optical depth measurements (e.g. Quaas et al., 2022; Hodnebrog et al., 2024).