Quantifying past changes in the rate of CO₂ accumulation in the atmosphere based on reconstructions using marine sediment proxies is complex as age model uncertainties, assumptions and shortcomings underlying proxy applications and sedimentary processes conspire to alter and confound rate estimates (Ajayi et al., 2020). Differential sediment mixing and bioturbation contribute to smooth and attenuate proxy records (Hupp and Kelly, 2020), thereby tending to underestimate maximum rates of change (Kemp et al., 2015). Considering the extent to which uncertainties can affect sediment-based rate estimates, and notwithstanding recent effort in minimizing their inherent contribution, there is generallylow to medium confidence in quantifying rates of change on a time scale less than a decade back thousands of years, and less than a millennium back millions of years in the past based on marine sediments.

In the past, atmospheric CO₂ concentrations reached much higher levels than present day (Cross-Chapter Box 2.1 and Figure 5.3). In particular, the Paleocene–Eocene thermal maximum (PETM), 55.9–55.7 Ma (Figure 5.3), provides some level of comparison with the current and projected anthropogenic increase in CO₂ emissions (Chapter 2). Atmospheric CO₂ concentrations increased from about 900 to around 2000 ppm in 3–20 kyr as a result of geological carbon release to the ocean–atmosphere system (Zeebe et al., 2016; Gutjahr et al., 2017; Cui and Schubert, 2018; Kirtland Turner, 2018). There is low to medium confidence in evaluations of the total amount of carbon released during the PETM, as proxy data constrained estimates vary from around 3000 to more than 7000 PgC, with methane hydrates, volcanic emissions, terrestrial and/or marine organic carbon, or some combination thereof, as the probable sources of carbon (Zeebe et al., 2009; Cui et al., 2011; Gutjahr et al., 2017; Elling et al., 2019; Jones et al., 2019; Haynes and Hönisch, 2020). Methane emissions related to hydrate/permafrost thawing and fossil carbon oxidation may have acted as positive feedbacks (Lunt et al., 2011; Armstrong McKay and Lenton, 2018; Lyons et al., 2019), as the inferred increase in atmospheric CO₂ can only account for approximately half of the reported warming (Zeebe et al., 2009). The estimated, time-integrated carbon input is broadly similar to the RCP8.5 extension scenario, although CO₂ emissions rates (0.3–1.5 Pg yr^–1) and by inference the rate of CO₂ accumulation in the atmosphere (4–42 ppm per century) during the PETM were at least 4–5 lower than during the modern era (from 1995 to 2014; Table 2.1; Zeebe et al., 2016; Gingerich, 2019).

Figure 5.3 | Atmospheric CO₂ concentrations and growth rates for the past 60 million years (Myr) and projections to 2100. (a) CO₂ concentrations data for the period 60 Myr to the time prior to 800 kyr (left column) are shown as the LOESS Fit and 68% range (data from Chapter 2) (Foster et al., 2017). Concentrations from 1750 and projections through 2100 are taken from Shared Socio-economic Pathways of IPCC AR6 (Meinshausen et al., 2017). (b) Growth rates are shown as the time derivative of the concentration time series. Inserts in (b) show growth rates at the scale of the sampling resolution. Further details on data sources and processing are available in the chapter data table (Table 5.SM.6).

Open figure

The last 50 Myr (50 million years) have been characterized by a gradual decline in atmospheric CO₂ levels at a rate of about 16 ppm Myr^–1 (Figure 5.3; Foster et al., 2017; Gutjahr et al., 2017). The exact cause of this long-term change in CO₂ remains uncertain, but may be related to an imbalance between long-term sources of CO₂ (volcanic outgassing) and long-term sinks (organic carbon burial and silicate weathering).

The most recent time interval when atmospheric CO₂ concentration was as high as 1000 ppm (i.e., similar to the end of 21st century projection for the high-end emissions scenario RCP8.5) was around 33.5 Ma, prior to the Eocene-Oligocene transition (Zhang et al., 2013; Anagnostou et al., 2016). Atmospheric CO₂ levels then reached a critical threshold (1000–750 ppm; DeConto et al., 2008) to allow for the development of permanent regional ice sheets on Antarctica, associated with changes in Southern Ocean hydrography, which would have increased deep ocean CO₂ storage (Leutert et al., 2020).

The most recent interval characterized by atmospheric CO₂ levels similar to modern (i.e., 360–420 ppm) was the mid-Pliocene Warm Period (MPWP, 3.3–3.0 Ma; Martínez-Botí et al., 2015; de la Vega et al., 2020) (Chapter 2). The relatively high atmospheric CO₂ concentration during the MPWP are related to vigorous ocean circulation and a rather inefficient marine biological carbon pump (Burls et al., 2017), which would have reduced deep ocean carbon storage. After the MPWP, atmospheric CO₂ concentrations declined gradually at a rate of 30 ppm Myr^–1 (Figure 5.3; de la Vega et al., 2020), as an increase in ocean stratification led to enhanced ocean carbon storage, allowing for major, sustained advances in Northern Hemisphere ice sheets, 2.7 Ma (Sigman et al., 2004; DeConto et al., 2008).

The Antarctic ice core record covering the past 800 kyr provides an important archive to explore the carbon–climate feedbacks prior to anthropogenic perturbations (Brovkin et al., 2016). Polar ice cores represent the only climatic archive from which past GHG concentrations can be directly measured. Major GHGs, CH₄, N₂O and CO₂ generally co-vary on orbital time scales (Loulergue et al., 2008; Lüthi et al., 2008; Schilt et al., 2010b; Chapter 2), with consistently higher atmospheric concentrations during warm intervals of the past, pointing to a strong sensitivity to climate (Figure 5.4). Modelling work suggests that the carbon cycle contributed to globalise and amplify changes in orbital forcing, which are pacing glacial–interglacial climate oscillations (Ganopolski and Brovkin, 2017), with ocean biogeochemistry and physics, terrestrial vegetation, peatland, permafrost and exchanges with the lithosphere including chemical weathering, volcanic activity, sediment burial and marine calcium carbonate compensation all playing a role in modulating the concentration of atmospheric GHGs.

Figure 5.4 | Atmospheric concentrations of CO₂ , CH₄ and N₂ O in air bubbles and clathrate crystals in ice cores (800,000 BCE to 1990 CE). Note the variable x-axis range and tick mark intervals for the three columns. Ice core data is over-plotted by atmospheric observations from 1958 to present for CO₂, from 1984 for CH₄ and from 1994 for N₂O. The time-integrated, millennial-scale linear growth rates for different time periods (800,000–0 BCE, 0–1900 CE and 1900–2017 CE) are given in each panel. For the BCE period, mean rise and fall rates are calculated for the individual slopes between the peaks (interglacials) and troughs (glacial periods), which are given in the panels in left column. The data for BCE period are used from the Vostok, EPICA, Dome C and WAIS ice cores (Petit et al., 1999; Monnin, 2001; Pépin et al., 2001; Raynaud et al., 2005; Siegenthaler et al., 2005; Loulergue et al., 2008; Lüthi et al., 2008; Schilt et al., 2010a). The data after 0–yr CE are taken mainly from Law Dome ice core analysis (MacFarling Meure et al., 2006). The surface observations for all species are taken from NOAA cooperative research network (Dlugokencky and Tans, 2019), where ALT, MLO and SPO stand for Alert (Canada), Mauna Loa Observatory, and South Pole Observatory, respectively. BCE = before current era, CE = current era. Further details on data sources and processing are available in the chapter data table (Table 5.SM.6).

Open figure

Since AR5, the number of ice core records and the temporal resolution of their data for the last 800 kyr have improved, in particular for the last 60 kyr. Additionally, the advent of isotopic measurements on GHGs extracted from air trapped in ice, allows for more robust source apportionments and inventory assessments. Therefore, the ensuing discussion focuses on these two specific aspects.

Major pre-industrial sources of CH₄ comprise wetlands (including subglacial environments) and biomass burning (Bock et al., 2010, 2017; Lamarche-Gagnon et al., 2019; Kleinen et al., 2020). Pre-industrial atmospheric N₂O concentrations were regulated by microbial production in marine and terrestrial environments and by photochemical removal in the stratosphere (Schilt et al., 2014; Battaglia and Joos, 2018b; H. Fischer et al., 2019). Pre-industrial atmospheric CO₂ concentrations were largely regulated by exchange with exogenic terrestrial and ocean carbon reservoirs. The imbalance between geological sources and sinks in the ocean–atmosphere–land biosphere system additionally plays an important role in modulating the air–sea partitioning of the active carbon inventory on multi-millennial time scales (Cartapanis et al., 2018).

Model-based estimates indicate that wetland CH₄ emissions were reduced by 24–40% during the Last Glacial Maximum (LGM) when compared to pre-industrial, while CH₄ emissions related to biomass burning (wildfires) decreased by 35–75% (Valdes et al., 2005; Hopcroft et al., 2017; Kleinen et al., 2020). N₂O emissions decreased by about 30% during the LGM based on data-constrained model estimates (Schilt et al., 2014; H. Fischer et al., 2019) owing to a combination of a weaker hydrological cycle and a generally better ventilated intermediate depth ocean relative to present, reducing (de)nitrification processes (Galbraith et al., 2013; Fischer et al., 2019).

During past ice ages, generally colder and drier climate conditions contributed to a substantial decline of the land biosphere carbon inventory, in particular in boreal peatlands (–300 PgC; Treat et al., 2019). Estimates assessing the glacial decrease in the global terrestrial biosphere carbon stock vary between –300 and –600 PgC (Ciais et al., 2012; Peterson et al., 2014; Menviel et al., 2017; Kleinen et al., 2020), possibly –850 PgC when accounting for ocean-sediment interactions and burial (Jeltsch-Thömmes et al., 2019), a considerable contraction when compared to the modern land biosphere stock. The large range of estimates reflects a yet limited understanding of how carbon cycle dynamics were altered by glacially perturbed nutrient fluxes and soil dynamics, as well as largely exposed shelf areas in the tropics as a result of lowered sea level. Recent estimates suggest deep-sea CO₂ storage during the last ice age exceeded modern values by as much as 750 – 950 PgC (Skinner et al., 2015, 2017; Buchanan et al., 2016; Anderson et al., 2019; Gottschalk et al., 2020b). A combination of increased CO₂ solubility associated with 2–3°C lower mean oceanic temperatures (Bereiter et al., 2018), increased the oceanic residence time of CO₂ (Skinner et al., 2017), altered oceanic alkalinity (Yu et al., 2010; Cartapanis et al., 2018). A generally more efficient marine biological carbon pump (BCP; Galbraith and Jaccard, 2015; Yu et al., 2019; Galbraith and Skinner, 2020) enhanced the partition CO₂ into the ocean interior, (although the relative contribution of each mechanism remains a matter of debate). Recent observationally constrained ESM results highlight that air–sea disequilibrium amplifies the effect of cooling and iron fertilization on glacial carbon storage (Khatiwala et al., 2019).

Ice core observations combined with model-based estimates thus reveal with high confidence that both terrestrial and marine CH₄ and N₂O emissions were reduced under glacial climate conditions. Multiple lines of evidence indicate with high confidence that enhanced storage of remineralized CO₂ in the ocean interior, owing to a combination of synergistic mechanisms, was sufficient to balance the removal of carbon from the atmosphere and the terrestrial biosphere reservoirs combined during the last ice age.

Vegetation regrowth and increased precipitation in wetland regions associated with the mid-deglacial Northern Hemisphere warming (referred to as the Bølling/Allerød (B/A) warm interval, 14.7–12.7 ka), in particular in the (sub)tropics, accounts for large increases in both CH₄ and N₂O emissions to the atmosphere (Baumgartner et al., 2014; Schilt et al., 2014; Bock et al., 2017; H. Fischer et al., 2019). Specifically, changes in CH₄ sources were steered by variations in vegetation productivity, source size area, temperatures and precipitation as modulated by insolation, local sea level changes and monsoon intensity (Bock et al., 2017; Kleinen et al., 2020). Changes in the CH₄ atmospheric sink term probably only played a secondary role in modulating atmospheric CH₄ inventories across the LDT (Hopcroft et al., 2017; Kleinen et al., 2020) Geological emissions, related to the destabilization of fossil (radiocarbon-dead) CH₄ sources buried in continental margins as a result of sudden warming, appear small (Bock et al., 2017; Petrenko et al., 2017; Dyonisius et al., 2020). Stable isotope analysis on N₂O extracted from Antarctic and Greenland ice reveal that marine and terrestrial emissions increased by 0.7 ± 0.3 and 1.7 ± 0.3 TgN, respectively, across the LDT (Fischer et al., 2019). During abrupt Northern Hemisphere warmings, terrestrial emissions responded rapidly to the northward displacement of the Intertropical Convergence Zone (ITCZ) associated with the resumption of the Atlantic meridional overturning circulation (AMOC; H. Fischer et al., 2019). About 90% of these step increases occurred rapidly, possibly in less than 200 years (Fischer et al., 2019). In contrast, marine emissions increased more gradually, modulated by global ocean circulation reorganization.

The gradual increase in atmospheric CO₂ across the LDT was punctuated by three centennial 10–13 ppm increments, coeval with 100–200 ppb increases in CH₄ (Marcott et al., 2014), reminiscent of similar oscillations reported for the last ice age associated with transient warming events (Dansgaard/Oeschger (DO) events; Ahn and Brook, 2014; Rhodes et al., 2017; Bauska et al., 2018) as well as previous deglacial transitions (Nehrbass-Ahles et al., 2020). The rate of change in atmospheric CO₂ accumulation during these transient events exceeds the averaged deglacial growth rates by at least 50% (Table 2.1, Figure 5.4). The early deglacial release of remineralized carbon from the ocean abyss coincided with the resumption of Southern Ocean overturning circulation (Skinner et al.,2010; Schmitt et al., 2012; Ferrari et al., 2014; Gottschalk et al., 2016, 2020a; Jaccard et al., 2016; Rae et al., 2018; Moy et al., 2019) and the concomitant reduction in the global efficiency of the marine BCP, associated, in part, with dwindling iron fertilization (Hain et al., 2010; Martínez-García et al., 2014; Jaccard et al., 2016) The two subsequent pulses, centred 14.8 and 12.9 ka, are associated with enhanced air–sea gas exchange in the Southern Ocean (T. Li et al., 2020), iron fertilization in the South Atlantic and North Pacific (Lambert et al., 2021) and rapid increase in soil respiration owing to the resumption of AMOC and associated southward migration of the ITCZ (Marcottet al., 2014; Bauska et al., 2018). Rapid warming of high northern latitudes contributed to thaw permafrost, possibly liberating labile organic carbon to the atmosphere (Köhler et al.,2014; Crichton et al., 2016; Winterfeld et al., 2018; Meyer et al., 2019). Ocean surface pH reconstructions indicate that the ocean was oversaturated with respect to the atmosphere during the early, mid-LDT (Martínez-Botí et al., 2015b; Shao et al., 2019; Shuttleworth et al., 2021), suggesting that ocean sources at that time may have been larger than terrestrial sources. Over the course of the LDT, the decrease in Northern Hemisphere permafrost carbon stocks has been more than compensated by an increase in the carbon stocks of mineral soils, peatland and vegetation (Lindgren et al., 2018; Jeltsch-Thömmes et al., 2019). The land biosphere was, on average, a net sink for atmospheric carbon and accumulated several hundred Gt of carbon over the LDT. Detailed investigations reveal that Antarctic air temperatures, and more generally Southern Hemisphere (30°S–60°S) proxy temperature reconstructions, led the rise inpCO₂ at the onset of the LDT, 18 ka ago, by several hundred years (Shakun et al., 2012; Chowdhry Beeman et al., 2019). Atmospheric CO₂ led reconstructed global average temperature by several centuries (Shakun et al., 2012), corroborating the importance of CO₂ as an amplifier of orbitally driven warming. During the LDT, the phasing between Antarctic air temperature and atmospheric GHG concentration changes was nearly synchronous, yet variable, owing to the complex nature of the mechanisms modulating the global carbon cycle (Chowdhry Beeman et al., 2019). Mean ocean temperature reconstructions, based on noble gas extracted from Antarctic ice are closely correlated with Antarctic air temperature and pCO₂ records, emphasizing the role the Southern Ocean is playing in modulating global climate variability (Bereiter et al., 2018; Baggenstos et al., 2019).

Enhanced mid-ocean ridge magmatism and/or hydrothermal activity modulated by sea level rise has recently been hypothesized to have contributed to the deglacial CO₂ rise (Crowley et al., 2015; Lund et al., 2016; Huybers and Langmuir, 2017; Stott et al., 2019b). While geological carbon release may have affected the ocean’s radiocarbon budget (Ronge et al., 2016; Rafter et al., 2019; Stott et al., 2019a), model results suggest that the potential contribution of geological carbon sources to the atmosphere remained small (Roth and Joos, 2012; Hasenclever et al., 2017).

Simulations of Earth models of intermediate complexity (EMIC) with coupled glacial–interglacial climate and the carbon cycle were able to reproduce first-order changes in the atmospheric CO₂ content for the first time in recent years (Ganopolski and Brovkin, 2017; Khatiwala et al., 2019). The most important processes accounting for the full deglacial CO₂ amplitude in the models include solubility changes, changes in oceanic circulation and marine carbonate chemistry. The effect of the terrestrial carbon cycle, variable volcanic outgassing and the temperature dependence on the oceanic remineralization length scale contribute less than 15 ppm CO₂ between the glacial and interglacial intervals of the cycles. However, details in the simulated response of the marine carbon cycle and atmospheric CO₂ concentrations to changes in ocean circulation depend to a large degree on model parametrization (Gottschalk et al., 2019).

Independent paleoclimatic evidence suggests with high confidence that marine and terrestrial CH₄ and N₂O emissions are highly sensitive to climate on (sub)centennial time scales. Limited, yet internally consistent ice core measurements indicate with medium confidence that pulsed geologic CH₄ release from continental margins associated with warming remained negligible across the LDT. Multiple lines of evidence suggest with high confidence that CO₂ was released from the ocean interior on centennial time scales during the LDT in response to, or associated with warming, contributing to the transition out of the last glacial stage to the current interglacial period.

Multiple lines of evidence inferred from marine sediment proxies indicate with low to medium confidence that the millennial rates of CO₂ concentration change in the atmosphere during the last 56 Myr were at least four to five times lower than during the last century (Figure 5.3). In spite of uncertainties in ice core reconstructions related to delayed enclosure of air bubbles, which tend to smooth the records, there is high confidence that the rates of atmospheric CO₂ and CH₄ change during the last century were at least 10 and 5 times faster, respectively, than the maximum centennial growth rate averages of those gases during the last 800 kyr (Fig. 5.4).

Atmospheric GHG concentrations were much less variable during the pre-industrial Holocene (from 11.7 ka to 1750 CE). Atmospheric CH₄ concentrations decreased at the beginning of the Holocene, consistent with a general weakening of boreal sources (Yang et al., 2017; Beck et al., 2018) and further decline during the mid-Holocene owing to a reduction in Southern Hemisphere emissions concomitant with a southward shift of the ITCZ (Singarayer et al., 2011; Beck et al., 2018). Atmospheric CH₄ concentrations increased about 5 ka, which prompted the hypothesis of an early anthropogenic influence related to land-use changes in South East Asia (Ruddiman et al., 2016). However, stable isotope compositions on CH₄ extracted from Greenland and Antarctic ice (Beck et al., 2018) reveal that natural emissions located in the southern tropics were responsible for the rise in atmospheric CH₄ concentrations, in line with model simulations (Singarayer et al., 2011) thus disputing the early anthropogenic influence on the global CH₄ budget. Atmospheric N₂O concentrations increased slightly (20 ppb) across the Holocene, associated with a gradual decline in its nitrogen stable isotope composition (H. Fischer et al., 2019). The combined signal is consistent with a small increase in terrestrial emissions, offset by a reduction in marine emissions (Schilt et al., 2010b; Fischer et al., 2019).

The early Holocene decrease in CO₂ concentration by about 5 ppm (Schmitt et al., 2012) has been attributed to post-glacial regrowth in terrestrial biomass and a gradual increase in peat reservoirs over the Holocene, resulting in the sequestration of several hundred PgC (Yu et al., 2010; Nichols and Peteet, 2019). Peat accumulation rates in boreal and temperate regions were higher under warmer summer conditions in the early to mid-Holocene (Loisel et al., 2014; Stocker et al., 2017). The 20 ppm gradual increase of atmospheric CO₂ starting 7 ka has been attributed to a decrease in natural terrestrial biomass due to climate change, carbonate compensation and enhanced shallow water carbonate deposition (Menviel and Joos, 2012; Brovkin et al., 2016), consistent with stable carbon isotope measurements on CO₂ extracted from Antarctic ice (Elsig et al., 2009; Schmitt et al., 2012). These isotopic measurements do not support an early anthropogenic influence on atmospheric CO₂ due to land-use change and forest clearing (Ruddiman et al., 2016). Recent paleoceanographic evidence suggests that remineralized carbon outgassing associated with increased Southern Ocean circulation and upwelling (Studer et al., 2018), possibly promoted by stronger Southern Hemisphere westerly winds (Saunders et al., 2018), could have additionally contributed to the late Holocene increase in atmospheric CO₂ concentrations. However, the role of these mechanisms remained insignificant in transient Holocene ESM simulations (Brovkin et al., 2019). Overall, as in AR5 (WGI, Chapter 5), there is medium confidence in the key drivers of the CO₂ increase between the early Holocene and the beginning of the industrial era, yet there is low confidence in the relative contributions of these drivers due to insufficient quantitative constraints on particular processes.

There are two anthropogenic sources of carbon dioxide (CO₂): fossil emissions and net emissions (including removals) resulting from land-use change and land management (also shown in this chapter as LULUCF: land use, land-use change, and forestry; in previous IPCC reports it has been termed forestry and other land use, FOLU). Fossil CO₂ emissions include the combustion of the fossil fuels coal, oil and gas, covering all sectors of the economy (electricity, transport, industrial, and buildings), fossil carbonates such as in cement manufacturing, and other industrial processes such as the production of chemicals and fertilizers (Figure 5.5a). Fossil CO₂ emissions are estimated by combining economic activity data and emissions factors, with different levels of methodological complexity (tiers) or approaches (e.g., IPCC Guidelines for National Greenhouse Gas Inventories). Several organizations or groups provide estimates of fossil CO₂ emissions, with each dataset having slightly different system boundaries, methods, activity data, and emissions factors (Andrew, 2020). Datasets cover different time periods, which can dictate the datasets and methods that are used for a particular application. The data reported here is from an annually updated data source that combines multiple sources to maximise temporal coverage (Friedlingstein et al., 2020). The uncertainty in global fossil CO₂ emissions is estimated to be ±5% (1 standard deviation).

Figure 5.5 | Global anthropogenic CO₂ emissions. (a) Historical trends of anthropogenic CO₂ emissions (fossil fuels and net land-use change, including land management, called LULUCF flux in the main text) for the period 1870 to 2019, with ‘others’ representing flaring, emissions from carbonates during cement manufacture. Data sources: (Boden et al., 2017; IEA, 2017; Andrew, 2018; BP, 2018; Le Quéré et al., 2018a; Friedlingstein et al., 2020). (b) The net land-use change CO₂ flux (PgC yr^–1) as estimated by three bookkeeping models and 16 Dynamic Global Vegetation Models (DGVMs) for the global annual carbon budget 2019 (Friedlingstein et al., 2020). The three bookkeeping models are from Hansis et al., 2015; Houghton and Nassikas, 2017; Gasser et al., 2020 and are all updated to 2019. Their average is used to determine the net land-use change flux in the annual global carbon budget (black line). The DGVM estimates are the result of differencing a simulation with and without land-use changes run under observed historical climate and CO₂, following the Trendy v9 protocol (https://sites.exeter.ac.uk/trendy/protocol/); they are used to provide an uncertainty range to the bookkeeping estimates (Friedlingstein et al., 2020). All estimates are unsmoothed annual data. Estimates differ in process comprehensiveness of the models and in definition of flux components included in the net land use change flux. Further details on data sources and processing are available in the chapter data table (Table 5.SM.6).

Open figure

Fossil CO₂ emissions have grown continuously since the beginning of the industrial era (Figure 5.5) with short intermissions due to global economic crises or social instability (Peters et al., 2012; Friedlingstein et al., 2020). In the most recent decade (2010–2019), fossil CO₂ emissions reached an average 9.6 ± 0.5 PgC yr^–1 and were responsible for 86% of all anthropogenic CO₂ emissions. In 2019, fossil CO₂ emissions were estimated to be 9.9 ±0.5 PgC yr^–1excluding carbonation (Friedlingstein et al., 2020), the highest on record. These estimates exclude the cement carbonation sink of around 0.2 PgC yr^–1. Fossil CO₂ emissions grew at 0.9% yr^–1 in the 1990s, increasing to 3.0% yr^–1 in the 2000s, and reduced to 1.2% from 2010 to 2019. The slower growth in fossil CO₂ emissions in the last decade is due to a slowdown in growth from coal use. CO₂ emissions from coal use grew at 4.8% yr^–1 in the 2000s, but slowed to 0.4% yr^–1 in the 2010s. CO₂ emissions from oil use grew steadily at 1.1% yr^–1 in both the 2000s and 2010s. CO₂ emissions from gas use grew at 2.5% yr^–1 in the 2000s and 2.4% yr^–1 in 2010s, but has shown signs of accelerated growth of 3% yr^–1 since 2015 (Peters et al., 2020). Direct CO₂ emissions from carbonates in cement production are around 4% of total fossil CO₂ emissions, and grew at 5.8% yr^–1 in the 2000s but a slower 2.4% yr^–1 in the 2010s. The uptake of CO₂ in cement infrastructure (carbonation) offsets about one half of the carbonate emissions from current cement production (Friedlingstein et al., 2020). These results are robust across the different fossil CO₂ emissions datasets, despite minor differences in levels and rates, as expected given the reported uncertainties (Andrew, 2020). During 2020, the COVID-19 pandemic led to a rapid, temporary decline in fossil CO₂ emissions, estimated to be around 7% based on a synthesis of four estimates. (Cross-Chapter Box 6.1; Forster et al., 2020; Friedlingstein et al., 2020; Le Quéré et al., 2020; Liu et al., 2020).

The global net flux from land-use change and land management is composed of carbon fluxes from land-use conversions, land management and changes therein (Pongratz et al., 2018) and is equivalent to the LULUCF fluxes from the agriculture, forestry and other land use (AFOLU) sector (Jia et al., 2019). It consists of gross emissions (loss of biomass and soil carbon in clearing or logging, harvested product decay, emissions from peat drainage and burning, degradation) and gross removals (CO₂ uptake in natural vegetation regrowing after harvesting or agricultural abandonment, afforestation). The LULUCF flux relates to direct human interference with terrestrial vegetation, as opposed to the natural carbon fluxes occurring due to interannual variability or trends in environmental conditions (in particular, climate, CO₂, and nutrient deposition) (Houghton, 2013).

Progress since AR5 and SRCCL (IPCC, 2019a) allows more accurate estimates of gross and net fluxes due to the availability of more models, model advancement in terms of inclusiveness of land-use practices, and advanced land-use forcings (Ciais et al., 2013; Klein Goldewijk et al., 2017; Hurtt et al., 2020). In addition, important terminological discrepancies were resolved. First, synergistic effects of land-use change and environmental changes have been identified as a key reason for the large discrepancies between model estimates of the LULUCF flux, explaining up to 50% of differences (Pongratz et al., 2014; Stocker and Joos, 2015; Gasser et al., 2020). Another reason for discrepancies relates to natural fluxes being considered as part of the LULUCF flux when occurring on managed land in the United Nations Framework Convention on Climate Change (UNFCCC) national GHG inventories; these fluxes are considered part of the natural terrestrial sink in global vegetation models and excluded in bookkeeping models (Grassi et al., 2018). LULUCF fluxes following national GHG inventories or Food and Agriculture Organization of the United Nations (FAO) datasets, including recent estimates (Tubiello et al., 2021), are thus excluded from our global assessment, but their comparison against the academic approach is available elsewhere – at the global scale (Jia et al., 2019) and European level (Petrescu et al., 2020).

Land-use-related component fluxes can be verified by the growing databases of global satellite-based biomass observations in combination with information on remotely sensed land cover change. However, they differ from bookkeeping and modelling with Dynamic Global Vegetation Models (DGVMs) in excluding legacy emissions from pre-satellite-era land-use change and land management, and neglecting soil carbon changes, often focusing on gross deforestation, not regrowth (Jia et al., 2019).

For the decade 2010–2019, average emissions were estimated at 1.6 ± 0.7 PgC yr^–1 (mean ± standard deviation, 1 sigma; Friedlingstein et al., 2020). Alikely general upward trend since 1850 is reversed during the second part of the 20th century (Figure 5.5b). Trends since the 1980s have low confidence because they differ between estimates, which is related, among other things, to Houghton and Nassikas (2017) using a different land-use forcing than Hansis et al. (2015) and the DGVMs. Higher emissions estimates are expected from DGVMs run under transient environmental conditions compared to bookkeeping estimates, because the DGVM estimate includes the loss of additional sink capacity. Because the transient setup requires a reference simulation without land-use change to separate anthropogenic fluxes from natural land fluxes, LULUCF estimates by DGVMs include the sink forests that would have developed in response to environmental changes on areas that in reality have been cleared (Pongratz et al., 2014). The agricultural areas that replaced these forests have a reduced residence time of carbon, lacking woody material, and thus provide a substantially smaller additional sink over time (Gitz and Ciais, 2003). The loss of additional sink capacity is growing in particular with atmospheric CO₂ and increases DGVM-based LULUCF flux estimates relative to bookkeeping estimates over time (Figure 5.5).

Gross emissions are on average two to three times larger than the net flux from LULUCF, increasing from an average of 3.5 ± 1.2 PgC yr^–1 for the decade of the 1960s to an average of 4.4 ± 1.6 PgC yr^–1 during 2010–2019 (Friedlingstein et al., 2020). Gross removals partly balance these gross emissions to yield the net flux from LULUCF and increase from –2.0 ± 0.7 PgC yr^–1 for the 1960s to –2.9 ± 1.2 PgC yr^–1 during 2010–2019. These large gross fluxes show the relevance of land management, such as harvesting or rotational agriculture, and the large potential to reduce emissions by halting deforestation and degradation.

More evidence on the pre-industrial LULUCF flux has emerged since AR5 in the form of new estimates of cumulative carbon losses until today, and of a better understanding of natural carbon cycle processes over the Holocene (Ciais et al., 2013). Cumulative carbon losses by land-use activities since the start of agriculture and forestry (pre-industrial and industrial era) have been estimated at 116 PgC based on global compilations of carbon stocks for soils (Sanderman et al., 2017) with about 70 PgC of this occurring prior to 1750, and for vegetation as 447 PgC (inner quartiles of 42 calculations: 375–525 PgC) (Erb et al., 2018). Emissions prior to 1750 can be estimated by subtracting the post-1750 LULUCF flux from Table 5.1 from the combined soil and vegetation losses until today; they would then amount to 328 (161–501) PgC assuming error ranges are independent. A share of 353 (310–395) PgC from prior to 1800 has indirectly been suggested as the difference between net biosphere flux and terrestrial sink estimates, which is compatible with ice-core records due to a low airborne fraction of anthropogenic emissions in pre-industrial times (Erb et al., 2018; see also Section 5.1.2.3). Low confidence is assigned to pre-industrial emissions estimates.

Since AR5, evidence emerged that the LULUCF flux might have been underestimated as DGVMs include anthropogenic land cover change, but often ignore land management practices not associated with a change in land cover; land management is more widely captured by bookkeeping models through use of observation-based carbon densities (Ciais et al., 2013; Pongratz et al., 2018). Sensitivity studies show that practices such as wood and crop harvesting increase global net LULUCF emissions (Arneth et al., 2017) and explain about half of the cumulative loss in biomass (Erb et al., 2018).

Atmospheric CO₂ concentration measurements in remote locations began in 1957 at the South Pole Observatory (SPO) and in 1958 at Mauna Loa Observatory (MLO), Hawaii, USA (Keeling, 1960) (Figure 5.6a). Since then, measurements have been extended to multiple locations around the world (Bacastow et al., 1980; Conway et al., 1994; Nakazawa et al., 1997). In addition, high-density global observations of total column CO₂ measurements by dedicated GHG-observing satellites began in 2009 (Yoshida et al., 2013; O’Dell et al., 2018). Annual mean CO₂ growth rates are observed to be 1.56 ± 0.18 ppm yr^–1 (average and range from 1 standard deviation of annual values) over the 61 years of atmospheric measurements (1959–2019), with the rate of CO₂ accumulation almost tripling from an average of 0.82 ± 0.29 ppm yr^–1 during the decade of 1960–1969 to 2.39 ± 0.37 ppm yr^–1 during the decade of 2010–2019 (Chapter 2). The latter agrees well with that derived for total column (XCO₂) measurements by the Greenhouse Gases Observing Satellite (GOSAT; Figure 5.6b). The interannual oscillations in monthly mean CO₂ growth rates (Figure 5.6b) show a close relationship with the El Niño–Southern Oscillation (ENSO) cycle (Figure 5.6b) due to the ENSO-driven changes in terrestrial and ocean CO₂ sources and sinks on the Earth’s surface (Section 5.2.1.4).

Figure 5.6 | Time series of CO₂ concentrations and related measurements in ambient air. (a) Concentration time series and MLO-SPO difference, (b) growth rates, (c) ¹⁴C and¹³C isotopes, and (d) O₂/N₂ ratio. The data for Mauna Loa Observatory (MLO) and South Pole Observatory (SPO) are taken from the Scripps Institution of Oceanography (SIO)/University of California, San Diego (Keeling et al., 2001). The global mean CO₂ are taken from National Oceanic and Atmospheric Administration (NOAA) cooperative network (as in Chapter 2), and Greenhouse Gases Observing Satellite (GOSAT) monthly mean XCO₂ (mixing ratio) time series are taken from National Institute for Environmental Studies (Yoshida et al., 2013). CO₂ growth rates are calculated as the time derivative of deseasonalized time series (Nakazawa et al., 1997). The D(O₂/N2) are expressed in per meg units (= (FF/M) × 10⁶, where FF = moles of O₂ consumed by fossil-fuel burning, M = 3.706 × 10¹⁹, total number of O₂ molecules in the atmosphere (Keeling and Manning, 2014). The¹⁴CO₂ time series at Barring Head, Wellington, New Zealand (BHD) is taken from GNS Science and NIWA (Turnbull et al., 2017). The multivariate ENSO index (MEI) is shown as the shaded background in panel (b); (warmer shade indicates El Niño). Further details on data sources and processing are available in the chapter data table (Table 5.SM.6).

Open figure

Multiple lines of evidence unequivocally establish the dominant role of human activities in the growth of atmospheric CO₂. First, the systematic increase in the difference between the MLO and SPO records (Figure 5.6a) is caused primarily by the increase in emissions from fossil fuel combustion in industrialized regions that are situated predominantly in the Northern Hemisphere (Ciais et al., 2019). Second, measurements of the stable carbon isotope in the atmosphere (d¹³C–CO₂) are more negative over time because CO₂ from fossil fuels extracted from geological storage is depleted in¹³C (Figure 5.6c; Rubino et al., 2013; Keeling et al., 2017). Third, measurements of the d(O₂/N₂) ratio show a declining trend because for every molecule of carbon burned, 1.17 to 1.98 molecules of oxygen (O₂) is consumed (Figure 5.6d; Ishidoya et al., 2012; Keeling and Manning, 2014). These three lines of evidence confirm unambiguously that the atmospheric increase of CO₂ is due to an oxidative process (i.e., combustion). Fourth, measurements of radiocarbon (¹⁴C–CO₂) at sites around the world (Levin et al., 2010; Graven et al., 2017; Turnbull et al., 2017) show a continued long-term decrease in the¹⁴C/¹²C ratio. Fossil fuels are devoid of¹⁴C and therefore fossil fuel-derived CO₂ additions decrease the atmospheric¹⁴C/¹²C ratio (Suess, 1955).

Over the past six decades, the fraction of anthropogenic CO₂ emissions that has accumulated in the atmosphere (referred to as airborne fraction) has remained near constant at approximately 44% (Figure 5.7) (Ballantyne et al., 2012; Ciais et al., 2019; Gruber et al., 2019b; Friedlingstein et al., 2020). This suggests that the land and ocean CO₂ sinks have continued to grow at a rate consistent with the growth rate of anthropogenic CO₂ emissions, albeit with large interannual and sub-decadal variability dominated by the land sinks (Figure 5.7).

Figure 5.7 | Airborne fraction and anthropogenic (fossil fuel and land-use change) CO₂ emissions. Data as in Section 5.2.1.1. The multivariate El Niño–Southern Oscillation (ENSO) index (shaded) and the major volcanic eruptions are marked along the x-axis. Further details on data sources and processing are available in the chapter data table (Table 5.SM.6).

Open figure

Since AR5, an alternative observable diagnostic to the airborne fraction has been proposed to understand the trends in land and ocean sinks in response to its driving atmospheric CO₂ concentrations (Raupach et al., 2014; Bennedsen et al., 2019). It is the sink rate that is defined as the combined ocean and land sink flux per unit of atmospheric excess of CO₂ above pre-industrial levels (Raupach et al., 2014). The sink rate has declined over the past six decades, which indicates that the combined ocean and land sinks are not growing as fast as the growth in atmospheric CO₂ (Raupach et al., 2014; Bennedsen et al., 2019). Possible explanations for the sink rate decline are that the land and/or ocean CO₂ sinks are no longer responding linearly with CO₂ concentrations or that anthropogenic emissions are slower than exponential (Figure 5.7 and Sections 5.2.1.3 and 5.2.1.4; Gloor et al., 2010; Raupach et al., 2014; Bennedsen et al., 2019). In addition, both diagnostics are influenced by major climate modes (e.g., ENSO) and volcanic eruptions that contribute to high interannual variability (Gloor et al., 2010; Frölicher et al., 2013; Raupach et al., 2014), suggesting high sensitivity to future climate change. Uncertain land-use change fluxes (Section 5.2.1.2) influence the robustness of the trends. Based on the airborne fraction (AF), it is concluded with medium confidence that both ocean and land CO₂ sinks have grown consistent with the rising of anthropogenic emissions. Further research is needed to understand the drivers of changes in the CO₂ sink rate.

Since AR5 and SROCC, major advances in globally coordinated ocean CO₂ observations (Surface Ocean CO₂ Atlas, SOCAT; and Global Ocean Data Analysis Project, GLODAP), the harmonization of ocean and coastal-observation-based products, atmospheric and oceanic inversion models and forced global ocean biogeochemical models (GOBMs) have increased the level of confidence in the assessment of trends and variability of air–sea fluxes and storage of CO₂ in the ocean during the historical period (1960–2018; see also Supplementary Materials 5.SM.1; Ciais et al., 2013; Bakker et al., 2016; Landschützer et al., 2016, 2020; Bindoff et al., 2019; DeVries et al., 2019; Gregor et al., 2019; Gruber et al., 2019a, b; Tohjima et al., 2019; Friedlingstein et al., 2020; Hauck et al., 2020; Olsen et al., 2020). A major advance since SROCC is that, for the first time, all six published observational product fluxes used in this assessment, are made more comparable using a common ocean and sea ice cover area, integration of climatological coastal fluxes scaled to increasing atmospheric CO₂ and an ensemble mean of ocean fluxes calculated from three re-analysis wind products (Supplementary Materials 5.SM.2; Landschützer et al., 2014, 2020; Rödenbeck et al., 2014; Zeng et al., 2014; Denvil-Sommer et al., 2019; Gregor et al., 2019; Iida et al., 2021). From a process point of view, the ocean uptake of anthropogenic carbon is a two-step set of abiotic processes that involves the exchange of CO₂, first across the air–sea boundary into the surface mixed layer, followed by its transport into the ocean interior where it is stored for decades to millennia, depending on the depth of storage (Gruber et al., 2019b). Two definitions of air–sea fluxes of CO₂ are used in this assessment for both observational products and models: S_ocean is the global mean ocean CO₂ sink and F_net denotes the net spatially varying CO₂ fluxes (Hauck et al., 2020). Adjustment of the mean global F_net for the pre-industrial sea-to-air CO₂ flux associated with land-to-ocean carbon flux term makes F_net comparable to S_ocean (Jacobson et al., 2007; Resplandy et al., 2018; Hauck et al., 2020).

There are multiple lines of observational and modelling evidence that support with high confidence the finding that, in the historical period (1960–2018), air–sea fluxes and storage of anthropogenic CO₂ are largely influenced by atmospheric CO₂ concentrations, physical ocean processes and physicochemical carbonate chemistry, which determines the unique properties of CO₂ in seawater (Chapter 9 and Cross-Chapter Box 5.3; Wanninkhof, 2014; DeVries et al., 2017; McKinley et al., 2017, 2020, Gruber et al., 2019a, b; Hauck et al., 2020). Here we assess three different approaches (Figures 5.8a,b and 5.9) that together provide high confidence that, during the historical period (1960–2018), the ocean carbon sink (S_ocean) and its associated ocean carbon storage have grown in response to global anthropogenic CO₂ emissions (Gruber et al., 2019a; Hauck et al., 2020; McKinley et al., 2020).

5.2.1.3.1 Ocean carbon fluxes and storage: Global multi-decadal trends

In the first assessment approach, the mean global multi-decadal (1960–2019) trends in the ocean sink (S_ocean) for CO₂ show a high degree of coherence across the nine GOBMs and sixpCO₂ -based observational product reconstructions (1987–2018) which, despite a temporary slowdown (or ‘hiatus’) in the 1990s, is also quasi-linear over that period (Figure 5.8a; Gregor et al., 2019; Hauck et al., 2020). This coherence between the GOBMs and observations-based reconstructions (1987–2018; r²=0.85) provides high confidence that the ocean sink (S_ocean in Section 5.2.1.5) evaluated from GOBMs (1960–2019) grew quasi-linearly from 1.0 ± 0.3 PgC yr^–1 to 2.5 ± 0.6 PgC yr^–1 between the decades 1960–1969 and 2010–2019 in response to global CO₂ emissions (Figure 5.8a; Table 5.1; Friedlingstein et al., 2020; Hauck et al., 2020). The cumulative ocean CO₂ uptake (105 ± 20 PgC) is 23% of total anthropogenic CO₂ emissions (450 ± 50 PgC) for the same period (Friedlingstein et al., 2020). Notwithstanding the high confidence in the magnitude of the annual to decadal trends for S_ocean, this assessment is moderated to mediumconfidence by the low confidence in the currently inadequately constrained uncertainties in the pre-industrial land-to-ocean carbon flux, the uncertain magnitude of winter outgassing from the Southern Ocean, and the uncertain effect of the ocean surface cool-skin, the effect of data sparsity, differences between wind products and the uncertain contribution from the changing land–ocean continuum on global and regional fluxes (Jacobson et al., 2007; Resplandy et al., 2018; Roobaert et al., 2018; Bushinsky et al., 2019; Hauck et al., 2020; Watson et al., 2020; Gloege et al., 2021). However, both GOBMs and pCO₂ -based observational products independently reveal a slowdown or ‘hiatus’ of the ocean sink in the 1990s, which provides a valuable constraint for model verification and leads to greater confidence in the model outputs (Figure 5.8a; Landschützer et al., 2016; Gregor et al., 2018; DeVries et al., 2019; Hauck et al., 2020). A number of studies point to the role of the Southern Ocean in the global ‘1990s hiatus’ in air–sea CO₂ fluxes, but provide different process-based explanations linking ocean temperature, mixing and meridional overturning circulation (MOC) responses to variability in large-scale climate systems, wind stress and volcanic activity, as well as the sensitivity of the air–sea CO₂ flux to small changes in the atmospheric forcing from anthropogenic CO₂ (Landschützer et al., 2016; DeVries et al., 2017; Bronselaer et al., 2018; Gregor et al., 2018; Gruber et al., 2019a; Keppler and Landschützer, 2019; McKinley et al., 2020; Nevison et al., 2020). Data sparsity in the Southern Ocean could also be a factor amplifying the global decadal perturbation of the 1990s (Gloege et al., 2021). Therefore, while there is high confidence in the 1990s hiatus of the global ocean sink for anthropogenic CO_2, and that the Southern Ocean makes an observable contribution to it, there is still low confidence in the attribution for the processes behind the 1990s hiatus (Section 5.2.1.3.2). Observed increases in the amplitude of the seasonal cycle of oceanpCO₂ and reductions in the mean global buffering capacity provide high confidence that the growing CO₂ sink is also beginning to drive observable large-scale changes in ocean carbonate chemistry (Jiang et al., 2019). However, there is medium confidence that these changes which, depending on the emissions scenario, could drive future ocean feedbacks, are still too small to emerge from the historical multi-decadal observed growth rate of S_ocean (Sections 5.1.2; 5.3.2 and 5.4.2, and Figure 5.8a; SROCC (Section 5.2.2.3.2; Bates et al., 2014; Sutton et al., 2016; Fassbender et al., 2017; Landschützer et al., 2018; Jiang et al., 2019). A recent model-based study suggests that re-emergence of previously stored anthropogenic CO₂ is changing the buffering capacity of the mixed layer and reducing the ocean sink for anthropogenic CO₂ during the historical period (Rodgers et al., 2020). This trend is not reflected in observations-based products (Figure 5.8a), so we attribute a low confidence.

Figure 5.8 | Multi-decadal trends for the ocean sink of CO₂ . (a) The multi-decadal (1960–2019) trends in the annual ocean sink (S_ocean) reconstructed from nine Global Ocean Biogeochemical Models (GOBM) forced with atmospheric re-analysis products (Hauck et al., 2020), six observationally based gap-filling products that reconstructed spatial and temporal variability in the ocean CO₂ flux from sparse observations of surface oceanpCO₂ (Supplementary Materials 5.SM.2). The trends in S_ocean were calculated from the mean annual GOBM outputs, and the observational products were used to provide confidence in the GOBM assessments (r²=0.85). Thick lines represent the multi-model mean. Observationally based products have been corrected for pre-industrial river carbon fluxes (0.62 PgC yr^–1) based on the average of estimates from Jacobson et al. (2007) and Resplandy et al. (2018). (b) Mean decadal constraints and their confidence intervals for global ocean sink (S_ocean) of anthropogenic CO₂ using multiple independent or quasi-independent lines of evidence or methods for the period 1990–2019 (see Supplementary Materials Tables 5.SM.1 and 5.SM.2 for magnitudes, uncertainties and published sources). Further details on data sources and processing are available in the chapter data table (Table 5.SM.6).

Open figure

The second assessment approach makes use of six independent methods to constrain the mean decadal ocean sink over the period 1990–2019 (Figure 5.8b). This provides a multi-decadal advance on the 1990–1999 decadal constraint from (Denman et al., 2007) that has been widely used as a model constraint for GOBMs used for the global carbon budget (Hauck et al., 2020). The medium confidence attributed by this assessment of the global multi-decadal trend (Figure 5.8a) is further supported by the broad agreement in magnitude and trend of the decadal mean ocean CO₂ uptake with assessments that also include additional observations-based, independent methods such as ocean CO₂ inversion and atmospheric CO₂ and O₂/N₂ measurements (Figure 5.8b; Supplementary Materials Tables 5.SM.1 and 5.SM.2).

Here we provide a third comparative assessment approach depicting the spatial coherence of ocean air–sea fluxes and storage rates of CO₂ as well as a quantitative assessment of both fluxes for the same period (1994–2007; Figure 5.9). Observation-based pCO₂ flux products show that emissions of natural CO₂ occur mostly in the tropics and high-latitude Southern Ocean, and that the uptake and storage of anthropogenic CO₂ occurs predominantly in the mid-latitudes (Chapter 9, Figure 5.9 and Cross-Chapter Box 5.3). Strong ocean CO₂ sink regions are those in the mid-latitudes associated with the cooling of poleward flowing subtropical surface waters as well as equatorward flowing sub-polar surface waters, both of which contribute to the formation of Mode, Intermediate and Deep water masses that transport anthropogenic CO₂ into the ocean interior on time scales of decades to centuries in both hemispheres (Section 9.2.2.3 and Figure 5.9; DeVries, 2014; Gruber et al., 2019b; Wu et al., 2019). The mean decadal scale magnitude and uncertainties of S_ocean from net air sea fluxes (F_net ) were calculated from an ensemble of six observational-based product reconstructions (Figure 5.9a) and the storage rates in the ocean interior derived from multiple ocean interior CO₂ datasets (Gruber et al., 2019b; Figure 5.9b). The cumulative CO₂ stored in the ocean interior from 1800 to 2007 has been estimated at 140 ±18 PgC (Gruber et al., 2019b). As reported in SROCC (Section 5.2.2.3.1; IPCC, 2019b), the net ocean CO₂ storage between 1994–2007 was 29 ± 4 PgC, which corresponds to a mean storage of 26 ± 5% of anthropogenic CO₂ emissions for that period (Gruber et al., 2019b). The resulting net annual storage rate of anthropogenic CO₂, equivalent to S_ocean for the period mid-1994 to mid-2007 is 2.2 ± 0.3 PgC yr^–1, which is in very close agreement with the top-down air–sea flux estimate of S_ocean of 2.1 ± 0.5 PgC yr^–1 from GOBMs and 1.9 ± 0.3PgC yr^–1 frompCO₂ -based observational products with the steady river carbon flux correction of 0.62 PgC yr^–1 for the same time period (Gruber et al., 2019b; Hauck et al., 2020). This close agreement between these independent ocean CO₂ sink estimates derived from air–sea fluxes and storage rates in the ocean interior support the medium confidence assessment that the ocean anthropogenic carbon storage rates continue to be determined by the ocean sink (S_ocean) in response to growing CO₂ emissions (Figure 5.9; McKinley et al., 2020).

Figure 5.9 | Comparative regional characteristics of the mean decadal (1994–2007) sea-air CO₂ flux (Fnet) and ocean storage of anthropogenic CO₂ . (a) Regional source–sink characteristics for contemporary ocean air–sea CO₂ fluxes (F_net ) derived from the ensemble of six observation-based products using Surface Ocean CO₂ (Atlas (SOCAT)v6 observational dataset (Landschützer et al., 2014; Rödenbeck et al., 2014; Zeng et al., 2014; Bakker et al., 2016; Denvil-Sommer et al., 2019; Gregor et al., 2019; Iida et al., 2021). Warm colours depict outgassing fluxes and black contours characterize the super-biomes defined from Fay and McKinley (2014) and adjusted by Gregor et al. (2019) also used to calculate the variability in regional flux anomalies (Supplementary Materials Figure 5.SM.1); (b) The regional characteristics of the storage fluxes of CO₂ in the ocean interior for the same period (Gruber et al., 2019b). The dots reflect ocean areas where the 1-sigma standard deviation of Fnet from the six observational-based product reconstructions is larger than the magnitude of the mean. This reflects source–sink transition areas where the mean Fnet is small and more strongly influenced by spatial and temporal variability across the products. Further details on data sources and processing are available in the chapter data table (Table 5.SM.6).

Open figure

Since the start of direct measurements of CH₄ in the atmosphere in the 1970s (Figure 5.13), the highest growth rate was observed from 1977 to 1986 at 18 ± 4 ppb yr^–1 (multi-year mean and 1 standard deviation) (Rice et al., 2016). This rapid CH₄ growth followed the green revolution with increased crop production and a fast rate of industrialization that caused rapid increases in CH₄ emissions from ruminant animals, rice cultivation, landfills, oil and gas industry and coal mining (Ferretti et al., 2005; Ghosh et al., 2015; Crippa et al., 2020). Due to increases in oil prices in the early 1980s, emissions from gas flaring declined significantly (Stern and Kaufmann, 1996). This explains the first reduction in CH₄ growth rates from 1985 to 1990 (Steele et al., 1992; Chandra et al., 2021). Further emissions reductions occurred following the Mt Pinatubo eruption in 1991 that triggered a reduction in CH₄ growth rate through a decrease in wetland emissions driven by lower surface temperatures due to the light scattering by aerosols (Bândă et al., 2016; Chandra et al., 2021). In the late 1990s through to 2006 there was a temporary pause in the CH₄ growth rate, with higher confidence on its causes than in AR5: emissions from the oil and gas sectors declined by about 10 Tg yr^–1through the 1990s, and atmospheric CH₄ loss steadily increased (Dlugokencky et al., 2003; Simpson et al., 2012; Crippa et al., 2020; Höglund-Isaksson et al., 2020; Chandra et al., 2021). The methane growth rate began to increase again at 7 ± 3 ppb yr^–1 during 2007–2016, the causes of which are highly debated since AR5 (Rigby et al., 2008; Dlugokencky et al., 2011; Dalsøren et al., 2016; Nisbet et al., 2016; Patra et al., 2016; Schaefer et al., 2016; Schwietzke et al., 2016; Turner et al., 2017; Worden et al., 2017; He et al., 2020); studies disagree on the relative contribution of thermogenic, pyrogenic and biogenic emission processes and variability in tropospheric OH concentration. The renewed CH₄ increase is accompanied by a reversal of d¹³C trend to more negative values post 2007; opposite to what occurred in the 200 years prior (Ferretti et al., 2005; Ghosh et al., 2015; Schaefer et al., 2016; Schwietzke et al., 2016; Nisbet et al., 2019), suggesting an increasing contribution from animal farming, landfills and waste, and a slower increase in emissions from fossil fuel exploitation since the early 2000s (Patra et al., 2016; Jackson et al., 2020; Chandra et al., 2021). Atmospheric concentrations of CH₄ reached 1866.3 ppb in 2019 (Figure 5.14). A comprehensive assessment of the CH₄ growth rates over the past four decades is presented in Cross-Chapter Box 5.2.

Figure 5.13 | Time series of CH₄ concentrations, growth rates and isotopic composition. (a) CH₄ concentrations; (b) CH₄ growth rates; (c) d¹³-CH₄. Data from selected site networks operated by the National Oceanic and Atmospheric Administration (NOAA; Dlugokencky et al., 2003), Advanced Global Atmospheric Gases Experiment (AGAGE; Prinn et al., 2018) and Portland Airport (PDX, Portland State University; Rice et al., 2016). To maintain clarity, data from many other measurement networks are not included here, and all measurements are shown in the World Metereological Organization X2004ACH₄ global calibration standard. Global mean values of XCH₄ (total-column), retrieved from radiation spectra measured by the Greenhouse Gases Observing Satellite (GOSAT) are shown in panels (a) and (b). Cape Grim Observatory (CGO; 41°S, 145°E) and Trinidad Head (THD; 41°N, 124°W) data are taken from the AGAGE network. NOAA global and northern hemispheric (NH) means for d¹³C are calculated from 10 and 6 sites, respectively. The PDX data adjusted to NH (period: 1977–2000) are merged with THD (period: 2001–2019) for CH₄ concentration and growth rate analysis, and PDX and NOAA NH means of d¹³C data are used for joint interpretation of long-term trends analysis. The multivariate El Niño–Southern Oscillation (ENSO) index (MEI) is shown in panel (b). Further details on data sources and processing are available in the chapter data table (Table 5.SM.6).

Open figure

The positive gradient between CH₄ at Cape Grim, Australia (41°S) and Trinidad Head, USA (41°N), and the bigger difference between Trinidad Head and global mean CH₄ compared to that between global mean CH₄ and Cape Grim, strongly suggest that the Northern Hemisphere is the dominant origin of anthropogenic CH₄ emissions (Figure 5.13). The loss rate of CH₄ in troposphere does not produce a large positive north–south hemispheric gradient in CH₄ due to parity in hemispheric mean OH concentration (Patra et al., 2014), or in the case of greater OH concentrations in the northern rather than the Southern Hemisphere as simulated by the chemistry-climate models (Naik et al., 2013). Coal mining contributed about 35% of the total CH₄ emissions from all fossil fuel-related sources. Top-down estimates of fossil fuel emissions (106 Tg yr^–1) are smaller than bottom-up estimates (115 Tg yr^–1) during 2008–2017 (Table 5.2). Inventory-based estimates suggest that CH₄ emissions from coal mining increased by 17 Tg yr^–1 between the periods 2002–2006 and 2008–2012, with a dominant contribution from China (Peng et al., 2016; Crippa et al., 2020; Höglund-Isaksson et al., 2020). Inventory-based estimates suggest that CH₄ emissions from coal mining increased by 17 Tg yr^–1 between the periods 2002–2006 and 2008–2012, with a dominant contribution from China (Peng et al., 2016; Crippa et al., 2020; Höglund-Isaksson et al., 2020). Recent country statistics and detailed inventory-based estimates show that CH₄ emissions from coal mining in China declined between 2012 and 2016 (Sheng et al., 2019; Gao et al., 2020), while atmospheric-based estimates suggest a continuation of CH₄ emissions growth but at a slower rate to the year 2015 (Miller et al., 2019) and 2016 (Chandra et al., 2021). Emissions from oil and gas extraction and use decreased in the 1980s and 1990s, but increased in the 2000s and 2010s (Dlugokencky et al., 1994; Stern and Kaufmann, 1996; Howarth, 2019; Crippa et al., 2020). The attribution to multiple CH₄ sources using spatially aggregated atmospheric d¹³C data remained underdetermined to infer the global total emissions from the fossil fuel industry, biomass burning and agriculture (Rice et al., 2016; Schaefer et al., 2016; Schwietzke et al., 2016; Worden et al., 2017; Thompson et al., 2018).

In the agriculture and waste sectors (Table 5.2), livestock production has the largest emissions source (109 Tg yr^–1 in 2008–2017) dominated by enteric fermentation by about 90%. Methane is formed during the storage of manure, when anoxic conditions are developed (Hristov et al., 2013). Emissions from enteric fermentation and manure have increased gradually from about 87 Tg yr^–1 in 1990–1999 to 109 Tg yr^–1 in 2008–2017 mainly due to the increase in global total animal numbers. Methane production in livestock rumens (cattle, goats, sheep, water buffalo) are affected by the type, amount and quality of feeds, energy consumption, animal size, health and growth rate, meat and milk production rate, and temperature (Broucek, 2014; S.R.O. Williams et al., 2020; SRCCLSection 5.4.3). Waste management and landfills produced 64 Tg yr^–1 in 2008–2017, with global emissions increasing steadily since the 1970s and, despite significant declines in the USA, western Europe and Japan (Crippa et al., 2020; Höglund-Isaksson et al., 2020).

Emissions from rice cultivation decreased from about 45 Tg yr^–1 in the 1980s to about 29 Tg yr^–1 in the decade 2000–2009, but increased again slightly to 31 Tg yr^–1 during 2008–2017, based on inventories data. However, ecosystem models showed a gradual increase with time due to climate change (limited evidence, low agreement) (Crippa et al., 2020; Höglund-Isaksson et al., 2020; Ito, 2020).

Biomass burning and biofuel consumption (including natural and anthropogenic processes) caused at least 30 Tg yr^–1emissions during 2008–2017 and constituted up to about 5% of global anthropogenic CH₄ emissions. Methane emissions from open biomass burning decreased during the past two decades mainly due to reduction of burning in savanna, grassland and shrubland (van der Werf et al., 2017; Worden et al., 2017). There is recent evidence from the tropics that fire occurrence is non-linearly related to precipitation, implying that severe droughts will increase CH₄ emissions from fires, particularly from the degraded peatlands (Field et al., 2016).

Freshwater wetlands are the single largest global natural source of CH₄ in the atmosphere, accounting for about 26% of the total CH₄ source (robust evidence, medium agreement). Progress has been made since AR5 (Ciais et al., 2013) in better constraining freshwater lake and river emissions and reducing double counting with wetland emissions. Bottom-up and top-down estimates for 2008–2017 are 149 and 180 Tg yr^–1, respectively, with a top-down uncertainty range of 159–199 Tg yr^–1 (Table 5.2). The large uncertainties stem from challenges in mapping wetland area and temporal dynamics to landscape estimates, and in scaling methane production, transport and consumption processes that are measured with small chambers or flux towers (Pham-Duc et al., 2017). Both the top-down and bottom-up estimates presented in Table 5.2 indicate little increase in wetland CH₄ emissions during the last three decades, with the new estimates being slightly smaller than in AR5 due to updated wetland maps and ecosystem model simulations (Melton et al., 2013; Poulter et al., 2017). Wetland emissions show strong interannual variability due to the changes in inundated land area, air temperature and microbial activity (Bridgham et al., 2013). Present terrestrial ecosystem model simulated CH₄ emissions variability does not produce strong correlation with the El Niño–Southern Oscillation (ENSO) cycle (Cross-Chapter Box 5.2, Figure 2), although observation evidence is emerging for lower CH₄ emissions during El Niños and greater emissions during La Niña (Pandey et al., 2017).

Trees in upland and wetland forests contribute to CH₄ emissions by abiotic production in the canopy, by the methanogenesis taking place in the stem, and by conducting CH₄ from soil into the atmosphere (Covey and Megonigal, 2019). There is emerging evidence of the important role of trees in transporting and conducting CH₄ from soils into the atmosphere, especially in tropics (Pangala et al., 2017), whereas direct production of CH₄₄ by vegetation only has a minor contribution (limited evidence, high agreement) (Bruhn et al., 2012; Covey and Megonigal, 2019). The contribution of trees in transporting CH₄ may further widen the gap between the bottom-up and top-down estimates in the global budget, particularly needing a re-assessment of emissions in the tropics and in forested wetlands of temperate and boreal regions (Pangala et al., 2017; Jeffrey et al., 2019; Welch et al., 2019; Sjögersten et al., 2020).

Microbial methane uptake by soil comprises up to 5% (30 Tg yr^–1) of the total CH₄ sink in 2008–2017 (Table 5.2). There is evidence from experimental and modelling studies of increasing soil microbial uptake due to increasing temperature (Yu et al., 2017), although evidence also exists for decreasing CH₄ consumption, possibly linked to precipitation changes (Ni and Groffman, 2018). The estimate of global methane loss by microbial oxidation in upland soils has been lowered marginally by 4 Tg yr^–1, compared to 34 Tg yr^–1 in AR5, for the period 2000–2009. Termites, an infraorder of insects (Isoptera) found in almost all land masses, emitted about 9 Tg yr^–1 of CH₄ in 2000–2009. Increased emissions from insects and other anthropods are projected (Brune, 2018).

In AR5, the ocean CH₄ emissions were reported together with geological emissions, summing up to 54 (33–75) Tg yr^–1. Coastal oceans, fjords and mud volcanos are major sources of CH₄ in the marine environment, but CH₄ flux measurements are sparse. Saunois et al. (2020) estimate that the oceanic budget, including biogenic, geological and hydrate emissions from coastal and open ocean, is 6 (range 4–10) Tg yr^–1 for the 2000s, which is in good agreement with an air–sea flux measurement-based estimate of 6–12 Tg yr^–1 (Weber et al., 2019). When estuaries are included, the total oceanic budget is 9–22 Tg yr^–1, with a mean value of 13 Tg yr^–1. A recent synthesis suggests that CH₄ emissions from shallow coastal ecosystems, particularly from mangroves, can be as high as 5–6 Tg yr^–1 (Al-Haj and Fulweiler, 2020). The reservoir emissions, including coastal wetlands and tidal flats, contribute up to 13 Tg yr^–1 (Borges and Abril, 2011; Deemer et al., 2016). Methane seepage from the Arctic shelf, possibly triggered by the loss of geological storage due to warming and thawing of permafrost and hydrate decomposition, has a wide estimated range of 0.0–17 Tg yr^–1 (Shakhova et al., 2010, 2014, 2017; Berchet et al., 2016); advanced eddy covariance measurements put the best estimate at about 3 Tg yr^–1 from the East Siberian Arctic shelf (Thornton et al., 2020). The current flux is expected to be a mix of pre-industrial and climate change-driven fluxes, CH₄ seepage is anticipated to increase in a warmer world (Dean et al., 2018).

All geological sources around the world, including the coastal oceans and fjords, are estimated to emit CH₄ in the range of 35–76 Tg yr^–1 (Etiope et al., 2019). There is evidence that the ventilation of geological CH₄ is likely to be smaller than 15 Tg yr^–1 (Petrenko et al., 2017; Hmiel et al., 2020). A lower geological CH₄ ventilation will reduce the gap between bottom-up and top-down estimates (Table 5.2), but widen the gap in the ratio of fossil fuel-derived sources to the biogenic sources for matching the D¹⁴C-CH₄ observations.

Inland water (lakes, rivers, streams, ponds, estuaries) emissions are proportionally the largest source of uncertainty in the CH₄ budget. Since AR5 (Ciais et al., 2013), the inland water CH₄ source has been revised from 8–73 Tg yr^–1 (1980s) to 117–212 Tg yr^–1 (2000s) with the availability of more observational data and improved areal estimates (Bastviken et al., 2011; Deemer et al., 2016; Stanley et al., 2016; DelSontro et al., 2018; Saunois et al., 2020). However, it is difficult to estimate bottom-up CH₄ emissions, due to the large spatial and temporal variation in lake and river CH₄ fluxes (Wik et al., 2016; Crawford et al., 2017; Natchimuthu et al., 2017), uncertainties in their global area (Allen and Pavelsky, 2018), a relatively small number of observations, and varying measurement methods – for example, those neglecting ebullition, varying upscaling methods, and lack of appropriate processes (Sanches et al., 2019; Engram et al., 2020; L. Zhang et al., 2020). Accordingly, there is no clear accounting of inland waters in top-down budgets, which is the main reason for the large gap in bottom-up and top-down estimates of ‘other sources’ in the CH₄ budget (Table 5.2). Despite recent progress in separating wetlands from inland waters, there is double-counting in the bottom-up estimates of their emissions (Thornton et al., 2016a). Although there is evidence that regional human activities and global warming both increase inland water CH₄ emissions (Beaulieu et al., 2019), the increase in the decadal emissions since AR5 (Ciais et al., 2013) rather reflect improvements in the estimate (medium confidence), due to updates in the datasets and new upscaling approaches (Saunois et al., 2020).

A summary of top-down and bottom-up estimates of CH₄ emissions and sinks for the period 2008–2017 is presented in Figure 5.14 (details in Table 5.2 and the associated text for the emissions). In addition to 483-682 Tg yr^–1loss of CH₄ in the troposphere by reaction with OH, 1–35 Tg yr^–1 of CH₄ loss is estimated to occur in the lower troposphere due to Cl but are not included in the top-down models as shown in Table 5.2 (Hossaini et al., 2016; Gromov et al., 2018; X. Wang et al., 2019). The decadal mean CH₄₄ burden/imbalance increased at the rate of 30, 12, 7 and 21 Tg yr^–1 in the 1980s (1980–1989), 1990s (1990–1999), 2000s (2000–2009) and the most recent decade (2008–2017), respectively (virtually certain), as can be estimated from observed atmospheric growth rate (Cross-Chapter Box 5.2, Figure 1).

Recent analysis using D¹⁴C-CH₄ in ice samples suggest that CH₄ emissions from fossil fuel exploitation are responsible for 30% of total CH₄ emissions (Lassey et al., 2007; Hmiel et al., 2020), which is largely inconsistent with sectorial budgets where fossil fuel emissions add up to 20% only (Ciais et al., 2013). However, recent model simulations produce fairly consistent d¹³C-CH₄ values and trends, as observed in the atmospheric samples using 20% fossil fuel emissions fraction (Ghosh et al., 2015; Warwick et al., 2016; Fujita et al., 2020; Strode et al., 2020). Further research is needed to clarify the relative roles of CH₄ emissions from fossil fuel exploitation and freshwater components. A key challenge is to accommodate the higher estimated emissions from these two components without a major increase in the sinks, in order to be consistent with the observed changes in the carbon and hydrogen isotopes.

Figure 5.14 | Global methane (CH₄ ) budget (2008–2017). Values and data sources as in Table 5.2 (in TgCH₄). The atmospheric stock is calculated from mean CH₄ concentration, multiplying a factor of 2.75 ± 0.015 Tg ppb^–1, which accounts for the uncertainties in global mean CH₄ (Chandra et al., 2021). Further details on data sources and processing are available in the chapter data table (Table 5.SM.6).

Open figure

The tropospheric abundance of N₂O was 332.1 ± 0.4 ppb in 2019 (Figure 5.15), which is 23% higher than pre-industrial levels of 270.1 ± 6.0 ppb (robust evidence, high agreement). Current estimates are based on atmospheric measurements with high accuracy and density (Francey et al., 2003; Elkins et al., 2018; Prinn et al., 2018), and pre-industrial estimates are based on multiple ice-core records Section 2.2.3.2.3). The average annual tropospheric growth rate was 0.85 ± 0.03 ppb yr^–1 during the period 1995 to 2019 (Figure 5.15a). The atmospheric growth rate increased by about 20% between the decade 2000–2009 and the most recent decade of 2010–2019 (0.95 ± 0.04 ppb yr^–1) (robust evidence, high agreement). The growth rate in 2010–2019 was also higher than during 1970–2000 (0.6–0.8 ppb yr^–1; Ishijima et al., 2007) and the 30-year period prior to 2011 (0.73 ± 0.03 ppb yr^–1), as reported by AR5. New evidence since AR5 (WGI, Section 6.4.3) confirms that, in the tropics and subtropics, large interannual variations in the atmospheric growth rate are negatively correlated with the multivariate ENSO index (MEI) and associated anomalies in land and ocean fluxes (Ji et al., 2019; Thompson et al., 2019; S. Yang et al., 2020) (Figure 5.15a).

Figure 5.15 | Changes in atmospheric nitrous oxide (N₂ O) and its isotopic composit ion since 1940. (a) Atmospheric N₂O abundance (parts per billion, ppb) and growth rat (ppb yr^–1); (b) δ¹⁵N of atmospheric N₂O; and (c) alpha-site¹⁵N–N₂O. Estimates are based on direct atmospheric measurements in the Advanced Global Atmospheric Gases Experiment (AGAGE), Commonwealth Scientific and Industrial Research Organisation (CSIRO), and National Oceanic and Atmospheric Administration (NOAA) networks (Prinn et al., 2000, 2018; Francey et al., 2003; Hall et al., 2007; Elkins et al., 2018), archived air samples from Cape Grim, Australia (Park et al., 2012), and firn air from the North Greenland Ice Core Project (NGRIP) Greenland and H72 Antarctica (Ishijima et al., 2007), Law Dome Antarctica (Park et al., 2012), as well as a collection of firn ice samples from Greenland (Prokopiou et al., 2017, 2018). Shading in (a) is based on the multivariate El Niño–Southern Oscillation (ENSO) index, with red indicating El Niño conditions (Wolter and Timlin, 1998). Further details on data sources and processing are available in the chapter data table (Table 5.SM.6).

Open figure

As assessed by SRCCL (IPCC, 2019a), combined firn, ice, air and atmospheric measurements show that the¹⁵N/¹⁴N isotope ratio (robust evidence, high agreement) and the predominant position of the¹⁵N atom in atmospheric N₂O (limited evidence, low agreement) in N₂O has changed since 1940 (Figure 5.15b, c) whereas they were relatively constant in the pre-industrial period (Ishijima et al., 2007; Park et al., 2012; Prokopiou et al., 2017, 2018). The SRCCL concluded that this change indicates a shift in the nitrogen-substrate available for denitrification, and the relative contribution of nitrification to the global N₂O source (robust evidence, high agreement), which are associated with increased fertilizer use in agriculture (Park et al., 2012; Snider et al., 2015; Prokopiou et al., 2018).

Since AR5 (WGI, Section 6.4.3), the mean atmospheric lifetime of N₂O has been revised to 116 ± 9 years (Prather et al., 2015). The small negative feedback of the N₂O lifetime to increasing atmospheric N₂O results in a slightly lower residence time (109 ± 10 years) of N₂O perturbations compared with that assessed by AR5 (118–131 years) (Prather et al., 2015). The dominant N₂O loss occurs through photolysis and oxidation by O(1D) radicals in the stratosphere and amounts to approximately 13.1 (12.4–13.6) TgN yr^–1 (Minschwaner et al., 1993; Prather et al., 2015; Tian et al., 2020).

The AR5 (WGI, Section 6.4.3) and SRCCL (Section 2.3.3) concluded that agriculture is the largest anthropogenic source of N₂O emissions. Since SRCCL (2.3.3), a new synthesis of inventory-based and modelling studies shows that the widespread use of synthetic fertilizers and manure on cropland and pasture, manure management and aquaculture resulted in 3.8 (2.5–5.8) TgN yr^–1 (average 2007–2016) (robust evidence, high agreement) (Table 5.3; Winiwarter et al., 2018; FAO, 2019; Janssens-Maenhout et al., 2019; Tian et al., 2020). Observations from field-measurements (Song et al., 2018), inventories (Wang et al., 2020) and atmospheric inversions (Thompson et al., 2019) further corroborate the assessment of SRCCL that there is a non-linear relationship between N₂O emissions and nitrogen input, implying an increasing fraction of fertilizer lost as N₂O with larger fertilizer excess (medium evidence, high agreement). Several studies using complementary methods indicate that agricultural N₂O emissions have increased by more than 45% since the 1980s (high confidence) (Figure 5.16 and Table 5.3; Davidson, 2009; Winiwarter et al., 2018; Janssens-Maenhout et al., 2019; Tian et al., 2020), mainly due to the increased use of nitrogen fertilizer and manure. N₂O emissions from aquaculture are among the fastest rising contributors of N₂O emissions, but their overall magnitude is still small in the overall N₂O budget (Tian et al., 2020).

Figure 5.16 | Decadal mean nitrous oxide (N₂ O) emissions for 2007–2016 and its change since 1850 based on process-model projections. The total effect, including that from anthropogenic nitrogen additions (atmospheric deposition, manure addition, fertilizer use and land-use), is evaluated against the background flux driven by changes in atmospheric carbon dioxide (CO₂) concentration, and climate change. Fluxes are derived from the N₂O model intercomparison project ensemble of terrestrial biosphere models (Tian et al., 2019) and three ocean biogeochemical models (Landolfi et al., 2017; Battaglia and Joos, 2018a; Buitenhuis et al., 2018). Further details on data sources and processing are available in the chapter data table (Table 5.SM.6).

Open figure

The principal non-agricultural anthropogenic sources of N₂O are industry, specifically chemical processing, wastewater, and the combustion of fossil fuels (Table 5.3). Industrial emissions of N₂O mainly due to nitric and adipic acid production have decreased in North America and Europe since the widespread installation of abatement technologies in the 1990s (Pérez-Ramίrez et al., 2003; Lee et al., 2011; Janssens-Maenhout et al., 2019). There is still considerable uncertainty in industrial emissions from other regions of the world with contrasting trends between inventories (Thompson et al., 2019). Globally, industrial emissions and emissions from fossil fuel combustion by stationary sources, such as power plants, as well as smaller emissions from mobile sources (e.g., road transport and aviation) have remained nearly constant between the 1980s and 2007–2016 (medium evidence, medium agreement) (Winiwarter et al., 2018; Janssens-Maenhout et al., 2019; Tian et al., 2020). Wastewater N₂O emissions, including those from domestic and industrial sources, have increased from 0.2 (0.1–0.3) TgN yr^–1 to 0.35 (0.2–0.5) TgN yr^–1 between the 1980s and 2007–2016 (Tian et al., 2020).

Biomass burning from crop residue burning, grassland, savannah and forest fires, as well as biomass burnt in household stoves, releases N₂O during the combustion of organic matter. Updated inventories since AR5 (WGI, Section 6.4.3) result in a lower range of the decadal mean emissions of 0.6 (0.5–0.8) TgN yr^–1 (van der Werf et al., 2017; Tian et al., 2020). The attribution of grassland, savannah or forest fires to natural or anthropogenic origins is uncertain, preventing a separation of the biomass burning source into natural and anthropogenic.

Since AR5 (WGI, Section 6.4.3), new estimates of the global ocean N₂O source derived from ocean biogeochemistry models are 3.4 (2.5–4.3) TgN yr^–1 for the period 2007–2016 (Figure 5.16; Manizza et al., 2012; Suntharalingam et al., 2012; Martinez-Rey et al., 2015; Landolfi et al., 2017; Buitenhuis et al., 2018; Tian et al., 2020). This is slightly lower than climatological estimates from empirically based methods and surface ocean data syntheses (Bianchi et al., 2012; S. Yang et al., 2020). Nitrous oxide processes in coastal upwelling zones continue to be poorly represented in global estimates of marine N₂O emissions (Kock et al., 2016), but may account for an additional 0.2–0.6 TgN yr^–1 of the global ocean source (Seitzinger et al., 2000; Nevison et al., 2004).

In the oxic ocean (>97% of ocean volume), nitrification is believed to be the primary N₂O source (Freing et al., 2012). In sub-oxic ocean zones (Section 5.3), where denitrification prevails, higher N₂O yields and turnover rates make these regions potentially significant sources of N₂O (Arévalo-Martínez et al., 2015; Babbin et al., 2015; Ji et al., 2015). The relative proportion of ocean N₂O from oxygen-minimum zones is highly uncertain (Zamora et al., 2012). Estimates derived from in situ sampling, particularly in the eastern tropical Pacific, suggest significant fluxes from these regions, and potentially account for up to 50% of the global ocean source (Codispoti, 2010; Arévalo-Martínez et al., 2015; Babbin et al., 2015). However, recent global-scale analyses estimate lower contributions (4–7%, Battaglia and Joos, 2018b; Buitenhuis et al., 2018). Further investigation is required to reconcile these estimates and provide improved constraints on the N₂O source from low-oxygen zones.

Atmospheric deposition of anthropogenic N on oceans can stimulate marine productivity and influence ocean emissions of N₂O. New ocean model analyses since AR5 (WGI, 6.4.3), suggest a relatively modest global potential impact of 0.01–0.32 TgN yr^–1 (pre-industrial to present-day) equivalent to 0.5–3.3% of the global ocean N₂O source (Suntharalingam et al., 2012; Jickells et al., 2017; Landolfi et al., 2017). However, larger proportionate impacts are predicted in nitrogen-limited coastal and inland waters downwind of continental pollution outflow, such as the Northern Indian Ocean (Jickells et al., 2017; Suntharalingam et al., 2019).

Inland waters and estuaries are generally sources of N₂O as a result of nitrification and denitrification of dissolved inorganic nitrogen, however, they can serve as N₂O sinks in specific conditions (Webb et al., 2019). Since AR5 (WGI, 6.4.3), improved emissions factors, including their spatio-temporal scaling, and consideration of transport within the aquatic system allows for better constraint of these emissions (Murray et al., 2015; Hu et al., 2016; Lauerwald et al., 2019; Maavara et al., 2019; Kortelainen et al., 2020; Yao et al., 2020). Despite uncertainties because of the side effects of canals and reservoirs on nutrient cycling, these advances permit attribution of a fraction of inland water N₂O emissions to anthropogenic sources (Tian et al., 2020), which contributes to the increased anthropogenic share of the global N₂O source in this report compared to AR5 (Ciais et al., 2013). As an indirect consequence of agricultural nitrogen use and waste-water treatment, the anthropogenic emissions from inland waters have increased by about a quarter (0.1 TgN yr^–1) between the 1980s and 2007–2016 (Tian et al., 2020).

Soils are the largest natural source of N₂O, arising primarily from nitrogen processing associated with microbial nitrification and denitrification (Table 5.3; Butterbach-Bahl et al., 2013; Snider et al., 2015). Under some conditions, soils can also act as a net sink of N₂O, but this effect is small compared to the overall source (Schlesinger, 2013). Since AR5 (WGI, Section 6.4.3), improved global process-based models (Tian et al., 2019) suggest a present-day source of 6.7 (5.3–8.1) TgN yr^–1 (2007–2016 average), which is consistent with the estimate in AR5. Process-based models and inventory-based methods show that increased N deposition has enhanced terrestrial N₂O emissions by 0.8 (0.4–1.4 TgN yr^–1) relative to approximately pre-industrial times, and by 0.2 (0.1–0.2) TgN yr^–1 between the 1980s and 2007–2016 (limited evidence, medium agreement) (Figure 5.16; Tian et al., 2019). This estimate is at the high end of the range reported in AR5 (WGI, Section 6.4.3). Model projections further show that global warming has led to increased soil N₂O emissions of 0.8 (0.3–1.3) TgN yr^–1 since approximately pre-industrial times, of which about half occurred since the 1980s (limited evidence, high agreement) (Tian et al., 2019, 2020).

The SRCCL assessed that deforestation and other forms of land-use change significantly alter terrestrial N₂O emissions through emission pulses following conversions, generally resulting in long-term reduced emissions in unfertilized ecosystems (medium evidence, high agreement). This conclusion is supported by a recent study demonstrating that the deforestation-pulse effect is offset by the effect of reduced area of mature tropical forests (Tian et al., 2020).

Uncertainties remain in process-based models with respect to their ability to capture the complicated responses of terrestrial N₂O emissions to rain pulses, freeze–thaw cycles and the net consequences of elevated levels of CO₂ accurately (Tian et al., 2019). Emerging literature suggests that permafrost thaw may contribute significantly to arctic N₂O emissions (Voigt et al., 2020), but these processes are not yet adequately represented in models and upscaling to large-scale remains a significant challenge.

The synthesis of bottom-up estimates of N₂O sources (Sections 5.2.3.2–5.2.3.4 and Figure 5.17) yields a global source of 17.0 (12.2 to 23.5) TgN yr^–1 for the years 2007–2016 (Table 5.3). This estimate is comparable to AR5, but the uncertainty range has been reduced primarily due to improved estimates of ocean and anthropogenic N₂O sources. Since AR5 (WGI, Section 6.4.3), improved capacity to estimate N₂O sources from atmospheric N₂O measurements by inverting models of atmospheric transport provides a new and independent constraint for the global N₂O budget (Saikawa et al., 2014; Thompson et al., 2019; Tian et al., 2020). The decadal mean source derived from these inversions is remarkably consistent with the bottom-up global N₂O budget for the same period, however, the split between land and ocean sources based on atmospheric inversions is less constrained, yielding a smaller land source of 11.3 (10.2 to 13.2) TgN yr^–1 and a larger ocean source of 5.7 (3.4 to 7.2) TgN yr^–1, respectively, compared to bottom-up estimates.

Figure 5.17 | Global nitrous oxide (N₂ O) budget (2007–2016). Values and data sources as in Table 5.3. The atmospheric stock is calculated from mean N₂O concentration, multiplying a factor of 4.79 ± 0.05 Tg ppb^–1 (Prather et al., 2012). Pool sizes for the other reservoirs are largely unknown. Further details on data sources and processing are available in the chapter data table (Table 5.SM.6).

Open figure

Supported by multiple studies and extensive observational evidence (Sections 5.2.3.2–5.2.3.4 and Figure 5.17), anthropogenic emissions contributed about 40% (7.3; uncertainty range: 4.2 to 11.4 TgN yr^–1) to the total N₂O source in 2007–2016 (high confidence). This estimate is larger than in AR5 (WGI, 6.4.3) due to a larger estimated effect of nitrogen deposition on soil N₂O emissions and the explicit consideration of the role of anthropogenic nitrogen in determining inland water and estuary emissions.

Based on bottom-up estimates, anthropogenic emissions from agricultural nitrogen use, industry and other indirect effects have increased by 1.7 (1.0 to 2.7) TgN yr^–1 between the decades 1980–1989 and 2007–2016, and are the primary cause of the increase in the total N₂O source (high confidence). Atmospheric inversions indicate that changes in surface emissions, rather than in the atmospheric transport or sink of N₂O, are the cause for the increased atmospheric growth rate of N₂O (robust evidence, high agreement) (Thompson et al., 2019). However, the increase of 1.6 (1.4 to 1.7) TgN yr^–1 in global emissions between 2000–2005 and 2010–2015 based on atmospheric inversions is somewhat larger than bottom-up estimates over the same period, primarily because of differences in the estimates of land-based emissions.

The Paleocene–Eocene Thermal Maximum (PETM) was an episode of global warming exceeding pre-industrial temperatures by 4°C–8°C (McInerney and Wing, 2011; Dunkley Jones et al., 2013) that occurred 55.9–55.7 Ma. The PETM involved a large pulse of geologic CO₂ released into the ocean–atmosphere system in 3–20 kyr (Zeebe et al., 2016; Gutjahr et al., 2017; Kirtland Turner et al., 2017; Kirtland Turner, 2018; Gingerich, 2019; Section 5.2.1.1). In response, observationally constrained model simulations report an increase in atmospheric CO₂ concentrations ranging from about 900 ppm to >2000 ppm (Chapter 2; Gutjahr et al., 2017; Cui and Schubert, 2018; Anagnostou et al., 2020). The PETM thus provides a test for our understanding of the ocean’s response to the increase in carbon (and heat) emissions over geologically short time scales.

A limited number of independent proxy records indicate that the PETM was associated with a surface ocean pH decline ranging from 0.15 to 0.30 units (Penman et al., 2014; Gutjahr et al., 2017; Babila et al., 2018). It was also accompanied by a rapid (<10 ka) shallowing of the carbonate saturation horizon, resulting in the widespread dissolution of sedimentary carbonate, followed by a gradual (100 kyr) recovery (Zachos et al., 2005; Bralower et al., 2018). The remarkable similarity among sedimentary records spanning a wide range of ecosystems suggests with medium confidence that the perturbation in the ocean carbonate saturation was global (Babila et al., 2018) and directly resulted from elevated atmospheric CO₂ levels. The degree of acidification is similar to the 0.4 pH unit decrease projected for the end of the 21st century under RCP8.5 (Gattuso et al., 2015) and is estimated to have occurred at a rate about one order of magnitude slower than the current rate of ocean acidification (Zeebe et al., 2016). There is low confidence in the inferred rates of ocean acidification inherent to the range of uncertainties affecting rates estimates based on marine sediments (Section 5.1.2.1).

Recent model outputs and globally distributed geochemical data reveal with medium confidence widespread ocean deoxygenation during the PETM (Dickson et al., 2012, 2014; Winguth et al., 2012; Chang et al., 2018; Remmelzwaal et al., 2019), with parts of the ocean potentially becoming drastically oxygen-depleted (anoxic; Yao et al., 2018; Clarkson et al., 2021). Deoxygenation affected the surface ocean globally (including the Arctic Ocean; Sluijs et al., 2006), due to vertical and lateral expansion of OMZs (Zhou et al., 2014) that resulted from warming and related changes in ocean stratification. Expansion of OMZs may have stimulated N₂O production through water-column (de)nitrification (Junium et al., 2018). The degree to which N₂O production impacted PETM warming, however, has not yet been established.

The feedbacks associated with recovery from the PETM are uncertain, yet could include drawdown associated with silicate weathering (Zachos et al., 2005) and regrowth of terrestrial and marine organic carbon stocks (Bowen and Zachos, 2010; Gutjahr et al., 2017).

The Last Deglacial Transition (LDT) is the best documented climatic transition in the past associated with a substantial atmospheric CO₂ rise ranging from 190 to 265 ppm between 18–11 ka (Marcott et al., 2014). The amplitude of the deglacial CO₂ rise is thus on the same order of magnitude as the increase since the Industrial Revolution.

Boron isotope ( δ¹¹B) data suggest a 0.15–0.05 unit decrease in sea surface pH (Hönisch and Hemming, 2005; Henehan et al., 2013) across the LDT, an average rate of decline of about 0.002 units per century compared with the current rate of more than 0.1 units per century (Bopp et al., 2013; Gattuso et al., 2015). Planktonic foraminiferal shell weights decreased by 40% to 50% (Barker and Elderfield, 2002), and coccolith mass decreased by about 25% (Beaufort et al., 2011) across the LDT. Independent proxy reconstructions thus highlight with high confidence that pH values decreased as atmospheric CO₂ concentrations increased across the LDT. There is, however, low confidence in the inferred rate of ocean acidification owing to multiple sources of uncertainties affecting rates estimates based on marine sediments (Section 5.1.2.1).

Geochemical and micropaleontological evidence suggest that intermediate-depth OMZs almost vanished during the Last Glacial Maximum (LGM) (Jaccard et al., 2014). However, multiple lines of evidence suggest with medium confidence that the deep (>1500 m) ocean became depleted in O₂ (concentrations were possibly lower than 50 μmol kg^–1) globally (Jaccard and Galbraith, 2012; Hoogakker et al., 2015, 2018; Gottschalk et al., 2016, 2020a; Anderson et al., 2019) as a combined result of sluggish ventilation of the ocean subsurface (Gottschalk et al., 2016, 2020a; Skinner et al., 2017) and a generally more efficient marine biological carbon pump (Buchanan et al., 2016; Yamamoto et al., 2019; Galbraith and Skinner, 2020).

During the LDT, deep ocean ventilation increased as Antarctic Bottom Water (AABW) (Skinner et al., 2010; Gottschalk et al., 2016; Jaccard et al., 2016) and subsequently the Atlantic meridional overturning circulation (McManus et al., 2004; Lippold et al., 2016) resumed, transferring previously sequestered remineralized carbon from the ocean interior to the upper ocean, and eventually the atmosphere (Skinner et al., 2010; Galbraith and Jaccard, 2015; Gottschalk et al., 2016; Ronge et al., 2016, 2020; Sikes et al., 2016; Rae et al., 2018), contributing to the deglacial CO₂ rise. Intermediate depths lost oxygen as a result of sluggish ventilation and increasing temperatures (decreasing saturation). As the world emerged from the last Glacial period, OMZs underwent a large volumetric increase at the beginning of the Bølling-Allerød (B/A), a northern-hemisphere wide warming event, 14.7 ka (Jaccard and Galbraith, 2012; Praetorius et al., 2015) with deleterious consequences for benthic ecosystems (e.g., Moffitt et al., 2015). These observations indicate with high confidence that the rate of warming, affecting the solubility of oxygen and upper water column stratification, coupled with changes in subsurface ocean ventilation, impose a direct control on the degree of ocean deoxygenation, implying a high sensitivity of ocean oxygen loss to warming. The expansion of OMZs contributed to a widespread increase in water column (de)nitrification (Galbraith and Kienast, 2013), which contributed substantially to enhanced marine N₂O emissions. Nitrogen stable isotope measurements on N₂O extracted from ice cores suggest that approximately one-third (of the order of 0.7 ± 0.3TgN yr^–1) of thedeglacial increase in N₂O emissions relates to oceanic sources (Schilt et al., 2014; H. Fischer et al., 2019).

Ocean pH time series are based on the reconstruction of coral boron isotope ratios ( δ¹¹B). A majority of coral δ¹¹B data have been generated from the western Pacific region with a few records from the Atlantic Ocean. Biweekly resolution paleo-pH records show monsoonal variation of about 0.5 pH unit in the South China Sea (Liu et al., 2014). Interannual ocean pH variability in the range of 0.07–0.16 pH unit characterizes southwest Pacific corals that are attributed to El Niño–Southern Oscillation (ENSO) (H.C. Wu et al., 2018) and river runoff (D’Olivo et al., 2015). Decadal (10-, 22- and 48-year) ocean pH variations in the south-west Pacific have been linked to the Inter-decadal Pacific Oscillation, causing variations of up to 0.30 pH unit in the Great Barrier Reef (Pelejero et al., 2005; Wei et al., 2009) but weaker (about 0.08 pH unit) in the open ocean (H.C. Wu et al., 2018). Decadal variations in the South China Sea pH changes of 0.10–0.20 have also been associated with the variation in the East Asian monsoon (Liu et al., 2014; Wei et al., 2015), as a weakening of the Asian winter monsoon leads to sluggish water circulation within the reefs, building up localised CO₂ concentration in the water due to calcification and respiration.

Since the beginning of the industrial period in the mid-19th century, coral δ¹¹B-derived ocean pH has decreased by 0.06–0.24 pH unit in the South China Sea (Liu et al., 2014; Wei et al., 2015) and 0.12 pH unit in the south-west Pacific (H.C. Wu et al., 2018). Since the mid-20th century, a distinct feature of coral δ¹¹B records relates to ocean acidification trends, albeit having a wide range of values: 0.12–0.40 pH unit in the Great Barrier Reef (Wei et al., 2009; D’Olivo et al., 2015), 0.05–0.08 pH unit in the north-west Pacific (Shinjo et al., 2013) and 0.04–0.09 pH unit in the Atlantic Ocean (Goodkin et al., 2015; Fowell et al., 2018). Concurrent coral carbon isotopic ( δ¹³C) measurements infer ocean uptake of anthropogenic CO₂ from the combustion of fossil fuel, based on the lower abundance of¹³C in fossil fuel carbon. Western Pacific coral records show depleted δ¹³C trends since the late 19th century that are more prominent since the mid-20th century (high confidence) (Pelejeroet al., 2005; Wei et al., 2009; Shinjo et al., 2013; Liu et al., 2014; Kubota et al., 2017; H.C. Wu et al., 2018).

Overall, many of the records show a highly variable seawater pH underlaying strong imprints of internal climate variability (high confidence) and, in most instances, superimposed on a decreasing δ¹¹B trend that is indicative of anthropogenic ocean acidification in recent decades (medium confidence). The robustness of seawater pH reconstructions is currently limited by the uncertainty on the calibration of The δ¹¹B proxy in different tropical coral species.

The SROCC (Section 5.2.2.3) indicated that it is virtually certain that the ocean has undergone acidification globally in response to ocean CO₂ uptake, and concluded that pH in open ocean surface water has changed by avirtually certain range of –0.017 to –0.027 pH units per decade since the late 1980s. Since SROCC, evidence of the progress of acidification across all regions of the oceans has been further strengthened by continued observations of seawater carbonate chemistry at ocean time series stations, and compiled shipboard studies providing temporally resolved and methodologically consistent datasets (Jiang et al., 2019) (Figure 5.20; Supplementary Material Table 5.SM.3; Section 2.3.3.5).

Figure 5.20 | Multi-decadal trends of pH (Total Scale) in surface layer at various sites of the oceans and a global distribution of annual mean pH adjusted to the year 2000. Time-series data of pH are from Dore et al. (2009), Olafsson et al. (2009), González-Dávila et al. (2010), Bates et al. (2014), Takahashi et al. (2014), Wakita et al. (2017), Merlivat et al. (2018), Ono et al. (2019), and Bates and Johnson (2020). Global distribution of annual mean pH have been evaluated from data of surface oceanpCO₂ measurements (Bakker et al., 2016; Jiang et al., 2019). Acronyms in panels: KNOT and K2 – Western Pacific subarctic gyre time series; HOT – Hawaii Ocean Time-series; BATS – Bermuda Atlantic Time-series Study; DYFAMED – Dynamics of Atmospheric Fluxes in the Mediterranean Sea; ESTOC – European Station for Time-series in the Ocean Canary Islands; CARIACO – Carbon Retention in a Colored Ocean Time-series. Further details on data sources and processing are available in the chapter data table (Table 5.SM.6).

Open figure

In the subtropical open oceans, decreases in pH have been reported with avery likely rate range from –0.016 to –0.019 pH units per decade since 1980s, which equates to approximately 4 % increase in hydrogen ion concentration ([H⁺]) per decade. Accordingly, the saturation state Ω (=[Ca²⁺][CO₃^2-]/K_sp) of seawater with respect to calcium carbonate mineral aragonite has been declining at rates ranging from –0.07 to –0.12 per decade (González-Dávila et al., 2010; Feely et al., 2012; Bates et al., 2014; Takahashi et al., 2014; Ono et al., 2019; Bates and Johnson, 2020; Supplementary Material Table 5.SM.3). These rates are consistent with the rates expected from the transient equilibration with increasing atmospheric CO₂ concentrations, but the variability of rate in decadal time scale has also been detected with robust evidence (Ono et al., 2019; Bates and Johnson, 2020). In the tropical Pacific, its central and eastern upwelling zones exhibited a faster pH decline of –0.022 to –0.026 pH unit per decade due to increased upwelling of CO₂ -rich sub-surface waters in addition to anthropogenic CO₂ uptake (Sutton et al., 2014; Lauvset et al., 2015). By contrast, warm pools in the western tropical Pacific exhibited slower pH decline of –0.010 to –0.013 pH unit per decade (Supplementary Material Table 5.SM.3; Lauvsetet al., 2015; Ishii et al., 2020). Observational and modelling studies (Nakano et al., 2015; Ishii et al., 2020) consistently suggest that slower acidification in this region is attributable to the anthropogenic CO₂ taken up in the extratropics around a decade ago and transported to the tropics via shallow meridional overturning circulations.

In open subpolar and polar zones, the very likely range (–0.003 to –0.026 pH unit per decade) and uncertainty (up to 0.010) observed in pH decline are larger than in the subtropics, reflecting the complex interplay between physical and biological forcing mechanisms (Olafsson et al., 2009; Midorikawa et al., 2012; Bates et al., 2014; Takahashi et al., 2014; Lauvset et al., 2015; Wakita et al., 2017; Merlivat et al., 2018). Nevertheless, the high agreement of pH decline among these available time-series studies leads to high confidence in the trend of acidification in these zones. In the Arctic Ocean, a temporally limited time series of carbonate chemistry measurements prevents drawing robust conclusions on ocean acidification trends. However, the carbonate saturation state (Ω) is generally low, and observational studies show with robust evidence that the recent extensive melting of sea ice leading to enhanced air–sea CO₂ exchange, large freshwater inputs, together with river discharge and glacial drainage, as well as the degradation of terrestrial organic matter in seawater, result in the decline of Ω of aragonite to undersaturation (Bates et al., 2009; Chierici and Fransson, 2009; Yamamoto-Kawai et al., 2009; Azetsu-Scott et al., 2010; Robbins et al., 2013; Fransson et al., 2015; Semiletov et al., 2016; Anderson et al., 2017; Qi et al., 2017; Beaupré-Laperrière et al., 2020; Y. Zhang et al., 2020; SROCCSection 3.2.1.2.4, IPCC, 2019b). The low saturation state of aragonite (Ω about 1) has also been observed in surface waters of the Antarctic coastal zone associated with freshwater input from glaciers (Mattsdotter Björk et al., 2014) and upwelling of deep water (Hauri et al., 2015) as well as along eastern boundary upwelling systems (Feely et al., 2016).

Overall, in agreement with SROCC, it is virtually certain from these observational studies that ocean surface waters undergo acidification globally with the CO₂ increase in the atmosphere. These sustained measurements over the past decades, and campaign studies of ocean carbonate chemistry, also highlight with robust evidence that trends of acidification have been modulated by the variability and changes in physical and chemical states of ocean, including those affected by the warming of the cryosphere, and need to be better understood.

Advances in observations and modelling for ocean physics and biogeochemistry and established knowledge of ocean carbonate chemistry show with very high confidence that anthropogenic CO₂ taken up into the ocean surface layer is further spreading into the ocean interior through ventilation processes, including vertical mixing, diffusion, subduction and meridional overturning circulations (Sections 2.3.3.5, 5.2.1.3 and 9.2.2.3; Sallée et al., 2012; Bopp et al., 2015; Nakano et al., 2015; Iudicone et al., 2016; Toyama et al., 2017; Pérez et al., 2018; Gruber et al., 2019b) and is causing acidification in the ocean interior. The net change in oxygen consumption by aerobic respiration of marine organisms further influences acidification by releasing CO₂ (Section 5.3.3.2; Chen et al., 2017; Breitburg et al., 2018; Robinson, 2019).

Observations over past decades of basin-wide and global syntheses of ocean interior carbon show that the extent of acidification due to anthropogenic CO₂ invasion tends to diminish with depth (very high confidence) (Section 5.2.1.3.3 and Figure 5.21; Woosley et al., 2016; Carter et al., 2017; Lauvset et al., 2020). The regions of deep convection such as subpolar North Atlantic and Southern Ocean present the deepest acidification detections below 2000 m (medium confidence). Mid-latitudinal zones within the subtropical cells and tropical regions present a relatively deep and shallow detection, respectively. A pH decrease has also been observed on the Antarctic continental shelf (Hauck et al., 2010; Williams et al., 2015). Acidification is also underway in the subsurface to intermediate layers of the Arctic Ocean due to the inflow of ventilated waters from the North Atlantic and the North Pacific (Qi et al., 2017; Ulfsbo et al., 2018).

Figure 5.21 | Spread of ocean acidification from the surface into the interior of ocean since pre-industrial times. (a) Map showing the three transects used to create the cross sections shown in (b) and (c); vertical sections of the changes in (b) pH and (c) saturation state of aragonite (Ω_arag) between 1800–2002 due to anthropogenic CO₂ invasion (colour). Contour lines are their contemporary values in 2002. The red transect begins in the Nordic Seas and then follows the GO-SHIP lines A16 southward in the Atlantic Ocean, SR04 and S04P westward in the Southern Ocean, and P16 northward in the Pacific Ocean. The purple line follows the GO-SHIP line I09 southward in the Indian Ocean. The green line on the smaller inset crosses the Arctic Ocean from the Bering Strait to North Pole along 175°W and from the North Pole to the Fram Strait along 5°E (Lauvset et al., 2020). Further details on data sources and processing are available in the chapter data table (Table 5.SM.6).

Open figure

A significant increase in acidification resulting from net metabolic CO₂ release coupled with ocean circulation changes has been shown with high confidence in large swathes of intermediate waters in the Pacific and Atlantic oceans (Dore et al., 2009; Byrne et al., 2010; Ríos et al., 2015; Chu et al., 2016; Carter et al., 2017; Lauvset et al., 2020). For example, ocean circulation contributes a pH change of –0.013 ± 0.013 to the overall observed change of –0.029 ± 0.014 for 1993–2013 at depths around 1000 m at 30°S–40°S in the South Atlantic ocean (Ríos et al., 2015). Long-term repeated observations in the North Pacific show a decline in dissolved oxygen (–4.0 μmol kg⁻¹ per decade at maximum) being sustained in the intermediate water since the 1980s (Takatani et al., 2012; Sasano et al., 2015). The amplification of acidification associated with the weakening ventilation is thought to have been occurring persistently. In contrast, for the North Pacific subtropical mode water, large decadal variability in pH and aragonite saturation state with amplitudes of about 0.02 and about 0.1, respectively, are superimposed on secular declining trends due to anthropogenic CO₂ invasion (Oka et al., 2019). This is associated with the variability in ventilation due to the approximately 50% variation in the formation volume of the mode water that is forced remotely by the Pacific Decadal Oscillation (Qiu et al., 2013; Oka et al., 2015).

These trends of acidification in the ocean interior lead to high confidence in shoaling of the saturation horizons of calcium carbonate minerals where Ω = 1. In the Pacific Ocean where the aragonite saturation horizon is shallower (a few hundred metres to 1200 m; Figure 5.21c), the rate of its shoaling is in the order of 1–2 m yr^–1 (Feely et al., 2012; Ross et al., 2020). In contrast, shoaling rates of 4 m yr^–1 to 1710 m for 1984–2008 and of 10–15 m yr^–1 to 2250 m for 1991–2016 have been observed in the Iceland sea and the Irminger sea, respectively (Olafsson et al., 2009; Pérez et al., 2018).

In summary, ocean acidification is spreading into the ocean interior. Its rates at depths are controlled by the ventilation of the ocean interior as well as anthropogenic CO₂ uptake at the surface, thereby diminishing with depth (very high confidence) (Figure 5.21). Variability in ocean circulation modulates the trend of ocean acidification at depths through the changes in ventilation and their impacts on metabolic CO₂ content. However, the large knowledge gap around ventilation changes leads to low confidence in their impacts in many ocean regions (Sections 5.3.3.2; 9.2.2.3 and 9.3.2).

As summarized in SROCC (Section 5.2.2.4), there is a growing consensus that between 1970 and 2010 the open ocean has very likely lost 0.5–3.3% of its dissolved oxygen in the upper 1000 m depth (Section 2.3.3.6; Helm et al., 2011; Ito et al., 2017; Schmidtko et al., 2017; Bindoff et al., 2019). Regionally, the equatorial and North Pacific, the Southern Ocean and the South Atlantic have shown the greatest oxygen loss of up to 30 mol m^–2 per decade (Schmidtko et al., 2017). Warming – via solubility reduction and circulation changes – mixing and respiration are considered the major drivers, with 50% of the oxygen loss for the upper 1000 m of the global oceans attributable to the solubility reduction (Schmidtko et al., 2017). Climate variability also modifies the oxygen loss on interannual and decadal time scales especially for the tropical ocean OMZs (Deutsch et al., 2011, 2014; Llanillo et al., 2013) and the North Pacific subarctic zone (Whitney et al., 2007; Sasano et al., 2018; Cummins and Ross, 2020). However, quantifying the oxygen decline and variability and attributing them to processes in different regions remains challenging (Levin, 2018; Oschlies et al., 2018). Earth system models (ESMs) in CMIP5 and CMIP6 corroborate the decline in ocean oxygen, and project a continuing and accelerating decline with a strong impact of natural climate variability under high-emissions scenarios (Bopp et al., 2013; Long et al., 2016; Kwiatkowski et al., 2020). However, CMIP5 models did not reproduce observed patterns for oxygen changes in the tropical thermocline, and generally simulated only about half the oxygen loss inferred from observations (Oschlies et al., 2018). CMIP6 models have a more realistic simulated mean state of ocean biogeochemistry than CMIP5 models due to improved ocean physical processes and better representation of biogeochemical processes (Séférian et al., 2020). Theyalso exhibit enhanced ocean warming as a result of an increase in the equilibrium climate sensitivity (ECS) of CMIP6 relative to CMIP5 models, which contributes to increased stratification and reduced subsurface ventilation (Sections 4.3.1, 4.3.4, 5.3.3.2, 7.4.2, 7.5.6, 9.2.1, and TS2.4). Consequently, CMIP6 model ensembles reproduce the ocean deoxygenation trend of −0.30 to −1.52 mmol m⁻³ per decade between 1970–2010 reported in SROCC (Section 5.2.2.4) with a verylikely range, and also project 32–71% greater subsurface (100–600 m) oxygen decline relative to their Representative Concentration Pathway (RCP) analogues in CMIP5, reaching to the likely range of decline of 6.4 ± 2.9 mmol m^–3under SSP1–2.6 and 13.3 ± 5.3 mmol m^–3under SSP5–8.5, from 1870–1899 to 2080–2099 (Kwiatkowski et al., 2020). It is concluded that the oxygen content of subsurface ocean is projected to transition to historically unprecedented condition with decline over the 21st century (medium confidence).

In oxygen-depleted waters, microbial processes (denitrification and anammox, i.e., anaerobic ammonium oxidation; Kuypers et al., 2005; Codispoti, 2007; Gruber and Galloway, 2008) remove fixed nitrogen, and when upwelled waters reach the photic zone, primary production becomes nitrogen-limited (Tyrrell and Lucas, 2002). However, in other oceanic regions, increased water-column stratification due to warming may reduce the amount of N₂O reaching the surface and thereby decrease N₂O flux to the atmosphere. Landolfi et al. (2017) suggest that, by 2100, under the RCP8.5 scenario, total N₂O production in the ocean may decline by 5% and N₂O emissions be reduced by 24% relative to the pre-industrial era due to decreased organic matter export and anthropogenic-driven changes in ocean circulation and atmospheric N₂O concentrations. Projected oxygen loss in the ocean is thought to result in an ocean-climate feedback through changes in the natural emissions of GHGs (low confidence).

The areas with relatively rapid oxygen decrease include OMZs in the tropical oceans, where oxygen content has been decreasing at a rate of 0.9–3.4 µmol kg^–1 per decade in the thermocline for the past five decades (Stramma et al., 2008). Low oxygen, low pH and shallow aragonite saturation horizons in the OMZs of the eastern boundary upwelling regions co-occur, affecting ecosystem structure (Chavez et al., 2008) and function in the water column, including the presently unbalanced nitrogen cycle (Paulmier and Ruiz-Pino, 2009). The coupling between upwelling, productivity, and oxygen depletion feeds back to biological productivity and the role of these regions as sinks or sources of climate active gases. When OMZ waters upwell and impinge on the euphotic zone, they release significant quantities of GHGs, including N₂O (0.81–1.35 TgN yr^–1), CH₄ (0.27–0.38 TgCH₄yr^–1), and CO₂ (yet to be quantified) to the atmosphere, exacerbating global warming (Paulmier et al., 2008; Naqvi et al., 2010; Kock et al., 2012; Arévalo-Martínez et al., 2015; Babbin et al., 2015; Farías et al., 2015). Modelling projectionssuggest a global decrease of 4–12% in oceanic N₂O emissions (from 3.71–4.03 TgN yr^–1 to 3.54–3.56 TgN yr^–1) from 2005 to 2100 under RCP8.5, despite a tendency to increased N₂O production in the OMZs, associated primarily with denitrification (Martinez-Rey et al., 2015). It is difficult to single out the contribution of nitrification and denitrification, which can occur simultaneously. A rigorous separation of these two processes would require more mechanistic parametrization, which has been hindered by the still large conceptual and parametric uncertainties (Babbin et al., 2015; Trimmer et al., 2016; Landolfi et al., 2017). Furthermore, the correlation between N₂O and oxygen varies with microorganisms present, nutrient concentrations, and other environmental variables (Voss et al., 2013).

In summary, total oceanic N₂O emissions were projected to decline by 4–12% from 2005–2100 (Martinez-Rey et al., 2015) and by 24% from the pre-industrial era to 2100 (Landolfi et al., 2017) under RCP8.5. However, there is low confidence in the reduction in N₂O emissions to the atmosphere, because of large conceptual and parametric uncertainties, a limited number of modelling studies that explored this process, and greater oxygen losses simulated in CMIP6 models than in CMIP5 models (Kwiatkowski et al., 2020).

Projections with CMIP5 ESMs, reported in AR5 (Section 6.4.4) and SROCC (Section 5.2.2.3; IPCC, 2019b), showed changes in global mean surface ocean pH from 1870–1899 to 2080–2099 of –0.14 ± 0.001 (inter-model standard deviation) under RCP2.6 and –0.38 ± 0.005 under RCP8.5 with pronounced regional variability (Bopp et al., 2013; Hurd et al., 2018). They also projected faster pH declines in mode waters below seasonal mixed layers (Resplandy et al., 2013; Watanabe and Kawamiya, 2017) as has been observed in the Atlantic (Salt et al., 2015) and in the Pacific (Carter et al., 2019), because of the net CO₂ release by respiration and lowering CO₂ buffering capacity of seawater. In these CO₂ concentration-driven simulations, the level of acidification in the surface ocean is primarily determined by atmospheric CO₂ concentration and regional seawater carbonate chemistry, thereby providing consistent projections across models. New projections with CMIP6 ESMs show greater surface pH decline of –0.16 ± 0.002 under the SSP1-2.6 and –0.44 ± 0.005 under SSP5-8.5 from 1870–1899 to 2080–2099 (Section 4.3.2.5 and Cross-Chapter Box 5.3; Kwiatkowski et al., 2020). The greater pH declines in CMIP6 are primarily a consequence of higher atmospheric CO₂ concentrations in SSPs than their CMIP5-RCP analogues (Kwiatkowski et al., 2020). Ocean acidification is also projected to occur with high confidence in the Abyssal Bottom Waters in regions such as the northern North Atlantic and the Southern Ocean (Sulpis et al., 2019), with the rates of global mean pH decline of –0.018 ± 0.001 under SSP1-2.6 and –0.030 ± 0.002 under SSP5-8.5 from 1870–1899 to 2080–2099 in CMIP6 (Kwiatkowski et al., 2020).

In surface ocean, changes in the amplitude of seasonal variations in pH are also projected to occur with high confidence. ESMs in CMIP6 show +73 ± 12% increase in the amplitude of seasonal variation in hydrogen ion concentration ([H⁺]) but 10 ± 5% decrease in the seasonal variation in pH (= -log [H⁺]) from 1995–2014 to 2080–2099 under SSP5-8.5. The simultaneous amplification of [H⁺] and attenuation of pH seasonal cycles is counterintuitive but is the consequence of a greater increase in the annual mean [H⁺] due to anthropogenic CO₂ invasion than the corresponding increase in its seasonal amplitude. These changes are consistent with the amplification/attenuation of the seasonal variation of +81 ±16% for [H⁺] and –16 ± 7% for pH from 1990–1999 to 2090–2099 under RCP8.5 in CMIP5 (Kwiatkowski and Orr, 2018).

The signal of ocean acidification in surface ocean is large and is projected to emerge beyond the range of natural variability within the time scale of a decade in all ocean basins (Schlunegger et al., 2019). There is high agreement among modelling studies that the largest pH decline and large-scale undersaturation of aragonite in surface seawater start to occur first in polar oceans (Orr et al., 2005; Steinacher et al., 2009; Hurd et al., 2018; Jiang et al., 2019). Under SSP5-8.5, the largest surface pH decline, exceeding 0.45 between 1995–2014 and 2080–2099, occurs in the Arctic Ocean (Kwiatkowski et al., 2020). The freshwater input from sea ice melt is an additional factor leading to a faster decline of aragonite saturation level than expected from the anthropogenic CO₂ uptake (Yamamoto et al., 2012). The increase in riverine and glacial discharges that provide terrigenous carbon, nutrients and alkalinity as well as freshwater are the other factors modifying the rate of acidification in the Arctic Ocean. However, their impacts have been projected in a limited number of studies with extensive knowledge gaps and model simplifications leading to low confidence in their impacts (Terhaar et al., 2019; Hopwood et al., 2020). In the Southern Ocean, the aragonite undersaturation starts in the 2030s in RCP8.5, and the area that experiences aragonite undersaturation for at least one month per year by 2100 is projected to be more than 95%. Under RCP2.6, short periods (less than one month) of aragonite undersaturation are expected to be found in less than 2% of the area during this century (Sasse et al., 2015; Hauri et al., 2016; Negrete-García et al., 2019). These long term projections are modified at interannual time scales by large-scale climate modes (Ríos et al., 2015) such as the ENSO and the Southern Annular Mode (Conrad and Lovenduski, 2015). In other regions, acidification trends are influenced by a range of processes such as changes in ocean circulation, temperature, salinity, carbon cycling, and the structure of the marine ecosystem. As, at present, models do not resolve fine-scale variability of these processes, current projections do not fully capture the changes that the marine environment will experience in the future (Takeshita et al., 2015; Turi et al., 2016).

Overall, with the rise of atmospheric CO₂, the physics of CO₂ transfer across the air–sea interface, the carbonate chemistry in seawater, the trends of ocean acidification being observed in the past decades (Section 5.3.3.2) and modelling studies described in this section, it is virtually certain that ocean acidification will continue to grow. However, the magnitude and sign (direction) of many of ocean carbon–climate feedbacks are still poorly constrained (Matear and Lenton, 2014, 2018), leading to low confidence in their significant and long-lasting impacts on ocean acidification.

Reversing the increase in atmospheric CO₂ concentrations through negative emissions (Section 5.6) will reverse ocean acidification at the sea surface (high confidence) but will not result in rapid amelioration of ocean acidification in the deeper ocean (Section 5.3.3.2). The ocean’s uptake of atmospheric CO₂ will start to decrease as atmospheric CO₂ decreases (Sections 5.4.5, 5.4.10 and 5.6.2.1; Mathesius et al., 2015; Tokarska and Zickfeld, 2015). However, because of the long time scales of the ocean turnover that transfers CO₂ from the upper to the deep ocean, excess carbon will continue to accumulate in the deep ocean even after a decrease in atmospheric CO₂ (Cao et al., 2014; Mathesius et al., 2015; Tokarska and Zickfeld, 2015; T. Li et al., 2020). There is thus high confidence that CO₂ emissions leave a long-term legacy in ocean acidification, and are therefore irreversible at multi-human generational scales, even with aggressive atmospheric CO₂ removal.

Observations and data products including models (Astor et al., 2013; Bakker et al., 2016; Kosugi et al., 2016; Vargas et al., 2016; Laruelle et al., 2017, 2018; Orselli et al., 2018; Roobaert et al., 2019; Cai et al., 2020; H. Sun et al., 2020) confirm the strong spatial and temporal variability in the coastal ocean surface carbonate chemistry and sea-air CO₂ fluxes (high agreement, robust evidence). The anthropogenic CO₂ -induced acidification is either mitigated or enhanced through biological processes; primary production removes dissolved CO₂ from the surface, and respiration adds CO₂ and consumes oxygen in the subsurface layers. The relative intensity of these processes is controlled by natural or anthropogenic eutrophication. Other drivers of variability include biological community composition, freshwater input from rivers or melting ice, sea ice cover and calcium carbonate precipitation/dissolution dynamics, coastal upwelling and regional circulation, and seasonal surface cooling (Fransson et al., 2015, 2017; Feely et al., 2018; Roobaert et al., 2019; Cai et al., 2020; Hauri et al., 2020; Monteiro et al., 2020b; H. Sun et al., 2020). Near-shore surface waters are often supersaturated with CO₂, regardless of the latitude, especially in highly populated areas receiving substantial amounts of domestic and industrial sewage (Chen and Borges, 2009). Nevertheless, thermal or haline-stratified eutrophic coastal areas may act as net atmospheric CO₂ sinks (Chou et al., 2013; Cotovicz Jr. et al., 2015). Continental shelves, excluding near-shore areas, act as CO₂ sinks at a rate of 0.2 ± 0.02 PgC yr^–1 (Laruelle et al., 2014; Roobaert et al., 2019), considering ice-free areas only. Under increasing atmospheric CO₂ and eutrophication, such ecosystems would be more vulnerable to ecological and seawater chemistry changes, impacting the local economy.

Since AR5, (Ciais et al., 2013) and in agreement with SROCC (IPCC, 2019b), there is now high agreement (robust evidence) that coastal ocean acidification, whether induced only by increasing atmospheric CO₂ or exacerbated by eutrophication or upwelling, has negative effects on specific groups of marine organisms such as reef-building corals, crabs, pteropods, and sessile fauna (AR6 WGII, Chapter 3; Dupont et al., 2010; Bindoff et al., 2019; Bednaršek et al., 2020; Osborne et al., 2020), especially when combined with stressors such as temperature and deoxygenation, and potentially increased bioavailability of toxic elements such as arsenic and copper (Millero et al., 2009; Boyd et al., 2015; Breitburg et al., 2018).

Since SROCC (Bindoff et al., 2019), there is further evidence that anthropogenic eutrophication via continental runoff and atmospheric nutrient deposition, and ocean warming are very likely the main drivers of deoxygenation in coastal areas (Levin and Breitburg, 2015; Levin et al., 2015; Royer et al., 2016; Breitburg et al., 2018; Cocquempot et al., 2019; Fagundes et al., 2020; Limburg et al., 2020). Increasing intensity and frequency of wind-driven upwelling is responsible for longer and more intense coastal hypoxia, fuelled by organic matter degradation from primary production(medium to high agreement, medium evidence) (Rabalais et al., 2010; Bakun et al., 2015; Varela et al., 2015; Fennel and Testa, 2019; Limburg et al., 2020). Locally, submarine groundwater discharge may enhance the eutrophication state (low agreement, limited evidence, Luijendijk et al., 2020). Since AR5 (Ciais et al., 2013) and SROCC (Bindoff et al., 2019) new observations and model studies confirm the trends in increasing coastal hypoxia caused by eutrophication, ocean warming and changes in circulation (Claret et al., 2018; Dussin et al., 2019; Limburg et al., 2020), as well as the ubiquitous impacts on marine organisms and fisheries (AR6 WGII Chapter 3; Carstensen and Conley, 2019; Fennel and Testa, 2019; Osma et al., 2020). Following open ocean deoxygenation trends since the 1950s, more than 700 coastal regions are being reported as hypoxic (dissolved oxygen concentration <2 mg O₂L^–1) (Limburg et al., 2020). Additionally, deoxygenation or increasing severe hypoxic periods in coastal areas may enhance the sea-to-air fluxes of N₂O and CH₄ especially through microbial-mediated processes in the water column–sediment interface (medium agreement) (Middelburg and Levin, 2009; Naqvi et al., 2010; Farías et al., 2015; Limburg et al., 2020).

There is high agreement (robust evidence) that heterogeneity implies different responses of coastal regions to increasing atmospheric CO₂, decreasing seawater pH and calcium carbonate saturation state, and deoxygenation (Duarte et al., 2013; Regnier et al., 2013; Breitburg et al., 2018; Laruelle et al., 2018; Carstensen and Duarte, 2019).

There is high agreement that long-time series of observations utilizing standard methods are needed to distinguish the climate change signal in seawater carbonate chemistry from the natural variability of coastal sites (Duarte et al., 2013; Salisbury and Jönsson, 2018; IOC, 2019; Sutton et al., 2019; Tilbrook et al., 2019; Turk et al., 2019). Despite the increasing availability of data and sea–air CO₂ flux budgets for the coastal ocean (Sections 5.3.5.1 and 5.2.3.1), additional long-term observations are required to constrain the global time of emergence of coastal acidification. There is high agreement (medium evidence) that, for the coastal subtropical to temperate north-east Pacific and north-west Atlantic, the mean time of emergence for acidification is above two decades (Sutton et al., 2019; Turk et al., 2019).

Observations and models predict an expansion and intensification of low-pH deep water intrusions for the north-east Pacific coastal upwelling area (high agreement, robust evidence) (Hauri et al., 2013; Feely et al., 2016; Cai et al., 2020). Areas such as the California Current System are naturally exposed to intrusions of low‐pH, high pCO_2sea deep waters from remineralization processes and anthropogenic CO₂ intrusion (Feely et al., 2008, 2010, 2018; Chan et al., 2019; Lilly et al., 2019; Cai et al., 2020).The eastern Pacific coastal upwelling displays seasonality in subsurface aragonite undersaturation as a consequence of the interplay between anthropogenic CO₂, respiration and intrusion of upwelling waters (Feely et al., 2008, 2010, 2016, 2018; Hauri et al., 2013; Vargas et al., 2016; Chan et al., 2019; Lilly et al., 2019). The coastal south-east Pacific upwelling combined with low-pH, low-alkalinity, organic matter-rich river inputs display extreme temporal variability in surface seawaterpCO₂ and low aragonite saturation (Vargas et al., 2016; Osma et al., 2020).

Temperate, non-upwelling coastal areas along the north-west Atlantic display a trend of decreasing seawater pH, mainly attributed to the combined effects of eutrophication and decreasing seawater buffering capacity (high agreement, robust evidence). Observations show an increasing north to south gradient of aragonite saturation state (Sutton et al., 2016; Fennel et al., 2019; Cai et al., 2020). Low alkalinity and total inorganic carbon concentration, combined with an ocean signal of acidification, diminishes the buffering capacity along the decreasing salinity gradient from the ocean to the coast. Models suggest that, in this area, the aragonite saturation is seasonally controlled by nutrient availability and primary production, supporting the finding that eutrophication is the main driver for exacerbating acidification (Cai et al., 2017, 2020). The coastal Gulf of Mexico is facing a parallel increase in bottom water acidification and deoxygenation off the Mississippi Delta driven by eutrophication (Cai et al., 2011; Laurent et al., 2017; Fennel et al., 2019).

Many coastal tropical areas are under heavy anthropogenic eutrophication induced by the effluents from large cities, or receive large riverine inputs of freshwater, nutrients, and organic matter (such as Amazon, Mississippi, Orinoco, Congo, Mekong, or Changjiang rivers). Under strong eutrophication, often sub-surface and bottom waters present pH values lower than average surface open ocean (about 8.0) because increased respiration decreases pH (high agreement, robust evidence), despite a net atmospheric CO₂ sink in shallow and vertically stratified coastal areas (Koné et al., 2009; Wallace et al., 2014; Cotovicz Jr. et al., 2015, 2018; Fennel and Testa, 2019; Lowe et al., 2019; Section 5.3.5.1).

There is medium evidence from observations and models that the coastal north-western Antarctic Peninsula (Southern Ocean) will experience calcium carbonate undersaturation by 2060, considering that anthropogenic emissions reach an atmospheric CO₂ concentration of about 500 pm at that date (Lencina-Avila et al., 2018; Monteiro et al., 2020a). The synergies among warming, melt water, sea-air CO₂ equilibrium and circulation may, to some extent, offset the coastal ocean acidification trends in Antarctica (Henley et al., 2020). In the coastal western Arctic Ocean, there is increasingrobust evidence that ocean acidification is driven by sea-air CO₂ fluxes and sea-ice melt, and increasing intrusions since the 1990s of low-alkalinity Pacific water, lowering aragonite saturation state (Qi et al., 2017, 2020; Cross et al., 2018). The Bering Sea (north-eastern Pacific) shows decreasing trends in calcium carbonate saturation, associated to the increasing atmospheric CO₂ uptake combined with riverine freshwater and carbon inputs (high agreement, robust evidence) (Pilcher et al., 2019; H. Sun et al., 2020).

The spatial distribution of hypoxic areas is highly heterogeneous in the coastal ocean, and there is high agreement, robust evidence that more severe hypoxia or anoxia is often associated with highly populated coastal areas,or local circulation and upwelling, and seasonal stratification leading to an accumulation of organic matter in subsurface waters (Ciais et al., 2013; Rabalais et al., 2014; M. Li et al., 2016; Breitburg et al., 2018; Bindoff et al., 2019; SROCC Chapter 5). The causes and trends of coastal deoxygenation can only be assessed by making available long-term time series combined with regional modelling (Fennel and Testa, 2019), as in the California current system (Wang et al., 2017), the East China Sea (Chen et al., 2007; Qian et al., 2017), the Namibian or along the north-western Atlantic shelves (Claret et al., 2018). Other coastal upwelling sites such as the Arabian Sea display seasonal hypoxia but no worsening trends (Gupta et al., 2016).

The Baltic Sea is the largest semi-enclosed sea where hypoxia is reported to have happened before the 1950s (Carstensen et al., 2014; Rabalais et al., 2014; Łukawska-Matuszewska et al., 2019). The frequency and volume of seawater inflow from the North Sea decreased after 1950, leading to an expansion of hypoxic areas from 40,000 to 60,000 km² in combination with increasing eutrophication (Carstensen et al., 2014). From the available observations, there is robust evidence that many areas in the Baltic Sea are experiencing deoxygenation despite efforts to reduce nutrient loads (Lennartz et al., 2014; Jokinen et al., 2018).

There is medium agreement (medium evidence) that simply reducing anthropogenic nutrient inputs may lead to less severe coastal hypoxic conditions, as observed in the coastal north-western Adriatic Sea (Djakovac et al., 2015). However, low-oxygen sediments may remain a long-term source of phosphorus and ammonium to the water column, and in this way fuelling primary production (Jokinen et al., 2018; Fennel and Testa, 2019; Limburg et al., 2020).

Plant productivity is highly dependent on local climate. In cold environments, warming has generally led to an earlier onset of the growing season, and with it an increase in early season vegetation productivity (e.g., Forkel et al., 2016). However, this trend is affected by the adverse effects of climate variability, and other emerging limitations on vegetation production by water, energy and nutrients, which may gradually reduce the effects of warming (Piao et al., 2017; Buermann et al., 2018; Liu et al., 2019). At centennial time scales, boreal forest expansion may act as a climate-driven carbon sink (Pugh et al., 2018).

In tropical and temperate environments, temperature simultaneously affects the metabolic rates of photosynthetic processes within leaf tissues, as well as the vapour pressure deficit that drives transpiration, its control by leaf stomata, and the resulting soil and plant tissue water content. Thus the direct effect of warming on photosynthesis can be positive, negative, or invariant depending on the environmental context (Lin et al., 2012; Yamori et al., 2014; Smith and Dukes, 2017; Grossiord et al., 2020). Observations and models suggest that the vapour pressure deficit effects are stronger than direct temperature effects on enzyme activities (Smith et al., 2020), and that acclimation of photosynthetic optimal temperature may mitigate productivity losses of tropical forests under climate change (Kattge and Knorr, 2007; Tan et al., 2017; Kumarathunge et al., 2019). Some models have begun to include these acclimation responses in photosynthesis and autotrophic respiration (Lombardozzi et al., 2015; Smith et al., 2015; Huntingford et al., 2017; Mercado et al., 2018).

The SRCCL assessed that climate change is playing an increasing role in determining wildfire regimes alongside human activity (medium confidence), with future climate variability expected to enhance the recurrence and severity of wildfires in many biomes, such as tropical rainforests (high confidence). Projections of increased fire weather in a warmer climate are widespread (Section 12.3.2.8) and may drive increased fire frequency and severity in several regions, including Arctic and boreal ecosystems (Gauthier et al., 2015; X.J. Walker et al., 2019), Mediterranean-type ecosystems (Turco et al., 2014; Jin et al., 2015), degraded tropical forests (Aragão et al., 2018), and tropical forest-savanna transition zones (Lehmann et al., 2014).

Wildfire is included in some CMIP6 ESMs (Table 5.4) and is thus only partially represented in estimates of carbon–climate feedbacks from these models. The CMIP5 ESMs that include fire project an 8–58% increase of fire carbon emissions under future scenarios, with higher emissions under higher warming scenarios; the ensemble spread is driven by differing factors such as population density, fire management, and other land-use processes (Kloster and Lasslop, 2017). Fire dynamics in CMIP6 models, as evaluated in land-only configurations of CMIP6-generation land surface models, also show large variations but better agreement with observations (Teckentrup et al., 2019; Hantson et al., 2020; Lasslop et al., 2020).

Climate change also drives changes to vegetation composition and ecosystem carbon storage through other disturbances such as forest dieback that lead to biome shifts in tropical forests (Cox et al., 2004; Jones et al., 2009; Brando et al., 2014; Le Page et al., 2017; Zemp et al., 2017), and temperate and boreal regions (Joos et al., 2001; Lucht et al., 2006; Scheffer et al., 2012; Lasslop et al., 2016). The AR5 assessed that large-scale loss of tropical forests due to climate change is unlikely (WGI, Section 6.4.9). Newer ecosystem modelling approaches that include a greater degree of ecosystem heterogeneity and diversity show a reduced sensitivity of such forest dieback-type changes (Levine et al., 2016; Sakschewski et al., 2016), supporting the AR5 assessment (Section 5.4.9). Beyond such biome shifts, observations of tropical forests also show that increasing tree mortality rates within tropical forests may reduce carbon turnover times and storage (Brienen et al., 2015), that increased tree mortality rates in tropical forests and elsewhere are expected with increased temperatures and vapour pressure deficit (Cross-Chapter Box 5.1; Allen et al., 2015; McDowell et al., 2018; Grossiord et al., 2020), and that these processes are not well represented in ESMs (Powell et al., 2013; Fisher et al., 2018). An ensemble of land models that includes ecological processes such as forest demography shows that changes to mortality may be a more important driver of carbon dynamics than changes to productivity (Friend et al., 2014).

Overall, climate change will force widespread increases in fire weather throughout the world (Section 12.3.2.8). Because of incomplete inclusion of fire in ESMs, a separate compilation of fire-driven carbon–climate feedback estimates is shown in Figure 5.29, based on results from Eliseev et al. (2014a) and Harrison et al. (2018). There is low agreement in magnitude and medium agreement in sign which leads to an assessment of medium confidence that fire represents a positive carbon–climate feedback, butvery low confidence in the magnitude of that feedback. Other disturbances such as tree mortality will increase across several ecosystems (medium agreement) with decreased vegetation carbon (medium confidence). However, the lack of model agreement and key process representation in ESMs leads to a low confidence assessment in the projected magnitude of this feedback.

Changes to soil carbon stocks in response to climate change are a potentially strong positive feedback (Cox et al., 2000). Since AR5 (WGI, Section 6.4.2), progress has been made in understanding soil carbon dynamics, and associated feedbacks. Advances include: (i) an increased understanding of and ability to quantify high-latitude soil carbon feedbacks (Box 5.1); (ii) increased understanding of the causes responsible for soil carbon persistence on long time scales, particularly the interactions between decomposers and soil organic matter and mineral assemblages (Kleber et al., 2007; Schmidt et al., 2011; Luo et al., 2016); and (iii) increased understanding of soil carbon dynamics in subsurface layers (Hicks Pries et al., 2017; Balesdent et al., 2018).

CMIP6 ESMs predict losses of soil carbon with warming, which are larger than climate-driven vegetation carbon losses (Arora et al., 2020). As in CMIP5 (Todd-Brown et al., 2013), there is also a large CMIP6 ensemble spread in climate-driven soil carbon changes, partially driven by a large spread in the current soil carbon stocks predicted by the models. In CMIP5 ESMs, much of the soil carbon losses with warming can be traced to decreased carbon inputs, with a weaker contribution from changing soil carbon lifetimes due to faster decomposition rates (Koven et al., 2015b), which may be an artefact of the lack of permafrost carbon (Box 5.1). Isotopic constraints suggest that CMIP5 ESMs systematically overestimated the transient sensitivity of soil¹⁴C responses to atmospheric¹⁴C changes, implying that the models respond too quickly to changes in either inputs or turnover times, and that therefore the soil contribution to all feedbacks may be weaker than currently projected (He et al., 2016). Using natural gradients of soil carbon turnover as a constraint on long-term responses to warming suggests that both CMIP5 and CMIP6 ESMs may systematically underestimate the temperature sensitivity at high latitudes, and may overestimate the temperature sensitivity in the tropics (Koven et al., 2017; Wieder et al., 2018; Varney et al., 2020), although experimental soil warming in tropical forests suggest high sensitivity of decomposition to warming in those regions as well (Nottingham et al., 2020).

Peat soils, where thick organic layers build up due to saturated and anoxic conditions, represent another possible source of carbon to the atmosphere. Peats could dry, and decompose or burn as a result of climate change in both high (Chaudhary et al., 2020) and tropical (Cobb et al., 2017) latitudes, and in combination with anthropogenic drainage of peatlands (Warren et al., 2017). Peat carbon dynamics are not included in the majority of CMIP6 ESMs. Soil microbial dynamics shift in response to temperature, giving rise to complex longer-term trophic effects that are more complex than the short-term sensitivity of decomposition to temperature. Such responses are observed in response to long-term warming experiments (Melillo et al., 2017). While most CMIP6 ESMs do not include microbial dynamics, simplified global soil models that do include such dynamics show greater uncertainty in projections of soil carbon changes, despite agreeing more closely with current observations, than the linear models used in most ESMs (Wieder et al., 2013; Guenet et al., 2018).

In nutrient-limited ecosystems, prolonged soil warming can induce a fertilization effect through increased decomposition, which increases nutrient availability and thereby vegetation productivity (Melillo et al., 2011). Models that include this process tend to show a weaker carbon–climate feedback than those that do not (Thornton et al., 2009; Zaehle et al., 2010; Wårlind et al., 2014; Meyerholt et al., 2020). In CMIP6, six out of 11 ESMs include a representation of the nitrogen cycle, and the mean of those models predicts a weaker carbon–climate feedback than the overall ensemble mean (Arora et al., 2020; Section 5.4.8). These models only partly account for the interactions of nutrient effects with other processes, such as shifts of vegetation zones under climate changes (Sakaguchi et al., 2016) leading to either changes in species composition or changes in plant tissue nutrient to carbon ratios (Thomas et al., 2015; Achat et al., 2016; Du et al., 2019).

The high agreement and multiple lines of evidence that warming increases decomposition rates lead to high confidence that warming will, overall, result in carbon losses relative to a constant climate and contribute to the positive carbon–climate feedback (Section 5.4.8). However, the wide spread in ESM projections and the lack of model representation of key processes that may amplify or mitigate soil carbon losses on longer time scales (including microbial dynamics, permafrost, peatlands, and nutrients) lead to low confidence in the magnitude of global soil carbon losses with warming.

The principal contribution to increasing global ocean carbon is the air–sea flux of CO₂, which changes the dissolved inorganic carbon (DIC) inventory (Section 5.4.2; Arora et al., 2020). The processes that influence the variability and trends of the ocean carbon–heat nexus are assessed in Cross-Chapter Box 5.3. Climate has three important impacts on the ocean uptake of anthropogenic CO₂ : (i) ocean warming reduces the solubility of CO₂, which increases pCO₂ and increases the stratification of the mixed layer, both acting as positive feedbacks weakening the ocean sink (Section 9.2.1 and Cross-Chapter Box 5.3; Arora et al., 2020); (ii) changing the temporal and spatial characteristics of wind stress and storms alters mixing – entrainment in, and across the bottom of, the mixed layer (Bronselaer et al., 2018); and (iii) warming and wind stress influence the large-scale meridional overturning circulation (MOC) circulation, which modifies the rate of ventilation, storage or outgassing of ocean carbon in the ocean interior (Section 5.2.3.1; Gruber et al., 2019b; Arora et al., 2020). The land-to-ocean riverine flux and the carbon burial in ocean sediments play a minor role (low confidence) (Arora et al., 2020). Based on high agreement of projections by coupled climate models, there is high confidence that the resultant climate–carbon cycle feedbacks are positive, but the extent of the ocean sink weakening is scenario dependent (Arora et al., 2020).

Regionally, the Southern Ocean is a major sink of anthropogenic CO₂ (Figure 5.8a), although challenges in modelling its circulation and Antarctic sea ice transport (Sections 3.4.1.2, 9.2.3.2 and 9.3.2) generate uncertainty in the response of its sink to future carbon–climate feedbacks. Increased freshwater input may cause a slowdown of the lower overturning circulation, leading to increased Southern Ocean biological carbon storage (Ito et al., 2015); alternatively, increased winds may intensify the overturning circulation, reducing the net CO₂ sink in the Southern Ocean (Bronselaer et al., 2018; Saunders et al., 2018). On centennial time scales, there is thus low confidence in the overall effect of intensifying winds in the Southern Ocean on CO₂ uptake.

While physical drivers control the present-day anthropogenic carbon sink, biological processes are responsible for the majority of the vertical gradient in DIC (natural carbon storage). A small fraction of the organic carbon fixed by primary production (PP) reaches the sea floor, where it can be stored in sediments on geological time scales, making the biological carbon pump (BCP) an important mechanism for very long-term CO₂ storage. Projected reductions in ocean ventilation (Section 9.2.1.4) would lengthen residence time and lead to DIC accumulating in the deep ocean due to organic carbon remineralization.

Since AR5 (Section 6.3.2.5.6), progress has been made in understanding the biological drivers of ocean carbon uptake in both coupled climate models and observations (SROCC, Section 5.2.2.6). Here we focus on potential feedbacks between biological processes and climate. In CMIP5 models, the direction of modelled PP in response to increased atmospheric CO₂ concentration and climate warming wasunclear (Taucher and Oschlies, 2011; Laufkötter et al., 2015). This remains the case in the CMIP6 models; inter-model uncertainty has increased in CMIP6 models, compared to CMIP5. The projected global multi-model mean change in PP in 13 models run under the SSP5−8.5 scenario is −3 ± 9% (2080–2099 mean values relative to 1870–1899 ± the inter-model standard deviation; Kwiatkowski et al., 2020). Under the low-emissions, high-mitigation scenario SSP1−2.6, the global change in PP is −0.56 ± 4%. Observations in the contemporary period provide little direct constraint on the modelled responses of PP to climate change, partly due to insufficiently long records (Henson et al., 2016). However, there is some indication of an emergent constraint on changes in tropical PP based on interannual variability derived from remote sensing (Section 5.4.6; Kwiatkowski et al., 2017).

In CMIP5 models run under RCP8.5, particulate organic carbon (POC) export flux is projected to decline by 1–12% by 2100 (Taucher and Oschlies, 2011; Laufkoetter et al., 2015). Similar values are predicted in 18 CMIP6 models, with declines of 2.5–21.5% (median –14%) or 0.2–2 GtC (median –0.8 GtC) between 1900 and 2100 under the SSP5-8.5 scenario. The mechanisms driving these changes vary widely between models due to differences in parametrization of particle formation, remineralization and plankton community structure.

Ocean warming reduces the vertical supply of nutrients to the upper ocean due to increasing stratification (Section 9.2.1.4) but may also act to alleviate seasonal light limitation. The projected effect is to decrease PP at low latitudes and increase PP at high latitudes (Kwiatkowski et al., 2020). Future changes to dust deposition due to desertification (Mahowald et al., 2017), alterations to the nitrogen cycle (Section 5.3.3.2; SROCC, Section 5.2.3.1.2), and reducing sea ice cover (Ardyna and Arrigo, 2020) all have the potential to alter PP regionally. Higher ocean temperatures tend to result in higher metabolic rates, although respiration may increase more rapidly than PP (Boscolo-Galazzo et al., 2018; Brewer, 2019; Cavan et al., 2019). Ocean warming and reduced PP are expected to result in lower zooplankton abundance, and the expansion of oxygen minimum zones (OMZs) may reduce the ability of zooplankton to remineralize POC, thus increasing the efficiency of the BCP and forming a negative climate feedback (Cavan et al., 2017). Increased microbial respiration due to warming may result in greater quantities of organic carbon transferred into the dissolved organic carbon pool (Jiao et al., 2014; Legendre et al., 2015; Roshan and DeVries, 2017) which, while increasing the residence time of carbon in the ocean, would ultimately reduce the sedimentary burial, and hence sequestration on geologic time scales (Olivarez Lyle and Lyle, 2006).

Most models project that smaller phytoplankton are favoured in future ocean conditions (medium confidence; Cabré et al., 2015; Fu et al., 2016; Flombaum et al., 2020) driven by warming water and/or changing nutrient availability, which would alter the magnitude and efficiency of the BCP by altering the sinking speed, respiration rate and aggregation/fragmentation of sinking particles. There is low confidence in the sign of the resulting feedback: regions in which small phytoplankton dominate may have a more efficient pump, although the total amount of organic carbon reaching the sea floor is lower (Herndl and Reinthaler, 2013; Bach et al., 2016; Richardson, 2019). Alternatively, an increase in small phytoplankton could result in a less efficient pump, due either to a greater fraction of PP being processed through the upper ocean microbial loop (Jiao et al., 2014) or generation of slower sinking particles (Guidi et al., 2009; Leung et al., 2021). Variable phytoplankton stoichiometry is predicted to increase the amount of carbon stored via the BCP relative to the amount of PP, so that fixed stoichiometry models (as in CMIP5) may underestimate cumulative ocean carbon uptake to 2100 by 0.5–3.5% (2–15 PgC; RCP8.5 scenario; Kwiatkowski et al., 2020). Other climate effects such as deoxygenation or ocean acidification could also result in alterations to the magnitude and efficiency of the BCP (Krumhardt et al., 2019; Raven et al., 2021; Taucher et al., 2021).

Based on high agreement across multiple lines of evidence and physical understanding there is high confidence that feedbacks to climate will arise from alterations to the magnitude and efficiency of the BCP changing PP, and the depth of remineralization. However, the complexity of the mechanisms involved in the export and remineralization of POC result in low confidence in the magnitude and sign of biological feedbacks to climate. Nevertheless, improved model representation of PP and the BCP is required (which requires better observational constraints), as the contribution of biological processes to CO₂ uptake is expected to become more significant with continued climate change (Hauck et al., 2015).

To give confidence in their projections, models need to be compared to the widest possible array of observational benchmarks. This is particularly the case for highly uncertain land carbon cycle feedbacks (Arora et al., 2013; Friedlingstein et al., 2014b). Land models within ESMs can be compared to multiple different datasets that test different aspects of the models. These include fluxes, such as gross carbon uptake, and states, such as leaf area and carbon stocks, which influence carbon fluxes and are diagnostic of carbon turnover times. Comparisons can also be made between between carbon and water cycles and other aspects of the terrestrial carbon cycle. To provide these multiple orthogonal constraints, a model benchmarking system – the international land model benchmarking (ILAMB) – has been developed (Collier et al., 2018).

Figure 5.22 shows an overview set of land (Figure 5.22a) and ocean (Figure 5.22b) benchmarks applied to both the CMIP5 and CMIP6 historical simulations. There is good evidence of an improvement in model performance from CMIP5 (in yellow) to CMIP6 (in green), in both the land and ocean, based on these benchmarks. The mean of the CMIP6 land models outperforms or performs equivalently to the mean of the CMIP5 land models on all available metrics.

Figure 5.22 | Overview scores for CMIP5 (left-hand side of table) and CMIP6 (right-hand side of table) Earth system models (ESMs), for multiple benchmarks against different datasets. (a) Benchmarking of ESM land models; (b) benchmarking of ocean models. Scores are relative to other models within each benchmark row, with positive scores indicating a better agreement with observations. Models included are only those from institutions that participated in both CMIP5 and CMIP6 carbon cycle experiments, in order to trace changes from one ensemble to the next. CMIP5 models are labels in blue and CMIP6 in red. Further details on data sources and processing are available in the chapter data table (Table 5.SM.6).

Open figure

This section evaluates concentration-driven historical simulations of changes in land and ocean cumulative carbon uptake, against observation-based estimates from the Global Carbon Project (GCP; Le Quéré et al., 2018a). For each model, common historical land-use changes were prescribed (Jones et al., 2016a).

Figure 5.23 shows global annual mean values from CMIP6 concentration-driven runs for 1850 to 2014. The ocean carbon cycle models reproduce historical carbon uptake well, with the model range for the global ocean carbon sink in 2014 (2.3–2.7 GtC yr^–1) clustering around the central GCP estimate of 2.6 ± 0.5 GtC yr^–1. Simulated cumulative ocean carbon uptake (1850–2014) ranges from 110 to 166 GtC, with a model mean of 131 ± 17 PgC, which is lower than the GCP estimate of 150 ± 25 GtC (Figure 5.23a). This suggests that CMIP6 models may slightly underestimate historical ocean carbon uptake (Watson et al., 2020).

Figure 5.23 | CMIP6 Earth system model (ESM) concentration-driven historical simulations for 1850 to 2014, compared to observation-based estimates from the global carbon project (GCP). (a) Cumulative ocean carbon uptake from 1850 (PgC); (b) cumulative land carbon uptake from 1850 (PgC). Only models that simulate both land and ocean carbon fluxes are shown here. Further details on data sources and processing are available in the chapter data table (Table 5.SM.6).

Open figure

The land carbon cycle components of historical ESM simulations show a larger range, with simulated cumulative land carbon uptake (1850–2014) spanning the range from –47 to +21 GtC, compared to the GCP estimate of –12 ± 50 GtC (Figure 5.23b). This range is due in part to the complications of simulating the difference between carbon uptake by intact ecosystems and the direct release of carbon due to land-use change (Hajima et al., 2020a). There is high confidence that the land continues to dominate the overall uncertainty in the projected response of the global carbon cycle to climate change.

This distinction between the relatively high fidelity with which the ocean carbon sink is simulated, and the much wider range of simulations of the land carbon sink, is also evident in the zonal distribution of the sinks (Figure 5.24). We compare the ESM simulations to estimates from three atmospheric inversion models: Copernicus Atmosphere Monitoring Service (CAMS; Chevallier et al., 2005), Carbon Tracker 2017 (Peters et al., 2007) and Model for Interdisciplinary Research on Climate Atmospheric Transport Model (MIROC-ATM4; Saeki and Patra, 2017). The ocean carbon cycle components of CMIP6 ESMs are able to simulate the tropical CO₂ source and mid-latitude CO₂ sink, with relatively small model spread (Figure 5.24a). The CMIP6 ensemble (red wedge) simulates a larger ocean carbon sink at 50°N and a weaker sink in the Southern Ocean, than the inversion estimate, but with some evidence of a reduction in these residual errors compared to CMIP5 (blue wedge). The spread in inversion fluxes arises primarily from differences in the atmospheric CO₂ measurement networks and from transport model uncertainties.

Figure 5.24 | Comparison of modelled zonal distribution of contemporary carbon sinks against atmospheric inversion estimates for 2000–2009: (a) ocean carbon uptake; (b) net land uptake. Latitude runs from 90°S (i.e., –90°N) to 90°N. Positive uptake represents a carbon sink to ocean/land while negative uptake represents a carbon source. The land uptake is taken as net biome productivity (NBP) and so includes net land-use change emissions. The bands show the mean ±1 standard deviation across the available inversions (black bands, 3 models), CMIP5 Earth system models (ESMs) (blue bands, 12 models for the ocean, 12 models for the land), and CMIP6 ESMs (red bands, 11 models for ocean, 10 models for land). Further details on data sources and processing are available in the chapter data table (Table 5.SM.6).

Open figure

It has been previously noted that AR5 models tended to overestimate land uptake in the tropics and underestimate uptake in the northern mid-latitudes, compared to inversion estimates. The inclusion of nitrogen limitations on CO₂ -fertilization within CMIP6 models was expected to reduce this discrepancy (Anav et al., 2013). There is indeed some evidence that the CMIP6 ensemble (red wedge in Figure 5.24b) captures the northern land carbon sink more clearly than CMIP5 (blue wedge in Figure 5.24b), but there remains a tendency for the ESMs to place more of the global land carbon sink in the tropics than the mid-latitudes, compared to the inversion estimates. Based on a consistent signal across CMIP6 ESMs, there is medium confidence that land carbon cycle models continue to underestimate the Northern Hemisphere land carbon sink, when compared to estimates from atmospheric inversion (Ciais et al., 2019).

Land and ocean carbon uptake are driven primarily by increases in atmospheric CO₂ (Figure 5.25). As a result, the evolution of land and ocean carbon sinks differs significantly between the SSP scenarios. Under scenarios that have greater increases in atmospheric CO₂ (such as SSP5-8.5 and SSP3-7.0) the absolute values of the sinks are larger, but the fraction of implied emissions taken up by the sinks declines through the 21^st century. By contrast, scenarios that assume CO₂ stabilization in the 21^st century (such as SSP1-2.6 or SSP2-4.5), have smaller absolute sinks, but these sinks take up an increasing fraction of the implied emissions (Figure 5.25d). These general principles apply to the ocean and land carbon sinks.

The concentration-driven CMIP6 ESMs agree well on the evolution of the global ocean carbon sink through the 21st century for four SSP scenarios (Figure 5.25). The five-year ensemble mean ocean sink declines to 0.6 ± 0.2 GtC yr^–1by 2100 under SSP1-2.6, and peaks around 2080 at 5.4 ± 0.4 GtC yr^–1under SSP5-8.5. Cumulative ocean carbon uptake from 1850 is projected to saturate at approximately 290 ± 30 GtC under SSP1-2.6, and to reach 520 ± 40 GtC by 2100 under SSP5-8.5 (Figure 5.25e).

The ensemble mean changes in land and ocean sinks are qualitatively similar, but the land shows much higher interannual variability in carbon uptake (Figure 5.25c) and also a much larger spread in the model projections of cumulative land carbon uptake (Figure 5.25f). The five-year ensemble mean net land carbon sink is projected to decline to 0.4 ± 1.0 GtC yr^–1by 2100 under SSP1-2.6, and to reach around 5.6 ± 3.7 GtC yr^–1under SSP5-8.5 (Figure 5.25c). Cumulative net land carbon uptake from 1850 is projected to saturate at approximately 150 ± 35 GtC under SSP1-2.6, and to reach 310 ± 130 GtC by 2100 under SSP5-8.5. Significant uncertainty remains in the future of the global land carbon sink, but there has been a notable reduction in the model spread from CMIP5 to CMIP6.

Figure 5.25 | Modelled evolution of the global land and ocean carbon sinks for 1900 to 2100 in concentration-driven CMIP6 Earth system model (ESM) scenario runs. (SSP1-2.6: blue; SSP2-4.5: orange; SSP3-7.0: red; SSP5-8.5: brown): (a) prescribed atmospheric CO₂ concentrations; (b) five-year running mean ocean carbon sink (GtC yr^–1); (c) five-year running mean net land carbon sink (GtC yr^–1); (d) inferred cumulative sink fraction of emissions from 1850; (e) change in ocean carbon storage from 1850 (GtC); (f) change in land carbon storage from 1850 (GtC). Thick lines represent the ensemble mean of the listed ESM runs, and the error bars represents ± 1 standard deviation about that mean. The grey wedges represent estimates from the global carbon project (GCP), assuming uncertainties in the annual mean ocean and net land carbon sinks of 0.5 GtC yr^–1 and 1 GtC yr^–1 respectively, and uncertainties in the changes in carbon stores (ocean, land and cumulative total emissions) of 25 GtC. The net land carbon sink is taken as net biome productivity (NBP) and so includes any modelled net land-use change emissions. Further details on data sources and processing are available in the chapter data table (Table 5.SM.6).

Open figure

Geographical patterns of carbon changes for four SSP scenarios are shown in Figure 5.26, with cleared areas (no diagonal lines) showing agreement on the sign of the change by at least 80% of the models. In all scenarios the ocean sink is strongest in the Southern Ocean and North Atlantic. The land carbon sink occurs primarily where there are present-day forests. In the mid- and high-northern latitudes, a carbon sink is projected as a result of the combined impacts of increasing CO₂ and warming (Section 5.4.5.5). Changes in land carbon storage in the tropics also depend strongly on the assumed rate of deforestation which varies in magnitude across the SSPs, from relatively low rates in SSP1-2.6 to relatively high rates in SSP3-7.0.

Figure 5.26 | Maps of net carbon changes under four Shared Socio-economic Pathway (SSP) scenarios, as evaluated from nine CMIP6 Earth system models. Uncertainty is represented using the simple approach (see Cross-Chapter Box Atlas.1 for more information). No overlay indicates regions with high model agreement, where ≥80% of models agree with the ensemble mean on the sign of change. Diagonal lines indicate regions with low model agreement, where <80% of models agree with the ensemble mean on the sign of change. On land, this is calculated as the time integral of net biome productivity (NBP), for the ocean it is the time-integral of air–sea carbon dioxide (CO₂) gas flux anomalies relative to the pre-industrial. Further details on data sources and processing are available in the chapter data table (Table 5.SM.6).

Open figure

In summary, oceanic and terrestrial carbon sinks are projected to continue to grow with increasing atmospheric concentrations of CO₂, but the fraction of emissions taken up by land and ocean is expected to decline as the CO₂ concentration increases (high confidence). In the ensemble mean, ESMs suggest approximately equal global land and ocean carbon uptake for each of the SSP scenarios. However, the range of model projections is much larger for the land carbon sink. Despite the wide range of model responses, uncertainty in atmospheric CO₂ by 2100 is dominated by future anthropogenic emissions rather than carbon–climate feedbacks (high confidence).

To diagnose the causes of the varying time-evolution of carbon sinks, the traditional linear feedback approach is adopted (Friedlingstein et al., 2003), as used previously to analyse C⁴MIP (Friedlingstein et al., 2006) and CMIP5 models (Arora et al., 2013). Changes in land carbon storage (ΔC_L) and changes in ocean carbon storage (ΔC_o) are decomposed into contributions arising from warming (ΔT) and increases in CO₂ (ΔCO2):

whereβ_L (β_o) and γ _L (γ _o) are coefficients that represent the sensitivity of land (ocean) carbon storage to changes in CO₂ and global mean temperature respectively. This feedback formalism is one of several that have been proposed for analysing climate–carbon cycle feedbacks (Lade et al., 2018).

This quasi-equilibrium framework is scenario dependent because of the time scales associated with land and ocean carbon uptake, as discussed in AR5 (WGI, Box 6.4). However, it is retained here for traceability with AR5. This approach has been used to define a number of emergent constraints on carbon cycle feedbacks (Section 5.4.6) and to reconstruct the transient climate response to cumulative CO₂ emissions (TCRE) (Jones and Friedlingstein, 2020), as in Section 5.5. To minimize the confounding effect of the scenario dependence, β and γvalues are diagnosed from idealized runs in which a 1% per year increase in atmospheric CO₂ concentration is prescribed, as for AR5 (WGI, Box 6.4; Arora et al., 2013). Values ofβare calculated from ‘biogeochemical’ runs in which the prescribed CO₂ increases do not affect climate, and these are then used to isolateγvalues in fully coupled runs where both climate and CO₂ change (Friedlingstein et al., 2003).

Table 5.5 shows the global land and global ocean values ofβ and γ for each of the CMIP6 ESMs (Arora et al., 2020). The last two rows show the ensemble means and standard deviation across the ensemble for CMIP6 and CMIP5. In both ensembles, the largest uncertainties are in the sensitivity of land carbon storage to CO₂ (β_L) and the sensitivity of land carbon storage to temperature (γ _L). The more widespread modelling of nitrogen limitations in CMIP6 was expected to lead to reductions in both of these feedback parameters. There is some evidence for that, with ensemble meanγ _L moving from –58 ± 38 GtC K^–1 to –33 ± 33 GtC K^–1. Between CMIP5 and CMIP6, there are also reductions in ensemble meanβ_o (0.82 to 0.77 GtC ppm^–1), β_L (0.93 to 0.89 GtC ppm^–1) and γ _o (–17.3 to –16.9 GtC K^–1), but these are progressively less significant compared to the model spread in each case.

In these idealized 1% per year CO₂ runs, the CMIP6 models show reasonable agreement on the patterns of carbon uptake and also on the separate impacts of CO₂ increase and climate change (Figure 5.27). For the ensemble mean, increasing atmospheric CO₂ increases carbon uptake by the oceans, especially in the Southern Ocean and the North Atlantic Ocean, and on the land, especially in tropical and boreal forests (β, Figure 5.27a). Climate change further enhances land carbon storage in the boreal zone, but has a compensating negative impact on the carbon sink in tropical and subtropical lands, and in the North Atlantic Ocean (γ , Figure 5.27b). Overall, the ensemble mean of the CMIP6 ESMs model indicates increasing carbon storage with CO₂ in almost all locations (Figure 5.27c).

Figure 5.27 | Maps of carbon-concentration and carbon–climate feedback terms, as well as net carbon changes under the idealized 1% per year carbon dioxide (CO₂ ) scenario, as evaluated from CMIP6 Earth system models (ESMs). Shown are the model means from nine CMIP6 ESMs. Uncertainty is represented using the simple approach (see Cross-Chapter Box Atlas.1 for more information): No overlay indicates regions with high model agreement, where ≥80% of models agree with the ensemble mean on the sign of change; diagonal lines indicate regions with low model agreement, where <80% of models agree with the ensemble mean on the sign of change. Also shown are zonal-mean latitude profiles of land (green) and ocean (blue) feedbacks. On the land, the zonal mean feedback for the mean of the ensemble of models that include nitrogen is shown as dashed lines, and for carbon-only models as dash-dotted lines, and the carbon–climate feedback from one permafrost-carbon enabled ESM is shown as a dotted line. Carbon changes are calculated as the difference between carbon stocks at different times on land and for the ocean as the time integral of atmosphere–ocean CO₂ flux anomalies relative to the pre-industrial. The denominator for gamma here is the global mean surface air temperature. Further details on data sources and processing are available in the chapter data table (Table 5.SM.6).

Open figure

Projecting abrupt changes is intrinsically difficult, because by definition abrupt changes occur in a small region of the parameter and/or forcing space. At the time of AR5 there was no available systematic study of abrupt changes or tipping points in ESMs. An analysis of ESMs since AR5 has identified a number of abrupt changes in the CMIP5 ensemble (Drijfhout et al., 2015; Bathiany et al., 2020). These include abrupt changes in tropical forests and high-latitude greening, permafrost thaw, and vegetation composition change (Bathiany et al., 2020). Most modelled abrupt changes were detected in boreal and tundra regions, with few models showing Amazon forest dieback (Bathiany et al., 2020).

Based on the evidence presented in this section, we conclude that abrupt changes and tipping points in the biogeochemical cycles lead to additional uncertainty in 21st century GHG concentrations changes. However, these are very likely to be small compared to the uncertainty associated with future anthropogenic emissions (high confidence).

The processes that translate emissions of CO₂ into a change in global temperature (terrestrial and oceanic carbon uptake, radiative forcing from CO₂, and ocean heat uptake) are governed by complex mechanisms that all evolve in time (Sections 3.5, 4.3, 4.5, 5.4, and 7.3, and Cross-Chapter Box 5.3; Gregory et al., 2009). Starting with an initial description in AR5 (M. Collins et al., 2013; T.F. Stocker et al., 2013), a body of literature has since expanded the understanding of physical mechanisms from which a simple proportional relationship between cumulative emissions of CO₂ and change in global temperature arises – expressed in either global mean surface temperature (GMST) or global surface air temperature (GSAT).

Studies have focused on two key features of the transient climate response to cumulative CO₂ emissions (TCRE) relationship: (i) why the relationship is nearly constant in time (Goodwin et al., 2015; MacDougall and Friedlingstein, 2015; Williams et al., 2016; Ehlert et al., 2017; Katavouta et al., 2018); and (ii) why, and under which conditions, the relationship is independent of the historical rate (or pathway) of CO₂ emissions (MacDougall, 2017; Seshadri, 2017).

There is increased confidence in the near-constancy of TCRE because of the variety of methods that have been used to examine this relationship: sensitivity studies with Earth system models of intermediate complexity (EMICs; Herrington and Zickfeld, 2014; Ehlert et al., 2017); theory-based equations used to examine ESM and EMIC output (Goodwin et al., 2015; R.G. Williams et al., 2016, 2017b); and simple analytical models that capture aspects of the TCRE relationship (MacDougall and Friedlingstein, 2015). All studies agree that the near-constancy of the TCRE arises from compensation between the diminishing sensitivity of radiative forcing to CO₂ at higher atmospheric concentration and the diminishing ability of the ocean to take up heat and carbon at higher cumulative emissions (Allen et al., 2009; Matthews et al., 2009; Frölicher and Paynter, 2015; Goodwin et al., 2015; Gregory et al., 2015; MacDougall and Friedlingstein, 2015; MacDougall, 2016; Tokarska et al., 2016; Ehlert et al., 2017).

The question of whether, and under which conditions, the TCRE relationship is independent of the historical rate of CO₂ emissions (also referred to as ‘pathway independence of TCRE’) has been examined by using simple mathematically tractable models (MacDougall, 2017; Seshadri, 2017). Based on the assumption that the cumulative fraction of carbon taken up by the terrestrial biosphere is constant, and that the climate feedback parameter and ocean heat uptake efficacy do not change in time, both studies agree that pathway independence is sensitive to the rate of CO₂ emissions, such that pathway independence is expected to break down at both very high and very low absolute CO₂ emissions rates (MacDougall, 2017; Seshadri, 2017). Note that, in pathways with strongly declining emissions, the cumulative sink fraction by the combined terrestrial biosphere and ocean is expected to increase (Figure 5.25). The studies also agree that no similar relationship analogous to TCRE can be expected for short-lived non-CO₂ forcers, for which the annual emissions are a closer proxy for the implied warming (M. Collins et al., 2013; Sections 6.4, 7.6). MacDougall (2017) suggests that two additional constraints are required to create pathway independence: first, the transport of heat and carbon into the deep ocean should be governed by processes with similar time scales; and second, the ratio of the net change in the atmospheric carbon pool to the net change in the ocean carbon pool should be close to the ratio of the enhanced longwave radiation to space (i.e., the radiative response of the surface) to ocean heat uptake. If these ratios are identical, then TCRE would be completely path independent (MacDougall, 2017). If the ratios are close but not identical, TCRE would be only approximately path independent over a wide range of cumulative emissions (Cross-Chapter Box 5.3; MacDougall, 2017).

The land carbon cycle does not appear to play a fundamental role in the origin of the linearity and path-independence of TCRE (Goodwin et al., 2015; MacDougall and Friedlingstein, 2015; Ehlert et al., 2017) but, in contrast to the ocean sink, dominates the uncertainty in the magnitude of TCRE by modulating the cumulative airborne fraction of carbon (Goodwin et al., 2015; Williams et al., 2016; Katavouta et al., 2018; Jones and Friedlingstein, 2020). Some terrestrial carbon cycle feedbacks (such as the permafrost carbon feedback; Section 5.4.8, Box 5.1) have the potential to alter both the linearity and pathway independence of TCRE, if such feedbacks significantly contribute carbon to the atmosphere (Sections 5.5.1.2.3 and 5.4.8, and Box 5.1; MacDougall and Friedlingstein, 2015). A recent study also shows how the value of TCRE can depend on the effect of ocean ventilation modulating ocean heat uptake (Katavouta et al., 2019).

The AR5 (M. Collins et al., 2013) assessed that the TCRE is likely to fall in the range of 0.8°C –2.5°C per 1000 PgC (or per exagrams of carbon, EgC^–1) for cumulative emissions up to 2000 PgC, based on multiple lines of evidence. These include estimates based on ESMs of varying complexity (Matthews et al., 2009; Gillett et al., 2013; Zickfeld et al., 2013), simple climate modelling approaches (Allen et al., 2009; Rogelj et al., 2012) or observational constraints and attributable warming (Gillett et al., 2013).

Since AR5, new studies have further expanded the evidence base for estimating the value of TCRE. These studies rely on ESMs or EMICs, observational constraints and concepts of attributable warming, or theoretically derived equations (see Table 5.7 for an overview). Several studies have endeavoured to partition the uncertainty in the value of TCRE into constituent sources. For example, TCRE can be decomposed into terms of TCR and the airborne fraction of anthropogenic CO₂ emissions over time (Allen et al., 2009; Matthews et al., 2009). These two terms are assessed individually (see Section 5.4 and Chapter 7, respectively) and allow the integration of evidence assessed elsewhere in the report into the assessment of TCRE (Section 5.5.1.4). Further studies use a variety of methods, including analysing the outputs from CMIP5 (R.G. Williams et al., 2017b) or CMIP6 (Arora et al., 2020; Jones and Friedlingstein, 2020), conducting perturbed parameter experiments with a single model (MacDougall et al., 2017), Monte-Carlo methods applied to a simple climate model (Spafford and Macdougall, 2020), or observations and estimates of the contribution of CO₂ and non-CO₂ forcers (Matthews et al., 2021). All of the studies agree that uncertainty in climate sensitivity – either equilibrium climate sensitivity (ECS) or transient climate response (TCR) – is among the most important contribution to uncertainty in TCRE, with uncertainty in the strength of the land carbon feedback and ocean heat uptake or ventilation having also been identified as crucial to uncertainty in TCRE (Matthews et al., 2009; Gillett et al., 2013; Ehlert et al., 2017; MacDougall et al., 2017; R.G. Williams et al., 2017a, 2020; Katavouta et al., 2019; Arora et al., 2020; Jones and Friedlingstein, 2020; Spafford and Macdougall, 2020). Finally, internal variability has been shown to affect the maximum accuracy of TCRE estimates by ±0.1°C per 1000 PgC (5–95% range; Tokarska et al., 2020).

Studies differ in how they define TCRE, in the methods they use, and their assumptions, such as the assumed climate sensitivity distribution or the choice of metrics of global temperature change (e.g., GMST or GSAT, see Table 5.7). This makes TCRE estimates from individual studies difficult to compare. The combined assessment of TCRE therefore takes advantage of the well-established decomposition of TCRE in two factors: the TCR and the AF (Section 5.5.1.3). This provides a TCRE assessment range for CO₂ -induced warming at the time of doubling CO₂ concentrations that builds on the broader Working Group 1 assessment. Expert judgement based on the airborne fraction range found in CMIP6 models (Arora et al., 2020; Jones and Friedlingstein, 2020) suggest a value of 53% with a 1-sigma range of ±6%, which is double the sigma range based on the spread of CMIP6 models only. Combining this range with the AR6 TCR assessment (Section 7.5, best estimate 1.8°C, 1.4°C–2.2°Clikely and 1.2°C–2.4°Cvery likely range) results in a 5–95% range of 1.0–2.3°C per 1000 PgC (0.27°C–0.63°C per 1000 GtCO₂). Based on expert judgement that accounts for the incomplete coverage of all Earth system components, this results in a consolidated assessment that TCRE would fall likely in the range of 1.0–2.3°C per 1000 PgC, with a best estimate of 1.65°C per 1000 PgC (0.45°C per 1000 GtCO₂). Warming here reflects the human-induced GSAT increase and assumes a normal distribution. Some studies using observational constraints support a lognormal shape for the TCRE distribution (Spafford and Macdougall, 2020), but such a distribution is currently not supported by the combined assessment of TCR and airborne fraction. Finally, this assessed TCRE range needs to be considered in combination with the ZEC (Section 4.7.1.1) when estimating the CO₂ -induced warming of low-emissions scenarios.

The AR6 Glossary (Annex VII) defines remaining carbon budgets as the maximum amount of cumulative net global anthropogenic CO₂ emissions expressed from a recent specified date that would result in limiting global warming to a given level with a given probability, taking into account the effect of other anthropogenic climate forcers, consistent with the definition used inSR1.5 (Rogelj et al., 2018b). Studies, however, apply a variety of definitions that result in published remaining carbon budget estimates informing to cumulative emissions at the time when global-mean temperature increase would reach, exceed, avoid, or peak at a given warming level with a given probability (M. Collins et al., 2013; T.F. Stocker et al., 2013; Clarke et al., 2014; Friedlingstein et al., 2014a; IPCC, 2014; Rogelj et al., 2016; Millar et al., 2017b).

This section provides an assessment of remaining carbon budgets consistent with the AR6 Glossary definition (Annex VII). Given that some feedbacks are time dependent, the values in this section apply to limiting warming over the 21st century, consistent with recent studies highlighting the usefulness of time-limited carbon budgets (Sanderson, 2020). Irrespective of the exact definition of the remaining carbon budget, the finding that higher cumulative CO₂ emissions lead to higher temperatures implies that annual net CO₂ emissions have to decline to close to zero in order to halt global warming, whether at 1.5°C, 2°C or another level (Allen et al., 2018).

Two approaches were used in AR5 to determine carbon budgets (M. Collins et al., 2013; T.F. Stocker et al., 2013; Clarke et al., 2014; IPCC, 2014; Rogelj et al., 2016). Working Group I (WGI) reported threshold exceedance budgets (TEB) that correspond to the amount of cumulative CO₂ emissions at the time a specific temperature threshold is exceeded, with a given probability in a particular greenhouse-gas and aerosol (pre-cursor) emissions scenario (M. Collins et al., 2013; IPCC, 2013b; T.F. Stocker et al., 2013). WGI also reported TEBs for the hypothetical case that only CO₂ would be emitted by human activities (M. Collins et al., 2013; IPCC, 2013b; T.F. Stocker et al., 2013). The AR5 Working Group III used threshold avoidance budgets (TAB) that correspond to the cumulative CO₂ emissions over a given time period of a subset of greenhouse-gas and aerosol (precursor) emissions scenarios in which global-mean temperature increase stays below a specific temperature threshold with at least a given probability (Clarke et al., 2014). The AR5 synthesis report used TABs defined until the time of peak warming over the 21st century (IPCC, 2014). Drawbacks have been identified for TEBs and TABs (Rogelj et al., 2016). TABs provide an estimate of the cumulative CO₂ emissions under pathways that have as a common characteristic the fact that they do not exceed a specific global warming threshold. However, the actual level of maximum warming can vary between pathways, leading to an unnecessary and poorly constrained spread in TAB estimates (Rogelj et al., 2016). Therefore, the TAB approach typically does not result in accurate projections of the remaining carbon budget. One drawback of TEBs is that they provide an estimate of the cumulative CO₂ emissions at the time global warming crosses a given threshold of interest in a specific emissions scenario – for example, most of the standard scenarios used in climate change research, such as the Representative Concentration Pathways (RCPs) or Shared Socio-economic Pathways (SSPs), exceed global warming of 1.5°C or 2°C (see Cross-Chapter Box 1.5) (M. Collins et al., 2013; T.F. Stocker et al., 2013; Friedlingstein et al., 2014a; Millar et al., 2017b). Because of potential variations in non-CO₂ warming at that point in time, or potential lags of about a decade in CO₂ warming (Joos et al., 2013; Ricke and Caldeira, 2014; Rogelj et al., 2015a, 2016, 2018b; Zickfeld and Herrington, 2015) TEBs also do not provide a precise estimate of the remaining carbon budget for limiting warming to a specific level.

Since the publication of AR5 (M. Collins et al., 2013), several new approaches have been proposed that provide a solution to the identified limitations of TABs and TEBs. Most of these approaches indirectly rely on the concept of TCRE (Section 5.5.1), for example, because they estimate modelled cumulative CO₂ emissions until a temperature threshold is crossed and use this budget to infer insights for pathways that attempt to limit warming to below this threshold and thus need to follow a different path (Friedlingstein et al., 2014a; Matthews et al., 2017; Millar et al., 2017b; Goodwin et al., 2018; Tokarska and Gillett, 2018). In this report, the assessment framework of SR1.5 for remaining carbon budgets is applied (Rogelj et al., 2018b, 2019). This framework allows integration of multiple lines of evidence to assess the contributions of five components that together result in a consolidated assessment of the remaining carbon budget (TCRE, historical human-induced warming, non-CO₂ warming, the ZEC, and adjustments due to additional Earth system feedbacks, see Section 5.5.2.2). It builds on the advances in estimating remaining carbon budgets or related quantities that have been published since AR5 (Rogelj et al., 2015a; Haustein et al., 2017; Matthews et al., 2017, 2021;Millar et al., 2017b; Gasser et al., 2018; Lowe and Bernie, 2018; Tokarska et al., 2018; Nicholls et al., 2020).

Recent studies suggest further changes to this framework by including non-linear adjustments to the TCRE contribution (Nicholls et al., 2020), or including non-CO₂ forcers in different ways by accounting for their different forcing effects (Matthews et al., 2021). Figure 5.31 provides a conceptual schematic of how the various individually assessed contributions are combined into a consolidated assessment of the remaining carbon budget. Together with estimates of historical CO₂ emissions to date (Section 5.2.1), these remaining carbon budgets provide the overall amount of cumulative CO₂ emissions consistent with limiting global warming to specific levels. A comparison with the approach applied in AR5 (M. Collins et al., 2013; Clarke et al., 2014) is available in SR1.5Section 2.2.2 (Rogelj et al., 2018b) as well as Box 5.2.

Figure 5.31 | Illustration of relationship between cumulative emissions of carbon dioxide (CO₂ ) and global mean surface air temperature (GSAT) increase (left) and conceptual schematic of the assessment of the remaining carbon budget from its constituting components (right). Carbon budgets consistent with various levels of additional warming are provided in Table 5.8 and should not be read from the illustrations in either panel. Left-hand panel: Historical data (thin black line data) shows historical CO₂ emissions as reported in Friedlingstein et al. (2020) together with the assessed global mean surface air temperature increase from 1850–1900 as assessed in Chapter 2 (Box 2.3, GSAT). The orange-brown range with its central line shows the estimated human-induced share of historical warming (Haustein et al., 2017). The vertical orange-brown line shows the assessed range of historical human-induced warming for the 2010–2019 period relative to 1850–1900 (Chapter 3). The grey cone shows the assessed range for the transient climate response to cumulative CO₂ emissions (TCRE) assessed to fall likely in the 1.0°C–2.3°C per 1000 PgC range (Section 5.5.1.4), starting from 2015. Thin coloured lines show CMIP6 simulations for the five scenarios of the AR6 core set (SSP1-1.9, sky blue; SSP1-2.6, dark blue; SSP2-4.5, yellow; SSP3-7.0, red; SSP5-8.5, maroon), starting from 2015. Diagnosed carbon emissions (Arora et al., 2020) are complemented with estimated land-use change emissions for each respective scenario (Gidden et al., 2019). Coloured areas show the Chapter 4 assessed very likely range of GSAT projections and thick coloured central lines the median estimate, for each respective scenario. These projections are expressed relative to the cumulative CO₂ emissions that are available for emissions-driven CMIP6 ScenarioMIP experiments for each respective scenario (Riahi et al., 2017; Rogelj et al., 2018a; Gidden et al., 2019). Right-hand panel: schematic illustration of assessment of remaining carbon budget based on multiple lines of evidence. The remaining allowable warming is estimated by combining the global warming limit of interest with the assessed historical human induced warming (Section 5.5.2.2.2), the assessed future potential non-CO₂ warming contribution (Section 5.5.2.2.3) and the zero emissions commitment (Section 5.5.2.2.4). Note that contributions in the right-hand panel are illustrative and contributions are not to scale. For example, the central ZEC estimate was assessed to be zero. The remaining allowable warming (vertical blue bar) is subsequently combined with the assessed TCRE (Sections 5.5.1.4 and 5.5.2.2.1) and contribution of unrepresented Earth system feedbacks in models used to estimate ZEC and TCRE (Section 5.5.2.2.5) to provide an assessed estimate of the remaining carbon budget (horizontal blue bar; see Table 5.8). Further details on data sources and processing are available in the chapter data table (Table 5.SM.6).

Open figure