Editing IPCC:AR6/WGI/Chapter-5 (section)

== 5.4 Biogeochemical Feedbacks on Climate Change ==

<div id="h1-5-siblings" class="h1-siblings"></div>

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 ( [[#5.4.5|Section 5.4.5]] ), emergent constraints on future projections ( [[#5.4.6|Section 5.4.6]] ), non-CO <sub>2</sub> feedbacks ( [[#5.4.7|Section 5.4.7]] ), combined feedback assessment ( [[#5.4.8|Section 5.4.8]] ), possible biogeochemical abrupt changes ( [[#5.4.9|Section 5.4.9]] ), long-term carbon cycle projections ( [[#5.4.10|Section 5.4.10]] ), and near-term prediction of ocean and land carbon sinks (5.4.11).

<div id="5.4.1" class="h2-container"></div>

<span id="direct-co-2-effect-on-land-carbon-uptake"></span>
=== 5.4.1 Direct CO <sub>2</sub> Effect on Land Carbon Uptake ===

<div id="h2-20-siblings" class="h2-siblings"></div>

The AR5 (WGI, Box 6.3) and SRCCL ( [[#IPCC--2019a|IPCC, 2019a]] ) concluded with ''high confidence'' that rising atmospheric CO <sub>2</sub> increases leaf-level photosynthesis. This effect is represented in all ESMs. New studies since AR5 add evidence that the leaf-level CO <sub>2</sub> fertilization is modulated by acclimation of photosynthesis to long-term CO <sub>2</sub> exposure, growth temperature, seasonal drought, and nutrient availability, but these effects are not yet routinely represented in ESMs ( [[#Smith--2013|Smith and Dukes, 2013]] ; [[#Baig--2015|Baig et al., 2015]] ; [[#Kelly--2016|Kelly et al., 2016]] ; [[#Drake--2017|Drake et al., 2017]] ; [[#Jiang--2020a|Jiang et al., 2020a]] ). Cross-Chapter Box 5.1 assesses multiple lines of evidence, which suggest that the ratio of plant CO <sub>2</sub> uptake to water loss – plant water-use efficiency (WUE) – increases in near proportionality to atmospheric CO <sub>2</sub> . Despite advances in the regional coverage of field experiments, observations of the consequences of CO <sub>2</sub> fertilization at ecosystem level are still scarce, in particular from outside the temperate zone ( [[#Song--2019|Song et al., 2019]] ). New syntheses since AR5 corroborate that the effect of elevated CO <sub>2</sub> on plant growth and ecosystem carbon storage is generally positive ( ''high confidence'' ), but is modulated by temperature, water and nutrient availability ( [[#Reich--2014|Reich et al., 2014]] ; [[#Obermeier--2017|Obermeier et al., 2017]] ; [[#Peñuelas--2017|Peñuelas et al., 2017]] ; [[#Hovenden--2019|Hovenden et al., 2019]] ; [[#Song--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--2014|Friend et al., 2014]] ; [[#Reich--2018|Reich et al., 2018]] ; A.P. [[#Walker--2019|]] [[#Walker--2019|Walker et al., 2019]] ; K. [[#Yu--2019|]] [[#Yu--2019|Yu et al., 2019]] ), and thus contribute strongly to uncertainty in ESM projections ( [[#Arora--2020|Arora et al., 2020]] ).

Field studies with elevated CO <sub>2</sub> have demonstrated that the initial stimulation of above-ground growth may decline if insufficient nutrients such as nitrogen or phosphorus are available ( [[#Finzi--2007|Finzi et al., 2007]] ; [[#Norby--2010|Norby et al., 2010]] ; [[#Hungate--2013|Hungate et al., 2013]] ; [[#Reich--2013|Reich and Hobbie, 2013]] ; [[#Talhelm--2014|Talhelm et al., 2014]] ; [[#Terrer--2018|Terrer et al., 2018]] ). Model-data syntheses have demonstrated that capturing the observed long-term effect of elevated CO <sub>2</sub> depends on the ability of models to predict the effect of vegetation on soil biogeochemistry ( [[#Zaehle--2014|Zaehle et al., 2014]] ; [[#Koven--2015b|Koven et al., 2015b]] ; [[#Medlyn--2015|Medlyn et al., 2015]] ; [[#Walker--2015|Walker et al., 2015]] ). Meta-analyses of CO <sub>2</sub> manipulation experiments point to increased soil microbial activity and accelerated turnover of soil organic matter ( [[#van%20Groenigen--2017|van Groenigen et al., 2017]] ) as a result of increased below-ground carbon allocation by plants ( [[#Song--2019|Song et al., 2019]] ), and increased root exudation or mycorrhizal activity due to enhanced plant nutrient requirements under elevated CO <sub>2</sub> ( [[#Drake--2011|Drake et al., 2011]] ; [[#Terrer--2016|Terrer et al., 2016]] ; [[#Meier--2017|Meier et al., 2017]] ). These effects are not considered in most ESMs. One global model that attempts to represent these processes suggests that elevated CO <sub>2</sub> -related carbon accumulation is reduced in soils but increased in vegetation relative to more conventional models ( [[#Sulman--2019|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--2018|Wang et al., 2018]] ; [[#Terrer--2019|Terrer et al., 2019]] ; [[#Du--2020|Du et al., 2020]] ). CO <sub>2</sub> experiments collectively show that soil phosphorus is an important constraint on the CO <sub>2</sub> fertilization effect on plant biomass ( [[#Terrer--2019|Terrer et al., 2019]] ; [[#Jiang--2020a|Jiang et al., 2020a]] ). For example, despite increases in photosynthesis after four years of CO <sub>2</sub> exposure, a free-air CO <sub>2</sub> enrichment experiment in a phosphorus-limited mature forest ecosystem did not find an increase in biomass production ( [[#Jiang--2020b|Jiang et al., 2020b]] ). The lack of free-air CO <sub>2</sub> enrichment experiments in phosphorus-limited tropical forests limits our understanding of the role of phosphorus availability in constraining the CO <sub>2</sub> fertilization effect globally ( [[#Norby--2016|Norby et al., 2016]] ; [[#Fleischer--2019|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 CO <sub>2</sub> ( [[#Goll--2012|Goll et al., 2012]] ; [[#Zhang--2014|Zhang et al., 2014]] ; X. [[#Yang--2019|Yang et al., 2019]] ). Insufficient data and uncertainties in the process formulation cause large uncertainty in the magnitude of this effect ( [[#Medlyn--2016|Medlyn et al., 2016]] ; [[#Fleischer--2019|Fleischer et al., 2019]] ).

Consistent with AR5 (WGI, Section 6.4.2), the CO <sub>2</sub> fertilization effect <sub></sub> 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--2020|Arora et al., 2020]] ). In the CMIP6 ensemble, the increase of land carbon storage due to CO <sub>2</sub> 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--2013|Todd-Brown et al., 2013]] ; [[#Friend--2014|Friend et al., 2014]] ; [[#Hajima--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--2020|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 CO <sub>2</sub> 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 C <sup>4</sup> MIP-CMIP6 ensemble that explicitly represents the effect of P availability on plant growth suggests the lowest carbon storage response to increasing CO <sub>2</sub> ( [[#Arora--2020|Arora et al., 2020]] ). The lower CO <sub>2</sub> effect due to decreased nutrient availability is generally consistent with analyses of the implicit nutrient limitation in CMIP5 simulations ( [[#Wieder--2015|Wieder et al., 2015]] ; [[#Zaehle--2015|Zaehle et al., 2015]] ) and independent assessments by stand-alone land models ( [[#Zaehle--2010|Zaehle et al., 2010]] ; [[#Wårlind--2014|Wårlind et al., 2014]] ; [[#Zhang--2014|Zhang et al., 2014]] ; [[#Goll--2017|Goll et al., 2017]] ; [[#Meyerholt--2020|Meyerholt et al., 2020]] ). The simulated effects are generally consistent with expectations based on independent observations ( [[#Walker--2021|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--2019|Song et al., 2019]] ; [[#Wieder--2019|Wieder et al., 2019]] ; [[#Meyerholt--2020|Meyerholt et al., 2020]] ).

Our understanding of the various biological processes that affect the strength of the CO <sub>2</sub> 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 CO <sub>2</sub> fertilization of photosynthesis acts as an important negative feedback on anthropogenic climate change, by reducing the rate at which CO <sub>2</sub> 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 CO <sub>2</sub> fertilization effect ( [[#Arora--2020|Arora et al., 2020]] ). The magnitude of the direct CO <sub>2</sub> effect on land carbon uptake, and its limitation by nutrients, remains uncertain.

<div id="5.4.2" class="h2-container"></div>

<span id="direct-co-2-effects-on-projected-ocean-carbon-uptake"></span>
=== 5.4.2 Direct CO <sub>2</sub> Effects on Projected Ocean Carbon Uptake ===

<div id="h2-21-siblings" class="h2-siblings"></div>

In AR5 (WGI, Section 6.4.2) there was ''high agreement'' that CMIP5 ESMs project continued ocean CO <sub>2</sub> 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 CO <sub>2</sub> , or in the inter-model spread, between the CMIP5 and CMIP6 era ( [[#Arora--2020|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 ( [[#5.4.5.3|Section 5.4.5.3]] ). CMIP6 models also show a similar time evolution of global ocean CO <sub>2</sub> uptake to CMIP5 models over the 21st century (Figure 5.25) with decreasing net ocean CO <sub>2</sub> uptake ratio to anthropogenic CO <sub>2</sub> 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--1957|Revelle and Suess, 1957]] ; [[#Egleston--2010|Egleston et al., 2010]] ), provides ''high confidence'' that the excess CO <sub>2</sub> dissolved in seawater leads to a non-linear reduction of the CO <sub>2</sub> buffering capacity, that is smaller dissolved inorganic carbon (DIC) increase with respect to ''p'' CO <sub>2</sub> increase along with the increase in cumulative ocean CO <sub>2</sub> uptake. Recent studies ( [[#Katavouta--2018|Katavouta et al., 2018]] ; [[#Jiang--2019|Jiang et al., 2019]] ; [[#Arora--2020|Arora et al., 2020]] ; [[#Rodgers--2020|Rodgers et al., 2020]] ) suggest with ''medium confidence'' that the decrease in the ocean CO <sub>2</sub> uptake ratio to anthropogenic CO <sub>2</sub> 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 CO <sub>2</sub> 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 CO <sub>2</sub> 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 ''p'' CO <sub>2</sub> , which impacts the mean annual air–sea fluxes ( [[#Hauck--2015|Hauck et al., 2015]] ; [[#Fassbender--2018|Fassbender et al., 2018]] ; [[#Landschützer--2018|Landschützer et al., 2018]] ; SROCC, [[#5.2.2.3|Section 5.2.2.3]] ). A larger amplification of the surface ocean ''p'' CO <sub>2</sub> seasonality occurs in the subtropics where ''p'' CO <sub>2</sub> seasonality is dominated by temperature seasonality, with the summer increase in the difference in ''p'' CO <sub>2</sub> <sup></sup> between surface water and the overlying atmosphere reaching 3μatm per decade between 1990 and 2030 under RCP8.5 ( [[#Schlunegger--2019|Schlunegger et al., 2019]] ; [[#Rodgers--2020|Rodgers et al., 2020]] ). In contrast, the impact of biological production on the seasonal cycle of ''p'' CO <sub>2</sub> in summer in the Southern Ocean strengthens the drawdown of CO <sub>2</sub> ( [[#Hauck--2015|Hauck et al., 2015]] ).

Overall, there is ''medium confidence'' on three outcomes in the ocean from projected CO <sub>2</sub> uptake under medium to high CO <sub>2</sub> concentration scenarios: (i) a weakening of the buffering capacity, which impacts the airborne fraction via the reduction of the ocean CO <sub>2</sub> buffering capacity due to cumulative ocean CO <sub>2</sub> uptake, which reduces the net ocean CO <sub>2</sub> uptake ratio to anthropogenic CO <sub>2</sub> emissions ( [[#Katavouta--2018|Katavouta et al., 2018]] ; [[#Arora--2020|Arora et al., 2020]] ; [[#Rodgers--2020|Rodgers et al., 2020]] ); (ii) an amplification of the seasonal cycle of CO <sub>2</sub> variables, which impacts both the ocean sink and ocean acidification ( [[#Hauck--2015|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 CO <sub>2</sub> ( [[#McNeil--2016|McNeil and Sasse, 2016]] ) (Cross-Chapter Box 5.3).

<div id="5.4.3" class="h2-container"></div>

<span id="climate-effect-on-land-carbon-uptake"></span>
=== 5.4.3 Climate Effect on Land Carbon Uptake ===

<div id="h2-22-siblings" class="h2-siblings"></div>

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 ( [[#5.4.5|Section 5.4.5]] ), continues to support a ''medium confidence'' assessment that future climate change will decrease land carbon uptake relative to the case with constant climate.

<div id="5.4.3.1" class="h3-container"></div>

<span id="plant-physiology"></span>
==== 5.4.3.1 Plant Physiology ====

<div id="h3-29-siblings" class="h3-siblings"></div>

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--2016|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--2017|Piao et al., 2017]] ; [[#Buermann--2018|Buermann et al., 2018]] ; [[#Liu--2019|Liu et al., 2019]] ). At centennial time scales, boreal forest expansion may act as a climate-driven carbon sink ( [[#Pugh--2018|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--2012|Lin et al., 2012]] ; [[#Yamori--2014|Yamori et al., 2014]] ; [[#Smith--2017|Smith and Dukes, 2017]] ; [[#Grossiord--2020|Grossiord et al., 2020]] ). Observations and models suggest that the vapour pressure deficit effects are stronger than direct temperature effects on enzyme activities ( [[#Smith--2020|Smith et al., 2020]] ), and that acclimation of photosynthetic optimal temperature may mitigate productivity losses of tropical forests under climate change ( [[#Kattge--2007|Kattge and Knorr, 2007]] ; [[#Tan--2017|Tan et al., 2017]] ; [[#Kumarathunge--2019|Kumarathunge et al., 2019]] ). Some models have begun to include these acclimation responses in photosynthesis and autotrophic respiration ( [[#Lombardozzi--2015|Lombardozzi et al., 2015]] ; [[#Smith--2015|Smith et al., 2015]] ; [[#Huntingford--2017|Huntingford et al., 2017]] ; [[#Mercado--2018|Mercado et al., 2018]] ).

<div id="5.4.3.2" class="h3-container"></div>

<span id="fire-and-other-disturbances"></span>
==== 5.4.3.2 Fire and Other Disturbances ====

<div id="h3-30-siblings" class="h3-siblings"></div>

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 ( [[IPCC:Wg1:Chapter:Chapter-12#12.3.2.8|Section 12.3.2.8]] ) and may drive increased fire frequency and severity in several regions, including Arctic and boreal ecosystems ( [[#Gauthier--2015|Gauthier et al., 2015]] ; X.J. [[#Walker--2019|]] [[#Walker--2019|Walker et al., 2019]] ), Mediterranean-type ecosystems ( [[#Turco--2014|Turco et al., 2014]] ; [[#Jin--2015|Jin et al., 2015]] ), degraded tropical forests ( [[#Aragão--2018|Aragão et al., 2018]] ), and tropical forest-savanna transition zones ( [[#Lehmann--2014|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--2017|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--2019|Teckentrup et al., 2019]] ; [[#Hantson--2020|Hantson et al., 2020]] ; [[#Lasslop--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--2004|Cox et al., 2004]] ; [[#Jones--2009|Jones et al., 2009]] ; [[#Brando--2014|Brando et al., 2014]] ; [[#Le%20Page--2017|Le Page et al., 2017]] ; [[#Zemp--2017|Zemp et al., 2017]] ), and temperate and boreal regions ( [[#Joos--2001|Joos et al., 2001]] ; [[#Lucht--2006|Lucht et al., 2006]] ; [[#Scheffer--2012|Scheffer et al., 2012]] ; [[#Lasslop--2016|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--2016|Levine et al., 2016]] ; [[#Sakschewski--2016|Sakschewski et al., 2016]] ), supporting the AR5 assessment ( [[#5.4.9|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--2015|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--2015|Allen et al., 2015]] ; [[#McDowell--2018|McDowell et al., 2018]] ; [[#Grossiord--2020|Grossiord et al., 2020]] ), and that these processes are not well represented in ESMs ( [[#Powell--2013|Powell et al., 2013]] ; [[#Fisher--2018|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--2014|Friend et al., 2014]] ).

Overall, climate change will force widespread increases in fire weather throughout the world ( [[IPCC:Wg1:Chapter:Chapter-12#12.3.2.8|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--2014a|Eliseev et al. (2014a)]] and [[#Harrison--2018|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, but ''very 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.

<div id="5.4.3.3" class="h3-container"></div>

<span id="soil-carbon"></span>
==== 5.4.3.3 Soil Carbon ====

<div id="h3-31-siblings" class="h3-siblings"></div>

Changes to soil carbon stocks in response to climate change are a potentially strong positive feedback ( [[#Cox--2000|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--2007|Kleber et al., 2007]] ; [[#Schmidt--2011|Schmidt et al., 2011]] ; [[#Luo--2016|Luo et al., 2016]] ); and (iii) increased understanding of soil carbon dynamics in subsurface layers ( [[#Hicks%20Pries--2017|Hicks Pries et al., 2017]] ; [[#Balesdent--2018|Balesdent et al., 2018]] ).

CMIP6 ESMs predict losses of soil carbon with warming, which are larger than climate-driven vegetation carbon losses ( [[#Arora--2020|Arora et al., 2020]] ). As in CMIP5 ( [[#Todd-Brown--2013|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--2015b|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 soil <sup>14</sup> C responses to atmospheric <sup>14</sup> C 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--2016|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--2017|Koven et al., 2017]] ; [[#Wieder--2018|Wieder et al., 2018]] ; [[#Varney--2020|Varney et al., 2020]] ), although experimental soil warming in tropical forests suggest high sensitivity of decomposition to warming in those regions as well ( [[#Nottingham--2020|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--2020|Chaudhary et al., 2020]] ) and tropical ( [[#Cobb--2017|Cobb et al., 2017]] ) latitudes, and in combination with anthropogenic drainage of peatlands ( [[#Warren--2017|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--2017|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--2013|Wieder et al., 2013]] ; [[#Guenet--2018|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--2011|Melillo et al., 2011]] ). Models that include this process tend to show a weaker carbon–climate feedback than those that do not ( [[#Thornton--2009|Thornton et al., 2009]] ; [[#Zaehle--2010|Zaehle et al., 2010]] ; [[#Wårlind--2014|Wårlind et al., 2014]] ; [[#Meyerholt--2020|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--2020|Arora et al., 2020]] ; [[#5.4.8|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--2016|Sakaguchi et al., 2016]] ) leading to either changes in species composition or changes in plant tissue nutrient to carbon ratios ( [[#Thomas--2015|Thomas et al., 2015]] ; [[#Achat--2016|Achat et al., 2016]] ; [[#Du--2019|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 ( [[#5.4.8|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.

<div id="box-5.1" class="h2-container box-container"></div>

'''Box 5.1 | Permafrost Carbon and Feedb''' '''acks to Climate'''

<div id="h2-23-siblings" class="h2-siblings"></div>

'''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--2014|Hugelius et al., 2014]] ; [[#Strauss--2017|Strauss et al., 2017]] ; [[#Mishra--2021|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--2014|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--2014|Loisel et al., 2014]] ; [[#Lindgren--2018|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 ( <sup>14</sup> C) signature of respiration at sites undergoing rapid permafrost thaw ( [[#Hicks%20Pries--2013|Hicks Pries et al., 2013]] ), of dissolved organic carbon in rivers draining watersheds with permafrost thaw ( [[#Vonk--2015|Vonk et al., 2015]] ; [[#Wild--2019|Wild et al., 2019]] ), and of methane (CH <sub>4</sub> ) produced in thawing lakes ( [[#Walter%20Anthony--2016|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 CO <sub>2</sub> emissions during early winter at Barrow, Alaska ( [[#Sweeney--2016|Sweeney et al., 2016]] ; [[#Webb--2016|Webb et al., 2016]] ; [[#Meredith--2019|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--2019|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--2018|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 CO <sub>2</sub> source for the tundra in Alaska ( [[#Commane--2017|Commane et al., 2017]] ), which is equivalent to 0.3 PgC yr <sup>–1</sup> when scaled up to the northern permafrost region ( ''low confidence'' ) ( [[#Meredith--2019|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 CO <sub>2</sub> 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--2013|Graven et al., 2013]] ; [[#Forkel--2016|Forkel et al., 2016]] ; [[#Takata--2017|Takata et al., 2017]] ; [[#Bruhwiler--2021|Bruhwiler et al., 2021]] ). Estimates of CO <sub>2</sub> 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--2016|Welp et al., 2016]] ; [[#Takata--2017|Takata et al., 2017]] ).

In addition to CO <sub>2</sub> , CH <sub>4</sub> 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 CH <sub>4</sub> emissions from diverse sources throughout the permafrost region. These include observed regional lake area change, which suggest a 1.6–5 Tg CH <sub>4</sub> yr <sup>–1</sup> increase over the last 50 years ( [[#Walter%20Anthony--2016|Walter Anthony et al., 2016]] ), ice-capped geological sources ( [[#Walter%20Anthony--2012|Walter Anthony et al., 2012]] ; [[#Kohnert--2017|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 CH <sub>4</sub> yr <sup>–1</sup> ( [[#Thornton--2016b|Thornton et al., 2016b]] ) <sup></sup> to 17 Tg CH <sub>4</sub> yr <sup>–1</sup> ( [[#Shakhova--2014|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 CH <sub>4</sub> concentration measurements ( [[#Berchet--2016|Berchet et al., 2016]] ).

Atmospheric measurements and inversions performed at the global and regional scales do not show any detectable trends in annual mean CH <sub>4</sub> emissions from the permafrost region over the past 30 years ( [[#Jackson--2020|Jackson et al., 2020]] ; [[#Saunois--2020|Saunois et al., 2020]] ; [[#Bruhwiler--2021|Bruhwiler et al., 2021]] ), consistent with atmospheric measurements in Alaska that showed no significant annual trends, despite significant increase in air temperature ( [[#Sweeney--2016|Sweeney et al., 2016]] ). Atmospheric inversions and biospheric models do not show any clear trends in CH <sub>4</sub> emissions for wetland regions of the high latitudes during the period 2000–2016 ( [[#Patra--2016|Patra et al., 2016]] ; [[#Poulter--2017|Poulter et al., 2017]] ; [[#Jackson--2020|Jackson et al., 2020]] ; [[#Saunois--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 CH <sub>4</sub> 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--2016a|Thornton et al., 2016a]] ; [[#Saunois--2020|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 CO <sub>2</sub> . There is ''robust evidence'' that some CH <sub>4</sub> 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 CO <sub>2</sub> and CH <sub>4</sub> 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 (N <sub>2</sub> O) release from soil ( [[#Abbott--2015|Abbott and Jones, 2015]] ; [[#Karelin--2017|Karelin et al., 2017]] ; [[#Wilkerson--2019|Wilkerson et al., 2019]] ), a previously unaccounted source. However, this release is unquantified at the pan-Arctic scale.

'''What does the paleo record 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--2014|Köhler et al., 2014]] ; [[#Crichton--2016|Crichton et al., 2016]] ; [[#Winterfeld--2018|Winterfeld et al., 2018]] ; [[#Meyer--2019|Meyer et al., 2019]] ), supporting the vulnerability of these areas to further warming ( [[#Strauss--2013|Strauss et al., 2013]] , 2017).

Radiogenic and stable isotopic measurements on CH <sub>4</sub> trapped in Antarctic ice support the view that CH <sub>4</sub> emissions from fossil carbon reservoirs, including permafrost and methane hydrates, remained small in response to the deglacial warming. Mass-balance calculations reveal that geological CH <sub>4</sub> emissions have not exceeded 19 Tg yr <sup>–1</sup> , highlighting that the deglacial increase in CH <sub>4</sub> emissions was predominantly related to contemporary CH <sub>4</sub> emissions from tropical wetlands and seasonally inundated floodplains ( [[#Bock--2017|Bock et al., 2017]] ; [[#Petrenko--2017|Petrenko et al., 2017]] ; [[#Dyonisius--2020|Dyonisius et al., 2020]] ). Isotopic constraints on CO <sub>2</sub> losses from permafrost with warming after the Last Glacial Maximum (LGM) are weaker than for CH <sub>4</sub> . 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--2012|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%20Anthony--2014|Walter Anthony et al., 2014]] ), and long-term accumulation of carbon in permafrost soils following LDT carbon loss ( [[#Lindgren--2018|Lindgren et al., 2018]] ), particularly in peatlands which accumulated carbon at a slow but persistent rate in warm paleoclimates ( [[#Treat--2019|Treat et al., 2019]] ).

In conclusion, several independent lines of evidence indicate that permafrost thaw did not release vast quantities of fossil CH <sub>4</sub> associated with the transient warming events of the LDT. This suggests that large emissions of CH <sub>4</sub> from old carbon sources will not occur in response to future warming ( ''medi'' ''um 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'' ) ( [[IPCC:Wg1:Chapter:Chapter-9#9.5.2|Section 9.5.2]] ), thus creating the potential for releasing CO <sub>2</sub> and CH <sub>4</sub> 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--2016a|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%20von%20Deimling--2015|Schneider von Deimling et al., 2015]] ; [[#Turetsky--2020|Turetsky et al., 2020]] ). Increased fire frequency and severity ( [[#Hu--2010|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--2013|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--2016|Salmon et al., 2016]] ), but also lead to biophysical feedbacks that further amplify warming ( [[#Myers-Smith--2011|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--2019|Meredith et al., 2019]] ; SROCC, Figure 3.11). Under lower emissions scenarios, [[#Schneider%20von%20Deimling--2015|Schneider von Deimling et al. (2015)]] estimated permafrost feedbacks of 20–58 PgC of CO <sub>2</sub> by 2100 under an RCP2.6 scenario, and 28–92 PgC of CO <sub>2</sub> under an RCP4.5 scenario.

This new assessment, based on studies included in or published since SROCC ( [[#Schaefer--2014|Schaefer et al., 2014]] ; [[#Koven--2015c|Koven et al., 2015c]] ; [[#Schneider%20von%20Deimling--2015|Schneider von Deimling et al., 2015]] ; [[#Schuur--2015|Schuur et al., 2015]] ; [[#MacDougall--2016a|MacDougall and Knutti, 2016a]] ; [[#Gasser--2018|Gasser et al., 2018]] ; [[#Yokohata--2020|Yokohata et al., 2020]] ), estimates that the permafrost CO <sub>2</sub> feedback per degree of global warming (Figure 5.29) is 18 [3.1 to 41, 5–95% range] PgC °C <sup>–1</sup> . The assessment is based on a wide range of scenarios evaluated at 2100, and an assessed estimate of the permafrost CH <sub>4</sub> -climate feedback at 2.8 [0.7 to 7.3] PgCeq °C <sup>–1</sup> (Figure 5.29). This feedback affects the remaining carbon budgets for climate stabilization and is included in their assessment ( [[#5.5.2|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%20von%20Deimling--2015|Schneider von Deimling et al., 2015]] ; [[#McGuire--2018|McGuire et al., 2018]] ). [[#Schneider%20von%20Deimling--2015|Schneider von Deimling et al. (2015)]] estimated that thawing permafrost could release 20–40 PgC of CO <sub>2</sub> in the period from 2100 to 2300 under an RCP2.6 scenario, and 115–172 PgC of CO <sub>2</sub> under an RCP8.5 scenario. The multi-model ensemble ( [[#McGuire--2018|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 CH <sub>4</sub> over the 21st century and 2800–7400 Tg CH <sub>4</sub> from 2100–2300 ( [[#Schneider%20von%20Deimling--2015|Schneider von Deimling et al., 2015]] ), and as 5300 Tg CH <sub>4</sub> over the 21st century and 16,000 Tg CH <sub>4</sub> from 2100–2300 ( [[#Turetsky--2020|Turetsky et al., 2020]] ). For RCP4.5, these numbers are 538–2356 Tg CH <sub>4</sub> until 2100 and 2000–6100 Tg CH <sub>4</sub> from 2100–2300 ( [[#Schneider%20von%20Deimling--2015|Schneider von Deimling et al., 2015]] ), and 4100 Tg CH <sub>4</sub> until 2100 and 10,000 Tg CH <sub>4</sub> from 2100–2300 ( [[#Turetsky--2020|Turetsky et al., 2020]] ).

A key uncertainty is whether permafrost carbon feedbacks scale roughly linearly with warming ( [[#Koven--2015c|Koven et al., 2015c]] ), or instead scale at a greater ( [[#MacDougall--2016b|MacDougall and Knutti, 2016b]] ; [[#McGuire--2018|McGuire et al., 2018]] ) or smaller rate (e.g., CH <sub>4</sub> emissions estimated by [[#Turetsky--2020|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--2015|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--2015a|Koven et al., 2015a]] ; [[#McGuire--2018|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 CO <sub>2</sub> released from permafrost will be 18 (3.1–41) PgC °C <sup>–1</sup> by 2100, with the relative contribution of CO <sub>2</sub> vs CH <sub>4</sub> remaining poorly constrained. Permafrost carbon feedbacks are included among the under-represented feedbacks quantified in Figure 5.29.

<div id="5.4.4" class="h2-container"></div>

<span id="climate-effects-on-future-ocean-carbon-uptake"></span>
=== 5.4.4 Climate Effects on Future Ocean Carbon Uptake ===

<div id="h2-24-siblings" class="h2-siblings"></div>

<div id="5.4.4.1" class="h3-container"></div>

<span id="physical-drivers-of-future-ocean-carbon-uptake-and-storage"></span>
==== 5.4.4.1 Physical Drivers of Future Ocean Carbon Uptake and Storage ====

<div id="h3-32-siblings" class="h3-siblings"></div>

The principal contribution to increasing global ocean carbon is the air–sea flux of CO <sub>2</sub> , which changes the dissolved inorganic carbon (DIC) inventory ( [[#5.4.2|Section 5.4.2]] ; [[#Arora--2020|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 CO <sub>2</sub> : (i) ocean warming reduces the solubility of CO <sub>2</sub> , which increases ''p'' CO <sub>2</sub> and increases the stratification of the mixed layer, both acting as positive feedbacks weakening the ocean sink ( [[IPCC:Wg1:Chapter:Chapter-9#9.2.1|Section 9.2.1]] and Cross-Chapter Box 5.3; [[#Arora--2020|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--2018|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 ( [[#5.2.3.1|Section 5.2.3.1]] ; [[#Gruber--2019b|Gruber et al., 2019b]] ; [[#Arora--2020|Arora et al., 2020]] ). The land-to-ocean riverine flux and the carbon burial in ocean sediments play a minor role ( ''low confidence'' ) ( [[#Arora--2020|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--2020|Arora et al., 2020]] ).

Regionally, the Southern Ocean is a major sink of anthropogenic CO <sub>2</sub> (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--2015|Ito et al., 2015]] ); alternatively, increased winds may intensify the overturning circulation, reducing the net CO <sub>2</sub> sink in the Southern Ocean ( [[#Bronselaer--2018|Bronselaer et al., 2018]] ; [[#Saunders--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 CO <sub>2</sub> uptake.

<div id="5.4.4.2" class="h3-container"></div>

<span id="biological-drivers-of-future-ocean-carbon-uptake"></span>
==== 5.4.4.2 Biological Drivers of Future Ocean Carbon Uptake ====

<div id="h3-33-siblings" class="h3-siblings"></div>

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 CO <sub>2</sub> storage. Projected reductions in ocean ventilation ( [[IPCC:Wg1:Chapter:Chapter-9#9.2.1.4|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, [[#5.2.2.6|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 CO <sub>2</sub> concentration and climate warming wasunclear ( [[#Taucher--2011|Taucher and Oschlies, 2011]] ; [[#Laufkötter--2015|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--2020|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--2016|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 ( [[#5.4.6|Section 5.4.6]] ; [[#Kwiatkowski--2017|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--2011|Taucher and Oschlies, 2011]] ; [[#Laufkoetter--2015|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 ( [[IPCC:Wg1:Chapter:Chapter-9#9.2.1.4|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--2020|Kwiatkowski et al., 2020]] ). Future changes to dust deposition due to desertification ( [[#Mahowald--2017|Mahowald et al., 2017]] ), alterations to the nitrogen cycle ( [[#5.3.3.2|Section 5.3.3.2]] ; SROCC, [[#5.2.3.1.2|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--2018|Boscolo-Galazzo et al., 2018]] ; [[#Brewer--2019|Brewer, 2019]] ; [[#Cavan--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--2017|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--2014|Jiao et al., 2014]] ; [[#Legendre--2015|Legendre et al., 2015]] ; [[#Roshan--2017|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%20Lyle--2006|Olivarez Lyle and Lyle, 2006]] ).

Most models project that smaller phytoplankton are favoured in future ocean conditions ( ''medium confidence'' ; [[#Cabré--2015|Cabré et al., 2015]] ; [[#Fu--2016|Fu et al., 2016]] ; [[#Flombaum--2020|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--2013|Herndl and Reinthaler, 2013]] ; [[#Bach--2016|Bach et al., 2016]] ; [[#Richardson--2019|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--2014|Jiao et al., 2014]] ) or generation of slower sinking particles ( [[#Guidi--2009|Guidi et al., 2009]] ; [[#Leung--2021|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--2020|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--2019|Krumhardt et al., 2019]] ; [[#Raven--2021|Raven et al., 2021]] ; [[#Taucher--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 CO <sub>2</sub> uptake is expected to become more significant with continued climate change ( [[#Hauck--2015|Hauck et al., 2015]] ).

<div id="5.4.5" class="h2-container"></div>

<span id="carbon-cycle-projections-in-earth-system-models"></span>
=== 5.4.5 Carbon Cycle Projections in Earth System Models ===

<div id="h2-25-siblings" class="h2-siblings"></div>

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 CO <sub>2</sub> ( [[#Friedlingstein--2014b|Friedlingstein et al., 2014b]] ) primarily associated with different magnitudes of carbon–climate and carbon-concentration feedbacks ( [[#Arora--2013|Arora et al., 2013]] ), but also exacerbated by differences in the simulation of the net carbon release from land-use change ( [[#Brovkin--2013|Brovkin et al., 2013]] ). A key deficiency of almost all CMIP5 ESMs was the neglect of nutrient limitations on CO <sub>2</sub> -fertilization of land plant photosynthesis ( [[#5.4.1|Section 5.4.1]] ; [[#Zaehle--2015|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--2020|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 CO <sub>2</sub> emissions (‘emissions-driven’ runs), in which case atmospheric CO <sub>2</sub> 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 CO <sub>2</sub> and resulting changes in climate, but do not feed back through changes in the atmospheric CO <sub>2</sub> concentration. Concentration-driven runs are used to diagnose the carbon emissions consistent with the Shared Socio-economic Pathways (SSPs) and other prescribed concentration scenarios ( [[#5.5|Section 5.5]] ). In this subsection we specifically analyse results from concentration-driven ESM projections.

<div id="5.4.5.1" class="h3-container"></div>

<span id="evaluation-of-the-contemporary-carbon-cycle-in-concentration-driven-runs"></span>
==== 5.4.5.1 Evaluation of the Contemporary Carbon Cycle in Concentration-driven Runs ====

<div id="h3-34-siblings" class="h3-siblings"></div>

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--2013|Arora et al., 2013]] ; [[#Friedlingstein--2014b|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--2018|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.

<div id="_idContainer064" class="Basic-Text-Frame"></div>

[[File:f34fe91fc558c40f0269ddc8c1522426 IPCC_AR6_WGI_Figure_5_22.png]]

'''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).

<div id="_idContainer062" class="_idGenObjectStyleOverride-1"></div>

'''Table 5.4 |''' '''Properties of the CMIP6 Earth system models (ESMs), focusing on the land and ocean carbon cycle components of these models''' '''''(Ar''''' or '''''a''''' '''''et''''' ''a'' l., 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).

{| class="wikitable"
|-
! 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

|-
| colspan="12"| '''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

|-
| colspan="12"| '''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

|}

<div id="5.4.5.2" class="h3-container"></div>

<span id="evaluation-of-historical-carbon-cycle-simulations-in-concentration-driven-runs"></span>
==== 5.4.5.2 Evaluation of Historical Carbon Cycle Simulations in Concentration-driven Runs ====

<div id="h3-35-siblings" class="h3-siblings"></div>

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%20Quéré--2018a|Le Quéré et al., 2018a]] ). For each model, common historical land-use changes were prescribed ( [[#Jones--2016a|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 <sup>–1</sup> ) clustering around the central GCP estimate of 2.6 ± 0.5 GtC yr <sup>–1</sup> . 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--2020|Watson et al., 2020]] ).

<div id="_idContainer066" class="Basic-Text-Frame"></div>

[[File:ea8311b6d9b7e8b3cca15fb75aa5f6cf IPCC_AR6_WGI_Figure_5_23.png]]

'''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--2020a|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.

<div id="5.4.5.3" class="h3-container"></div>

<span id="evaluation-of-latitudinal-distribution-of-simulated-carbon-sinks"></span>
==== 5.4.5.3 Evaluation of Latitudinal Distribution of Simulated Carbon Sinks ====

<div id="h3-36-siblings" class="h3-siblings"></div>

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--2005|Chevallier et al., 2005]] ), Carbon Tracker 2017 ( [[#Peters--2007|Peters et al., 2007]] ) and Model for Interdisciplinary Research on Climate Atmospheric Transport Model (MIROC-ATM4; [[#Saeki--2017|Saeki and Patra, 2017]] ). The ocean carbon cycle components of CMIP6 ESMs are able to simulate the tropical CO <sub>2</sub> source and mid-latitude CO <sub>2</sub> 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 CO <sub>2</sub> measurement networks and from transport model uncertainties.

<div id="_idContainer068" class="Basic-Text-Frame"></div>

[[File:241de5eb0312fe9520a7064d39e739db IPCC_AR6_WGI_Figure_5_24.png]]

'''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 CO <sub>2</sub> -fertilization within CMIP6 models was expected to reduce this discrepancy ( [[#Anav--2013|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--2019|Ciais et al., 2019]] ).

<div id="5.4.5.4" class="h3-container"></div>

<span id="coupled-climatecarbon-cycle-projections"></span>
==== 5.4.5.4 Coupled Climate–Carbon Cycle Projections ====

<div id="h3-37-siblings" class="h3-siblings"></div>

Land and ocean carbon uptake are driven primarily by increases in atmospheric CO <sub>2</sub> (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 CO <sub>2</sub> (such as SSP5-8.5 and SSP3-7.0) the absolute values of the sinks are larger, but the fraction of implied emissions taken up by the sinks declines through the 21 <sup>st</sup> century. By contrast, scenarios that assume CO <sub>2</sub> stabilization in the 21 <sup>st</sup> 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 <sup>–1</sup> by 2100 under SSP1-2.6, and peaks around 2080 at 5.4 ± 0.4 GtC yr <sup>–1</sup> under 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 <sup>–1</sup> by 2100 under SSP1-2.6, and to reach around 5.6 ± 3.7 GtC yr <sup>–1</sup> under 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.

<div id="_idContainer070" class="_idGenObjectStyleOverride-1"></div>

[[File:5b10d45d4115f53c9bd7f695963d249b IPCC_AR6_WGI_Figure_5_25.png]]

'''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 CO <sub>2</sub> concentrations; '''(b)''' five-year running mean ocean carbon sink (GtC yr <sup>–1</sup> ); '''(c)''' five-year running mean net land carbon sink (GtC yr <sup>–1</sup> ); '''(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 <sup>–1</sup> and 1 GtC yr <sup>–1</sup> 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 CO <sub>2</sub> and warming ( [[#5.4.5.5|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.

<div id="_idContainer072" class="_idGenObjectStyleOverride-1"></div>

[[File:2a259b3269d37076a2119b78309d917a IPCC_AR6_WGI_Figure_5_26.png]]

'''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 &lt;80% of models agree with the ensemble mean on the sign of change. On land, this is calculated as the time integral of net biome productivity (NBP), for the ocean it is the time-integral of air – sea carbon dioxide (CO <sub>2</sub> ) <sub></sub> 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 CO <sub>2</sub> , but the fraction of emissions taken up by land and ocean is expected to decline as the CO <sub>2</sub> 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 CO <sub>2</sub> by 2100 is dominated by future anthropogenic emissions rather than carbon–climate feedbacks ( ''hi'' ''gh confidence'' ).

<div id="5.4.5.5" class="h3-container"></div>

<span id="linear-feedback-analysis"></span>
==== 5.4.5.5 Linear Feedback Analysis ====

<div id="h3-38-siblings" class="h3-siblings"></div>

To diagnose the causes of the varying time-evolution of carbon sinks, the traditional linear feedback approach is adopted ( [[#Friedlingstein--2003|Friedlingstein et al., 2003]] ), as used previously to analyse C <sup>4</sup> MIP ( [[#Friedlingstein--2006|Friedlingstein et al., 2006]] ) and CMIP5 models ( [[#Arora--2013|Arora et al., 2013]] ). Changes in land carbon storage ( ''Δ'' ''C'' <sub>L</sub> ) and changes in ocean carbon storage ( Δ ''C'' <sub>o</sub> ) are decomposed into contributions arising from warming ( ''Δ'' ''T'' ) and increases in CO <sub>2</sub> ( Δ ''CO'' 2 ):

[[File:fd37bf72ee47cd82827d75d6c0fa6132 IPCC_AR6_WGI_Formula_Chapter_5_5455_1.jpg]]

where ''β'' <sub>L</sub> ( β <sub>o</sub> ) and γ <sub>L</sub> ( γ <sub>o</sub> ) are coefficients that represent the sensitivity of land (ocean) carbon storage to changes in CO <sub>2</sub> and global mean temperature respectively. This feedback formalism is one of several that have been proposed for analysing climate–carbon cycle feedbacks ( [[#Lade--2018|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 ( [[#5.4.6|Section 5.4.6]] ) and to reconstruct the transient climate response to cumulative CO <sub>2</sub> emissions (TCRE) ( [[#Jones--2020|Jones and Friedlingstein, 2020]] ), as in [[#5.5|Section 5.5]] . To minimize the confounding effect of the scenario dependence, β and γ values are diagnosed from idealized runs in which a 1% per year increase in atmospheric CO <sub>2</sub> concentration is prescribed, as for AR5 (WGI, Box 6.4; [[#Arora--2013|Arora et al., 2013]] ). Values of β are calculated from ‘biogeochemical’ runs in which the prescribed CO <sub>2</sub> increases do not affect climate, and these are then used to isolate γ values in fully coupled runs where both climate and CO <sub>2</sub> change ( [[#Friedlingstein--2003|Friedlingstein et al., 2003]] ).

Table 5.5 shows the global land and global ocean values of β and γ for each of the CMIP6 ESMs ( [[#Arora--2020|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 CO <sub>2</sub> ( β <sub>L</sub> ) and the sensitivity of land carbon storage to temperature ( γ <sub>L</sub> ). 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 mean γ <sub>L</sub> moving from –58 ± 38 GtC K <sup>–1</sup> to –33 ± 33 GtC K <sup>–1</sup> . Between CMIP5 and CMIP6, there are also reductions in ensemble mean β <sub>o</sub> (0.82 to 0.77 GtC ppm <sup>–1</sup> ), β <sub>L</sub> (0.93 to 0.89 GtC ppm <sup>–1</sup> ) and γ <sub>o</sub> (–17.3 to –16.9 GtC K <sup>–1</sup> ), but these are progressively less significant compared to the model spread in each case.

<div id="_idContainer073" class="Basic-Text-Frame"></div>

'''Table 5.5 |''' '''Diagnosed global feedback parameters for CMIP6 ESMs based on 1% per year runs to 4×CO''' <sub>2</sub> (Arora et al., 2020). The last two rows show the mean and standard deviation across the CMIP6 and CMIP5 models, respectively.

{| class="wikitable"
|-
!
! colspan="2"| '''Land F''' '''eedback Factors'''

! colspan="2"| '''Ocean F''' '''eedback Factors'''

|-
! '''Model Name'''

! β '''L''' '''(PgC ppm''' –1 ''')'''

! γ '''L''' '''(PgC K''' –1 ''')'''

! β '''o''' '''(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 CO <sub>2</sub> runs, the CMIP6 models show reasonable agreement on the patterns of carbon uptake and also on the separate impacts of CO <sub>2</sub> increase and climate change (Figure 5.27). For the ensemble mean, increasing atmospheric CO <sub>2</sub> 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 CO <sub>2</sub> in almost all locations (Figure 5.27c).

<div id="_idContainer075" class="_idGenObjectStyleOverride-1"></div>

[[File:01a1dba8ed4789b745ddb11da3734ff4 IPCC_AR6_WGI_Figure_5_27.png]]

'''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 (CO''' <sub>2</sub> ''') 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 &lt;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 CO <sub>2</sub> 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).

<div id="5.4.6" class="h2-container"></div>

<span id="emergent-constraints-to-reduce-uncertainties-in-projections"></span>
=== 5.4.6 Emergent Constraints to Reduce Uncertainties in Projections ===

<div id="h2-26-siblings" class="h2-siblings"></div>

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--2019|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|Cox, 2019]] ). Figure 5.28 shows some key published emergent constraints on the carbon cycle in ESMs.

<div id="_idContainer077" class="Basic-Text-Frame"></div>

[[File:11335dd156d2654759d973cf07e96dad IPCC_AR6_WGI_Figure_5_28.png]]

'''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 (CO <sub>2</sub> ) concentration by 2060 under the RCP8. 5 emissions scenario against the simulated CO <sub>2</sub> in 2010 ( [[#Friedlingstein--2014b|Friedlingstein et al., 2014b]] ; [[#Hoffman--2014|Hoffman et al., 2014]] ); '''(b)''' sensitivity of tropical land carbon to warming ( γ <sub>LT</sub> ) against the sensitivity of the atmospheric CO <sub>2</sub> growth-rate to tropical temperature variability ( [[#Cox--2013|Cox et al., 2013]] ; [[#Wenzel--2014|Wenzel et al., 2014]] ); '''(c)''' sensitivity of extratropical (30°N–90°N) gross primary production to a doubling of atmospheric CO <sub>2</sub> against the sensitivity of the amplitude of the CO <sub>2</sub> seasonal cycle at Kumkahi, Hawaii to global atmospheric CO <sub>2</sub> concentration ( [[#Wenzel--2016|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--2019|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--2017|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|Cox (2019)]] . Further details on data sources and processing are available in the chapter data table (Table 5.SM.6).

<div id="5.4.7" class="h2-container"></div>

<span id="climate-feedbacks-from-ch-4-and-n-2-o"></span>
=== 5.4.7 Climate Feedbacks from CH <sub>4</sub> and N <sub>2</sub> O ===

<div id="h2-27-siblings" class="h2-siblings"></div>

Sources and sinks of CH <sub>4</sub> and N <sub>2</sub> O respond both directly and indirectly to atmospheric CO <sub>2</sub> 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--2017|Gasser et al., 2017]] ; [[#Lade--2018|Lade et al., 2018]] ; [[#Denisov--2019|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 [[IPCC:Wg1:Chapter:Chapter-7#7.4.1.1|Section 7.4.1.1]] , for CH <sub>4</sub> and N <sub>2</sub> O biogeochemical feedbacks. The strength of the feedbacks is estimated in a linear framework ( [[#Gregory--2009|Gregory et al., 2009]] ), using the radiative forcing equations for CO <sub>2</sub> , CH <sub>4</sub> and N <sub>2</sub> O ( [[#Etminan--2016|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--2010|Arneth et al., 2010]] ; [[#Thornhill--2021|Thornhill et al., 2021]] ). These assessed feedback parameters are used in [[IPCC:Wg1:Chapter:Chapter-7#7.4.2.5|Section 7.4.2.5]] .

The CH <sub>4</sub> 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). CH <sub>4</sub> emissions from wetlands and landfills generally increase with warming due to enhanced decomposition with higher temperatures, thereby potentially providing a positive CH <sub>4</sub> feedback on climate ( [[#Dean--2018|Dean et al., 2018]] ). The contribution of wetlands to interannual variability of atmospheric CH <sub>4</sub> 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 CH <sub>4</sub> growth rate is complex ( [[#Pison--2013|Pison et al., 2013]] ; [[#Nisbet--2016|Nisbet et al., 2016]] ; X. [[#Zhang--2020|]] [[#Zhang--2020|]] [[#Zhang--2020|Zhang et al., 2020]] ). As assessed by SROCC ( [[#IPCC--2019b|IPCC, 2019b]] ), there is ''high agreement'' across model simulations that wetlands CH <sub>4</sub> emissions will increase in the 21 <sup>st</sup> century, but ''low agreement'' in the magnitude of the change ( [[#Denisov--2013|Denisov et al., 2013]] ; [[#Shindell--2013|Shindell et al., 2013]] ; B.D. [[#Stocker--2013|]] [[#Stocker--2013|Stocker et al., 2013]] ; [[#Zhang--2017|Zhang et al., 2017]] ; [[#Koffi--2020|Koffi et al., 2020]] ). Climate change increases wetland emissions ( [[#Gedney--2004|Gedney et al., 2004]] , 2019; [[#Volodin--2008|Volodin, 2008]] ; [[#Ringeval--2011|Ringeval et al., 2011]] ; [[#Denisov--2013|Denisov et al., 2013]] ; [[#Shindell--2013|Shindell et al., 2013]] ) and gives rise to an estimated wetland CH <sub>4</sub> –climate feedback of 0.03 ± 0.01 W m <sup>–2</sup> °C <sup>–1</sup> (mean ± 1 standard deviation; ''limited evidence'' , ''high agreement'' ) ( [[#Arneth--2010|Arneth et al., 2010]] ; [[#Shindell--2013|Shindell et al., 2013]] ; B.D. [[#Stocker--2013|]] [[#Stocker--2013|Stocker et al., 2013]] ; [[#Zhang--2017|Zhang et al., 2017]] ). The effect of rising CO <sub>2</sub> on productivity, and therefore on the substrate for methanogenesis, can further increase the projected increase in wetland CH <sub>4</sub> emissions ( [[#Ringeval--2011|Ringeval et al., 2011]] ; [[#Melton--2013|Melton et al., 2013]] ). Model projections accounting for the combined effects of CO <sub>2</sub> and climate change suggest a potentially larger climate feedback (0.01–0.16 W m <sup>–2</sup> °C <sup>–1</sup> ) ( ''limited evidence'' , ''low agreement'' ) ( [[#Gedney--2019|Gedney et al., 2019]] ; [[#Thornhill--2021|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--2016|Stepanenko et al., 2016]] ; [[#Donis--2017|Donis et al., 2017]] ; [[#Beaulieu--2019|Beaulieu et al., 2019]] ). Methane emissions from thermokarst ponds and wetlands resulting from permafrost thaw are estimated to contribute an additional CH <sub>4</sub> -climate feedback of 0.01 [0.003 to 0.04, 5–95% range] W m <sup>–2</sup> °C <sup>–1</sup> ( ''limited evidence, l'' ''ow agreement'' ).

Methane release from wildfires may increase by up to a factor of 1.5 during the 21 <sup>st</sup> century ( [[#Eliseev--2014a|Eliseev et al., 2014a]] , b; [[#Kloster--2017|Kloster and Lasslop, 2017]] ). However, given the contemporary estimate for CH <sub>4</sub> from wildfires of no more than 16 TgCH <sub>4</sub> yr <sup>–1</sup> ( [[#van%20der%20Werf--2017|van der Werf et al., 2017]] ; [[#Saunois--2020|Saunois et al., 2020]] ), this feedback is small, adding no more than 40 ppb to the atmospheric CH <sub>44</sub> by the end of the 21 <sup>st</sup> century ( ''medium confidence'' ). Methane emissions from pan-Arctic freshwater bodies is also estimated to increase by 16 TgCH <sub>4</sub> yr <sup>–1</sup> in the 21 <sup>st</sup> century ( [[#Tan--2015|Tan and Zhuang, 2015]] ). Emissions from subsea and permafrost methane hydrates are not expected to change substantially in the 21 <sup>st</sup> century ( [[#5.4.9.1.3|Section 5.4.9.1.3]] ).

Land biosphere models show ''high agreement'' that long-term warming will increase N <sub>2</sub> O release from terrestrial ecosystems ( [[#Xu-Ri--2012|Xu-Ri et al., 2012]] ; B.D. [[#Stocker--2013|]] [[#Stocker--2013|Stocker et al., 2013]] ; [[#Zaehle--2013|Zaehle, 2013]] ; [[#Tian--2019|Tian et al., 2019]] ). A positive land N <sub>2</sub> O climate feedback is consistent with paleo-evidence based on reconstructed and modelled emissions during the last deglacial period ( [[#Schilt--2014|Schilt et al., 2014]] ; H. [[#Fischer--2019|]] [[#Fischer--2019|Fischer et al., 2019]] ; [[#Joos--2020|Joos et al., 2020]] ). The response of terrestrial N <sub>2</sub> O emissions to atmospheric CO <sub>2</sub> increase and associated warming is dependent on nitrogen availability ( [[#van%20Groenigen--2011|van Groenigen et al., 2011]] ; [[#Butterbach-Bahl--2013|Butterbach-Bahl et al., 2013]] ; [[#Tian--2019|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--2010|Elberling et al., 2010]] ; [[#Voigt--2017|Voigt et al., 2017]] ). There is ''medium confidence'' that the land N <sub>2</sub> O climate feedback is positive, but ''low confidence'' in the magnitude (0.02 ± 0.01 W m <sup>–2</sup> °C <sup>–1</sup> ).

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

Based-on these studies, there is ''medium confidence'' that the combined climate feedback parameter for CH <sub>4</sub> and N <sub>2</sub> O is positive, but there is ''low confidence'' in the magnitude of the estimate (0.05 [0.02 to 0.09] W m <sup>–2</sup> °C <sup>–1</sup> , 5–95% range).

<div id="5.4.8" class="h2-container"></div>

<span id="combined-biogeochemical-climate-feedback"></span>
=== 5.4.8 Combined Biogeochemical Climate Feedback ===

<div id="h2-28-siblings" class="h2-siblings"></div>

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 ( [[#5.4.5.5|Section 5.4.5.5]] ), independent estimates of CO <sub>2</sub> emissions due to permafrost thaw (Box 5.1) and fire ( [[#5.4.3.2|Section 5.4.3.2]] ), natural CH <sub>4</sub> and N <sub>2</sub> O emissions ( [[#5.4.7|Section 5.4.7]] ), and aerosol and atmospheric chemistry (Section 6.3.6). We derive a physical climate feedback parameter α , as defined in [[IPCC:Wg1:Chapter:Chapter-7#7.4.1.1|Section 7.4.1.1]] , for CO <sub>2</sub> -based feedbacks using the linear framework proposed by [[#Gregory--2009|Gregory et al. (2009)]] , using the radiative forcing equations for CO <sub>2</sub> ( [[#Etminan--2016|Etminan et al., 2016]] ).

<div id="_idContainer081" class="_idGenObjectStyleOverride-1"></div>

[[File:5351138eee306fa6bb1a9910c191dbd9 IPCC_AR6_WGI_Figure_5_29.png]]

'''Figure 5.29 |''' '''Estimates of the biogeochemical climate feedback parameter''' ( α ''')''' . The parameter α (W m <sup>−2</sup> °C <sup>−1</sup> ) for a feedback variable ''x'' is defined as

<code> α </code> <sub>x</sub> =

δ <code> N </code> d <code> x </code> / δ <code> x </code> d <code> T </code>

where

δ <code> N </code> / δ <code> x </code>

is the change in top-of-atmosphere energy balance in response to a change in ''x'' induced by a change in surface temperature ( ''T'' ), as in [[IPCC:Wg1:Chapter:Chapter-7#7.4.1.1|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--2013|Arora et al., 2013]] ) and CMIP6 ( [[#Arora--2020|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) [[#5.4.3.2|Section 5.4.3.2]] ( [[#Eliseev--2014a|Eliseev et al., 2014a]] ; [[#Harrison--2018|Harrison et al., 2018]] ); (b) [[#5.4.3.3|Section 5.4.3.3]] ( [[#Schneider%20von%20Deimling--2012|Schneider von Deimling et al., 2012]] ; [[#Burke--2013|Burke et al., 2013]] , 2017b; [[#Koven--2015a|Koven et al., 2015a]] , c; [[#MacDougall--2016b|MacDougall and Knutti, 2016b]] ; [[#Gasser--2018|Gasser et al., 2018]] ; [[#Kleinen--2018|Kleinen and Brovkin, 2018]] ), where the estimates from [[#Burke--2013|Burke et al., 2013]] have been constrained as assessed in their study (c) [[#5.4.7|Section 5.4.7]] ( [[#Schneider%20von%20Deimling--2012|Schneider von Deimling et al., 2012]] , 2015; [[#Koven--2015c|Koven et al., 2015c]] ; [[#Turetsky--2020|Turetsky et al., 2020]] ); (d) [[#5.4.7|Section 5.4.7]] ( [[#Arneth--2010|Arneth et al., 2010]] ; [[#Denisov--2013|Denisov et al., 2013]] ; [[#Shindell--2013|Shindell et al., 2013]] ; B.D. [[#Stocker--2013|]] [[#Stocker--2013|Stocker et al., 2013]] ; [[#Zhang--2017|Zhang et al., 2017]] ); (f) [[#5.4.7|Section 5.4.7]] ( [[#Xu-Ri--2012|Xu-Ri et al., 2012]] ; B.D. [[#Stocker--2013|]] [[#Stocker--2013|Stocker et al., 2013]] ; [[#Zaehle--2013|Zaehle, 2013]] ; [[#Tian--2019|Tian et al., 2019]] ); (g) [[#5.4.7|Section 5.4.7]] ( [[#Martinez-Rey--2015|Martinez-Rey et al., 2015]] ; [[#Landolfi--2017|Landolfi et al., 2017]] ; [[#Battaglia--2018b|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 CO <sub>2</sub> (–1.13 ± 0.45 W m <sup>–2</sup> °C <sup>–1</sup> , mean and 1 standard-deviation range) is dominated by the contribution of the CO <sub>2</sub> -induced increase of ocean and land carbon storage (–1.46 ± 0.41 W m <sup>–2</sup> °C <sup>–1</sup> , corresponding to a β <sub>L+O</sub> of 1.66 ± 0.31 PgC ppm <sup>–1</sup> ), with smaller contributions from the carbon cycle’s response to climate (0.24 ± 0.17 W m <sup>–2</sup> °C <sup>–1</sup> , corresponding to γ <sub>L+O</sub> of –50 ± 34 PgC °C <sup>–1</sup> ), and emissions from permafrost thaw (0.09 [0.02 to 0.20] W m <sup>–2</sup> °C <sup>–1</sup> , corresponding to γ of –18 [3 to 41] PgC °C <sup>–1</sup> , mean and 5–95% range) (Figure 5.29a). This estimate does not include an estimate of the fire-related CO <sub>2</sub> feedback (range: 0.01–0.06 W m <sup>–2</sup> °C <sup>–1</sup> ), as only ''limited evidence'' was available to inform its assessment. The sum (mean and 5–95th percentile range) of feedbacks from natural emissions of CH <sub>4</sub> including permafrost thaw, and N <sub>2</sub> O (0.05 [0.02 to 0.09] W m <sup>–2</sup> °C <sup>–1</sup> ), and feedbacks from aerosol and atmospheric chemistry (–0.20 [–0.41 to 0.01] W m <sup>–2</sup> °C <sup>–1</sup> ) leads to an estimate of the non-CO <sub>2</sub> biogeochemical feedback parameter of –0.15 [–0.36 to +0.06] W m <sup>–2</sup> °C <sup>–1</sup> . There is ''low confidence'' in the estimate of the non-CO <sub>2</sub> 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.

CO <sub>2</sub> and non-CO <sub>2</sub> biogeochemical feedbacks are an important component of the assessment of TCRE and the remaining carbon budget ( [[#5.5|Section 5.5]] ). The feedbacks of the carbon cycle of CO <sub>2</sub> 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-CO <sub>2</sub> biogeochemical feedback, can be converted into a carbon-equivalent feedback term ( γ ; [[#5.4.5.5|Section 5.4.5.5]] , 7.6) by reverse application of the linear feedback approximation ( [[#Gregory--2009|Gregory et al., 2009]] ). The contributions of non-CO <sub>2</sub> biogeochemical feedbacks combine to a linear feedback term of 30 ± 27 PgCeq °C <sup>–1</sup> (1 standard deviation range, 111 ± 98 Gt CO <sub>2-</sub> eq °C <sup>–1</sup> ), including a feedback term of –11 [–18 to –5] PgCeq °C <sup>–1</sup> (5–95% range, –40 [–62 to –18] Gt CO <sub>2-</sub> eq °C <sup>–1</sup> ) from natural CH <sub>4</sub> and N <sub>2</sub> O sources. The biogeochemical feedback from permafrost thaw leads to a combined linear feedback term of –21 ± 12 PgCeq °C <sup>–1</sup> (1 standard deviation range –77 ± 44 Gt CO <sub>2-</sub> eq °C <sup>–1</sup> ). For the integration of these feedbacks in the assessment of the remaining carbon budget ( [[#5.5.2|Section 5.5.2]] ), two individual non-CO <sub>2</sub> 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 [[#5.5.2|Section 5.5.2]] is assessed at a reduction of 7 ± 27 PgCeq °C <sup>–1</sup> (1 standard deviation range, –26 ± 97 PgCeq °C <sup>–1</sup> ). For the same reasons as for the feedback terms expressed in W m <sup>–2</sup> °C <sup>–1</sup> (see above), there is overall ''low confidence'' in the magnitude of these feedbacks.

<div id="5.4.9" class="h2-container"></div>

<span id="abrupt-changes-and-tipping-points"></span>
=== 5.4.9 Abrupt Changes and Tipping Points ===

<div id="h2-29-siblings" class="h2-siblings"></div>

The applicability of the linear feedback framework ( [[#5.4.5.5|Section 5.4.5.5]] ) suggests that large-scale biogeochemical feedbacks are approximately linear in the forcing from changes in CO <sub>2</sub> and climate. Nevertheless, regionally the biosphere is known to be capable of producing abrupt changes or even ‘tipping points’ ( [[#Higgins--2012|Higgins and Scheiter, 2012]] ; [[#Lasslop--2016|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 ( [[IPCC:Wg1:Chapter:Chapter-1#1.4.5|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--2008|Lenton et al., 2008]] ; [[#Steffen--2018|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.

<div id="_idContainer082" class="_idGenObjectStyleOverride-1"></div>

'''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 21 <sup>st</sup> 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 <sup>–1</sup> ).

{| class="wikitable"
|-
! '''Abrupt Chang''' '''e/Tipping Point'''

! '''Key Region(s)'''

! '''Probability to Occur in the 21st Century'''

! '''Maximum CO''' <sub>2</sub> '''or CH''' <sub>4</sub> '''Release in the 21st Century'''

! '''Principal Development Time Scale'''

! '''Maximum CO''' <sub>2</sub> '''or CH''' <sub>4</sub> '''Rate of Change Over the 21st Century'''

! '''(Ir)reversibility'''

|-
| Tropical forests dieback ( [[#5.4.9.1.1|Section 5.4.9.1.1]] )

| Amazon watershed

| Low

| &lt;200 PgC as CO <sub>2</sub> ( [[#5.4.9.1.1|Section 5.4.9.1.1]] ; ''medium confidence'' )

| Multi-decadal

| CO <sub>2</sub> : &lt;0.5 ppm yr <sup>–1</sup>

| 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

| &lt;27 Pg ( [[#5.4.9.1.2|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 ( [[#5.4.9.1.2|Section 5.4.9.1.2]] )

| Pan-Arctic

| High

| up to 240 PgC of CO <sub>2</sub> and up to 5300 Tg of CH <sub>4</sub> [[#5.4.8.1.2|Section 5.4.8.1.2]] ; ''low confidence'' )

| Multi-decadal

| CO <sub>2</sub> : ≤1 ppm yr <sup>–1</sup>

CH <sub>4</sub> : ≤10 ppb yr <sup>–1</sup>

| Irreversible at centennial time scales ( ''high confidence'' )

|-
| Methane release from clathrates ( [[#5.4.9.1.3|Section 5.4.9.1.3]] )

| Oceanic shelf

| Very low

| ''very likely'' small ( [[#5.4.9.1.3|Section 5.4.9.1.3]] )

| Multi-millennium

| CH <sub>4</sub> : ≤0.2 ppb yr <sup>–1</sup>

| Irreversible at multi-millennium time scales ( ''medium confidence'' )

|}

<div id="5.4.9.1" class="h3-container"></div>

<span id="assessment-of-biogeochemical-tipping-points"></span>
==== 5.4.9.1 Assessment of Biogeochemical Tipping Points ====

<div id="h3-39-siblings" class="h3-siblings"></div>

<div id="5.4.9.1.1" class="h4-container"></div>

<span id="forest-dieback"></span>
===== 5.4.9.1.1 Forest dieback =====

<div id="h4-5-siblings" class="h4-siblings"></div>

Published examples of abrupt biogeochemical changes in models include tropical rain forest dieback ( [[#Cox--2004|Cox et al., 2004]] ; [[#Jones--2009|Jones et al., 2009]] ; [[#Brando--2014|Brando et al., 2014]] ; [[#Le%20Page--2017|Le Page et al., 2017]] ; [[#Zemp--2017|Zemp et al., 2017]] ), and temperate and boreal forest dieback ( [[#Joos--2001|Joos et al., 2001]] ; [[#Lucht--2006|Lucht et al., 2006]] ; [[#Scheffer--2012|Scheffer et al., 2012]] ; [[#Lasslop--2016|Lasslop et al., 2016]] ; [[#5.4.3|Section 5.4.3]] ). Such transitions may be related to: (i) large-scale changes in mean climate conditions crossing particular climate thresholds ( [[#Joos--2001|Joos et al., 2001]] ; [[#Cox--2004|Cox et al., 2004]] ; [[#Lucht--2006|Lucht et al., 2006]] ; [[#Hirota--2011|Hirota et al., 2011]] ; [[#Scheffer--2012|Scheffer et al., 2012]] ; [[#Le%20Page--2017|Le Page et al., 2017]] ; [[#Zemp--2017|Zemp et al., 2017]] ); (ii) temperature and precipitation extremes ( [[#Staver--2011|Staver et al., 2011]] ; [[#Higgins--2012|Higgins and Scheiter, 2012]] ; [[#Scheffer--2012|Scheffer et al., 2012]] ; [[#Pavlov--2015|Pavlov, 2015]] ; [[#Zemp--2017|Zemp et al., 2017]] ); or (iii) possible enhancement and intermittency in fire activity ( [[#Staver--2011|Staver et al., 2011]] ; [[#Higgins--2012|Higgins and Scheiter, 2012]] ; [[#Lasslop--2016|Lasslop et al., 2016]] ; [[#Brando--2020|Brando et al., 2020]] ). Simulated changes in forest cover are a combination of the effects of CO <sub>2</sub> on photosynthesis and water-use efficiency ( [[#5.4.1|Section 5.4.1]] ), and the effects of climate change on photosynthesis, respiration and disturbance ( [[#5.4.3|Section 5.4.3]] ). In ESMs, direct CO <sub>2</sub> 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 CO <sub>2</sub> effects overwhelming the negative effects of climate change ( [[#Huntingford--2013|Huntingford et al., 2013]] ; [[#Drijfhout--2015|Drijfhout et al., 2015]] ; [[#Boulton--2017|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--2015|Reyer et al., 2015]] ; [[#Levine--2016|Levine et al., 2016]] ; [[#Sakschewski--2016|Sakschewski et al., 2016]] ), and because of possible acclimation of vegetation to warming ( [[#Good--2011|Good et al., 2011]] , 2013; [[#Lloret--2012|Lloret et al., 2012]] ; [[#Mercado--2018|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 ( [[#5.4.3.2|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--2016|Nobre et al., 2016]] ; [[#Le%20Page--2017|Le Page et al., 2017]] ).

To estimate an upper limit on the impact of Amazon forest dieback on atmospheric CO <sub>2</sub> , we consider the ''very unlikely'' limiting case of negligible direct-CO <sub>2</sub> effects ( [[#5.4.1|Section 5.4.1]] ). Emergent constraint approaches ( [[#5.4.6|Section 5.4.6]] ) may be used to estimate an overall loss of tropical land carbon due to climate change alone, of around 50 PgC per °C of tropical warming ( [[#Cox--2013|Cox et al., 2013]] ; [[#Wenzel--2014|Wenzel et al., 2014]] ). This implies an upper limit to the release of tropical land carbon of &lt;200 PgC over the 21 <sup>st</sup> century (assuming tropical warming of &lt;4°C '','' and no CO <sub>2</sub> -fertilization), which translates to dCO <sub>2</sub> /dt &lt;0.5 ppm yr <sup>–1</sup> .

Boreal forest dieback is not expected to change the atmospheric CO <sub>2</sub> concentration substantially because forest loss at the south is partly compensated by: (i) temperate forest invasion into previously boreal areas; and (ii) boreal forest gain at the north ( [[#Friend--2014|Friend et al., 2014]] ; [[#Kicklighter--2014|Kicklighter et al., 2014]] ; [[#Schaphoff--2016|Schaphoff et al., 2016]] ) ( ''medium confidence'' ). An upper estimate of this magnitude, based on statistical modelling of climate change alone, is of 27 Pg vegetation carbon loss in the southern boreal forest, which is roughly balanced by gains in the northern zone ( [[#Koven--2013|Koven, 2013]] ). Carbon release from vegetation and soil due to wildfires in boreal regions ( [[#Eliseev--2014b|Eliseev et al., 2014b]] ; [[#Turetsky--2015|Turetsky et al., 2015]] ; X.J. [[#Walker--2019|]] [[#Walker--2019|Walker et al., 2019]] ) is also not expected to change this estimate substantially because of its small present-day value of about 0.2 PgC yr <sup>–1</sup> ( [[#van%20der%20Werf--2017|van der Werf et al., 2017]] ), and because of ''likely'' increases in precipitation in boreal regions ( [[IPCC:Wg1:Chapter:Chapter-4#4.5.1|Section 4.5.1]] ).

<div id="5.4.9.1.2" class="h4-container"></div>

<span id="biogenic-emissions-following-permafrost-thaw"></span>
===== 5.4.9.1.2 Biogenic emissions following permafrost thaw =====

<div id="h4-6-siblings" class="h4-siblings"></div>

There is large uncertainty in release of GHGs from permafrost in the 21st century. The largest of these estimates implies tens to hundreds of gigatons of carbon released in the form of CO <sub>2</sub> (Box 5.1) and CH <sub>4</sub> emissions up to 100 TgCH <sub>4</sub> yr <sup>–1</sup> (Box 5.1). A carbon dioxide release of such magnitude would lead to an increase in the CO <sub>2</sub> accumulation rate in the atmosphere of ≤1 ppm yr <sup>–1</sup> . These emissions develop at a multi-decadal time scale. Assuming a CH <sub>4</sub> lifetime in the atmosphere of the order of 10 years and the associated feedback parameter of 1.34 ± 0.04 (Section 6.2.2.1), this would increase the atmospheric CH <sub>4</sub> content by about 500 ppb over the century, corresponding to a rate of ≤10 ppb yr <sup>–1</sup> . Irrespective of its origin, additional CH <sub>4</sub> accumulation of such a magnitude is not expected to modify the temperature response to anthropogenic emissions by more than a few tenths of a °C ( [[#Gedney--2004|Gedney et al., 2004]] ; [[#Eliseev--2008|Eliseev et al., 2008]] ; [[#Denisov--2013|Denisov et al., 2013]] ). Emissions from permafrost thawing are assessed in Box 5.1.

<div id="5.4.9.1.3" class="h4-container"></div>

<span id="methane-release-from-clathrates"></span>
===== 5.4.9.1.3 Methane release from clathrates =====

<div id="h4-7-siblings" class="h4-siblings"></div>

The total global clathrate reservoir is estimated to contain 1500–2000 PgC ( [[#Archer--2009|Archer et al., 2009]] ; [[#Ruppel--2017|Ruppel and Kessler, 2017]] ), held predominantly in ocean sediments, with only an estimated 20 PgC in and under permafrost ( [[#Ruppel--2015|Ruppel, 2015]] ). The present-day CH <sub>4</sub> release from shelf clathrates is &lt;10 TgCH <sub>4</sub> yr <sup>–1</sup> ( [[#Kretschmer--2015|Kretschmer et al., 2015]] ; [[#Saunois--2020|Saunois et al., 2020]] ). Despite polar amplification (Chapter 7), substantial releases from the permafrost-embedded subsea clathrates is ''very unlikely'' ( [[#Minshull--2016|Minshull et al., 2016]] ; [[#Malakhova--2017|Malakhova and Eliseev, 2017]] , 2020). This is consistent with an overall small release of CH <sub>4</sub> from the shelf clathrates during the last deglacial transition, despite large reorganizations in climate state ( [[#Bock--2017|Bock et al., 2017]] ; [[#Petrenko--2017|Petrenko et al., 2017]] ; [[#Dyonisius--2020|Dyonisius et al., 2020]] ). The long time scales associated with clathrate destabilization makes it ''unlikely'' that CH <sub>4</sub> release from the ocean to the atmosphere will deviate markedly from the present-day value through the 21st century ( [[#Hunter--2013|Hunter et al., 2013]] ), corresponding to no more than additional 20 ppb of atmospheric CH <sub>4</sub> (i.e., &lt;0.2 ppb yr <sup>–1</sup> ). Another possible source of CH <sub>4</sub> is gas clathrates in deeper terrestrial permafrost and below it ( [[#Buldovicz--2018|Buldovicz et al., 2018]] ; [[#Chuvilin--2018|Chuvilin et al., 2018]] ), which may have caused recent craters in the north of Russia ( [[#Arzhanov--2016|Arzhanov et al., 2016]] , 2020; [[#Arzhanov--2017|Arzhanov and Mokhov, 2017]] ; [[#Kizyakov--2017|Kizyakov et al., 2017]] , 2018). Land clathrates are formed at depths greater than 200 m ( [[#Ruppel--2017|Ruppel and Kessler, 2017]] ; [[#Malakhova--2020|Malakhova and Eliseev, 2020]] ), which precludes a substantial response to global warming over the next few centuries and associated emissions.

Thus, it is ''very unlikely'' that CH <sub>4</sub> emissions from clathrates will substantially warm the climate system over the next few centuries.

<div id="5.4.9.2" class="h3-container"></div>

<span id="abrupt-changes-detected-in-earth-system-model-projections"></span>
==== 5.4.9.2 Abrupt Changes Detected in Earth System Model Projections ====

<div id="h3-40-siblings" class="h3-siblings"></div>

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

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

<div id="5.4.10" class="h2-container"></div>

<span id="long-term-response-past-2100"></span>
=== 5.4.10 Long-term Response Past 2100 ===

<div id="h2-30-siblings" class="h2-siblings"></div>

The AR5 assessed with ''very high confidence'' that the carbon cycle in the ocean and on land will continue to respond to climate change and rising atmospheric CO <sub>2</sub> concentrations created during the 21 <sup>st</sup> century (WGI, Chapter 6, Executive Summary). Since AR5, experiments with the Community Earth System Model version 1 (CESM1) under the RCP8.5 extension scenario to 2300, suggest that both land and ocean carbon–climate feedbacks strengthen in time, land and ocean carbon-concentration feedbacks weaken, and the relative importance of ocean sinks versus land sinks increases ( [[#Randerson--2015|Randerson et al., 2015]] ). Under high emissions scenarios, this relative strengthening of land carbon–climate feedbacks leads the terrestrial biosphere to shift from sink to source at some point after 2100 in all of the CMIP5 ESMs and CMIP5-era Earth system models of intermediate complexity (EMICs) ( [[#Tokarska--2016|Tokarska et al., 2016]] ). The strengthening of land and ocean carbon–climate feedbacks projected beyond 2100 under high emissions scenarios offsets the declining climate sensitivity to incremental increases of CO <sub>2</sub> , leading to a net strengthening of carbon cycle feedbacks, as measured by the gain parameter, from one century to the next ( [[#Randerson--2015|Randerson et al., 2015]] ).

Figure 5.30 shows carbon cycle changes to 2300 under three SSP scenarios with long-term extensions: SSP5-8.5, SSP5-3.4-overshoot, and SSP1-2.6, for four CMIP6 ESMs and one EMIC. Under all three scenarios, all five models project a reversal of the terrestrial carbon cycle from a sink to a source. However, the reasons for these reversals under very high emissions and low/negative emissions are very different. Under the SSP5-8.5 scenario, the terrestrial carbon–climate feedback is projected to strengthen, while the carbon-concentration feedbacks weaken after emissions peak at 2100, which together drives the land to become a net carbon source after 2100 ( [[#Tokarska--2016|Tokarska et al., 2016]] ). The difference in both timing and magnitude of this transition across the ensemble, leads to an assessment of ''medium confidence'' in the likelihood and ''low confidence'' in the timing and strength, of the land transitioning from a net sink to a net source under such a scenario. Based on ''high agreement'' across all available models, we assess with ''high confidence'' that the ocean sink strength would weaken but not reverse under a long-term high emissions scenario. In the SSP5-3.4-overshoot scenario, both the terrestrial and ocean reservoirs act as transient carbon sources during the overshoot period, when net anthropogenic CO <sub>2</sub> emissions are negative and CO <sub>2</sub> concentrations are falling, and then revert to near-zero (land) or weak sink (ocean) fluxes after stabilization of atmospheric CO <sub>2</sub> . The SSP1-2.6 scenario, characterized by lower peak CO <sub>2</sub> concentrations, a smaller overshoot, and much less carbon loss from land-use change, shows instead a relaxation towards a neutral biosphere on land, and a sustained weak sink in the ocean (see also [[#5.6.2.2.1.2|Section 5.6.2.2.1.2]] ).

<div id="_idContainer084" class="Basic-Text-Frame"></div>

[[File:7f24b0a55ec5d7c4544d68574abfbdd2 IPCC_AR6_WGI_Figure_5_30.png]]

'''Figure 5.30 |''' '''Trajectories of carbon cycle dynamics for models beyond 2100.''' Shown are three scenarios: SSP5-8.5; SSP5-3.4-overshoot; and SSP1-2.6, from four ESMs (CanESM5, UKESM1, CESM2-WACCM, IPSL-CM6a-LR) and one EMIC (UVIC-ESCM, [[#Mengis--2020|Mengis et al., 2020]] ) for which extensions beyond 2100 are available. Solid lines represent the median flux value across the ensemble, and shading represents 15 <sup>th</sup> –85 <sup>th</sup> percentiles across the ensemble. Further details on data sources and processing are available in the chapter data table (Table 5.SM.6).

<div id="5.4.11" class="h2-container"></div>

<span id="near-term-prediction-of-ocean-and-land-carbon-sinks"></span>
=== 5.4.11 Near-term Prediction of Ocean and Land Carbon Sinks ===

<div id="h2-31-siblings" class="h2-siblings"></div>

The AR5 (WGI, [[IPCC:Wg1:Chapter:Chapter-11#11.3.2|Section 11.3.2]] ) assessed near-term climate predictability based on ESMs initialized from the observed climate state. Since AR5, a growing number of prediction systems have been developed based on ESMs that include the ocean and land carbon cycle components. Predictability of key physical climate variables (assessed in Chapter 4) provides a platform to establish predictive skill for interannual variations in the strength of the natural carbon sinks in response to internal climate variability. In most systems the carbon cycle components are only indirectly initialized and respond to the initialized climate variations ( [[#Li--2019|Li et al., 2019]] ). This subsection synthesizes information on predictability of the land and ocean carbon sinks using both the idealized potential predictability and the actual predictability skill measures.

Longer-term memory residing in the ocean enables predictability of the ocean carbon sink ( [[#McKinley--2017|McKinley et al., 2017]] ; [[#Li--2018|Li and Ilyina, 2018]] ). The predictive horizon of the globally integrated air–sea CO <sub>2</sub> fluxes has been assessed in perfect-model frameworks that are based on an idealized ensemble of simulations in which each ensemble member serves as a verification, while no observations are assessed. Perfect-model studies provide an estimate of the upper range of potential predictability for the integrated air–sea CO <sub>2</sub> fluxes of about two years globally and up to a decade in some regions ( [[#Séférian--2018a|Séférian et al., 2018a]] ; [[#Spring--2020|Spring and Ilyina, 2020]] ). Evidence is also emerging for predictive skill of the global air–sea CO <sub>2</sub> fluxes of up to six years based on prediction systems initialized with observed physical climate states ( [[#Ilyina--2021|Ilyina et al., 2021]] ), with a potential for even longer-term regional predictability in some regions, including the North Atlantic and subpolar Southern Ocean (H. [[#Li--2016|]] [[#Li--2016|]] [[#Li--2016|Li et al., 2016]] ; [[#Lovenduski--2019a|Lovenduski et al., 2019a]] ).

Models suggest that predictability of the air–sea CO <sub>2</sub> flux is related to predictability of ocean biogeochemical state variables such as dissolved inorganic carbon (DIC) and total alkalinity (TA; [[#Lovenduski--2019a|Lovenduski et al., 2019a]] ), as well as the mixed layer depth (H. [[#Li--2016|]] [[#Li--2016|]] [[#Li--2016|Li et al., 2016]] ). Temperature variations largely control shorter-term predictability of the ocean carbon sink, while longer-term predictability is related to non-thermal drivers such as ocean circulation and biology ( [[#Li--2019|Li et al., 2019]] ). Although there is a substantial spatial heterogeneity, initialized predictions suggest stronger multi-year variations of the air–sea CO <sub>2</sub> flux and generally tend to outperform uninitialized simulations on the global scale ( [[#Li--2019|Li et al., 2019]] ). The predictive skill of air–sea CO <sub>2</sub> flux shows a consistent spatial pattern in different models, despite the wide range of techniques used to assimilate observational information ( [[#Regnier--2013|Regnier et al., 2013]] ). ESM-based prediction systems also demonstrate predictability of other marine biogeochemical properties such as net primary production ( [[#Séférian--2014|Séférian et al., 2014]] ; [[#Yeager--2018|Yeager et al., 2018]] ; [[#Park--2019|Park et al., 2019]] ) and seawater pH ( [[#Brady--2020|Brady et al., 2020]] ).

Seasonal predictability of air-land CO <sub>2</sub> flux up to 6–8 months is driven by the state of El Niño–Southern Oscillation (ENSO) ( [[#Zeng--2008|Zeng et al., 2008]] ; [[#Betts--2018|Betts et al., 2018]] ). Fewer land carbon initialized predictions are available from decadal prediction systems, yet they tend to outperform the uninitialized simulations in capturing the major year-to-year variations, as indicated by higher correlations with the global carbon budget estimates. There is growing evidence that the potential predictive skill of air-land CO <sub>2</sub> flux is maintained out to a lead-time of two years ( [[#Lovenduski--2019b|Lovenduski et al., 2019b]] ); this predictability horizon is also supported by perfect model studies ( [[#Séférian--2018a|Séférian et al., 2018a]] ; [[#Spring--2020|Spring and Ilyina, 2020]] ). The origins of this interannual predictability are not yet fully understood. However, they seem to be associated with the oscillatory behaviour of ENSO ( [[#Séférian--2014|Séférian et al., 2014]] ) and the drivers of terrestrial carbon flux predictability, such as ecosystem respiration and gross primary production ( [[#Lovenduski--2019a|Lovenduski et al., 2019a]] ). Initialized simulations suggest that observed variability in the land carbon sink is improved through initialization of prediction systems with the observed state of the physical climate.

The predictability horizon of variations in atmospheric CO <sub>2</sub> growth rate is not yet fully established in the literature. However, predictive skill of the land and ocean carbon sinks show a potential to establish predictability of variations in atmospheric CO <sub>2</sub> up to two years in advance in the initialized prediction systems, with an upper bound of up to three years in a perfect-model study ( [[#Spring--2020|Spring and Ilyina, 2020]] ); this skill is primarily limited by the terrestrial carbon sink predictability.

<div id="5.5" class="h1-container"></div>

<span id="remaining-carbon-budgets"></span>