Chapter 5: Global Carbon and other Biogeochemical Cycles and Feedbacks

Coordinating Lead Authors:

Josep G. Canadell (Australia), Pedro M.S. Monteiro (South Africa)

Lead Authors:

Marcos H. Costa (Brazil), Leticia Cotrim da Cunha (Brazil), Peter M. Cox (United Kingdom), Alexey V. Eliseev (Russian Federation), Stephanie Henson (United Kingdom), Masao Ishii (Japan), Samuel Jaccard (Switzerland), Charles Koven (United States of America), Annalea Lohila (Finland), Prabir K. Patra (Japan/India), Shilong Piao (China), Joeri Rogelj (United Kingdom/Belgium), Stephen Syampungani (Zambia), Sönke Zaehle (Germany), Kirsten Zickfeld (Canada/Germany)

Contributing Authors:

Georgii A. Alexandrov (Russian Federation), Govindasamy Bala (India/United States of America), Laurent Bopp (France), Lena Boysen (Germany), Long Cao (China), Naveen Chandra (Japan/India), Philippe Ciais (France), Sergey N. Denisov (Russian Federation), Frank J. Dentener (EU, The Netherlands), Hervé Douville (France), Amanda Fay (United States of America), Piers Forster (United Kingdom), Baylor Fox-Kemper (United States of America), Pierre Friedlingstein (United Kingdom), Weiwei Fu (United States of America/China), Sabine Fuss (Germany), Véronique Garçon (France), Bettina Gier (Germany), Nathan P. Gillett (Canada), Luke Gregor (Switzerland/South Africa), Karsten Haustein (United Kingdom/Germany), Vanessa Haverd (Australia), Jian He (United States of America/China), Helene T. Hewitt (United Kingdom), Forrest M. Hoffman (United States of America), Tatiana Ilyina (Germany), Robert B. Jackson (United States of America), Christopher Jones (United Kingdom), David P. Keller (Germany/United States of America), Lester Kwiatkowski (France/United Kingdom), Robin D. Lamboll (United Kingdom/United States of America, United Kingdom), Xin Lan (United States of America/China), Charlotte Laufkötter (Switzerland/Germany), Corinne Le Quéré (United Kingdom), Andrew Lenton (Australia), Jared Lewis (Australia/New Zealand), Spencer Liddicoat (United Kingdom), Laura Lorenzoni (United States of America/Venezuela), Nicole Lovenduski (United States of America), Andrew H. MacDougall (Canada), Sabine Mathesius (Canada/Germany), H. Damon Matthews (Canada), Malte Meinshausen (Australia/Germany), Igor I. Mokhov (Russian Federation), Vaishali Naik (United States of America), Zebedee R. J. Nicholls (Australia), Intan Suci Nurhati (Indonesia), Michael O’Sullivan (United Kingdom), Glen Peters (Norway), Julia Pongratz (Germany), Benjamin Poulter (United States of America), Jean-Baptiste Sallée (France), Marielle Saunois (France), Edward A.G. Schuur (United States of America), Sonia I. Seneviratne (Switzerland), Ann Stavert (Australia), Parvadha Suntharalingam (United Kingdom/United States of America), Kaoru Tachiiri (Japan), Jens Terhaar (Switzerland/Germany), Rona Thompson (Norway, Luxembourg/New Zealand), Hanqin Tian (United States of America), Jocelyn Turnbull (New Zealand), Sergio M. Vicente-Serrano (Spain), Xuhui Wang (China), Rik Wanninkhof (United States of America), Philip Williamson (United Kingdom)

Review Editors:

Victor Brovkin (Germany/Russian Federation), Richard A. Feely (United States of America)

Chapter Scientist:

Alice D. Lebehot (South Africa/France)

Figure 5.1

Figure 5.2

Figure 5.3

Figure 5.4

Figure 5.5

Figure 5.6

Figure 5.7

Figure 5.8

Figure 5.9

Figure 5.10

Figure 5.11

Figure 5.12

Figure 5.13

Figure 5.14

Figure 5.15

Figure 5.16

Figure 5.17

Figure 5.18

Figure 5.19

Figure 5.20

Figure 5.21

Figure 5.22

Figure 5.23

Figure 5.24

Figure 5.25

Figure 5.26

Figure 5.27

Figure 5.28

Figure 5.29

Figure 5.30

Figure 5.31

Figure 5.32

Figure 5.33

Figure 5.34

Figure 5.35

Figure 5.36

Figure 5.37

Cross-Chapter Box 5.2, Figure 1

Cross-Chapter Box 5.2, Figure 2

Cross-Chapter Box 5.3, Figure 1

FAQ 5.1 Figure 1

FAQ 5.2, Figure 1

FAQ 5.3, Figure 1

FAQ 5.4, Figure 1

This chapter should be cited as:

Canadell, J.G., P.M.S. Monteiro, M.H. Costa, L. Cotrim da Cunha, P.M. Cox, A.V. Eliseev, S. Henson, M. Ishii, S. Jaccard, C. Koven, A. Lohila, P.K. Patra, S. Piao, J. Rogelj, S. Syampungani, S. Zaehle, and K. Zickfeld, 2021: Global Carbon and other Biogeochemical Cycles and Feedbacks. In Climate Change 2021: The Physical Science Basis. Contribution of Working Group I to the Sixth Assessment Report of the Intergovernmental Panel on Climate Change [Masson-Delmotte, V., P. Zhai, A. Pirani, S.L. Connors, C. Péan, S. Berger, N. Caud, Y. Chen, L. Goldfarb, M.I. Gomis, M. Huang, K. Leitzell, E. Lonnoy, J.B.R. Matthews, T.K. Maycock, T. Waterfield, O. Yelekçi, R. Yu, and B. Zhou (eds.)]. Cambridge University Press, Cambridge, United Kingdom and New York, NY, USA, pp. 673–816, doi: 10.1017/9781009157896.007.

Executive Summary

It is unequivocal that the increases in atmospheric carbon dioxide (CO2), methane (CH4) and nitrous oxide (N2O) since the pre-industrial period are caused by human activities. The accumulation of GHGs in the atmosphere is determined by the balance between anthropogenic emissions, anthropogenic removals, and physical-biogeochemical source and sink dynamics on land and in the ocean. This chapter assesses how physical and biogeochemical processes of the carbon and nitrogen cycles affect the variability and trends of GHGs in the atmosphere as well as ocean acidification and deoxygenation. It identifies physical and biogeochemical feedbacks that have affected (or could affect) future rates of GHG accumulation in the atmosphere, and therefore, influence climate change and its impacts. This chapter also assesses the remaining carbon budget to limit global warming within various goals, as well as the large-scale consequences of carbon dioxide removal (CDR) and solar radiation modification (SRM) on biogeochemical cycles. {Figures 5.1, 5.2}.

The Human Perturbation of the Carbon and Biogeochemical Cycles

Global mean concentrations for well-mixed GHGs (CO2 , CH4 and N2 O) in 2019 correspond to increases of about 47%, 156%, and 23%, respectively, above the levels in 1750 (representative of the pre-industrial era) (high confidence). Current atmospheric concentrations of the three GHGs are higher than at any point in the last 800,000 years, and in 2019 reached 409.9 parts per million (ppm) of CO2, 1866.3 parts per billion (ppb) of CH4, and 332.1 ppb of N2O (very high confidence). Current CO2 concentrations in the atmosphere are also unprecedented in the last 2 million years (high confidence). In the past 60 million years, there have been periods in Earth’s history when CO2 concentrations were significantly higher than at present, but multiple lines of evidence show that the rate at which CO2 has increased in the atmosphere during 1900–2019 is at least 10 times faster than at any other time during the last 800,000 years (high confidence), and 4–5 times faster than during the last 56 million years (low confidence). {5.1.1, 2.2.3; Figures 5.3, 5.4; Cross-Chapter Box 2.1}

Contemporary Trends of Greenhouse Gases

It is unequivocal that the increase of CO2 , CH4 , and N2 O in the atmosphere over the industrial era is the result of human activities (very high confidence). This assessment is based on multiple lines of evidence including atmospheric gradients, isotopes, and inventory data. During the last measured decade, global average annual anthropogenic emissions of CO2, CH4, and N2O, reached the highest levels in human history at 10.9 ± 0.9 petagrams of carbon per year (PgC yr–1, 2010–2019), 335–383 teragrams of methane per year (TgCH4yr–1, 2008–2017), and 4.2–11.4 teragrams of nitrogen per year (TgN yr–1, 2007–2016), respectively (high confidence). {5.2.1, 5.2.2, 5.2.3, 5.2.4; Figures 5.6, 5.13, 5.15}.

The CO2 emitted from human activities during the decade of 2010–2019 (decadal average 10.9 ± 0.9PgC yr–1) was distributed between three Earth system components: 46% accumulated in the atmosphere (5.1 ± 0.02PgC yr–1), 23% was taken up by the ocean (2.5 ± 0.6PgC yr–1) and 31% was stored by vegetation in terrestrial ecosystems (3.4 ± 0.9PgC yr–1) (high confidence). Of the total anthropogenic CO2 emissions, the combustion of fossil fuels was responsible for 81–91%, with the remainder being the net CO2 flux from land-use change and land management (e.g., deforestation, degradation, regrowth after agricultural abandonment, and peat drainage). {5.2.1.2, 5.2.1.5; Table 5.1; Figures 5.5, 5.7, 5.12}

Over the past six decades, the average fraction of anthropogenic CO2 emissions that has accumulated in the atmosphere (referred to as the airborne fraction) has remained nearly constant at approximately 44%. The ocean and land sinks of CO2 have continued to grow over the past six decades in response to increasing anthropogenic CO2 emissions (high confidence). Interannual and decadal variability of the regional and global ocean and land sinks indicate that these sinks are sensitive to climate conditions and therefore to climate change (high confidence). {5.2.1.1, 5.2.1.2, 5.2.1.3, 5.2.1.4; Figures 5.7, 5.8, 5.10}

Recent observations show that ocean carbon processes are starting to change in response to the growing ocean sink, and these changes are expected to contribute significantly to future weakening of the ocean sink under medium- to high-emissions scenarios. However, the effects of these changes are not yet reflected in a weakening trend of the contemporary (1960–2019) ocean sink (high confidence). {5.1.2, 5.2.1.3, 5.3.2.1; Figures 5.8, 5.20; Cross-Chapter Box 5.3}

Atmospheric concentration of CH4 grew at an average rate of 7.6 ± 2.7 ppb yr–1 for the last decade (2010–2019), with a faster growth of 9.3 ± 2.4 ppb yr–1 over the last six years (2014–2019) (high confidence). The multi-decadal growth trend in atmospheric CH4 is dominated by anthropogenic activities (high confidence), and the growth since 2007 is largely driven by emissions from both fossil fuels and agriculture (dominated by livestock) (medium confidence). The interannual variability is dominated by El Niño–Southern Oscillation cycles, during which biomass burning and wetland emissions, as well as loss by reaction with tropospheric hydroxyl radical (OH) play an important role. {5.2.2; Figures 5.13, 5.14; Table 5.2; Cross-Chapter Box 5.2}

Atmospheric concentration of N2 O grew at an average rate of 0.85 ± 0.03 ppb yr–1 between 1995 and 2019, with a further increase to 0.95 ± 0.04 ppb yr–1 in the most recent decade (2010–2019). This increase is dominated by anthropogenic emissions, which have increased by 30% between the 1980s and the most recent observational decade (2007–2016) (high confidence). Increased use of nitrogen fertilizer and manure contributed to about two-thirds of the increase during the 1980–2016 period, with the fossil fuels/industry, biomass burning, and wastewater accounting for much of the rest (high confidence). {5.2.3; Figures 5.15, 5.16, 5.17}

Ocean Acidification and Ocean Deoxygenation

Ocean acidification is strengthening as a result of the ocean continuing to take up CO2 from human-caused emissions (very high confidence). This CO2 uptake is driving changes in seawater chemistry that result in the decrease of pH and associated reductions in the saturation state of calcium carbonate, which is a constituent of skeletons or shells of a variety of marine organisms. These trends of ocean acidification are becoming clearer globally, with avery likely rate of decrease in pH in the ocean surface layer of 0.016 to 0.020 per decade in the subtropics and 0.002 to 0.026 per decade in subpolar and polar zones since the 1980s. Ocean acidification has spread deeper in the ocean, surpassing 2000 m depth in the northern North Atlantic and in the Southern Ocean. The greater projected pH declines in Coupled Model Intercomparison Project Phase 6 (CMIP6) models are primarily a consequence of higher atmospheric CO2 concentrations in the Shared Socio-economic Pathways (SSPs) scenarios than their Coupled Model Intercomparison Project Phase 5 (CMIP5) Representative Concentration Pathway (RCP) analogues. {5.3.2.2, 5.3.3.1; 5.3.4.1; Figures 5.20, 5.21}

Ocean deoxygenation is projected to continue to increase with ocean warming (high confidence). Earth system models (ESMs) project a 32–71% greater subsurface (100–600 m) oxygen decline, depending on the scenario, than reported in the Special Report on the Ocean and Cryosphere (SROCC) for the period 2080–2099. This is attributed to the effect of larger surface warming in CMIP6 models, which increases ocean stratification and reduces ventilation (medium confidence). There is low confidence in the projected reduction of oceanic N2O emissions under high-emissions scenarios because of greater oxygen losses simulated in ESMs in CMIP6, uncertainties in the process of oceanic N2O emissions, and a limited number of modelling studies available. {5.3.3.2; 7.5}

Future Projections of Carbon Feedbacks on Climate Change

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). 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 uncertainties related to carbon–climate feedbacks (high confidence). {5.4.5; Figure 5.25, 5.26}

Increases in atmospheric CO2 lead to increases in land carbon storage through CO2 fertilization of photosynthesis and increased water use efficiency (high confidence). However, the overall change in land carbon also depends on land-use change and on the response of vegetation and soil to continued warming and changes in the water cycle, including increased droughts in some regions that will diminish the sink capacity. Climate change alone is expected to increase land carbon accumulation in the high latitudes (not including permafrost) and also to lead to a counteracting loss of land carbon in the tropics (medium confidence, Figure 5.25). More than half of the latest CMIP6 ESMs include nutrient limitations on the carbon cycle, but these models still project increasing tropical land carbon (medium confidence) and increasing global land carbon (high confidence) through the 21st century. {5.4.1, 5.4.3, 5.4.5; Figure 5.27; Cross-Chapter Box 5.1}

Future trajectories of the ocean CO2 sink are strongly emissions-scenario dependent (high confidence). Emissions scenarios SSP4-6.0 and SSP5-8.5 lead to warming of the surface ocean and large reductions of the buffering capacity, which will slow the growth of the ocean sink after 2050. Scenario SSP1-2.6 limits further reductions in buffering capacity and warming, and the ocean sink weakens in response to the declining rate of increasing atmospheric CO2. There is low confidence in how changes in the biological pump will influence the magnitude and direction of the ocean carbon feedback. {5.4.2, 5.4.4, Cross-Chapter Box 5.3}

Beyond 2100, land and ocean may transition from being a carbon sink to a source under either very high emissions or net negative emissions scenarios, but for different reasons. Under very high emissions scenarios such as SSP5-8.5, ecosystem carbon losses due to warming lead the land to transition from a carbon sink to a source (medium confidence), while the ocean is expected to remain a sink (high confidence). For scenarios in which CO2 concentration stabilizes, land and ocean carbon sinks gradually take up less carbon as the increase in atmospheric CO2 slows down. In scenarios with moderate net negative CO2 emissions, and CO2 concentrations declining during the 21st century (e.g., SSP1-2.6), the land sink transitions to a net source in decades to a few centuries after CO2 emissions become net negative, while the ocean remains a sink (low confidence). Under scenarios with large net negative CO2 emissions and rapidly declining CO2 concentrations (e.g., SSP5-3.4-OS (overshoot)), both land and ocean switch from a sink to a transient source during the overshoot period (medium confidence). {5.4.10, 5.6.2.1.2; Figures 5.30, 5.33}

Thawing terrestrial permafrost will lead to carbon release (high confidence), but there is low confidence in the timing, magnitude and the relative roles of CO2 versus CH4 as feedback processes. CO2 release from permafrost is projected to be 3–41 PgC per 1°C of global warming by 2100, based on an ensemble of models. However, the incomplete representation of important processes such as abrupt thaw, combined with weak observational constraints, only allowlow confidence in both the magnitude of these estimates and in how linearly proportional this feedback is to the amount of global warming. It is very unlikely that gas clathrates in terrestrial and subsea permafrost will lead to a detectable departure from the emissions trajectory during this century. {5.4.9; Box 5.1}

The net response of natural CH4 and N2 O sources to future warming will be increased emissions (medium confidence). Key processes include increased CH4 emissions from wetlands and permafrost thaw, as well as increased soil N2O emissions in a warmer climate, while ocean N2O emissions are projected to decline at centennial time scale. The magnitude of the responses of each individual process and how linearly proportional these feedbacks are to the amount of global warming is known with low confidence due to incomplete representation of important processes in models combined with weak observational constraints. Models project that, over the 21st century, the combined feedback of 0.02–0.09 W m–2°C–1is comparable to the effect of a CO2 release of 5–18 petagrams of carbon equivalent per °C (PgCeq °C–1) (low confidence). {5.4.7, 5.4.8; Figure 5.29}

The response of biogeochemical cycles to the anthropogenic perturbation can be abrupt at regional scales, and irreversible on decadal to century time scales (high confidence). The probability of crossing uncertain regional thresholds (e.g., high severity fires, forest dieback) increases with climate change (high confidence). Possible abrupt changes and tipping points in biogeochemical cycles lead to additional uncertainty in 21st century GHG concentrations, but these are very likely to be smaller than the uncertainty associated with future anthropogenic emissions (high confidence). {5.4.9}

Remaining Carbon Budgets to Climate Stabilization

There is a near-linear relationship between cumulative CO2 emissions and the increase in global mean surface air temperature (GSAT) caused by CO2 over the course of this century for global warming levels up to at least 2°C relative to pre-industrial (high confidence). Halting global warming would thus require global net anthropogenic CO2 emissions to become zero. The ratio between cumulative CO2 emissions and the consequent GSAT increase, which is called the transient climate response to cumulative emissions of CO2 (TCRE), likely falls in the 1.0°C–2.3°C per 1000 PgC range. The narrower range compared to the IPCC Fifth Assessment Report (AR5) is due to a better integration of evidence across the science in this assessment. Beyond this century, there is low confidence that the TCRE remains an accurate predictor of temperature changes in scenarios of very low or net negative CO2 emissions because of uncertain Earth system feedbacks that can result in further warming or a path-dependency of warming as a function of cumulative CO2 emissions. {5.4, 5.5.1}

Mitigation requirements over this century for limiting maximum warming to specific levels can be quantified using a carbon budget that relates cumulative CO2 emissions to global mean temperature increase (high confidence). For the period 1850–2019, a total of 655 ± 65 PgC (2390 ± 240 GtCO2, likely range) of anthropogenic CO2 has been emitted. Remaining carbon budgets (starting from 1 January 2020) for limiting warming to 1.5°C, 1.7°C, and 2.0°C are 140 PgC (500 GtCO2), 230 PgC (850 GtCO2) and 370 PgC (1350 GtCO2), respectively, based on the 50th percentile of TCRE. For the 67th percentile, the respective values are 110 PgC (400 GtCO2), 190 PgC (700 GtCO2) and 310 PgC (1150 GtCO2). These remaining carbon budgets may vary by an estimated ± 60 PgC (220 GtCO2) depending on how successfully future non-CO2 emissions can be reduced. Since AR5 and the Special Report on Global Warming of 1.5°C (SR1.5), estimates have undergone methodological improvements, resulting in larger, yet consistent estimates. {5.5.2, 5.6; Figure 5.31; Table 5.8}

Several factors affect the precise value of remaining carbon budgets, including estimates of historical warming, future emissions from thawing permafrost, and variations in projected non-CO2 warming. Remaining carbon budget estimates can increase or decrease by 150 PgC (likely range; 150 PgC equals 550 GtCO2) due to uncertainties in the level of historical warming, and by an additional ± 60 PgC (±220 GtCO, likely range) due to geophysical uncertainties surrounding the climate response to non-CO2 emissions such as CH4, N2O, and aerosols. Permafrost thaw is included in the estimates, together with other feedbacks that are often not captured by models. Despite the large uncertainties surrounding the quantification of the effects of additional Earth system feedback processes, such as emissions from wetlands and permafrost thaw, these feedbacks represent identified additional amplifying risk factors that scale with additional warming and mostly increase the challenge of limiting warming to specific temperature thresholds. These uncertainties do not change the basic conclusion that global CO2 emissions would need to decline to at least net zero to halt global warming. {5.4, 5.5.2}

Biogeochemical Implications of Carbon Dioxide Removal and Solar Radiation Modification

Land- and ocean-based carbon dioxide removal (CDR) methods have the potential to sequester CO2 from the atmosphere, but the benefits of this removal would be partially offset by CO2 release from land and ocean carbon stores (very high confidence). The fraction of CO2 removed that remains out of the atmosphere, a measure of CDR effectiveness, decreases slightly with increasing amount of removal (medium confidence) and decreases strongly if CDR is applied at lower CO2 concentrations (medium confidence). {5.6.2.1; Figures 5.32, 5.33, 5.34}

The century-scale climate–carbon cycle response to a CO2 removal from the atmosphere is not always equal and opposite to the response to a CO2 emission (medium confidence). For simultaneously cumulative CO2 emissions and removals of greater than or equal to 100 PgC, CO2 emissions are 4 ± 3% more effective at raising atmospheric CO2 than CO2 removals are at lowering atmospheric CO2. The asymmetry originates from state-dependencies and non-linearities in carbon cycle processes and implies that an extra amount of CDR is required to compensate for a positive emission of a given magnitude to attain the same change in atmospheric CO2. The net effect of this asymmetry on the global surface temperature is poorly constrained due to low agreement between models (low confidence). {5.6.2.1; Figure 5.35}

Wide-ranging side effects of CDR methods have been identified that can either weaken or strengthen the carbon sequestration and cooling potential of these methods and affect the achievement of sustainable development goals (high confidence). Biophysical and biogeochemical side effects of CDR methods are associated with changes in surface albedo, the water cycle, emissions of CH4 and N2O, ocean acidification and marine ecosystem productivity (high confidence). These side effects and associated Earth system feedbacks can decrease carbon uptake and/or change local and regional climate, and in turn limit the CO2 sequestration and cooling potential of specific CDR methods (medium confidence). Deployment of CDR, particularly on land, can also affect water quality and quantity, food production and biodiversity, with consequences for the achievement of related sustainable development goals (high confidence). These effects are often highly dependent on local context, management regime, prior land use, and scale of deployment (high confidence). A wide range of co-benefits are obtained with methods that seek to restore natural ecosystems or improve soil carbon (high confidence). The biogeochemical effects of terminating CDR are expected to be small for most CDR methods (medium confidence). {5.6.2.2; Figure 5.36; Cross-Chapter Box 5.1}

Solar radiation modification (SRM) would increase the global land and ocean CO2 sinks (medium confidence) but would not stop CO2 from increasing in the atmosphere, thus exacerbating ocean acidification under continued anthropogenic emissions (high confidence). SRM acts to cool the planet relative to unmitigated climate change, which would increase the land sink by reducing plant and soil respiration and slow the reduction of ocean carbon uptake due to warming (medium confidence). SRM would not counteract or stop ocean acidification (high confidence). The sudden and sustained termination of SRM would rapidly increase global warming, with the return of positive and negative effects on the carbon sinks (very high confidence). {4.6.3; 5.6.3}

5.1 Introduction

The physical and biogeochemical controls of greenhouse gases (GHGs) is a central motivation for this chapter, which identifies biogeochemical feedbacks that have led or could lead to a future acceleration, slowdown or abrupt transitions in the rate of GHG accumulation in the atmosphere, and therefore of climate change. A characterization of the trends and feedbacks lead to improved quantification for the remaining carbon budgets for climate stabilization, and the responses of the carbon cycle to atmospheric carbon dioxide removal (CDR), which is embedded in many of the mitigation scenarios, to achieve the goals of the Paris Agreement.

Changes in the abundance of well-mixed GHGs – carbon dioxide (CO2), methane (CH4) and nitrous oxide (N2O) – in the atmosphere play a large role in determining the Earth’s radiative properties and its climate in the past, the present and the future (Chapters 2, 4, 6 and 7). Since 1950, the increase in atmospheric GHGs has been the dominant cause of the human-induced climate change (Section 3.3). While the main driver of changes in atmospheric GHGs over the past 200 years relates to the direct emissions from human activities, the net accumulation of GHGs in the atmosphere is controlled by biogeochemical source-sink dynamics of carbon that exchange between multiple reservoirs on land, oceans and atmosphere. The combustion of fossil fuels and land-use change for the period 1750–2019 released an estimated 700 ± 75 PgC (1 PgC = 1015g of carbon) into the atmosphere, of which less than half remains in the atmosphere today (Sections 5.2.1.2; 5.2.1.5) (Friedlingstein et al., 2020). This emphasizes the central role of terrestrial and ocean CO2 sinks in regulating its atmospheric concentration (Ballantyne et al., 2012; W. Li et al., 2016; Le Quéré et al., 2018a; Ciais et al., 2019; Gruber et al., 2019b; Friedlingstein et al., 2020).

The chapter covers three dominant GHGs in the human perturbation of the Earth’s radiation budget for which high-quality records exist: CO2, CH4 and N2O (Figure 5.1).

Figure 5.1 | Visual guide to Chapter 5.

(Section 5.1 (this section) provides the time context on how unique current and future scenarios of GHGs atmospheric concentrations and growth rates are in the Earth’s history. It also introduces the main processes involved in carbon–climate feedbacks, followed by an assessment of what can be learned from the paleo record towards a better understanding of contemporary and future GHGs–climate dynamics and their response to different mitigation trajectories.

(Section 5.2 covers the state of the carbon cycle and other biogeochemical cycles, and global budgets of CO2, CH4 and N2O for the industrial era (since 1750). The section emphasizes the last 60-year period for which high-resolution observations are available and the most recent decade for comprehensive GHG budgets. Significant advances have taken place since the IPCC Fifth Assessment Report (AR5), particularly in constraining the annual-to-decadal variability of the ocean and land carbon sources and sinks, and in revealing the sensitivity of carbon pools to current and future climate changes. There has been an important increase in modelling capability of the three GHGs, for land and oceans, atmospheric and ocean observations, and remote sensing products that have enabled researchers to constrain the causes of the observed trends and variability.

(Section 5.3 builds on the Special Report on the Ocean and Cryosphere (SROCC) covering the change in ocean acidification due to oceanic CO2 uptake across the paleo, historical periods and future projections using Coupled Model Intercomparison Project Phase 6 (CMIP6), with consequences for marine life (assessed in the Sixth Assessment Report Working Group II, AR6 WGII) and biogeochemical cycles. The section also assesses changes in deoxygenation of the oceans due to warming, increased stratification of the surface ocean, and slowing of the meridional overturning circulation.

(Section 5.4 covers the future projections of biogeochemical cycles and their feedbacks to the climate system fully utilizing the database of the concentration-driven CMIP6. Since AR5, Earth system models (ESMs) have made progress towards including more complex carbon cycle and associated biogeochemical processes that enable exploring a range of possible future carbon–climate feedbacks and their influences on the climate system. The section addresses uncertainties and limits of our models to predict future dynamics for GHG emissions trajectories, as well as new understanding on processes involved in carbon–climate feedbacks and the possibility for rapid and abrupt changes brought by non-linear dynamics.

(Section 5.5 covers the development of the total and remaining carbon budgets to climate stabilization targets and the associated transient climate response to cumulative CO2 emissions. The section shows the progress made since AR5 (IPCC, 2013a) and the Special Report on Global Warming of 1.5°C (IPCC, 2018), particularly on key components required to estimate the remaining carbon budget, including the transient response to cumulative emissions of CO2, the zero emissions commitment, the projected non-CO2 warming, and the unrepresented Earth system feedbacks.

(Section 5.6 assesses the impacts of CDR and solar radiation modification for the purpose of climate change mitigation on the global carbon cycle, building from the assessment in the IPCC Special Report on Climate Change and Land (SRCCL). It includes an overview of the major CDR options and potential collateral biogeochemical effects beyond the intended climate change mitigation strategies. The potential capacity to deliver atmospheric reductions and the socio-economic feasibility of such options are assessed in detail in AR6 working group III (WGIII).

Finally, Section 5.7 highlights the knowledge gaps as limits to the assessment. The assessment would have been strengthened had those gaps not existed.

5.1.1 The Physical and Biogeochemical Processes in Carbon–Climate Feedbacks

The influence of anthropogenic CO2 emissions and emissions scenarios on the carbon–climate system is the primary driver of ocean and terrestrial sinks as the major negative feedbacks that determine the atmospheric CO2 levels, which then drive climate feedbacks through radiative forcing (Figure 5.2) (Friedlingstein et al., 2006; Jones et al., 2013; Jones and Friedlingstein, 2020). Biogeochemical feedbacks follow as an outcome of both carbon and climate forcing on the physics and the biogeochemical processes of the ocean and terrestrial carbon cycles (Figure 5.2) (Katavouta et al., 2018; Williams et al., 2019; Jones and Friedlingstein, 2020). Together, these carbon–climate feedbacks can amplify or suppress climate change by altering the rate at which CO2 builds up in the atmosphere through changes in the land and ocean sources and sinks (Figure 5.2; C.D. Jones et al., 2013; Raupach et al., 2014; Williams et al., 2019). These changes depend on the, often non-linear, interaction of the drivers (CO2 and climate) and processes in the ocean and land as well as the emissions scenarios (Figure 5.2; Sections 5.4 and 5.6) (Raupach et al., 2014; Schwinger et al., 2014; Williams et al., 2019). There is high confidence that carbon–climate feedbacks and their century scale evolution play a critical role in two linked climate metrics that have significant climate and policy implications: (i) the fraction of anthropogenic CO2 emissions that remains in the atmosphere, the so-called airborne fraction of CO2 (AF; Section 5.2.1.2, Figures 5.2 and 5.7, and FAQ 5.1); and (ii) the quasi-linear trend characteristic of the transient temperature response to cumulative CO2 emissions (TCRE; Section 5.5; MacDougall, 2016; Williams et al., 2016; Jones and Friedlingstein, 2020) and other GHGs (CH4 and N2O). This chapter assesses the implications of these issues from the perspective of carbon cycle processes (Figure 5.2) in Section 5.2 (historical and contemporary), Section 5.3 (changing carbonate chemistry), Section 5.4 (future projections), Section 5.5 (remaining carbon budget) and Section 5.6 (response to carbon dioxide removal and solar radiation modification).

Figure 5.2 | Key compartments, processes and pathways that govern historical and future CO2 concentrations and carbon–climate feedbacks through the coupled Earth system. The anthropogenic CO2 emissions, including land-use change, are partitioned via negative feedbacks (turquoise dotted arrows) between the ocean (23%), the land (31%) and the airborne fraction (46%) of anthropogenic CO2 that sets the changing CO2 concentration in the atmosphere (2010–2019; Table 5.1). This regulates most of the radiative forcing that creates the heat imbalance that drives the climate feedbacks to the ocean (blue) and land (green). Positive feedbacks (red arrows) result from processes in the ocean and on land (red text). Positive feedbacks are influenced by both carbon-concentration and carbon–climate feedbacks simultaneously. Additional biosphere processes have been included, but these have an as-yet-uncertain feedback impact (blue-dotted arrows). CO2 removal from the atmosphere into the ocean, land and geological reservoirs, necessary for negative emissions, has been included (grey arrows). Although this schematic is built around CO2 (the dominant greenhouse gas), some of the same processes also influence the fluxes of CH4 and N2O and the strength of the positive feedbacks from the terrestrial and ocean systems.

The airborne fraction is an important constraint for adjustments in carbon–climate feedbacks and reflects the partitioning of CO2 emissions between reservoirs by the negative feedbacks, which were 31% on land and 23% in the ocean for the decade 2010–2019 and also dominated the historical period (Figure 5.2; Table 5.1) (Friedlingstein et al., 2020). During the period 1959–2019, the airborne fraction has largely followed the growth in anthropogenic CO2 emissions with a mean of 44% and a large interannual variability (Ballantyne et al., 2012; Ciais et al., 2019; Friedlingstein et al., 2020, Section 5.2.1.2; Table 5.1). The negative feedback to CO2 concentrations is associated with its impact on the air–sea and air–land CO2 exchange through strengthening of partial pressure of CO2 (pCO2) gradients as well as the internal processes that enhance uptake. Two of these key processes are the buffering capacity of the ocean and the CO2 fertilization effect on gross primary production (Sections 5.4.1–5.4.4).

Positive and negative climate and carbon feedbacks involve: (i) fast processes on land and oceans at time scales from minutes to years, such as photosynthesis, soil respiration, net primary production, shallow ocean physics and air–sea fluxes; and (ii) slower processes taking from decades to millennia, such as changing ocean buffering capacity, ocean ventilation, vegetation dynamics, permafrost changes, peat formation and decomposition (Figure 5.2; Ciais et al., 2013; Forzieri et al., 2017; Williams et al., 2019). Depending on the particular combination of driver process and response dynamics, they behave as positive or negative feedbacks that amplify or dampen the magnitude and rates of climate change, respectively (Cox et al., 2000; Friedlingstein et al., 2003, 2006; Hauck and Völker, 2015; Williams et al., 2019); red and turquoise arrows in Figure 5.2 and Table 5.1).

Carbon cycle feedbacks co-exist with climate (heat and moisture) feedbacks (Cross-Chapter Boxes 5.1 and 5.3), which together drive contemporary (Section 5.2) and future (Section 5.4) carbon–climate feedbacks (Williams et al., 2019). The excess heat generated by radiative forcing from increasing concentration of atmospheric CO2 and other GHGs is mostly taken up by the ocean (>90%) and the residual balance partitioned between atmospheric, terrestrial and ice melting (Cross-Chapter Box 9.2; Frölicher et al., 2015). The combined effect of these two large-scale negative feedbacks of CO2 and heat are reflected in the TCRE (Section 5.5 and Cross-Chapter Box 5.3), which points to a quasi-linear and quasi-emission-path independent relationship between cumulative emissions of CO2 and global warming, which is used as the basis to estimate the remaining carbon budget (Section 5.5; MacDougall and Friedlingstein, 2015; MacDougall, 2017; Bronselaer and Zanna, 2020; Jones and Friedlingstein, 2020). There is still low confidence on the relative roles and importance of the ocean and terrestrial carbon processes on TCRE variability and uncertainty on centennial time scales (MacDougall, 2016; MacDougall et al., 2017; Williams et al., 2017; Katavouta et al., 2018, 2019; Jones and Friedlingstein, 2020) (Sections 5.5.1.1, 5.5.1.2).

The combined effects of climate and CO2 concentration feedbacks on the global carbon cycle are projected by ESMs to modify both the processes and natural reservoirs of carbon on a regional and global scale that may result in positive feedbacks (red arrows in Figure 5.2), which could weaken the major terrestrial and ocean sinks and disrupt the airborne fraction and TCRE under medium- to high-emissions scenarios (Section 5.4.5 and Figure 5.25).

5.1.2 Paleo Trends and Feedbacks

Paleoclimatic proxy records extend beyond the variability of recent decadal climate oscillations and thus provide an independent perspective on feedbacks between climate and carbon cycle dynamics. According to reconstructions, these past changes were slower than the current anthropogenic ones, so they cannot provide an unequivocal comparison. Nonetheless, they can help appraise sensitivities and point towards potentially dominant mechanisms of change (Tierney et al., 2020) on (sub)centennial to (multi)millennial time scales.

The AR5 (WGI, Chapter 5) concluded with medium confidence that atmospheric CO2 concentrations reached 350–450 ppm during the mid-Pliocene (3.3–3.0 million years ago (Ma)), and possibly 1000 ppm during the Early Eocene (52–48 Ma). The AR5 (WGI, Chapter 5) also concluded with very high confidence that the current rates of increases in CO2, CH4 and N2O atmospheric concentrations were unprecedented with respect to the ice core record covering the last deglacial transition (LDT, 18–11 ka) and with medium confidence that the rate of change of the reconstructed GHG rise was also unprecedented compared to the lower resolution of the records of the past 800 kyr.

5.1.2.1 Cenozoic Proxy CO2 Record

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).

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).

5.1.2.2 Glacial–Interglacial Greenhouse Gas Records

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).

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).

5.1.2.3 Holocene Changes

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.

5.2 Historical Trends, Variability and Budgets of CO2, CH4 and N2O

This section assesses the trends and variability in atmospheric accumulation of the three main greenhouse gases (GHGs) – CO2, CH4 and N2O – their ocean and terrestrial sources and sinks as well as their budgets during the Industrial Era (1750–2019). Emphasis is placed on the more recent contemporary period (1959–2019) where understanding is increasingly better constrained by atmospheric, ocean and land observations. The section also assesses our increased understanding of the anthropogenic forcing and processes driving the trends, as well as how variability at the seasonal to decadal scales provide insights on the mechanism governing long-term trends and emerging biogeochemical–climate feedbacks with their regional characteristics.

5.2.1 CO2: Trends, Variability and Budget

5.2.1.1 Anthropogenic CO2 emissions

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).

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).

5.2.1.2 Atmosphere

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).

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).

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.

5.2.1.3 Ocean Carbon Fluxes and Storage

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: Socean is the global mean ocean CO2 sink and Fnet denotes the net spatially varying CO2 fluxes (Hauck et al., 2020). Adjustment of the mean global Fnet for the pre-industrial sea-to-air CO2 flux associated with land-to-ocean carbon flux term makes Fnet comparable to Socean (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 (Socean) 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 (Socean) 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 (Socean 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 Socean, 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 CO2, 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 Socean (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 (Socean) 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 Socean 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 (Socean) 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).

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 Socean from net air sea fluxes (Fnet ) 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 Socean 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 Socean 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 (Socean) 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 airsea CO2 fluxes (Fnet ) 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).

5.2.1.3.2 Ocean carbon fluxes and storage: Regional and global variability

The intent of this assessment is to show how global variability can be regionally forced (Gregor et al., 2019; Landschützer et al., 2019; Hauck et al., 2020). Since AR5 and SROCC, advances in global ocean CO2 flux products, GOBMs and atmospheric inversion models have strengthened confidence in the assessment of how ocean regions influence mean global variability and trends of ocean CO2 air–sea fluxes (Fnet ; see Supplementary Materials Figure 5.SM.1; Ciais et al., 2013; Landschützer et al., 2014, 2015; Rödenbeck et al., 2014; McKinley et al., 2017; Bindoff et al., 2019; Gregor et al., 2019; Friedlingstein et al., 2020; Hauck et al., 2020). The coherence in the regional variability of the anomalies in Fnet from three independent lines of evidence support with high confidence that the non-steady state global interannual-decadal variability of Fnet has clear regional influences (Gregor et al., 2019; Landschützer et al., 2019). The tropical oceans contribute the most to the global mean interannual variability (Supplementary Materials Figure 5.SM.1d). The high latitude oceans, particularly the Southern Ocean, contribute the most to the global-scale decadal variability (Supplementary Materials Figure 5.SM5.1b,c; (Landschützer et al., 2016, 2019; Gregor et al., 2019; Gruber et al., 2019a; Hauck et al., 2020). The influence of the Southern Ocean on the global mean decadal variability and the 1990s hiatus is supported by the highest regional–global correlation coefficients (Supplementary Materials Figures 5.SM.1a,c). In contrast, the equatorial oceans’ influence on global mean Fnet has a low correlation because, notwithstanding the coherence in interannual variability, it does not show the same global mean trend of strengthening sink in response to growing global emissions (Supplementary Materials Figure 5.SM.1d; Gregor et al., 2019). All regions, except the equatorial ocean, contribute to varying extents to the multi-decadal trend of growth in the global ocean sink (Supplementary Materials Figure 5.SM.1). Data sparseness in the high latitudes and the relatively short length of the observational records leads to low confidence in the attribution of the processes that link regional–global variability to climate (Landschützer et al., 2019; Gloege et al., 2021).

Regional decadal-scale anomalies in the variability of ocean CO2 storage have also emerged, probably associated with changes in the MOC, which may influence the global variability in Fnet (Chapter 9; DeVries et al., 2017). In the interior of the Indian and Pacific sectors of the Southern Ocean, and the North Atlantic, the increase in the CO2 inventory from 1994 to 2007 was about 20% smaller than expected from the atmospheric CO2 increase during the same period and the anthropogenic CO2 inventory in 1994 (Sabine eta al., 2004; Gruber et al., 2019a). There is medium confidence that the ocean CO2 inventory strengthened again in the decade 2005–2015 (DeVries et al., 2017). In the North Atlantic, a low rate of anthropogenic CO2 storage at 1.9 ± 0.4 PgC per decade during the time period of 1989–2003 increased to 4.4 ± 0.9 PgC per decade during 2003–2014. This is associated with changing ventilation patterns driven by the North Atlantic Oscillation (Woosley et al., 2016). In the Pacific sector of the Southern Ocean, the rate of anthropogenic CO2 storage also increased from 8.8 ± 1.1 (1σ) PgC per decade during 1995–2005 to 11.7 ± 1.1 PgC per decade during 2005–2015 (Carter et al., 2019). However, in the Subantarctic Mode Water of the Atlantic sector of the Southern Ocean, the storage rate of the anthropogenic CO2 was rather lower after 2005 than before (Section 9.2.3.2; Tanhua et al., 2017; Bindoff et al., 2019). These changes have been predominantly ascribed to the impact of changes in the MOC on the transport of anthropogenic CO2 into the ocean interior due to regional climate variability, in addition to the increase in the atmospheric CO2 concentration (Section 9.2.3.1; Wanninkhof et al., 2010; Pérez et al., 2013; DeVries et al., 2017, 2019; Gruber et al., 2019b; McKinley et al., 2020). However,the low frequency of carbon observations in the interior of the vast ocean leads to medium confidence in the assessment of temporal variability in the rate of regional ocean CO2 storage and its controlling mechanisms.

In summary, multiple lines of observational and modelling evidence provide high confidence in the finding that the ocean sink for anthropogenic CO2 has increased quasi-linearly over the past 60 years in response to growing global emissions of anthropogenic CO2, with a mean fraction of 23% of total emissions. The high confidence assessment is moderated to medium confidence due to a number of ocean CO2 flux terms yet to be adequately constrained. Observed changes in the variability of oceanpCO2 and observed 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. 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 Socean.

5.2.1.4 Land CO2 Fluxes: Historical and Contemporary Variability and Trends

5.2.1.4.1 Trend in land–atmosphere CO2 exchange

The global net land CO2 sink is assessed to have grown over the past six decades (Sarmiento et al., 2010; Ballantyne et al., 2017; Le Quéré et al., 2018b; Ciais et al., 2019; Friedlingstein et al., 2020) (high confidence). Estimated as residual from the mass balance budget of fossil fuel CO2 emissions minus atmospheric CO2 growth and the ocean CO2 sink, the global net land CO2 sink (including both land CO2 sink and net land-use change emissions) increased from 0.3 ± 0.6 PgC yr–1 during the 1960s to 1.8 ± 0.8 PgC yr–1 during the 2010s (Friedlingstein et al., 2020). An increasing global net land CO2 sink since the 1980s (Figure 5.10) was consistently suggested both by atmospheric inversions (e.g., Peylin et al., 2013) and by DGVMs (e.g., Sitch et al., 2015; Friedlingstein et al., 2019). The Northern Hemisphere contributes more to the net increase in the land CO2 sink compared to the Southern Hemisphere (Ciais et al., 2019), and boreal and temperate forests probably contribute the most (Tagesson et al., 2020). Attributing an increased net land CO2 sink to finer regional scales remains challenging, but inversions of satellite-based column CO2 products that have emerged since AR5 are a promising tool to further constrain regional land-atmosphere CO2 exchange (Ciais et al., 2013; Houweling et al., 2015; Reuter et al., 2017; O’Dell et al., 2018; Palmer et al., 2019).

Figure 5.10 | Trends of the net land CO2 sink and related vegetation observations during 1980–2019. (a) Net land CO2 sink. The residual net land CO2 sink is estimated from the global CO2 mass balance (fossil fuel emissions minus atmospheric CO2 growth rate and ocean CO2 sink). Inversions indicate the net land CO2 sink estimated by an ensemble of four atmospheric inversions. Dynamic Global Vegetation Models (DGVMs) indicate the mean net land CO2 sink estimated by 17 dynamic global vegetation models driven by climate change, rising atmospheric CO2, land-use change and nitrogen deposition change (for carbon-nitrogen models). The positive values indicate net CO2 uptake from the atmosphere. (b) Normalized difference vegetation index (NDVI). The anomaly of global area-weighted NDVI observed by Advanced Very High Resolution Radiometer (AVHRR) and MODIS satellite sensors. AVHRR data are accessible during 1982–2016 and MODIS data are accessible during 2000–2018. (c) Near-infrared reflectance of vegetation (NIRv) and contiguous solar-induced chlorophyll fluorescence (CSIF). The standardized anomaly of area-weighted NIRv during 2001–2018 (Badgley et al., 2017) and CSIF during 2000–2018 (Zhang et al., 2018). (d) Gross primary production (GPP). The GPP from Cheng et al. (2017), DGVMs and MODIS GPP product (MOD17A3). GPP from Cheng et al. (2017) is based on an analytical model driven by climate change, rising atmospheric CO2, AVHRR leaf area index datasets and evapotranspiration datasets. GPP from DGVMs is the ensemble mean global GPP estimated by the same 17 DGVMs that provide the net land CO2 sink estimates. Shaded area indicates 1–σ inter-model spread except for atmospheric inversions, whose ranges were used due to limited number of models. Further details on data sources and processing are available in the chapter data table (Table 5.SM.6).

Carbon uptake by vegetation photosynthesis exerts a first-order control over the net land CO2 sink. Several lines of evidence show enhanced vegetation photosynthesis over the past decades (medium to high confidence) (Figure 5.10), including increasing satellite-derived vegetation greenness (e.g., see Chapter 2; Mao et al., 2016; Zhu et al., 2016; Jia et al., 2019) and satellite-derived photosynthesis indicators (e.g., Badgley et al., 2017; Zhang et al., 2018), change in atmospheric concentration of carbonyl sulphide (Campbell et al., 2017), enhanced seasonal CO2 amplitude (Graven et al., 2013; Forkel et al., 2016), observation-driven inference of increasing photosynthesis CO2 uptake based mostly on enhanced water use efficiency (Cheng et al., 2017), and DGVM simulated increase of photosynthesis CO2 uptake (Anav et al., 2015).

Substantial progress has been made since AR5 on attributing change of the global net land CO2 sink. Increasing global net land CO2 sink since the 1980s is mainly driven by the fertilization effect from rising atmospheric CO2 concentrations (Schimel et al., 2015; Sitch et al., 2015; Fernández-Martínez et al., 2019; O’Sullivan et al., 2019; Tagesson et al., 2020; Walker et al., 2021) (medium confidence). Increasing nitrogen deposition (de Vries et al., 2009; Devaraju et al., 2016; Huntzinger et al., 2017) or the synergy between increasing nitrogen deposition and atmospheric CO2 concentration (O’Sullivan et al., 2019) could have also contributed to the increasing global net land CO2 sink. The effects of climate change alone on the global net land CO2 sink is so divergent that even the signs (directions) of the effects are not the same across DGVMs (e.g., Huntzinger et al., 2017).

Lower fire emissions of CO2 and enhanced vegetation carbon uptake due to reduced global burned area have contributed to the increasing global net land CO2 sink in the recent decade (Arora and Melton, 2018; Yin et al., 2020) (low to medium confidence). Satellite observations reveal a declining trend in global burned area by about 20% over past two decades (Andela et al., 2017; Earl and Simmonds, 2018; Forkel et al., 2019), a trend most pronounced in regions like northern Africa (Forkel et al., 2019; Zubkova et al., 2019; Bowman et al., 2020) and Mediterranean Europe (Turco et al., 2016). However, burned area trends are highly heterogeneous regionally with increasing trends reported in regions like western United States (Holden et al., 2018; Abatzoglou et al., 2019). Some regions (e.g., Amazon basin and Australia) experienced record-breaking fire events in 2019 and 2020 (e.g., Boer et al., 2020), whose effects on burned area trends remain to be explored. The burned area trends were primarily attributed to both human-induced climate change and human activities (Jolly et al., 2015; Andela et al., 2017; Holden et al., 2018; Turco et al., 2018; Teckentrup et al., 2019; Bowman et al., 2020), as well as changing frequency of lightning in the boreal region (Veraverbeke et al., 2017). In addition to changes in the burned area, fire dynamics could affect the trend in land-atmosphere CO2 exchange indirectly through increasing concentration of air pollutants (see Section 6.3.4 for impacts of ozone and aerosol on the carbon cycle; Yue and Unger, 2018; Lasslop et al., 2019).

Significant uncertainties remain for the land CO2 sink partition of processes due to challenges in reconciling multiple-scale evidence from experiments to the globe (Fatichi et al., 2019; Walker et al., 2021), due to large spatial and inter-model differences in diagnosing dominant driving factors affecting the net land CO2 sink (Huntzinger et al., 2017; Fernández-Martínez et al., 2019), and due to model deficiency in process representations (He et al., 2016). Nitrogen dynamics, a major gap in DGVMs identified in AR5, have now been incorporated in about half of the DGVMs contributing to the carbon budget of the Global Carbon Project (GCP) (see Le Quéré et al. (2018a) for model characteristics) and a growing number of ESMs (Arora et al., 2020). However, as the representations of carbon–nitrogen interactions vary greatly among models, large uncertainties remain on how nitrogen cycling regulates the response of ecosystem carbon uptake to higher atmospheric CO2 (Walker et al., 2015; Wieder et al., 2019; Davies-Barnard et al., 2020; Meyerholt et al., 2020; see Section 5.4.1). Fire modules have been incorporated into 10 of 16 DGVMs contributing to the global carbon budget (Le Quéré et al., 2018a), and a growing number of models have representations of human ignitions and fire suppression processes (Rabin et al., 2017; Teckentrup et al., 2019). There are also growing DGVM developments to include management practices (Pongratz et al., 2018) and the effects of secondary forest regrowth (Pugh et al., 2019), though models still under-represent intensively managed ecosystems, such as croplands and managed forests (Guanter et al., 2014; Thurner et al., 2017). Processes that have not yet played a significant role in the land CO2 sink of the past decades but can grow in importance, include permafrost (Box 5.1) and peatlands dynamics (Dargie et al., 2017; Gibson et al., 2019), have also been incorporated in some DGVMs (Koven et al., 2015b; Burke et al., 2017a; Guimberteau et al., 2018). Growing numbers and varieties of Earth observations are being jointly used to drive and benchmark models, helping to further identify missing key processes or mechanisms that are poorly represented in the current generation of DGVMs (e.g., Collier et al., 2018).

5.2.1.4.2 Interannual variability in land–atmosphere CO2 exchange

The AR5 stated that the interannual variability of the atmospheric CO2 growth rate is dominated by tropical land ecosystems. A set of new satellite measurements applied to assess the variability of the tropical land carbon balance since AR5 (Ciais et al., 2013) confirm this statement, including satellite column CO2 measurements, estimating the recent anomalous land–atmosphere CO2 exchange induced by El Niño at continental scale (e.g., J. Liu et al., 2017; Palmer et al., 2019), and L-band vegetation optical depth, estimating tropical above-ground biomass carbon stock changes (Fan et al., 2019). In addition, based onmedium evidence and medium agreement between studies with DGVMs and atmospheric inversions, semi-arid ecosystems over the tropical zones have a larger contribution to interannual variability in global land–atmosphere CO2 exchange than moist tropical forest ecosystems (low to medium confidence) (Poulter et al., 2014; Ahlstrom et al., 2015; Piao et al., 2020).

Understanding the mechanisms driving interannual variability in the carbon cycle has the potential to provide insights into whether and to what extent the carbon cycle can affect the climate (carbon–climate feedback), with particular interests over the highly climate-sensitive tropical carbon cycle (e.g., Cox et al., 2013; X. Wang et al., 2014; Fang et al., 2017; Jung et al., 2017; Humphrey et al., 2018; Malhi et al., 2018; see Section 5.4). Consistent findings from studies with atmospheric inversions, satellite observations and DGVMs (e.g., Malhi et al., 2018; Rödenbeck et al., 2018) lead to high confidence that the tropical net land CO2 sink is reduced under warmer and drier conditions, particularly during El Niño events. Interannual variations in tropical land-atmosphere CO2 exchange are significantly correlated with anomalies of tropical temperature, water availability and terrestrial water storage (X. Wang et al., 2014; Jung et al., 2017; Humphrey et al., 2018; Piao et al., 2020), whose relative contribution are difficult to separate due to covariations between these climatic factors. At continental scale, the dominant climatic driver of interannual variations of tropical land-atmosphere CO2 exchange was temperature variations (Figure 5.11; Piao et al., 2020), which could partly result from the spatial compensation of the water availability effects on land-atmospheric CO2 exchange (Jung et al., 2017).

Figure 5.11 | Interannual variation in detrended anomalies of the net land CO2 sink and land surface air temperature during 1980–2019. Correlation coefficients between the net land CO2 sink anomalies and temperature anomalies are show on the right bar plots. The net land CO2 sink is estimated by four atmospheric inversions (blue) and 15 Dynamic Global Vegetation Models (DGVMs) (green), respectively (Friedlingstein et al., 2020). Solid blue and green lines show model mean detrended anomalies of the net land CO2 sink. The ensemble mean of DGVMs is bounded by the 1–σ inter-model spread in each large latitude band (North 30°N–90°N, Tropics 30°S–30°N, South 90°S–30°S) and the globe. The ensemble mean of atmospheric inversions is bounded by model spread. For each latitudinal band, the anomalies of the net land CO2 sink and temperature (orange) were obtained by removing the long-term trend and seasonal cycle. A 12-month running mean was taken to reduce high-frequency noise. The bars in the right panels show correlation coefficients between the net land CO2 sink anomalies and temperature anomalies for each region. ** indicates P<0.01; * indicates P<0.05. The grey shaded area shows the intensity of El Niño–Southern Oscillation (ENSO) as defined by the Niño 3.4 index. Two volcanic eruptions (El Chichón and Mount Pinatubo) are indicated with light blue dashed lines. Temperature data are from the Climatic Research Unit (CRU), University of East Anglia (Harris et al., 2014). Anomalies were calculated following Patra et al. (2005), but using a 12-month low-pass filter and detrended to obtain interannual variations. Further details on data sources and processing are available in the chapter data table (Table 5.SM.6).

Cross-Chapter Box 5.1 | Interactions Between the Carbon and Water Cycles, Particularly Under Drought Conditions

Contributors: Josep G. Canadell (Australia), Philippe Ciais (France), Hervé Douville (France), Sabine Fuss (Germany), Robert Jackson (United States of America), Annalea Lohila (Finland), Shilong Piao (China), Sonia I. Seneviratne (Switzerland), Sergio M. Vicente-Serrano (Spain), Sönke Zaehle (Germany)

This box presents an assessment of interactions between the carbon and water cycles that influence the dynamics of the biosphere and its interaction with the climate system. It also highlights carbon–water trade-offs arising from the use of land-based climate change mitigation options. Individual aspects of the interactions between the carbon and water cycles are addressed in separate chapters (Sections 5.2.1, 5.4.1, 8.2.3, 8.3.1, 8.4.1 and 11.6). The influence of wetlands and dams on methane emissions is assessed elsewhere (Sections 5.2.2, 5.4.7 and 8.3.1), as well as the consequences of permafrost thawing (Section 9.5.2 and Box 5.1) and/or increased flooding (Sections 8.4.1, 11.5 and 12.4) on wetland extent in the northern high latitudes and wet tropics.

Does elevated CO2 alleviate the impacts of drought?

Increasing atmospheric CO2 concentration enhances leaf photosynthesis and drives a partial closure of leaf stomata, leading to higher water-use efficiency (WUE) at the leaf canopy and ecosystem scales (Norby and Zak, 2011; De Kauwe et al., 2013; Fatichi et al., 2016; Knauer et al., 2017; Mastrotheodoros et al., 2017). Since AR5 (Box 6.3), a growing body of evidence from tree-ring and carbon isotopes further confirms an increase of plant water-use efficiency over decadal to centennial time scales, with some evidence for a stronger enhancement of photosynthesis compared to stomatal reductions (Frank et al., 2015; Guerrieri et al., 2019; Adams et al., 2020).

Multiple lines of evidence suggest that WUE has increased in near proportionality to atmospheric CO2 (high confidence) at a rate generally consistent with Earth system models (ESMs), despite variation in the WUE response to CO2 (De Kauwe et al., 2013; Frank et al., 2015; Keeling et al., 2017; Lavergne et al., 2019; Walker et al., 2021). Both field-scale CO2 enrichment experiments and process models show the effect of physiologically induced water savings, particularly under water-limiting conditions (De Kauwe et al., 2013; Farrior et al., 2015; Lu et al., 2016; Roy et al., 2016). Plants can also benefit from reduced drought stress due to enhanced CO2 without ecosystem-scale water savings (Jiang et al., 2021). To some extent, this increased WUE offsets the effects of enhanced vapour pressure deficit (VPD) on plant transpiration (Bobich et al., 2010; Creese et al., 2014; Jiao et al., 2019), but will have limited effect on ameliorating plant water stress during extreme drought events (Xu et al., 2016; Menezes-Silva et al., 2019; L. Liu et al., 2020), when leaf stomata are governed primarily by soil moisture (Roy et al., 2016).

Leaf stomata closure can have large effects on land freshwater availability because of reduced plant transpiration, leading in some regions to higher soil moisture and runoff (Roderick et al., 2015; Milly and Dunne, 2016; Y. Yang et al., 2019). However, increased water availability is often not realized because other CO2 physiological effects that enhance ecosystem evapotranspiration might offset the gains. These effects include plant growth and leaf area expansion (Ainsworth and Long, 2005; Ukkola et al., 2016; McDermid et al., 2021), lengthening of the vegetative growing season (Frank et al., 2015; Lian et al., 2021), and the effects of stomatal closure on near-surface atmosphere that leads to increased air temperature and VPDs (Berg et al., 2016; Vogel et al., 2018; Zhou et al., 2019; Grossiord et al., 2020).

ESMs show no consensus about the net hydrological response to physiological CO2 effects. Some studies show water savings as a consequence of the CO2 effects on leaf stomata closure (Swann et al., 2016; Lemordant et al., 2018), while other studies show that increased leaf area offsets the gains from increased WUE (Mankin et al., 2019). However, these projections are subject to ESM uncertainties to quantify transpiration (Lian et al., 2021), among them the correct representations of plant hydraulic architecture such as changes in xylem anatomical properties and deep rooting (Nie et al., 2013; L. Liu et al., 2020).

In conclusion, it is very likely that elevated CO2 leads to increased WUE at the leaf level, concurrent with enhanced photosynthesis. Increased CO2 concentrations alleviate the effects of water deficits on plant productivity (medium confidence) but there is low confidence for its role under extreme drought conditions. There is low confidence that increased WUE by vegetation will substantially reduce global plant transpiration and diminish the frequency and severity of soil moisture and streamflow deficits associated with the radiative effect of higher CO2 concentrations.

How does drought affect the terrestrial CO2 sink?

Water availability controls the spatial distribution of photosynthesis – gross primary productivity (GPP) – over a larger part of the globe (Beer et al., 2010) and, at local scale, drought decreases GPP more than respiration (Schwalm et al., 2012) over most ecosystem types. This makes water availability a major climatic driver of variability in net ecosystem exchange (Jung et al., 2017; Humphrey et al., 2018). In addition to suppressing photosynthesis, field evidence suggests that droughts reduce the land CO2 sink, also through increasing forest mortality and promoting wildfire (Allen et al., 2015; Brando et al., 2019; Abram et al., 2021).

At the global scale, interannual variability in the atmospheric CO2 growth rate and global-scale terrestrial water storage from satellite show that a lower global net land CO2 sink is associated with below-average terrestrial water storage (Humphrey et al., 2018). Atmospheric inversions based on surface and satellite column CO2 measurements show significant carbon release during drought events in pan-tropic areas (Phillips et al., 2009; Gatti et al., 2014; J. Liu et al., 2017; Palmer et al., 2019). Regional extreme droughts in the mid-latitudes also decrease GPP and land CO2 sink (Ciais et al., 2005; Wolf et al., 2016; W. Peters et al., 2020; Flach et al., 2021). Droughts are not compensated by equivalent wet anomalies because of the non-linear response of the terrestrial carbon uptake to soil moisture (Green et al., 2019).

Uncertainties remain on the magnitude of sensitivity of the land carbon fluxes to droughts. Global studies indicate stronger control of soil moisture to variations in satellite proxies of GPP than VPD (Stocker et al., 2019; L. Liu et al., 2020). However, given that VPD increases exponentially with atmospheric warming, some studies suggest that VPD in stomatal regulation will become increasingly more important under a warmer climate (Novick et al., 2016; Grossiord et al., 2020). It is difficult to isolate the relative contributions of warmer temperature, higher VPD and lower soil moisture. This is because land-atmosphere feedbacks cause a simultaneous increase of plant evaporative demand and of root zone water deficit impairing plant root uptake (Berg et al., 2016). These physiological responses can be further compounded by drought legacies (Anderegg et al., 2015), changes in structure and population dynamics due to forest mortality (McDowell et al., 2020), disturbances associated with drought (fire, insects damage; Anderegg et al., 2020) and possible trade-offs between resistance and resilience (X. Li et al., 2020). Nonetheless, ESMs suggest that increased drought effects under very high levels of global warming (about 4°C at the end of the 21st century) contribute to the reduced efficiency of the land sink (Green et al., 2019).

In conclusion, there is high confidence that the global net land CO2 sink is reduced on interannual scale when regional-scale reductions in water availability associated with droughts occur, particularly in tropical regions. There is also high confidence that the global land sink will become less efficient due to soil moisture limitations and associated drought conditions in some regions for high-emissions scenarios, specially under global warming above 4°C. However, there is low confidence on how these water cycle feedbacks will play out in lower emissions scenarios (at 2°C global warming or lower) due to uncertainties in regional rainfall changes and the balance between the CO2 fertilization effect, through WUE, and the radiative impacts of greenhouse gases.

What are the limits of carbon dioxide removal from a water cycle perspective?

Carbon dioxide removal (CDR) options based on terrestrial carbon sinks will require the appropriation of significant amounts of water at the landscape level. Most mitigation pathways that seek to limit global warming to 1.5°C or less than 2°C require the removal of about 30 to 300 GtC from the atmosphere by 2100 (Rogelj et al., 2018b). Bioenergy with carbon capture and storage (BECCS), and afforestation/reforestation are the dominant CDR options used in climate stabilization scenarios, implying large requirements for land and water (Section 5.6; Beringer et al., 2011; Boysen et al., 2017b; Fajardy and Mac Dowell, 2017; Jans et al., 2018; Séférian et al., 2018b; Yamagata et al., 2018; Stenzel et al., 2019). A review of freshwater requirements for irrigating biomass plantations shows a range between 15 and 1250 km3 per GtC of biomass harvest. This is equivalent to a water requirement of 99–8250 km3 for the median BECCS deployment of around 3.3 GtC yr−1 (Smith et al., 2016) in <2°C-scenarios (Stenzel et al., 2021), assuming that biomass is converted to electricity, which is substantially less efficient than converting biomass to heat. These large ranges are the result of different assumptions about the type of biomass and yield improvements, management, and land availability. The use of alternative feedstocks, such as wastes, residues and algae, would lead to smaller water requirements (Smith et al., 2019).

Most of the water consumed in BECCS is used to grow the feedstock, with carbon capture and storage constituting a smaller portion across all crops (Rosa et al., 2020), with an estimated evaporative loss of 260 km3yr−1 for 3.3 GtC yr−1 (Smith et al., 2016). The same authors also estimate water use for CDR through afforestation at 1040 km3yr−1 for 3.3 GtC yr−1, including interception and transpiration, adjusted for the original land cover’s water use.

The impacts of different CDR options on the water cycle depend crucially on regional climate, prior land cover, and scale of deployment (Trabucco et al., 2008). Extensive irrigation for afforestation in drier areas will have larger downstream impacts than in wetter regions, with the difference in water use between the afforested landscapes and its previous vegetation determining the level of potential impacts on evapotranspiration and runoff (Jackson et al., 2005; Teuling et al., 2017). Afforestation and reforestation sometimes enhances precipitation through atmospheric feedbacks such as increased convection, at least in the tropics (Ellison et al., 2017) and the increase in precipitation can, in some regions, even cancel out the increased evapotranspiration (Li et al., 2018).

In conclusion, extensive deployment of BECCS and afforestation/reforestation will require larger amounts of freshwater resources than used by the previous vegetation, altering the water cycle at regional scales (high confidence). Consequences of high water consumption on downstream uses, biodiversity, and regional climate depend on prior land cover, background climate conditions, and scale of deployment (high confidence). Therefore, a regional approach is required to determine the efficacy and sustainability of CDR projects.

5.2.1.5 CO2 Budget

The global CO2 budget (Figure 5.12) encompasses all natural and anthropogenic CO2 sources and sinks. Table 5.1 shows the perturbation of the global carbon mass balance between reservoirs since the beginning of the industrial era, circa 1750.

Table 5.1 | Global anthropogenic CO2 budget accumulated since the Industrial Revolution (onset in 1750) and averaged over the 1980s, 1990s, 2000s, and 2010s. By convention, a negative ocean or land to atmosphere CO2 flux is equivalent to a gain of carbon by these reservoirs. The table does not include natural exchanges (e.g., rivers, weathering) between reservoirs. Uncertainties represent the 68% confidence interval (Friedlingstein et al., 2020).

1750–2019

Cumulative

(PgC)

1850–2019

Cumulative

(PgC)

1980–1989

Mean Annual Growth Rate

(PgC yr–1)

1990–1999

Mean Annual Growth Rate

(PgC yr–1)

2000–2009

Mean Annual Growth Rate

(PgC yr–1)

2010–2019

Mean Annual Growth Rate

(PgC yr–1)

Emissions

Fossil fuel combustion and cement production

445 ± 20

445 ± 20

5.4 ± 0.3

6.3 ± 0.3

7.7 ± 0.4

9.4 ± 0.5

Net land-use change

240 ± 70

210 ± 60

1.3 ± 0.7

1.4 ± 0.7

1.4 ± 0.7

1.6 ± 0.7

Total emissions

685 ± 75

655 ± 65

6.7 ± 0.8

7.7 ± 0.8

9.1 ± 0.8

10.9 ± 0.9

Partition

Atmospheric increase

285 ± 5

265 ± 5

3.4 ± 0.02

3.2 ± 0.02

4.1 ± 0.02

5.1 ± 0.02

Ocean sink

170 ± 20

160 ± 20

1.7 ± 0.4

2.0 ± 0.5

2.1 ± 0.5

2.5 ± 0.6

Terrestrial sink

230 ± 60

210 ± 55

2.0 ± 0.7

2.6 ± 0.7

2.9 ± 0.8

3.4 ± 0.9

Budget imbalance

0

20

–0.4

–0.1

0

–0.1

Figure 5.12 | Global carbon (CO2 ) budget (2010–2019). Yellow arrows represent annual carbon fluxes (in PgC yr–1) associated with the natural carbon cycle, estimated for the time prior to the industrial era, around 1750. Pink arrows represent anthropogenic fluxes averaged over the period 2010–2019. The rate of carbon accumulation in the atmosphere is equal to net land-use change emissions, including land management (called LULUCF in the main text) plus fossil fuel emissions, minus land and ocean net sinks (plus a small budget imbalance, Table 5.1). Circles with yellow numbers represent pre-industrial carbon stocks in PgC. Circles with pink numbers represent anthropogenic changes to these stocks (cumulative anthropogenic fluxes) since 1750. Anthropogenic net fluxes are reproduced from Friedlingstein et al. (2020). The relative change of gross photosynthesis since pre-industrial times is based on 15 DGVMs used in Friedlingstein et al. (2020). The corresponding emissions by total respiration and fire are those required to match the net land flux, exclusive of net land-use change emissions which are accounted for separately. The cumulative change of anthropogenic carbon in the terrestrial reservoir is the sum of carbon cumulatively lost by net land-use change emissions, and net carbon accumulated since 1750 in response to environmental drivers (warming, rising CO2, nitrogen deposition). The adjusted gross natural ocean–atmosphere CO2 flux was derived by rescaling the value in Figure 1 of Sarmiento and Gruber (2002) of 70 PgC yr–1by the revised estimate of the bomb radiocarbon (14C) inventory in the ocean. The original bomb14C inventory yielded an average global gas transfer velocity of 22 cm hr–1; the revised estimate is 17cm hr–1leading to 17/22*70=54. Dissolved organic carbon reservoir and fluxes from Hansell et al. (2009). Dissolved inorganic carbon exchanges between surface and deep ocean, subduction and obduction from Levy et al. (2013). Export production and flux from (Boyd et al., 2019). Net primary production (NPP) and remineralization in surface layer of the ocean from Kwiatkowski et al. (2020); Séférian et al. (2020). Deep ocean reservoir from Keppler et al. (2020). Anthropogenic carbon reservoir in the ocean is from Gruber et al. (2019b) extrapolated to 2015. Fossil fuel reserves are from BGR (2020); fossil fuel resources are 11,490 PgC for coal, 6,780 PgC for oil and 365 PgC for natural gas. Permafrost region stores are from Hugelius et al. (2014); Strauss et al. (2017); Mishra et al. (2021) (see also Box 5.1) and soil carbon stocks outside of permafrost region from Batjes (2016); Jackson et al. (2017). Biomass stocks (range of seven estimates) are from Erb et al. (2018). Sources for the fluxes of the land–ocean continuum are provided in main text and adjusted within the ranges of the various assessment to balance the budget (Section 5.2.1.5).

Since AR5 (Ciais et al., 2013), a number of improvements have led to the more constrained carbon budget presented here. Some new additions include: (i) the use of independent estimates for the residual carbon sink on natural terrestrial ecosystems (Le Quéré et al., 2018a); (ii) improvements in the estimates of emissions from cement production (Andrew, 2019) and the sink associated with cement carbonation (Cao et al., 2020); (iii) improved and new emissions estimates from forestry and other land use (Hansis et al., 2015; Gasser et al., 2020); (iv) the use of ocean observation-based sink estimates and a revised river flux partition between hemispheres (Friedlingstein et al., 2020); and (v) the expansion of constraints from atmospheric inversions, based on surface networks and the use of satellite retrievals.

The budget, based on the annual assessment by the GCP (Friedlingstein et al., 2020), uses independent estimates of all major flux components: fossil fuel and carbonate emissions (EFOS), CO2 fluxes from land use, land-use change, and forestry (ELULUCF), the growth rate of CO2 in the atmosphere (Gatm), and the ocean (Socean) and natural land (Sland) CO2 sinks. An imbalance term (BImb) is required to ensure mass balance of the source and sinks that have been independently estimated: EFOS+ ELULUCF= Gatm+ Socean+ Sland. + BImb. All estimates are reported with 1 standard deviation (±1σ, 1 sigma) representing a likelihood of 68%.

Over the past decade (2010–2019), 10.9 ± 0.9 PgC yr–1 were emitted from human activities, which were distributed between three Earth system components: 46% accumulated in the atmosphere (5.1 ± 0.02 PgC yr–1), 23% was taken up by the ocean (2.5 ± 0.6 PgC yr–1) and 31% was stored by vegetation in terrestrial ecosystems (3.4 ± 0.9 PgC yr–1) (Table 5.1). There is a budget imbalance of 0.1 PgCyr–1 which is within the uncertainties of the other terms. Over the industrial era (1750–2019), the total cumulative CO2 fossil fuel and industry emissions were 445 ± 20 PgC, and the LULUCF flux (= net land-use change in Figure 5.12) was 240 ± 70 PgC (medium confidence). The equivalent total emissions (685 ± 75 PgC) was distributed between the atmosphere (285 ± 5 PgC), oceans (170 ± 20 PgC) and land (230 ± 60 PgC; Table 5.1), with a budget imbalance of 20 PgC. This budget (Table 5.1) does not explicitly account for source/sink dynamics due to carbon cycling in the land–ocean aquatic continuum comprising freshwaters, estuaries, and coastal areas. Natural and anthropogenic transfers of carbon from soils to freshwater systems are significant (2.4–5.1 PgC yr–1) (Regnier et al., 2013; Drake et al., 2018). Some of the carbon is buried in freshwater bodies (0.15 PgC) (Mendonça et al., 2017), and a significant proportion returns to the atmosphere via outgassing from lakes, rivers and estuaries (Raymond et al., 2013; Regnier et al., 2013; Lauerwald et al., 2015). The net export of carbon from the terrestrial domain to the open oceans is estimated to be 0.80 PgC yr–1 (medium confidence), based on the average of (Jacobson et al., 2007; Resplandy et al., 2018) and corrected to account for 0.2 PgC buried in ocean floor sediments. These terms are included in Figure 5.12. Inclusion of other smaller fluxes could further constrain the carbon budget (Ito, 2019; Friedlingstein et al., 2020).

5.2.2 Methane (CH4): Trends, Variability and Budget

Methane is a much more powerful greenhouse gas than CO2 (Chapter 7) and participates in tropospheric chemistry (Chapter 6). The CH4 variability in the atmosphere is mainly the result of the net balance between the sources and sinks on the Earth’s surface and chemical losses in the atmosphere. Atmospheric transport evens out the regional CH4 differences between different parts of the Earth’s atmosphere. The steady-state lifetime is estimated to be 9.1 ± 0.9 years (Section 6.3.1 and Table 6.2). About 90% of the loss of atmospheric CH4 occurs in the troposphere by reaction with hydroxyl radical (OH), 5% by bacterial soil oxidation, and the rest 5% by chemical reactions with OH, excited state oxygen (O1D), and atomic chlorine (Cl) in the stratosphere (Saunois et al., 2020). Methane has large emissions from natural and anthropogenic origins, but a clear demarcation of their nature is difficult because of the use and conversions of the natural ecosystem for human activities. The largest natural sources are from wetlands, freshwater and geological process, while the largest anthropogenic emissions are from enteric fermentation and manure treatment, landfills and waste treatment, rice cultivation and fossil fuel exploitation (Table 5.2). In the past two centuries, CH4 emissions have nearly doubled, predominantly human driven since 1900, and persistently exceeded the losses (virtually certain), thereby increasing the atmospheric abundance as evidenced from the ice core and firn air measurements (Ferretti et al., 2005; Ghosh et al., 2015).

Table 5.2 | Global CH4 budget. Sources and sinks of CH4 for the two most recent decades for wich data is available, from bottom-up and top-down estimations (in Tg CH4 yr–1). The data are updated from Saunois et al. (2020), for the bottom-up anthropogenic emissions (FAO, 2019; US EPA, 2019; Crippa et al., 2020; Höglund-Isaksson et al., 2020), top-down geological emissions (Schwietzke et al., 2016; Petrenko et al., 2017; Hmiel et al., 2020), and top-down sinks from seven selected inverse models. The means (min-max) with outliers removed from the range and the means are given. Outliers defined as >75th percentile + 3 × the interquartile range or <25th percentile – 3 × the interquartile range. The top-down budget imbalances are calculated for each model separately and averaged. Note also the round-off error for the sources and sinks, which sometimes leads to last digit mismatch in the sums. For detailed information on datasets, see further details on data table 5.SM.6.

This section discusses both bottom-up and top-down estimates of emissions and sinks. Bottom-up estimates are based on empirical upscaling of point measurements, emissions inventories and dynamical model simulations, while top-down estimates refer to those constrained by atmospheric measurements and chemistry-transport models in inversion systems. Since AR5, a larger suite of atmospheric inversions using both in situ and remote sensing measurement have led to better understanding of the regional CH4 sources (Cross-Chapter Box 5.2). New ice core measurements of14C-CH4 are used for estimating the geological sources of CH4 (Table 5.2). Compared to the SRCCL (IPCC, 2019a; Jia et al., 2019), we provide a whole atmospheric sources-sinks budget consisting of all emissions and losses.

5.2.2.1 Atmosphere

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).

5.2.2.2 Anthropogenic Methane (CH4) Emissions

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).

5.2.2.3 Land Biospheric Emissions and Sinks

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).

5.2.2.4 Ocean and Inland Water Emissions and Sinks

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).

5.2.2.5 Methane (CH4) Budget

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).

Cross-Chapter Box 5.2 | Drivers of Atmospheric Methane Changes During 1980–2019

Contributors: Prabir K. Patra (Japan/India), Josep G. Canadell (Australia), Frank J. Dentener (European Union, The Netherlands), Xin Lan (United States of America/China), Vaishali Naik (United States of America)

The atmospheric methane (CH4) growth rate has varied widely over the past three decades, and the causes have been extensively studied since AR5. The mean growth rate decreased from 15 ± 5 ppb yr–1 in the 1980s to 0.48 ± 3.2 ppb yr–1 during 2000–2006 (the so-called quasi-equilibrium phase) and returned to an average rate of 7.6 ± 2.7 ppb yr–1 in the past decade (2010–2019) (based on data in Figure 5.14). Atmospheric CH4 grew faster (9.3 ± 2.4 ppb yr–1) over the last six years (2014–2019) – a period with prolonged El Niño conditions, which contributed to high CH4 growth rates consistent with behaviour during previous El Niño events (Figure 5.14b). Because of large uncertainties in both the emissions and sinks of CH4, it has been challenging to quantify accurately the methane budget and ascribe reasons for the growth over 1980–2019. In the context of CH4 emissions mitigation, it is critical to understand if the changes in growth rates are caused by emissions from human activities or by natural processes responding to changing climate. If CH4 continues to grow at rates similar to those observed over the past decade, it will contribute to decadal scale climate change and hinder the achievement of the long-term temperature goals of the Paris Agreement (Section 7.3.2.2; Nisbet et al., 2019).

Cross-Chapter Box 5.2, Figure 1 shows the decadal CH4 budget derived from the Global Carbon Project (GCP)-CH4 synthesis for 1980s, 1990s and 2000s (Kirschke et al., 2013), and for 2010–2017 (Saunois et al., 2020). The imbalance of the sources and sinks estimated by atmospheric inversions (dark blue bars) can be used to explain the changes in CH4 concentration increase rates between the decades (Table 5.2).

Since AR5, many studies have discussed the role of different source categories in explaining the increase in CH4 growth rate since 2007 and a coincident decrease of d13C–CH4 and dD–CH4 isotopes (Figure 5.13; Rice et al., 2016). Both13C and D are enriched in mass-weighted average source signatures for CH4 emissions from thermogenic sources (e.g., coal mining, oil and gas industry) and pyrogenic (biomass burning) sources, and depleted in biogenic (e.g., wetlands, rice paddies, enteric fermentation, landfill and waste) sources. Proposed hypotheses for CH4 growth (2007–2017) are inconclusive and vary from a concurrent decrease in thermogenic and increase in wetland and other biogenic emissions (Nisbet et al., 2016; Schwietzke et al., 2016), an increase in emissions from agriculture in the tropics (Schaefer et al., 2016), a concurrent reduction in pyrogenic emissions and an increase in thermogenic emissions (Worden et al., 2017), or an emissions increase from biogenic sources and a slower increase in emissions from thermogenic sources compared to inventory emissions (Patra et al., 2016; Thompson et al., 2018; Jackson et al., 2020; Chandra et al., 2021).

Cross-Chapter Box 5.2, Figure 1 | Methane sources and sinks for four decades from atmospheric inversions with the budget imbalance (source–sink; dark blue bars) (plotted on the left y-axis). Top-down analysis from Kirschke et al. (2013); Saunois et al. (2020). The global CH4 concentration seen in the black line (plotted on the right y-axis), representing National Oceanic and Atmospheric Administration (NOAA) observed global monthly mean atmospheric CH4 in dry-air mole fractions for 1983–2019 (Chapter 2, Annex V). Natural sources include emissions from natural wetlands, lakes and rivers, geological sources, wild animals, termites, wildfires, permafrost soils, and oceans. Anthropogenic sources include emissions from enteric fermentation and manure, landfills, waste and wastewater, rice cultivation, coal mining, oil and gas industry, biomass and biofuel burning. The top-down total sink is determined from global mass balance that includes chemical losses due to reactions with hydroxyl (OH), atomic chlorine (Cl), and excited atomic oxygen (O1D), and oxidation by bacteria in aerobic soils (Table 5.2). Further details on data sources and processing are available in the chapter data table (Table 5.SM.6).

A few studies emphasize the role of chemical destruction by hydroxyl (OH; the primary sink of methane), in driving changes in the growth of atmospheric methane abundance, in particular after 2006 (Rigby et al., 2017; Turner et al., 2017). Studies applying three-dimensional atmospheric inversion (McNorton et al., 2018), simple multi-species inversion (Thompson et al., 2018), as well as empirical methods using a variety of observational constraints based on OH chemistry (Nicely et al., 2018; Patra et al., 2021), do not find trends in OH large enough to explain the methane changes post-2006. On the contrary, global chemistry–climate models based on fundamental principles of atmospheric chemistry and known emissions trends of anthropogenic non-methane short-lived climate forcers simulate an increase in OH over this period (Zhao et al., 2019; Stevenson et al., 2020; see Section 6.2.3). These contrasting lines of evidence suggest that OH changes may have had a small moderating influence on methane growth since 2007 (low confidence).

Cross-Chapter Box 5.2 Figure 2 shows that modelled wetland emissions anomalies for all regions did not exhibit statistically significant trends (high agreement between models, medium evidence). Thus, the inter-decadal difference of total CH4 emissions derived from inversion models and wetland emissions, arises mainly from anthropogenic activities. The time series of regional emissions suggest that progress towards atmospheric CH4 quasi-equilibrium was primarily driven by reductions in anthropogenic (fossil fuel exploitation) emissions in Europe, Russia and temperate North America over 1988–2000. In the global totals, emissions equalled loss in the early 2000s. The growth since 2007 is driven by increasing agricultural emissions from East Asia (1997–2017), West Asia (2005–2017), Brazil (1988–2017) and Northern Africa (2005–2017), and fossil fuel exploitations in temperate North America (2010–2017; Lan et al., 2019; Crippa et al., 2020; Höglund-Isaksson et al., 2020; Jackson et al., 2020; Chandra et al., 2021).

Cross-Chapter Box 5.2, Figure 2 | Anomalies in global and regional methane (CH4) emissions for 1988–2017. The map in the centre shows mean CH4 emissions for 2010–2016. Multi-model mean (line) and 1-s standard deviations (shaded) for 2000–2017 are shown for 9 surface CH4 and 10 satellite XCH4 inversions, and 22 wetland models or model variants that participated in GCP-CH4 budget assessment (Saunois et al., 2020). The results for the period before 2000 are available from two inversions, one using 19 sites (Chandra et al., 2021; also used for the 2010–2016 mean emissions map) and one for global totals (Bousquet et al., 2006). The long-term mean values for 2010–2016 (common for all GCP–CH4 inversions), as indicated within each panel separately, are subtracted from the annual-mean time series for the calculation of anomalies for each region. Further details on data sources and processing are available in the chapter data table (Table 5.SM.6).

There is evidence from emissions inventories at country level and regional scale inverse modelling that CH4 growth rate variability between 1988 and 2017 is closely linked to anthropogenic activities (medium agreement). Isotopic composition observations and inventory data suggest that concurrent emissions changes from both fossil fuels and agriculture are playing roles in the resumed CH4 growth since 2007 (high confidence). Shorter-term decadal variability is predominantly driven by the influence of El Niño–Southern Oscillation on emissions from wetlands and biomass burning (Cross-Chapter Box 5.2, Figure 2), and loss due to OH variations (medium confidence), but lacking quantitative contribution from each of the sectors. By synthesizing all available information regionally from a priori (bottom-up) emissions, satellite and surface observations, including isotopic information, and inverse modelling (top-down), the capacity to track and explain changes in, and drivers of, natural and anthropogenic CH4 regional and global emissions has improved since AR5, but fundamental uncertainties related to OH variations remain unchanged.

5.2.3 N2O: Trends, Variability and Budget

In natural ecosystems, nitrous oxide (N2O) is primarily produced as a by-product during the remineralization of organic matter via the primary processes of nitrification and denitrification (Butterbach-Bahl et al., 2013; Voss et al., 2013). The net N2O production is highly sensitive to local environmental conditions such as temperature, oxygen concentrations, pH and the concentrations of ammonium and nitrate, among others, causing strong variability of N2O emissions in time and space, even at small scales. Changes in the atmospheric abundance of N2O result largely from the balance of the net N2O sources on land and ocean, and the photochemical destruction of N2O in the stratosphere.

Since AR5 (WGI, Section 6.4.3), improved understanding of N2O sources allows for a more comprehensive assessement of the global N2O budget (Table 5.3). This progress is based on extended atmospheric observations (Francey et al., 2003; Elkins et al., 2018; Prinn et al., 2018), improved atmospheric N2O inversions (Saikawa et al., 2014; Thompson et al., 2019), updated and expanded inventories of N2O sources (Winiwarter et al., 2018; Janssens-Maenhout et al., 2019), as well as improved bottom-up estimate of freshwater, ocean and terrestrial sources (Martinez-Rey et al., 2015; Landolfi et al., 2017; Buitenhuis et al., 2018; Lauerwald et al., 2019; Maavara et al., 2019; Tian et al., 2019).

The human perturbation of the natural nitrogen cycle through the use of synthetic fertilizers and manure, as well as nitrogen deposition resulting from land-based agriculture and fossil fuel burning has been the largest driver of the increase in atmospheric N2O of 31.0 ± 0.5 ppb (10%) between 1980 and 2019 (robust evidence, high agreement) (Tian et al., 2020). The long atmospheric lifetime of N2O implies that it will take more than a century before atmospheric abundances stabilize after the stabilization of global emissions. The rise of atmospheric N2O is of concern, not only because of its contribution to the anthropogenic radiative forcing (Chapter 7) but also because of the importance of N2O in stratospheric ozone loss (Ravishankara et al., 2009; Fleming et al., 2011; W. Wang et al., 2014).

5.2.3.1 Atmosphere

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).

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).

5.2.3.2 Anthropogenic N2O Emissions

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).

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-Ramrez 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.

5.2.3.3 Emissions from Ocean, Inland Water Bodies and Estuaries

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).

5.2.3.4 Emissions and Sinks in Non-agricultural Land

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.

5.2.3.5 N2O Budget

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).

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.

Table 5.3 | Global N2 O budget (units TgN yr–1) averaged over the 1980s, 1990s, 2000s as well as the recent decade starting in 2007. Uncertainties represent the assessed range of source/sink estimates. All numbers are reproduced from Tian et al. (2020) based on a compilation of inventories, bottom-up models, as well as atmospheric inversions. For detailed information on datasets, see Data Table 5.SM.6.

AR6 1980–1989

(TgN yr–1)

AR6 1990–1999

(TgN yr–1)

AR6 2000–2009

(TgN yr–1)

AR6 (2007–2016)

(TgN yr–1)

AR5 (2006–2011)

(TgN yr–1)

Bottom-up Budget

Anthropogenic sources

Fossil fuel combustion and Industry

0.9 (0.8 to 1.1)

0.9 (0.9 to 1.0)

1.0 (0.8 to 1.0)

1.0 (0.8 to 1.1)

0.7 (0.2 to 1.8)

Agriculture (incl. aquaculture)

2.6 (1.8 to 4.1)

3.0 (2.1 to 4.8)

3.4 (2.3 to 5.2)

3.8 (2.5 to 5.8)

4.1 (1.7 to 4.8)

Biomass and biofuel burning

0.7 (0.7 to 0.7)

0.7 (0.6 to 0.8)

0.6 (0.6 to 0.6)

0.6 (0.5 to 0.8)

0.7 (0.2 to 1.0)

Wastewater

0.2 (0.1 to 0.3)

0.3 (0.2 to 0.4)

0.3 (0.2 to 0.4)

0.4 (0.2 to 0.5)

0.2 (0.1 to 0.3)

Inland water, estuaries, coastal zones

0.4 (0.2 to 0.5)

0.4 (0.2 to 0.5)

0.4 (0.2 to 0.6)

0.5 (0.2 to 0.7)

Atmospheric nitrogen deposition on ocean

0.1 (0.1 to 0.2)

0.1 (0.1 to 0.2)

0.1 (0.1 to 0.2)

0.1 (0.1 to 0.2)

0.2 (0.1 to 0.4)

Atmospheric nitrogen deposition on land

0.6 (0.3 to 1.2)

0.7 (0.4 to 1.4)

0.7 (0.4 to 1.3)

0.8 (0.4 to 1.4)

0.4 (0.3 to 0.9)

Other indirect effects from CO2, climate and land-use change

0.1 (–0.4 to 0.7)

0.1 (–0.5 to 0.7)

0.2 (–0.4 to 0.9)

0.2 (–0.6 to 1.1)

Total anthropogenic

5.6 (3.6 to 8.7)

6.2 (3.9 to 9.6)

6.7 (4.1 to 10.3)

7.3 (4.2 to 11.4)

6.3 (2.6 to 9.2)

Natural sources and sinks

Rivers, estuaries, and coastal zones

0.3 (0.3 to 0.4)

0.3 (0.3 to 0.4)

0.3 (0.3 to 0.4)

0.3 (0.3 to 0.4)

0.6 (0.1 to 2.9)

Open oceans

3.6 (3.0 to 4.4)

3.5 (2.8 to 4.4)

3.5 (2.7 to 4.3)

3.4 (2.5 to 4.3)

3.8 (1.8 to 9.4)

Soils under natural vegetation

5.6 (4.9 to 6.6)

5.6 (4.9 to 6.5)

5.6 (5.0 to 6.5)

5.6 (4.9 to 6.5)

6.6 (3.3 to 9.0)

Atmospheric chemistry

0.4 (0.2 to 1.2)

0.4 (0.2 to 1.2)

0.4 (0.2 to 1.2)

0.4 (0.2 to 1.2)

0.6 (0.3 to 1.2)

Surface sink

–0.01 (–0.3 to 0)

–0.01 (–0.3 to 0)

–0.01 (–0.3 to 0)

–0.01 (–0.3 to 0)

–0.01 (–1 to 0)

Total natural

9.9 (8.512.2)

9.8 (8.3–12.1)

9.8 (8.212.0)

9.7 (8.012.0)

11.6 (5.5–23.5)

Total bottom-up source

15.5 (12.1 to 20.9)

15.9 (12.2 to 21.7)

16.4 (12.3 to 22.4)

17.0 (12.2 to 23.5)

17.9 (8.1 to 30.7)

Observed growth rate

3.7 (3.7 to 3.7)

4.5 (4.3 to 4.6)

3.6 (3.5 to 3.8)

Inferred stratospheric sink

12.9 (12.2-13.5)

13.1 (12.4–13.6)

14.3 (4.3 to 28.7)

Atmospheric inversion

Atmospheric loss

12.1 (11.4 to 13.3)

12.4 (11.7 to 13.3)

Total source

15.9 (15.1 to 16.9)

16.9 (15.9 to 17.7)

Imbalance

3.6 (2.2 to 5.7)

4.2 (2.4 to 6.4)

5.2.4 The Relative Importance of CO2, CH4, and N2O

The total influence of anthropogenic greenhouse gases (GHGs) on the Earth’s radiative balance is driven by the combined effect of those gases, and the three most important – carbon dioxide (CO2), methane (CH4), nitrous oxide (N2O) – were discussed in the previous sections. This section compares the balance of the sources and sinks of these three gases and their regional net flux contributions to the radiative forcing. CO2 has multiple residence times in the atmosphere – from one year to many thousands of years (Box 6.1 in Ciais et al., 2013) – and N2O has a mean lifetime of 116 years. They are both long-lived GHGs, while CH4 has a lifetime of 9.1 years and is considered a short-lived GHG (see Chapter 2 for lifetime of GHGs, Chapter 6 for CH4 chemical lifetime, and Chapter 7 for effective radiative forcing of all GHGs).

Figure 5.18 shows the contribution to radiative forcing of CO2, CH4, N2O, and the halogenated species since the 1900s to more recent decades. For the period 1960–2019, the relative contribution to the total effective radiative forcing (ERF) was 63% for CO2, 11% for CH4, 6% for N2O, and 17% for the halogenated species (Chapter 7; Figure 5.18). The systematic decline in the relative contribution to ERF for CH4 since 1850 is caused by a slower increase rate of CH4 in the recent decades, at 6, 10 and 5 ppb yr–1 during 1850–2019, 1960–2019 and 2000–2019, respectively, in comparison with the increasing rate of CO2 (at 0.7, 1.6 and 2.2 ppm yr–1, respectively) and N2O (at 0.4, 0.7 and 0.9 ppb yr–1, respectively; Figure 5.4). Owing to the shorter lifetime of CH4, the effect of a reduction in the emissions increase rate on the ERF increase is evident at inter-decadal time scales.

Figure 5.18 | Contributions of carbon dioxide (CO2 ), methane (CH4 ), nitrous oxide (N2 O) and halogenated species to the total effective radiative forcing (ERF) increases in 2019 since 1850, 1960 and 2000, respectively. ERF data are taken from Annex III (based on calculations from Chapter 7). Note that the sum of the ERFs exceeds 100% because there are negative ERFs due to aerosols and clouds. Further details on data sources and processing are available in the chapter data table (Table 5.SM.6).

Atmospheric abundance of GHGs is proportional to their emissions-loss budgets in the Earth’s environment. There are multiple metrics to evaluate the relative importance of different GHGs for the global atmospheric radiation budget and the socio-economic impacts (Section 7.6). Metrics for weighting emissions are further developed in AR6 WGIII. Figure 5.19 shows the regional emissions of the three main GHGs. For North Asia, Europe, Temperate North America and West Asia, the most dominant GHG source is CO2 (high confidence) (Figure 5.19) while, for East Asia, South Asia, South East Asia, Tropical South America, Temperate North America and Central Africa, the source is CH4 (Figure 5.19). The N2O emissions are dominant in regions with intense use of nitrogen fertilizers in agriculture. Only boreal North America showed net sinks of CO2, while close to flux neutrality is observed for North Asia, Southern Africa, and Australasia. Persistent emissions of CO2 are observed for Tropical and South America, northern Africa, and South East Asia (medium confidence). The medium confidence arises from large uncertainties in the estimated non-fossil fuel CO2 fluxes over these regions due to the lack of high-quality atmospheric measurements.

Figure 5.19 | Regional distributions of net fluxes of carbon dioxide (CO2), methane (CH4), nitrous oxide (N2O) on the Earth’s surface. The region divisions, shown as the shaded map, are made based on ecoclimatic characteristics of the land. The fluxes include those from anthropogenic activities and natural causes that result from responses to anthropogenic greenhouse gases and climate change (feedbacks) as in the three budgets shown in Sections 5.2.1.5, 5.2.2.5, and 5.2.3.5. The CH4and N2O emissions are weighted by arbitrary factors of 50 and 500, respectively, for depiction by common y-axes. Fluxes are shown as the mean of the inverse models as available from Thompsonet al. (2019); Friedlingsteinet al. (2020); Saunoiset al. (2020). Further details on data sources and processing are available in the chapter data Table (Table 5.SM.6).

5.3 Ocean Acidification and Deoxygenation

The surface ocean has absorbed a quarter of all anthropogenic CO2 emissions, mainly through physical–chemical processes (McKinley et al., 2016; Gruber et al., 2019b; Friedlingstein et al., 2020). Once dissolved in seawater, CO2 reacts with water and forms carbonic acid, which in turn dissociates, leading to a decrease in the concentration of carbonate (CO3–2) ions, and increasing both bicarbonate (HCO3) and hydrogen (H+) ion concentration. This process has caused a shift in the carbonate chemistry towards a less basic state, commonly referred to as ‘ocean acidification’ (Caldeira and Wickett, 2003; Orr et al., 2005; Doney et al., 2009). Although the societal concern regarding ocean acidification is relatively recent (about the last 20 years), the physical–chemical basis for the ocean absorption (sink) of atmospheric CO2 has been discussed much earlier by Revelle and Suess (1957). The AR5 and SROCC assessments were of robust evidence that the H+ion concentration is increasing in the surface ocean, thereby reducing seawater pH (= -log [H+]) (Section 2.3.4.1; Orr et al., 2005; Feely et al., 2009; Ciais et al., 2013; Bindoff et al., 2019), and there is high confidence that ocean acidification is impacting marine organisms (Bindoff et al., 2019).

Ocean oxygen decline, or deoxygenation, is driven by changes in ocean ventilation and solubility (Bindoff et al., 2019). It is virtually certain that anthropogenic forcing has made a substantial contribution to the ocean heat content increase over the historical period (Bindoff et al., 2019; IPCC, 2019c; Chapter 9, Section 2.3.3.1), strengthening upper water column stratification. Ocean warming decreases the solubility of dissolved oxygen in seawater, and it contributes to about 15% of the dissolved oxygen decrease in the oceans according to estimates based on solubility and the recent SROCC assessment (medium confidence), especially in sub-surface waters, between 100–600 m depth (Helm et al., 2011; Schmidtko et al., 2017; Breitburg et al., 2018; Oschlies et al., 2018; SROCC, Section 5.3.1). Stratification reduces the ventilation flux into the ocean interior, contributing to most of the remaining ocean deoxygenation (Schmidtko et al., 2017; Breitburg et al., 2018; Section 3.6.2). Deoxygenation may enhance emissions of nitrous oxide, especially from oxygen minimum zones (OMZs) or hypoxic coastal areas (Breitburg et al., 2018; Oschlies et al., 2018). Since SROCC (Bindoff et al., 2019), CMIP6 model simulation results agree with the reported 2% loss (4.8 ± 2.1 Pmoles O2) in total dissolved oxygen in the upper ocean layer (100–600 m) for the 1970–2010 period (Helm et al., 2011; Ito et al., 2017; Schmidtko et al., 2017; Kwiatkowski et al., 2020; Section 2.3.4.2). The response of marine organisms to the coupled effects of ocean warming, acidification and deoxygenation occur at different metabolic levels on different groups, and include respiratory stress and reduction of thermal tolerance (Gruber, 2011; Bindoff et al., 2019; IPCC, 2019c; Kawahata et al., 2019). An assessment of these effects on marine biota is found in WGII AR6 Chapter 2.

This section assesses past events of ocean acidification and deoxygenation (Section 5.3.1), the historical trends and spatial variability for the upper ocean (Section 5.3.2) and the ocean interior (Section 5.3.3). Future projections for ocean acidification and the drivers in the coastal ocean are assessed in Sections 5.3.4 and 5.3.5, respectively.

5.3.1 Paleoclimate Context

5.3.1.1 Paleocene–Eocene Thermal Maximum

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).

5.3.1.2 Last Deglacial Transition

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).

5.3.2 Historical Trends and Spatial Characteristics in the Upper Ocean

5.3.2.1 Reconstructed Centennial Ocean Acidification Trends

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.

5.3.2.2 Observations of Ocean Acidification over Recent Decades

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).

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+][CO32-]/Ksp) 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.

5.3.3 Ocean Interior Change

5.3.3.1 Ocean Memory: Acidification in the Ocean Interior

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 (Ωarag) 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).

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).

5.3.3.2 Ocean Deoxygenation and its Implications for Greenhouse Gases

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).

5.3.4 Future Projections for Ocean Acidification

5.3.4.1 Future Projections with Earth System Models (ESMs)

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.

5.3.4.2 Reversal of Ocean Acidification by Carbon Dioxide Removal

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.

5.3.5 Coastal Ocean Acidification and Deoxygenation

The coastal ocean, from the shoreline to the isobath of 200 m, is highly heterogeneous due to the complex interplay between physical, biogeochemical and anthropogenic factors (Gattuso et al., 1998; Chen and Borges, 2009; Dürr et al., 2011; Laruelle et al., 2014; McCormack et al., 2016). These areas, according to SROCC (Bindoff et al., 2019) are, with high confidence, already affected by ocean acidification and deoxygenation. This section assesses the drivers and spatial variability of acidification and deoxygenation based on new observations and data products.

5.3.5.1 Drivers

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).

5.3.5.2 Spatial Characteristics

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 pCO2sea 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).

5.4 Biogeochemical Feedbacks on Climate Change

This section covers biogeochemical feedbacks on climate change, which represent one of the largest sources of uncertainty in climate change projections. The relevant processes are discussed (Sections 5.4.1 to 5.4.4), prior to discussing the simulation and projection of the carbon cycle in Earth system models (Section 5.4.5), emergent constraints on future projections (Section 5.4.6), non-CO2 feedbacks (Section 5.4.7), combined feedback assessment (Section 5.4.8), possible biogeochemical abrupt changes (Section 5.4.9), long-term carbon cycle projections (Section 5.4.10), and near-term prediction of ocean and land carbon sinks (5.4.11).

5.4.1 Direct CO2 Effect on Land Carbon Uptake

The AR5 (WGI, Box 6.3) and SRCCL (IPCC, 2019a) concluded with high confidence that rising atmospheric CO2 increases leaf-level photosynthesis. This effect is represented in all ESMs. New studies since AR5 add evidence that the leaf-level CO2 fertilization is modulated by acclimation of photosynthesis to long-term CO2 exposure, growth temperature, seasonal drought, and nutrient availability, but these effects are not yet routinely represented in ESMs (Smith and Dukes, 2013; Baig et al., 2015; Kelly et al., 2016; Drake et al., 2017; Jiang et al., 2020a). Cross-Chapter Box 5.1 assesses multiple lines of evidence, which suggest that the ratio of plant CO2 uptake to water loss – plant water-use efficiency (WUE) – increases in near proportionality to atmospheric CO2. Despite advances in the regional coverage of field experiments, observations of the consequences of CO2 fertilization at ecosystem level are still scarce, in particular from outside the temperate zone (Song et al., 2019). New syntheses since AR5 corroborate that the effect of elevated CO2 on plant growth and ecosystem carbon storage is generally positive (high confidence), but is modulated by temperature, water and nutrient availability (Reich et al., 2014; Obermeier et al., 2017; Peñuelas et al., 2017; Hovenden et al., 2019; Song et al., 2019). Plant carbon allocation, changes in plant community composition, disturbance, and natural plant mortality are important processes affecting the magnitude of the response, but are currently poorly represented in models (De Kauwe et al., 2014; Friend et al., 2014; Reich et al., 2018; A.P. Walker et al., 2019; K. Yu et al., 2019), and thus contribute strongly to uncertainty in ESM projections (Arora et al., 2020).

Field studies with elevated CO2 have demonstrated that the initial stimulation of above-ground growth may decline if insufficient nutrients such as nitrogen or phosphorus are available (Finzi et al., 2007; Norby et al., 2010; Hungate et al., 2013; Reich and Hobbie, 2013; Talhelm et al., 2014; Terrer et al., 2018). Model-data syntheses have demonstrated that capturing the observed long-term effect of elevated CO2 depends on the ability of models to predict the effect of vegetation on soil biogeochemistry (Zaehle et al., 2014; Koven et al., 2015b; Medlyn et al., 2015; Walker et al., 2015). Meta-analyses of CO2 manipulation experiments point to increased soil microbial activity and accelerated turnover of soil organic matter (van Groenigen et al., 2017) as a result of increased below-ground carbon allocation by plants (Song et al., 2019), and increased root exudation or mycorrhizal activity due to enhanced plant nutrient requirements under elevated CO2 (Drake et al., 2011; Terrer et al., 2016; Meier et al., 2017). These effects are not considered in most ESMs. One global model that attempts to represent these processes suggests that elevated CO2 -related carbon accumulation is reduced in soils but increased in vegetation relative to more conventional models (Sulman et al., 2019).

Our understanding of the effects of phosphorus limitation is less developed than for nitrogen, but a growing body of literature suggests that it is just as important, particularly in regions with highly weathered soils (Wang et al., 2018; Terrer et al., 2019; Du et al., 2020). CO2 experiments collectively show that soil phosphorus is an important constraint on the CO2 fertilization effect on plant biomass (Terrer et al., 2019; Jiang et al., 2020a). For example, despite increases in photosynthesis after four years of CO2 exposure, a free-air CO2 enrichment experiment in a phosphorus-limited mature forest ecosystem did not find an increase in biomass production (Jiang et al., 2020b). The lack of free-air CO2 enrichment experiments in phosphorus-limited tropical forests limits our understanding of the role of phosphorus availability in constraining the CO2 fertilization effect globally (Norby et al., 2016; Fleischer et al., 2019). Models accounting for the effects of phosphorus availability, in addition to nitrogen, generally show an even stronger reduction of the response of ecosystem carbon storage to elevated CO2 (Goll et al., 2012; Zhang et al., 2014; X. Yang et al., 2019). Insufficient data and uncertainties in the process formulation cause large uncertainty in the magnitude of this effect (Medlyn et al., 2016; Fleischer et al., 2019).

Consistent with AR5 (WGI, Section 6.4.2), the CO2 fertilization effect is the dominant cause for the projected increase in land carbon uptake between 1860 and 2100 in ESMs (Figures 5.26 and 5.27, and Table 5.5; Arora et al., 2020). In the CMIP6 ensemble, the increase of land carbon storage due to CO2 fertilization is a global phenomenon but is strongest in the tropics (Figure 5.26). The resulting increase of productivity is a key driver of increases in vegetation and soil carbon storage. However, consistent with earlier findings (Todd-Brown et al., 2013; Friend et al., 2014; Hajima et al., 2014), processes affecting vegetation carbon-use efficiency and turnover, such as allocation changes, mortality, and vegetation structural changes, as well as the pre-industrial soil carbon turnover time, also play an important role (Arora et al., 2020).

As a major advance since AR5 (WGI, Section 6.4.2), six out of 11 models in the C4MIP-CMIP6 ensemble account for nitrogen cycle dynamics over land (Table 5.4). On average, these models exhibit a 25–30% lower CO2 fertilization effect on land carbon storage, compared to models that do not account for nitrogen cycle dynamics (Figure 5.29 and Table 5.5). The only model in the C4MIP-CMIP6 ensemble that explicitly represents the effect of P availability on plant growth suggests the lowest carbon storage response to increasing CO2 (Arora et al., 2020). The lower CO2 effect due to decreased nutrient availability is generally consistent with analyses of the implicit nutrient limitation in CMIP5 simulations (Wieder et al., 2015; Zaehle et al., 2015) and independent assessments by stand-alone land models (Zaehle et al., 2010; Wårlind et al., 2014; Zhang et al., 2014; Goll et al., 2017; Meyerholt et al., 2020). The simulated effects are generally consistent with expectations based on independent observations (Walker et al., 2021). However, the magnitude of nutrient feedbacks in these models is poorly constrained by observations, owing to the limited geographic distribution of available observations and the uncertain scaling of results obtained from manipulation experiments to transient system dynamics (Song et al., 2019; Wieder et al., 2019; Meyerholt et al., 2020).

Our understanding of the various biological processes that affect the strength of the CO2 fertilization effect on photosynthesis and its impact on carbon storage in vegetation and soils, (in particular regarding the limitations imposed by nitrogen and phosphorus availability), has developed since AR5 (WGI, Box 6.2). Based on consistent behaviour across all CMIP6 ESMs, there is high confidence that CO2 fertilization of photosynthesis acts as an important negative feedback on anthropogenic climate change, by reducing the rate at which CO2 accumulates in the atmosphere. Since AR5 (WGI, Box 6.2), an increasing number of CMIP6 ESMs account for nutrient cycles. The consistent results found in their model projections suggests with high confidence that limited nutrient availability will limit the CO2 fertilization effect (Arora et al., 2020). The magnitude of the direct CO2 effect on land carbon uptake, and its limitation by nutrients, remains uncertain.

5.4.2 Direct CO2 Effects on Projected Ocean Carbon Uptake

In AR5 (WGI, Section 6.4.2) there was high agreement that CMIP5 ESMs project continued ocean CO2 uptake through to 2100, with higher uptake corresponding to higher concentration or emissions pathways. There has been no significant change in the magnitude of the sensitivity of ocean carbon uptake to increasing atmospheric CO2, or in the inter-model spread, between the CMIP5 and CMIP6 era (Arora et al., 2020). The analysis from emissions and concentration-driven CMIP5 model projections show that the ocean sink stops growing beyond 2050 across all emissions scenarios (Section 5.4.5.3). CMIP6 models also show a similar time evolution of global ocean CO2 uptake to CMIP5 models over the 21st century (Figure 5.25) with decreasing net ocean CO2 uptake ratio to anthropogenic CO2 emissions under SSP5-8.5.

The projected weakening of ocean carbon uptake is driven by a combination of decreasing carbonate buffering capacity and warming, which are positive feedbacks under weak to no mitigation scenarios (SSP4 and 5). In high mitigation scenarios (SSP1-2.6), weakening ocean carbon uptake is driven by decreasing emissions (Cross-Chapter Box 5.3). The detailed understanding of carbonate chemistry in seawater that has accumulated over more than half a century (e.g., Revelle and Suess, 1957; Egleston et al., 2010), provides high confidence that the excess CO2 dissolved in seawater leads to a non-linear reduction of the CO2 buffering capacity, that is smaller dissolved inorganic carbon (DIC) increase with respect to pCO2 increase along with the increase in cumulative ocean CO2 uptake. Recent studies (Katavouta et al., 2018; Jiang et al., 2019; Arora et al., 2020; Rodgers et al., 2020) suggest with medium confidence that the decrease in the ocean CO2 uptake ratio to anthropogenic CO2 emissions, under low to no mitigation scenarios over the 21st century, is predominantly attributable to the ocean carbon-concentration feedback through the reduction of the seawater CO2 buffering capacity, but with contributions from physical drivers such as warming and wind stress (medium confidence) and biological drivers (low confidence) (Sections 5.2.1.3.3 and 5.4.4).

Projected increases in ocean DIC due to anthropogenic CO2 uptake amplify the sensitivity of carbonate system variables to perturbations of DIC in the surface ocean, for example via the amplitude of the seasonal cycle of pCO2, which impacts the mean annual air–sea fluxes (Hauck et al., 2015; Fassbender et al., 2018; Landschützer et al., 2018; SROCC, Section 5.2.2.3). A larger amplification of the surface oceanpCO2 seasonality occurs in the subtropics where pCO2 seasonality is dominated by temperature seasonality, with the summer increase in the difference inpCO2 between surface water and the overlying atmosphere reaching 3μatm per decade between 1990 and 2030 under RCP8.5 (Schlunegger et al., 2019; Rodgers et al., 2020). In contrast, the impact of biological production on the seasonal cycle of pCO2 in summer in the Southern Ocean strengthens the drawdown of CO2 (Hauck et al., 2015).

Overall, there is medium confidence on three outcomes in the ocean from projected CO2 uptake under medium to high CO2 concentration scenarios: (i) a weakening of the buffering capacity, which impacts the airborne fraction via the reduction of the ocean CO2 buffering capacity due to cumulative ocean CO2 uptake, which reduces the net ocean CO2 uptake ratio to anthropogenic CO2 emissions (Katavouta et al., 2018; Arora et al., 2020; Rodgers et al., 2020); (ii) an amplification of the seasonal cycle of CO2 variables, which impacts both the ocean sink and ocean acidification (Hauck et al., 2015); (iii) a decrease in the aragonite and calcite saturation levels in the ocean, which negatively impacts the calcification rates of marine organisms (high confidence) and forms a negative feedback on the uptake of CO2 (McNeil and Sasse, 2016) (Cross-Chapter Box 5.3).

5.4.3 Climate Effect on Land Carbon Uptake

The AR5 assessed with medium confidence that future climate change will decrease land carbon uptake relative to the case with constant climate, but with a poorly constrained magnitude (AR5 WGI, Chapter 6, Executive Summary). Ongoing uncertainty in the magnitude and geographic pattern of the feedbacks (Section 5.4.5), continues to support amedium confidence assessment that future climate change will decrease land carbon uptake relative to the case with constant climate.

5.4.3.1 Plant Physiology

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).

5.4.3.2 Fire and Other Disturbances

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.

5.4.3.3 Soil Carbon

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.

Box 5.1 | Permafrost Carbon and Feedbacks to Climate

What is permafrost carbon and why should we be concerned about it?

Soils in the Arctic and other cold regions contain perennially frozen layers, known as permafrost. Soils in the northern permafrost region store a large amount of organic carbon, estimated at 1460–1600 PgC across surface soils and deeper deposits (Hugelius et al., 2014; Strauss et al., 2017; Mishra et al., 2021). Of that carbon, permafrost soils and deposits store 1070–1360 PgC, of which 300–400 PgC are in the first metre, and the rest at depth. The remaining 280–340 PgC are in permafrost-free soils within the permafrost region. These carbon deposits have accumulated over thousands of years due to the slow rates of organic matter decomposition in frozen and/or waterlogged soil layers, but these frozen soils are highly decomposable upon thaw (Schädel et al., 2014).

Is permafrost carbon already thawing and emitting greenhouse gases?

The permafrost region was a historic carbon sink over centuries to millennia (high confidence) (Loisel et al., 2014; Lindgren et al., 2018). Currently though, thawing soils due to anthropogenic warming are losing carbon from the decomposition of old frozen organic matter, as found via carbon 14 (14C) signature of respiration at sites undergoing rapid permafrost thaw (Hicks Pries et al., 2013), of dissolved organic carbon in rivers draining watersheds with permafrost thaw (Vonk et al., 2015; Wild et al., 2019), and of methane (CH4) produced in thawing lakes (Walter Anthony et al., 2016).

Despite accumulating evidence of increased carbon losses, it is difficult to scale up site- and ecosystem-level measurements to assess the net carbon balance over the entire permafrost region, due to the high spatial heterogeneity, the strong seasonal cycles, and the difficulty in monitoring these regions consistently across the year. The Special Report on Ocean and the Cryosphere in a Changing Climte (SROCC) assessed with high confidence that ecosystems in the permafrost region act as carbon sinks during the summer growing season, and that wintertime carbon losses are significant, consistent with a multi-decadal small increase in CO2 emissions during early winter at Barrow, Alaska (Sweeney et al., 2016; Webb et al., 2016; Meredith et al., 2019). These findings have been further strengthened by recent comprehensive synthesis of in-situ wintertime flux observations that show large carbon losses during the non-growing season (Natali et al., 2019). Increased autumn and winter respiration are a key large-scale fingerprint of top-down permafrost thaw predicted by ecosystem models (Parazoo et al., 2018). However, the length of these wintertime observational records is too short to unequivocally determine whether winter carbon losses are higher now than they used to be. One study inferred a multi-year net CO2 source for the tundra in Alaska (Commane et al., 2017), which is equivalent to 0.3 PgC yr–1 when scaled up to the northern permafrost region (low confidence) (Meredith et al., 2019).

Since AR5, evidence of a more active carbon cycle in the northern high-latitude regions has also been observed through the increased amplitude of CO2 seasonal cycles. However, the relative roles of local sources versus influence from mid-latitudes makes it difficult to infer changes to Arctic ecosystems from these observations (Graven et al., 2013; Forkel et al., 2016; Takata et al., 2017; Bruhwiler et al., 2021). Estimates of CO2 fluxes with atmospheric inversion models showed an enhanced seasonal cycle amplitude but no significant trends in annual total fluxes, in agreement with flux tower measurements over one decade (2004–2013) (Welp et al., 2016; Takata et al., 2017).

In addition to CO2, CH4 emissions from the northern permafrost region contribute to the global methane budget, but evidence as to whether these emissions have increased from thawing permafrost is mixed. The SROCC assigned low confidence to the degree of recent additional CH4 emissions from diverse sources throughout the permafrost region. These include observed regional lake area change, which suggest a 1.6–5 Tg CH4yr–1 increase over the last 50 years (Walter Anthony et al., 2016), ice-capped geological sources (Walter Anthony et al., 2012; Kohnert et al., 2017), and shallow Arctic Ocean shelves. The shallow subsea emissions are particularly uncertain due to the wide range of estimates (3 Tg CH4yr–1 (Thornton et al., 2016b) to 17 Tg CH4yr–1 (Shakhova et al., 2014)), and the lack of a baseline with which to infer any changes; however, the upper half of this range in flux estimates is inconsistent with the atmospheric inversions constrained by the pan-Arctic CH4 concentration measurements (Berchet et al., 2016).

Atmospheric measurements and inversions performed at the global and regional scales do not show any detectable trends in annual mean CH4 emissions from the permafrost region over the past 30 years (Jackson et al., 2020; Saunois et al., 2020; Bruhwiler et al., 2021), consistent with atmospheric measurements in Alaska that showed no significant annual trends, despite significant increase in air temperature (Sweeney et al., 2016). Atmospheric inversions and biospheric models do not show any clear trends in CH4 emissions for wetland regions of the high latitudes during the period 2000–2016 (Patra et al., 2016; Poulter et al., 2017; Jackson et al., 2020; Saunois et al., 2020). Large uncertainties on wetland extent and limited data constraints place low confidence in these modelling approaches.

The SROCC also assessed with high confidence that CH4 fluxes have been under-observed due to their high variability at multiple scales in both space and time, and that there is a persistent mismatch between top-down and bottom-up methane budgets, with emissions calculated by upscaling ground observations typically higher than emissions inferred from large-scale atmospheric observations (Thornton et al., 2016a; Saunois et al., 2020).

In conclusion, there is high confidence that the permafrost region has acted as a historic carbon sink over centuries to millennia, and high confidence that some permafrost regions are currently net sources of CO2. There is robust evidence that some CH4 emissions sources for some regions have increased over the past decades (medium confidence). For the northern permafrost-wide region, no multi-decadal trend has been detected on CO2 and CH4 fluxes but, given the low resolution and sparse observations of current observations and modelling sytems, we place low confidence in this statement.

Since AR5, there have been new studies showing that permafrost thaw also leads to nitrous oxide (N2O) release from soil (Abbott and Jones, 2015; Karelin et al., 2017; Wilkerson et al., 2019), a previously unaccounted source. However, this release is unquantified at the pan-Arctic scale.

What does the paleorecord tell us?

Large areas of Alaska and Siberia are underlain by frozen, glacial-age, ice- and carbon-rich deposits, and many of these areas show evidence of thermokarst processes during Holocene warm periods. Rapid warming of high northern latitudes contributed to permafrost thaw, liberating labile organic carbon tothe atmosphere (Köhler et al., 2014; Crichton et al., 2016; Winterfeld et al., 2018; Meyer et al., 2019), supporting the vulnerability of these areas to further warming (Strauss et al., 2013, 2017).

Radiogenic and stable isotopic measurements on CH4 trapped in Antarctic ice support the view that CH4 emissions from fossil carbon reservoirs, including permafrost and methane hydrates, remained small in response to the deglacial warming. Mass-balance calculations reveal that geological CH4 emissions have not exceeded 19 Tg yr–1, highlighting that the deglacial increase in CH4 emissions was predominantly related to contemporary CH4 emissions from tropical wetlands and seasonally inundated floodplains (Bock et al., 2017; Petrenko et al., 2017; Dyonisius et al., 2020). Isotopic constraints on CO2 losses from permafrost with warming after the Last Glacial Maximum (LGM) are weaker than for CH4. While the biosphere as a whole held less carbon during the LGM than the pre-industrial, that change in stocks was smaller than the change in plant productivity, and so carbon losses at high latitudes may have been offset by increased tropical productivity in response to warming during the Last Deglacial Transition (LDT; Ciais et al., 2012). There is also paleoclimate evidence for processes that mitigate carbon losses with warming on longer time scales, such as longer-term carbon accumulation in lake deposits following thermokarst thaw (Walter Anthony et al., 2014), and long-term accumulation of carbon in permafrost soils following LDT carbon loss (Lindgren et al., 2018), particularly in peatlands which accumulated carbon at a slow but persistent rate in warm paleoclimates (Treat et al., 2019).

In conclusion, several independent lines of evidence indicate that permafrost thaw did not release vast quantities of fossil CH4 associated with the transient warming events of the LDT. This suggests that large emissions of CH4 from old carbon sources will not occur in response to future warming (medium confidence).

What level of emissions do we expect in the future?

Near-surface permafrost is projected to decrease significantly under future global warming scenarios (high confidence) (Section 9.5.2), thus creating the potential for releasing CO2 and CH4 to the atmosphere, and act as a positive carbon–climate feedback.

The processes that govern permafrost carbon loss are grouped into gradual and abrupt mechanisms. Gradual processes include the deepening of the seasonally thawed active layer into perennially frozen permafrost layers and lengthening of the thawed season within the active layer, which increases the amount of organic carbon that is thawed and the duration of thaw. Abrupt thaw processes include ice-wedge polygon degradation, hillslope collapse, thermokarst lake expansion and draining, all of which are processes largely occurring in regions with very high soil carbon content (Olefeldt et al., 2016a, b). Abrupt thaw processes can contribute up to half of the total net greenhouse gas release from permafrost loss, the rest attributed to gradual thaw (Schneider von Deimling et al., 2015; Turetsky et al., 2020). Increased fire frequency and severity (Hu et al., 2010) also contributes to abrupt emissions and the removal of the insulating cover which leads to an acceleration of permafrost thaw (Genet et al., 2013). Ecological feedbacks can both mitigate and amplify carbon losses: nutrient release from increased organic matter decomposition can drive vegetation growth that partially offsets soil carbon losses (Salmon et al., 2016), but also lead to biophysical feedbacks that further amplify warming (Myers-Smith et al., 2011).

Through the Coupled Model Intercomparison Project Phase 5 (CMIP5), Earth system models (ESMs) had not included permafrost carbon dynamics. This remains largely true in Coupled Model Intercomparison Project Phase 6 (CMIP6), with most models not representing permafrost carbon processes, a small number representing the active-layer thickening effect on decomposition (Table 5.4), and no ESMs representing thermokarst or fire-permafrost-carbon interactions. The CMIP6 ensemble mean predicts a negative carbon–climate feedback in the permafrost region. However, those that do include permafrost carbon show a positive carbon–climate feedback in the permafrost region (Figure 5.27). Given the current limited ESM capacity to assess permafrost feedbacks, estimates in this report are based on published permafrost-enabled land surface model results.

The SROCC assessed that warming under a high-emissions scenario (RCP8.5 or similar) would result in a loss of permafrost carbon by 2100 of 10s to 100s of PgC, with a maximum estimate of 240 PgC and a best estimate of 92 ± 17 PgC (Meredith et al., 2019; SROCC, Figure 3.11). Under lower emissions scenarios, Schneider von Deimling et al. (2015) estimated permafrost feedbacks of 20–58 PgC of CO2 by 2100 under an RCP2.6 scenario, and 28–92 PgC of CO2 under an RCP4.5 scenario.

This new assessment, based on studies included in or published since SROCC (Schaefer et al., 2014; Koven et al., 2015c; Schneider von Deimling et al., 2015; Schuur et al., 2015; MacDougall and Knutti, 2016a; Gasser et al., 2018; Yokohata et al., 2020), estimates that the permafrost CO2 feedback per degree of global warming (Figure 5.29) is 18 [3.1 to 41, 5–95% range] PgC °C–1. The assessment is based on a wide range of scenarios evaluated at 2100, and an assessed estimate of the permafrost CH4-climate feedback at 2.8 [0.7 to 7.3] PgCeq °C–1 (Figure 5.29). This feedback affects the remaining carbon budgets for climate stabilization and is included in their assessment (Section 5.5.2).

Beyond 2100, models suggest that the magnitude of the permafrost carbon feedback strengthens considerably over the period 2100–2300 under a high-emissions scenario (Schneider von Deimling et al., 2015; McGuire et al., 2018). Schneider von Deimling et al. (2015) estimated that thawing permafrost could release 20–40 PgC of CO2 in the period from 2100 to 2300 under an RCP2.6 scenario, and 115–172 PgC of CO2 under an RCP8.5 scenario. The multi-model ensemble (McGuire et al., 2018) projects a much wider range of permafrost soil carbon losses of 81–642 PgC (mean 314 PgC) for an RCP8.5 scenario from 2100 to 2300, and of a gain of 14 PgC to a loss of 54 PgC (mean loss of 17 PgC) for an RCP4.5 scenario over the same period.

Methane release from permafrost thaw (including abrupt thaw) under a high-warming RCP8.5 scenario has been estimated at 836–2614 Tg CH4 over the 21st century and 2800–7400 Tg CH4 from 2100–2300 (Schneider von Deimling et al., 2015), and as 5300 Tg CH4 over the 21st century and 16,000 Tg CH4 from 2100–2300 (Turetsky et al., 2020). For RCP4.5, these numbers are 538–2356 Tg CH4 until 2100 and 2000–6100 Tg CH4 from 2100–2300 (Schneider von Deimling et al., 2015), and 4100 Tg CH4 until 2100 and 10,000 Tg CH4 from 2100–2300 (Turetsky et al., 2020).

A key uncertainty is whether permafrost carbon feedbacks scale roughly linearly with warming (Koven et al., 2015c), or instead scale at a greater (MacDougall and Knutti, 2016b; McGuire et al., 2018) or smaller rate (e.g., CH4 emissions estimated by Turetsky et al., 2020). It alsoremains unclear whether the permafrost carbon pool represents a coherent global tipping element of the Earth system with a single abrupt threshold (Drijfhout et al., 2015) at a given level of global warming, or a local scale tipping point without abrupt thresholds when aggregated across the pan-Arctic region, as is suggested by recent model results (e.g., Koven et al., 2015a; McGuire et al., 2018).

In conclusion, thawing terrestrial permafrost will lead to carbon release under a warmer world (high confidence). However, there is low confidence on the timing, magnitude and linearity of the permafrost climate feedback owing to the wide range of published estimates and the incomplete knowledge and representation in models of drivers and relationships. It is projected that CO2 released from permafrost will be 18 (3.1–41) PgC °C–1by 2100, with the relative contribution of CO2 vs CH4 remaining poorly constrained. Permafrost carbon feedbacks are included among the under-represented feedbacks quantified in Figure 5.29.

5.4.4 Climate Effects on Future Ocean Carbon Uptake

5.4.4.1 Physical Drivers of Future Ocean Carbon Uptake and Storage

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.

5.4.4.2 Biological Drivers of Future Ocean Carbon 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).

5.4.5 Carbon Cycle Projections in Earth System Models

This section summarizes future projections of land and ocean carbon sinks from the latest ESMs. ESMs are the basis for century time-scale projections (Chapter 4), and for detection and attribution studies (Chapter 3). These models aim to simulate the evolution of the carbon sources and sinks on land and in the ocean, in addition to the physical components of the climate system. ESMs include interactions between many of the processes and feedbacks described in Sections 5.4.1 to 5.4.4.

ESMs are now integral to the Coupled Model Intercomparison Project. Model output data from CMIP5 was analysed in AR5, while data from CMIP6 forms the basis for the analysis presented in this subsection. The CMIP5 ESMs discussed in AR5 (WGI, Section 6.4.2) produced a wide range of projections of future CO2 (Friedlingstein et al., 2014b) primarily associated with different magnitudes of carbon–climate and carbon-concentration feedbacks (Arora et al., 2013), but also exacerbated by differences in the simulation of the net carbon release from land-use change (Brovkin et al., 2013). A key deficiency of almost all CMIP5 ESMs was the neglect of nutrient limitations on CO2-fertilization of land plant photosynthesis (Section 5.4.1; Zaehle et al., 2015).

Some CMIP6 models considered in this report now include nitrogen limitations on land vegetation growth, along with many other added processes compared to CMIP5. Table 5.4 summarizes characteristics of the land and ocean carbon cycle models used in CMIP6 ESMs (Arora et al., 2020). In CMIP6, most ocean carbon cycle models (8 of 11) track three or more limiting nutrients (most often nitrogen, phosphorus, silicon, iron), and include two or more phytoplankton types. More than half of the land carbon cycle models (6 of 11) now include an interactive nitrogen cycle, and almost half (5 of 11) represent forest fires. However, even for CMIP6, very few models explicitly represent vegetation dynamics (3 of 11) or permafrost carbon (2 of 11). Despite these remaining limitations, the carbon cycle components of CMIP6 represent an advance on those in CMIP5, as they represent additional important processes (e.g., nitrogen limitations on the land carbon sink, and iron limitations on ocean ecosystems).

ESMs can be driven by anthropogenic CO2 emissions (‘emissions-driven’ runs), in which case atmospheric CO2 concentration is a predicted variable; or by prescribed time-varying atmospheric concentrations (‘concentration-driven’ runs). In concentration-driven runs, simulated land and ocean carbon sinks respond to the prescribed atmospheric CO2 and resulting changes in climate, but do not feed back through changes in the atmospheric CO2 concentration. Concentration-driven runs are used to diagnose the carbon emissions consistent with the Shared Socio-economic Pathways (SSPs) and other prescribed concentration scenarios (Section 5.5). In this subsection we specifically analyse results from concentration-driven ESM projections.

5.4.5.1 Evaluation of the Contemporary Carbon Cycle in Concentration-driven Runs

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).
Table 5.4 | Properties of the CMIP6 Earth system models (ESMs), focusing on the land and ocean carbon cycle components of these models(Aroraet al., 2020). Characteristics listed under each ESM are: number of vegetation carbon pools (veg C pools); number of soil and litter carbon pools (dead C pools); number of Plant Functional Types (PFTs); whether wildfire is represented (fire); whether vegetation dynamics is represented (dynamic veg); whether permafrost carbon is represented (permafrost C); whether the nitrogen cycle is represented (nitrogen cycle); the number of phytoplankton types (phytoplankton); the number of zooplankton types (zooplankton); and the list of ocean nutrients represented (limiting nutrients).

Modelling Group

CSIRO

BCC

CCCma

CESM

CNRM

GFDL

IPSL

JAMSTEC

MPI

NorESM2-LM

UK

ESM

ACCESS-ESM1.5

BCC-CSM2-MR

CanESM5

CESM2

CNRM-ESM2-1

GFDL-ESM4

IPSL-CM6A-LR

MIROC-ES2L

MPI-ESM1.2-LR

NorESM2-LM

UKESM1-0-LL

Land carbon/biogeochemistry component

Model name

CABLE2.4

CASA-CNP

BCC-AVIM2

CLASS-CTEM

CLM5

ISBA-CTRIP

LM4p1

ORCHIDEE ( 2)

MATSIRO (phys)

VISIT-e (BGC)

JSBACH3.2

CLM5

JULES-ES-1.0

Veg C pools

3

3

3

22

6

6

8

3

3

3

3

Dead C pools

6

8

2

7

7

4

3

6

18

7

4

PFTS

13

16

9

22

16

6

15

13

12

21

13

Fire

No

No

No

Yes

Yes

Yes

No

No

Yes

Yes

No

Dynamic Veg

No

No

No

No

No

Yes

No

No

Yes

No

Yes

Permafrost C

No

No

No

Yes

No

No

No

No

No

Yes

No

Nitrogen cycle

Yes

No

No

Yes

No

No

No

Yes

Yes

Yes

Yes

Ocean carbon/biogeochemistry component

Model name

WOMBAT

MOM4_L40

CMOC (biol)

MARBL

PISCESv2-gas

COBALTv2

PISCES-v2

OECO2

HAMOCC6

HAMOCC5.1

MEDUSA-2.1

Phytoplankton

1

0

1

3

2

3

2

2

2

1

2

Zooplankton

1

0

1

1

2

3

2

1

1

1

2

Limiting nutrients

P, Fe

P

N

N, P, Si, Fe

N, P, Si, Fe

N, P, Si, Fe

N, P, Si, Fe

N, P, Fe

N, P, Si, FE

N, P, Si, Fe

N, Si, Fe

5.4.5.2 Evaluation of Historical Carbon Cycle Simulations in Concentration-driven Runs

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).

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.

5.4.5.3 Evaluation of Latitudinal Distribution of Simulated Carbon Sinks

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).

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).

5.4.5.4 Coupled Climate–Carbon Cycle Projections

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 21st century. By contrast, scenarios that assume CO2 stabilization in the 21st 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).

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 airsea 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).

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).

5.4.5.5 Linear Feedback Analysis

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 (ΔCL) and changes in ocean carbon storage (ΔCo) are decomposed into contributions arising from warming (ΔT) and increases in CO2 (ΔCO2):

whereβL (βo) 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 (βL) 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 meanβo (0.82 to 0.77 GtC ppm–1), βL (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.

Table 5.5 | Diagnosed global feedback parameters for CMIP6 ESMs based on 1% per year runs to 4×CO2 (Aroraet al., 2020). The last two rows show the mean and standard deviation across the CMIP6 and CMIP5 models, respectively.

Land Feedback Factors

Ocean Feedback Factors

Model Name

βL(PgC ppm–1)

γ L(PgC K–1)

βo(PgC p_uo c;hnjppm–1)

γ o(PgC K–1)

ACCESS-ESM1.5

0.37

–21.1

0.90

–23.8

CanESM5

1.28

16.0

0.77

–14.7

CESM2

0.90

–21.6

0.71

–10.9

CNRM-ESM2-1

1.36

–83.1

0.70

–9.4

IPSL-CM6A-LR

0.62

–8.7

0.76

–13.0

MIROC-ES2L

1.12

–69.6

0.73

–22.3

MPI-ESM1.2-LR

0.71

–5.2

0.77

–20.1

NOAA-GFDL-ESM4

0.93

–80.1

0.84

–21.7

NorESM2-LM

0.85

–21.0

0.78

–19.6

UKESM1-0-LL

0.75

–38.4

0.75

–14.1

CMIP6 Model Mean

0.89±0.30

–33.3±33.8

0.77±0.06

–16.9±5.1

CMIP5 Model Mean

0.93±0.49

–57.9±38.2

0.82±0.07

–17.3±3.8

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).

5.4.6 Emergent Constraints to Reduce Uncertainties in Projections

Emergent constraints are based on relationships between observable aspects of the current or past climate (such as trends or variability), and uncertain aspects of future climate change (such as the strength of particular feedbacks). These relationships are evident across an ensemble of models. When combined with an observational estimate of the trend or variability in the real climate, such emergent relationships can yield ‘emergent constraints’ on future climate change (Hall et al., 2019). At the time of AR5 (WGI, Section 9.8.3), there had been relatively few applications of the technique to constrain carbon cycle sensitivities, but there have been many studies published since (e.g., the summary in Cox, 2019). Figure 5.28 shows some key published emergent constraints on the carbon cycle in ESMs.

Figure 5.28 | Examples of emergent constraints on the carbon cycle in Earth system models (ESMs), reproduced from previously published studies: (a) projected global mean atmospheric carbon dioxide (CO2) concentration by 2060 under the RCP8. 5 emissions scenario against the simulated CO2 in 2010 (Friedlingstein et al., 2014b; Hoffman et al., 2014); (b) sensitivity of tropical land carbon to warming (γ LT) against the sensitivity of the atmospheric CO2 growth-rate to tropical temperature variability (Cox et al., 2013; Wenzel et al., 2014); (c) sensitivity of extratropical (30°N–90°N) gross primary production to a doubling of atmospheric CO2 against the sensitivity of the amplitude of the CO2 seasonal cycle at Kumkahi, Hawaii to global atmospheric CO2 concentration (Wenzel et al., 2016); (d) change in high-latitude (30°N–90°N) gross primary production versus trend in high-latitude leaf area index or ‘greenness’ (Winkler et al., 2019); (e) sensitivity of the primary production of the Tropical Ocean to climate change versus its sensitivity to El Niño–Southern Oscillation (ENSO)-driven temperature variability (Kwiatkowski et al., 2017); (f) global ocean carbon sink in the 2090s versus the current-day carbon sink in the Southern Ocean. In each case, a red dot represents a single ESM projection, the grey bar represents the emergent relationship between the y-variable and the x-variable, the blue bar represents the observational estimate of the x-axis variable, and the green bar represents the resulting emergent constraint on the y-axis variable. The thicknesses represent ± one standard error in each case. Figure after Cox (2019). Further details on data sources and processing are available in the chapter data table (Table 5.SM.6).

5.4.7 Climate Feedbacks from CH4 and N2O

Sources and sinks of CH4 and N2O respond both directly and indirectly to atmospheric CO2 concentration and climate change, and thereby give rise to additional biogeochemical feedbacks in the climate system, which may amplify or attenuate climate–carbon cycle feedbacks (Gasser et al., 2017; Lade et al., 2018; Denisov et al., 2019). Many of these of feedbacks are only partially understood, and thus were only partially addressed in AR5 (WGI, Sections 6.3.3, 6.3.4 and 6.4.7). Since AR5, a growing body of estimates from ESMs, as well as independent modelling and observation-based studies, enable improved estimates of the associated feedbacks.

The goal of this section is to assess the climate feedback parameters α , as it is defined in Section 7.4.1.1, for CH4 and N2O biogeochemical feedbacks. The strength of the feedbacks is estimated in a linear framework (Gregory et al., 2009), using the radiative forcing equations for CO2, CH4 and N2O (Etminan et al., 2016). In addition to estimates from ESMs, the feedback parameter α is estimated from independent estimates of surface emission climate sensitivities and atmospheric box models, following (Arneth et al., 2010; Thornhill et al., 2021). These assessed feedback parameters are used in Section 7.4.2.5.

The CH4 feedbacks may arise from changing wetland emissions (including rice farming) and from sources that are expected to grow under climate change (e.g., related to permafrost thaw, fires, and freshwater bodies). CH4 emissions from wetlands and landfills generally increase with warming due to enhanced decomposition with higher temperatures, thereby potentially providing a positive CH4 feedback on climate (Dean et al., 2018). The contribution of wetlands to interannual variability of atmospheric CH4 is shaped by the different impacts of temperature and precipitation anomalies on wetland emissions (e.g., during El Niño episodes) and therefore the relationship between climate anomalies and the wetland contribution to the CH4 growth rate is complex (Pison et al., 2013; Nisbet et al., 2016; X. Zhang et al., 2020). As assessed by SROCC (IPCC, 2019b), there is high agreement across model simulations that wetlands CH4 emissions will increase in the 21st century, butlow agreement in the magnitude of the change (Denisov et al., 2013; Shindell et al., 2013; B.D. Stocker et al., 2013; Zhang et al., 2017; Koffi et al., 2020). Climate change increases wetland emissions (Gedney et al., 2004, 2019; Volodin, 2008; Ringeval et al., 2011; Denisov et al., 2013; Shindell et al., 2013) and gives rise to an estimated wetland CH4–climate feedback of 0.03 ± 0.01 W m–2°C–1 (mean ± 1 standard deviation; limited evidence, high agreement) (Arneth et al., 2010; Shindell et al., 2013; B.D. Stocker et al., 2013; Zhang et al., 2017). The effect of rising CO2 on productivity, and therefore on the substrate for methanogenesis, can further increase the projected increase in wetland CH4 emissions (Ringeval et al., 2011; Melton et al., 2013). Model projections accounting for the combined effects of CO2 and climate change suggest a potentially larger climate feedback (0.01–0.16 W m–2°C–1) (limited evidence, low agreement) (Gedney et al., 2019; Thornhill et al., 2021). Methane release from wetlands depends on the nutrient availability for methanogenic and methanotrophic microorganisms that can further modify this feedback (Stepanenko et al., 2016; Donis et al., 2017; Beaulieu et al., 2019). Methane emissions from thermokarst ponds and wetlands resulting from permafrost thaw are estimated to contribute an additional CH4-climate feedback of 0.01 [0.003 to 0.04, 5–95% range] W m–2°C–1 (limited evidence, low agreement).

Methane release from wildfires may increase by up to a factor of 1.5 during the 21st century (Eliseev et al., 2014a, b; Kloster and Lasslop, 2017). However, given the contemporary estimate for CH4 from wildfires of no more than 16 TgCH4yr–1 (van der Werf et al., 2017; Saunois et al., 2020), this feedback is small, adding no more than 40 ppb to the atmospheric CH44 by the end of the 21st century (medium confidence). Methane emissions from pan-Arctic freshwater bodies is also estimated to increase by 16 TgCH4yr–1 in the 21st century (Tan and Zhuang, 2015). Emissions from subsea and permafrost methane hydrates are not expected to change substantially in the 21st century (Section 5.4.9.1.3).

Land biosphere models show high agreement that long-term warming will increase N2O release from terrestrial ecosystems (Xu-Ri et al., 2012; B.D. Stocker et al., 2013; Zaehle, 2013; Tian et al., 2019). A positive land N2O climate feedback is consistent with paleo-evidence based on reconstructed and modelled emissions during the last deglacial period (Schilt et al., 2014; H. Fischer et al., 2019; Joos et al., 2020). The response of terrestrial N2O emissions to atmospheric CO2 increase and associated warming is dependent on nitrogen availability (van Groenigen et al., 2011; Butterbach-Bahl et al., 2013; Tian et al., 2019). Model-based estimates do not account for the potentially strong emissions increases in boreal and arctic ecosystems associated with future warming and permafrost thaw (Elberling et al., 2010; Voigt et al., 2017). There is medium confidence that the land N2O climate feedback is positive, but low confidence in the magnitude (0.02 ± 0.01 W m–2°C–1).

Climate change will also affect N2O production in the ocean (Codispoti, 2010; Freing et al., 2012; Bopp et al., 2013; Rees et al., 2016; Breider et al., 2019). Model projections in the 21st century show a 4–12% decrease in ocean N2O emissions under RCP8.5 due to a combination of factors, including increased ocean stratification, decreased ocean productivity, and the impact of increasing atmospheric N2O abundance on the air–sea flux, corresponding to an ocean N2O climate feedback of –0.008 ± 0.002 W m–2°C–1 (limited evidence, high agreement) (Martinez-Rey et al., 2015; Landolfi et al., 2017; Battaglia and Joos, 2018b). On millennial time scales, the ocean N2O climate feedback may be positive, owing to ocean deoxygenation and long-term increases in remineralization (Battaglia and Joos, 2018b).

Based-on these studies, there is medium confidence that the combined climate feedback parameter for CH4 and N2O is positive, but there is low confidence in the magnitude of the estimate (0.05 [0.02 to 0.09] W m–2°C–1, 5–95% range).

5.4.8 Combined Biogeochemical Climate Feedback

This section assesses the magnitude of the combined biogeochemical feedback in the climate system (Figure 5.29) by integrating evidence from: carbon-cycle projections represented in Earth system models (Section 5.4.5.5), independent estimates of CO2 emissions due to permafrost thaw (Box 5.1) and fire (Section 5.4.3.2), natural CH4 and N2O emissions (Section 5.4.7), and aerosol and atmospheric chemistry (Section 6.3.6). We derive a physical climate feedback parameter α , as defined in Section 7.4.1.1, for CO2 -based feedbacks using the linear framework proposed by Gregory et al. (2009), using the radiative forcing equations for CO2 (Etminan et al., 2016).

Figure 5.29 | Estimates of the biogeochemical climate feedback parameter ( α ). The parameter α (W m−2°C−1) for a feedback variable x is defined as
αx =
δN dx / δx dT
where
δN / δx
is the change in top-of-atmosphere energy balance in response to a change inx induced by a change in surface temperature (T), as in Section 7.4.1.1. (a) Synthesis of biogeochemical feedbacks from panels (b) and (c). Orange (blue) bars correspond to positive (negative) feedbacks increasing (decreasing) radiative forcing at the top of the atmosphere. Bars denote the mean and the error bar represents the 5–95% range of the estimates; (b) carbon-cycle feedbacks as estimated by coupled carbon-cycle climate models in the CMIP5 (Arora et al., 2013) and CMIP6 (Arora et al., 2020) ensembles, where dots represent single model estimates, and filled (open) circles are those estimates which do (not) include the representation of a terrestrial nitrogen cycle; (c) Estimates of other biogeochemical feedback mechanisms based on various modelling studies. Dots represent single estimates, and coloured bars denote the mean of these estimates with no weighting being made regarding the likelihood of any single estimate, and error bars the 5–95% range derived from these estimates. Results in panel (c) have been compiled from (a)Section 5.4.3.2 (Eliseev et al., 2014a; Harrison et al., 2018); (b)Section 5.4.3.3 (Schneider von Deimling et al., 2012; Burke et al., 2013, 2017b; Koven et al., 2015a, c; MacDougall and Knutti, 2016b; Gasser et al., 2018; Kleinen and Brovkin, 2018), where the estimates from Burke et al., 2013 have been constrained as assessed in their study (c)Section 5.4.7 (Schneider von Deimling et al., 2012, 2015; Koven et al., 2015c; Turetsky et al., 2020); (d)Section 5.4.7 (Arneth et al., 2010; Denisov et al., 2013; Shindell et al., 2013; B.D. Stocker et al., 2013; Zhang et al., 2017); (f)Section 5.4.7 (Xu-Ri et al., 2012; B.D. Stocker et al., 2013; Zaehle, 2013; Tian et al., 2019); (g)Section 5.4.7 (Martinez-Rey et al., 2015; Landolfi et al., 2017; Battaglia and Joos, 2018b). (h) Section 6.3, Table 6.9 mean and the 5–95%range the assessed feedback parameter. Further details on data sources and processing are available in the chapter data table (Table 5.SM.6).

The climate feedback parameter for CO2 (–1.13 ± 0.45 W m–2°C–1, mean and 1 standard-deviation range) is dominated by the contribution of the CO2 -induced increase of ocean and land carbon storage (–1.46 ± 0.41 W m–2°C–1, corresponding to aβL+O of 1.66 ± 0.31 PgC ppm–1), with smaller contributions from the carbon cycle’s response to climate (0.24 ± 0.17 W m–2°C–1, corresponding toγ L+O of –50 ± 34 PgC °C–1), and emissions from permafrost thaw (0.09 [0.02 to 0.20] W m–2°C–1, corresponding toγof –18 [3 to 41] PgC °C–1, mean and 5–95% range) (Figure 5.29a). This estimate does not include an estimate of the fire-related CO2 feedback (range: 0.01–0.06 W m–2°C–1), as onlylimited evidence was available to inform its assessment. The sum (mean and 5–95th percentile range) of feedbacks from natural emissions of CH4 including permafrost thaw, and N2O (0.05 [0.02 to 0.09] W m–2°C–1), and feedbacks from aerosol and atmospheric chemistry (–0.20 [–0.41 to 0.01] W m–2°C–1) leads to an estimate of the non-CO2 biogeochemical feedback parameter of –0.15 [–0.36 to +0.06] W m–2°C–1. There is low confidence in the estimate of the non-CO2 biogeochemical feedbacks, due to the large range in the estimates of α for some individual feedbacks (Figure 5.29c), which can be attributed to the diversity in how models account for these feedbacks, limited process-level understanding, and the existence of known feedbacks where there is insufficient evidence to assess the feedback strength.

CO2 and non-CO2 biogeochemical feedbacks are an important component of the assessment of TCRE and the remaining carbon budget (Section 5.5). The feedbacks of the carbon cycle of CO2 and climate are implicitly taken account in the TCRE assessment, because they are represented in the various underlying lines of evidence. Other feedback contributions, such as the non-CO2 biogeochemical feedback, can be converted into a carbon-equivalent feedback term (γ; Section 5.4.5.5, 7.6) by reverse application of the linear feedback approximation (Gregory et al., 2009). The contributions of non-CO2 biogeochemical feedbacks combine to a linear feedback term of 30 ± 27 PgCeq °C–1 (1 standard deviation range, 111 ± 98 Gt CO2- eq °C–1), including a feedback term of –11 [–18 to –5] PgCeq °C–1 (5–95% range, –40 [–62 to –18] Gt CO2- eq °C–1) from natural CH4 and N2O sources. The biogeochemical feedback from permafrost thaw leads to a combined linear feedback term of –21 ± 12 PgCeq °C–1 (1 standard deviation range –77 ± 44 Gt CO2- eq °C–1). For the integration of these feedbacks in the assessment of the remaining carbon budget (Section 5.5.2), two individual non-CO2 feedbacks (tropospheric ozone, and methane lifetime) are captured in the AR6-calibrated emulators (Box 7.1). Excluding those two contributions, the resulting combined linear feedback term for application in Section 5.5.2 is assessed at a reduction of 7 ± 27 PgCeq °C–1 (1 standard deviation range, –26 ± 97 PgCeq °C–1). For the same reasons as for the feedback terms expressed in W m–2°C–1 (see above), there is overall low confidence in the magnitude of these feedbacks.

5.4.9 Abrupt Changes and Tipping Points

The applicability of the linear feedback framework (Section 5.4.5.5) suggests that large-scale biogeochemical feedbacks are approximately linear in the forcing from changes in CO2 and climate. Nevertheless, regionally the biosphere is known to be capable of producing abrupt changes or even ‘tipping points’ (Higgins and Scheiter, 2012; Lasslop et al., 2016). Abrupt change is defined as a change in the system that is substantially faster than the typical rate of the changes in its history (Section 1.4.5). A related matter is a tipping point: a critical threshold beyond which a system reorganizes, often abruptly and/or irreversibly. Possible abrupt changes in the Earth system include those related to ecosystems and biogeochemistry (Lenton et al., 2008; Steffen et al., 2018): tropical and boreal forest dieback; and release of greenhouse gases (GHGs) from permafrost and methane clathrates (Table 5.6). In this section we therefore focus on estimating upper limits on the possible impact of abrupt changes on the evolution of atmospheric GHGs out to 2100, for comparison to the impact of direct anthropogenic emissions.

Table 5.6 | Examples of possible biogeochemical abrupt changes and tipping points in the Earth system. The fourth and sixth columns provide upper estimates of the impact of each example on the evolution of atmospheric GHGs in the 21st century. These upper estimates are therefore very unlikely but provide a useful comparison to the impact of direct anthropogenic emissions (currently 2.5 ppm yr–1).

Abrupt Change/Tipping Point

Key Region(s)

Probability to Occur in the 21st Century

Maximum CO2or CH4Release in the 21st Century

Principal Development Time Scale

Maximum CO2or CH4Rate of Change Over the 21st Century

(Ir)reversibility

Tropical forests dieback (Section 5.4.9.1.1)

Amazon watershed

Low

<200 PgC as CO2 (Section 5.4.9.1.1; medium confidence)

Multi-decadal

CO2 : <0.5 ppm yr–1

Irreversible at multi-decadal scale (medium confidence)

Boreal forests dieback (Sections 5.4.9.1.1, 5.4.3.2)

Boreal Eurasia and North America

Low

<27 Pg (Section 5.4.9.1.2; medium confidence)

Multi-decadal

Small (low confidence)

Irreversible at multi-decadal scale (medium confidence)

Biogenic emissions from permafrost thaw (Section 5.4.9.1.2)

Pan-Arctic

High

up to 240 PgC of CO2 and up to 5300 Tg of CH4Section 5.4.8.1.2; low confidence)

Multi-decadal

CO2 : ≤1 ppm yr–1

CH4: ≤10 ppb yr–1

Irreversible at centennial time scales (high confidence)

Methane release from clathrates (Section 5.4.9.1.3)

Oceanic shelf

Very low

very likely small (Section 5.4.9.1.3)

Multi-millennium

CH4: ≤0.2 ppb yr–1

Irreversible at multi-millennium time scales (medium confidence)

5.4.9.1 Assessment of Biogeochemical Tipping Points

5.4.9.1.1 Forest dieback

Published examples of abrupt biogeochemical changes in models include tropical rain forest dieback (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 forest dieback (Joos et al., 2001; Lucht et al., 2006; Scheffer et al., 2012; Lasslop et al., 2016; Section 5.4.3). Such transitions may be related to: (i) large-scale changes in mean climate conditions crossing particular climate thresholds (Joos et al., 2001; Cox et al., 2004; Lucht et al., 2006; Hirota et al., 2011; Scheffer et al., 2012; Le Page et al., 2017; Zemp et al., 2017); (ii) temperature and precipitation extremes (Staver et al., 2011; Higgins and Scheiter, 2012; Scheffer et al., 2012; Pavlov, 2015; Zemp et al., 2017); or (iii) possible enhancement and intermittency in fire activity (Staver et al., 2011; Higgins and Scheiter, 2012; Lasslop et al., 2016; Brando et al., 2020). Simulated changes in forest cover are a combination of the effects of CO2 on photosynthesis and water-use efficiency (Section 5.4.1), and the effects of climate change on photosynthesis, respiration and disturbance (Section 5.4.3). In ESMs, direct CO2 effects tend to enhance forest growth, but the impacts of climate change vary between being predominantly negative in the tropics and predominantly positive in the boreal zone (Figure 5.27).

Most ESMs project continuing carbon accumulation in tropical forests as a result of direct CO2 effects overwhelming the negative effects of climate change (Huntingford et al., 2013; Drijfhout et al., 2015; Boulton et al., 2017). In the real world, forests may be less vulnerable to climate changes than those modelled in ESMs because of the greater plant trait diversity, which confers additional resilience (Reyer et al., 2015; Levine et al., 2016; Sakschewski et al., 2016), and because of possible acclimation of vegetation to warming (Good et al., 2011, 2013; Lloret et al., 2012; Mercado et al., 2018). On the contrary, forests may be more vulnerable in the real world due to indirect climate change effects such as insect outbreaks and diseases not considered here (Section 5.4.3.2) or model limitations in representing the effects disturbances such as wildfire and droughts. In general, forests are most vulnerable when climate change is combined with increased rates of direct deforestation (Nobre et al., 2016; Le Page et al., 2017).

To estimate an upper limit on the impact of Amazon forest dieback on atmospheric CO2